Comprehensive Classification of Retinal Bipolar Neurons by Single-Cell Transcriptomics

doi:10.1016/j.cell.2016.07.054

. 2016 Aug 25;166(5):1308-1323.e30.

doi: 10.1016/j.cell.2016.07.054.

Comprehensive Classification of Retinal Bipolar Neurons by Single-Cell Transcriptomics

Affiliations

¹ Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA.
² Department of Genetics, Harvard Medical School, Boston, MA 02115, USA; Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA.
³ Center for Brain Science and Department of Molecular and Cellular Biology, Harvard University, Cambridge, MA 02130, USA.
⁴ Department of Genetics, Harvard Medical School, Boston, MA 02115, USA; Stanley Center for Psychiatric Research, Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA; Program in Medical and Population Genetics, Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA.
⁵ Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA; Stanley Center for Psychiatric Research, Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA.
⁶ Department of Genetics, Harvard Medical School, Boston, MA 02115, USA; Stanley Center for Psychiatric Research, Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA.
⁷ Department of Genetics, Harvard Medical School, Boston, MA 02115, USA; Department of Ophthalmology, Harvard Medical School, Boston, MA 02115, USA; Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA. Electronic address: cepko@genetics.med.harvard.edu.
⁸ Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA; Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA; Department of Biology and Koch Institute, MIT, Cambridge, MA 02139, USA. Electronic address: aregev@broadinstitute.org.
⁹ Center for Brain Science and Department of Molecular and Cellular Biology, Harvard University, Cambridge, MA 02130, USA. Electronic address: sanesj@mcb.harvard.edu.

PMID: 27565351
PMCID: PMC5003425
DOI: 10.1016/j.cell.2016.07.054

Comprehensive Classification of Retinal Bipolar Neurons by Single-Cell Transcriptomics

Karthik Shekhar et al. Cell. 2016.

. 2016 Aug 25;166(5):1308-1323.e30.

doi: 10.1016/j.cell.2016.07.054.

Authors

Affiliations

¹ Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA.
² Department of Genetics, Harvard Medical School, Boston, MA 02115, USA; Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA.
³ Center for Brain Science and Department of Molecular and Cellular Biology, Harvard University, Cambridge, MA 02130, USA.
⁴ Department of Genetics, Harvard Medical School, Boston, MA 02115, USA; Stanley Center for Psychiatric Research, Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA; Program in Medical and Population Genetics, Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA.
⁵ Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA; Stanley Center for Psychiatric Research, Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA.
⁶ Department of Genetics, Harvard Medical School, Boston, MA 02115, USA; Stanley Center for Psychiatric Research, Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA.
⁷ Department of Genetics, Harvard Medical School, Boston, MA 02115, USA; Department of Ophthalmology, Harvard Medical School, Boston, MA 02115, USA; Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA. Electronic address: cepko@genetics.med.harvard.edu.
⁸ Broad Institute of Harvard and MIT, Cambridge, MA 02142, USA; Howard Hughes Medical Institute, Chevy Chase, MD 20815, USA; Department of Biology and Koch Institute, MIT, Cambridge, MA 02139, USA. Electronic address: aregev@broadinstitute.org.
⁹ Center for Brain Science and Department of Molecular and Cellular Biology, Harvard University, Cambridge, MA 02130, USA. Electronic address: sanesj@mcb.harvard.edu.

PMID: 27565351
PMCID: PMC5003425
DOI: 10.1016/j.cell.2016.07.054

Abstract

Patterns of gene expression can be used to characterize and classify neuronal types. It is challenging, however, to generate taxonomies that fulfill the essential criteria of being comprehensive, harmonizing with conventional classification schemes, and lacking superfluous subdivisions of genuine types. To address these challenges, we used massively parallel single-cell RNA profiling and optimized computational methods on a heterogeneous class of neurons, mouse retinal bipolar cells (BCs). From a population of ∼25,000 BCs, we derived a molecular classification that identified 15 types, including all types observed previously and two novel types, one of which has a non-canonical morphology and position. We validated the classification scheme and identified dozens of novel markers using methods that match molecular expression to cell morphology. This work provides a systematic methodology for achieving comprehensive molecular classification of neurons, identifies novel neuronal types, and uncovers transcriptional differences that distinguish types within a class.

PubMed Disclaimer

Figures

None — Histogram of the fraction of ‘T’ bases in the last UMI position for each cell barcode. Data from replicate 1 of Batch 1 of Drop-seq was used. For most cell barcodes (green), the fraction of ‘T’ at position 8 of the UMI barcode is drawn from a normal distribution centered around 0.25, consistent with a uniform distribution for all 4 bases. For a small number of the cell barcodes (~5%), a fixation of ‘T’ at the last UMI position is observed (blue).

