Transcriptional Code and Disease Map for Adult Retinal Cell Types

r e so u r c e

Transcriptional code and disease map for adult retinal cell types

Sandra Siegert1,7, Erik Cabuy1,7, Brigitte Gross Scherf1, Hubertus Kohler1, Satchidananda Panda2, Yun-Zheng Le3,4, Hans Jörg Fehling5, Dimos Gaidatzis1,6, Michael B Stadler1,6 & Botond Roska1

Brain circuits are assembled from a large variety of morphologically and functionally diverse cell types. It is not known how the intermingled cell types of an individual adult brain region differ in their expressed genomes. Here we describe an atlas of cell type transcriptomes in one brain region, the mouse retina. We found that each adult cell type expressed a specific set of genes, including a unique set of transcription factors, forming a ‘barcode’ for cell identity. Cell type transcriptomes carried enough information to categorize cells into morphological classes and types. Several genes that were specifically expressed in particular retinal circuit elements, such as inhibitory neuron types, are associated with eye diseases. The resource described here allows gene expression to be compared across adult retinal cell types, experimenting with specific transcription factors to differentiate stem or somatic cells to retinal cell types, and predicting cellular targets of newly discovered disease-associated genes.

The brain is composed of many neuronal cell types that are deter- across the retina with a mosaic-like distribution (Supplementary Text mined during development by a dynamic transcriptional program1–5. and Supplementary Fig. 1). The cellular architecture of the retina is In adults, neurons sampled from different brain areas such as the highly conserved among mammals14–17. cortex, cerebellum and hippocampus maintain differences in their By constructing and analyzing a transcriptome atlas for retinal cell expressed genomes6,7. However, the extent to which intermingled cell types, we show that adult retinal cell types have highly diverse gene types within a particular brain region differ in their transcriptomes expression patterns. Our data uncover a transcription factor code is not understood6,8. for the cell types studied. Mapping known disease-associated genes Dissecting cell type transcriptomes within brain areas could also shed to retinal cell types revealed that inhibitory cells, as well as retinal light on the relationship between cell type and disease. Human genetic microglia, are cellular targets of inherited diseases. studies have identified hundreds of gene mutations correlated to diseases © 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature of the nervous system. Although the affected brain regions can be deter- RESULTS mined in human subjects or mutant mice, the expression of disease- Retina cell type transcriptomes associated genes had not been systematically mapped to regional cell We assembled a library of 22 transgenic mouse lines18 in which each line npg types. This mapping is important because understanding disease mecha- had a group of retinal cells marked with fluorescent proteins (Fig. 1b, nisms, as well the design of therapeutic strategies, may vary according Supplementary Table 1 and Supplementary Text). We generated the to how widely the disease-associated gene is expressed across cell types. library with the goal of having some mouse lines in which single retinal Recent studies have demonstrated the feasibility of reprogramming stem cell types and others in which combinations of types from a single class cells and somatic cells to become neuronal cell types by expressing cell were labeled. The library had mouse lines with labeled cells represent- type–specific transcription factors9–11. Knowing these factors and ing each of the six retinal cell classes. Retinal cells were character- having reference transcriptomes for the different neuronal cell types of ized by physiological recording and immunohistochemical staining a brain region would facilitate cell-type engineering. (Supplementary Table 1 and Supplementary Figs. 2–6). We isolated The retina offers opportunities to investigate the relationship 200 fluorescent protein–labeled retinal cells (“cell groups”) from at least between the cellular elements of neuronal circuits and the genes that three different mice of each mouse line by fluorescence-activated cell they express12,13. On the basis of morphological and physiological sorting5,19,20 (Supplementary Figs. 7–9). The transcripts of each cell criteria, cells can be grouped into six classes: photoreceptor, hori- group of these biological triplicates were independently amplified in zontal, bipolar, amacrine, ganglion and non-neuronal cells2 (Fig. 1a). batches. Each batch contained an internal control cell group from the Each class can be further subdivided into cell types; these spread Arc line (Supplementary Fig. 7 and Supplementary Text). The Arc

1Neural Circuit Laboratories, Friedrich Miescher Institute for Biomedical Research, Basel, Switzerland. 2Regulatory Biology Laboratory, Salk Institute for Biological Studies, La Jolla, California, USA. 3Department of Medicine, University of Oklahoma Health Sciences Center, Oklahoma City, Oklahoma, USA. 4Harold Hamm Oklahoma Diabetes Center, University of Oklahoma Health Sciences Center, Oklahoma City, Oklahoma, USA. 5Institute of Immunology, University Clinics Ulm, Ulm, Germany. 6Swiss Institute of Bioinformatics, Basel, Switzerland. 7Present addresses: The Picower Institute, Massachusetts Institute for Technology, Cambridge, Massachusetts, USA (S.S.), Reliable Cancer Therapies, Energy-based Therapies, Strombeek-Bever, Belgium (E.C.). Correspondence should be addressed to B.R. ([email protected]). Received 18 August 2011; accepted 20 December 2011; published online 22 January 2012; doi:10.1038/nn.3032

nature NEUROSCIENCE VOLUME 15 | NUMBER 3 | MARCH 2012 487 r e so u r c e

cell group is a GABAergic, ON amacrine cell population18. A reason Linear RNA amplification is desirable for quantitative analysis of RNA we used the Arc cells as internal control was that we established the content. To test whether amplifications were indeed linear, we exam- protocols for dissociation, sorting and amplification with this line. ined the relationship between gene expression values and the amounts

a b Photoreceptors Horizontal cells b2 Chrnb4 d4 Gja10 Photo- Outer nuclear layer receptors (ONL) Horizontal cells Bipolar Inner nuclear layer ONL cells (INL)

Amacrine cells Glial INL cells

Ganglion Ganglion cell layer GCL cells (GCL)

Bipolar cells Amacrine cells mGluR6 Kcng4 Arc Igfbp2 Rgs5 Crh

ONL

INL

GCL

Pcp2 Lhx4 ChAT Chrna3 Fam81a Fbxo32 Ier5

ONL

INL

GCL

Ganglion cells Microglia

ONL INL npg

INL

GCL GCL

Mean Mean Gene R Gene R c expression expression Fabp7 256 0.997 Htr2b 120 0.992 2.0 Olfr1372-ps1 199 0.989 Rprml 81 0.989 Pde6c 314 0.989 Frmd7 53 0.986 Gnat2 1821 0.988 1.5 P2rx2 53 0.982 Mogat1 569 0.984 Pxmp2 72 0.980 1.5 Osgep 469 0.979 Slc18a3 1615 0.978 Gulo 44 0.977 Pomc 116 0.971 Otop3 45 0.976 Cmtm8 396 0.960 Ppm1j 293 0.976 1.0 Gabrd 189 0.949 1.0 Clca3 43 0.975 Tgfb3 108 0.941

Expression value/ 0.5

mean expression value 0.5

Chrnb4 ChAT 0 0 0 50 100 150 200 200 150 100 50 0 Number of cells in the mixture Figure 1 Retinal inventory for cell type comparative transcriptome analysis. (a) Schematic overview of the retina. (b) Immunohistochemical staining of vibratome sections from the retinas of mice with fluorescent protein expression in cell groups. Blue, DAPI; green, fluorescently labeled cells; purple, the stratum marker choline acetyltransferase (ChAT; arrows). Scale bars, 10 µm. (c) Expression value changes of ten cone photoreceptor (Chrnb4)-specific and starburst cell (ChAT)-specific genes for graduated variations in ratios of the two cell groups in a mixture.

