FUNCTIONAL GENOMICS OF NERVOUS SYSTEM DEVELOPMENT AND 
DISEASE
by
MICHAEL RYAN MILLER
A DISSERTATION
Presented to the Department of Biology
and the Graduate School of the University of Oregon
in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
December 2011
DISSERTATION APPROVAL PAGE
Student: Michael Ryan Miller
Title: Functional Genomics of Nervous System Development and Disease
This dissertation has been accepted and approved in partial fulfillment of the 
requirements for the Doctor of Philosophy degree in the Department of Biology by:
Dr. John Postlethwait Chair
Dr. Chris Doe Advisor
Dr. Bruce Bowerman Member
Dr. Patrick Phillips Member
Dr. Tom Stevens Outside Member
and
Kimberly Andrews Espy Vice President for Research & Innovation/Dean of the 
Graduate School
Original approval signatures are on file with the University of Oregon Graduate School.
Degree awarded December 2011
ii
© 2011 Michael Ryan Miller
iii
DISSERTATION ABSTRACT
Michael Ryan Miller
Doctor of Philosophy
Department of Biology
December 2011
Title: Functional Genomics of Nervous System Development and Disease
The goal of functional genomics is to elucidate the relationship between an 
organism's genotype and phenotype. A key characteristic of functional genomics is the 
use of genome-wide approaches as opposed to more traditional single-gene approaches. 
Genome-wide expression profiling is used to investigate the dynamic properties of 
transcriptomes, provides insights into how biological functions are encoded in genomes, 
and is an important technique in functional genomics. This dissertation describes the use 
of genome-wide expression profiling and other functional genomics techniques to address 
a variety of biological questions related to development and disease of the nervous 
system. Our results reveal novel and important insights into nervous system development 
and disease and demonstrate the power of functional genomics approaches for the study 
of nervous system biology. This dissertation also describes a novel technique called TU-
tagging that facilitates cell type-specific RNA isolation from intact complex tissues. The 
isolation of RNA from specific cell types within a complex tissue is a major limiting 
factor in the application of genome-wide expression profiling, and TU-tagging can be 
used to address a wide array of interesting and important biological questions. 
This dissertation includes previously published and unpublished co-authored 
material.
iv
CURRICULUM VITAE 
NAME OF AUTHOR: Michael Ryan Miller 
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED: 
University of Oregon, Eugene
DEGREES AWARDED: 
Doctor of Philosophy, Biology, 2011, University of Oregon 
Bachelor of Science, Biology, 2006, University of Oregon 
AREAS OF SPECIAL INTEREST: 
Genetics and Genomics
PROFESSIONAL EXPERIENCE: 
Graduate researcher, Department of Biology, University of Oregon, Eugene, 
2006-2011 
Teaching assistant, Department of Biology, University of Oregon, Eugene, 2006-
2007 
Undergraduate researcher, Department of Biology, University of Oregon, Eugene, 
2004-2006
GRANTS, AWARDS, AND HONORS: 
Aquaculture Workshop Student Travel Award, International Plant and Animal 
Genome Conference XIX, 2011 
Pete von Hippel Graduate Education Endowment Award, University of Oregon 
Institute of Molecular Biology, 2010 
Best Student Poster Presentation, Society for Developmental Biology Northwest 
Regional Meeting, 2009 
Graduate Research Fellowship, National Science Foundation, 2008-2011 
v
Integrative Graduate Education and Research Traineeship Program Associate 
Award, National Science Foundation, 2007-2010 
Developmental Biology Traineeship, National Institutes of Health, 2007-2008 
Graduate Teaching Fellowship, University of Oregon, 2006-2007
PUBLICATIONS:
Carney, T.D., Miller, M.R., Robinson, K.J., Bayraktar, O.A., Osterhout, J.A., Doe, C.Q. 
Functional genomics identifies neural stem cell sub-type expression profiles and 
genes regulating neuroblast homeostasis. Dev. Biol. Under review. 
Miller, M.R., Brunelli, J.P., Wheeler, P.A., Liu, S., Rexroad, C.E. 3rd, Palti, Y., Doe, 
C.Q., Thorgaard, G.H. A conserved haplotype controls parallel adaptation in 
geographically distant salmonid populations. Mol. Ecol. In press. 
O'Rourke, S.M., Yochem, J., Connolly, A.A., Price, M.H., Carter, L., Lowry, J.B., 
Turnbull, D.W., Kamps-Hughes, N., Stiffler, N., Miller, M.R., Johnson, E.A., 
Bowerman, B. Rapid mapping and identification of mutations in Caenorhabditis 
elegans by RAD mapping and genomic interval pull-down sequencing. Genetics. 
In press. 
Liu, C., Sage, J.C., Miller, M.R., Verhaak, R.G.W., Hippenmeyer, S., Vogel, H., Forman, 
O., Bronson, R.T., Nishiyama, A., Luo, L., Zong, H., 2011. Mosaic analysis with 
double markers reveals tumor cell of origin in glioma. Cell 146, 209-221. 
Tran, K.D., Miller, M.R., Doe, C.Q., 2010. Recombineering Hunchback identifies two 
conserved domains required to maintain neuroblast competence and specify early-
born neuronal identity. Development 137, 1421-1430. 
Miller, M.R., Robinson, K.J., Cleary, M.D., Doe, C.Q., 2009. TU-tagging: cell type-
specific RNA isolation from intact complex tissues. Nat. Methods 6, 439-441. 
Yokoi, H., Yan, Y.L., Miller, M.R., BreMiller, R.A., Catchen, J.M., Johnson, E.A., 
Postlethwait, J.H., 2009. Expression profiling of zebrafish sox9 mutants reveals 
that Sox9 is required for retinal differentiation. Dev. Biol. 329, 1-15. 
Lewis, Z.A., Shiver, A.L., Stiffler, N., Miller, M.R., Johnson, E.A., Selker, E.U., 2007. 
High density detection of restriction site associated DNA (RAD) markers for 
rapid mapping of mutated loci in Neurospora. Genetics 177, 1163-1171. 
Miller, M.R., Atwood, T.S., Eames, B.F., Eberhart, J.K., Yan, Y.L., Postlethwait, J.H., 
Johnson, E.A., 2007. RAD marker microarrays enable rapid mapping of zebrafish 
mutations. Genome Biol. 8, R105. 
vi
Miller, M.R., Dunham, J.P., Amores, A., Cresko, W.A., Johnson, E.A., 2007. Rapid and 
cost-effective polymorphism identification and genotyping using restriction site 
associated DNA (RAD) markers. Genome Res. 17, 240-248.
Johnson, E.A., Liu, G., Miller, M.R., 2006. Methods of mapping polymorphisms and 
polymorphism microarrays. Patent Pub. No. WO/2006/122215.
vii
ACKNOWLEDGMENTS
I thank B. Barut, S. Johnson, F. Pignoni, J. Rawls, J. Skeath, U. Walldorf, L. Zon, 
the Bloomington Stock Center, the Developmental Studies Hybridoma Bank, the Vienna 
Drosophila RNAi, and the Zebrafish International Resource Center for providing 
reagents; J. Boone, J. Dugas, J. Gleason, A. Henner, P.K. Loi, J. Murphy, A. Rapp, A. 
Selix, A. Starks, and M. Valle for technical support; B. Bowerman, C. Canestro, C. Doe, 
R. Galvao, T. Herman, W. Hong, M. Kohwi, S. Lai, Z. Lewis, S. O'Rourke, M. Raff, R. 
Read, B. Tasic, X. Wu, and Y. Zhu for critical comments on chapters of this dissertation; 
the European Molecular Biology Organization, Howard Hughes Medical Institute, 
Human Frontier Science Program Organization, National Center for Research Resources, 
National Institutes of Health, National Science Foundation, Pew Charitable Trusts, and 
Swiss National Science Foundation for financial support; my co-authors on chapters of 
this dissertation for the exciting and fruitful collaborations; and my dissertation advisory 
committee and other members of the University of Oregon Biology Department for 
invaluable advice and support throughout my tenure as a student.
viii
TABLE OF CONTENTS
Chapter
I. INTRODUCTION...................................................................................................
II. EXPRESSION PROFILING OF ZEBRAFISH SOX9 MUTANTS REVEALS 
THAT SOX9 IS REQUIRED FOR RETINAL DIFFERENTIATION ..................
Introduction.......................................................................................................
Methods.............................................................................................................
Results...............................................................................................................
Discussion..........................................................................................................
III. FUNCTIONAL GENOMICS IDENTIFIES NEURAL STEM CELL SUB-
TYPE EXPRESSION PROFILES AND GENES REGULATING 
NEUROBLAST HOMEOSTASIS .........................................................................
Introduction.......................................................................................................
Methods.............................................................................................................
Results...............................................................................................................
Discussion..........................................................................................................
IV. MOSAIC ANALYSIS WITH DOUBLE MARKERS REVEALS TUMOR 
CELL OF ORIGIN IN GLIOMA ...........................................................................
Introduction.......................................................................................................
Methods.............................................................................................................
Results...............................................................................................................
Discussion..........................................................................................................
V. TU-TAGGING: CELL TYPE-SPECIFIC RNA ISOLATION FROM INTACT 
COMPLEX TISSUES.............................................................................................
Introduction.......................................................................................................
Methods.............................................................................................................
Results...............................................................................................................
Discussion..........................................................................................................
Page
1
4
4
7
11
23
31
31
35
38
45
50
50
53
68
79
85
85
86
91
94
ix
Chapter
APPENDICES
 A. TABLES............................................................................................................
 B. FIGURES..........................................................................................................
REFERENCES CITED.................................................................................................
Page
96
98
128
x
LIST OF FIGURES
Figure
1. Differential regulation of col11a2 expression by sox9a and sox9b.........................
2. Expression of candidate sox9 down-stream targets analyzed by in situ 
hybridization analysis in 2 dpf embryos..................................................................
3. Candidate target genes calb2a and calb2b are co-orthologs of tetrapod Calb2 
and their expression was lost in the sox9b mutant retina........................................
4. Candidate genes were expressed in different layers in the developing retina.........
5. Expression of candidate genes in retina at 3 dpf.....................................................
6. Fluorescent antibody staining of retinal cell types in wild-type and sox9 single 
and double mutant larvae.........................................................................................
7. Expression of candidate genes in the pectoral fin bud............................................
8. Using ectopic self-renewal mutants for expression profiling of neuroblasts..........
9. Results of cluster analysis.......................................................................................
10. Differential expression of genes excluded from type II neuroblasts.......................
11. Identification of a cluster with type II-biased expression.......................................
12. Gene ontology terms enriched in each group..........................................................
13. RNAi screen identifies neuroblast homeostasis genes............................................
14. MADM-based glioma model allows phenotypic analysis at single-cell 
resolution.................................................................................................................
15. MADM-mediated sporadic concurrent inactivation of p53 and NF1 in 
embryonic NSCs reveals the entire process of gliomagenesis................................
16. Analysis at a pretransforming stage of gliomagenesis suggests that OPCs rather 
than NSCs serve as the cell of origin.......................................................................
17. MADM-generated glioma cells exhibit many OPC features..................................
Page
98
99
101
103
104
106
107
108
109
110
111
112
114
115
116
117
119
xi
Figure
18. Spatial analyses of early lesions based on perineuronal cytoarchitecture as a 
landmark suggest that gliomas initiate at brain regions away from the SVZ..........
19. OPCs can be directly transformed into malignant glioma.......................................
20. Comparative pathological analyses of NSC- and NG2-Cre induced tumors 
suggest that tumor cell morphology is highly dependent on the location rather 
than initially mutated cell types...............................................................................
21. TU-tagging: overview and cell type-specific labeling............................................
22. Cell type-specific RNA isolation and analysis........................................................
Page
121
123
125
126
127
xii
LIST OF TABLES
Table
1. Candidate genes for down-stream targets of sox9a and sox9b in zebrafish............
2. Mutants affecting brain neuroblast numbers used in this study...............................
3. Representation of cell type-specific genes within microarray groups A, B, and 
C..............................................................................................................................
Page
96
96
97
xiii
CHAPTER I
INTRODUCTION
This dissertation includes previously published and unpublished co-authored 
material. Chapter II was previously published with co-authors Hayato Yokio, Yi-Lin Yan, 
Ruth BreMiller, Julian Catchen, Eric Johnson, and John Postlethwait. Chapter III is 
unpublished material with co-authors Travis Carney, Kristin Robinson, Omer Bayraktar, 
Jessica Osterhout, and Chris Doe. Chapter IV was previously published with co-authors 
Chong Liu, Johnathan Sage, Roel Verhaak, Simon Hippenmeyer, Hannes Vogel, Oded 
Foreman, Roderick Bronson, Akiko Nishiyama, Liqun Luo, and Hui Zong. Chapter V 
was previously published with co-authors Kristin Robinson, Michael Cleary, and Chris 
Doe.
The availability of complete genome sequences for many organisms has led to the 
emergence of a discipline called functional genomics. The overarching goal of functional 
genomics is to understand the biological function of sequence elements and in turn 
elucidate the relationship between an organism's genotype and phenotype. This lofty goal 
requires the development and application of global experimental approaches, and one 
characteristic of functional genomics is the use of genome-wide approaches as opposed to 
more traditional single-gene approaches. Genome-wide gene-expression profiling is one 
of the most important and widely-used techniques in functional genomics and is used to 
investigate the dynamic properties of cellular transcriptomes. This technique provides a 
wealth of data and can be used investigate diverse biological processes.
Chapters II, III, and IV of this dissertation describe the use of genome-wide 
expression profiling and other functional genomics techniques to address a variety of 
1
biological questions related to development and disease of the nervous system. Chapter II 
describes an investigation of the mechanism of action of the Sox9 transcription factor 
which is mutated in humans with the severe disorder campomelic dysplasia. We used 
genome-wide expression profiling followed by other methods to compare wild-type 
zebrafish embryos to mutants lacking Sox9 activity and discovered novel roles for Sox9 
in regulating Müller glia and photoreceptor cell numbers and organizing the neural retina. 
Chapter III describes an investigation of the genes regulating lineage pattern, tumor 
susceptibility, and homeostasis in neural stem cells (NSCs) of the Drosophila central 
nervous system (CNS). We used genome-wide expression profiling and found that the 
differential expression of surprisingly few genes can have profound effects on NSC 
lineage pattern and tumor susceptibility. We also performed an RNAi-based functional 
screen of genes identified with the expression profiling and identified many novel 
regulators of NSC homeostasis. Chapter IV describes an investigation of the cell type of 
origin of cancer in a mouse model of the deadly brain cancer malignant glioma. We used 
a variety of experiments, including genome-wide expression profiling, and found that 
oligodendrocyte precursor cells (OPCs) as opposed to other NSC-derived lineages or 
NSCs themselves were the glioma cell of origin. These three studies reveal novel and 
important insights into nervous system development and disease and demonstrate the 
power of functional genomics approaches for the study of nervous system biology. 
Chapter V of this dissertation describes a novel technique called TU-tagging that 
facilitates cell type-specific RNA isolation from intact complex tissues. We demonstrated 
that a combination of spatially restricted uracil phosphoribosyltransferase (UPRT) 
expression with 4-thiouracil delivery can be used to label and purify RNA from specific 
2
cell types within the Drosophila CNS. In this technique, 4-thiouracil is modified and 
subsequently incorporated into newly synthesized RNA only in cells expressing UPRT. 
Thio-substituted nucleotides are not a natural component of nucleic acids, and the 
resulting thio-labeled RNA can be readily tagged and purified. Therefore, even if RNA is 
isolated from the whole organism, RNA from the cells expressing UPRT can be 
recovered by purifying labeled RNA. This method is useful for isolating RNA from cell 
types that are difficult to isolate by dissection or dissociation methods and should work in 
many organisms, including mammals and other vertebrates. The isolation of RNA from 
specific cell types within a complex tissue is a major limiting factor in the application of 
genome-wide expression profiling, and TU-tagging can be used to address a wide array 
of interesting and important biological questions.
3
CHAPTER II
EXPRESSION PROFILING OF ZEBRAFISH SOX9 MUTANTS REVEALS THAT 
SOX9 IS REQUIRED FOR RETINAL DIFFERENTIATION
This work was published in volume 329 of the journal Developmenal Biology in 
May 2009. Hayato Yokio and myself performed the microarray-based expression 
analysis. Yi-Lin Yan, Hayato Yokio, and Ruth BreMiller performed the in situ 
hybridization and immunohistochemistry analysis. Julian Catchen performed the 
phylogenetic analysis. John Postlethwait and Eric Johnson were the principle 
investigators for this work. 
Introduction
Haploinsufficiency for SOX9 results in the human disease campomelic dysplasia 
(CD), a syndrome characterized by skeletal abnormalities and male-to-female sex 
reversal (Foster et al., 1994; Wagner et al., 1994). This clinical phenotype shows that 
Sox9 function is important for skeletal development and male sex differentiation, and 
these processes have been foci of Sox9 research (Bi et al., 1999; Marshall and Harley, 
2000; Vidal et al., 2001). CD patients show various other clinical features, including 
micrognathia, cleft palate, hypoplastic lungs, defects in olfactory tissues, and heart and 
renal malformations (Houston et al., 1983). Most patients die in early infancy because of 
respiratory problems, and some survivors show hearing loss and myopia (Mansour et al., 
2002). These descriptions identify various essential functions of SOX9, for example, in 
neural crest cells (NCC), heart, lung, ear, and other tissues. Intensive analyses of Sox9 
using conditional knockout mice and other model organisms have also assigned functions 
to Sox9, including the differentiation of neural crest cells (Spokony et al., 2002; Mori-
4
Akiyama et al., 2003; Saint-Germain et al., 2004; Cheung et al., 2005; Sakai et al., 2006), 
glial cells in the spinal cord (Stolt et al., 2003), paneth cells in the intestine (Bastide et al., 
2007; Mori-Akiyama et al., 2007), notochord (Barrionuevo et al., 2006), heart valves 
(Akiyama et al., 2004; Lincoln et al., 2007), pancreas (Seymour et al., 2007), otic vesicle 
(Barrionuevo et al., 2008) and gonad (Chaboissier et al., 2004). Sox9 is now recognized 
as a multi-functional gene that plays essential roles in various tissues during vertebrate 
development.
Sox9 encodes a transcription factor of the Sry-related HMG box (Sox) family that 
binds to cis-regulatory DNA elements to control the transcription of down-stream target 
genes. Genes shown to be regulated by Sox9 include Col2a1 and Col11a2 in the skeleton 
and Amh/Mis in the gonad (Bell et al., 1997; Lefebvre et al., 1997; Ng et al., 1997; 
Bridgewater et al., 1998; De Santa Barbara et al., 1998; Arango et al., 1999; Bi et al., 
1999). Recent microarray analysis using cultured human cells revealed that the calcium 
binding proteins S100A1 and S100B are transcriptional targets of SOX9 and its 
coactivators SOX5 and SOX6 (Saito et al., 2007). Given the multi-functionality of Sox9, 
however, most of its targets remain unidentified. Identification of Sox9 target genes will 
improve our understanding of the biological mechanisms by which Sox9 functions to 
regulate development.
Zebrafish has two copies of Sox9, called sox9a and sox9b (Chiang et al., 2001). 
These co-orthologs of mammalian Sox9 resulted from a whole genome duplication event 
that occurred after teleost lineage segregated from the tetrapod lineage and before the 
diversification of teleost fish (Amores et al., 1998; Postlethwait et al., 1998; Taylor et al., 
2003; Jaillon et al., 2004). After the genome duplication event, duplicate gene copies 
5
often evolved in a paralog-specific manner involving, at least in part, subfunction 
partitioning (Force et al., 1999; Postlethwait et al., 2004). The expression patterns of 
sox9a and sox9b overlap in some regions and are gene-specific in other domains, and the 
sum of their expression patterns is similar to the mouse Sox9 expression pattern (Chiang 
et al., 2001; Cresko et al., 2003; Yan et al., 2005).
The zebrafish mutation jellyfish (jef), which disrupts sox9a, causes skeletal defects 
due to abnormal cartilage formation (Yan et al., 2002). The zebrafish sox9b deletion 
mutant shows reduction of cartilage, and the sox9a;sox9b double mutant shows more 
severe defects revealing synergistic, additive and redundant functions of the two 
zebrafish co-orthologs of Sox9 (Yan et al., 2005). The partitioning of Sox9 subfunctions 
between the two zebrafish co-orthologs facilitates the analysis of Sox9 functions that 
were difficult to analyze in mammals due to pleiotropy (Yan et al., 2005). In mouse, 
heterozygous mutants (Sox9+/−) showed skeletal abnormalities and cleft palate, and die 
perinatally because of haploinsufficiency, as do most human CD patients (Bi et al., 2001); 
thus, homozygous mutant mouse embryos are not available. Fortunately, tissue-specific 
conditional knockout systems allow the investigation of Sox9 function in specific tissues 
(Kist et al., 2002; Chaboissier et al., 2004). But if Sox9 acts in a tissue that has not been 
examined by tissue-specific knockouts, we will not learn the full role of Sox9 in 
development. Making use of the zebrafish sox9a and sox9b mutants, whose heterozygotes 
are viable and fertile, we performed microarray analysis to compare the expression 
profile between homozygous mutant and wild-type embryos. Microarray is a powerful 
technique that enables us to conduct a genome wide screening of changes in gene 
expression between different biological groups, however at the same time, any detected 
6
changes in gene expression may be due to secondary effects, including indirect non-cell-
autonomous effects.
We evaluated potential Sox9 targets identified by microarray by comparing the 
expression patterns of putative targets with the expression patterns of sox9a and sox9b. 
We also tested for reduced expression of putative targets in sox9 mutant embryos by in 
situ hybridization. Validated downstream candidate targets of sox9 included previously 
known targets such as collagen type II alpha-1a (col2a1a) and collagen type XI alpha-2 
(col11a2) (Bell et al., 1997; Lefebvre et al., 1997; Ng et al., 1997; Bridgewater et al., 
1998; Bi et al., 1999), and novel targets, including cone rod homeobox (crx), retinoschisis  
1 (rs1), calbindin 2 (calb2) genes and other genes expressed in developing retina. The 
demonstration that sox9b mutant embryos have eye defects supports the role of sox9 
candidate targets. Sox9 is expressed in the retina of developing mouse embryos 
(Ihanamäki et al., 2002), but its function in the vertebrate retina has not been 
characterized. Thus, microarray analysis using zebrafish mutants coupled to mutant 
phenotypes, uncovered a previously unidentified function of Sox9. We also present here 
the detailed analyses of eye defects observed in the sox9b mutants and show the 
importance of Sox9 in retinal differentiation.
Methods
Zebrafish
Zebrafish were maintained under standard conditions (Westerfield, 2000) and 
embryos were staged as described (Kimmel et al., 1995). The sox9a mutation jeftw37, the 
sox9b deletion mutant b971, and the sox9atw37;sox9bb971 double mutant stock have been 
7
described previously (Piotrowski et al., 1996; Yan et al., 2002; Yan et al., 2005). Embryos 
were collected from matings of sox9a;sox9b double heterozygotes. At 2 days post 
fertilization (dpf), sox9b mutant embryos are distinguished from wild-type embryos by a 
curly tail phenotype, and sox9a;sox9b double mutant embryos are distinguished by lack 
of otic vesicles, while sox9a mutant embryos are morphologically indistinguishable from 
wild-type embryos until 3 dpf. All work with vertebrate animals was approved by the 
University of Oregon Institutional Animal Care and Use Committee.
RNA extraction
Total RNA was extracted from embryos using TRI reagent. RNA quantity was 
measured using NanoDrop ND-1000 spectrophotometer. Approximately 1 μg of total 
RNA was extracted from individual embryos at 2 dpf.
Microarray analysis
Zebrafish DNA microarray containing 16,399 65-mer oligonucleotides 
representing 16,228 gene clusters and 171 control features was kindly provided by Steve 
Johnson and John Rawls (Rawls et al., 2004). The gene clusters arise from various 
developmental stages and organs, including maternal RNAs, shield, early gastrulation and 
segmentation stage, and various adult tissues and organs, including brain, heart, kidney, 
retina, ovary and testis.
For each experiment, 20 μg of total RNA from wild-type or mutant embryos was 
used to synthesize cDNA labeled with Cy3 or Cy5-dUTP using SuperScript III reverse 
transcriptase. The Cy3 labeled cDNA was mixed with Cy5 labeled cDNA and purified 
with QIAquick PCR purification kit, dried in a speed vac and resuspended in 
hybridization buffer (50% formamide, 5 × SSC, 1% SDS, 1 mg/ml calf thymus DNA). 
8
After denaturing, labeled cDNA was applied onto the microarray and covered with a 
coverslip, incubated in a hybridization chamber, and hybridized in a 42 °C water bath 
overnight. Following the hybridization, the microarray was washed with an SSC series 
and dried by centrifugation. The microarray was immediately scanned using Genepix 
4000B scanner. Three sets of experiments with dye-swap, in total six hybridizations, were 
conducted using independently isolated RNA samples from three different sets of animals 
as biological replicates. Each microarray was scanned with two or three different laser 
intensities, depending on the variation in signal intensities, to recover data of wider range 
of gene expression.
GenePix Pro 3.0 image analysis software was used to measure fluorescence signal 
intensities of the array elements. Ratios were normalized to the overall difference in 
signal intensities from each channel. Microarray data were transferred to Microsoft 
Excel, which was also used for calculations and data sorting. The “Median of Ratios” 
(MR, the median of Cy3 and Cy5 ratios taken across the array element) was used for the 
analyses. Each microarray has two or three scan results (MRs) obtained with different 
laser intensities, from which the “Median of the MR” (MeMR) and the “Maximum of the 
MR” (MaMR) were calculated. Data from each hybridization were merged and sorted 
according to the Median of the MeMR. The probability of rejecting the hypothesis that 
gene expression was not changed between wild-type and mutant (log2 = 0) was 
calculated by two tailed Student's t-test. Candidate genes were picked from the list of the 
summary result of the microarray. The results from each single experiment (MeMR and 
MaMR) and scanned images were also considered. Genes that showed inconsistent 
results in the dye-swap experiment were removed from further analyses.
9
Cloning of candidate genes and whole-mount in situ hybridization
Candidate genes were cloned by RT-PCR, except for klf2b, which was kindly 
provided by Bruce Barut and Len Zon (Oates et al., 2001), and calb2a and calb2b, which 
were obtained from the I.M.A.G.E. Consortium EST collection. PCR primers were 
designed based on previous descriptions or genome information from the Ensembl 
database (Flicek et al., 2008). PCR products were cloned into pCR4-TOPO vector. DIG-
labeled RNA probe was synthesized with T3 or T7 RNA polymerase. Whole-mount in 
situ hybridization was performed as described in Yan et al. (2005).
Phylogenetic analysis
Amino acid sequences were aligned using CLUSTAL X (Thompson et al., 1997) 
and the alignment was further checked by eye using Seaview software (Galtier et al., 
1996). A neighbor-joining (NJ) tree (Saitou and Nei, 1987) was plotted using NJplot 
(Perrière and Gouy, 1996).
Sectioning of the whole-mount in situ samples and in situ hybridization on sections
Samples from whole-mount in situ hybridization were embedded in paraffin and 
sectioned in the Histology Facility of the University of Oregon. In situ hybridization on 
cryosections were performed as described (Rodríguez-Marí et al., 2005).
Immunohistochemistry and confocal observation
Retinal Müller glial cells were labeled by anti-carbonic anhydrase antibody 1:200 
(Jensen et al., 2001), the photoreceptor layer was labeling using zpr-3 antibody 1:200 
from the Zebrafish International Resource Center (ZIRC) (Jensen et al., 2001), and 
amacrine cells were labeled by anti-GABA antibody 1:1000 (Sandell et al., 1994). 
Nuclear layers of the retina were stained by the nuclear dye TO-PRO-3 iodide. 
10
Fluorescent images were analyzed using Bio-Rad Radiance 2100MP confocal microscope 
system.
Results
Identification of Sox9 down-stream targets
To identify genes under control of Sox9 during vertebrate embryogenesis, we 
compared the gene expression profile of wild-type and sox9a;sox9b double mutant 
zebrafish embryos at 2 dpf using a zebrafish oligonucleotide microarray (Rawls et al., 
2004). The 2 dpf time point is late enough so that double mutant embryos are 
morphologically distinguishable from wild-type and single mutant embryos, but early 
enough so that most secondary effects will not have yet taken place; nevertheless, some 
secondary downstream genes may also be detected because sox9 genes express as early 
as 1 dpf (Chiang et al., 2001). The microarray represents 16,228 gene clusters from EST 
collections of various embryonic stages and a variety of adult organs, including brain, 
eye, heart, kidney, liver, testis and ovary. To minimize erroneous expression results, three 
sets of experiments with dye-swap were conducted using independently isolated RNA 
samples. Data were analyzed and sorted with GenePix Pro software and Microsoft Excel, 
and candidates were picked according to the criteria described in Methods. Because Sox9 
is a transcription activator (Südbeck et al., 1996), we focused on genes that are down-
regulated in mutant embryos.
The b971 mutant allele of sox9b used in this study deletes sox9b and some 
surrounding genes including sox8 (Yan et al., 2005). Because the morpholino knock 
down of sox9b phenocopies the sox9b971 defects, the phenotype observed in sox9b971 at 
11
this developmental stage is due solely to loss of sox9b function (Yan et al., 2005). The 
sox8 gene is not expressed at this time in development (Yan et al., 2005). The Ensembl 
zebrafish genome database (Flicek et al., 2008) was used to evaluate the genomic 
position of candidate genes. Although pvalb1 (parvalbumin 1) and pvalb4, pvalb8 
consistently appeared as down-regulated genes, they are removed by the sox9b971 
deletion, and so are excluded from the candidates.
To test whether candidates derived from the microarray analysis were valid, we 
utilized in situ hybridization in wild-type and mutant embryos to evaluate potential Sox9 
targets. We cloned the following candidate genes: calbindin 2a (calb2a), calbindin 2b 
(calb2b), collagen type II alpha-1a (col2a1a), collagen type XI alpha-2 (col11a2), cone 
rod homeobox (crx), crystallin alpha A (cryaa), crystallin beta B1 (crybb1), crystallin  
gamma M2c (crygm2c), kruppel like factor 2b (klf2b), neurod, retinoschisis (X-linked,  
juvenile) 1 (rs1), sox4a, syndecan 2, visual system homeobox 1 (vsx1) and unknown 
ESTs, AI585100, AI722369, AI942960 and AW777642. The expression of candidate 
genes was compared among wild-type, sox9a mutant, sox9b mutant and sox9a;sox9b 
double mutant embryos at 2 dpf, the same stage used for the microarray analysis. We 
describe below genes identified in the microarray experiments whose expression change 
was verified in sox9 mutant embryos. Candidate genes could be categorized into two 
major groups: skeletogenesis and retinal development (Table 1; see Appendix A for all 
tables).
col11a2 expression is regulated by both zebrafish Sox9 co-orthologs
The collagen genes col2a1a and col11a2 showed 3.148 and 1.983 times higher 
expression in wild-type compared with double mutants, respectively, and dye-swap 
12
experiments confirmed results. Because their mouse orthologs are known as direct targets 
of Sox9 (Bell et al., 1997; Lefebvre et al., 1997; Ng et al., 1997; Bridgewater et al., 1998; 
Bi et al., 1999; Liu et al., 2000), and because col2a1a expression is known to be reduced 
in sox9 mutants (Yan et al., 2005), these results validate our experimental strategy. Here, 
we tested whether col11a2 is a target of sox9a and/or sox9b by analyzing the expression 
of col11a2 in wild-type and sox9 mutants.