**Figure 1. Clustering of bipolar cells by Drop-seq**
(A) Sketch of retinal cross-section depicting major resident cell classes. Rod and cone photoreceptors detect and transduce light stimuli into chemical signals, relaying this information to rod and cone bipolar cells (BCs), respectively (turquoise and purple/orange). BCs synapse on retinal ganglion cells (whose axons form the optic nerve) in the inner plexiform layer (IPL) at varying depths that depend on the BC type. (B) Overview of experimental strategy. Retinas from *Vsx2*-GFP mice were dissociated, followed by FAC sorting for GFP+ cells. Single cell libraries were prepared using Drop-seq and sequenced. Raw reads were processed to obtain a digital expression matrix (genes x cells). PCA, followed by graph clustering was used to partition cells into clusters, and identify cluster-specific markers, which were validated *in vivo* using methods that detect gene expression and cellular morphology in combination. (C)–(E) 2D visualization of single cell clusters using tSNE. Individual points correspond to single cells colored according to clusters identified by the **(C)** Louvain-Jaccard, and **(D)** Infomap algorithms, and numbered in decreasing order of size. Arrows in panels (C), (D) indicate a Louvain-Jaccard BC cluster that was partitioned by Infomap (examined in Figure 5). Panel (E) shows the clustering output of Infomap when applied on cells from a single Drop-seq experiment (50% of the dataset). The tSNE representation was only used for visualization, and not for defining clusters. (F) Gene expression patterns (columns) of major retinal class markers (left panels) and known BC type markers (right panels) in BC (upper panels) and non-BC clusters (lower panels) based on the clusters in panel C. Clusters with cell-doublet signatures, and that contained < 50 cells are not shown. Putative cell type assignments, based on the expression of known genes, are indicated on the right (see Table S2). Nomenclature for BC types 1 and 5 is based on results in Figures 3 and 4. The size of each circle depicts the percentage of cells in the cluster in which the marker was detected (≥1 UMI), and its color depicts the average transcript count in expressing cells (nTrans). MG = Müller glia, AC = amacrine cells, PR = photoreceptors). (G) Hierarchical clustering of average gene signatures of BC clusters based (euclidean distance metric, average linkage). The confidence level of each split was assessed using bootstrap (**Methods and Resources**). Relatedness between clusters was used in prospective cluster assignment to BC type in panel F. See also Figures S1–S3 and Tables S1–S2

**Figure 2. Validation of markers for six BC types**
(A) Representative markers (columns) enriched in BC clusters (rows) predicted and validated in this study. Representation as in Figure 1F. (**B–D**) Validation of RBC-specific genes *Vstm2b*, *Casp7*, and *Rpa1* by FISH combined with PKCα immunostaining, which marks RBCs. (**E–I**) Validation of new markers of BC3A, BC3B, BC4, BC6, and BC7 against cell morphology. Leftmost panels show representative drawings of these types based on EM reconstructions (Helmstaedter et al., 2013), middle panels show lentiviral labeling of single BCs combined with FISH for the indicated gene in retinal cross sections. Dashed lines are drawn from calretinin antibody labeling within sublaminae (S) 2, 3, and 4. Insets show localization of FISH signal within virus-labeled cell bodies. Rightmost panels show FISH labeling of cell bodies on retinal whole mounts. To reduce background puncta in the GFP+ lentivirus labeled cells, an outlier removal noise filter was applied (**Methods and Resources**). Scale bars indicate 20 μm for main panels and 10 μm for insets. See also Figures S4–S5 and Tables S3–S4.

**Figure 3. BC1B is a non-canonical bipolar type**
(A) Expression patterns of known BC and amacrine cell genes across BC1A, BC1B, and BC2 clusters, plus new BC1A and BC1B markers. Representation as in Figure 1F. (B) *Pcdh17* (BC1A marker) and *Wls* (BC1B marker) label distinct populations of CFP-positive cells in the MitoP line. *Pcdh17* labels cells with a bipolar morphology positioned in the bipolar cell layer (BCL), whereas *Wls* labels cells that lack an upward process and are positioned in the amacrine cell layer (ACL) (dashed grey line denotes the division between these two layers). Insets show example cells with or without an upward process. (C) BC1B cells (Vsx2+ Ppp1r17−) are distinct from nGnG amacrine cells (Vsx2− Ppp1r17+) (**D–E**) Lentiviral labeling and Vsx2 immunostaining shows that BC1B cells lack an upward process and laminate in S1 (D), in contrast to other bipolar types that laminate at a similar depth in the IPL (E). (F) Representative drawings based on EM reconstructions of BC1A, BC2, and BC1B (the latter identified post-hoc from (Helmstaedter et al., 2013)) (**G–I**) BC1B cells lose their apical process, and translocate to the ACL. G. BC1B cells (CFP+ Lhx3+) predominatly have a bipolar morphology at P6, by P17 most have become unipolar. See representative images in H and diagram of developmental events in I. Scale bars indicate 20 μm for main panels and 10 μm for insets. See also Figure S6.