488 VOLUME 15 | NUMBER 3 | MARCH 2012 nature NEUROSCIENCE r e so u r c e

of RNA in a cell group (see Supplementary Text). We mixed varying The linearity and repeatability of amplification allows quantitative ratios of green fluorescent protein (GFP)-positive cones from the Chrnb4 analysis of the cell group transcriptomes. mouse line and red fluorescent protein (RFP)-positive starburst amacrine cells from the ChAT line to yield a total of 200 cells (Supplementary A genetic barcode for retinal cell types Fig. 10). Gene expression in the mixtures was analyzed by exploiting We then asked whether there are only graded or combinatorial differ- the finding that both cones and starburst cells each express ~20 genes ences between cell type transcriptomes or whether a set of genes exists at least threefold higher than in any other cell group. Both for genes for each cell type that is only expressed in that type. First, we organ- enriched in cones and those enriched in starburst cells, gene expression ized the cell groups into the six cell classes and ranked the genes for values increased linearly with an increase in the number of cones or each class according to a specificity ratio (s.r.); that is, the ratio of the starburst cells in the mixtures (Fig. 1c). The linear correlation coefficients mean expression within the class compared with the maximal expres- were independent of the extent to which genes were expressed in sion across all other classes. For each class, we found transcripts that the pure cone and starburst cell groups (Supplementary Fig. 10). were enriched in that class (Fig. 2a). To quantify class enrichment, This suggests that the expression values a Photo- Bipolar Amacrine Ganglion b obtained were proportional to the RNA con- receptor cell cell cell 183 Pde6b tent. As a further independent test of propor- Rcvrn Krt18 tionality, we estimated the cell type composition Pde6g Gpr65 Fscn2 60 of three independent mixtures. This estimate Fabp4 Sag was based on a linear algorithm that compared 2610034M16Rik 1700008G05Rik Rtbdn cone-specific gene expression in a mixture and C79127 4930430E16Rik 30 in a pure cone sample (Supplementary Fig. 10). Aipl1 Specificity ratio Rdh12 Mosc1 The mean error was 12 ± 3% (s.d.) when ten Mpp4 20 Guca1a genes were used to estimate the composition Nrn1 Mosc1 10 7 Gabrr2 Slc32a1 5-fold of the mixture (Supplementary Fig. 10). Srd5a2l2 3 Wfdc10 0 2.5-fold Lhx1 The mean and the s.d. of the prediction error Onecut1 Photo- Horizontal Bipolar Amacrine Ganglion Microglia ENSMUSG00000054450 receptor cell cell cell cell decreased with an increase in the number of Gja10 Ddo cone-specific genes used for the prediction Sucnr1 Arhgap15 (Supplementary Fig. 10 and Supplementary Sept4 Ndrg1 Text). The linearity of amplification allowed Mtnr1a c Tpm3 R = –0.82 (one outlier removed) Angpt2 us to separate the transcriptomes of cell types 432 Photoreceptor Gabrr2 Horizontal cell when a cell group contained more than one cell Tmem215 Bipolar cell Gabrr1 40 Amacrine cell type, as well as to compensate for transcrip- Accn3 Cacna2d4 Ganglion cell tome contamination by rod transcriptomes Cabp5 Microglia Traf5 30 (Supplementary Text). Compensation for Gstp2 Otx2 rod contamination was necessary because we BC030499 Khdrbs3 20 Nrn1l detected rod-specific genes in the transcrip- Lrtm1 Scgn 10 tome of every cell group (Supplementary Frmd3 © 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature 7

Fig. 10). This is probably because ~80% of A Slc32a1 21 Tmem41a Number of cell class−specific genes 0 mouse retinal cells are rods . We assessed Tcfap2a Gad1 1 2 7 13 17 repeatability by correlating the expression values Id4 GABAergic Diversity of cell classes npg across the internal-control Arc cell groups and Myf6 ENSMUSG00000070886 across all biological triplicates (Supplementary Necab3 6330527O06Rik Plekhf1 Fig. 11). The high correlation (0.97 ± 0.002, Fgfr4 Slc6a9 n = 36 pairs for Arc cell groups and 0.97 ± 0.003, Cabp1

Glycinergic Htra1 n = 78 pairs for the triplicates, s.e.m.) suggested 2310040A07Rik Ncald that cell sorting and transcript amplification were Traip 4930444P10Rik reliable procedures (Supplementary Fig. 11). Nrn1 S100a10 Rbpms Fbxo2 Figure 2 Transcriptome comparisons of cell Nefl 1 b2.1st Normalized Speci city Tppp3 Photo- 2 b2.2nd expression ratio (s.r.) groups that belong to a cell class. (a) Cluster b2.single EG435376 receptor 3 value heat map of all cell classes. Amacrine cells (A) Sncg 4 Chrnb4 >15 Rbpms2 Horizontal 5 d4 1 H Gja10 are further subdivided into GABA- and Rlbp1l2 cell ≤ 15 Slc17a6 1 mGluR6 0.8 Bipolar 2 Kcng4 glycine-releasing. Purple indicates high gene Chrnb3 10 BC089491 cell 3 Pcp2 0.6 4 Lhx4 expression. P-value and specificity ratio (s.r.) Cpne9 0.4 8 Tubb3 1 Arc.1st are color-coded as indicated in the figure. 2 Arc.2nd 0.2 6 Gpr65 3 Arc.3rd (b) Specificity ratios for genes in a cell class Ccl3 4 Igfbp2 0 4 Chi3l3 Amacrine 5 Rgs5 3 that passed a threshold of 2.5 (open circles). C3ar1 cell 6 Crh Cd48 7 ChAT Signi cance 2 The gene with the highest specificity ratio is Ncf4 8 Chrna3 P < 9 Fam81a Tlr7 0.001 1.75 named. Specificity ratio values with a cross Gm885 10 Fbxo32 showed expression below the threshold. Ccl4 11 Ier5 0.01 1.50 Tlr2 1 PV Cd86 Ganglion 2 Drd4 0.05 1.25 (c) Correlation analysis of the number of cell 3 Grik4 Fcgr4 cell Not 4 Opn4 1.10 class-specific genes against number of cells AF251705 significant P2ry6 Microglia 1 Csf2rb2 within a cell class in our data set. For the 6330407A03Rik 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 9 1011 1 2 3 4 1

P correlation analysis, microglia is the outlier. s.r.