In situ hybridization experiments showed that at 2 dpf, col11a2 was expressed in 
neurocranial cartilages (eg. ethmoid plate), pharyngeal cartilages (eg, mandibular 
cartilage), otic vesicle, and pectoral fin bud (Figs. 1A, E; see Appendix B for all figures). 
In sox9a mutants, expression of col11a2 disappeared from the neurocranium and 
pharyngeal arches, and was reduced in the opercle (Figs. 1B, F). This expression pattern 
was consistent with the previously described sox9a mutant phenotype in which all 
neurocranial cartilage elements and most pharyngeal arch cartilages were missing from 
sox9a mutants; dermal bone elements, however, were present but small in sox9a mutant 
embryos (Piotrowski et al., 1996; Yan et al., 2002; Yan et al., 2005). The expression of 
col11a2 was also affected in sox9b mutants, but in a fashion different than in sox9a 
mutants: col11a2 expression was greatly reduced in the mandibular and ceratohyal 
cartilages in sox9b mutants, but was retained in the ethmoid plate and trabeculae of the 
neurocranium (Figs. 1C, G), consistent with the sox9b mutant phenotype (Yan et al., 
2005). In sox9a;sox9b double mutants, expression of col11a2 disappeared in all 
neurocranial and pharyngeal cartilages. Because the otic vesicle is usually missing in 
double mutants (Yan et al., 2005), col11a2 expression in the otic vesicle vanished in 
sox9a;sox9b double mutant embryos. Expression of col11a2 in the pectoral fin bud was 
13
greatly reduced in the sox9a;sox9b double mutant (Figs. 1D, H).
Expression patterns of the two zebrafish sox9 genes can help explain the observed 
phenotypic differences between the sox9a and sox9b mutants: at 2 dpf, sox9a is expressed 
in the ethmoid plate, mandibular and ceratohyal (Fig. 1I), whereas sox9b is expressed in 
the mandibular and ceratohyal cartilages, but not in the ethmoid plate (Fig. 1J). These 
expression patterns are consistent with the col11a2 expression phenotype in the mutants: 
col11a2 expression in the neurocranium depends on sox9a function, but expression in the 
pharyngeal cartilages depends on both sox9a and sox9b. We conclude that both sox9a and 
sox9b regulate col11a2 in a tissue-specific fashion related to the sox9 genes' expression 
domains. Thus, the regulatory relationship between Sox9 and Col11a2 is conserved from 
teleost fish to mammals.
Regulation of retinal genes by Sox9
A group of genes expressed in the eye was down regulated in mutant embryos at 2 
dpf. Because the size of the eye was reduced in sox9b and sox9a;sox9b double mutant 
embryos compared to wild-type and sox9a mutant embryos, we sought to test the 
hypothesis that sox9b is required for eye development.
In situ hybridization experiments showed broad expression of crx throughout 
wild-type retinas at 2 dpf (Fig. 2A). The expression of crx was severely affected in retinas 
of sox9b mutants and sox9a;sox9b double mutant embryos (Figs. 2B, C), except for a 
small patch of cells in the ventral part of the retina (Figs. 2B, C). Crx is a member of the 
otd/Otx family of paired-like homeobox genes and is essential for retinal photoreceptor 
cell differentiation and maintenance in mouse (Furukawa et al., 1997; Furukawa et al., 
1999). Zebrafish crx is expressed in the epiphysis and in proliferating retinal progenitors 
14
(Liu et al., 2001) and may function in the differentiation of photoreceptors and other cells 
of the retina (Shen and Raymond, 2004).
In wild-type embryos (Fig. 2D), rs1 was expressed strongly throughout the entire 
retina (Thisse and Thisse, 2004). In contrast, in sox9b mutants and double mutants (Figs. 
2E, F), rs1 expression was retained only in the ventral part of the retina in the same 
domain as found previously for crx (Figs. 2E, F). Rs1 is a secreted protein with putative 
functions in cell adhesion and/or signaling (Wu et al., 2005; Molday et al., 2007). 
Mutations in the human RS1 gene cause X-linked retinoschisis, a disease involving 
intraretinal splitting due to degeneration (Sauer et al., 1997). Zebrafish rs1 was cloned in 
the Zebrafish Gene Collection Project (ZGC), and the ZFIN database reported its 
expression pattern to include a few cells in branchial arches, epiphysis, and dorsal 
telencephalon in addition to the retina (Thisse and Thisse, 2004).
Consistent with previous work (Passini et al., 1997; Passini et al., 1998), we found 
expression of vsx1 in the developing retina of wild-type embryos (Fig. 2G). In the retina 
of sox9b and sox9a;sox9b double mutant embryos, expression of vsx1 was down-
regulated, with the dorsal and caudal portions of the eye most severely reduced (Figs. 2H, 
I). Vsx1 is a paired-like homeobox protein, first isolated from goldfish retina (Levine et 
al., 1997). Gene knockout analyses in mouse showed that Vsx1 is required for retinal 
cone bipolar cell differentiation (Chow et al., 2004; Ohtoshi et al., 2004).
Our in situ hybridization analyses (Fig. 2J) confirmed that neurod is expressed in 
both outer and inner nuclear layers of the retina and in the telencephalon in wild-type 
embryos (Korzh et al., 1998; Mueller and Wullimann, 2002; Rauch et al., 2003; 
Ochocinska and Hitchcock, 2007). Expression of neurod in the retina was severely 
15
reduced in sox9b and sox9a;sox9b double mutant embryos (Figs. 2K, L). Because the 
expression of neurod was normal in the telencephalon, we conclude that sox9b is 
important for neurod expression in the retina but not in other parts of the nervous system. 
NeuroD is a basic helix–loop–helix (bHLH) transcription factor that can convert 
epidermal cells into neurons (Lee et al., 1995) and activate insulin gene transcription 
(Naya et al., 1995; Chae et al., 2004).
Our results confirmed expression of sox4a in the retina and developing brain of 
wild-type embryos (Fig. 2M) (Rauch et al., 2003; Thisse and Thisse, 2004), but showed 
that sox4a was not expressed in the retina of sox9b mutants and sox9a;sox9b double 
mutants; in contrast to the retina, expression of sox4a in the telencephalon and 
diencephalon was not affected in mutant animals (Figs. 2N, O). Sox4 is a Sox family 
transcription factor belonging to the SoxC group (Sox4, -11, and -12), while Sox8, -9, and 
-10 are in the SoxE group (Bowles et al., 2000). Sox4 is required for development of 
lymphocytes, heart and pancreas (Schilham et al., 1996; Ya et al., 1998; Wilson et al., 
2005). Zebrafish has two co-orthologs of Sox4 called sox4a and sox4b, that appear to 
have evolved in a duplication event at the base of the teleost radiation (Mavropoulos et 
al., 2005). In zebrafish, sox4a is widely expressed in the central nervous system including 
telencephalon, diencephalon, midbrain, rhombomeres, and retina (Rauch et al., 2003; 
Thisse and Thisse, 2004).
These expression results show that the full retinal expression of crx, rs1, vsx1, 
neurod and sox4a depends on sox9b but not sox9a function. This conclusion is consistent 
with the expression pattern of zebrafish sox9a and sox9b: at 2 dpf, sox9b is expressed in 
the ciliary marginal zone of the retina (Figs. 2S, T) but sox9a is not (Figs. 2P, Q) (Chiang 
16
et al., 2001; Cresko et al., 2003; Yan et al., 2005).
Co-orthologs of Calb2 are down-regulated in the sox9b mutant retina
The calcium binding proteins calbindin 2a (calb2a, previously known as calb2 or 
calretinin) and its closely related paralog calbindin 2b (calb2b, previously known as 
calbindin 2-like) were down-regulated in sox9b and sox9a;sox9b double mutant embryos. 
These two genes have not previously been well characterized in zebrafish. Our 
phylogenetic analysis showed that Calb2a and Calb2b proteins branch as sisters of the 
tetrapod Calb2 clade with high bootstrap support (Fig. 3A).
To evaluate the evolutionary origin of the zebrafish calb2 genes, we analyzed 
conserved synteny around calb2a and calb2b using our orthology and conserved synteny 
database based on reciprocal best blast hit (RBH) analysis (Catchen et al., 2008). Red 
dots in Fig. 3B represent the chromosomal positions of zebrafish orthologs of a 20 Mb 
segment of human chromosome 16 (Hsa16) surrounding the CALB2 gene. Orthologous 
chromosome segments appear as clusters of red dots. This analysis showed that a 
chromosomal segment of zebrafish chromosome 7 (Dre7) around calb2a and a segment 
of Dre18 around calb2b correspond to duplicated copies of the region surrounding 
CALB2 in Hsa16 (Fig. 3B). These results show that calb2a and calb2b are co-orthologs 
of tetrapod Calb2 and likely arose in the genome duplication that preceded the teleost 
radiation (Amores et al., 1998; Postlethwait et al., 1998; Jaillon et al., 2004).
Our expression analysis showed that in wild-type embryos, calb2a was expressed 
in the retina and in neurons of the cranial ganglia and in rhombomeres of the hindbrain 
(Fig. 3C), confirming directly submitted data in ZFIN (Rauch et al., 2003; Thisse and 
Thisse, 2004). The expression of calb2b appeared in the ganglion cell layer of the retina, 
17
in the olfactory placode, tegmentum, midbrain-hindbrain boundary (MHB) and ventral 
hindbrain (Fig. 3F), which confirms data in ZFIN (Rauch et al., 2003; Thisse and Thisse, 
2004). Both calb2 genes were expressed in developing retina, but their expression 
patterns differed: calb2a was expressed widely in the retina including the ganglion cell 
layer, but calb2b was expressed only in the ganglion cell layer of the retina (Figs. 3C, F). 
In other parts of the nervous system, calb2a was expressed in cranial ganglia and 
rhombomeres of the hindbrain, whereas calb2b was expressed in the olfactory placode, 
tegmentum, MHB and ventral hindbrain. In mouse, high-throughput gene expression 
analyses by RNA in situ hybridization showed that Calb2 is expressed in developing 
retina, olfactory bulb, tegmentum and hindbrain (Visel et al., 2004). Although description 
of the Calb2 expression pattern in mouse is not detailed enough to know its cellular 
distribution within the retina, this domain appears to be ancient and conserved in mouse 
Calb2 and partitioned into zebrafish co-orthologs calb2a and calb2b.
Expression analysis in sox9 mutant embryos verified our microarray results. In 
sox9b mutants and in sox9a;sox9b double mutants, calb2a expression did not appear in 
the eye, although other calb2a expression domains, including the cranial ganglia and 
rhombomeres of the hindbrain, were not affected (Figs. 3D, E). Likewise, in sox9b 
mutant embryos and in sox9a;sox9b double mutant embryos, calb2b expression was 
missing from the retina, although other expression domains were not affected (Figs. 3G, 
H).
These results show that sox9b regulates expression of both calb2a and calb2b in 
the developing retina. As mentioned above, calb2a and calb2b are teleost fish co-
orthologs of the tetrapod Calb2 gene. Although the expression patterns of calb2a and 
18
calb2b have diverged, expression of both genes in the retina depends on sox9b, 
suggesting that this relationship likely existed in the pre-duplication ancestral gene.
sox9b controls genes expressed in several retinal layers
As Fig. 2 and Fig. 3 show, a number of genes expressed in the retina were down-
regulated in sox9b mutants. To learn which specific retinal cell types are regulated by 
Sox9, we investigated detailed expression domains in tissue sections. In wild-type 
embryos, vsx1 was expressed in the outer part of the inner nuclear layer (Fig. 4A), but in 
sox9b mutants and in sox9a;sox9b double mutants, the expression domain was reduced 
dorsally (Figs. 4B, C). Expression of rs1 was observed in the outer part of the inner 
nuclear layer and the outer nuclear layer in the wild-type retina (Fig. 4D), while in mutant 
embryos, rs1 expression was reduced (Fig. 4E) or not detected (Fig. 4F). Expression of 
calb2a was observed in the inner nuclear layer (Fig. 4G) and calb2b was expressed in the 
ganglion cell layer (Fig. 4J). These expression domains were both severely reduced in 
mutant embryos with a small portion of ventral expression occasionally remaining (Figs. 
4H, I, K, L). These results show that sox9b plays important roles in several different 
retinal cell layers, including the ganglion cell layer, inner nuclear layer and outer nuclear 
layer.
Retinal differentiation in zebrafish sox9 mutants in later development
Expression profiling of wild-type and sox9a;sox9b double mutants identified 
seven genes expressed in the developing retina as downstream of Sox9 and hence 
potentially early Sox9 targets. As shown previously, sox9b is expressed in the developing 
eye by 24 hpf (Chiang et al., 2001; Cresko et al., 2003; Yan et al., 2005), and continues to 
be expressed in the retina at 2 dpf (Figs. 2S, T), so it is reasonable to conclude that Sox9 
19
functions in retinal differentiation by regulating down-stream target genes in the zebrafish 
embryo.
To learn how mutant phenotypes change after 2 dpf, we examined the expression 
of putative sox9 targets in 3 dpf embryos. At 3 dpf, calb2a was expressed in the ganglion 
cell layer and inner nuclear layer (Fig. 5Aa), and calb2b was expressed in the ganglion 
cell layer and in certain cells in the inner nuclear layer of wild-type embryos (Fig. 5Ba). 
The expression of neither calb2a nor calb2b was affected in sox9a mutants, but 
expression of both genes was reduced in the dorsal and ventral retina in sox9b mutants 
and in sox9a;sox9b double mutants (Figs. 5Ac, Ad, Bc, Bd). The outer and inner nuclear 
layers both expressed crx (Fig. 5Ca). In sox9b mutants and sox9a;sox9b double mutants, 
the same layers expressed crx, but expression was reduced in the most dorsal and most 
ventral expression domains (Figs. 5Cc, Cd). Expression of neurod was observed in the 
outer nuclear layer and in a few cells in the inner nuclear layer (Fig. 5Da), as previously 
reported (Ochocinska and Hitchcock, 2007). Expression of neurod was not affected in 
sox9a mutants (Fig. 5Db), but expression of neurod in the outer nuclear layer was less 
clear in sox9b and sox9a;sox9b double mutants, perhaps because retinal layers did not 
form properly in these mutants (Figs. 5Dc, Dd). The outer and inner nuclear layers both 
expresed rs1, except for the ciliary marginal zone (Fig. 5Ea). The expression was reduced 
only in the dorsal-most and ventral-most expression domains in sox9b mutants and in 
sox9a;sox9b double mutants (Figs. 5Ec, Ed). Expression of sox4a appeared in a 
proportion of inner nuclear layer cells and in some cells in the dorsal and ventral parts of 
the ganglion cell layer in wild types and sox9a mutants (Figs. 5Fa, Fb). In sox9b and 
double mutants, however, sox4a expression in the inner nuclear layer appeared to be a bit 
20
stronger than in the wild-type retina (Figs. 5Fc, Fd), suggesting an expansion of the 
sox4a+ cell type. Sections showed that vsx1 was expressed in the inner nuclear layer (Fig. 
5Ga), but expression in the dorsal and ventral parts of the retina was reduced in sox9b 
and in sox9a;sox9b double mutants (Figs. 5Gc, Gd).
Expression of col2a1a, a known target of Sox9 in cartilaginous tissue, appeared in 
the iris (Fig. 5Ha). In contrast to cartilaginous tissue, col2a1a expression in the iris was 
retained in sox9a mutants, sox9b mutants, and in sox9a;sox9b double mutant embryos 
(Figs. 5Hb, Hc, Hd). These results showed that the regulatory relationship of Sox9 → 
Col2a1 that is well established for cartilage does not hold for the iris.
Taken together, expression analysis on tissue sections suggest that sox9b, but not 
sox9a, is required for retinal development. The requirement for Sox9 is most evident 
around 2 dpf and becomes milder at 3 dpf in sox9b mutants and in sox9a;sox9b double 
mutants, suggesting that other gene(s) can partially compensate for the lack of sox9b 
function on the expression of the candidate genes at 3 dpf. Because candidate genes 
expressed in or near the ciliary marginal zone were sensitive to loss of sox9b function at 3 
dpf, sox9b may function in cell proliferation and/or retinal layer organization.
Independent of the microarray analysis, we had noticed that sox9b mutants have 
smaller eyes than wild types. We investigated eye development in wild-types and in 
sox9a, sox9b and double mutants to elucidate the function of Sox9 in retinal 
differentiation in larval fish. By 5 dpf, most cell types in the retina have already 
differentiated (Fadool and Dowling, 2008). Antibody staining for specific cell layers 
revealed that cells expressing the Müller glial cell marker carbonic anhydrase (Figs. 6A, 
E, I, M) (Jensen et al., 2001) and the photoreceptor cell marker zpr-3 (Figs. 6B, F, J, N) 
21
(Jensen et al., 2001) are reduced in number and disorganized in the retinas of sox9b 
mutants and sox9a;sox9b double mutants (Figs. 6I, J, M, N). In contrast, sox9a mutant 
retinas had no obvious defects, even though sox9a is expressed in the inner nuclear layer 
at 68 hpf (Fig. 2R). These results on 5dpf animals show that sox9b is required for 
differentiation of the Müller glial cell layer and photoreceptor cell layer in zebrafish. 
Glial cells are required for proper migration of neurons, so the disorganized retina might 
be secondary due to a primary defect in Müller glial cells. Sox9 is involved in glial cell 
development in mouse spinal cord (Stolt et al., 2003), and here we showed that sox9b is 
required for retinal Müller glial cells. Although different glial types are involved, the role 
of Sox9 in glial cell development may be conserved between Müller glia and glia of the 
central nervous system.
Expression of klf2b and EST AI722369 in the pectoral fin bud disappeared in sox9 
mutants
The microarray experiments showed that klf2b was down regulated 2.344 fold in 
double mutants. Expression of klf2b was observed in the central mesenchyme of the 
pectoral fin bud and in the adjacent cleithrum of the shoulder girdle (Figs. 7A, C), as 
previously described (Oates et al., 2001; Thisse et al., 2001). In double sox9 mutants, 
expression of klf2b in the fin bud disappeared but expression in the cleithrum was not 
affected (Figs. 7B, D). Consistent with these defects, sox9a and sox9b were expressed in 
the pectoral fin bud, but the cleithrum expressed neither sox9a nor sox9b (Yan et al., 
2005). This result is consistent with the appendage defect observed in the sox9 mutant 
pectoral fin: only the cleithrum was retained while other fin structures were greatly 
reduced in sox9a;sox9b double mutants (Yan et al., 2005).
22
An EST of unknown function (AI722369) was 5.673 times down regulated in 
double mutant embryos. The AI722369 nucleotide sequence has more than 98% identity 
over about 500 nucleotides to sequences assigned to at least five different chromosomes 
in the Ensembl genome database, suggesting that related sequences are dispersed in the 
zebrafish genome. AI722369 is similar to the uncharacterized human hypothetical protein 
LOC100131068. Despite appearing to be a repeated element, AI722369 was expressed in 
a strongly specific manner in time and space: in the pectoral fin bud in two parallel 
domains in the position of presumptive pectoral fin muscle (Fig. 7E). Furthermore, 
AI722369 expression disappeared from sox9a;sox9b double mutants (Fig. 7F), suggesting 
that it is downstream of Sox9 activity. In the wild-type pectoral apparatus, sox9a is 
expressed strongly in the scapulocoracoid of the pectoral girldle but was weaker in the 
blade of the fin bud and did not appear in the cleithrum of the pectoral girdle. 
Reciprocally, sox9b is expressed strongly in the chondrogenic core of the fin bud but not 
in the scapulocoracoid or the cleithrum of the pectoral girdle (Yan et al., 2005) (Figs. 7G, 
H). In sox9a;sox9b double mutants, both the scapulocoracoid and the extension of the fin 
bud are missing or severely reduced (Yan et al., 2005). We conclude that the reduced 
expression of klf2b and AI722369 in mutant embryos arises from fin bud hypogenesis 
downstream of Sox9 and so these genes may not be direct targets of Sox9.
Discussion
Because Sox9 encodes a transcription factor that functions by regulating the 
transcription of down-stream target genes, the identification of targets is an appropriate 
way to understand the various roles of Sox9 in development. Although several 
23
downstream targets of Sox9 are known, many others remain to be identified. A genome-
wide screening for the identification of targets provides a fruitful method for recognizing 
previously unknown targets, and hence for identifying unrecognized functions of Sox9. 
To implement this strategy, we compared the gene expression profile of wild-type and 
sox9-deficient mutant embryos using a microarray representing 16,228 gene clusters. 
Because Sox9 generally acts as a transcriptional activator (Südbeck et al., 1996), we 
focused on genes whose expression was down-regulated in mutant embryos. We 
identified several types of candidate target genes, including genes expressed in cartilage, 
retina, and pectoral fin bud. A comparison of sox9 expression patterns in the retina to the 
phenotypes of sox9 mutants and the expression patterns of candidate targets in wild-type 
and sox9 mutant embryos uncovered a previously unrecognized requirement for Sox9 in 
development of the retina.
Down-stream targets of Sox9
Microarray is a powerful tool to identify genes whose expression is different 
among experimental groups (Schena et al., 1995; Hoheisel, 2006). Expression differences 
between groups could arise due either to direct, primary effects, or to subsequent, 
secondary effects. A direct target of Sox9 should be co-expressed with sox9 in space. A 
secondary target would be turned on in response to a primary target in a regulatory 
cascade, and so might be first expressed well after sox9 or be expressed in cells that do 
not express sox9. To confirm microarray results, we compared the expression patterns of 
candidate genes and sox9 genes among wild-type, sox9a, sox9b and sox9a;sox9b double 
mutant embryos by whole mount in situ hybridization. Our microarray experiments 
identified potential Sox9 target genes in two major developmental processes, 
24
skeletogenesis and retinogenesis, as well as several genes likely to represent secondary 
effects.
Targets involved in skeletogenesis
Microarray analysis showed that the collagen genes col2a1a and col11a2 were 
down-regulated in mutants. Because the mouse and human orthologs of col2a1a and 
col11a2 are well characterized direct targets of Sox9 (Bell et al., 1997; Lefebvre et al., 
1997; Ng et al., 1997; Bridgewater et al., 1998; Bi et al., 1999; Sekiya et al., 2000), these 
genes validate the method. Indeed, it has already been shown in zebrafish embryos that 
the expression of col2a1a is down-regulated in sox9a, sox9b and double mutant embryos 
(Yan et al., 2005).
This study presents the first description of col11a2 expression in zebrafish. At 
2dpf, col11a2 was expressed in neurocranial cartilages, pharyngeal cartilages, otic vesicle 
and pectoral fin bud. This expression pattern is consistent with mouse Col11a2, which is 
expressed in cartilaginous tissues, such as chondrocranium, hyoid, Meckel's, and limb 
cartilages (Sugimoto et al., 1998). Our results showed that the expression of zebrafish 
col11a2 was affected differently in sox9a, sox9b, and sox9a;sox9b double mutant 
embryos. Thus, we conclude that sox9a and sox9b both regulate col11a2 expression in 
tissues in which each sox9 gene is expressed. After the sox9 gene duplication event, 
regulatory elements controlling the expression of sox9 co-orthologs apparently 
partitioned in a tissue-specific fashion, but the Sox9 proteins encoded by sox9a and 
sox9b retained the ability to regulate col11a2, presumably by binding to regulatory sites 
for the col11a2 gene. Other known or potential Sox9 targets, like aggrecan, hyaluronan 
and proteoglycan link protein 1b (hapln1b), runx2a and runx2b were not spotted on the 
25
array.
Targets involved in retinogenesis
Microarray analysis detected down-regulation of genes expressed in the retina, 
including calb2a, calb2b, crx, neurod, rs1, sox4a and vsx1; whole mount in situ 
hybridization experiments confirmed reduced expression of these genes in mutant retinas. 
Our analysis revealed the detailed expression pattern of calb2a, calb2b, rs1 and sox4a in 
developing retina, which had not previously been characterized in detail, although some 
data had been deposited in the high-throughput expression database in ZFIN (Rauch et 
al., 2003; Thisse and Thisse, 2004).
In situ hybridization results for calb2b were similar to the previously reported 
antibody staining for the Calretinin, which labels ganglion, amacrine and bipolar cells 
(Bernardos et al., 2005). Although calretinin is an alternative name for calb2a in 
zebrafish, the popularly used anti-Calretinin polyclonal antibody (AB5054) seems likely 
to recognize also Calb2b in zebrafish. Mouse Calretinin (Calb2) is a marker for ganglion 
and amacrine cells in retinas, and the same cell types express calb2b in zebrafish. 
Expression results suggest evolutionary models: the ancestral expression domain of 
Calb2 was in ganglion and amacrine cells of the retina, and that this pattern was 
preserved by Calb2 in mouse and by calb2b in zebrafish, while zebrafish calb2a might 
have acquired an expanded expression domain in the inner nuclear layer of the retina. 
Another possible model would be that the ancestral Calb2 was expressed in a pattern that 
equals the sum of the zebrafish calb2a and calb2b domains, that these domains 
partitioned between two zebrafish calb2 genes, and that the mouse lineage preserved only 
the ganglion and amacrine expression domains.
26
Human RS1 is associated with X-linked retinoschisis, a common form of macular 
degeneration (Sauer et al., 1997). Rs1 encodes a protein containing a single discoidin 
domain shared with receptors for collagen (Vogel, 1999; Vogel et al., 2006). Although it 
has not been shown that Rs1 protein interacts with collagen, Rs1 is interesting as a 
candidate target for Sox9 because the collagens Col2a, Col9a1 and Col11a2 are known as 
Sox9 targets (Bell et al., 1997; Lefebvre et al., 1997; Ng et al., 1997; Bridgewater et al., 
1998; Bi et al., 1999; Sekiya et al., 2000; Zhang et al., 2003; Genzer and Bridgewater, 
2007). Mouse Rs1 is down-regulated in the retina of Crx knockout mice (Livesey et al., 
2000; Blackshaw et al., 2001). The down-regulation of zebrafish rs1 in sox9 mutant 
retinas could thus either be a secondary effect due to the primary down-regulation of crx, 
or it might directly result from regulation by Sox9.
In zebrafish, crx is expressed in proliferating retinal progenitors and its expression 
was diminished in sox9b mutants. Our preliminary comparative genomic analyses 
identified a conserved non-coding element (CNE) near the crx gene that has Sox9 
consensus binding sites (data not shown). Although we haven't confirmed its enhancer 
activity on crx gene expression, it is possible that Sox9 binds to these cis-elements and 
directly activates crx gene expression.
Our result showed that expression of vsx1 became down-regulated in sox9b 
mutants, and it has been published that expression of vsx1 also down-regulates in 
zebrafish crx morpholino injected embryos (Shen and Raymond, 2004). Thus, the 
reduction in vsx1 expression in sox9b mutant embryos could result from either a direct 
effect of sox9b loss or a secondary effect caused by the reduced crx activity that is 
associated with loss of sox9b. Both crx and vsx1 have predicted Sox9 binding sites (data 
27
not shown), as expected by the hypothesis that they are directly regulated by Sox9.
A novel function for Sox9 in retinal development
The retina of zebrafish, human, and other vertebrates consists of seven major cell 
types: six types of neurons and a glial cell type, all of which differentiate from the neural 
epithelium (Malicki, 2000; Stenkamp, 2007). In zebrafish, retinal neurogenesis proceeds 
sequentially; ganglion cells are generated from 24 to 36 hpf, cells of the inner nuclear 
layer arise from 36 to 48 hpf, and outer nuclear layer cells are born from 48 to 60 hpf 
(Stenkamp, 2007). Differentiation initiates in the ventral retina and then spreads dorsally 
around over time (Fadool and Dowling, 2008; Stenkamp, 2007). One of the zebrafish 
sox9 genes, sox9b, begins to be expressed in the eye by 24 hpf (Chiang et al., 2001), and 
at 2 dpf, localizes to the ciliary marginal zone of the retina, where retinal progenitors 
proliferate. In the retina of sox9b mutants, expression of candidate genes was severely 
reduced at 2 dpf in the outer nuclear layer, inner nuclear layer and ganglion nuclear layer, 
suggested that sox9b is required for differentiation of most of the retina. Although the 
expression of candidate genes became milder at 3 dpf, gene expression close to the ciliary 
marginal zone was still reduced, and Müller glial cells and photoreceptor cells were 
diminished in number and were disorganized in location at 5 dpf. Considering the 
expression pattern of sox9b, the expression of candidate genes in sox9b mutants, and the 
phenotype observed in the retina of sox9b mutants, we conclude that sox9b is essential 
for retinal development. The requirement of sox9b for retinal development is most 
prominent at 2 dpf, and it becomes less essential at 3 dpf but still required for 
organization and proliferation of Müller glial cells and photoreceptor cells; thus, these 
results uncover a novel function for Sox9.
28
Evolution of functions of zebrafish Sox9 co-orthologs
After gene duplication, preserved duplicates often experience reciprocal loss of 
gene subfunctions (Force et al., 1999; Postlethwait et al., 2004), for example, the 
zebrafish Sox9 co-orthologs, sox9a and sox9b, appear to have partitioned the ancestral 
craniofacial skeleton expression domain between the two genes. In other domains, both 
duplicates preserve ancestral functions, for example, both sox9a and sox9b are expressed 
in the otic vesicle. Because Col11a2 is a direct target of Sox9 (Bridgewater et al., 1998; 
Liu et al., 2000), and also because col11a2 was isolated as a candidate sox9 target in this 
study, the expression of col11a2 would be a good marker for visualizing functions of 
sox9, and for tracing the partitioning of subfunctions between sox9a and sox9b.
In the neurocranium, the ethmoid plate expresses sox9a but not sox9b, and the 
ethmoid plate is deleted in sox9a mutants but not in sox9b mutants; thus, our results show 
that col11a2 expression in the ethmoid plate depends on sox9a but not sox9b. On the 
other hand, wild-type pharyngeal arches express both sox9a and sox9b, with sox9a in the 
mesenchyme and sox9b in the epithelium (Yan et al., 2005), but neither sox9a nor sox9b 
single mutant embryos express col11a2 in pharyngeal arches. This result shows that the 
activity of both Sox9 co-orthologs is required in the pharyngeal arches to regulate 
col11a2. Both sox9a and sox9b are expressed in the otic vesicles, but in different, though 
overlapping domains (Chiang et al., 2001; Yan et al., 2005). Likewise, sox9a and sox9b 
single mutant embryos both express col11a2 in the otic vesicle, but in somewhat different 
patterns, and the function of both genes must be missing to ensure that the entire ear is 
gone. This result shows that, like the pharyngeal arches, the two Sox9 co-orthologs likely 
induce col11a2 expression in the domains of the otic vesicle in which they are expressed. 
29
We conclude that the regulatory relationship between sox9 genes and col11a2 differs 
among tissues: sox9a is sufficient for col11a2 expression in the ethmoid plate, both sox9a 
and sox9b are essential in pharyngeal arches, and each of the duplicates controls col11a2 
expression in different parts of the otic vesicles.
Although Sox9 function in the eye is not yet well characterized in mouse or 
human, in mouse, Sox9 protein is expressed in eye (Ihanamäki et al., 2002), and human 
patients heterozygous for dominant SOX9 mutant alleles sometime experience reduced 
eye function (Mansour et al., 2002). These results suggest that the requirement of Sox9 in 
retinal development is conserved among vertebrates.
We propose that a requirement for Sox9 function in retinal development already 
existed in the last common ancestor of tetrapods and teleost fish, and that after the 
genome duplication in the teleost fish lineage, retinal function mostly partitioned to 
sox9b, which now acts to control cell numbers and organization during retinal 
development by activating, directly or indirectly, genes identified by our expression 
profiling study. Future experiments should be directed towards the identification of 
specific DNA elements responsible for this newly discovered role of Sox9.
30
CHAPTER III
FUNCTIONAL GENOMICS IDENTIFIES NEURAL STEM CELL SUB-TYPE 
EXPRESSION PROFILES AND GENES REGULATING NEUROBLAST 
HOMEOSTASIS
This work was submitted to the journal Developmenal Biology in September 2011 
and is currently 'under review'. Myself, Kristin Robinson, and Travis Carney performed 
the microarray-based expression analysis. Travis Carney, Jessica Osterhout, and myself 
performed the RNAi analysis. Omer Bayraktar performed the immunohistochemistry 
analysis. Chris Doe was the principle investigators for this work. 
Introduction
Drosophila neuroblasts are a powerful model system for understanding the 
molecular control of stem cell self-renewal versus differentiation. The majority of 
neuroblasts (type I neuroblasts) repeatedly divide asymmetrically with respect to size and 
fate to self-renew and produce a smaller daughter cell called a ganglion mother cell 
(GMC) that divides only once to produce two post-mitotic neurons or glia (reviewed in: 
Chia et al., 2008; Doe, 2008; Knoblich, 2008). Neuroblast/GMC fate differences are due 
in part to the asymmetric partitioning of proteins into the GMC during neuroblast cell 
division. These factors include the transcription factor Prospero (Pros), the Notch 
inhibitor Numb, and the translational repressor Brain tumor (Brat) (Betschinger et al., 
2006; Broadus et al., 1998; Hirata et al., 1995; Knoblich et al., 1995; Lee et al., 2006c; 
Spana and Doe, 1995). Proper segregation of Pros, Numb, and Brat into the GMC require 
the scaffolding protein Miranda (Mira) and the WD40-domain protein Lethal giant larvae 
(Lgl) (Betschinger et al., 2006; Lee et al., 2006c; Ohshiro et al., 2000; Peng et al., 2000). 
31
In the GMC, Pros enters the nucleus and promotes cell cycle exit and differentiation by 
directly activating differentiation genes and repressing self-renewal and cell cycle 
regulatory genes (Choksi et al., 2006; Li and Vaessin, 2000). This process allows for a 
single neuroblast to generate a lineage of many differentiated neurons and glia and for a 
relatively small number of neuroblasts to generate the thousands of cells found in the 
central nervous system of the adult fly.
Recently, "type II" neuroblasts were identified in the larval brain which divide 
asymmetrically to produce small transit-amplifying progenitors (intermediate neural 
progenitors, INPs). INPs themselves undergo molecularly asymmetric cell divisions to 
generate 4-6 GMCs, each of which typically generates two post-mitotic neurons or glia 
(Bello et al., 2008; Boone and Doe, 2008; Bowman et al., 2008; Izergina et al., 2009; 
Viktorin et al., 2011). Six type II neuroblasts inhabit the dorso-medial region of the lobe 
and are designated DM1-6, and two occupy more lateral positions (Bayraktar et al., 2010; 
Izergina et al., 2009). Type II neuroblasts behave in a manner similar to mammalian 
neural stem cells in that they generate transit-amplifying INPs. Transit-amplifying 
progenitors are important in the development of the nervous system in mammals as well 
as in flies because they permit the rapid amplification of neuronal progeny (Bello et al., 
2008; Boone and Doe, 2008; Bowman et al., 2008; Merkle and Alvarez-Buylla, 2006; 
Morrison and Kimble, 2006). Thus, while there are ~95 type I neuroblasts per larval brain 
lobe and only eight type II neuroblasts, these few lineages produce a considerable 
fraction – approximately a quarter – of the neurons of the adult brain (Izergina et al., 
2009). Type II neuroblasts and INPs have thus become a model for the study of transit-
amplifying neural progenitors, and determining how these cells are specified and 
32
maintained may shed light on the function of mammalian neural stem cells.
Type II neuroblasts are known to differ from type I neuroblasts in several ways. 
First, they generate INPs and thus make much larger lineages than type I neuroblasts 
(Bello et al., 2008; Boone and Doe, 2008; Bowman et al., 2008). Second, they are the 
sole contributors of the intrinsic neurons of the adult central complex (Bayraktar et al., 
2010; Izergina et al., 2009). Third, they are more susceptible to tumor formation (Boone 
and Doe, 2008; Bowman et al., 2008; Ouyang et al., 2011; Weng et al., 2010). Numerous 
genotypes have been identified that cause the production of ectopic larval brain 
neuroblasts, and several of these specifically affect type II neuroblasts. For example, 
mutations in brat lead to "overgrowth" of just the type II neuroblasts, and mutations in lgl 
affect type II much more strongly than type I neuroblasts. On the other hand, loss of 
Aurora-A (Aur) or neuroblast misexpression of membrane-tethered atypical protein 
kinase C (aPKCCAAX) leads to ectopic type I and type II neuroblasts (Bowman et al., 
2008). It is likely that the lack of Pros in the new-born INP renders this daughter cell 
sensitive to the loss of a second growth inhibitor/differentiation factor, making it easier 
for this cell to revert to a type II neuroblast identity.
In spite of the marked differences between type I and type II neuroblasts in 
proliferative potential and susceptibility to tumor formation, only two molecular 
differences are known: type II neuroblasts lack the transcription factors Pros and Asense 
(Ase), while both are present in all type I neuroblasts (Bello et al., 2008; Boone and Doe, 
2008; Bowman et al., 2008). It is currently unknown how many other genes are regulated 
differentially between type I and type II neuroblasts, and which of them regulates each 
distinct aspect of type I and type II function. In order to discover such transcriptional 
33
differences, relatively pure populations of each neuroblast sub-type must be isolated from 
which to extract RNA. Complicating these efforts, the Drosophila central nervous system 
contains only a small number of neuroblasts which are dispersed throughout a complex 
population of thousands of neurons and glia, making it difficult to physically separate 
neuroblast sub-types from each other and from other cell types. Thus, comparing the 
transcriptional outputs of neuroblast sub-types is technically challenging due to the 
difficulty of isolating cell type-specific RNA. 
In order to enrich for each type of neuroblast, here we make use of published and 
unpublished mutants in which type I and type II neuroblasts exhibit differential 
overproliferation phenotypes. We perform microarray-based whole-genome 
transcriptional profiling to compare each of these different mutant brains to wild-type; 
thus we are able to probe the transcriptional differences not only between each mutant 
and wild-type, but also between type I and type II neuroblasts. We identify only a small 
number of genes exhibiting transcriptional differences between type I and type II 
neuroblasts, providing a highly specific group of genes to screen for a function in 
establishing each type of neuroblast. We identify a large group of genes which are likely 
expressed in neuroblasts but not neurons, and we verify the neuroblast function of a 
subset of these genes using an RNAi-based targeted loss of function screen. Using this 
approach we identify 84 genes required to maintain neuroblast numbers in larval brains, 
all of which have conserved mammalian orthologs.
34
Methods
Fly stocks
All fly stocks used in this study have been previously described except for lgl-  
lgdd7, in which the lgl locus has been spontaneously lost (Jason Boone, unpublished data) 
from the lgdd7 chromosome (Jaekel and Klein, 2006). Other fly stocks used were: lgl334  
(Rolls et al., 2003); aurA8839 (Lee et al., 2006a); lgl334;pins62 and UAS-aPKCCAAX (Lee et 
al., 2006b); brat11 (Lee et al., 2006c); wor-gal4 (Albertson et al., 2004); R9D11-Gal4 and 
R19H09-Gal4 (Bayraktar et al., 2010); UAS-mCD8::GFP and UAS-Dicer2 (Bloomington 
Stock Center).
Microarray analysis
Mutant larvae were dissected at 144 hours after larval hatching (ALH); wild-type 
heterozygous larvae that were at the same developmental stage were dissected at 96 hours 
ALH as wild-type controls. The only exceptions were the aPKCCAAX experiments, in 
which the experimental larvae were raised at 30°C and exhibited enlarged brains packed 
with ectopic neuroblasts, and genetically identical control larvae were raised at room 
temperature, where the ectopic neuroblast phenotype is much weaker. Total RNA was 
extracted from larval brain lobes using TRIzol extraction methods according to the 
manufacturer’s instructions. First strand synthesis and amplification of Cy3- and Cy5-
labeled RNA were accomplished using the Agilent Low Input Quick Amp Labeling Kit. 
Four experimental replicates were performed for each mutant genotype: two standard 
replicates (Cy5-labeled mutant RNA and Cy3-labeled wild-type RNA) and two dye-
swapped replicates. Exceptions were the aPKCCAAX experiments, in which one standard 
and two swapped replicates were used for clustering, and the lgl experiments, in which 
35
three standard and one swapped replicates were used. Hybridization was performed as 
previously described (Miller et al., 2009), except for each replicate, 825 ng of both 
mutant and wild-type RNA were mixed and hybridized to Agilent microarrays. The slides 
were scanned using an Axon GenePix 4000B, and GenePix software was used for feature 
extraction.
Cluster analysis
We selected genes for cluster analysis if, in at least one of the mutant genotypes, 
their average transcript levels over all biological replicate experiments deviated from the 
wild-type RNA sample by greater than two-fold. This criterion resulted in the selection of 
2781 genes. To investigate the reproducibility of replicate experiments of the same 
mutant genotype and the overall transcriptional similarities between experiments, we first 
performed cluster analysis without averaging individual replicate experiments. This 
analysis allowed us to determine whether, for each experiment, the clustering analysis 
grouped replicates of the same genotype together. To identify groups of genes with 
similar transcript patterns over the different mutant genotypes, we averaged replicates 
and performed cluster analysis on these averages in order to avoid artificial gene 
clustering relationships due to technical noise between experimental replicates. After 
performing cluster analysis, we found that a large group of genes with increased 
expression in mutant brains clustered with high correlation. Group A was defined by a 
tree branch that represented a large decrease in correlation, from >0.6 to ~0.38. We 
similarly found that a long branch in the genes with reduced expression in mutant brains 
caused a decrease in correlation that passed the .6 cutoff (from >0.7 to ~0.54); thus we 
considered all the genes clustering below this branch to comprise group C. The remainder 
36
of the genes exhibited variable expression patterns and did not cluster with high 
correlation, and was defined as group B.
RNAi screen
Knockdowns were performed on group A genes with annotated human orthologs 
and transgenic RNAi stocks available from the Vienna Drosophila RNAi Center (VDRC). 
We used these lines to knock down genes cell type-specifically by crossing RNAi line 
males to wor-Gal4UAS-Dicer2 virgins. The progeny of these crosses were raised at 30°C 
and scored for lethality. Each knock-down was performed at least twice to judge the 
consistency of the phenotype. For those RNAi constructs that caused lethality, we 
performed crosses again and dissected brains from wandering third instar larvae. We took 
confocal stacks of these brains (see below) and determined the number of central brain 
neuroblasts per brain lobe using antibodies against the neuroblast-specific proteins 
Deadpan (Dpn) and Mira. Optic lobe neuroblasts were excluded from these counts based 
on their small size, tight clustering, and stereotyped lateral position in the brain lobe. 
Fixation, antibody staining, and confocal microscopy
The following antibodies were used for immunohistochemical staining of larval 
brains: guinea pig anti-Dpn, 1:2000 (J. Skeath); rat anti-Dpn, 1:1-1:50 (Doe lab); rat anti-
Elav, 1:50 (Developmental Studies Hybridoma Bank 7E8A10); rabbit anti-Ase, 1:2000 
(Brand et al., 1993); rabbit anti-Optix, 1:500 (Kenyon et al., 2005); rabbit anti-Rx, 1:2000 
(Davis et al., 2003); chicken anti-GFP, 1:1000 (Aves Laboratories); mouse anti-Pros, 
1:1000 (MR1A). Brains were dissected in Schneider’s medium, fixed in PBST 
(phosphate-buffered saline + 0.1% Triton-X100) with 4% formaldehyde for 20 minutes, 
rinsed for 30 minutes in PBST, blocked for 30 minutes using PBSBT (PBST + 1% bovine 
37
serum albumin) or PBST + 5% normal goat serum. Brains were incubated in primary 
antibody overnight at 4°C with rocking, and then rinsed in PBST + block for 1 hour. 
Brains were incubated with secondary antibodies (1:500; Molecular Probes or Jackson 
Immunoresearch) for 2 hours at room temperature with rocking, and then rinsed for 1 
hour with PBST and stored in Vectashield until microscopy could be performed. 
Microscopy images were taken using a Bio-Rad Radiance or Zeiss700 confocal 
microscope.
Results
Transcriptional profiling of larval brains containing ectopic type I or type II 
neuroblasts
We analyzed six different genotypes that generate ectopic neuroblasts in the third 
instar larval brain (Table 2). The brat and lgl single mutants produce primarily ectopic 
type II neuroblasts, whereas aur mutation or misexpression of membrane-tethered aPKC 
is reported to generate ectopic type I and type II neuroblasts (Bowman et al., 2008). We 
also analyzed lgl lgd and lgl pins double mutants, both of which produce large numbers 
of ectopic neuroblasts of unknown type (Lee et al., 2006a; Lee et al., 2006b; Wang et al., 
2006; Jason Boone, unpublished data). We stained these brains for the pan-neuroblast 
marker Dpn and neuronal marker Elav to determine the total number of ectopic 
neuroblasts and remaining number of neurons, showing that there is a graded increase in 
the number of neuroblasts per brain from lgl (the fewest ectopic neuroblasts) to lgl pins 
(almost entirely neuroblasts; Figs. 8A and 8B). As type I and type II neuroblasts can be 
distinguished by the presence of Ase only in type I neuroblasts (Figs. 8C and 8D), we 
38
also stained the brains for Dpn and Ase to determine the proportion of ectopic type I/type 
II neuroblasts. We confirmed that the brat single mutant generates primarily ectopic type 
II Ase- neuroblasts while aur and aPKCCAAX brains contain more type I Ase+ neuroblasts, 
although there is also an increase in type II neuroblasts (Fig. 8E).  Moreover, we found 
that the lgl lgd double mutant is strongly enriched for type II neuroblasts, and lgl pins 
brains contain both neuroblast types with an enrichment of type I neuroblasts. We noted 
that in lgl pins brains, distinct regions of type I and type II neuroblast overproliferation 
are discernible based on the lack of Ase and Pros in ectopic cells derived from type II 
neuroblasts (Fig. 8E and inset). We conclude that the six genotypes used here exhibit a 
range of type I/type II differential overproliferation phenotypes, with brat, lgl, and lgl lgd 
brains representing enriched pools of type II neuroblasts, and with aur, aPKCCAAX, and lgl  
pins being more enriched for type I neuroblasts.
Next, we used transcriptional profiling of the larval brain lobes from each of these 
six genotypes to identify (a) genes differentially regulated in type I vs. type II 
neuroblasts, that may function in establishing the striking differences between these two 
types of progenitors, and (b) genes expressed in all neuroblasts, that may function to 
regulate self-renewal or asymmetric cell division. For each experiment, we isolated RNA 
from mutant and wild-type brain lobes, amplified and fluorescently labeled each RNA 
sample, and hybridized them directly against each other to microarrays representing the 
entire complement of protein-coding Drosophila genes with at least two-fold redundancy 
(Fig. 8F). We used cluster analysis to group genes according to transcriptional pattern 
similarities in the different experiments. Genes exhibiting no change between wild-type 
and mutant were not included in the cluster analysis (see Methods). Biological replicates 
39
and dye-swap experiments cluster much more closely to one another than to replicates for 
any other mutant (data not shown); this demonstrates that our data are highly 
reproducible and that each mutant exhibits a distinct transcriptional profile.
We sorted genes into three groups based on their transcript pattern in the six 
genotypes (Fig. 9A). Group A contains 1045 genes with elevated expression in the mutant 
genotypes (Fig. 9B); these genes are likely to be expressed in neuroblasts and are good 
candidates for regulating neuroblast function (see below). Group B contains 467 genes 
that have variable expression between mutants. Group C contains 1269 genes with 
decreased transcript levels in each mutant (Figs. 9A and 9B); thus the genes in this large 
group are good candidates for genes expressed in neurons or glia but not in neuroblasts. 
Interestingly, we did not see clustering of the genotypes that generate ectopic type II 
neuroblasts (lgl, lgl lgd, and brat) – note that the dendrogram at the top of Fig. 9A shows 
that each of these genotypes has a more closely related genotype that generates ectopic 
type I neuroblasts – suggesting type I and type II neuroblasts are much more similar 
transcriptionally than different.
Identification of genes transcribed preferentially in type I or type II neuroblasts
To identify genes expressed differentially between type I and type II neuroblasts, 
we looked for genes clustered with pros and ase, the only two genes known to be 
differentially expressed in type II neuroblasts. We found that pros and ase reside together 
in a small sub-cluster of only 11 genes within group B (Fig. 10A). This sub-cluster as a 
whole exhibits reduced expression in brat, lgl, and lgl lgd mutants and enrichment in aur,  
aPKCCAAX, and lgl pins; remarkably, no other sub-cluster exhibits such a pattern. This 
suggests that the other nine genes in the cluster may also be specifically expressed in type 
40
I neuroblasts, like pros and ase, and that these are potentially the only genes that exhibit 
this unique pattern.
To test whether other genes in the small pros/ase cluster are also expressed in type 
I neuroblasts but not type II neuroblasts, we obtained an antibody to a candidate from this 
cluster, Retinal homeobox (Rx), a homeodomain-containing transcription factor (Davis et 
al., 2003; Eggert et al., 1998). We found that Rx is completely absent from type II 
neuroblasts, similar to Pros and Ase; Rx is detected in several type I neuroblasts as well 
as in a subset of differentiated type II progeny (Figs. 10B and 10C). Consistent with this 
expression pattern, we found that brat mutants, which overproduce type II neuroblasts, 
show a loss of Rx staining (Fig. 10D). In contrast, lgl pins mutants, which have ectopic 
type I neuroblasts, show territories of strong Rx expression which is confined to Pros+ 
(likely type I-originating) cells (Fig. 10E). The fact that only a small patch of lgl pins 
mutant brain tissue is Rx+ is probably because Rx is normally expressed in a subset of 
type I neuroblasts. We conclude that Rx, like Pros and Ase, is expressed in type I but not 
type II neuroblasts. Thus, most or all of the 11 genes in the pros/ase sub-cluster may be 
expressed in type I but not type II neuroblasts.
We next wanted to find genes expressed in type II neuroblasts but not type I 
neuroblasts, as there are currently no known markers specifically expressed in type II 
neuroblasts. We reasoned that transcripts expressed in type II neuroblasts should be 
enriched in genotypes that overproduce type II neuroblasts: brat, lgl and lgl lgd. We 
found one small cluster enriched in two of the three mutants (brat and lgl lgd) (Fig. 11A). 
This cluster contains just six genes, all encoding transcription factors. To verify the 
expression pattern of this gene cluster, we examined the expression of one gene product, 
41
Optix. Optix is a conserved homeodomain-containing transcription factor required for 
eye development (Kenyon et al., 2005; Seimiya and Gehring, 2000; Toy et al., 1998). 
Consistent with our microarray data, we found that most of the Optix expression in the 
brain is indeed restricted to type II lineages; four of the six dorso-medial type II 
neuroblasts (DM1, 2, 3, and 6) express Optix, as do most of the INPs, GMCs, and 
neurons in these lineages (Figs. 11B and 11C). The other two dorso-medial type II 
lineages (DM4 and 5) exhibit some expression of Optix in a subset of neuronal progeny, 
but it is absent from the neuroblasts and INPs in these lineages (Figs. 11B and 11C and 
not shown). In addition, a single dorsal type I neuroblast expresses Optix (Fig. 11C). 
Inspection of mutant brains further confirmed the type II-biased expression of Optix, in 
that brat mutant brains exhibit a marked increase in Optix+ neuroblasts (Fig. 11D), and in 
lgl pins, the increase in Optix is almost exclusively in a Pros- (type II-originating) region 
of the brain (Fig. 11E). Our results indicate that our clustering relationships can be used 
to predict type I/type II expression bias with good accuracy. We conclude that Optix is 
primarily expressed in type II but not type I neuroblasts, and that Optix and the other five 
genes in this cluster are excellent candidates for regulators of type II neuroblast identity.
Identification of genes predicted to be expressed in neuroblasts but not neurons
To determine whether the ectopic neuroblasts in the mutant brains express wild-
type neuroblast genes, we tested whether genes known to be expressed primarily or 
exclusively in neuroblasts are found in group A; indeed all such positive control genes 
(with the exception of the type II-negative pros and ase genes) are represented in group 
A, including worniu (wor), deadpan, and CyclinE (Table 3). Conversely, neuronal and 
glial genes are excluded from group A and found in group C [e.g. elav, glial cells missing 
42
(gcm), and reversed polarity (repo)] (Table 3). In addition, there is a good correlation 
between the number of neuroblasts in each mutant brain and the level of enrichment 
shown for group A genes (Figs. 8B and 8B). Thus, the genes in group A are likely to be 
expressed in both type I and type II neuroblasts, but not in neurons or glia. Conversely, 
genes in group C are likely to be expressed in differentiated neurons or glia, but not in 
neuroblasts. We conclude that ectopic neuroblasts are similar transcriptionally to wild-
type neuroblasts, and thus the mutant genotypes represent an enriched source of 
neuroblast-expressed mRNA.
We next determined the gene ontology (GO) terms that represent the neuroblast-
enriched group A genes and the neuron/glia-enriched group C genes. We found that group 
A genes are strongly enriched for several GO terms, including cell cycle and ribosome 
biosynthesis – processes expected in neuroblasts that must repeatedly divide and grow 
(Figs. 12A and 12B). For example, a small sub-cluster in group A exhibits a very 
significant enrichment for genes involved in DNA replication (p<10-15); this process is not 
significantly represented in any other cluster (p>.01 for all other group A genes 
combined; Fig. 12A). Conversely, we found group C to be significantly enriched for the 
GO terms morphogenesis, signal transduction, and differentiation – all expected for post-
mitotic neurons and glia (Fig. 12B). In addition, neuropeptide signaling and cell 
morphogenesis are both significantly enriched (p<10-20 and p<10-21, respectively) in 
distinct sub-clusters (Fig. 12C). We conclude that group A is enriched for genes that are 
expressed in neuroblasts but not differentiated neurons and glia, and group C is primarily 
composed of genes expressed in post-mitotic neurons and glia.
43
Functional analysis of genes predicted to be expressed in neuroblasts but not 
neurons
To determine if group A genes are required for neuroblast survival, proliferation, 
or self-renewal, we performed RNAi knock-down experiments. We selected genes for 
which transgenic RNAi stocks were available from the Vienna Drosophila RNAi Center, 
and we further restricted our analysis to those genes with human orthologs in order to 
enhance the relevance of our study to issues of human stem cell function (Fig. 13). This 
resulted in our analyzing 691 RNAi lines representing 595 genes. We reasoned that loss 
of function of genes with critical functions in neuroblasts would cause defective central 
nervous system development and eventual lethality, as seen in other genotypes which 
affect neuroblast function. Thus we screened for lethality in the progeny of males from 
each RNAi line crossed to wor-Gal4 UAS-Dicer2 flies at 30°C [wor-Gal4 drives 
expression in neuroblasts (Albertson et al., 2004), while Dicer2 improves RNAi efficacy 
(Dietzl et al., 2007)]. Of the 691 RNAi lines tested, 195 (28%) cause lethality or semi-
lethality (Fig. 13A). We found that of the genes for which we tested multiple RNAi lines, 
84% exhibit the same lethality phenotype for both lines. Few RNAi lines cause 
embryonic lethality at 30°C, and in these cases larval stages were obtained by setting up 
crosses at 18°C and shifting larvae to 30°C after embryogenesis. The lack of a lethal 
phenotype in 72% of the lines may be due to either inefficient RNAi knock-down of gene 
expression or the non-essential function of the gene in larval neuroblasts. Hence we 
restricted our subsequent analysis to the 28% of lines with a lethal or semi-lethal 
phenotype.
We tested each of the lethal or semi-lethal genes for a change in neuroblast 
44
number, reasoning that genes expressed in neuroblasts but not neurons may play a role in 
neuroblast survival, quiescence, identity, asymmetric division, or self-renewal. We 
performed the same RNAi experiments as above and determined the number of Dpn+ 
Mira+ central brain neuroblasts (optic lobe neuroblasts were not assayed). We found that 
nearly one half of lethal genes (86) cause a significant change in central brain neuroblast 
numbers (Fig. 13B, Fig. 13C, and data not shown). A majority of these changes are 
decreases in neuroblast number, as expected based on the predicted expression of these 
genes in neuroblasts. Two genes known to regulate neuroblast numbers were detected in 
the screen, mira and aurora borealis, thereby validating this approach. Importantly, all of 
these genes have clear mammalian orthologs. We conclude that our RNAi-based 
screening method has yielded a list of 84 new candidates for regulating neuroblast self-
renewal. 
Discussion
It has previously been shown that co-clustering of genes in expression profiling 
data is likely to reflect physical or genetic interactions (Ge et al., 2001; Jansen et al., 
2002) and participation in the same pathways (van Noort et al., 2003). Our results are 
consistent with these conclusions. For example, we identified a small group of 11 genes 
containing the only two genes known to be expressed in type I but not type II neuroblasts, 
and showed that a third gene has a similar pattern of expression – thus all genes in this 
cluster are likely to be expressed in type I but not type II neuroblasts. Furthermore, the 
strong enrichment of GO terms in small sub-clusters within both group A and group C 
(Fig. 12) indicates that genes within these clusters are likely to share similar functions or 
45
processes.
Differences between type I/type II neuroblasts are caused by a small number of 
genes
At the outset of this study, we expected to find a large group of genes that were 
differentially expressed in type II versus type I neuroblasts, because these neuroblasts 
have such strikingly different cell lineages. However, we were only able to identify a few 
gene clusters that were differentially regulated in such a type I/type II consistent manner 
– the 11 genes in the pros/ase cluster depleted in type II neuroblasts and the six genes 
enriched in type II neuroblasts (Figs. 10A and 11A). This suggests that the small number 
of genes that we identified may play a disproportionately large role in generating 
differences between type I and type II neuroblasts. Might pros and ase be the only genes 
regulating type I/type II differences? Both Ase and Pros can promote cell cycle exit 
(Choksi et al., 2006; Dominguez and Campuzano, 1993; Li and Vaessin, 2000; Southall 
and Brand, 2009; Wallace et al., 2000), which may result in the Ase+ Pros+ type I progeny 
taking a GMC identity and undergoing just one terminal division and the Ase- Pros- type 
II progeny taking an INP identity and continuing to proliferate. Indeed, the misexpression 
of either Ase or low levels of Pros in type II neuroblasts is sufficient to cause the loss of 
INPs and/or their premature cell cycle exit, thereby decreasing lineage size toward the 
size of type I neuroblasts (Bayraktar et al., 2010; Bowman et al., 2008). However, it is 
unclear what is required to fully transform these cells into type I neuroblasts; addressing 
this question will require additional molecular markers (some provided by our work here) 
and tracing the axon projections of the progeny of these "transformed" neuroblasts (e.g. 
do they now fail to make intrinsic neurons of the adult central complex?). The fact that 
46
mutants in ase and pros do not transform type I neuroblasts into type II neuroblasts 
(Bowman et al., 2008; Weng et al., 2010) indicates that other genes, perhaps some in the 
pros/ase cluster described here, are also important for specification of type I neuroblast 
identity.
Group A: candidate neuroblast-specific genes and neuroblast homeostasis regulators
We found that the neuroblasts in each mutant have remarkably similar expression 
profiles, as shown by the extensive list of similarly expressed genes in group A and by the 
list of genes with depleted expression in mutant brains, represented by group C. We 
believe that these categories provide lists of genes that are representative of those 
expressed in neuroblasts and neurons, respectively, based on all known neuroblast-
specific genes showing up in group A and all known neuron- or glial-specific genes being 
excluded from group A.
Our RNAi-based screen helped to substantiate this claim, in that a substantial 
percentage of group A genes caused lethality when subjected to neuroblast-specific 
knock-down. We do not believe that off-target effects lent a significant amount of error to 
these lethality data for two reasons: (1) a similar percentage of lethal and non-lethal 
RNAi lines (about 30%) had more than one non-specific target, which indicates that the 
observed lethality was due to specific target knock-down; and (2) a majority of genes 
(about 85%) caused the same lethality phenotype when targeted with multiple 
independent RNAi lines. Interestingly, we found that many RNAi lines caused lethality 
with no concomitant change in neuroblast numbers (Fig. 13). We believe this to be due to 
neuroblast defects which disrupt normal brain development without causing neuroblast 
loss, per se. For instance, neuroblast failure to make the proper number or type of 
47
progeny might be expected to cause such a phenotype. It will be interesting to investigate 
the specific effects these essential genes have on neuroblast function. We note that all of 
the putative regulators of neuroblast homeostasis identified here have mammalian 
orthologs; these genes are excellent candidates for regulating self-renewal of mammalian 
neural stem cells.
Group B: expression in subsets of neuroblasts or neurons?
Group B genes apparently are not expressed in all neuroblasts like the group A 
genes, nor in all neurons or glia like group C genes. However, group B genes are more 
likely to be expressed in subsets of neurons, not neuroblasts, because group B genes as a 
whole have an over-representation of GO terms more similar to group C than to group A 
(Fig. 12B). Why then are group B genes excluded from group C, the neuron cluster? One 
possible explanation is that different neuroblast lineages are affected in each mutant, and 
thus different subsets of neurons are missing in each mutant. If different neuroblast 
lineages express different genes (which seems likely), then each mutant would be missing 
a unique subset of neural differentiation genes, leading to the cluster being excluded from 
group C. This model raises the intriguing possibility that group B sub-clusters may 
represent lineage-specific genes.
It is also possible that the mutant genotypes themselves may cause unique 
transcriptional differences, leading to a cluster of genes in group B. For example, several 
small sub-clusters in group B are expressed differently only in aPKCCAAX brains (data not 
shown). These transcriptional differences are not correlated with the number of type I or 
type II neuroblasts. Instead, these genes appear to be differentially expressed in response 
to elevated aPKC. Drosophila aPKC has been best studied as a component of the apical 
48
complex in mitotic neuroblasts, and its capacity for causing ectopic self-renewal has been 
shown to be reliant on both its catalytic activity and its membrane localization (Atwood 
and Prehoda, 2009; Lee et al., 2006b). However, aPKC has been ascribed a role in 
neuroblast proliferation as well as in polarity (Chabu and Doe, 2008; Rolls et al., 2003), 
and a vertebrate homolog, PKC-ζ, was shown to possess a nuclear role in both 
proliferation of neural progenitors and neuronal cell fate specification (Sabherwal et al., 
2009). These observations are consistent with a role of aPKC in causing transcriptional 
differences. 
Conclusions
Our findings highlight the importance of expression profiling of multiple 
genotypes. This method allowed us to get a more reliable picture of the group A genes 
expressed in neuroblasts, because genes with lineage-specific or genetic background-
specific changes in expression appeared to be focused into group B, where they do not 
interfere with the clustering of groups A and C. In addition, we identified two small sub-
clusters of genes in group B that are excellent candidates for being preferentially 
expressed in type I or type II neuroblasts, for which there have been few examples to 
date. Finally, we conclude that group A genes are likely to be expressed in neuroblasts, 
and our functional studies have identified 84 genes that are conserved in mammals and 
required for regulating neuroblast numbers in Drosophila. Future phenotypic analysis in 
Drosophila will determine whether these genes regulate neuroblast survival, quiescence, 
asymmetric cell division, and/or self-renewal. Future studies on the expression and 
function of orthologous genes in mouse neural progenitors and human stem cells (IP or 
neural) will reveal whether they have conserved roles from flies to mammals.