nature NEUROSCIENCE VOLUME 15 | NUMBER 3 | MARCH 2012 489 r e so u r c e

we defined a term, “class-specific genes.” A gene was classified as expressed the Rhox4 family of transcription factors with a specificity class-specific if its s.r. was greater than 2.5 and the difference between ratio of at least 2.7: Rhox4a (s.r. = 3.8), Rhox4b (s.r. = 4.3), Rhox4c expression inside and outside the class was significant (P < 0.05). (s.r. = 3.7), Rhox4f (s.r. = 4.1) and Rhox4g (s.r. = 2.7). Crh amacrine Both the magnitude of the s.r. and the number of class-specific genes cells specifically expressed Ascl1 (s.r. = 3.0); starburst amacrine cells, varied markedly across cell classes, ranging from 3 to 183 and 0 to Sox2 (s.r. = 3.7); Opn4 ganglion cells, Eomes (s.r. = 2.7); microglia, 432, respectively (Fig. 2b,c). Myc (s.r. = 9.7), Sfpi1 (s.r. = 9) and Hhex (s.r. = 7.9). To analyze the source of this variability, we examined the relation- Furthermore, we observed transcription factors that were enriched ship between the number of cell class–specific genes and the diversity in cell classes or in combinations of classes (Supplementary Fig. 18): of the classes. The diversity of a class was defined as the estimated Rax (s.r. = 4.5) was expressed in photoreceptors, Vsx2 (s.r. = 1.3) in number of cell types in the class across our mouse library. We found a bipolar cells, and Otx2 (s.r. = 1.7) in photoreceptors and bipolar cells. negative correlation between diversity and the number of class-specific To determine whether transcription factor expression pattern alone genes (Fig. 2c). For example, horizontal cells and non-neuronal cells is enough to categorize cells into the corresponding morphological each consisted of a single cell type and had 42 and 432 class-specific classes and types, we performed hierarchical clustering using only the genes, respectively. In contrast, amacrine cells consisted of many cell list of expression values for all transcription factors. Transcription types and lacked class-specific genes. This negative correlation sug- factor–based clustering separated cell groups into relevant biological gests that the less diverse a cell population, the higher the number of classes, and it also separated most cell groups into the correct biologi- genes expressed only in that population. Therefore, we compared the cal triplicates (Fig. 4c). transcriptomes of ‘pure’ cell groups that contained only one or two cell types (Fig. 3a,b and Supplementary Figs. 12–14). We found a Cell type–specific pathways median of 20 group-specific genes (s.r. > 2.5, P < 0.05) in the pure Our cell-type comparative gene expression atlas may expose evolution- groups (Supplementary Fig. 15). Although we did not find any class- ary conserved, cell type–specific pathways. Melanopsin-containing specific genes for the amacrine cell class, pure amacrine cell groups ganglion cells, uniquely labeled in Opn4-mice, are thought to be had a median of 17 specific genes (see Supplementary Text). To ancient vertebrate photoreceptors25,26 and may have a phototrans- validate cell type–specific expression patterns, we performed in situ duction cascade similar to that of rhabdomeric photoreceptors of hybridization in combination with GFP staining. The in situ probes invertebrates27. We searched the transcriptome of the Opn4 cell group colocalized with the GFP-positive cells (Supplementary Fig. 16). for vertebrate homologs of Drosophila melanogaster phototransduc- In addition, we compared cell type–specific genes proposed in the tion genes28 (Supplementary Fig. 19). Gna14, Dgka, Dgkg and Cds2 literature with our data set and found that the majority of the published were enriched in the Opn4 cell groups; however, expression of the pre- cell type–specific genes were also cell type–specifically expressed in dicted guanine nucleotide binding protein (Gnaq)29,30 was not found, our data set (Supplementary Text). These results suggest that a retinal suggesting a different G-protein partner of Opn4 (Supplementary cell type can be identified by many cell type–specific transcripts. The Fig. 19). In Drosophila, transient receptor potential (Trp) channels collection of such transcripts can be thought of as a genetic barcode provide the photocurrent for vision28. We detected Trpc3, Trpc6 and for retinal cell types. Trpc7 (ref. 31) in the Opn4 cell group. Are the transcriptome differences robust enough to allow categori- In addition to the phototransduction pathway, we investigated zation of cell groups into the corresponding, morphologically defined more than 150 molecular pathways for over-representation in retinal cell classes and biological triplicates in an unbiased way? To address cell groups. With the exception of microglia, we found no pathway © 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature this question, we performed divisive hierarchical clustering of the enrichment in retinal cells (data not shown). However, when the dif- measured transcriptomes (Supplementary Text). For each gene, we ferent channel and receptor genes were mapped to cell groups, each calculated three measures of the variability of expression across cell cell group was found to have a distinct channel-receptor fingerprint npg groups: variance, entropy and the analysis of variance (ANOVA) (Supplementary Fig. 20). Therefore, this analysis of various gene P-value (see Supplementary Text). The 1,000 most heterogeneously groups suggests a sophisticated fine-tuning of excitability to a par- expressed genes according to each measure were selected as inputs ticular task of a cell type in the network and, with regard to photo to the clustering algorithm. Independent of the measure of variabil- transduction, supports a view that melanopsin-containing cells have ity, microglia were separated from the remaining retinal cells first a phototransduction cascade similar to that of invertebrates. (Fig. 3c and Supplementary Fig. 17); followed by the photoreceptors, with rods and cones appearing on different branches; horizontal cells; Disease map for retinal cell types bipolar cells; and a large group containing amacrine and ganglion The existence of many adult cell type–specific genes, and the special- cells. With few exceptions, the final cluster on each branch was the ized role of each cell type in retinal image processing, raises the ques- biological triplicate of each cell group. Thus, the cell group transcription of whether mutated genes observed in hereditary eye diseases tomes carry enough information to separate them into biologically are expressed in certain cell types. Although several gene mutations relevant cell classes and types. that affect photoreceptors and bipolar cells have been described, it is not known whether inhibitory amacrine cells are disease targets. Transcription factor code We mapped all of the ~200 known genes whose mutation has been Transcription factors are known to be key for the specification of neu- shown to cause, or correlate with, eye diseases to the 22 retinal groups ronal cell fate during development22–24. We asked whether transcrip- of this study (Fig. 5 and Supplementary Fig. 21). As expected, most tion factors are part of the adult cell type barcode. Almost all pure cell of the disease-associated genes were found to be expressed only in groups expressed at least one transcription factor specifically (Fig. 4): photoreceptors (n = 48) or were expressed in many cell types in the adult rod photoreceptors expressed Nr2e3 (s.r. = 114), Nrl (s.r. = 4.5) retina (n = 112). However, 23 gene transcripts were enriched in non- and Csda (s.r. = 3.6); cones, En2 (s.r. = 2.8); horizontal cells, Lhx1 photoreceptor cell types (s.r. > 1.75). (s.r. = 11.7) and Onecut1 (s.r. = 8.5); rod bipolar cells, Sebox (s.r. = Notably, some inner retinal cells such as starburst amacrine cells, as 5.7). Rgs5 and Crh cell groups (Crh cells are a subset of Rgs5 cells) well as the non-neuronal microglia cells, specifically expressed several

490 VOLUME 15 | NUMBER 3 | MARCH 2012 nature NEUROSCIENCE r e so u r c e

disease-associated genes. Starburst amacrine cells are part of the idiopathic congenital nystagmus34, was exclusively expressed in our direction-selective circuits in the retina32 and control reflex eye move- data set in those starburst cells. Clrn1 (s.r. = 2.8), which is mutated in ments33. Frmd7 (s.r. = 4.8), a gene implicated in the eye motion disease type-3 Usher syndrome35, was also highly enriched in starburst cells.

a Photo- Bipolar Amacrine Ganglion b receptor cell cell cell Pde6b Pde6b 328 Nr2e3 Cnga1 114 Pde6a Gpr65 A930006D01Rik 62 Pde6c Rp1h Sh2d1a Guca1b Faim OTTMUSG00000007987 Gnb1 Fscn2 Reep6 Specifitity ratio 30 Nxnl1 Slc18a3 Nrl Mosc1 Crygd Rpe65 P s.r. 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 91011 1 2 3 4 1 20 Wfdc1 Chrna3 11100 ENSMUS Photo- Bipolar Amacrine Ganglion Otor17l16Rik G000000 Opn4 receptor cell cell cell 10 Lect1 Sebox Myf6 66798 7 Rhox4b Pde6c Gcnt3 Cntn6Pnlip 5-fold 3 2.5-fold Opn1sw 0 Fabp7 Gnat2 b2 Crh Ier5 PV Olfr1372-ps1 Pcp2 Lhx4 Rgs5 Drd4 Gja10 Kcng4 ChAT Grik4 Opn4 Opn1mw mGluR6 Chrna3Fam81aFbxo32 Csf2rb2 Pde6h Chrnb4-d4 Arc-Igfbp2 Clca3 Agr2 Arr3 Gulo c Height 0 0.1 0.2 0.3 0.4 Otop3 Ppm1j Csf2rb2_a Mogat1 Csf2rb2_b Csf2rb2_c Gja10_c Hspb6 Gja10_a 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 9 1011 1 2 3 4 1 Gja10_b P s.r. Lhx4_a Kcng4_a Kcng4_b Photo- Bipolar Amacrine Ganglion Kcng4_c Lhx4_b Lhx4_c receptor cell cell cell Pcp2_a Pcp2_b Pcp2_c Slc18a3 mGluR6_a mGluR6_b Igfbp6 mGluR6_c ChAT_c Slc10a4 ChAT_a ChAT_b P2rx2 Chrna3_b Chrna3_a Gabrd Chrna3_c Arc.1st_a Cmtm8 Arc.1st_b Arc.1st_c Pomc Fam81a_c Fbxo32_a Dok5 Fbxo32_b Fbxo32_c Kcnf1 Fam81a_a Fam81a_b Frmd7 Ier5_c Ier5_a Sftpc Ier5_b Drd4_a Rprml Crh_b Crh_c