49
CHAPTER IV
MOSAIC ANALYSIS WITH DOUBLE MARKERS REVEALS TUMOR CELL OF 
ORIGIN IN GLIOMA
This work was published in volume 146 of the journal Cell in July 2011. Chong 
Liu performed the immunohistochemistry analysis. Chong Liu, myself, and Roel Verhaak 
performed the microarray-based expression analysis. Chong Liu and Johnathan Sage 
performed the quantitative PCR analysis. Simon Hippenmeyer contributed MADM 
mouse strains. Hui Zong, Liqun Luo, Akiko Nishiyama, Roderick Bronson, Oded 
Foreman, and Hannes Vogel were the principle investigators for this work. 
Introduction
Cancer is a disease of genetic mosaicism because cancerous cells harbor genetic 
mutations that are absent in normal cells within the same individual. In familial cancer 
patients, even though initial mutations exist in every cell, in most cases, only specific cell 
types can progress into malignancy. Those cell types are called cancer cell of origin. Such 
a cell type-specific susceptibility implies the existence of a permissive or even 
synergizing signaling context in the cell of origin for particular genetic mutations to cause 
cancer formation. Therefore, identification of the cancer cell of origin would provide 
critical insights for understanding tumorigenic mechanisms and for designing rational 
therapeutic strategies.
Despite such importance, identification of the cell of origin for most cancers has 
been a daunting task (Visvader, 2011). The reliability of revealing cell identity solely 
based on molecular and cellular analyses of late stage tumors is often confounded not 
only by infiltrated bystander cells, but also by the acquired plasticity often found in 
50
terminal cancerous cells. To circumvent these issues, genetically engineered mouse 
models have been widely used to determine the tumorigenic potential of a specific cell 
type by initiating mutations with a cell type-specific Cre transgene. However, it is critical 
to note that cells initially acquiring mutations (cell of mutation) may not be the cell of 
origin. When mutations are introduced in stem/progenitor cells, it is extremely difficult to 
distinguish whether initial mutant cells directly transform or whether they merely pass on 
mutations to more restricted progeny that then transform. In the latter scenario, the 
mutated stem/progenitor cell is simply the cell of mutation, and the transforming progeny 
is the actual cell of origin (Visvader, 2011).
The cell of origin for malignant glioma, a type of deadly brain cancer, remains 
controversial. Successful isolation of tumor cells with stem cell features (known as 
cancer stem cells) from human gliomas (Singh et al., 2004) implies neural stem cells 
(NSCs) as the cell of origin. However, such NSC-like features of malignant glioma cells 
could be acquired during transformation rather than reflect the nature of the original cell 
type (Visvader, 2011). Further evidence supporting the NSC origin of glioma was 
obtained from mouse genetic studies. For example, the inactivation of tumor suppressor 
genes (TSGs) p53 and neurofibromatosis 1 (NF1) or the expression of a mutant form of 
p53 in NSCs consistently led to glioma formation in mouse models, and the physical 
locations of tumors appeared to associate with the subventricular zone (SVZ), where 
adult NSCs reside (Alcantara Llaguno et al., 2009; Wang et al., 2009; Zhu et al., 2005). 
However, other studies suggest that NSC-derived progeny such as astrocytes or 
oligodendrocyte precursor cells (OPCs) might directly transform (Bachoo et al., 2002; 
Lindberg et al., 2009; Persson et al., 2010). This unresolved controversy partially stems 
51
from the distinct oncogenic mutations used in these models that make the direct 
comparison difficult and, more importantly, from the lack of high-resolution analyses of 
cellular aberrations during the transforming process.
Mosaic analysis with double markers (MADM), a mouse genetic mosaic system 
(Zong et al., 2005), could, in principle, be used to analyze aberrations in individual cell 
lineages prior to the final transformation and should thus be suitable for identifying 
cancer cell of origin. Via Cre/loxP-mediated mitotic interchromosomal recombination, 
MADM generates a small number of homozygous mutant cells, thus mimicking the 
sporadic loss of heterozygosity (LOH) of TSGs in human cancers (Knudson, 1971). 
MADM also permanently labels these mutant cells with green fluorescent protein (GFP) 
and their sibling wild-type (WT) cells with red fluorescent protein (RFP) within an 
otherwise unlabeled heterozygous mouse (Fig. 14A). The single-cell resolution benefited 
from the sparse labeling (0.1%–1% or much lower) (Zong et al., 2005) enables one to 
track mutant cells throughout the entire process of tumorigenesis. The sibling red WT 
cells serve as internal controls for green mutant cells, thereby greatly facilitating detailed 
analyses of cellular aberrations of all lineages in their native environment. In summary, 
MADM can provide features that are indispensable for a robust analytical paradigm to 
identify a cell of origin.
Here, we report the application of MADM to glioma modeling. After initiating 
p53/NF1 mutations sporadically in neural stem cells (NSCs), we analyzed mutant NSCs 
and all of their progeny at premalignant stages. We found dramatic overexpansion and 
aberrant growth specifically in OPCs, but not in NSCs or other lineages. Upon tumor 
formation, marker staining and transcriptome analysis confirmed the OPC nature of 
52
tumor cells. Finally, introducing the same mutations into OPCs consistently led to 
gliomagenesis. Our findings reveal OPCs as the cell of origin for glioma even when 
initial p53/NF1 mutations occur in NSCs, thus resolving the current controversy by 
distinguishing cancer cell of mutation from cell of origin. Importantly, although our 
studies focused on glioma, the analytical paradigm with MADM that is developed here 
could be applied to identify cellular origins for many other cancers.
Methods
Mouse lines
Mouse lines used for NSC-Cre glioma model: a pair of MADM mice (TG11, 
GT11) (Hippenmeyer et al., 2010), hGFAP-Cre (Zhuo et al., 2001) or Nestin-Cre 
(Petersen et al., 2002), p53KO (Jacks et al., 1994) and NF1flox (Zhu et al., 2001). Mouse 
lines used for NG2-Cre glioma model: a pair of MADM mice (TG11ML, GT11ML, 
unpublished lines), NG2-Cre (Zhu et al., 2008), p53KO (Jacks et al., 1994) and NF1flox 
(Zhu et al., 2001). TG11ML and GT11ML mouse lines have three mutually exclusive 
loxP variants in MADM cassettes, instead of single loxP sites in TG11 and GT11 mouse 
lines. This design was used to increase the efficiency of inter-chromosomal 
recombination induced by Cre transgenes.
BrdU labeling
BrdU was administered by i.p. injection (50mg/kg body weight) for short term 
labeling (1.5hours) or by drinking water (1mg/ml) for long term labeling (7 days). Mice 
were sacrificed right after the treatment and brain tissue was collected for the further 
analysis.
53
Pathology
Tissues used for pathological analysis were collected from O.C.T.-embedded 
brains. The tissue was thawed and the O.C.T. was removed by washing with cold PBS. 
The tissues were then preserved in 4% PFA until being processed for paraffin embedding 
and H&E staining.
Immunohistochemistry
Most immunohistochemistry in our work involved MADM staining (for GFP and 
RFP), which occupies the commonly used green (488nm laser) and red (543 nm laser) 
channels in confocal imaging. Considering this, marker staining other than MADM was 
applied in either the Far-red channel (635 nm laser) or ultraviolet (UV) channel (405 nm 
laser), or both. If only a single marker was stained along with MADM (which we named 
three-channel staining, see below), we usually stained this marker with the appropriate 
secondary antibody to visualize it in the Far-red channel (such as Alexa Fluor 647 or Cy5 
conjugated secondary antibodies). If more than one marker besides GFP and RFP was 
used (we named it as four-channel staining), the fourth would be the UV-channel which 
was visualized by using appropriate biotin-conjugated secondary antibody to recognize 
the primary antibody and then Alexa Fluor 405 conjugated streptavidin to develop the 
fluorescent signal.
Occasionally, we performed staining for three markers along with GFP and RFP 
(such as those in Figs. 16, we named it as “4+1” channel staining). In this way, we 
borrowed the red channel from RFP for visualizing the fifth marker. Although this fifth 
marker staining cannot be visualized in either red WT or double-color cells, it helps us 
identify marker expression in green mutant cells, especially within tumor region or 
54
among the pre-transforming OPC population where cells are almost exclusively green or 
colorless. For brain tissue containing grafted tumor cells (secondary tumors) the red 
channel was saved for staining cell markers other than RFP because no RFP labled WT 
cells existed in the tumor.
Staining procedures
For three-channel staining with MADM and one more marker, tissue sections 
were air-dried at room temperature for at least 1 hr after cryosectioning. Tissue sections 
were then re-hydrated and washed three times (10 min each) with phosphate-buffered 
saline (PBS) to remove sucrose prior to incubation with blocking/ permeablizing buffer 
(5% normal donkey serum and 0.3% Triton X-100 in PBS) for 30 min at room 
temperature to prevent non-specific binding. After this step, the sections were incubated 
overnight at 4°C with appropriate primary antibodies that were diluted in blocking/ 
permeablizing buffer.
The sections were washed three times (10 min each) with PBT (0.3% Triton X-
100 in PBS) before incubation overnight at 4°C with appropriate fluorophore-conjugated 
secondary antibodies that had been diluted in PBT. Sections were washed three times in 
PBT (10 min each) and followed by one wash in PBS (5 min) before being mounted with 
anti-fade mounting gel mounting medium. To visualize nuclei, the slides were incubated 
with DAPI solution (25ng/ml in 1X PBS) for 10 30min before the last PBS wash.∼
For four-channel staining with MADM and two more markers, the staining 
procedure is the same as for three-channel staining, excepting for the visualization of the 
UV-channel signal by fluorophore Alexa Fluor 405. Alexa Fluor 405-conjugated donkey 
antibody was not commercially available when this work was performed. Alternatively, 
55
we incubated the sections with Biotin conjugated species-specific donkey secondary 
antibody (For the UV channel) together with the fluorophore-conjugated secondary 
antibodies for the other three channels overnight at 4°C. After a brief wash in PBT (10 
min), the sections were incubated with Alexa Fluor 405-conjugated Streptavidin (1:200) 
in PBT for 2 hr at room temperature. The slides were mounted in anti-fade gel mounting 
medium after three washes in PBT (10 min each) and one wash in PBS (5 min).
For staining including BrdU as a marker, slides used for BrdU detection were 
treated with HCl prior to primary antibody incubation. Briefly, after three washes in PBS, 
the sections were submerged in 2N HCl (in PBS) for 30 min at 37°C. The sections were 
then washed three times with PBS to remove HCl prior to incubation with the blocking 
solution.
For staining including Nestin as a marker, we found that Mouse anti-Nestin 
monoclonal antibody generated high background signal in mouse tissue. To remove the 
background signal, we used Donkey-anti-mouse monovalent Fab fragment to block the 
tissue sections prior to primary antibody incubation. Briefly, after incubation in blocking 
solution for 1 hr, the slides were washed three times in PBT (2 min each) before 
incubation with Fab Donkey-anti-mouse IgG (H+L) (1:10 diluted in 1X PBT) overnight 
at 4°C. The slides were then washed three times in PBT (10 min each) prior to incubation 
with anti-Nestin primary antibody.
For tissue culture staining, cells used for immunohistochemistry were cultured on 
poly-D-lysine coated 12mm glass coverslips. For staining, after gently removing the 
medium from the tissue culture plates, 0.5ml 4% PFA was added to each well and the 
cells were fixed at room temperature for 15 min. The coverslips were then washed one 
56
time with 1ml PBS, dried at room temperature for 30min and were kept in 1X PBS until 
used for staining. To stain surface antigen O4, Triton X-100 was omitted during the entire 
process. Otherwise, 0.1% Triton X-100 was used for permeabilization.
Cells were incubated in blocking solution (10% NDS in PBS with or without 
0.1% Triton X-100) for 1 hr at room temperature, and then incubated with primary 
antibody in blocking solution for 1 hr prior to three washes with PBS (with or without 
0.1% Triton X-100, 10 min each). Cells were then incubated with appropriate secondary 
antibody (10% NDS in PBS with or without 0.1% Triton X-100) for 45 min at room 
temperature followed by 3 washes in PBS (with or without 0.1% Triton X-100, 10 min 
each). Before being mounted with anti-fade gel mounting medium, cells were incubated 
with DAPI for 5 min and washed one more time with PBS.
Brain regions sampled for quantification
In the present study, we define 23 distinct adult brain regions in the 
nonneurogenic parenchyma of sagittal brain sections for cell quantification. Selected 
regions cover major representative brain structures encompassing both gray and white 
matter, including the cerebral cortex (four random fields from rostral to caudal), the 
corpus callosum (four random fields from rostral to caudal), the striatum (two random 
fields), the ventral forebrain (ventral to striatum, two random fields), the thalamus (two 
random fields), the hypothalamus (two random fields), the fornix, the anterior 
commissure, the olfactory bulb (two random fields) and the midbrain (two random 
fields). Cerebellum was not included in the counting scheme because no tumor was found 
in this region. We also define 5 regions close to the lateral ventrical wall of adult P60 
brains used for counting both slow and fast dividing cells in neurogenic SVZ, including 
57
the dorsal lateral angle SVZ (S1) and the lateral wall SVZ (S2-S5).
Quantification of G/R Ratios from P5 to P60 mutant-MADM mouse model 
presented in Fig. 15
Fixed brains were cryosectioned sagittally. Three sections (20 μm in thickness) at 
least 200 μm apart from each other (N = 3 brains for each age) were stained and 
processed for data collection. To quantify MADM labeled cells and BrdU+ cells in the 
same brain, we stained the slides with MADM together with BrdU and DAPI. For each 
brain section, 9 regions were collected, including three in the cerebral cortex (Figs. 14A-
14C), three in the corpus callosum (Figs. 15A, 15B and 15D), and three in the ventral 
side of the brain (Figs. 16A, 17A/20A, and 17B/20B). Each image was taken by a 40X 
objective with 1X digital zoom (0.1mm2 each image) by 2 μm optical sectioning 
(scanning depth 6 μm). Only the cells with clearly visible nuclei were counted.
Quantification of neurons, astrocytes, OPCs, and oligodendrocytes in nonneurogenic 
parenchyma of P60 mutant-MADM mouse brains presented in Fig. 16A
To count the cell number and G/R ratios of four distinct cell types in Fig. 16A, 
P60 brains (N = 3 brains, 4 slices from each brain) were sectioned sagittally in 14 μm (for 
neurons, oligodendrocytes and OPCs) or 30 μm (for astrocytes) slices. Cell types were 
identified by MADM-labeling and cellular markers (NeuN for neurons, CC1 for 
oligodendrocytes and PDGFRα for OPCs) except for astrocytes. Because Myc (to 
visualize RFP, which was fused with a Myc-tag at the C-terminal (Hippenmeyer et al., 
2010)) and Aldolase C (for astrocytes) antibodies were both raised from goat, we omitted 
the Myc antibody in such staining. We found that the live color of RFP was stable for 
several months and strong enough in its fluorescence intensity even without antibody 
58
staining. Therefore omitting myc staining did not affect the identification of red or double 
color cells.
We quantified neurons from 14 regions; and astrocytes and oligodendrocytes from 
16 regions. Only cells with clearly visible cell bodies were counted. OPC quantification 
was performed together with BrdU long-term labeling quantification (see below). Each 
image was taken using a 40X objective with 1X digital zoom (0.1mm2 area) by 1 μm 
optical sectioning (scanning depth 10 μm). The number of red and green cells for each 
marker was summed together from all the regions of the four slices to calculate the 
average G/R ratio for the respective cell type in that brain. This value was averaged from 
three brains and presented as ± SEM.
Quantification of BrdU-positive OPCs in the nonneurogenic parenchyma of P60 
mutant-MADM brain
Brain tissues were sectioned sagittally in 14 μm thick slices and stained with 
MADM plus PDGFRα (Far-red channel) and BrdU (UV channel). We collected images 
from 21 regions. We processed each image in two ways for different purposes. First, we 
counted all the PDGFRα+ cells, including red, green, yellow and non-labeled cells. This 
count was used to calculate the G/R ratio of OPCs and the percentage of BrdU labeling 
among OPCs with distinct genotypes. Second, we focused on BrdU+ cells. BrdU+ cells 
will be either PDGFRα+ or PDGFRα- and present one of four possible colors. To confirm 
that PDGFRα- BrdU+ mutant cells also belong to the OPC-oligodendrocyte lineage, we 
co-stained adjacent slices with MADM plus PDGFRα (far-red channel), BrdU (UV 
channel) and Olig2 (Red channel) (Fig. 16).
59
Quantification of BrdU+ cells in the neurogenic subventricular zone of P60 mutant-
MADM
To identify the SVZ, we sectioned brains coronally. Four sections (30 μm in 
thickness, at least 200 μm apart) from each brain (three brains in total) were counted. The 
SVZ was divided into dorsal lateral angle SVZ (Region S1) and the lateral wall SVZ 
(region S2-S5). We used a 40X objective with 2X digital zoom to collect each image 
(imaging area is 0.025mm2 each). Bilateral SVZs in each brain section were scanned. 
Therefore, ten images were collected from each brain slice. The cell number from 40 
images derived from a single brain was summed to calculate the G/R ratio and the 
percentage of mutant, WT and heterozygous cells among all BrdU positive cells in that 
brain. These values were averaged from N = 3 brains and presented as ± SEM.
Characterization of NG2-Cre MADM mice
NG2-Cre mediated MADM-labeling in the SVZ was analyzed in coronal 
cryosections (20 μm) at P360 (N = 4). Non-adjacent brain sections separated by 60 μm 
were collected from the rostral to caudal axis in the forebrain (i.e., collected one from 
every four sections), and stained with GFAP and DAPI. The SVZ was defined on each 
section based on DAPI staining until the SVZ region could not be clearly identified. We 
used a 40X objective with 2X digital zoom to scan for MADM labeled cells within the 
SVZ and then performed multiple optical layer confocal sectioning (0.5 μm for each step) 
to examine whether the MADM-labeled cells expressed GFAP.
To examine MADM-labeled cells in the olfactory bulb (OB), bilateral OBs were 
coronally sectioned. 100 continuous sections were collected and immunostained with 
anti-NeuN antibody. The granule cell layer (GCL) was identified by DAPI staining. Four 
60
fields in the GCL of each OB section were randomly chosen using a 40X objective with 
2X digital zoom. If any MADM labeled cells were found, further digital zoom (up to 4X) 
combined with multiple optical sectioning (0.5 μm for each step) was performed to 
carefully examine whether these cells co-localized with NeuN staining. 800 fields were 
scanned from the bilateral OBs of a single mouse (N = 4 mice).
Enrichment of OPC-like glioma cells
OPC-like tumor cells were purified from fresh glioma tissues as previously 
described (Cahoy et al., 2008) with minor modification. To prepare BSL-1 plates, two 
Petri dishes (100 × 10 mm) were incubated with BSL-1 solution (25 μg BSL-1 in 10ml 
DPBS for each plate). The solution was swirled carefully to ensure that each whole Petri 
dish was completely coated with solution prior to incubation overnight at 4°C. The next 
morning, the BSL-1 solution was removed from the plates and 20ml 0.2% BSA (in 
DPBS) was added to block the plates at room temperature until panning was performed 
(usually more than 4 hr).
To prepare PDGFRα immunopanning plates, two Petri dishes (100 × 10 mm) were 
each incubated with 10 ml of a secondary antibody solution containing Tris buffer 
solution (50 mM pH8.5 Tris-Cl) and 75 μg (30 μl) goat anti-rat IgG H+L chain (112-005-
167) overnight at 4°C. The next morning, the secondary antibody solution was removed 
and the plates were washed three times with 1X DPBS. Subsequently, a primary antibody 
solution containing 10 μl Rat-anti-PDGFRα antibody in 10ml DPBS with 0.1% BSA was 
added to each plate. The plates were then incubated at room temperature until used and 
washed directly before immunopanning.
61
Cell dissociation
All dissection tools were sterilized by soaking in Cidex for 10 min and then rinsed 
with autoclaved distilled water prior to dissection. Tools were kept in 70% ethanol 
between usages. Tumor-carrying mice were euthanized by cervical dislocation, sprayed 
down with 70% ethanol and then wiped with Betadine to remove loose fur. The head was 
cut off with a pair of large dissection scissors and then soaked in 70% ethanol for 45 s to 
1 min. The skin down the midline of the head was sliced through with a razor blade. The 
skull was cut open using a pair of small dissection scissors. Brain tissue was then lifted 
from the skull and washed once with sterilized cold PBS before being embedded into 
chilled but still liquid 4% low-melting temperature agarose (prepared in 1X PBS) on ice. 
After the agarose became solid, the brain was sliced into 1mm thick coronal sections and 
kept in cold SMEM in a Petri Dish on ice. Under a fluorescence dissection scope, tumors 
were easily identified because of the strong MADM fluorescence. The tumor region was 
dissected out with a 22.5° straight microsurgical knife. Adjacent brain slices containing 
some tumor tissue were preserved in 4% PFA immediately after dissection. These slices 
were used for immunohistochemistry and H&E staining for tumor pathology.
The remaining dissected tumor tissue was transferred to a 60 mm Petri dish 
containing 5ml Papain solution and diced into 1 mm3 pieces. The Papain solution∼  
includes 1X EBSS, 0.5mM EDTA, 10mM HEPES, 26mM NaHCO3, 0.16mg/ml L-
Cysteine, 20Unit/ml Papain, 1mg/ml DNase and was carefully balanced in 5% CO2 at 
37°C for at least 30 min before use.
The tumor tissue was incubated in Papain solution at 37°C 5% CO2 for 90 min 
(gently agitated every 15 min). Afterwards it was transferred into a sterile 15ml Conical 
62
polystyrene tube and allowed to settle to the bottom of the tube (we used a 10ml 
serological pipette to transfer the tissue to avoid losing the cells at this step). The Papain 
solution was then pipetted off and 2ml Diluted trypsin inhibitor buffer was added. Diluted 
trypsin inhibitor includes 1X EBSS, 10mM HEPES, 26mM NaHCO3, 1mg/ml 
ovomucoid, 1mg/ml BSA and was carefully balanced in 5% CO2 at 37°C for at least 30 
min before use. After the tissue settled down ( 2 min), the diluted trypsin inhibitor∼  
buffer was replaced by 2ml fresh diluted trypsin inhibitor buffer. The tissue was then 
triturated 8 times with a 5ml Pasteur pipette. After the tissue settled, 1ml of supernatant 
(containing dissociated cells) was carefully transferred to a new 15ml conical polystyrene 
tube and set aside. 1ml fresh diluted trypsin inhibitor buffer was added to the original 
tissue, and the trituration procedure was repeated. After the tissue settled, 1ml of 
supernatant was removed and combined with the other set aside supernatant. The 
trituration was repeated until all the diluted trypsin inhibitor buffer was used (10ml total). 
The last two rounds of trituration were performed with a 1ml pipette tip to ensure all the 
visible tissue chunks were fully dissociated into single cells.
Once the tissue was dissociated, 4ml standard trypsin inhibitor solution (10mg/ml 
BSA in diluted tripsin inhibitor buffer) was placed in a new 15ml conical polystyrene 
tube. Dissociated cell suspension was carefully layered on top of standard trypsin 
inhibitor solution. The layered solution was centrifuged at 220 g for 15 min followed by 
immediate removal of the supernatant. Cells were re-suspended in 6ml panning solution 
(0.02% BSA, 1mg/ml DNase in 1X DPBS) using a 5ml pipette and centrifuged at 220 g 
for 10 min. Cells were then re-suspended in 12 ml Panning buffer and passed through a 
70 μm cell strainer before performing immunopanning.
63
Immunopanning
The first BSL-1 coated plate was washed 2X with PBS and 1X with DPBS. 100 μl 
of single cell suspension was set aside for cell number counting and cell culture. The 
remaining cell suspension was poured onto the first BSL-1 plate and incubated at room 
temperature for 15 min. During that incubation time, we counted the living cell number 
using the Trypan Blue method and a hemacytometer. The second BSL-1 coated plate was 
washed 2X with PBS and 1X with DPBS before the supernatant from the first plate was 
transferred to it.
During the second BSL-1 panning plate, one primary antibody coated plate was 
washed 3X with 20ml DPBS. The supernatant from the second BSL-1 plate was 
transferred to the washed primary antibody plate and incubated for 45min, and then the 
supernatant was transferred to the second primary antibody coated. During the second 
primary antibody panning plate, the first primary antibody panning plate was washed 6-8 
times with DPBS to remove the non-binding cells. At the end of the last wash the plate 
was checked under the microscope to visualize floating cells. Additional washes were 
performed until there were almost zero and certainly less than 5% floating cells.
Cells were collected by incubating the first primary antibody panning plate with 
2ml pre-balanced TrypLE in 5% CO2 at 37°C for 7 min. The reaction was then stopped 
by adding 2ml trypsin inhibitor solution (2mg/ml Ovomucoid, 0.4% BSA in 1X DPBS). 
Tumor cells were dislodged from the plate surface by squirting with a 1 ml pipette tip. We 
collected the cells and also the supernatant from the second primary antibody plate and 
transferred them separately to new 15ml conical polystyrene tubes which were pre-coated 
with 0.4% BSA at 37°C for 1 hr to prevent adherence of cells to the tubes. After 
64
centrifugation at 220 g for 10min, the cells were washed one time with NSC medium (see 
cell culture section for recipes) before re-suspension in the same medium for further use 
(RNA extraction or allografting, see below).
WT neurosphere culture
The fetal NSCs were enriched from two independent litters of E15.5 WT 
embryonic brains according to the method described previously (Pollard et al., 2006) with 
minor modification. The forebrains (removed cerebella, olfactory bulbs and midbrains) 
were dissociated with Papain solution (as described above) and triturated into single cells. 
The dissociated cells were adjusted to 100,000 cells/ml and plated onto Poly (2-
hydroxyethyl methacrylate) pre-treated 6-well tissue culture plates. The cells were 
cultured in NSC medium containing Neurobasal medium with B27 supplement (50X) 
plus Penicillin/ Streptomycin solution (100X), GlutaMAX (100X), 50ng/ml EGF and 
10ng/ml FGF.
After three days of culture in 5% CO2 at 37°C, the primary neurospheres were 
pooled together and passaged twice as follows. We collected neurospheres in a 15ml 
conical polystyrene tube and let them settle for 10 min before removing the supernatant. 
The neurospheres were washed once with DPBS and incubated with pre-warmed 1X 
TrypLE for 20 min at 37°C. TrypLE was inactivated with equal volume 2X trypsin 
inhibitor solution (2mg/ml Ovomucoid, 0.4% BSA in DPBS) and the spheres were then 
dissociated into single cells by triturating 40 times with a 1ml pipette tip. Cells were 
centrifuged at 220xg for 10 min. The supernatant was removed and the cells were washed 
once with NSC medium. Cells were passed through a cell strainer and plated onto a new 
plate and cultured to form secondary neurospheres. Three to four days later secondary 
65
neurospheres were dissociated into single cells as above. Dissociated cells were expanded 
in NSC medium using 1X adherent 25cm2 PDL-coated flask for 4 days and passaged one 
more time in 3X 25cm2 PDL-coated flasks. The expanded cells were trypsinized off the 
flasks, washed, and frozen in 15% DMSO (in NSC medium) in liquid nitrogen for further 
use or directly lysed in TRIzol for RNA analysis.
Real-time quantitative PCR
Total RNA was extracted by TRIzol followed by phenol/chloroform extraction 
and ethanol precipitation (Miller et al., 2009). cDNA was then synthesized using 
Superscript III and Oligo (dT) Primers. qRT-PCR primers were designed using 
MacVector v10.0 and Primer 3 Plus to cover two exons flanking a > 500bp intronic 
region. Gene amplification was tested by at least two distinct functional primer sets. Only 
primer sets with the best performance for each gene were used for further analysis.
qPCR was done using an ABI 7900HT Real-Time PCR System for 40 cycles by 
denaturing at 95°C for 15 s, annealing at 58°C for 30 s and elongating at 72°C for 30 s, 
followed by the default dissociation curve program. PCR amplification was performed in 
RT2 Real-Time SYBR Green/Rox PCR Master Mix. Relative levels of cDNA were 
measured by using control primers for β-Actin (ActB) and/or Glyceraldehyde-3-
Phosphate Dehydrogenase (Gapdh) to normalize gene expression for each sample. CT 
values were measured within the geometric amplification phase and averaged for 
triplicate reactions. Reactions with a CT value of 35 or higher were considered null. 
Values for individual genes were normalized by subtracting the average of the internal 
control gene(s) Gapdh and/or ActB.
66
Microarray and gene expression analysis
44K Mouse Development Oligo Microarrays from Agilent Technologies were 
used for microarray analysis. For each experiment, total RNA was fluorescently labeled 
and hybridized directly against a common reference sample generated from the RNA pool 
of four WT P17 mouse brain neocortexs. After scanning the images by using a GenePix 
4000B scanner, the intensity values were extracted by using the Agilent Feature 
Extraction Software. In this study, we only focused on the channel of the tumor samples 
to extract their intensity values, but ignored the reference channel signals. Probe 
sequences were downloaded from the Agilent website and aligned to a transcript 
sequence database consisting of 49,040 Mus musculus genes from Ensembl57. After 
removing probes that did not align, or aligned with more than 5 mismatches (n = 19,194), 
or probes that aligned to multiple genes (n = 2,283), 23,006 probes representing 17,585 
genes remained. Expression profiles were quantile normalized and log transformed. 
Expression values for genes represented by multiple probes were collapsed by taking the 
mean value of the set of probes.
Data from Cahoy et al. (2008) were downloaded and pre-processed as described 
previously (Verhaak et al., 2010). The data set contains 38 mouse samples, which 
clustered into 8 clusters after selecting 2,000 variably expressed genes. Gene sets were 
defined by selecting the 250 genes with the highest absolute t test statistic value, except 
for the OPC cluster with only two cluster members, where fold change values were used. 
The clusters used in the present study contained the following samples: (1) Postnatal day 
16 myelin oligodendrocytes (n = 4); (2) Postnatal day 6 to 7 OPCs (n = 2); (3) Postnatal 
day 7 neurons (n = 4);. (4) Postnatal day 17-30 astrocytes (n = 6). Gene sets for four 
67
previously described GBM subtypes (Verhaak et al., 2010) were defined by selecting the 
top 250 marker genes as provided, and by mapping the human genes to mouse genes 
using Ensembl57.
Single sample GSEA has been described elsewhere (Barbie et al., 2009; Verhaak 
et al., 2010). In short, genes were ranked by their expression values. The empirical 
cumulative distribution functions (ECDF) of both the genes in the signature as well as the 
remaining genes were calculated. An enrichment score was obtained by a sum of the 
difference between a weighted ECDF of the genes in the signature and the ECDF of the 
remaining genes. This calculation was repeated for all signatures and samples. A positive 
score indicates a high ranking of up-genes in the signature, and low ranking of down-
genes in the signature. A negative value does not indicate the opposite, but rather a lack 
of effect.
Results
Establishment of an MADM-based genetic mosaic model for glioma
MADM-based cancer modeling needs three prerequisites: cancer-causing gene 
mutation(s), MADM cassettes that reside centromerically to the prospective mutated 
genes, and a Cre transgene that expresses in a certain tissue or organ. To establish an 
MADM-based glioma model, we decided to inactivate p53 and NF1, both of which are 
among the most frequently mutated genes in human glioma patients (McLendon et al., 
2008; Parsons et al., 2008) and have been used to model glioma in mice (Reilly et al., 
2000; Zhu et al., 2005). We also engineered MADM cassettes into the Hipp11 genomic 
locus (Fig. 14A) proximal to the p53 and NF1 chromosomal locations on mouse Chr. 11 
68
(Hippenmeyer et al., 2010). To induce MADM-based recombination, we chose hGFAP- 
or Nestin-Cre transgenes (termed NSC-Cre hereafter) that are expressed in both 
embryonic and adult NSCs (Petersen et al., 2002; Zhuo et al., 2001), thereby generating 
MADM-labeled cells in all NSCs-derived lineages: neurons, astrocytes, 
oligodendrocytes, and OPCs (Fig. 14B). With the ability to identify MADM-labeled cell 
types, we can readily analyze the aberrant growth of GFP+ mutant cells by directly 
comparing to their RFP+ WT counterparts within each lineage in the same mouse brain.
MADM-based glioma model offers the opportunity to trace the entire tumorigenic 
process
After recombining mutant alleles of p53 and NF1 with MADM alleles to generate 
a mutant-MADM mouse model, we analyzed overall expansion of green mutant cells at 
different ages. In contrast to the comparable distribution of green and red cells in WT-
MADM brains in which all labeled cells are wild-type (Fig. 15A, left), we observed a 
progressive overrepresentation of green mutant cells in mutant-MADM brains from 
postnatal day 5 (P5) to P60 (Fig. 15A, top three of the right column) and eventually the 
formation of GFP+ full-blown tumors at 5 months of age (Fig∼ . 15A, bottom-right).
Because mutant (green) and wild-type (red) sibling cells originate from the same 
mother cell in equal numbers initially (Fig. 14A), the ratio of green to red cell numbers 
(referred to as the G/R ratio hereafter) allows us to quantitatively evaluate the extent of 
mutant cell expansion. A G/R ratio equal or close to 1 indicates no growth advantage of 
the mutant population, whereas a G/R ratio > 1 indicates a growth advantage of mutant 
over WT cells (referred to as “overexpansion” hereafter). From P5 to P60, the average 
G/R ratio in mutant-MADM brains increased from 4 to 30 (Fig∼ ∼ . 15B), indicating a 
69
continuous overexpansion of mutant cells. However, the increase of G/R ratio appeared 
to reach a plateau between P30 and P60, suggesting that overexpansion of mutant cells 
had largely stopped between those ages. In addition to quantifying cell numbers, we also 
compared the proliferative status between mutant and WT cells with a brief BrdU pulse 
to label cells undergoing DNA replication. At all time points analyzed, the percentage of 
BrdU+ cells in the mutant population was significantly higher than that in the WT 
population (Fig. 15C). Furthermore, whereas WT cells largely ceased to proliferate at 
early postnatal age (P10), some mutant cells ( 0.3% among all mutant cells population)∼  
continued to divide and contributed to the entire population of BrdU+ cells in the brain 
parenchyma at P60 (Fig. 15D). Thus, mutant cells showed not only elevated, but also 
prolonged proliferative capacity compared to WT cells at a stage before tumor formation 
(termed the “pretransforming” stage hereafter).
At later ages (4–5 months), all mutant-MADM mice developed brain tumors with 
strong Ki67 staining (Fig. 15A, bottom-right). Tumors were invariably GFP+ (n = 28), 
indicating that they originated from MADM-induced mutant cells. Although 
transcriptome profiles of these tumors were very similar (see below), pathological 
analysis revealed great heterogeneity in their appearance: some showed typical astrocytic 
features; some showed malignant glioma features such as necrosis, multinuclear giant 
cells, and perivascular and perineuronal satellitosis; and most were highly anaplastic 
(data not shown, Figs. 17B–17D, and Fig. 18D). Taken together, MADM allows us to 
trace mutant cells at all gliomagenic stages and provides the analytical accessibility 
between initial mutations and the final transformation.
70
Mutant OPCs manifest dramatic overexpansion in pretransforming MADM brains
We next sought to determine the identity of glioma cell of origin by separately 
determining the G/R ratio of NSCs and of each NSC-derived cell type at P60, when 
mutant cell expansion had largely ceased but tumors were yet to arise. We predicted that, 
at this ostensibly dormant stage in tumorigenesis, the mutant cell type capable of 
transformation should manifest significant overexpansion and maintain sustained 
proliferative activity. We first quantified the G/R ratios of all four NSC-derived cell types 
(Fig. 16A). The average G/R ratio from different brain regions showed slight reduction of 
mutant neurons and only minor expansion of mutant astrocytes and oligodendrocytes. In 
stark contrast, the G/R ratio of OPCs was > 130, significantly higher than those of the 
other three cell types (Fig. 16A and data not shown). Notably, mutant OPCs appeared to 
have impaired differentiation potentials to give rise to mature oligodendrocytes, as the 
G/R ratio of oligodendrocytes was 10 times less than that of OPCs (Fig∼ . 16A). 
Moreover, brief BrdU pulses at P60 exclusively labeled mutant OPCs in brain 
parenchyma, although the total number was low, suggesting that mutant OPCs were 
progressing but were still relatively dormant at this age (Fig. 16B, left). Remarkably, 
despite such a dramatic overrepresentation of mutant OPCs, we did not observe any 
pathological features in these brains by conventional pathological analysis (data not 
shown), demonstrating the capability of MADM to probe into a previously inaccessible 
stage during tumor development.
Mutant OPCs constitute the majority of the proliferation pool in pretransforming 
MADM brains
It seems paradoxical that we found nearly full penetrance of tumor formation in 
71
our model at 4–5 months of age, yet dividing cells at P60 were so rare (Fig. 15C and Fig. 
16B, left). Considering that pretransforming cells could undergo a prolonged cell cycle, 
we evaluated the slow-dividing cell population by extending the duration of BrdU 
administration and observed that 30% of all mutant OPCs were able to incorporate BrdU 
(Fig. 16B, right, and data not shown), suggesting that many mutant OPCs were dividing 
slowly. We noticed that a small fraction of BrdU-positive mutant cells in the brain 
parenchyma did not express the OPC marker PDGFRα (Figs. 16B, right, and 16C). To 
determine whether these were cells from other lineages or oligodendrocytes differentiated 
from recently divided OPCs, we costained brain sections adjacent to those shown in Fig. 
16C with antibodies against BrdU, PDGFRα, and Olig2, a “pan-oligo” marker that 
expresses in both OPCs and oligodendrocytes. Using a “4+1 channel” staining scheme 
(Figs. 16C–16E), we found that all BrdU+, PDGFRα-negative cells examined expressed 
Olig2 (Fig. 16D), suggesting that they were newly differentiated oligodendrocytes from 
initially BrdU-labeled mutant OPCs. Taken together, our detailed analysis reveals that, 
among all mutant cell types, OPCs are the only proliferative population in the brain 
parenchyma at pretransforming stages.
Although we did not observe the same extent of overexpansion in other mutant 
cell types as that in mutant OPCs at P60, it could be a result of increased cell death in 
neuronal and astrocytic lineage following the overexpansion of NSCs. Therefore, it is 
critical to directly analyze the proliferative activity of adult NSCs within the SVZ 
(Doetsch et al., 1999). After 7 days of BrdU feeding of mutant-MADM mice, we 
quantified all BrdU+ cells in the SVZ, which include type B, C, and A cells (Doetsch et 
al., 1999). In stark contrast to our findings in the brain parenchyma, where 85% of BrdU-
72
positive cells were mutant (Fig. 16G, left), only 1% of BrdU-positive cells were mutant∼  
in the SVZ (Fig. 16G, right). More importantly, the G/R ratio of BrdU+ cells in adult 
SVZ was not significantly different from 1 (p = 0.39), indicating that there was no 
overexpansion of mutant adult NSCs (Fig. 16F). Furthermore, we did not observe any 
overexpansion of mutant granule neurons in the olfactory bulb, which are the main 
progeny generated by adult NSCs (Fig. 16F). Therefore, p53 and NF1 mutations seemed 
unable to enhance the proliferative activity of adult NSCs. Taken together, detailed 
comparison between mutant and WT cell behaviors in NSCs and all progeny lineages 
strongly suggests that NSCs function as the cell of mutation but fail to directly transform, 
whereas OPCs function as the cell of origin for glioma.
Tumor cells in fully developed gliomas exhibit salient OPC features
Following the investigation at pretransforming stages, we next analyzed the 
expression pattern of cell-specific markers in malignant tumors generated in the MADM-
glioma model (Figs. 17A–17D). Under low magnification, we observed prominent 
enrichment of NSC markers Nestin and Sox2 and the astrocytic marker GFAP, but not 
neuronal or oligodendrocyte markers in the tumor region (Fig. 17E). This observation is 
consistent with previous findings in both mouse models and human patients (Alcantara 
Llaguno et al., 2009; Louis et al., 2007; Wang et al., 2009; Zhu et al., 2005) and has been 
considered as evidence to support the NSC origin of gliomas. However, it is important to 
note that these tumors also showed enriched expression of OPC markers, such as Olig2, 
PDGFRα, NG2, CD9, and O4 (Fig. 17E and data not shown). The expression of some 
OPC markers was further confirmed by quantitative RT-PCR from crude tumor samples 
(Fig. 17F).
73
To clarify whether these markers were expressed by tumor or bystander cells, we 
analyzed the tumor sections at higher magnification and found that GFP+ proliferating 
tumor cells expressed many markers for OPCs (Figs. 17G-I), but not for neurons or 
astrocytes (Figs. 17J and 17K). In some tumors, there were a few GFP+ cells expressing 
oligodendrocyte marker CC-1, but not Ki67 (Fig. 17L), which could be either residential 
cells or oligodendrocytes differentiated from tumor cells. To extend the molecular 
characterization of the cellular identity, we next performed global transcriptome 
comparison between tumor samples and all four neuroglial cell types (Cahoy et al., 2008) 
and confirmed that tumor cells closely resembled OPCs, but not any other neuroglial cell 
types (Fig. 17M). Importantly, regardless of diverse pathological features, all tumor 
samples shared almost identical molecular profiles (Fig. 17M and data not shown), 
suggesting a common cell of origin for these morphologically heterogeneous tumors.
In line with immunofluorescent staining (Fig. 17E), we observed elevated Nestin 
level in some, but not all, tumors by transcriptome and qRT-PCR analyses (data not 
shown). However, it is unclear whether Nestin is expressed in OPC-like tumor cells or in 
bystander cells. To clarify this problem, we enriched for OPC-like tumor cells by using 
an immunopanning method widely used to purify WT OPCs based on their surface 
expression of PDGFRα (Cahoy et al., 2008). The transcriptome of purified OPC-like 
tumor cells displayed an even higher extent of similarity to that of normal OPCs (Fig. 
17M), demonstrating that these cells contributed to the OPC signatures of glioma 
samples. qRT-PCR results showed that, in addition to consistent expression of OPC 
markers, purified OPC-like cells from some tumors expressed Nestin at a comparable 
level to NSC-derived neurospheres (data not shown). This finding was further confirmed 
74
by marker costaining of the tumor mass, showing that some GFP+ dividing tumor cells 
coexpressed both PDGFRα and Nestin (data not shown). Importantly, regardless of their 
Nestin expression levels, these panned tumor cells, when orthotopically allografted into 
NOD-SCID mouse brains, effectively initiated secondary tumors (data not shown) that 
recapitulated both histological (data not shown) and molecular features (Figs. 17F and 
17M) of their primary tumors. Taken together, these data demonstrate that highly 
proliferating tumor cells manifested salient OPC features. Although Nestin was expressed 
in some tumor cells, the fact that primary tumor cells with either undetectable or high 
levels of Nestin expression could efficiently initiate new tumors implies that Nestin 
expression appears irrelevant to tumorigenicity and is likely acquired during 
transformation. Therefore, analyses of tumor cells further support OPCs, rather than 
NSCs, neurons, or astrocytes, as the cell of origin for glioma in p53, NF1 mutation-driven 
glioma model.
Localization of earliest neoplastic lesions suggests a gray matter origin of glioma
The analyses of lesion locations at early stages of transformation could provide 
great insights for understanding the cancer cell of origin. We first analyzed morphological 
or cyto-architectural changes of mutant cells in fully developed tumors, reasoning that 
some features could serve as landmarks of transformation, thereby helping to determine 
the locations of early lesions. In all malignant tumors, even in the smallest lesions 
identifiable by classic pathology, MADM staining revealed a prominent cyto-
architectural feature in which multiple glioma cells wrapped around a single neuronal cell 
body (Fig. 18C), a salient pathological feature known as perineuronal satellitosis in 
human glioma (Fig. 18D). It should be noted that normal OPCs are often juxtaposed to 
75
neuron cell bodies (data not shown), implicating that perineuronal satellitosis in tumors 
might stem from this unique cyto-architecture of OPCs.
When we compared OPC::neuron association from pretransforming to tumor-
bearing brains, we found that the number of mutant OPCs for each accompanying neuron 
changes from mostly one (occasionally two) prior to transformation (Figs. 18A and 18E ) 
to frequently three or higher in any Ki67-positive lesions (Figs. 18B, 18C, and 18E). 
Therefore, we used “three or more mutant OPCs per neuron” as a landmark to examine a 
cohort of brains for initiating lesions that cover a period from right after the dormant 
stage to around the time when tumors can readily be identified (Figs. 18F–18K). In 2.5- 
to 3.5-month old mice, we observed the earliest detectable lesions in cortical gray matters 
(Fig. 18F, P75–P90 and top of P105 column). In further developed lesions, tumor cells 
were more frequently found to overlap with white matter tracks and the SVZ (Fig. 18F, 
bottom of P105, P120, and > P120 columns). This observation indicates that previous 
findings of glioma in white matter tracks and the SVZ (Persson et al., 2010; Zhu et al., 
2005) might be the path or endpoint, rather than the starting point, of tumor cell 
migration and further argues against NSCs and for OPCs as the cell of origin.
OPCs can be directly transformed into malignant glioma by p53/NF1 mutations
Our results thus far point toward the following scenario for glioma development 
in our model: NSCs carrying p53/NF1 mutations give rise to mutant OPCs, which 
multiply, progress, and eventually transform into malignancies. However, it remains 
unclear whether mutations must occur in NSCs or whether OPCs can be directly 
transformed by the same set of mutations. To address this question, we next used NG2-
Cre that has been reported to specifically express in OPCs, but not NSCs (Komitova et 
76
al., 2009; Zhu et al., 2008), to perform MADM analysis. We first verified that NG2-Cre 
exclusively labels OPCs and oligodendrocytes, but not mature astrocytes or neurons in 
brain parenchyma of WT-MADM mice (Figs. 19A–19D). Next, we examined the entire 
SVZ of WT-MADM mice and never found any GFP- or RFP-labeled cells coexpressing 
GFAP (data not shown). Lastly, we thoroughly examined the olfactory bulbs (OB) from 
four WT-MADM mice and did not find any NeuN+ MADM-labeled green or red cells 
(data not shown), further supporting the absence of NG2-Cre expression in NSCs.
In stark contrast to the extremely sparse labeling in WT-MADM mouse brains 
induced by NG2-Cre (Fig. 19E), the green mutant OPCs in mutant-MADM mice 
populated the entire brain (Fig. 19F) and reached an average G/R ratio of more than 300 
(Fig. 19G) at P60. Consistent with findings from the NSC-Cre-based MADM model, the 
differentiation process of mutant OPCs was hampered (Figs. 19G and 19H). These data 
demonstrate that introduction of NF1 and p53 gene mutations directly into OPCs is 
sufficient to drive overproliferation and to impair OPC differentiation.
After 8 months of age, we found GFP+ tumors in brains of NG2-Cre-induced 
mutant-MADM mice at nearly full penetrance (Fig. 19I, Fig. 19J, and data not shown). 
These tumors showed marker staining and transcriptome profiles indistinguishable from 
NSC-Cre-induced malignant gliomas (Figs. 17F, Fig. 17M, and data not shown). 
Dissociated cells from these tumors could effectively initiate secondary tumors in 
allografting assays (data not shown), demonstrating the malignancy of these OPC-
initiated tumors.
Despite their indistinguishable marker expression patterns, transcriptome profiles, 
and tumor-initiating ability, we observed that tumors induced by NSC-Cre tended to 
77
reside in the brain parenchyma, whereas those induced by NG2-Cre were more frequently 
found around the ventral hypothalamic region and often invaded into the subarachnoid 
space (Fig. 19I, Fig. 20B, and data not shown). However, tumors induced by the two Cre 
lines at equivalent locations showed indistinguishable pathological features (Figs. 20A 
and 20B). Moreover, although tumor cells with subarachnoid invasion were 
morphologically distinct from those in the brain parenchyma, transplanting them into the 
brain parenchyma of NOD-SCID mice resulted in secondary tumors that were 
morphologically indistinguishable from those primary tumors in the brain parenchyma 
(Fig. 20C). These data strongly suggest that the variation in tumor pathology is merely an 
effect of tumor location. The differences in latency and spatial distribution between NG2-
Cre- and NSC-Cre-induced tumors most likely are attributed to technical reasons, such as 
the much lower expression level of NG2-Cre transgene or possible spatial biases of these 
Cre lines. It is also possible that these two models have distinct noncell-autonomous cues 
for tumor formation. For instance, NSC-Cre generates heterozygous astrocytes or 
neurons, whereas NG2-Cre does not, which might function as special niches to facilitate 
tumor formation and to bias the tumor location. Nevertheless, all lines of evidences, 
including marker expression patterns (Figs. 17F–17L and data not shown), transcriptome 
profiles (Fig. 17M), tumorigenic capacity, and pathology (Fig. 20 and data not shown), 
indicate that tumors from both models are intrinsically identical. Therefore, our data 
demonstrate that OPCs can be directly transformed into malignant glioma as a result of 
concurrent p53/NF1 mutations.
Finally, we investigated the human relevance of our studies by comparing the 
transcriptome profile of tumor cells in our model with the molecular signatures of human 
78
glioma samples. Among four subtypes of human malignant glioma revealed by recent 
studies from the Cancer Genome Atlas (TCGA) (Verhaak et al., 2010), both NSC-Cre- 
and NG2-Cre-initiated glioma models match well with the proneural subtype (Fig. 17M), 
which has the poorest responses to chemo- and radiotherapies (Verhaak et al., 2010). 
Therapeutic strategies designed according to the intrinsic OPC nature should have 
promising effectiveness for proneural or even other subtypes of glioma.
Discussion
In this study, we used the mouse genetic mosaic system termed MADM to track 
the entire tumorigenic process, with the aim of identifying the cell of origin for glioma 
when initial concurrent mutations of p53 and NF1 occur in embryonic NSCs. Taking 
advantage of the in vivo cellular resolution and WT internal control cells afforded by 
MADM, we analyzed lineage-specific cellular aberrations at premalignant stages and 
observed dramatic overexpansion and elevated proliferative activity in the OPC lineage 
but minimal abnormalities in NSCs and all other progeny lineages. Furthermore, marker 
staining, cyto-architecture, and transcriptome analyses all point to OPCs as the 
transforming cell type for malignant glioma. Finally, direct introduction of p53/NF1 
mutations into OPCs led to the formation of glioma that were indistinguishable from 
those initiated from NSCs, based on both transcriptome and pathological analyses. 
Therefore, our studies clarified that, for the glioma model induced by p53 and NF1, 
though NSCs could serve as the cell of mutation, OPCs serve as the actual cell of origin.
MADM as a tool to distinguish cancer cell of origin and cell of mutation
The identification of the cancer cell of origin is crucial but often controversial, 
79
even with the use of genetic modified mouse models (Visvader, 2011). Conceptually, 
caution must be taken that cells acquiring initial mutations (cell of mutation) may not 
directly transform into malignancy because their transforming potential could be 
manifested by their progeny cell types. For example, when Patched mutation was 
introduced into NSCs in a mouse model of medulloblastoma, malignant transformation 
only occurred after NSCs gave rise to lineage-restricted granular neuron precursors 
(GNPs) (Schüller et al., 2008; Yang et al., 2008). In this scenario, the transforming 
potential of NSCs (cell of mutation) is only manifested by mutant progeny GNPs, the cell 
of origin that possesses the conducive signaling context. Therefore, to unambiguously 
identify cancer cell of origin, one has to genetically dissect the transforming potentials of 
all progeny derived from the initial mutated cell type. This could be feasible if highly 
specific Cre-transgenic lines are available, such as in the case of medulloblastoma studies 
(Schüller et al., 2008; Yang et al., 2008). However, such a strategy cannot be generalized 
to other cancer models due to the lack of specific Cre-transgenic mice in every lineage 
for most tissues. An alternative approach is to analyze aberrant growth in all cell lineages 
derived from initial mutant cells during pretransforming stages (Visvader, 2011), which is 
now possible with 
MADM-based cancer models
Our study extends rather than contradicts previous reports. Limited by the cellular 
resolution at pretransforming stages, previous studies demonstrated the transforming 
potential of NSCs but did not clearly distinguish cell of mutation from cell of origin 
(Alcantara Llaguno et al., 2009; Wang et al., 2009; Zhu et al., 2005). MADM, on the 
other hand, offers a robust analytical paradigm for the identification of the cell of origin 
80
with both permanently GFP-labeled mutant cells and RFP-labeled WT internal control 
cells immediately after the initial mutations occurred. In principle, MADM could be used 
for all types of cancer, as long as oncogenic mutations for a particular type of cancer, a 
tissue-specific stem cell Cre-transgenic line, and cellular markers of each lineage within 
that tissue are available.
Our studies emphasize the importance of intersection between genetic mutations 
and the signaling context within the cell of origin. We show here that OPCs are 
particularly sensitive to p53/NF1 mutations, whereas NSCs and other brain cell types are 
much less responsive. Interestingly, it has been reported that the NF1 mutation alone can 
promote OPCs to proliferate in mice and zebrafish (Bennett et al., 2003; Lee et al., 2010), 
suggesting that NF1 plays a pivotal role in regulating OPC proliferation with conserved 
mechanisms across species. Such intersections should be further exploited to help 
understanding of the molecular mechanism of OPC transformation, which should provide 
critical insights for developing effective therapeutic strategies. Notably, whereas OPCs 
appear to be particularly responsive to NF1 mutation, other cell types, including NSCs, 
could be specifically vulnerable to other genetic lesions. It would be very interesting to 
investigate whether different mutations could transform distinct cell of origins, which 
might account for the great heterogeneity of human glioma.
OPC is an important glioma cell of origin with underappreciated proliferative and 
plastic potentials
One of the notions that support the prevalent view of NSCs as the glioma cell of 
origin is the persistent proliferative activity of NSCs in the brain during the entire life, 
which makes them susceptible to oncogenic mutations. Additionally, NSCs share many 
81
cellular properties with glioma cancer stem cells, such as the capability to self-renew, the 
potential to differentiate into multiple cell lineages, and the expression of some common 
cellular markers such as Nestin and Sox2. However, a large body of work has revealed 
that many of these features are also shared by OPCs. First, OPCs in fact represent the 
largest proliferative pool in the brain parenchyma for both rodents and human (Dawson et 
al., 2003; Geha et al., 2010). Second, OPCs at early developmental stages have been 
found to express many commonly used stem cell markers, including Nestin and Sox2. 
Third, OPCs isolated from rat optic nerve can be readily reprogrammed into a multipotent 
NSC-like status under in vitro conditions, which become self-renewable and can 
differentiate into astrocytes and neurons in addition to oligodendrocytes (Kondo and 
Raff, 2000). Intriguingly, OPCs from human subcortical white matter intrinsically behave 
like NSCs without reprogramming (Nunes et al., 2003), raising the possibility that 
primate OPCs are more plastic than their rodent counterparts. Taken together, the intrinsic 
nature of OPCs renders them great susceptibility to oncogenic mutations, which could be 
harnessed to devise effective therapeutic strategies.
Previously, it has been reported that the injection of PDGF-BB-expressing virus 
into the corpus callosum, the expression of PDGF-BB under CNPase promoter, and the 
overexpression of v-erbB under human S100β promoter all led to glioma formation with 
OPC features (Assanah et al., 2006; Lindberg et al., 2009; Persson et al., 2010). Although 
providing circumstantial evidence, these studies have faced challenges to unequivocally 
pinpoint OPCs as the cell of origin for malignant glioma (Visvader, 2011) due to noncell-
autonomous effects caused by the secretion of PDGF-BB (Assanah et al., 2006; Lindberg 
et al., 2009) and the use of non-OPC-specific promoter to drive the initial tumorigenic 
82
event (Persson et al., 2010). By circumventing all of these caveats, our studies now firmly 
establish that OPCs serve as the cell of origin for malignant glioma with relevant genetic 
mutations. For clinical applications, intrinsic OPC properties should be fully exploited to 
design effective therapies for glioma, especially for the proneural subtype. For example, 
understanding the unique proliferative capacity of OPCs could help devise treatments to 
stall tumor progression; deciphering the migration mechanism of OPCs could enhance 
the effectiveness of surgery by preventing the infiltration of tumor cells into the entire 
brain; and probing into the differentiation process of OPCs could facilitate the design of 
differentiation therapy strategies.
The application of MADM beyond cancer cell of origin studies
Compared to conventional models, the MADM-based cancer model bears 
significant advantages for exploring critical cancer biology problems beyond the cell of 
origin. Its reliable labeling of tumor cells at the single-cell resolution should help to 
distinguish bona fide tumor cells from bystander cells in a complex tumor mass, which is 
highly valuable for dissecting tumor architecture, tracking metastasized tumor cells, and 
studying tumor-niche interactions. Importantly, the availability of a built-in internal 
control enables quantitative comparison between mutant and WT sibling cells. Such 
rigorous phenotypic analyses within the same animal largely remove variations often 
introduced by comparing phenotypes between individual animals, thereby helping to 
identify subtle but important phenotypes that could be easily missed by using 
conventional approaches.
Using Drosophila genetic mosaic system, combined with the application of 
positive marking with gene manipulation at the mosaic level, helped to elucidate many 
83
fundamental questions in cancer biology, such as identifying important oncogenes and 
TSGs by forward genetic screening (Potter et al., 2001; Tapon et al., 2001) and 
deciphering the complicated tumorigenic processes such as metastasis (Pagliarini and Xu, 
2003). Along this same principle, we expect that a further modified MADM system could 
address even more sophisticated problems in cancer biology beyond the capability of the 
current one. For instance, by replacing fluorescent reporters with a transcription factor 
such as tTA or rtTA to specifically manipulate gene expression in mosaic mutant cells, 
one could model “second hit” events or therapeutic strategies and gain deep insights into 
the molecular network for cancer formation. By incorporating novel magnetic resonance 
imaging reporters such as H-ferritin (Genove et al., 2005) into MADM cassettes, one 
could noninvasively trace tumor initiation, progression and metastasis in real time. Also, 
by combining with transposon-based mutagenesis such as Sleeping Beauty (Collier et al., 
2005) or PiggyBac (Ding et al., 2005), the MADM system should be able to perform 
forward genetic screening for novel recessive TSGs in mouse somatic cells without 
involving a large number of animals.
Concluding remarks
In summary, our study demonstrates the importance of analyzing the entire 
process of tumor development for identifying cancer cell of origin. MADM-based tumor 
modeling can be applied in principle to any other tumor type and should help to resolve 
many important problems in cancer biology. Broadly, the ability to perform sporadic 
single-cell genetic manipulations in unambiguously labeled cells should make MADM an 
invaluable analytical tool for fields such as developmental biology and neuroscience that 
rely heavily on in vivo analyses.
84
CHAPTER V
TU-TAGGING: CELL TYPE-SPECIFIC RNA ISOLATION FROM INTACT 
COMPLEX TISSUES
This work was published in the journal Nature Methods in June 2009. Myself and 
Kristin Robinson performed the microarray-based and immunohistochemistry expression 
analyses. Kristin Robinson and Michael Cleary performed the northern-blot analysis. 
Chris Doe was the principle investigators for this work. 
Introduction
Cell type–specific gene expression is a defining feature of multicellular organisms 
(Tomancak et al., 2007). The analysis of cell type–specific transcriptomes can provide 
insight into the mechanisms used to generate cellular diversity (Wang et al., 2009) as well 
as help determine the underlying cause of disease (Schadt et al., 2003). Although a few 
methods are available for cell type–specific RNA isolation (Roy et al., 2002; Tanke and 
van der Keur, 1993; Doyle et al., 2008; Heiman et al., 2008) each has constraints, and 
researchers are often limited by their ability to isolate RNA from cell types of interest 
(Nelson et al., 2006). Thus, developing new methods for cell type–specific RNA isolation 
is an important goal for genomic analysis of development and disease.
We previously had shown that the Toxoplasma gondii nucleotide salvage enzyme 
uracil phosphoribosyltransferase (UPRT) can be used to biosynthetically label newly 
synthesized RNA in vivo (Cleary et al., 2005). Under natural conditions, UPRT couples 
ribose-5-phosphate to the N1 nitrogen of uracil to yield uridine monophosphate (UMP), 
which is subsequently incorporated into RNA. When the modified uracil analog 4-
thiouracil is provided to UPRT as a substrate, the resultant product is also incorporated 
85
into RNA, and this incorporation has little effect of cellular physiology  (Cleary et al., 
2005). Thio-substituted nucleotides are not a natural component of nucleic acids, and the 
resulting thio-labeled RNA can be readily tagged and purified using commercially 
available reagents. Owing to the ability to use this method to separate newly synthesized 
RNA from bulk cellular RNA, we and others have used it to measure RNA synthesis and 
decay rates (Cleary et al., 2005; Dolken et al., 2008).
Here we describe a different use for 4-thiouracil UPRT-based biosynthetic 
labeling that we call 'TU-tagging'. We reasoned that by spatially restricting UPRT 
expression in a multicellular organism, 4-thiouracil will be modified and subsequently 
incorporated into newly synthesized RNA only in cells expressing UPRT. Thus, even if 
RNA is isolated from the whole organism, RNA from the cells expressing UPRT can be 
recovered by purifying labeled RNA (Fig. 21A). This method would be particularly 
useful for isolating RNA from cell types that are difficult to isolate by dissection or 
dissociation methods, such as subsets of neurons or glia in the central nervous system.
Methods
UPRT and 4-thiouracil toxicity tests
We crossed UAS-UPRT lines to various GAL4 lines to express UPRT in several 
cell types (see below) and ubiquitously (using tubulin-GAL4). In no case did we observe 
lethality, developmental defects or delay in the time to pupariation and eclosion (data not 
shown). We next monitored development of larvae ubiquitously expressing UPRT when 
grown on food containing 4-thiouracil (4TU). We observed no effect on central nervous 
system development (developmental timing or neuroanatomy) in larvae grown on 4TU 
86
for 48 h or less; we then used an 8 h or less feeding interval for all experiments. Longer 
exposure to 4TU (with or without UPRT expression) produced slight developmental 
delays but did not affect viability.
Fly stocks
We used standard methods to clone the T. gondii UPRT coding sequence into 
pUAST to generate the UAS-UPRT and UAS-HA-UPRT plasmids, and obtained 
independent viable insertions on the X, II or III chromosome. Three UAS-UPRT lines 
consistently had low background expression and high GAL4-induced expression: UAS-
HA:UPRT2.1 (chromosome II), UAS-UPRT3.1 (chromosome III) and UAS-HA:UPRT3.2 
(chromosome III). We crossed one of these three lines to the following previously 
described GAL4 lines for all experiments (chromosome in parentheses) (i) worniu-
GAL4(II) × UAS-HA:UPRT3.2; (ii) OK107-GAL4(IV) × UAS-HA:UPRT3.2; (iii) c179-
GAL4(II) × UAS-HA:UPRT3.2; (iv) elav-GAL4(III) × UAS-HA:UPRT3.2; (v) tubulin-
GAL4(III) × UAS-HA-UPRT3.2; (vi) prospero-GAL4(III) × UAS-UPRT3.1; (vii) twist-
GAL4(II);how24B-GAL4(III) × UAS-UPRT3.1; and (viii) repo-GAL4(III) × UAS-
HA:UPRT3.2.