© 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature Crh_a Sox2 Rgs5_c Rgs5_a Htr2b Rgs5_b Igfbp2_c Pxmp2 Igfbp2_a 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 9 1011 1 2 3 4 1 Igfbp2_b P s.r. Arc.2nd_b Photoreceptor Arc.2nd_a Arc.2nd_c Photo- Bipolar Amacrine Ganglion Arc.3rd_a Arc.3rd_b Horizontal cell npg receptor cell cell cell Arc.3rd_c Drd4_b Drd4_c Opn4 PV_a Bipolar cell PV_b Nppb PV_c Opn4_a Amacrine cell Adcyap1 Opn4_b Opn4_c 1700011I03Rik Grik4_b Grik4_a Ganglion cell Prph Grik4_c b2.2nd_c b2.2nd_a Ctxn3 b2.2nd_b Microglia b2.1st_a Cd24a b2.1st_b b2.1st_c Prkcq b2.single_b b2.single_a Eomes b2.single_c d4_a Genes: ANOVA P-value Cdkn1c d4_b d4_c Cluster method: median Gm687 Chrnb4_b Chrnb4_a Trhde Chrnb4_c Distance: correlation Igf1 Rbms3 Gm527 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 9 10 11 1 2 3 4 1 P s.r.

Photo-Horizontal Bipolar Amacrine Ganglion Micro- receptor cell cell cell cell glia 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 91011 1 2 3 4 1 d4 PV Crh Ier5 Drd4 Lhx4 Pcp2 Rgs5 Grik4 ChAT Opn4 Gja10 b2.1st Igfbp2 Kcng4 b2.2nd Arc.1st Arc.3rd Chrnb4 Chrna3 Fbxo32 Arc.2nd Csf2rb2 Fam81a mGluR6 b2.single

Normalized Signi cance expression value 1 0.8 0.6 0.4 0.2 0 P < 0.001 0.01 0.05 Not significant Speci city ratio (s.r.) >15 ≤ 15 10 8 6 4 3 2 1.75 1.50 1.25 1.10 Figure 3 Transcriptome comparisons of cell groups. (a) Cluster heat map of four cell groups. Selected cell groups are labeled with a gray bar. Purple indicates high gene expression. P-value and specificity ratio (s.r.) are color-coded as indicated in the figure. (b) Specificity ratios for genes in a cell group that passed a threshold of 2.5 (open circles). The gene with the highest specificity ratio is named. Specificity ratio values with a cross showed expression below the threshold. (c) Hierarchical clustering of the cell groups for the top 1,000 genes that passed the ANOVA-test. “_a”, “_b”, and “_c” indicate the first, second, or third cell group of a biological triplicate, respectively.

nature NEUROSCIENCE VOLUME 15 | NUMBER 3 | MARCH 2012 491 r e so u r c e

We found microglia, a retinal macrophage, to be enriched in transcripts Müller cells and astrocytes, and it is likely that these cell types also associated with age-related macular degeneration (Cx3cr1 (ref. 36), specifically express a number of disease genes. Finally, given that retinal s.r. = 8.6; Cfh (ref. 37), s.r. = 2.5; Apoe (ref. 38), s.r. = 3.0) and glau- abnormalities have been observed in Alzheimer’s disease, it is note- coma (Tlr4 (ref. 39), s.r. = 3.3; Tnf (ref. 40), s.r. = 2.9; Supplementary worthy that one of the Alzheimer-associated genes, CD33 (refs. 41,42; Fig. 21). Note that the current transcriptome atlas does not contain s.r. = 13.8), is uniquely expressed in our data set in retinal microglia.

a Photo- Horizontal Bipolar Ganglion Micro- b receptor cell cell Amacrine cell cell glia

Nr2e3 Nrl Hmgb2 Csda Nr2e3 Lhx1 Sap30 Polr3k E2f5 100 Hist1h1c 800 En2 Hopx 80 Rxrg Lhx1 600 Onecut1 Rpa1 60 Vsx1 Sebox 400 Prdm5 40 Lhx4 Irx3 Nr4a2 200 Tcfap2a 20 Expression value Fezf2 Rhox4b Rhox4f 0 0 Rhox4c Rhox4a 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 Rhox4e Rhox4g Tox3 Ascl1 Sebox Lhx4 Gbx2 Sox2 Neurog2 120 Neurod6 60 Tcf4 Myf6 100 Pou4f1 50 Isl2 Pou3f2 80 Tcfap2d 40 Irx6 Pou6f2 60 30 Eomes Msc Sfpi1 40 20 Myc Hhex 20 10 Atf3 Batf3 Irf8 0 0 Lmo2 Mafb 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 Batf Aim2 Cebpa Runx1 Nfe2l2 Sox2 Myc P 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 9 10 11 1 2 3 4 1 s.r. 100 150 80 Height c 0 0.05 0.10 0.15 0.20 60 100

b2.1st_b b2.2nd_c 40 b2.single_b b2.1st_a 50

© 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature b2.single_c b2.1st_c b2.single_a 20 b2.2nd_a b2.2nd_b Chrnb4_a Chrnb4_b 0 0 Chrnb4_c d4_c d4_a 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 d4_b Csf2rb2_a npg Csf2rb2_b Csf2rb2_c Chrna3_a ChAT_b ChAT_a ChAT_c mGluR6_a Pcp2_c Photo- Horizontal Bipolar Amacrine Ganglion Micro- Pcp2_a Pcp2_b receptor cell cell cell cell glia Kcng4_c Kcng4_a Kcng4_b mGluR6_b 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 91011 1 2 3 4 1 mGluR6_c Lhx4_a

Lhx4_b d4 PV Crh

Lhx4_c Ier5 Drd4 Lhx4 Pcp2

Ier5_a Rgs5 Grik4 Opn4 ChAT Gja10 b2.1st Igfbp2 Ier5_b Kcng4 b2.2nd Arc.1st Arc.3rd Chrnb4 Chrna3

Gja10_a Fbxo32 Arc.2nd Csf2rb2 mGluR6 Fam81a

Gja10_b b2.single Gja10_c Crh_a PV_a PV_b PV_c Drd4_c Drd4_b Grik4_b Normalized Signi cance Grik4_a Grik4_c expression value 1 0.8 0.6 0.4 0.2 0 P < 0.001 0.01 0.05 Not Opn4_a Opn4_b significant Opn4_c Arc.1st_a Arc.1st_b Speci city Arc.1st_c Rgs5_a >15 ≤ 15 10 8 6 4 3 2 1.75 1.50 1.25 1.10 Rgs5_b ratio (s.r.) Rgs5_c Arc.2nd_c Arc.2nd_a Arc.2nd_b Arc.3rd_b Arc.3rd_a Igfbp2_b Igfbp2_a Arc.3rd_c Igfbp2_c Fam81a_c Fbxo32_a Fbxo32_b Fbxo32_c Fam81a_a Fam81a_b Ier5_c Chrna3_b Chrna3_c Distance: correlation Drd4_a Crh_b Crh_c Cluster method: median Figure 4 Cell group–specific transcription factors. (a) Cluster heat map of transcription factors that show expression in distinct cell groups. Purple indicates high gene expression. P-value and specificity ratio (s.r.) are color-coded as indicated in the figure. (b) Bar charts of individual transcription factors. The color-coded bar represents the cell group in which expression is highest. Error bars show s.e.m. (c) Hierarchical clustering of the cell groups based only on transcription factors. “_a”, “_b”, and “_c” indicate the first, second, or third cell group of a biological triplicate, respectively.