4TU treatment and RNA extraction
To treat embryos with 4TU, the embryos were dechorionated in bleach, washed, 
rinsed with isopropanol, blotted dry and submerged in octane for 3 min (all in the basket). 
Embryos were blotted and air dried for ~3 min until soft to touch. Embryos were 
transferred to Schneider's media containing 1.0 mM 4TU for 2 h at 25 °C or 30 °C, 
blotted dry, moved using a paintbrush to an Eppendorf tube containing 1 × PBS with 1% 
tween (PT) and centrifuged at 6,000g for 30 s. PT was removed and the embryos were 
87
homogenized in Trizol and stored at −80 °C until RNA purification. To treat larvae with 
4TU, larvae of the desired stage were placed on mocha food caps (20 ml H20, 0.4 g 
sucrose, 0.18 g agar, 1 g dextrose and 0.5 g brewers yeast) containing 0.5 mM 4TU for 
the indicated time at 30 °C, homogenized in Trizol and stored at −80 °C until RNA 
purification. To treat adult flies with 4TU, they were starved for ~16 h, placed on mocha 
food containing 1.0 mM 4TU for 6–8 h at 25 °C, homogenized in Trizol and stored at −80 
°C until RNA purification. A more detailed protocol is available on request.
Total RNA was extracted from Trizol using standard methods with the following 
additional steps: an initial centrifugation at 12,000g for 10 min at 4 °C to remove 
insoluble material followed by a 5-min incubation at room temperature (20–25 °C) to 
ensure complete dissociation of RNA-bound proteins. Only RNA samples with 
absorbance 260/280 ratios of ≥ 2.0 were used for subsequent biotinylation and 
purification steps. In all cases, RNA samples were resuspended at a final concentration of 
≥0.4 μg μl−1.
Purification of TU-tagged RNA
Detailed methods for biotinylating and purifying thio-tagged RNAs have been 
published (Cleary et al., 2005; Zeiner et al., 2008), and detailed protocols including the 
most recent improvements are available upon request. Relevant changes to previous 
protocols are summarized here. Biotinylation of RNA was performed using EZ-Link 
biotin-HPDP, as previously described. Biotinylation reactions contained 10 mM Tris (pH 
7.4), 1 mM EDTA and 2 μl of a 1 mg ml−1 biotin-HPDP solution (in 
dimethylformamide) per 2 μg of RNA. The reaction volume was adjusted with water so 
that the concentration of biotin-HPDP was equal to 30% of the final reaction volume. 
88
Biotinylation reactions were incubated in the dark for 3 h at 25 °C before RNA 
precipitation. Biotinylated RNA was detected by blotting and probing with streptavidin-
horseradish peroxidase as previously described. Purification of biotinylated TU-tagged 
RNA was performed as previously described with the following modifications: 2 μl of 
MPG streptavidin beads were used per microgram of input RNA. The input RNA was 
always at a concentration of 0.5 μg μl−1. After blocking with yeast tRNA and washing, 
beads were resuspended in the input RNA sample plus a volume of MPG buffer (1 M 
NaCl, 10 mM EDTA and 100 mM Tris (pH 7.4) in RNase-free H2O) equal to one-third 
the input RNA volume. Beads plus RNA were incubated at room temperature for 20 min 
before collecting the non-bound sample and washing with MPG buffer (one 5-min wash 
at room temperature, two 1-min washes at room temperature, one 1-min wash in 65 °C 
MPG buffer and one final 1-min wash at room temperature). After the removal of as 
much MPG buffer as possible, TU-tagged RNA was eluted by incubating the beads for 10 
min in freshly prepared 5% 2-mercaptoethanol. RNA was precipitated using isopropanol 
and linear acrylamide. After resuspending RNA in water, samples were placed in the 
magnetic stand again to remove any remaining MPG beads. For the purification 
procedures, input amounts of biotinylated RNA were 14–20 μg.
Microarray analysis
We used the Agilent eArray platform to design a custom D. melanogaster 
oligonucleotide microarray representing 14,141 unique genes from the Flybase release 
5.4 genome. Fifty nanograms of TU-tagged and 200 ng untagged RNA were fluorescently 
labeled and hybridized directly against each other to Agilent microarrays. The microarray 
experiments were performed according to Agilent's protocol (Version 5.5, February 2007) 
89
and scanned using an Axon GenePix 4000B scanner. Fluorescent ratios for each 
microarray element were recovered and normalized using GenePix Pro 6.0.
Normalizing for transcript uracil number
We downloaded the dmel-all-transcript-r5.4.fasta file from Flybase, which 
contained the sequences of every predicted transcript from genome release 5.4. We wrote 
a Perl script to count the number of uracils in a transcript (transcript uracil number) for 
each transcript. In cases where there was alternative splicing, we averaged the uracil 
number over the multiple isoforms. This was necessary because, for the most part, our 
microarray did not distinguish between different isoforms of the same gene. To normalize 
the data, these uracil counts were plotted against the observed microarray ratios using 
OpenOffice.org Spreadsheet and the regression equation was determined. This equation 
was used to calculate the expected ratio for each transcript based on the transcript's uracil 
number. For each transcript, the normalized ratio was calculated by subtracting the 
expected ratio from the observed ratio. After plotting the initial normalized ratios, we 
noticed a group of transcripts, all with low uracil numbers, that had very high normalized 
ratios. Upon investigation we noticed that these transcripts were missing annotations for 
either or both untranslated regions. Thus, our transcript uracil counts for these transcripts 
were lower than their actual uracil number, leading to incorrect normalization. Therefore, 
transcripts with missing untranslated region annotations were excluded from 
normalization and subsequent analysis. We note that normalization errors owing to 
transcript misannotation will decrease as transcriptome annotations improve.
90
Results
To test the ability to biosynthetically label RNA in Drosophila, we delivered 4-
thiouracil and monitored its incorporation into RNA by purifying total RNA, performing 
thio-biotin coupling and using streptavidin–horseradish peroxidase to detect labeled RNA 
(Methods). Wild-type larvae or adult flies, or larvae or adult flies containing only the 
GAL-4–inducible transgene UAS-UPRT (upstream activating sequence–UPRT) but no 
source of GAL4, had very small amounts of labeled RNA when fed 4-thiouracil (Fig. 
21B). In contrast, 4-thiouracil–fed larvae or adult flies containing both GAL4 and UAS-
UPRT transgenes expressed UPRT in the cell types containing GAL4 (data not shown) 
and showed robust RNA labeling (Fig. 21B). Similarly, embryos soaked in 4-thiouracil–
containing medium had robust RNA labeling only when both UAS-UPRT and GAL4 were 
present (Fig. 21C). We conclude that the combination of UPRT and 4-thiouracil can be 
used to biosynthetically label RNA in Drosophila embryos, larvae and adults.
To determine the limits of sensitivity, we fed larvae 4-thiouracil and expressed 
UPRT in different subsets of the larval brain. When UPRT was expressed in about 2,000 
neurons in the entire larva, we detected small amounts of labeled RNA (Fig. 21B). When 
we reduced the number of UPRT-expressing cells to about 250 neural progenitors per 
larva, we observed no detectable signal over background (Fig. 21D). However, a simple 
dissection of the intact larval brain before RNA purification yielded excellent signal and 
dramatically reduced background RNA labeling (Fig. 21D). We conclude that the amount 
of TU-tagged RNA correlated well with the number of cells expressing UPRT; that 
simple tissue isolation can reduce background and increase sensitivity; and that non–
central nervous system tissue contributed to nearly all of the low-level background RNA 
91
labeling in larvae.
To confirm that thio-labeled RNA was from UPRT-expressing cells and 
demonstrate the utility of TU-tagging for cell type–specific RNA isolation, we purified 
TU-tagged and untagged RNA and compared them using microarrays. We isolated RNA 
from larval glia, which are present in low numbers, are highly dispersed and have a 
complex cell morphology (Fig. 22A), making them one of the most difficult cell types to 
isolate by dissection or dissociation methods. We used reversed polarity (repo)-GAL4 to 
drive expression of UAS-UPRT specifically in glial cells of the larval brain (Fig. 22A). 
We purified TU-tagged and untagged RNA from 72–96-h larval brains and hybridized 
them to custom Agilent microarrays (Methods). We detected signal for 7,354 of the 
14,141 genes present on the microarray. If TU-tagged RNA was primarily from UPRT-
expressing cells, known larval glia–specific genes should be among the most enriched in 
the TU-tagged relative to untagged RNA. There are four genes known to be expressed 
specifically in larval glia but not neurons or trachea of the brain. All four were enriched 
greater than twofold in the TU-tagged RNA (Fig. 22B) and three of the four were among 
the 5% most enriched genes ana (3.54-fold, top 0.3%, 19/7,354), repo (2.85-fold, top 
1.6%, 115/7,354) and moody (2.75-fold, top 1.9%, 141/7,354), which are expressed in all 
larval glia (Bainton et al., 2005; Ebens et al., 1993), and ced-6 (2.13-fold, top 7.6%, 
561/7,354), which is expressed in a subset of larval glia (Awasaki et al., 2006). Thus, 
known larval glia–specific genes were enriched in the TU-tagged RNA. We conclude that 
TU-tagging can effectively isolate glia-specific RNA from whole brain tissue without 
prior cell dissociation.
We next tested whether the number of uracils in a transcript influenced the extent 
92
of enrichment we observed, because long transcripts containing many uracils are 
expected to be labeled at a higher frequency than short transcripts with few uracils. Thus 
we plotted the number of uracils in a transcript against the observed microarray ratio for 
each transcript (Methods). There was a notable positive correlation (Fig. 22C). 
Transcripts with many uracils were often among the most enriched transcripts overall, 
even when they were not enriched relative to other transcripts with a similar number of 
uracils (Fig. 22C), whereas transcripts with few uracils that were clearly enriched relative 
to other transcripts with a similar number of uracils were unlikely to be among the most 
enriched transcripts overall (Fig. 22C). We conclude that the number of uracils in a 
transcript is a source of bias in TU-tagging experiments.
To remove this bias on the microarray results, we used the regression equation to 
calculate the expected TU-tagged to untagged ratio and subtracted it from the observed 
ratio for each transcript (Methods). This normalization procedure successfully removed 
the bias in data that resulted from the number of uracils in a transcript (Fig. 22D) and had 
a large impact on which genes were present in the top 1% or 5% most enriched genes 
(Fig. 22E). We conclude that enrichment bias owing to the number of uracils in a 
transcript can be removed using a simple normalization procedure and that normalization 
has a large impact on which genes are considered most enriched.
We next determined how the normalization affected the enrichment of the known 
larval glia–specific genes. Without normalization three of four previously known larval 
glia–specific genes were within the top 5% of enriched genes (Fig. 22C). After 
normalization all four were within the top 3.2% of enriched genes: ana, top 0.2% 
(12/6,167); repo, top 1.6% (99/6,167); moody, top 2.4% (150/6,167); and ced-6, top 3.2% 
93
(235/6,167) (Fig. 22D). Although the characterization of new glia genes was beyond the 
scope of this work, these results suggest that the other highly enriched genes in this 
dataset are excellent candidates for regulating aspects of larval glia biology. We conclude 
that normalizing for transcript uracil number improves TU-tagging data analysis and that 
normalization should be useful for other 4-thiouracil/UPRT–based methods.
Discussion
An important property of the TU-tagging method is that only newly synthesized 
RNAs are labeled. Thus, the percentage of labeled cellular RNA will depend on the 
duration of labeling. Although long labeling periods should work well for isolating the 
majority of RNA present in a particular cell type, short labeling periods could be used to 
detect changes in gene expression at successive time points in specific cell types, because 
newly synthesized RNA could be separated from bulk cellular RNA. This would be 
useful for studying rapid changes in gene expression after a particular developmental or 
physiological event.
The TU-tagging method is likely to work well in other systems including 
vertebrates. There is very little UPRT-independent 'background' incorporation of 4-
thiouracil into RNA in mouse or human cell culture lines (Cleary et al., 2005; Rinn et al., 
2008). Spatial control of UPRT in vertebrates could be achieved by using transgenes, as 
in Drosophila, or by retroviral delivery, electroporation or mRNA injection. Thus, at least 
Drosophila, mice and humans appear suitable for TU-tagging experiments. It is likely 
that TU-tagging can be used for cell type–specific RNA isolation in many multicellular 
organisms and should be particularly useful for the study of development, neurobiology 
94
and disease.
95
APPENDIX A
TABLES
Table 1. Candidate genes for down-stream targets of sox9a and sox9b in zebrafish.
Chr, Chromosome location; PA, Pharyngeal arch; Ref, References; PF, Pectoral fin; M, 
Muscle; N.A., not available. 1, (Liu et al., 2001); 2, (Shen and Raymond, 2004); 3, 
(Passini et al., 1997); 4, (Passini et al., 1998); 5, (Korzh et al., 1998); 6, (Mueller and 
Wullimann, 2002); 7, (Ochocinska and Hitchcock, 2007); 8, (Mavropoulos et al., 2005); 
9, (Rauch et al., 2003); 10, (Thisse and Thisse, 2004); 11, (Oates et al., 2001); 12, (Thisse 
et al., 2001); 13, (Yan et al., 1995); 14, (Yan et al., 2005).
Table 2. Mutants affecting brain neuroblast numbers used in this study.
96
Table 3. Representation of cell type-specific genes within microarray groups A, B, and C.
97
APPENDIX B
FIGURES
Fig. 1. Differential regulation of col11a2 expression by sox9a and sox9b. (A–D) Lateral 
view and (E–H) ventral view of 2 dpf embryos. Expression of col11a2 appeared in 
neurocranium (ethmoid plate, arrowhead), pharyngeal arches (mandibular and ceratohyal, 
arrows), otic vesicle and in the pectoral fin bud (A, E). Expression in neurocranium and 
pharyngeal arches was lost in the sox9a mutant (B, F). In the sox9b mutant, expression 
disappeared from the pharyngeal cartilages, but expression in the neurocranium 
(arrowhead) remained (C, G). Expression almost disappeared in sox9a;sox9b double 
mutant embryos (D, H). At 2dpf, sox9a was expressed in the neurocranium (arrowhead) 
and pharyngeal arches (arrows) in locations corresponding to col11a2 expression (I), but 
sox9b was only expressed in mandibular and ceratohyal (arrows), not in the ethmoid plate 
and trabeculae (J). These results confirmed microarray results showing that col11a2 is 
down-regulated in sox9 mutant embryos, and revealed that sox9a and sox9b regulate 
col11a2 expression in a tissue specific manner according to sox9 expression patterns. ch, 
ceratohyal; ep, ethmoid plate; ma, mandibular; ov, otic vesicle; op, opercle; pf, pectoral 
fin bud.
98
99
Fig. 2. Expression of candidate sox9 down-stream targets analyzed by in situ 
hybridization analysis in 2 dpf embryos, the same stage as used for microarray analysis. 
Anterior to the left. Lateral views of crx (A–C), rs1 (D–F) and vsx1 (G–I) show that they 
are strongly expressed in developing retina; crx and rs1 but not vsx1 were expressed in 
the outer nuclear layer. Expression was severely reduced in sox9b mutants and double 
mutants; only a small ventral-nasal patch of retinal expression (black arrowhead) was 
detected for crx (B, C) and rs1 (E, F), and ventro-nasal expression in the retina (white 
arrowhead) was observed for vsx1 (H, I). Ventral view of neurod (J–L) and sox4a (M–O) 
showing neurod expression in the outer nuclear layer and inner nuclear layer of the retina 
and telencephalon (J) and sox4a expression in inner part of retina, telencephalon and 
diencephalon (M). Retinal expression of neurod was reduced in sox9b and double mutant 
embryos, but the expression in the telencephalon was not affected in these mutants, 
demonstrating specificity (K, L). Expression of sox4a in the retina was lost in sox9b 
mutants and double mutants, whereas expression in telencephalon and diencephalon was 
not affected in the mutants (N, O). At 2 dpf, sox9b (S, T), but not sox9a (P, Q), was 
expressed in retina, consistent with the reduced gene expression pattern shown above. 
Lateral view (P, S) and ventral view (Q, T). At 68 hpf, sox9a is expressed in the inner 
nuclear layer (R), and sox9b is expressed in ciliary marginal zone (U). These results 
confirmed the microarray result that crx, rs1, vsx1, neurod and sox4a are down-regulated 
in sox9b mutant embryos, and revealed that sox9b is required for expression of these 
candidate genes in retina at 2 dpf. cmz, ciliary marginal zone; d, diencephalon; onl, outer 
nuclear layer; pa, pharyngeal arches; r, retina; t, telencephalon; tc, tectum.
100
Fig. 3. Candidate target genes calb2a and calb2b are co-orthologs of tetrapod Calb2 and 
their expression was lost in the sox9b mutant retina. (A) Phylogenetic analysis of 
Calbindin family proteins using the neighbor-joining method. Numbers on the branches 
are bootstrap values for the group in a thousand runs. Zebrafish Calb2a and Calb2b 
branched together as co-orthologs of tetrapod Calb2 with very high bootstrap support. 
Ascidian Calbindin-related protein was used as an outgroup. Cin: ascidian, Ciona 
intestinalis; Dre: zebrafish, Danio rerio; Hsa: human, Homo sapiens; Mmu: mouse, Mus 
musculus. (B) A conserved synteny around Human CALB2 on Homo sapiens 
chromosome 16 (Hsa16) with zebrafish calb2a on Danio rerio chromosome 7 (Dre7) and 
calb2b on Dre18. Red dots represent zebrafish genes orthologous to human genes in the 
position 60–80 Mb of Hsa16. Dorsal view of calb2a (C–E) and calb2b (F–H) expression. 
Expression of calb2a appeared in the retina, cranial ganglia and hindbrain neurons (C). 
Expression in the retina was not observed in sox9b mutants (D) and double mutants (E), 
but other expression domains were not affected, demonstrating specificity (D, E). The 
expression of calb2b was observed in the olfactory placode, tegmentum and ganglion cell 
layer (F). Expression in the retina was not detected in sox9b mutants (G) and double 
mutants (H), but expression in the other tissues was not affected (G, H). These results 
showed that zebrafish calb2a and calb2b are co-orthologs of the vertebrate Calb2 gene, 
that they are differently expressed in retina, and that both genes depend on sox9b 
101
function. cg, cranial ganglia; gcl, ganglion cell layer; op, olfactory placode; r, retina; tg, 
tegmentum.
102
Fig. 4. Candidate genes were expressed in different layers in the developing retina. (A–
C), vsx1; (D–F), rs1; (H–I), calb2a; (J–L), calb2b. Expression of vsx1 was observed in 
the outer part of the inner nuclear layer (A). Expression was reduced especially in the 
dorsal part in sox9b mutants (B) and double mutants (C). In wild-type embryos, rs1 was 
expressed in the outer part of the inner nuclear layer and in the outer nuclear layer (D). In 
sox9b mutants and sox9a;sox9b double mutant embryos, expression was severely reduced 
or nearly gone except for the ventral ciliary marginal zone (E, F). Expression of calb2a 
was observed in the inner nuclear layer (G) and calb2b was expressed in the ganglion cell 
layer (J). In each case, expression was severely reduced in sox9b mutants (H, K) and 
sox9a;sox9b double mutant embryos (I, L). White bar in (A, D), inlo, outer part of inner 
nuclear layer; yellow bar in (D), onl, outer nuclear layer; green bar in (G), inl, inner 
nuclear layer; orange bar in (J), gcl, ganglion cell layer.
103
104
Fig. 5. Expression of candidate genes in retina at 3 dpf. In situ hybridization on 
histological sections with calb2a (Aa, wild-type; Ab, sox9a mutant; Ac, sox9b mutant; 
Ad, double mutant), calb2b (Ba–Bd), crx (Ca–Cd), neurod (Da–Dd), rs1 (Ea–Ed), sox4a 
(Fa–Fd), vsx1 (Ga–Gd) and col2a1a (Ha–Hd). Expression of calb2a appeared in the 
ganglion cell layer and inner nuclear layer (Aa); calb2b was expressed in the ganglion 
cell layer and a cell population in the inner part of the inner nuclear layer (Ba); crx was 
expressed in the outer part of the inner nuclear layer and outer nuclear layer (Ca); neurod 
was expressed in the outer nuclear layer and a cell population in the inner nuclear layer 
(Da); rs1 was expressed in the outer part of the inner nuclear layer and the outer nuclear 
layer (Ea); sox4a was expressed in a cell population in the inner nuclear layer (Fa); and 
vsx1 was expressed in the outer part of the inner nuclear layer (Ga). The ciliary marginal 
zone did not express calb2a, calb2b, neurod, rs1 or sox4a (asterisk). In sox9b and double 
mutants, expression in dorsal margin and ventral margin was reduced for all seven genes. 
The expression of sox4a in the inner nuclear layer was stronger in sox9b and sox9a;sox9b 
double mutants than in wild-type and sox9a mutants (Fa–Fd). Expression of col2a1a 
appeared in iris (Ha), and was not affected in sox9 mutant embryos (Hb–Hd). No obvious 
defect was observed in the sox9a mutant retina. cmz, ciliary marginal zone; gcl, ganglion 
cell layer; inl, inner nuclear layer; inli, inner part of inner nuclear layer; inlo, outer part of 
inner nuclear layer; ir, iris; onl, outer nuclear layer.
105
Fig. 6. Fluorescent antibody staining of retinal cell types in wild-type and sox9 single and 
double mutant larvae. (A–P) 5 dpf and (Q–T) 4 dpf. Eye morphology in sox9a mutants 
was generally similar to that in wild-type siblings. In contrast, sox9b mutants and 
sox9a;sox9b double mutants had smaller eyes and eye morphology was less organized 
than in wild-type siblings and sox9a mutants. (A, E, I, M) Retinal Müller glial cells 
labeled by anti-carbonic anhydrase showed that sox9b mutants (I) and sox9a;sox9b 
double mutants (M) had fewer Müller glial cells compared to wild-types (A) and sox9a 
(E) siblings. (B, F, J, N) Photoreceptor cells labeled by zpr-3 showed that fewer and 
thinner photoreceptor cells were present in sox9b mutants (J) and sox9a;sox9b double 
mutants (N) compared to wild-type (B) and sox9a (F) siblings. (C, G, K, O) Nuclear 
layers of the retina stained by nuclear dye (TO-PRO-3 iodide) showed that all the nuclear 
layers were present in sox9 mutants. (D, H, L, P) green: anti-carbonic anhydrase, red: zpr-
3, and blue: nuclear staining. (Q–T) Retinal amacrine cells labeled by anti-GABA: green, 
photoreceptor cell labeled by zpr-3: red, and nuclear layers of the retina staining by 
nuclear dye (TO-PRO-3 iodide): blue. The results showed that sox9a mutants (R), sox9b 
mutants (S) and sox9a;sox9b double mutants (T) have fewer organized amacrine cells and 
photoreceptor cells compared to wild-type siblings (Q). Scale bar: 100 μm.
106
Fig. 7. Expression of candidate genes in the pectoral fin bud. (A, B) Lateral view and (C, 
D) dorsal view of klf2b expression. In wild-type embryos, klf2b was expressed in the 
mesenchyme of the pectoral fin bud and in the adjacent cleithrum (A, C). Expression in 
the pectoral fin bud was not detected in double mutants (B, D). Expression of klf2b in the 
cleithrum was not affected in the mutant. Expression of EST AI722369 was observed in 
the presumptive muscle in the pectoral fin bud (E). AI722369 expression did not appear 
in sox9a;sox9b double mutant embryos (F). Lateral view of sox9a (G) and sox9b (H) 
expression. At this stage, sox9a was expressed in an L-shape in the pectoral fin bud (G), 
and sox9b expression appeared more broadly in the pectoral fin bud (H). cl, cleithrum; 
co, scapulocoracoid; ed, endochondral disc; pf, pectoral fin bud.
107
Fig. 8. Using ectopic self-renewal mutants for expression profiling of neuroblasts. Six 
genotypes were used which are known to cause expansions in the number of neuroblasts 
in Drosophila larval central brains. (A) Single-slice confocal images of wild-type (120 
hours ALH) and mutant (144 hours ALH) brain lobes stained for Dpn (neuroblast marker) 
and Elav (neuronal marker). (B) The variable level of ectopic neuroblast number per 
brain lobe of each mutant genotype (n=2 for each genotype). (C) Schematic of wild-type 
type I and type II neuroblast divisions. Type I neuroblasts have nuclear Ase as well as 
diffuse cytoplasmic Pros, which is asymmetrically segregated into the GMC upon 
neuroblast division. Type II neuroblasts lack both Pros and Ase, both of which are 
expressed in INPs; Pros is then segregated asymmetrically into the GMC upon INP 
division. GMCs divide to generate Elav+ neurons. (D) High magnification image of a 
Dpn+ Ase+ type I neuroblast (white arrowhead) and a Dpn+ Ase- type II neuroblast (yellow 
arrowhead) in the dorso-medial region of a wild-type brain. The type II neuroblast can be 
unambiguously identified based on the presence of GFP driven by R19H09-Gal4 
(Bayraktar et al., 2010). (E) Mutant brains (120 hours ALH) stained with anti-Dpn (to 
mark all neuroblasts) and anti-Ase (which only marks type I neuroblasts). Inset in the Ase 
panel of the lgl pins brain shows that the Pros staining pattern in the same brain matches 
very closely to the Ase pattern. (F) Schematic of the methodology used here. Scale bars: 
10 µm in (D); 100 µm in (A and E).
108
Fig. 9. Results of cluster analysis. (A) Cluster analysis-categorized genes with expression 
changes in mutant compared to wild-type brains, divided into three groups (A, B, and 
C).The dendrogram at the top is labeled according to the mutant genotype; Roman 
numerals indicate the neuroblast subtype(s) enriched in each mutant (I = type I; II = type 
II). (B) Log2 expression changes (mutant/wild-type) averaged over all genes in each 
group.
109
Fig. 10. Differential expression of genes excluded from type II neuroblasts. (A) Position 
within group B of sub-cluster containing pros and ase as well as nine other genes with 
unknown expression patterns. Log2 enrichment over wild-type is shown, averaged over 
all genes in the sub-cluster. (B) Confocal image of a wild-type brain lobe at 120 hours 
ALH. Multiple type I neuroblasts, four type II lineages, and three type II neuroblasts are 
visible. (C) Enlargement of the region boxed in (B). All neuroblasts are Dpn+; a subset of 
type I neuroblasts are Rx+, while type II neuroblasts are Rx-. INPs are also Rx-. Rx is 
expressed in a subset of neuronal progeny in both type I and type II lineages. (D) brat 
mutant brain lobe (120 hours ALH) contains many Dpn+ neuroblasts, but these cells do 
not express Rx. Rx is expressed in a few of the Pros+ cells, all of which in this focal plane 
are neurons and do not express Dpn. (E) lgl pins brain lobe (120 hours ALH) in which Rx 
is expressed in a subset of ectopic Dpn+ neuroblasts. Rx expression is limited to cells 
expressing Pros, which in lgl pins also express Ase and are likely derived from expansion 
of type I neuroblasts. The Pros- regions (type II-derived) are entirely Rx-. White arrow: 
Rx- type I neuroblast; white arrowheads: Rx+ type I neuroblasts; yellow arrows: Rx- type 
II neuroblasts; mCD8::GFP driven by R9D11-Gal4 marks a subset of type II lineages, but 
not the type II neuroblasts themselves (Bayraktar et al., 2010). Shown outlined here with 
yellow dashed lines are several dorso-medial type II lineages [DM 2, 3, 4, 5, and 6 in (B); 
DM 2, 3, and 4 in (C)]. Scale bars: 100 µm in (B), (D), and (E); 10 µm in (C).
110
Fig. 11. Identification of a cluster with type II-biased expression. (A) Position in group B 
of a sub-cluster in which genes are expressed higher in brat and lgl lgd than in other 
genotypes. Enrichment shown is averaged over all genes in the sub-cluster. (B) Wild-type 
brain lobe (120 hours ALH). Visible are multiple type I neuroblasts as well as several 
type II neuroblasts and their lineages. Most type I neuroblasts are Optix-, while four of 
the six dorso-medial type II neuroblasts are Optix+ (DM1,2,3,6); Optix is absent from 
type II neuroblasts DM 4&5 and their INPs, but present in a subset of their progeny. (C) 
Enlargement of the box in (B) shows both Optix+ and Optix- type I and type II 
neuroblasts. Shown are type II lineages DM 2, 3, and 4. Optix is nearly absent from the 
entire DM4 lineage. (D) brat brain (120 hours ALH) shows that Optix is expressed in a 
dorso-medial region in which nearly all cells are Dpn+ ectopic neuroblasts. (E) lgl pins 
brain (120 hours ALH) exhibits Optix expression primarily in Dpn+ Pros- regions (type II-
derived ectopic neuroblasts). White arrows: Optix- type I neuroblasts; white arrowheads: 
Optix+ type I neuroblasts; yellow arrow: Optix- type II neuroblast; yellow arrowheads: 
Optix+ type II neuroblasts. GFP driven by R9D11-Gal4 marks dorso-medial type II 
lineages, but not the type II neuroblasts themselves. Shown outlined here with yellow 
dashed lines are several dorso-medial type II lineages [DM 2, 3, 4, 5, and 6 in (B); DM 2, 
3, and 4 in (C)]. Scale bars: 100 µm in (B), (D), and (E); 10 µm in (C).
111
Fig. 12. Gene ontology terms enriched in each group. (A) Group A, the “neuroblast 
cluster” with three sub-clusters marked in which the indicated GO annotations are 
significantly enriched compared to all Drosophila genes. Each value in red indicates the 
enrichment of the GO term in all group A genes excluding the adjacent sub-cluster. (B) 
Chart depicting the percent of all Drosophila genes characterized by select GO 
annotations as well as percent of genes in each group with those annotations. Asterisks 
indicate significant enrichment of GO term compared with all Drosophila genes (*: 
112
p<.05; **: p<.001). (C) Group C, the “neuron cluster” with sub-clusters labeled 
indicating significantly enriched GO terms; each value in red indicates the enrichment of 
the GO term in all group C genes except the adjacent sub-cluster.
113
Fig. 13. RNAi screen identifies neuroblast homeostasis genes. (A) Flowchart describing 
the selection of 691 RNAi lines used in this screen. (B) Neuroblast number gain and loss 
phenotypes of the 179 genes for which RNAi knock-down caused lethality. (C) 
Neuroblast numbers per brain lobe of the genes which were assayed for neuroblast 
number phenotype.
114
Fig. 14. MADM-based glioma model allows phenotypic analysis at single-cell resolution. 
(A) Scheme of MADM-based glioma modeling. Inset illustrates how MADM 
concurrently mutates and labels cells. (B) Representative confocal images show hGFAP-
Cre-induced MADM labeling of four NSC-derived cell types in a 2-month-old WT-
MADM mouse. Arrows point to MADM-labeled cells expressing corresponding markers. 
Scale bars, 20 μm; inset, 5 μm.
115
Fig. 15. MADM-mediated sporadic concurrent inactivation of p53 and NF1 in embryonic 
NSCs reveals the entire process of gliomagenesis. (A) Sagittal sections from brains of 
MADM mice at indicated ages. In WT-MADM mice (left column), both green and red 
cells are WT. In mutant-MADM mice (right column), green cells are p53 and NF1 double 
null; red cells, WT. Tumor boundary is demarcated with dashed line. Scale bars, 2 mm; 
insets, 100 μm. Ki67 staining shows that tumor cells are highly proliferative. (B) 
Systematic quantification of G/R ratios in mutant-MADM brains from P5 to P60. Total 
cell numbers counted are shown in parentheses. Each number is the sum from three 
brains. Error bars represent standard error of the mean (SEM). (C) The percentage of 
BrdU+ cells in WT and mutant cell populations in the brain parenchyma of mutant-
MADM mice at indicated ages. Error bars represent SEM. (D) The proportion of BrdU+ 
cells with genotypes “−/−,” “+/−,” and “+/+” in the brain parenchyma of mutant-MADM 
mice at indicated ages. “+/−” includes both double-colored and colorless BrdU+ cells. In 
(C) and (D), BrdU was administered 1.5 hr prior to sacrifice. Error bars ± SEM; n = 3 
mice. p < 0.05, p < 0.01, paired t test.∗ ∗∗
116
Fig. 16. Analysis at a pretransforming stage of gliomagenesis suggests that OPCs rather 
than NSCs serve as the cell of origin. (A) Average G/R ratios of each cell type in brain 
parenchyma of P60 mutant-MADM mice. p < 0.05, p < 0.01, paired t test. #p <∗ ∗∗  
0.0001, one-way ANOVA. Error bars represent SEM. (B) (Left chart) With a single BrdU 
injection (1.5 hr prior to sacrifice) at P60, BrdU-positive mutant (−/−) cells in the brain 
parenchyma consist entirely of OPCs (PDGFRα+). (Right chart) Upon BrdU 
administration by drinking water for 1 week, all BrdU-positive mutant (−/−) cells in the 
brain parenchyma belong to the oligodendrocytic lineage (Olig2+), and the majority of 
them are OPCs (PDGFRα+). (C–E) “4+1” channel staining shows that all BrdU+ mutant 
OPCs in a P60 mutant-MADM mouse brain belong to the oligodendrocyte lineage. (C) 
Without Olig2 staining, all MADM-labeled BrdU+ cells are mutant (green). Notably, 
some BrdU+ mutant cells are PDGFRα negative (marked with cyan circles). (D) An 
adjacent section stained with Olig2 together with MADM, BrdU, and PDGFRα shows 
that all mutant BrdU+ cells have red nuclei, indicating positive staining of Olig2. (E) 
Representative magnified confocal images show: mutant cells without Olig2 staining 
117
(left), mutant cells with Olig2 staining in red channel (middle), and heterozygous yellow 
cells (right). The orthogonal z axis is shown on the side of each panel. Scale bars: (C and 
D) 20 μm; insets, 5 μm; (E) 5 μm. (F) G/R ratios of BrdU+ cells in the SVZ and NeuN+ 
cells in the olfactory bulb (OB) from P60 mutant-MADM mice. BrdU was given in 
drinking water for 7 days. (G) Quantification of cells with indicated genotypes among all 
BrdU+ cells in the non-SVZ brain parenchyma (left chart) or the SVZ (right chart) after 
one week of BrdU administration. n = 3 mice for all quantification in (A), (B), (F) and 
(G).  Error bars represent ± SEM. Total cell numbers counted in (A), (B), (F), and (G) are 
shown in parentheses. Each number is the sum from three brains.
118
Fig. 17. MADM-generated glioma cells exhibit many OPC features. (A) Representative 
image of a mutant-MADM brain carrying a GFP+ glioma. (B–D) Adjacent H&E staining 
of tumor regions shows typical glioma features, including necrotic areas (“N” in B), 
multinucleated giant cells (C), and perivascular satellitosis (D). Scale bars: (A) 2 mm; (C 
and D) 200μm. Magnification in (B), 400×. (E) Representative low-magnification images 
show elevated expression of a panel of well-established glioma markers in tumor regions. 
119
All staining was done with adjacent sections from the same tumor. Tumor boundary is 
demarcated by dashed lines. T, tumor mass. Scale bars, 100 μm except for CD9 staining, 
for which it is 50 μm. (F) Quantitative RT-PCRs confirm the overexpression of OPC 
markers in MADM-generated glioma. (G–L) Confocal images at high magnification 
show that proliferating (Ki67+) green tumor cells express markers for OPCs (G–I, 
pointed by arrows), but not for other cell types (J–L). The signals of cell type marker 
staining in the right column of (J–L) were converted to red for better examination of their 
colocalization with GFP. Some Ki67-negative green cells were CC1+ (circled in L). Scale 
bars, 20 μm. (M) Transcriptome comparison between tumor samples and four neuroglial 
cell types (top four rows) and the four subtypes of human GBMs defined by TCGA 
(bottom four rows) with the single sample Gene Set Enrichment Analysis (ssGSEA) 
method. N and O represent tumor samples from mutant-MADM mice induced by NSC-
Cre (Nestin-Cre or hGFAP-Cre) and by NG2-Cre, respectively. PDGFRα+ indicates 
primary tumor cells enriched by anti-PDGFRα immunopanning method. Red to blue 
indicates significantly similar to dissimilar.
120
Fig. 18. Spatial analyses of early lesions based on perineuronal cytoarchitecture as a 
landmark suggest that gliomas initiate at brain regions away from the SVZ. (A–C) 
Immunofluorescent staining of mutant-MADM brains at distinct tumorigenic stages. 
Neuronal nuclei were stained with NeuN and are marked as “ .” Arrows point to∗  
perineuronal pretransforming OPCs or tumor cells. The proliferating status of 
perineuronal mutant cells is shown by Ki67 staining (yellow arrows in the bottom row). 
(D) H&E staining of the adjacent section of (C) shows perineuronal satellitosis. (E) 
121
Proportion of perineuronal structures with distinct mutant OPC-to-neuron ratios (O:N) 
either in pretransforming MADM mutant brains (P30 and P60, n = 3 brains each) or in 
tumors (T, n = 4). (F) Schematic summary of lesion sites (green spots), which are defined 
by O:N ≥ 3 perineuronal structure together with MADM labeling, Ki67 staining, and also 
pathology in most cases. Brains that are devoid of any detectable lesions are shown in 
light gray. The analysis is based on a cohort of mutant-MADM mice induced by hGFAP-
Cre. (G–K) Representative brain images from (F) with gliomas from medium to small 
sizes. Insets show Ki67 staining of the tumor regions. The glioma identity in these brains 
was confirmed by pathological criteria except for (K), in which the perineuronal structure 
of ≥ 3 mutant OPCs suggests that it should be a lesion at its early stage. Scale bars: (A–
D) 20 μm; (G–K) left column, 2 mm; middle and right columns, 50 μm.
122
Fig. 19. OPCs can be directly transformed into malignant glioma. (A–D) In the MADM 
system, NG2-Cre transgene labels OPCs and oligodendrocytes, but not astrocytes or 
neurons. Arrows point to MADM-labeled cells expressing indicated markers. (E–H) 
Mutant OPCs overexpand at pretransforming stages. Brain sections from P60 WT-
MADM (E) or mutant-MADM (F) mice induced by NG2-Cre. (G) G/R ratios within each 
cell lineage in P60 Mutant-MADM brains induced by NG2-Cre. NA, not applicable. 
Error bars represent ± SEM. Total cell number being counted is shown in parentheses. 
(H) Percentage of OPCs (PDGFRα+) versus oligodendrocytes (CC1+) within mutant and 
WT cell populations. n = 3 mice in (G) and (H). p < 0.05, t test.∗  (I and J) Representative 
gross images of malignant glioma in NG2-Cre-induced mutant-MADM mice, either 
123
locating around hypothalamus with invasion into the subarachnoid space (I) or residing 
within the brain parenchyma (J). Arrows in (I) point to tumor cells spreading along the 
meninges. Ki67 staining in insets shows that tumor cells are highly proliferative. Scale 
bars, 2mm; inset, 100 μm.
124
Fig. 20. Comparative pathological analyses of NSC- and NG2-Cre induced tumors 
suggest that tumor cell morphology is highly dependent on the location rather than 
initially mutated cell types. (A and B) Regardless of the Cre lines used, tumors at the 
same location exhibit indistinguishable pathological features. (C) Transplantation of 
tumor cells from NG2-Cre-induced glioma with subarachnoid invasion (top row) into the 
brain parenchyma of NOD/SCID mice to generate secondary tumors (bottom row). 
Pathological features of the primary and the secondary tumors mimic tumor features in 
(B) and (A), respectively. Tumor boundaries are demarcated by dashed lines. The 
magnification of images in the middle columns is 400×. Images from the right column 
are 2.5× digital zoom-in of the corresponding middle-column ones.
125
Fig. 21. TU-tagging: overview and cell type-specific labeling. (A) TU-tagging procedure. 
(B) TU-tagging in larvae and adult flies. RNA from larvae expressing GAL4 in none of 
the cells (−), all neurons (elav), muscle cells (c179) or the mushroom body (OK107) and 
either with or without UAS-UPRT, and from adults of the indicated genotypes were 
electrophoresed and stained with ethidium bromide to detect all RNA (top) and 
streptavidin-horseradish peroxidase to detect thio-RNA (bottom). (C) The 0–16-h 
embryos of the indicated genotypes were treated with 4-thiouracil for 2 h. RNA was 
analyzed as indicated above. (D) Comparison of RNA labeling before tissue isolation 
(whole larvae) and after tissue isolation (brain only). twi/how, mesoderm/muscle; pros, 
neural; and wor, neuroblasts.
126
Fig. 22. Cell type-specific RNA isolation and analysis. (A) Single confocal section 
through a brain lobe from a 96-h after larval hatching (ALH) larvae expressing 
hemagglutinin-tagged UPRT (HA-UPRT) in all glia (repo-GAL4 UAS-HA-UPRT) stained 
for HA-UPRT (detected with an HA antibody) and glial nuclei (detected with Repo 
antibody). Scale bar, 20 μm (B) Fold enrichment of the indicated larval glia–specific 
genes. (C) Average microarray ratios from two glia TU-tagging experiments plotted 
against the number of uracils in the transcript. Dashed red line indicates cutoff for top 5% 
enriched genes. Green dots, previously known larval glia–specific genes; vertical 
arrowhead, possible false positives; and horizontal arrowhead, possible false negatives. 
(D) TU-tagging microarray ratios after normalization and removal of transcripts with 
missing untranslated region annotations. (E) Comparison of top 1% and 5% enriched 
genes before and after normalization for uracil number.
127
REFERENCES CITED
Akiyama, H., Chaboissier, M.C., Behringer, R.R., Rowitch, D.H., Schedl, A., Epstein, 
J.A., de Crombrugghe, B., 2004. Essential role of Sox9 in the pathway that 
controls formation of cardiac valves and septa. Proc. Natl. Acad. Sci. U.S.A. 101, 
6502-6507.
Albertson, R., Chabu, C., Sheehan, A., Doe, C.Q., 2004. Scribble protein domain 
mapping reveals a multistep localization mechanism and domains necessary for 
establishing cortical polarity. J. Cell. Sci. 117, 6061-6070.
Alcantara Llaguno, S., Chen, J., Kwon, C.H., Jackson, E.L., Li, Y., Burns, D.K., Alvarez-
Buylla, A., Parada, L.F., 2009. Malignant astrocytomas originate from neural 
stem/progenitor cells in a somatic tumor suppressor mouse model. Cancer Cell 
15, 45-56.
Almeida, M.S., Bray, S.J., 2005. Regulation of post-embryonic neuroblasts by Drosophila 
Grainyhead. Mech. Dev. 122, 1282-1293.
Amores, A., Force, A., Yan, Y.L., Joly, L., Amemiya, C., Fritz, A., Ho, R.K., Langeland, 
J., Prince, V., Wang, Y.L., et al., 1998. Zebrafish hox clusters and vertebrate 
genome evolution. Science 282, 1711-1714.
Arango, N.A., Lovell-Badge, R., Behringer, R.R., 1999. Targeted mutagenesis of the 
endogenous mouse Mis gene promoter: in vivo definition of genetic pathways of 
vertebrate sexual development. Cell 99, 409-419.
Ashraf, S.I., Ganguly, A., Roote, J., Ip, Y.T., 2004. Worniu, a Snail family zinc-finger 
protein, is required for brain development in Drosophila. Dev. Dyn. 231, 379-386.
Assanah, M., Lochhead, R., Ogden, A., Bruce, J., Goldman, J., Canoll, P., 2006. Glial 
progenitors in adult white matter are driven to form malignant gliomas by 
platelet-derived growth factor-expressing retroviruses. J. Neurosci. 26, 6781-
6790.
Atwood, S.X., Prehoda, K.E., 2009. aPKC phosphorylates Miranda to polarize fate 
determinants during neuroblast asymmetric cell division. Curr. Biol. 19, 723-729.
Awasaki, T., Tatsumi, R., Takahashi, K., Arai, K., Nakanishi, Y., Ueda, R., Ito, K., 2006. 
Essential role of the apoptotic cell engulfment genes draper and ced-6 in 
programmed axon pruning during Drosophila metamorphosis. Neuron 50, 855-
867.
Bachoo, R.M., Maher, E.A., Ligon, K.L., Sharpless, N.E., Chan, S.S., You, M.J., Tang, 
Y., DeFrances, J., Stover, E., Weissleder, R., et al., 2002. Epidermal growth factor 
receptor and Ink4a/Arf: convergent mechanisms governing terminal 
differentiation and transformation along the neural stem cell to astrocyte axis. 
Cancer Cell 1, 269-277.
128
Bainton, R.J., Tsai, L.T.Y., Schwabe, T., DeSalvo, M., Gaul, U., Heberlein, U., 2005. 
moody encodes two GPCRs that regulate cocaine behaviors and blood-brain 
barrier permeability in Drosophila. Cell 123, 145-156.
Barbie, D.A., Tamayo, P., Boehm, J.S., Kim, S.Y., Moody, S.E., Dunn, I.F., Schinzel, A. 
C., Sandy, P., Meylan, E., Scholl, C., et al., 2009. Systematic RNA interference 
reveals that oncogenic KRAS-driven cancers require TBK1. Nature 462, 108-112.
Barrionuevo, F., Naumann, A., Bagheri-Fam, S., Speth, V., Taketo, M.M., Scherer, G., 
Neubüser, A., 2008. Sox9 is required for invagination of the otic placode in mice. 
Dev. Biol. 317, 213-224.
Barrionuevo, F., Taketo, M.M., Scherer, G., Kispert, A., 2006. Sox9 is required for 
notochord maintenance in mice. Dev. Biol. 295, 128-140.
Bastide, P., Darido, C., Pannequin, J., Kist, R., Robine, S., Marty-Double, C., Bibeau, F., 
Scherer, G., Joubert, D., Hollande, F., et al., 2007. Sox9 regulates cell 
proliferation and is required for Paneth cell differentiation in the intestinal 
epithelium. J. Cell Biol. 178, 635-648.
Bayraktar, O.A., Boone, J.Q., Drummond, M.L., Doe, C.Q., 2010. Drosophila type II 
neuroblast lineages keep Prospero levels low to generate large clones that 
contribute to the adult brain central complex. Neural Dev. 5, 26.
Bell, D.M., Leung, K.K., Wheatley, S.C., Ng, L.J., Zhou, S., Ling, K.W., Sham, M.H., 
Koopman, P., Tam, P.P., Cheah, K.S., 1997. SOX9 directly regulates the type-II 
collagen gene. Nat. Genet. 16, 174-178.
Bello, B.C., Izergina, N., Caussinus, E., Reichert, H., 2008. Amplification of neural stem 
cell proliferation by intermediate progenitor cells in Drosophila brain 
development. Neural Dev. 3, 5.
Bennett, M.R., Rizvi, T.A., Karyala, S., McKinnon, R.D., Ratner, N., 2003. Aberrant 
growth and differentiation of oligodendrocyte progenitors in neurofibromatosis 
type 1 mutants. J. Neurosci. 23, 7207-7217.
Bernardos, R.L., Lentz, S.I., Wolfe, M.S., and Raymond, P.A., 2005. Notch-Delta 
signaling is required for spatial patterning and Müller glia differentiation in the 
zebrafish retina. Dev. Biol. 278, 381-395.
Betschinger, J., Mechtler, K., Knoblich, J.A., 2006. Asymmetric segregation of the tumor 
suppressor brat regulates self-renewal in Drosophila neural stem cells. Cell 124, 
1241-1253.
Bi, W., Deng, J.M., Zhang, Z., Behringer, R.R., de Crombrugghe, B., 1999. Sox9 is 
required for cartilage formation. Nat. Genet. 22, 85-89.
129
Bi, W., Huang, W., Whitworth, D.J., Deng, J.M., Zhang, Z., Behringer, R.R., de 
Crombrugghe, B., 2001. Haploinsufficiency of Sox9 results in defective cartilage 
primordia and premature skeletal mineralization. Proc. Natl. Acad. Sci. U.S.A. 98, 
6698-6703.
Bier, E., Vaessin, H., Younger-Shepherd, S., Jan, L.Y., Jan, Y.N., 1992. deadpan, an 
essential pan-neural gene in Drosophila, encodes a helix-loop-helix protein 
similar to the hairy gene product. Genes Dev. 6, 2137-2151.
Blackshaw, S., Fraioli, R.E., Furukawa, T., Cepko, C.L., 2001. Comprehensive analysis 
of photoreceptor gene expression and the identification of candidate retinal 
disease genes. Cell 107, 579-589.
Boone, J.Q., Doe, C.Q., 2008. Identification of Drosophila type II neuroblast lineages 
containing transit amplifying ganglion mother cells. Dev. Neurobiol. 68, 1185-
1195.
Bowles, J., Schepers, G., Koopman, P., 2000. Phylogeny of the SOX family of 
developmental transcription factors based on sequence and structural indicators. 
Dev. Biol. 227, 239-255.
Bowman, S.K., Rolland, V., Betschinger, J., Kinsey, K.A., Emery, G., Knoblich, J.A., 
2008. The tumor suppressors Brat and Numb regulate transit-amplifying 
neuroblast lineages in Drosophila. Dev. Cell 14, 535-546.
Brand, M., Jarman, A.P., Jan, L.Y., Jan, Y.N., 1993. asense is a Drosophila neural 
precursor gene and is capable of initiating sense organ formation. Development 
119, 1-17.
Bridgewater, L.C., Lefebvre, V., de Crombrugghe, B., 1998. Chondrocyte-specific 
enhancer elements in the Col11a2 gene resemble the Col2a1 tissue-specific 
enhancer. J. Biol. Chem. 273, 14998-15006.
Broadus, J., Fuerstenberg, S., Doe, C.Q., 1998. Staufen-dependent localization of 
prospero mRNA contributes to neuroblast daughter-cell fate. Nature 391, 792-
795.
Cahoy, J.D., Emery, B., Kaushal, A., Foo, L.C., Zamanian, J.L., Christopherson, K.S., 
Xing, Y., Lubischer, J.L., Krieg, P.A., Krupenko, S.A., et al., 2008. A 
transcriptome database for astrocytes, neurons, and oligodendrocytes: a new 
resource for understanding brain development and function. J. Neurosci. 28, 264-
278.
Caldwell, M.C., Datta, S., 1998. Expression of cyclin E or DP/E2F rescues the G1 arrest 
of trol mutant neuroblasts in the Drosophila larval central nervous system. Mech. 
Dev. 79, 121-130.
130
Catchen, J.M., Conery, J.S., Postlethwait, J.H., 2008. Inferring ancestral gene order. 
Methods Mol. Biol. 452, 365-383.
Chaboissier, M.C., Kobayashi, A., Vidal, V.I.P., Lützkendorf, S., van de Kant, H.J.G., 
Wegner, M., de Rooij, D.G., Behringer, R.R., Schedl, A., 2004. Functional 
analysis of Sox8 and Sox9 during sex determination in the mouse. Development 
131, 1891-1901.
Chabu, C. Doe, C.Q., 2008. Dap160/intersectin binds and activates aPKC to regulate cell 
polarity and cell cycle progression. Development 135, 2739-2746.
Chae, J.H., Stein, G.H., Lee, J.E., 2004. NeuroD: the predicted and the surprising. Mol. 
Cells 18, 271-288.
Cheung, M., Chaboissier, M.C., Mynett, A., Hirst, E., Schedl, A., Briscoe, J., 2005. The 
transcriptional control of trunk neural crest induction, survival, and delamination. 
Dev. Cell 8, 179-192.
Chia, W., Somers, W.G., Wang, H., 2008. Drosophila neuroblast asymmetric divisions: 
cell cycle regulators, asymmetric protein localization, and tumorigenesis. J. Cell 
Biol. 180, 267-272.
Chiang, E.F., Pai, C.I., Wyatt, M., Yan, Y.L., Postlethwait, J., Chung, B., 2001. Two sox9 
genes on duplicated zebrafish chromosomes: expression of similar transcription 
activators in distinct sites. Dev. Biol. 231, 149-163.
Choksi, S.P., Southall, T.D., Bossing, T., Edoff, K., de Wit, E., Fischer, B.E., van 
Steensel, B., Micklem, G., Brand, A.H., 2006. Prospero acts as a binary switch 
between self-renewal and differentiation in Drosophila neural stem cells. Dev. 
Cell 11, 775-789.
Chow, R.L., Volgyi, B., Szilard, R.K., Ng, D., McKerlie, C., Bloomfield, S.A., Birch, 
D.G., McInnes, R.R., 2004. Control of late off-center cone bipolar cell 
differentiation and visual signaling by the homeobox gene Vsx1. Proc. Natl. Acad. 
Sci. U.S.A. 101, 1754-1759.
Cleary, M.D., Meiering, C.D., Jan, E., Guymon, R., Boothroyd, J.C., 2005. Biosynthetic 
labeling of RNA with uracil phosphoribosyltransferase allows cell-specific 
microarray analysis of mRNA synthesis and decay. Nat. Biotechnol. 23, 232-237.
Collier, L.S., Carlson, C.M., Ravimohan, S., Dupuy, A.J., Largaespada, D.A., 2005. 
Cancer gene discovery in solid tumours using transposon-based somatic 
mutagenesis in the mouse. Nature 436, 272-276.
Cresko, W.A., Yan, Y.L., Baltrus, D.A., Amores, A., Singer, A., Rodríguez-Marí, A., 
Postlethwait, J.H., 2003. Genome duplication, subfunction partitioning, and 
lineage divergence: Sox9 in stickleback and zebrafish. Dev. Dyn. 228, 480-489.
131
Davis, R.J., Tavsanli, B.C., Dittrich, C., Walldorf, U., Mardon, G., 2003. Drosophila 
retinal homeobox (drx) is not required for establishment of the visual system, but 
is required for brain and clypeus development. Dev. Biol. 259, 272-287.
Dawson, M.R.L., Polito, A., Levine, J.M., Reynolds, R., 2003. NG2-expressing glial 
progenitor cells: an abundant and widespread population of cycling cells in the 
adult rat CNS. Mol. Cell. Neurosci. 24, 476-488.
Dietzl, G., Chen, D., Schnorrer, F., Su, K.C., Barinova, Y., Fellner, M., Gasser, B., 
Kinsey, K., Oppel, S., Scheiblauer, S., et al., 2007. A genome-wide transgenic 
RNAi library for conditional gene inactivation in Drosophila. Nature 448, 151-
156.
Ding, S., Wu, X., Li, G., Han, M., Zhuang, Y., Xu, T., 2005. Efficient transposition of the 
piggyBac (PB) transposon in mammalian cells and mice. Cell 122, 473-483.
Doe, C.Q., 2008. Neural stem cells: balancing self-renewal with differentiation. 
Development 135, 1575-1587.
Doetsch, F., Caillé, I., Lim, D.A., García-Verdugo, J.M., Alvarez-Buylla, A., 1999. 
Subventricular zone astrocytes are neural stem cells in the adult mammalian brain. 
Cell 97, 703-716.
Domínguez, M., Campuzano, S., 1993. asense, a member of the Drosophila achaete-scute 
complex, is a proneural and neural differentiation gene. EMBO J. 12, 2049-2060.
Doyle, J.P., Dougherty, J.D., Heiman, M., Schmidt, E.F., Stevens, T.R., Ma, G., Bupp, S., 
Shrestha, P., Shah, R.D., Doughty, M.L., et al., 2008. Application of a 
translational profiling approach for the comparative analysis of CNS cell types. 
Cell 135, 749-762.
Dölken, L., Ruzsics, Z., Rädle, B., Friedel, C.C., Zimmer, R., Mages, J., Hoffmann, R., 
Dickinson, P., Forster, T., Ghazal, P., et al., 2008. High-resolution gene expression 
profiling for simultaneous kinetic parameter analysis of RNA synthesis and decay. 
RNA 14, 1959-1972.
Ebens, A.J., Garren, H., Cheyette, B.N., Zipursky, S.L., 1993. The Drosophila 
anachronism locus: a glycoprotein secreted by glia inhibits neuroblast 
proliferation. Cell 74, 15-27.
Eggert, T., Hauck, B., Hildebrandt, N., Gehring, W.J., Walldorf, U., 1998. Isolation of a 
Drosophila homolog of the vertebrate homeobox gene Rx and its possible role in 
brain and eye development. Proc. Natl. Acad. Sci. U.S.A. 95, 2343-2348.
Eisen, M.B., Spellman, P.T., Brown, P.O., Botstein, D., 1998. Cluster analysis and 
display of genome-wide expression patterns. Proc. Natl. Acad. Sci. U.S.A. 95, 
14863-14868.
132
Fadool, J.M., Dowling, J. E., 2008. Zebrafish: a model system for the study of eye 
genetics. Prog. Retin. Eye Res. 27, 89-110.
Flicek, P., Aken, B.L., Beal, K., Ballester, B., Caccamo, M., Chen, Y., Clarke, L., Coates, 
G., Cunningham, F., Cutts, T., et al., 2008. Ensembl 2008. Nucleic Acids Res. 36, 
D707-714.
Force, A., Lynch, M., Pickett, F.B., Amores, A, Yan, Y.L., Postlethwait, J.H., 1999. 
Preservation of duplicate genes by complementary, degenerative mutations. 
Genetics 151, 1531-1545.
Foster, J.W., Dominguez-Steglich, M.A., Guioli, S., Kwok, C., Weller, P.A., Stevanović, 
M., Weissenbach, J., Mansour, S., Young, I.D., Goodfellow, P.N., 1994. 
Campomelic dysplasia and autosomal sex reversal caused by mutations in an 
SRY-related gene. Nature 372, 525-530.
Furukawa, T., Morrow, E.M., Li, T., Davis, F.C., Cepko, C.L., 1999. Retinopathy and 
attenuated circadian entrainment in Crx-deficient mice. Nat. Genet. 23, 466-470.
Furukawa, T., Morrow, E.M., Cepko, C.L., 1997. Crx, a novel otx-like homeobox gene, 
shows photoreceptor-specific expression and regulates photoreceptor 
differentiation. Cell 91, 531-541.
Galtier, N., Gouy, M., Gautier, C., 1996. SEAVIEW and PHYLO_WIN: two graphic tools 
for sequence alignment and molecular phylogeny. Comput. Appl. Biosci. 12, 543-
548.
Ge, H., Liu, Z., Church, G.M., Vidal, M., 2001. Correlation between transcriptome and 
interactome mapping data from Saccharomyces cerevisiae. Nat. Genet. 29, 482-
486.
Geha, S., Pallud, J., Junier, M.P., Devaux, B., Leonard, N., Chassoux, F., Chneiweiss, H., 
Daumas-Duport, C., Varlet, P., 2010. NG2+/Olig2+ cells are the major cycle-
related cell population of the adult human normal brain. Brain Pathol. 20, 399-
411.
Genove, G., DeMarco, U., Xu, H., Goins, W.F., Ahrens, E.T., 2005. A new transgene 
reporter for in vivo magnetic resonance imaging. Nat. Med. 11, 450-454.
Genzer, M.A., Bridgewater, L.C., 2007. A Col9a1 enhancer element activated by two 
interdependent SOX9 dimers. Nucleic Acids Res. 35, 1178-1186.
Heiman, M., Schaefer, A., Gong, S., Peterson, J.D., Day, M., Ramsey, K.E., Suárez-
Fariñas, M., Schwarz, C., Stephan, D.A., Surmeier, D.J., et al., 2008. A 
translational profiling approach for the molecular characterization of CNS cell 
types. Cell 135, 738-748.
133
Hippenmeyer, S., Youn, Y.H., Moon, H.M., Miyamichi, K., Zong, H., Wynshaw-Boris, 
A., Luo, L., 2010. Genetic mosaic dissection of Lis1 and Ndel1 in neuronal 
migration. Neuron 68, 695-709.
Hirata, J., Nakagoshi, H., Nabeshima, Y., Matsuzaki, F., 1995. Asymmetric segregation of 
the homeodomain protein Prospero during Drosophila development. Nature 377, 
627-630.
Hoheisel, J.D., 2006. Microarray technology: beyond transcript profiling and genotype 
analysis. Nat. Rev. Genet. 7, 200-210.
Hosoya, T., Takizawa, K., Nitta, K., Hotta, Y., 1995. glial cells missing: a binary switch 
between neuronal and glial determination in Drosophila. Cell 82, 1025-1036.
Houston, C.S., Opitz, J.M., Spranger, J.W., Macpherson, R.I., Reed, M.H., Gilbert, E.F., 
Herrmann, J., Schinzel, A., 1983. The campomelic syndrome: review, report of 17 
cases, and follow-up on the currently 17-year-old boy first reported by Maroteaux 
et al in 1971. Am. J. Med. Genet. 15, 3-28.
Ihanamäki, T., Säämänen, A.M., Suominen, J., Pelliniemi, L.J., Harley, V., Vuorio, E., 
Salminen, H., 2002. Expression of Sox9 and type IIA procollagen during ocular 
development and aging in transgenic Del1 mice with a mutation in the type II 
collagen gene. Eur. J. Ophthalmol. 12, 450-458.
Izergina, N., Balmer, J., Bello, B., Reichert, H., 2009. Postembryonic development of 
transit amplifying neuroblast lineages in the Drosophila brain. Neural. Dev. 4, 44.
Jacks, T, Remington, L., Williams, B.O., Schmitt, E.M., Halachmi, S., Bronson, R.T., 
Weinberg, R.A., 1994. Tumor spectrum analysis in p53-mutant mice. Curr. Biol. 
4, 1-7.
Jaekel, R., Klein, T., 2006. The Drosophila Notch inhibitor and tumor suppressor gene 
lethal (2) giant discs encodes a conserved regulator of endosomal trafficking. Dev. 
Cell 11, 655-669.
Jaillon, O., Aury, J.M., Brunet, F., Petit, J.L., Stange-Thomann, N., Mauceli, E., Bouneau, 
L., Fischer, C., Ozouf-Costaz, C., Bernot, A., et al., 2004. Genome duplication in 
the teleost fish Tetraodon nigroviridis reveals the early vertebrate proto-
karyotype. Nature 431, 946-957.
Jansen, R., Greenbaum, D., Gerstein, M., 2002. Relating whole-genome expression data 
with protein-protein interactions. Genome Res. 12, 37-46.
Jensen, A.M., Walker, C., Westerfield, M., 2001. mosaic eyes: a zebrafish gene required 
in pigmented epithelium for apical localization of retinal cell division and 
lamination. Development 128, 95-105.
134
Kenyon, K.L., Li, D.J., Clouser, C., Tran, S., Pignoni, F., 2005. Fly SIX-type 
homeodomain proteins Sine oculis and Optix partner with different cofactors 
during eye development. Dev. Dyn. 234, 497-504.
Kimmel, C.B., Ballard, W.W., Kimmel, S.R., Ullmann, B., Schilling, T.F., 1995. Stages of 
embryonic development of the zebrafish. Dev. Dyn. 203, 253-310.
Kist, R., Schrewe, H., Balling, R., Scherer, G., 2002. Conditional inactivation of Sox9: a 
mouse model for campomelic dysplasia. Genesis 32, 121-123.