492 VOLUME 15 | NUMBER 3 | MARCH 2012 nature NEUROSCIENCE r e so u r c e

a b Photo- Horizontal Bipolar Amacrine Ganglion Micro-

receptor cell cell cell cell glia 700 2,500 Bbs12 Unc119 Parl 600 Pgk1 2,000 Rd3 500 Unc119 Sag 400 1,500 Guca1a 300 Rom1 1,000 Prph2 Rpgrip1 200 Pde6g Expression value 500 Tulp1 100 Oat Col2a1 0 0 Prom1 Ttc8 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 1 2 3 4 5 H1 2 3 4 1 2 3 4 5 6 7 8 9 10111 2 3 4 1 Bbs5 Pde6a Cacna2d4 Ndp Rdh12 120 60 Nrl Hspa1a 100 50 Bbs12 Rs1 Bbs7 80 40 Optn Fscn2 60 30 Nr2e3 Pde6b Guca1b 40 20 Cnga1 Gnat1 20 10 Rho Grm6 Trpm1 0 0 Cacna1f 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 Otx2 Cln3 Crx Frmd7 Clrn1 Grk1 Bbs4 Bbs2 60 30 Rbp3 Opn1mw 50 25 Opn1sw Gnat2 Pcdh15 40 20 Pde6c Cngb3 30 15 Adrb1 Bbs9 20 10 Aipl1 Arl6 Efemp1 10 5 Ndp Rbp4 0 0 Vegfc Clrn1 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 41 Frmd7 Cyp1b1 Rgr Adrb2 Rlbp1 Kcnj13 200 Rdh5 Rpe65 60 Rgr Tnf 150 50 Apoe Cfh 40 Tap1 100 Cx3cr1 30 Adrb2 Tlr4 Mertk 20 50 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6* 7 8 9 10 11 1 2 3 4 1 10

0 0 Photo- Horizontal Bipolar Amacrine Ganglion Micro- 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 receptor cell cell cell cell glia © 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature Normalized Cd33 1 2 3 4 5 H 1 2 3 4 1 2 3 4 5 6 7 8 9 10 11 1 2 3 4 1 expression 1 0.8 0.6 0.4 0.2 0 value 200 d4 PV Crh Ier5 Drd4 Lhx4 Pcp2 Rgs5 Grik4 Opn4 150 ChAT Gja10 b2.1st Igfbp2 Kcng4 b2.2nd Arc.1st Arc.3rd Chrna3 Chrnb4 Fbxo32 Arc.2nd npg Csf2rb2 mGluR6 Fam81a b2.single 100

0 1 2 3 4 5H1 2 3 4 1 2 3 4 5 6 7 8 910111 2 3 4 1 Figure 5 Retinal disease-associated genes in adult cell types. (a) Cluster heat map of disease-associated genes that showed expression values >20 and passed the threshold of 1.75. Purple indicates high gene expression. The Crh cell group (no. 6 in amacrine cells; asterisk) has been removed from the analysis, and the plot area is white (see Supplementary Text). (b) Bar charts of individual disease genes. The color-coded bar represents the cell type in which expression is highest. Error bars, s.e.m.

DISCUSSION data/index.php), where the expression values for any chosen gene The analysis of our cell-type comparative transcriptome atlas suggests present on the gene array can be compared across cell types. This that genetic differences between adult retinal cell types are not only gene-by-gene search can be useful, for example, when a retinal pheno graded or combinatorial but rather that cell types also harbor many type is observed in a knockout mouse line and the cell type that is cell type–specific genes. Whether other individual adult brain regions responsible for the functional change is not known. Whole genome are built in a similar, ‘retina like’ fashion, where the genetic diversity or exome sequencing of individuals with a given disease is rapidly is very high or, alternatively, are composed of cell types with more providing new disease-associated genes, and the gene-by-gene search graded and combinatorial genetic differences is not yet known. may allow identification of cellular disease targets in the retina. Our data can be useful in the following ways. First, they allow Second, the atlas enables correlation of cellular physiology with quantitative comparison of gene expression across different cell types gene expression. The cell types in the mouse lines described here in the adult retina (Supplementary Figs. 12–23). To make this pro can be targeted for physiological recordings18. If a physiologically cess more accessible, we created a website (http://www.fmi.ch/roska. relevant current is isolated and characterized, the transcriptome of

nature NEUROSCIENCE VOLUME 15 | NUMBER 3 | MARCH 2012 493 r e so u r c e

the recorded cell type can help to narrow down the genes responsible sorting. S.P. provided the Opn4-Cre mouse. Y.-Z.L. provided b2-Cre and d4-Cre for the current. This could substantially reduce the time required to mice. H.J.F. provided the Rosa26-LSL-RFP reporter mouse. D.G. normalized exon find the relevant knockout mouse lines for loss-of-function studies. array data. M.B.S. assisted with gene array data analysis and provided scripts for pairwise correlation analysis and hierarchical clustering. B.R. helped with data In this respect, the Cre recombinase–expressing mouse lines (Online analysis; designed experiments; and wrote the manuscript. Methods and Supplementary Table 1) used in our study could be of use, since conditional virus-mediated or plasmid-mediated knock- COMPETING FINANCIAL INTERESTS down is fast and efficient. A particularly productive avenue for cor- The authors declare no competing financial interests. relating physiology with gene expression would be the physiological characterization of cell types in cell type–specific transcription factor Published online at http://www.nature.com/natureneuroscience/. knockouts. Cell type–specific transcription factors may ‘equip’ retinal Reprints and permissions information is available online at http://www.nature.com/ cells with particular physiological properties. reprints/index.html. Third, in recent years notable progress has been made in creating retinal cell types, or even whole retinas43, from embryonic stem cells, progenitor cells44 and induced pluripotent stem cells43,45. The genetic 1. Kim, D.S. et al. Identification of molecular markers of bipolar cells in the murine retina. J. Comp. Neurol. 507, 1795–1810 (2008). identity of engineered cells can be correlated with those of the corre- 2. Trimarchi, J.M. et al. Molecular heterogeneity of developing retinal ganglion and sponding natural adult cells described here. Such ‘genetic benchmarking’ amacrine cells revealed through single cell gene expression profiling. J. Comp. Neurol. 502, 1047–1065 (2007). could provide a rational endpoint for cell engineering. Cell type–specific 3. Kay, J.N., Voinescu, P.E., Chu, M.W. & Sanes, J.R. Neurod6 expression defines new transcription factors, conversely, could offer opportunities for develop- retinal amacrine cell subtypes and regulates their fate. Nat Neurosci. 14, 965–972 ing new cell lines for bipolar cells or interneurons derived from somatic (2011). 4. Cherry, T.J., Trimarchi, J.M., Stadler, M.B. & Cepko, C.L. Development and cells, embryonic stem cells or induced pluripotent stem cells. diversification of retinal amacrine interneurons at single cell resolution. Proc. Natl. Fourth, as retinal gene therapy becomes possible, the targeting of Acad. Sci. USA 106, 9495–9500 (2009). genetic material to specific retinal cell types becomes an important 5. Arlotta, P. et al. Neuronal subtype-specific genes that control corticospinal motor neuron development in vivo. Neuron 45, 207–221 (2005). issue. Conserved genomic regions around cell type–specific genes can 6. Sugino, K. et al. Molecular taxonomy of major neuronal classes in the adult mouse be a first starting point for searching for cell type–specific enhanc- forebrain. Nat. Neurosci. 9, 99–107 (2006). 7. Doyle, J.P. et al. Application of a translational profiling approach for the comparative ers. The plots of specificity versus expression value (Supplementary analysis of CNS cell types. Cell 135, 749–762 (2008). Fig. 15) are likely to aid the search for efficient enhancers. Because the 8. Nelson, S.B., Sugino, K. & Hempel, C.M. The problem of neuronal cell types: a cellular infrastructure of the retina is highly conserved among mam- physiological genomics approach. Trends Neurosci. 29, 339–345 (2006). 9. Son, E.Y. et al. Conversion of mouse and human fibroblasts into functional spinal mals, vectors with cell type–specific enhancers may allow the genetic motor neurons. Cell Stem Cell 9, 205–218 (2011). manipulation of cell types across species. Finally, the transcriptome 10. Caiazzo, M. et al. Direct generation of functional dopaminergic neurons from mouse atlas described here is open-ended in the sense that, as new mouse and human fibroblasts. Nature 476, 224–227 (2011). 11. Kim, J. et al. Functional integration of dopaminergic neurons directly converted lines become available, more and more cell groups can be added to from mouse fibroblasts. Cell Stem Cell 9, 413–419 (2011). it. The comparability of a newly added transcriptome with the older 12. Roesch, K. et al. The transcriptome of retinal Müller glial cells. J. Comp. Neurol. 509, 225–238 (2008). ones is ensured by the internal control Arc line. 13. Corbo, J.C., Myers, C.A., Lawrence, K.A., Jadhav, A.P. & Cepko, C.L. A typology of photoreceptor gene expression patterns in the mouse. Proc. Natl. Acad. Sci. USA 104, Methods 12069–12074 (2007). 14. Gollisch, T. & Meister, M. Eye smarter than scientists believed: neural computations Methods and any associated references are available in the online in circuits of the retina. Neuron 65, 150–164 (2010). © 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature version of the paper at http://www.nature.com/natureneuroscience/. 15. Sanes, J.R. & Zipursky, S.L. Design principles of insect and vertebrate visual systems. Neuron 66, 15–36 (2010). 16. Wässle, H. Parallel processing in the mammalian retina. Nat. Rev. Neurosci. 5, Accession codes. Gene Expression Omnibus: GSE33089 (the original 747–757 (2004). npg data without processing). The rod contamination–corrected data can 17. Masland, R.H. The fundamental plan of the retina. Nat. Neurosci. 4, 877–886 (2001). be downloaded from http://www.fmi.ch/roska.data/index.php. 18. Siegert, S. et al. Genetic address book for retinal cell types. Nat. Neurosci. 12, 1197–1204 (2009). Note: Supplementary information is available on the Nature Neuroscience website. 19. Okaty, B.W., Sugino, K. & Nelson, S.B. Cell type-specific transcriptomics in the brain. J. Neurosci. 31, 6939–6943 (2011). Acknowledgments 20. Okaty, B.W., Sugino, K. & Nelson, S.B. A quantitative comparison of cell-type- We thank S. Djaffer, J. Jüttner, J. Hall and Y. Shimada for technical assistance, specific microarray gene expression profiling methods in the mouse brain. PLoS 6, e16493 (2011). E. Oakeley for comments on the experimental design, D. Balya for providing help ONE 21. Jeon, C.J., Strettoi, E. & Masland, R.H. The major cell populations of the mouse with programming, and K. Farrow, S. Rompani, K. Yonehara, V. Busskamp, retina. J. Neurosci. 18, 8936–8946 (1998). S. Oakeley, P. King, A. Matus, S. Arber and F. Rijli for comments on the manuscript. 22. Livesey, F.J. & Cepko, C.L. Vertebrate neural cell-fate determination: lessons from We thank Z. Raics for making the webpage, and L. Kus and S. Gong (Rockefeller the retina. Nat. Rev. Neurosci. 2, 109–118 (2001). University) for help and for BACs from the GENSAT project. We thank 23. Agathocleous, M. & Harris, W.A. From progenitors to differentiated cells in the I. Provencio (University of Virginia) for providing the melanopsin antibody. vertebrate retina. Annu. Rev. Cell Dev. Biol. 25, 45–69 (2009). The study was supported by Friedrich Miescher Institute funds, Alcon award, 24. Dasen, J.S. & Jessell, T.M. Hox networks and the origins of motor neuron diversity. a National Center of Competence in Research Genetics grant, a European Curr. Top. Dev. Biol. 88, 169–200 (2009). 25. Arendt, D. Evolution of eyes and photoreceptor cell types. 47, Research Council grant, a Swiss-Hungarian grant, and RETICIRC, TREATRUSH, Int. J. Dev. Biol. 563–571 (2003). SEEBETTER and OPTONEURO grants from the European Union to B.R. 26. Koyanagi, M., Kubokawa, K., Tsukamoto, H., Shichida, Y. & Terakita, A. Cephalochordate melanopsin: evolutionary linkage between invertebrate visual cells AUTHOR CONTRIBUTIONS and vertebrate photosensitive retinal ganglion cells. Curr. Biol. 15, 1065–1069 S.S. designed experiments; characterized mouse lines with immunohistochemistry (2005). and electrophysiology; performed confocal microscopy, image processing and 27. Isoldi, M.C., Rollag, M.D., Castrucci, A.M.D. & Provencio, I. Rhabdomeric phototransduction initiated by the vertebrate photopigment melanopsin. quantification; dissociated retinas; performed fluorescence-activated cell sorting; Proc. Natl. Acad. Sci. USA 102, 1217–1221 (2005). normalized gene array data; analyzed gene and exon array data; created figures; 28. Hardie, R.C. Phototransduction in Drosophila melanogaster. J. Exp. Biol. 204, and wrote the manuscript. E.C. performed RNA isolation, amplification and gene 3403–3409 (2001). profiling for gene and exon arrays. B.G.S. generated BAC transgenic mouse lines 29. Lee, Y.J. et al. The Drosophila Dgq gene encodes a Gα-protein that mediates and performed in situ hybridization. H.K. assisted with fluorescence-activated cell phototransduction. Neuron 13, 1143–1157 (1994).

494 VOLUME 15 | NUMBER 3 | MARCH 2012 nature NEUROSCIENCE r e so u r c e

30. Terakita, A., Hariyama, T., Tsukahara, Y., Katsukura, Y. & Tashiro, H. Interaction of 38. Klaver, C.C. et al. Genetic association of apolipoprotein E with age-related macular GTP-binding protein Gq with photoactivated rhodopsin in the photoreceptor- degeneration. Am. J. Hum. Genet. 63, 200–206 (1998). membranes of crayfish. FEBS Lett. 330, 197–200 (1993). 39. Shibuya, E. et al. Association of Toll-like receptor 4 gene polymorphisms with normal 31. Sekaran, S. et al. 2-Aminoethoxydiphenylborane is an acute inhibitor of directly tension glaucoma. Invest. Ophthalmol. Vis. Sci. 49, 4453–4457 (2008). photosensitive retinal ganglion cell activity in vitro and in vivo. J. Neurosci. 27, 40. Tezel, G. TNF-alpha signaling in glaucomatous neurodegeneration. Prog. Brain Res. 3981–3986 (2007). 173, 409–421 (2008). 32. Briggman, K.L., Helmstaedter, M. & Denk, W. Wiring specificity in the direction- 41. Naj, A.C. et al. Common variants at MS4A4/MS4A6E, CD2AP, CD33 and EPHA1 selectivity circuit of the retina. Nature 471, 183–188 (2011). are associated with late-onset Alzheimer’s disease. Nat. Genet. 43, 436–441 33. Yoshida, K. et al. A key role of starburst amacrine cells in originating retinal directional (2011). selectivity and optokinetic eye movement. Neuron 30, 771–780 (2001). 42. Hollingworth, P. et al. Common variants at ABCA7, MS4A6A/MS4A4E, EPHA1, 34. Tarpey, P. et al. Mutations in FRMD7, a newly identified member of the FERM CD33 and CD2AP are associated with Alzheimer’s disease. Nat. Genet. 43, family, cause X-linked idiopathic congenital nystagmus. Nat. Genet. 38, 429–435 (2011). 1242–1244 (2006). 43. Eiraku, M. et al. Self-organizing optic-cup morphogenesis in three-dimensional 35. Geng, R. et al. Usher syndrome IIIA gene clarin-1 is essential for hair cell function culture. Nature 472, 51–56 (2011). and associated neural activation. Hum. Mol. Genet. 18, 2748–2760 (2009). 44. MacLaren, R.E. et al. Retinal repair by transplantation of photoreceptor precursors. 36. Chen, J., Connor, K.M. & Smith, L.E. Overstaying their welcome: defective CX3CR1 Nature 444, 203–207 (2006). microglia eyed in macular degeneration. J. Clin. Invest. 117, 2758–2762 (2007). 45. Lamba, D.A. et al. Generation, purification and transplantation of photoreceptors 37. Klein, R.J. et al. Complement factor H polymorphism in age-related macular derived from human induced pluripotent stem cells. PLoS ONE 5, e8763 degeneration. Science 308, 385–389 (2005). (2010). © 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature npg