Knoblich, J.A., Jan, L.Y., Jan, Y.N., 1995. Asymmetric segregation of Numb and 
Prospero during cell division. Nature 377, 624-627.
Knoblich, J.A., 2008. Mechanisms of asymmetric stem cell division. Cell 132, 583-597.
Knudson, A.G., Jr., 1971. Mutation and cancer: statistical study of retinoblastoma. Proc. 
Natl. Acad. Sci. U.S.A. 68, 820-823.
Komitova, M., Zhu, X., Serwanski, D.R., Nishiyama, A., 2009. NG2 cells are distinct 
from neurogenic cells in the postnatal mouse subventricular zone. J. Comp. 
Neurol. 512, 702-716.
Kondo, T., Raff, M., 2000. Oligodendrocyte precursor cells reprogrammed to become 
multipotential CNS stem cells. Science 289, 1754-1757.
Korzh, V., Sleptsova, I., Liao, J., He, J., Gong, Z., 1998. Expression of zebrafish bHLH 
genes ngn1 and nrd defines distinct stages of neural differentiation. Dev. Dyn. 
213, 92-104.
Lee, C.Y., Andersen, R.O., Cabernard, C., Manning, L., Tran, K.D., Lanskey, M.J., 
Bashirullah, A., Doe, C.Q., 2006. Drosophila Aurora-A kinase inhibits neuroblast 
self-renewal by regulating aPKC/Numb cortical polarity and spindle orientation. 
Genes Dev. 20, 3464-3474.
Lee, C.Y., Robinson, K.J., Doe, C.Q., 2006. Lgl, Pins and aPKC regulate neuroblast self-
renewal versus differentiation. Nature 439, 594-598.
Lee, C.Y., Wilkinson, B.D., Siegrist, S.E., Wharton, R.P., Doe, C.Q., 2006. Brat is a 
Miranda cargo protein that promotes neuronal differentiation and inhibits 
neuroblast self-renewal. Dev. Cell 10, 441-449.
Lee, J.E., Hollenberg, S.M., Snider, L., Turner, D.L., Lipnick, N., Weintraub, H., 1995. 
Conversion of Xenopus ectoderm into neurons by NeuroD, a basic helix-loop-
helix protein. Science 268, 836-844.
135
Lee, J.S., Padmanabhan, A., Shin, J., Zhu, S., Guo, F., Kanki, J.P., Epstein, J.A., Look, 
A.T., 2010. Oligodendrocyte progenitor cell numbers and migration are regulated 
by the zebrafish orthologs of the NF1 tumor suppressor gene. Hum. Mol. Genet. 
19, 4643-4653.
Lefebvre, V., Huang, W., Harley, V.R., Goodfellow, P.N., de Crombrugghe, B., 1997. 
SOX9 is a potent activator of the chondrocyte-specific enhancer of the pro 
alpha1(II) collagen gene. Mol. Cell. Biol. 17, 2336-2346.
Levine, E.M., Passini, M., Hitchcock, P.F., Glasgow, E., Schechter, N., 1997. Vsx-1 and 
Vsx-2: two Chx10-like homeobox genes expressed in overlapping domains in the 
adult goldfish retina. J. Comp. Neurol. 387, 439-448.
Li, L., Vaessin, H., 2000. Pan-neural Prospero terminates cell proliferation during 
Drosophila neurogenesis. Genes Dev. 14, 147-151.
Ligon, K.L., Huillard, E., Mehta, S., Kesari, S., Liu, H., Alberta, J.A., Bachoo, R.M., 
Kane, M., Louis, D.N., Depinho, R.A., et al., 2007. Olig2-regulated lineage-
restricted pathway controls replication competence in neural stem cells and 
malignant glioma. Neuron 53, 503-517.
Lincoln, J., Kist, R., Scherer, G., Yutzey, K.E., 2007. Sox9 is required for precursor cell 
expansion and extracellular matrix organization during mouse heart valve 
development. Dev. Biol. 305, 120-132.
Lindberg, N., Kastemar, M., Olofsson, T., Smits, A., Uhrbom, L., 2009. Oligodendrocyte 
progenitor cells can act as cell of origin for experimental glioma. Oncogene 28, 
2266-2275.
Liu, Y., Li, H., Tanaka, K., Tsumaki, N., Yamada, Y., 2000. Identification of an enhancer 
sequence within the first intron required for cartilage-specific transcription of the 
alpha2(XI) collagen gene. J. Biol. Chem. 275, 12712-12718.
Liu, Y., Shen, Y., Rest, J.S., Raymond, P.A., Zack, D.J., 2001. Isolation and 
characterization of a zebrafish homologue of the cone rod homeobox gene. Invest. 
Ophthalmol. Vis. Sci. 42, 481-487.
Livesey, F.J., Furukawa, T., Steffen, M.A., Church, G.M., Cepko, C.L. (2000). 
Microarray analysis of the transcriptional network controlled by the photoreceptor 
homeobox gene Crx. Curr. Biol. 10, 301-310.
Louis, D.N., Ohgaki, H., Wiestler, O.D., Cavenee, W.K., 2007. WHO Classification of 
Tumours of the Central Nervous System 4th ed. WHO Press, Geneva, 
Switzerland.
Malicki, J., 2000. Harnessing the power of forward genetics--analysis of neuronal 
diversity and patterning in the zebrafish retina. Trends Neurosci. 23, 531-541.
136
Mansour, S., Offiah, A.C., McDowall, S., Sim, P., Tolmie, J., Hall, C., 2002. The 
phenotype of survivors of campomelic dysplasia. J. Med. Genet. 39, 597-602.
Marshall, O.J., Harley, V.R., 2000. Molecular mechanisms of SOX9 action. Mol. Genet. 
Metab. 71, 455-462.
Mavropoulos, A., Devos, N., Biemar, F., Zecchin, E., Argenton, F., Edlund, H., Motte, P., 
Martial, J.A., Peers, B., 2005. sox4b is a key player of pancreatic alpha cell 
differentiation in zebrafish. Dev. Biol. 285, 211-223.
McLendon, R., Friedman, A., Bigner, D., Van Meir, E.G., Brat, D.J., Mastrogianakis, M., 
Olson, J.J., Mikkelsen, T., Lehman, N., Aldape, K., 2008. Comprehensive 
genomic characterization defines human glioblastoma genes and core pathways. 
Nature 455, 1061-1068.
Merkle, F.T., Alvarez-Buylla, A., 2006. Neural stem cells in mammalian development. 
Curr. Opin. Cell Biol. 18, 704-709.
Miller, M.R., Robinson, K.J., Cleary, M.D., Doe, C.Q., 2009. TU-tagging: cell type-
specific RNA isolation from intact complex tissues. Nat. Methods 6, 439-441.
Molday, L.L., Wu, W.W.H., Molday, R.S., 2007. Retinoschisin (RS1), the protein 
encoded by the X-linked retinoschisis gene, is anchored to the surface of retinal 
photoreceptor and bipolar cells through its interactions with a Na/K ATPase-
SARM1 complex. J. Biol. Chem. 282, 32792-32801.
Mori-Akiyama, Y., Akiyama, H., Rowitch, D.H., de Crombrugghe, B., 2003. Sox9 is 
required for determination of the chondrogenic cell lineage in the cranial neural 
crest. Proc. Natl. Acad. Sci. U.S.A. 100, 9360-9365.
Mori-Akiyama, Y., van den Born, M., van Es, J.H., Hamilton, S.R., Adams, H.P., Zhang, 
J., Clevers, H., de Crombrugghe, B., 2007. SOX9 is required for the 
differentiation of paneth cells in the intestinal epithelium. Gastroenterology 133, 
539-546.
Morrison, S.J., Kimble, J., 2006. Asymmetric and symmetric stem-cell divisions in 
development and cancer. Nature 441, 1068-1074.
Mueller, T., Wullimann, M.F., 2002. Expression domains of neuroD (nrd) in the early 
postembryonic zebrafish brain. Brain Res. Bull. 57, 377-379.
Naya, F.J., Stellrecht, C.M., Tsai, M.J., 1995. Tissue-specific regulation of the insulin 
gene by a novel basic helix-loop-helix transcription factor. Genes Dev. 9, 1009-
1019.
Nelson, S.B., Hempel, C., Sugino, K., 2006. Probing the transcriptome of neuronal cell 
types. Curr. Opin. Neurobiol. 16, 571-576.
137
Ng, L.J., Wheatley, S., Muscat, G.E., Conway-Campbell, J., Bowles, J., Wright, E., Bell, 
D.M., Tam, P.P., Cheah, K.S., Koopman, P., 1997. SOX9 binds DNA, activates 
transcription, and coexpresses with type II collagen during chondrogenesis in the 
mouse. Dev. Biol. 183, 108-121.
van Noort, V., Snel, B., Huynen, M.A., 2003. Predicting gene function by conserved co-
expression. Trends Genet. 19, 238-242.
Nunes, M.C., Roy, N.S., Keyoung, H.M., Goodman, R.R., McKhann, G., 2nd, Jiang, L., 
Kang, J., Nedergaard, M., Goldman, S.A., 2003. Identification and isolation of 
multipotential neural progenitor cells from the subcortical white matter of the 
adult human brain. Nat. Med. 9, 439-447.
Oates, A.C., Pratt, S.J., Vail, B., Yan Y.L., Ho, R.K., Johnson, S.L., Postlethwait, J.H., 
Zon, L.I., 2001. The zebrafish klf gene family. Blood 98, 1792-1801.
Ochocinska, M.J. Hitchcock, P.F., 2007. Dynamic expression of the basic helix-loop-
helix transcription factor neuroD in the rod and cone photoreceptor lineages in the 
retina of the embryonic and larval zebrafish. J. Comp. Neurol. 501, 1-12.
Ohshiro, T., Yagami, T., Zhang, C., Matsuzaki, F., 2000. Role of cortical tumour-
suppressor proteins in  asymmetric division of Drosophila neuroblast. Nature 
408, 593-596.
Ohtoshi, A., Wang, S.W., Maeda, H., Saszik, S.M., Frishman, L J., Klein, W.H., 
Behringer, R.R., 2004. Regulation of retinal cone bipolar cell differentiation and 
photopic vision by the CVC homeobox gene Vsx1. Curr. Biol. 14, 530-536.
Ouyang, Y., Petritsch, C., Wen, H., Jan, L., Jan, Y.N., Lu, B., 2011. Dronc caspase exerts 
a non-apoptotic function to restrain phospho-Numb-induced ectopic neuroblast 
formation in Drosophila. Development 138, 2185-2196.
Pagliarini, R.A., Xu, T., 2003. A genetic screen in Drosophila for metastatic behavior. 
Science 302, 1227-1231.
Parmentier, M.L., Woods, D., Greig, S., Phan, P.G., Radovic, A., Bryant, P., O’Kane, C.J., 
2000. Rapsynoid/partner of inscuteable controls asymmetric division of larval 
neuroblasts in Drosophila. J. Neurosci. 20, RC84.
Parsons, D.W., Jones, S., Zhang, X., Lin, J.C.H., Leary, R.J., Angenendt, P., Mankoo, P., 
Carter, H., Siu, I.M., Gallia, G.L., et al., 2008. An integrated genomic analysis of 
human glioblastoma multiforme. Science 321, 1807-1812.
Passini, M.A., Kurtzman, A.L., Canger, A.K., Asch, W.S., Wray, G.A., Raymond, P.A., 
Schechter, N., 1998. Cloning of zebrafish vsx1: expression of a paired-like 
homeobox gene during CNS development. Dev. Genet. 23, 128-141.
138
Passini, M.A., Levine, E.M., Canger, A.K., Raymond, P.A., Schechter, N., 1997. Vsx-1 
and Vsx-2: differential expression of two paired-like homeobox genes during 
zebrafish and goldfish retinogenesis. J. Comp. Neurol. 388, 495-505.
Peng, C.Y., Manning, L., Albertson, R., Doe, C.Q., 2000. The tumour-suppressor genes 
lgl and dlg regulate basal protein targeting in Drosophila neuroblasts. Nature 408, 
596-600.
Perrière, G., Gouy, M., 1996. WWW-query: an on-line retrieval system for biological 
sequence banks. Biochimie 78, 364-369.
Persson, A.I., Petritsch, C., Swartling, F.J., Itsara, M., Sim, F.J., Auvergne, R., 
Goldenberg, D.D., Vandenberg, S.R., Nguyen, K.N., Yakovenko, S., et al., 2010. 
Non-stem cell origin for oligodendroglioma. Cancer Cell 18, 669-682.
Petersen, P.H., Zou, K., Hwang, J.K., Jan, Y.N., Zhong, W., 2002. Progenitor cell 
maintenance requires numb and numblike during mouse neurogenesis. Nature 
419, 929-934.
Piotrowski, T., Schilling, T.F., Brand, M., Jiang, Y.J., Heisenberg, C.P., Beuchle, D., 
Grandel, H., van Eeden, F.J., Furutani-Seiki, M., Granato, M., et al., 1996. Jaw 
and branchial arch mutants in zebrafish II: anterior arches and cartilage 
differentiation. Development 123, 345-356.
Pollard, S.M., Conti, L., Sun, Y., Goffredo, D., Smith, A., 2006. Adherent neural stem 
(NS) cells from fetal and adult forebrain. Cereb. Cortex 16 Suppl 1, i112-120.
Postlethwait, J.H., Yan, Y.L., Gates, M.A., Horne, S., Amores, A., Brownlie, A., 
Donovan, A., Egan, E.S., Force, A., Gong, Z., et al., 1998. Vertebrate genome 
evolution and the zebrafish gene map. Nat. Genet. 18, 345-349.
Postlethwait, J.H., Amores, A., Cresko, W., Singer, A., Yan, Y.L., 2004. Subfunction 
partitioning, the teleost radiation and the annotation of the human genome. Trends 
Genet. 20, 481-490.
Potter, C.J., Huang, H., Xu, T., 2001. Drosophila Tsc1 functions with Tsc2 to antagonize 
insulin signaling in regulating cell growth, cell proliferation, and organ size. Cell 
105, 357-368.
Rauch, G.J., Lyons, D.A., Middendorf, I., Friedlander, B., Arana, N., Reyes, T., Talbot, 
W.S., 2003. Submission and Curation of Gene Expression Data.
Rawls, J.F., Samuel, B.S., Gordon, J.I., 2004. Gnotobiotic zebrafish reveal evolutionarily 
conserved responses to the gut microbiota. Proc. Natl. Acad. Sci. U.S.A. 101, 
4596-4601.
139
Reilly, K.M., Loisel, D.A., Bronson, R.T., McLaughlin, M.E., Jacks, T., 2000. Nf1;Trp53 
mutant mice develop glioblastoma with evidence of strain-specific effects. Nat. 
Genet. 26, 109-113.
Rinn, J.L., Wang, J.K., Allen, N., Brugmann, S.A., Mikels, A.J., Liu, H., Ridky, T.W., 
Stadler, H.S., Nusse, R., Helms, J.A., et al., 2008. A dermal HOX transcriptional 
program regulates site-specific epidermal fate. Genes Dev. 22, 303-307.
Robinow, S., White, K., 1988. The locus elav of Drosophila melanogaster is expressed in 
neurons at all developmental stages. Dev. Biol. 126, 294-303.
Rodríguez-Marí, A., Yan, Y.L., Bremiller, R.A., Wilson, C., Cañestro, C., Postlethwait, 
J.H., 2005. Characterization and expression pattern of zebrafish Anti-Müllerian 
hormone (Amh) relative to sox9a, sox9b, and cyp19a1a, during gonad 
development. Gene Expr. Patterns 5, 655-667.
Rolls, M.M., Albertson, R., Shih, H.P., Lee, C.Y., Doe, C.Q., 2003. Drosophila aPKC 
regulates cell polarity and cell proliferation in neuroblasts and epithelia. J. Cell 
Biol. 163, 1089-1098.
Roy, P.J., Stuart, J.M., Lund, J., Kim, S.K., 2002. Chromosomal clustering of muscle-
expressed genes in Caenorhabditis elegans. Nature 418, 975-979.
Sabherwal, N., Tsutsui, A., Hodge, S., Wei, J., Chalmers, A.D., Papalopulu, N., 2009. The 
apicobasal polarity kinase aPKC functions as a nuclear determinant and regulates 
cell proliferation and fate during Xenopus primary neurogenesis. Development 
136, 2767-2777.
Saint-Germain, N., Lee, Y.H., Zhang, Y., Sargent, T.D., Saint-Jeannet, J.P., 2004. 
Specification of the otic placode depends on Sox9 function in Xenopus. 
Development 131, 1755-1763.
Saito, T., Ikeda, T., Nakamura, K., Chung, U., Kawaguchi, H., 2007. S100A1 and S100B, 
transcriptional targets of SOX trio, inhibit terminal differentiation of 
chondrocytes. EMBO Rep. 8, 504-509.
Saitou, N., Nei, M., 1987. The neighbor-joining method: a new method for reconstructing 
phylogenetic trees. Mol. Biol. Evol. 4, 406-425.
Sakai, D., Suzuki, T., Osumi, N., Wakamatsu, Y., 2006. Cooperative action of Sox9, 
Snail2 and PKA signaling in early neural crest development. Development 133, 
1323-1333.
Sandell, J.H., Martin, S.C., Heinrich, G., 1994. The development of GABA 
immunoreactivity in the retina of the zebrafish (Brachydanio rerio). J. Comp. 
Neurol. 345, 596-601.
140
De Santa Barbara, P., Bonneaud, N., Boizet, B., Desclozeaux, M., Moniot, B., Sudbeck, 
P., Scherer, G., Poulat, F., Berta, P., 1998. Direct interaction of SRY-related 
protein SOX9 and steroidogenic factor 1 regulates transcription of the human 
anti-Müllerian hormone gene. Mol. Cell. Biol. 18, 6653-6665.
Sauer, C.G., Gehrig, A., Warneke-Wittstock, R., Marquardt, A., Ewing, C.C., Gibson, A., 
Lorenz, B., Jurklies, B., Weber, B.H., 1997. Positional cloning of the gene 
associated with X-linked juvenile retinoschisis. Nat. Genet. 17, 164-170.
Schadt, E.E., Monks, S.A., Friend, S.H., 2003. A new paradigm for drug discovery: 
integrating clinical, genetic, genomic and molecular phenotype data to identify 
drug targets. Biochem. Soc. Trans. 31, 437-443.
Schena, M., Shalon, D., Davis, R.W., Brown, P.O., 1995. Quantitative monitoring of gene 
expression patterns with a complementary DNA microarray. Science 270, 467-
470.
Schilham, M.W., Oosterwegel, M.A., Moerer, P., Ya, J., de Boer, P.A., van de Wetering, 
M., Verbeek, S., Lamers, W.H., Kruisbeek, A.M., Cumano, A., et al., 1996. 
Defects in cardiac outflow tract formation and pro-B-lymphocyte expansion in 
mice lacking Sox-4. Nature 380, 711-714.
Schüller, U., Heine, V.M., Mao, J., Kho, A.T., Dillon, A.K., Han, Y.G., Huillard, E., Sun, 
T., Ligon, A.H., Qian, Y., et al., 2008. Acquisition of granule neuron precursor 
identity is a critical determinant of progenitor cell competence to form Shh-
induced medulloblastoma. Cancer Cell 14, 123-134.
Seimiya, M., Gehring, W.J., 2000. The Drosophila homeobox gene optix is capable of 
inducing ectopic eyes by an eyeless-independent mechanism. Development 127, 
1879-1886.
Sekiya, I., Tsuji, K., Koopman, P., Watanabe, H., Yamada, Y., Shinomiya, K., Nifuji, A., 
Noda, M., 2000. SOX9 enhances aggrecan gene promoter/enhancer activity and is 
up-regulated by retinoic acid in a cartilage-derived cell line, TC6. J. Biol. Chem. 
275, 10738-10744.
Seymour, P.A., Freude, K.K., Tran, M.N., Mayes, E.E., Jensen, J., Kist, R., Scherer, G., 
Sander, M., 2007. SOX9 is required for maintenance of the pancreatic progenitor 
cell pool. Proc. Natl. Acad. Sci. U.S.A. 104, 1865-1870.
Shen, Y., Raymond, P.A., 2004. Zebrafish cone-rod (crx) homeobox gene promotes 
retinogenesis. Dev. Biol. 269, 237-251.
Singh, S.K., Hawkins, C., Clarke, I.D., Squire, J.A., Bayani, J., Hide, T., Henkelman, 
R.M., Cusimano, M.D., Dirks, P.B., 2004. Identification of human brain tumour 
initiating cells. Nature 432, 396-401.
141
Southall, T.D., Brand, A.H., 2009. Neural stem cell transcriptional networks highlight 
genes essential for nervous system development. EMBO J. 28, 3799-3807.
Spana, E.P., Doe, C.Q., 1995. The prospero transcription factor is asymmetrically 
localized to the cell cortex during neuroblast mitosis in Drosophila. Development 
121, 3187-3195.
Spokony, R.F., Aoki, Y., Saint-Germain, N., Magner-Fink, E., Saint-Jeannet, J.P., 2002. 
The transcription factor Sox9 is required for cranial neural crest development in 
Xenopus. Development 129, 421-432.
Staugaitis, S.M., Zerlin, M., Hawkes, R., Levine, J.M., Goldman, J.E., 2001. Aldolase 
C/zebrin II expression in the neonatal rat forebrain reveals cellular heterogeneity 
within the subventricular zone and early astrocyte differentiation. J. Neurosci. 21, 
6195-6205.
Stenkamp, D.L., 2007. Neurogenesis in the fish retina. Int. Rev. Cytol. 259, 173-224.
Stolt, C.C., Lommes, P., Sock, E., Chaboissier, M.C., Schedl, A., Wegner, M., 2003. The 
Sox9 transcription factor determines glial fate choice in the developing spinal 
cord. Genes Dev. 17, 1677-1689.
Sugimoto, M., Kimura, T., Tsumaki, N., Matsui, Y., Nakata, K., Kawahata, H., Yasui, N., 
Kitamura, Y., Nomura, S., Ochi, T., 1998. Differential in situ expression of 
alpha2(XI) collagen mRNA isoforms in the developing mouse. Cell Tissue Res. 
292, 325-332.
Südbeck, P., Schmitz, M.L., Baeuerle, P.A., Scherer, G., 1996. Sex reversal by loss of the 
C-terminal transactivation domain of human SOX9. Nat. Genet. 13, 230-232.
Tanke, H.J., van der Keur, M., 1993. Selection of defined cell types by flow-cytometric 
cell sorting. Trends Biotechnol. 11, 55-62.
Tapon, N., Ito, N., Dickson, B.J., Treisman, J.E., Hariharan, I.K., 2001. The Drosophila 
tuberous sclerosis complex gene homologs restrict cell growth and cell 
proliferation. Cell 105, 345-355.
Taylor, J.S., Braasch, I., Frickey, T., Meyer, A., Van de Peer, Y., 2003. Genome 
duplication, a trait shared by 22000 species of ray-finned fish. Genome Res. 13, 
382-390.
Thisse, B., Thisse, C., 2004. Fast release clones: a high throughput expression analysis.
Thisse, B., Pflumio, S., Fürthauer, M., Loppin, B., Heyer, V., Degrave, A., Woehl, R., 
Lux, A., Steffan, T., Charbonnier, X.Q., et al., 2001. Expression of the zebrafish 
genome during embryogenesis.
142
Thompson, J.D., Gibson, T.J., Plewniak, F., Jeanmougin, F., Higgins, D.G., 1997. The 
CLUSTAL_X windows interface: flexible strategies for multiple sequence 
alignment aided by quality analysis tools. Nucleic Acids Res. 25, 4876-4882.
Tomancak, P., Berman, B.P., Beaton, A., Weiszmann, R., Kwan, E., Hartenstein, V., 
Celniker, S.E., Rubin, G.M., 2007. Global analysis of patterns of gene expression 
during Drosophila embryogenesis. Genome Biol. 8, R145.
Toy, J., Yang, J.M., Leppert, G.S., Sundin, O.H., 1998. The optx2 homeobox gene is 
expressed in early precursors of the eye and activates retina-specific genes. Proc. 
Natl. Acad. Sci. U.S.A. 95, 10643-10648.
Uv, A.E., Harrison, E.J., Bray, S.J., 1997. Tissue-specific splicing and functions of the 
Drosophila transcription factor Grainyhead. Mol. Cell. Biol. 17, 6727-6735.
Verhaak, R.G.W., Hoadley, K.A., Purdom, E., Wang, V., Qi, Y., Wilkerson, M.D., Miller, 
C.R., Ding, L., Golub, T., Mesirov, J.P., et al., 2010. Integrated genomic analysis 
identifies clinically relevant subtypes of glioblastoma characterized by 
abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell 17, 98-110.
Vidal, V.P., Chaboissier, M.C., de Rooij, D.G., Schedl, A., 2001. Sox9 induces testis 
development in XX transgenic mice. Nat. Genet. 28, 216-217.
Viktorin, G., Riebli, N., Popkova, A., Giangrande, A., Reichert, H., 2011. Multipotent 
neural stem cells generate glial cells of the central complex through transit 
amplifying intermediate progenitors in Drosophila brain development. Dev. Biol. 
356, 553-565.
Visel, A., Thaller, C., Eichele, G., 2004. GenePaint.org: an atlas of gene expression 
patterns in the mouse embryo. Nucleic Acids Res. 32, D552-556.
Visvader, J.E., 2011. Cells of origin in cancer. Nature 469, 314-322.
Vogel, W., 1999. Discoidin domain receptors: structural relations and functional 
implications. FASEB J. 13 Suppl, S77-82.
Vogel, W.F., Abdulhussein, R., Ford, C.E., 2006. Sensing extracellular matrix: an update 
on discoidin domain receptor function. Cell. Signal. 18, 1108-1116.
Wagner, T., Wirth, J., Meyer, J., Zabel, B., Held, M., Zimmer, J., Pasantes, J., Bricarelli, 
F.D., Keutel, J., Hustert, E., et al., 1994. Autosomal sex reversal and campomelic 
dysplasia are caused by mutations in and around the SRY-related gene SOX9. Cell 
79, 1111-1120.
Wallace, K., Liu, T.H., Vaessin, H., 2000. The pan-neural bHLH proteins DEADPAN and 
ASENSE regulate mitotic activity and cdk inhibitor dacapo expression in the 
Drosophila larval optic lobes. Genesis 26, 77-85.
143
Wang, H., Somers, G.W., Bashirullah, A., Heberlein, U., Yu, F., Chia, W., 2006. Aurora-A 
acts as a tumor suppressor and regulates self-renewal of Drosophila neuroblasts. 
Genes Dev. 20, 3453-3463.
Wang, Y., Yang, J., Zheng, H., Tomasek, G.J., Zhang, P., McKeever, P.E., Lee, E.Y.H.P., 
Zhu, Y., 2009. Expression of mutant p53 proteins implicates a lineage relationship 
between neural stem cells and malignant astrocytic glioma in a murine model. 
Cancer Cell 15, 514-526.
Wang, Z., Gerstein, M., Snyder, M., 2009. RNA-Seq: a revolutionary tool for 
transcriptomics. Nat. Rev. Genet. 10, 57-63.
Weng, M., Golden, K.L., Lee, C.Y., 2010. dFezf/Earmuff maintains the restricted 
developmental potential of intermediate neural progenitors in Drosophila. Dev. 
Cell 18, 126-135.
Westerfield, M., 2000. The Zebrafish Book: A Guide for the Laboratory Use of Zebrafish 
(Danio rerio) University of Oregon Press, Eugene, OR.
Wilson, M.E., Yang, K.Y., Kalousova, A., Lau, J., Kosaka, Y., Lynn, F.C., Wang, J., 
Mrejen, C., Episkopou, V., Clevers, H.C., et al., 2005. The HMG box transcription 
factor Sox4 contributes to the development of the endocrine pancreas. Diabetes 
54, 3402-3409.
Wu, W.W.H., Wong, J.P., Kast, J., Molday, R.S., 2005. RS1, a discoidin domain-
containing retinal cell adhesion protein associated with X-linked retinoschisis, 
exists as a novel disulfide-linked octamer. J. Biol. Chem. 280, 10721-10730.
Xiong, W.C., Okano, H., Patel, N.H., Blendy, J.A., Montell, C., 1994. repo encodes a 
glial-specific homeo domain protein required in the Drosophila nervous system. 
Genes Dev. 8, 981-994.
Ya, J., Schilham, M.W., de Boer, P.A., Moorman, A.F., Clevers, H., Lamers, W.H., 1998. 
Sox4-deficiency syndrome in mice is an animal model for common trunk. Circ. 
Res. 83, 986-994.
Yan, Y.L., Hatta, K., Riggleman, B., Postlethwait, J.H., 1995. Expression of a type II 
collagen gene in the zebrafish embryonic axis. Dev. Dyn. 203, 363-376.
Yan, Y.L., Miller, C.T., Nissen, R.M., Singer, A., Liu, D., Kirn, A., Draper, B., 
Willoughby, J., Morcos, P.A., Amsterdam, A., et al., 2002. A zebrafish sox9 gene 
required for cartilage morphogenesis. Development 129, 5065-5079.
Yan, Y.L., Willoughby, J., Liu, D., Crump, J.G., Wilson, C., Miller, C.T., Singer, A., 
Kimmel, C., Westerfield, M., Postlethwait, J.H., 2005. A pair of Sox: distinct and 
overlapping functions of zebrafish sox9 co-orthologs in craniofacial and pectoral 
fin development. Development 132, 1069-1083.
144
Yang, Z.J., Ellis, T., Markant, S.L., Read, T.A., Kessler, J.D., Bourboulas, M., Schüller, 
U., Machold, R., Fishell, G., Rowitch, D.H., et al., 2008. Medulloblastoma can be 
initiated by deletion of Patched in lineage-restricted progenitors or stem cells. 
Cancer Cell 14, 135-145.
Zeiner, G.M., Cleary, M.D., Fouts, A.E., Meiring, C.D., Mocarski, E.S., Boothroyd, J.C., 
2008. RNA analysis by biosynthetic tagging using 4-thiouracil and uracil 
phosphoribosyltransferase. Methods Mol. Biol. 419, 135-146.
Zhang, P., Jimenez, S.A., Stokes, D.G., 2003. Regulation of human COL9A1 gene 
expression. Activation of the proximal promoter region by SOX9. J. Biol. Chem. 
278, 117-123.
Zhu, X., Bergles, D.E., Nishiyama, A., 2008. NG2 cells generate both oligodendrocytes 
and gray matter astrocytes. Development 135, 145-157.
Zhu, Y., Romero, M.I., Ghosh, P., Ye, Z., Charnay, P., Rushing, E.J., Marth, J.D., Parada, 
L.F., 2001. Ablation of NF1 function in neurons induces abnormal development of 
cerebral cortex and reactive gliosis in the brain. Genes Dev. 15, 859-876.
Zhu, Y., Guignard, F., Zhao, D., Liu, L., Burns, D.K., Mason, R.P., Messing, A., Parada, 
L.F., 2005. Early inactivation of p53 tumor suppressor gene cooperating with NF1 
loss induces malignant astrocytoma. Cancer Cell 8, 119-130.
Zhuo, L., Theis, M., Alvarez-Maya, I., Brenner, M., Willecke, K., Messing, A., 2001. 
hGFAP-cre transgenic mice for manipulation of glial and neuronal function in 
vivo. Genesis 31, 85-94.
Zong, H., Espinosa, J.S., Su, H.H., Muzumdar, M.D., Luo, L., 2005. Mosaic analysis 
with double markers in mice. Cell 121, 479-492.
145