nature NEUROSCIENCE VOLUME 15 | NUMBER 3 | MARCH 2012 495 ONLINE METHODS and cell debris discarded by setting a second gate. Finally, after determination Mice. Mouse lines are summarized in Supplementary Table 1a. Mice obtained of the duration of the events (pulse width), a third gate was selected that served from Mutant Mouse Regional Resource Centers (MMRRC) were FVB/N-Swiss to exclude cell doublets or clumps. Using these three gates, cells were sorted in Webster hybrids that are known to harbor a recessive mutation causing photo single-cell drop mode. A total of 200 cells were sorted at 25 °C into a low-binding receptor degeneration. These were backcrossed into the C57BL/6J background tube (Eppendorf) containing 60 µl cell lysis buffer (TRI reagent, Sigma-Aldrich). (RCC) and the F2 generation was used for transcriptome analysis. All other mouse In the case of the PV, Drd4, Grik4 and Opn4 mouse lines, up to six retinas were lines had already been in the C57BL/6J background for several previous genera- combined to obtain the 200 cells. The tube was vortexed and then centrifuged tions. For cell sorting, Cre and CreERT2-expressing mouse lines were crossed to for a few seconds before incubation at 25 °C for 15 min. Cell lysates were then reporter lines. The PV-Cre mouse line was crossed to the ThyStopY reporter line. stored at –80 °C until further processing. The percentage of dead cells in the Other Cre-expressing mice were crossed to the Rosa26-LSL-RFP reporter line. To sorted population was estimated using propidium iodide, a DNA-binding dye activate CreERT2, Kcng4-CreERT2 × Rosa26-LSL-RFP mice were injected twice that is membrane-impermeable and excluded from viable cells. Propidium iodide intraperitoneally with tamoxifen in corn oil (Sigma, 0.15 mg per kilogram body (30 µl at 25 µg ml−1 in PBS, Sigma-Aldrich) was added to the sorted cells from weight) with an 8-h interval between the injections. The average age of mice was five different mouse lines and the number of propidium iodide–positive cells 66 d, and sexes were chosen randomly. For in situ hybridization, ChAT-Cre mice quantified: 0.05 ± 0.03% (s.e.m.) of the cells were positive. were crossed to the ThyStopY reporter. All animal procedures were performed in accordance with standard ethical guidelines (European Communities Guidelines RNA isolation, amplification and gene profiling. RNA amplification was per- on the Care and Use of Laboratory Animals: 86/609/EEC) and were approved by formed in batches (Supplementary Fig. 7). To avoid apparent differences between the Veterinary Department of the Canton of Basel-Stadt, Switzerland. cell-type transcriptomes caused by variable amplification across batches, RNA samples of cell groups belonging to the same biological triplicate were amplified Generation of transgenic mouse lines. Gja10-Cre, mGluR6-Cre and Kcng4- in different batches. Each batch included the Arc cell group as a reference group, CreERT2 mice were generated by bacterial artificial chromosome (BAC)-based which allowed comparison of the repeatability of amplifications across all batches. transgenesis46,47 (Supplementary Table 2). The purified BAC was injected into We performed 189 amplifications and gene chip hybridizations: 24 for testing the C57BL/6 blastocysts. linearity of amplification, 78 for comparing the transcriptomes of adult cell types and 87 for determining the time course of transcriptome changes across postnatal Two-photon targeted patch clamp recordings. The method has been development. Total RNA from each cell group was isolated using the PicoPure described18. RNA isolation kit (Arcturus) with the following modifications. The tubes were incubated at 42 °C for 10 min and vortexed at low speed. Total RNA was treated Immunohistochemistry. Retinas were dissected from the eyecup, fixed in 4% with 10 units of DNase I (Qiagen) for 15 min to remove any remaining genomic (wt/vol) paraformaldehyde in PBS for 20–30 min and washed overnight in DNA and eluted into 8 µl of elution buffer. The RNA quality was assessed using PBS. All further procedures have been described18. The primary antibodies are RNA 6000 PicoChips with an Agilent 2100 bioanalyzer. No traces of DNA were described in Supplementary Table 3. Donkey secondary antibodies were pur- detected; no degradation of ribosomal RNA was recorded. The rRNA represented chased from Invitrogen (Alexa Fluor 405, Alexa Fluor 488, Alexa Fluor 555, 90–95% of total RNA, and the 1.9–2.2 ratio of 28S/18S recorded indicated that Alexa Fluor 633) or from Jackson Laboratory (Cy2, Cy3, Cy5) and used at a the extractions met the criteria for downstream genetic analysis. concentration of 1:200. Total RNA from each sample was reverse transcribed using 4 µM T7- (dT)24/T7-(dN)6 primer mix (Affymetrix) and 150 units Superscript II reverse Microscopy. Confocal three-dimensional scans were carried out with an LSM 510 transcriptase (Invitrogen) in a volume of 10 µl. Synthesis of second-strand Meta Axioplan 2 or LSM 700 Axio Imager Z2 laser scanning confocal microscope cDNA was performed by adding 4 mM dNTPs, 6 units DNA polymerase I (Zeiss) using Plan-Apochromat 63× 1.4 numerical aperture, Plan-Apochromat and 0.4 units RNase H in 20 µl reaction volume. cRNA was produced by 20× 0.8 numerical aperture and EC Plan-Neofluoar 40× 1.30 numerical aperture in vitro transcription with a T7 RNA polymerase at 37 °C for 14 h using the

© 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature Oil M27 immersion objective lenses at four excitation laser lines. Images were MEGAscript T7 kit (Ambion) as per the manufacturer’s instructions. For the analyzed with Imaris (Bitplane). second cycle, first-strand cDNA was synthesized using 0.2 µg random prim- ers from 9 µl purified cRNA. The second-strand cDNA was produced using Retina dissociation. Because the expression of several retinal genes depends on the 10 µM T7-(dN)6 primer and 40 units of DNA polymerase at 16 °C for 2 h, npg phase of the animal’s circadian rhythm48, retina dissociation and cell sorting were after which 10 units T4 DNA polymerase (Invitrogen) were added and the always performed between 9:30 and 11:30 a.m. Retinal cells were dissociated using incubation continued for a further 10 min. The cDNA was in vitro transcribed a protocol modified from refs. 49 and 50 with a process time of only 15 min. Before with a T7 RNA polymerase at 37 °C for 16 h. The single-stranded cDNA was dissociation, 40 µl papain (Roche, 10 mg ml−1, 30 U per mg protein) was activated synthesized using 10 µg purified cRNA in the presence of 4 µg random prim- with 40 µl cysteine/ethylenediaminetetraacetic acid mix (pH 6.0–7.0, 25 mM cysteine ers, 0.2 M dithiothreitol, 12 mM dNTP + dUTP and 750 units Superscript II from Sigma, 5 mM EDTA from Amresco) in 320 µl HEPES buffer (Fluka; 100 mM (Roche Diagnostics) in a total volume of 20 µl. The cRNA was hydrolyzed in Hanks’ balanced salt solution (HBSS, Sigma)) for 15 min at 37 °C. The retina with 2 units of RNase H at 37 °C for 40 min. The sense cDNA was purified was removed from the eyecup in HBSS and incubated in the activated papain and eluted in 28 µl of elution buffer. Amplified products were purified using solution for 5 min at 37 °C. For PV, Drd4, Grik4 and Opn4 mouse lines, the the GeneChip cDNA Sample Cleanup Module (Affymetrix) with a 6.000g cen- incubation time was extended to 7 min. The cells were centrifuged for 2 min at trifugation speed during the first two steps. To improve the recovery from the 200g and then washed with 1 ml HBSS containing 2% (vol/vol) heat-inactivated columns, water or elution buffer was spun into the matrix at 50g and the mix- FCS. For retinal tissue dissociation, 800 µl HBSS containing 2% FCS was added ture left to stand for 4 min before centrifugation at 16,000g. The quantity and and the suspension gently triturated 10–15 times with a fire-polished Pasteur purity of the cRNA and cDNA produced during the first and second rounds pipette. The suspension was filtered through a 70-µm cup Filcon (BD) into a were evaluated using a NanoDrop ND-1000 spectrophotometer (Nanodrop 5-ml polypropylene round-bottom Falcon tube (BD). Technologies). The cDNA was then fragmented by uracil DNA glycosylase and apurinic/apyrimidic endonuclease 1 (APE 1) and biotin-labeled with termi- Fluorescence-activated cell sorting. Cells from the retina of the right eye were nal deoxynucleotidyl transferase using the GeneChip WT terminal labeling sorted and the retinal pattern of fluorescent protein expression was confirmed kit (Affymetrix). Hybridization was performed with 5 µg biotinylated target, from the left eye using immunohistochemical staining. Cell sorting was per- which was incubated with the GeneChip Gene 1.0 ST array (Affymetrix) at formed using a MoFlo sorter (Cytomation) with HQ515/30 (GFP) and HQ616/26 45 °C for 16 h. The GeneChip Mouse Exon 1.0 ST array was used for the (RFP) bandpass filters. After recording several thousand events, the fluores- developmental study. After hybridization, nonspecifically bound nucleotides cence gate was set using Summit v. 4.3.01 build 2449 software (Supplementary were removed by washing and the specifically bound target was detected using Figs. 7–9) and a population of events selected. Size and granularity of this popu- the GeneChip Hybridization, Wash and Stain Kit and the GeneChip Fluidics lation were determined using the forward and side scatter values, respectively, Station 450 (Affymetrix). The arrays were scanned using a GeneChip Scanner

nature NEUROSCIENCE doi:10.1038/nn.3032 3000 7G (Affymetrix) and the final output Affymetrix CEL data file format tyramide signal amplification and the GFP antibody steps, all procedures were acquired using a GeneChip Command Console Software (Affymetrix). carried out in a Ventana Discovery XT machine (Ventana Medical Systems) using the Research FISH XT procedure (Supplementary Table 4). Before addition of Array data analysis. A detailed description of the data normalization, estimation the probe, the slides were blocked with 4% (wt/vol) blocking solution (Molecular of the linearity, mixture component prediction and general data analysis may be Probes) in PBS for 4 min. The first antibody (mouse anti-digoxin–biotin, 1:500; found in the Supplementary Text. Jackson), was incubated for 60 min at 37 °C on the slides. The secondary antibody (streptavidin–horseradish peroxidase, 1:50; Tyramide Signal Amplification Kit, In situ hybridization. RNA isolation was performed with TRIzol LS reagent Molecular Probes) was incubated for 32 min at 37 °C. All further steps were car- (Invitrogen) according to the protocol with the following modification: the RNA ried out according to the manufacturer’s protocol. Alexa Fluor 568 dye (1:100) was precipitated overnight at −20 °C. cDNA was synthesized using 5 µg RNA and was added for 12 min to fluorescently label tyramide, followed by washing with the oligo-(dT)20 primer according to the ThermoScript RT-PCR kit (Invitrogen). PBS. For the GFP antibody staining, the sample was blocked with normal donkey cDNA was diluted 1:2 with diethylpyrocarbonate (DEPC)-treated H2O and 2 µl serum for 30 min, incubated with the primary antibody sheep anti-GFP (1:100) cDNA used for PCR with GoTaq Polymerase (Promega) (Supplementary for 2 h at 25 °C and washed with PBS. The secondary antibody, donkey anti- Table 4). The PCR product was cleaned using the Qiagen PCR purification kit sheep–Alexa 488 (1:200) was added together with DAPI for 1 h and the sample and sequenced using the forward primer. For in vitro transcription, 1 µg of the washed with PBS. The preparation was mounted on slides with ProLong Gold PCR product was incubated with Sp6 polymerase (NEB) at 40 °C for 2 h. After the antifade reagent (Invitrogen). in vitro transcription, 2 µl DNase (RNase free, Roche) was added and the preparation incubated at 37 °C for 15 min. The probe was mixed with 100 µl Tris-EDTA 46. Lee, E.C. et al. A highly efficientEscherichia coli-based chromosome engineering system buffer (prepared in DEPC-treated H2O, pH 8.0) and 10 µl 4 M LiCl (prepared adapted for recombinogenic targeting and subcloning of BAC DNA. Genomics 73, in DEPC-treated H2O). Ethanol (100%, 400 µl) was added to the sample for 56–65 (2001). precipitation overnight at −20 °C. After centrifugation at 16,000g and 4 °C for 47. Liu, P., Jenkins, N.A. & Copeland, N.G. A highly efficient recombineering-based 10 min, the probe was washed with 70% (vol/vol) ethanol and centrifuged again method for generating conditional knockout mutations. Genome Res. 13, 476–484 (2003). at 16,000g and 4 °C for 2 min. The pellet was air-dried and resuspended in 48. Caputto, B.L. & Guido, M.E. Immediate early gene expression within the visual 100 µl DEPC-treated H2O. For in situ hybridization, retinas were prepared 1 d system: light and circadian regulation in the retina and the suprachiasmatic nucleus. before the hybridization. All solutions were kept at 4 °C. The retinas were fixed for Neurochem. Res. 25, 153–162 (2000). 30 min in paraformaldehyde in PBS, washed twice in PBS for 1 h and incubated 49. Matsuda, T. & Cepko, C.L. Electroporation and RNA interference in the rodent retina in vivo and in vitro. Proc. Natl. Acad. Sci. USA 101, 16–22 (2004). in 30% (wt/vol) sucrose in PBS overnight. Each retina was then embedded in 50. Morrow, E.M., Belliveau, M.J. & Cepko, C.L. Two phases of rod photoreceptor Shandon M-2 embedding matrix (Thermo Electron), cryosectioned into 12-µm- differentiation during rat retinal development. J. Neurosci. 18, 3738–3748 thick slices (Microm HM560, Leica) and dried at 60 °C for 2–3 h. Except for the (1998). © 2012 Nature America, Inc. All rights reserved. America, Inc. © 2012 Nature npg

doi:10.1038/nn.3032 nature NEUROSCIENCE