**Figure 4. Four BC5 types with distinct morphology and gene expression**
(A) Expression patterns of known and novel BC5 genes across BC5A-BC5D clusters. Representation as in Figure 1F. (**B–E**) FISH+lentiviral labeling for BC5A-BC5D markers from A. Insets show localization of FISH within GFP+ cell body. Noise reduction was applied to GFP+ lentivirus labeled cells as in Figure 2. (**F–K**) BC5 types labeled in transgenic lines that report on genes highlighted in **A. F**. *Kcng4*-cre;stop-YFP retina whole-mounts labeled with GFP and Sox6 show co-localization in BC5A cells G. *Kcng4*-cre;*Cdh9*-lacZ retinal cross section labeled for lacZ and cre show near complete co-localization. H. BC5D cells are GFP+ and PkarIIβ− in *Kirrel3*-GFP retinas. GFP-low and PkarIIβ+ cells correspond to BC3B J. A density recovery profile (DRP) shows that BC5D cells are uniformly spaced compared to a density matched random population I. *Kcng4*-cre;*Htr3a*-GFP retinal cross section labeled for GFP and cre show near complete co-localization. J. *Kcng4*-cre;stop-YFP;*Cntn5*-lacZ retinas labeled for YFP, lacZ, and Nfia combinatorially mark BC5A-5C. (**L,M)** *Kcng4*-cre;*Cntn5*-lacZ retinas infected with AAV-stop-GFP marks the morphology BC5A (green and orange cells) and BC5D (cyan cell). L. Terminals are shown from the side (left) and *en face* (right). M. GFP and lacZ labeling of axon stalks distinguishes BC5A (lacZ+) from BC5D (lacZ−). Dashed lines drawn from choline acetyltransferase (ChAT) labeling of S2 and S4. Scale bars indicate 20 μm for main panels and 10 μm for insets. (N) Bulk RNA-seq of FAC sorted *Htr3a*-GFP cells shows BC5A and BC5D markers robustly expressed, but BC5B and BC5C markers are absent.

**Figure 5. BC8 and BC9 are closely related but separable by unsupervised methods**
(A) A magnified view of cluster 15 on the tSNE map in Figure 1C shows two subpopulations. Individual cells are colored by expression levels of ON cone BC genes (*Grm6*, *Isl1*, *Scgn*) and Cluster 15 enriched genes (*Cpne9*, *Spock3*, *Seripini1*). (B) BC9 cells labeled by the Clm-1 transgenic line are *Cpne9*⁺ (C) Single cells labeled by lentivirus combined with FISH. *Cpne9* is expressed in some but not all BCs with wide axonal arbors laminating at low IPL depth, consistent with the presence of two populations, BC9 (*Cpne9⁺*) and BC8 (*Cpne9⁻*). Insets show FISH and GFP labeling of the cell body. Noise reduction applied to GFP+ lentivirus labeled cells as in Figure 2. (**D–E**) Soma spacing in *Cpne9* labeled retinal whole-mounts is indicative of *Cpne9* marking a single type. E shows a density recovery profile derived from whole-mount (D), revealing uniform spacing with an exclusion zone of 14.3 μm, which is absent in density matched simulations of randomly distributed, non-overlapping cells of similar size. (F) Retinal whole-mounts with double FISH labeling for *Cpne9* and *Seripini1*, reveals two populations, *Cpne9⁺ Seripini1⁺* (BC9, indicated by solid outlines) and *Cpne9*⁻ *Seripini*⁺ (BC8, indicated by dashed outlines). Scale bars indicate 20 μm for main panels and 10 μm for insets.

**Figure 6. Drop-seq transcriptomes provides insights into BC function**
(A) Representative drawings of BC types validated in this study, drawn from EM reconstructions (Helmstadter et al., 2013). (B) Hierarchical clustering of BC clusters, similar to Figure 1G, now with identities of the BC1s, BC5s, and BC8 and 9 resolved based on results from Figures 3, 4, and 5, respectively. (C) Enrichment patterns of Gene Ontology (GO) categories in BC clusters based on the GO-PCA algorithm. Rows correspond to significantly enriched GO terms, while columns correspond to random averages of single-cell gene expression signatures arranged by cluster (200 per cluster, averaging was performed to mitigate single-cell noise). (**D–H**) Dotplots of functionally and developmentally relevant genes expressed by BC types. Representation as in Figure 1F. D. Glutamate receptors and ON pathway components. E Acetylcholine, GABA, and glycine receptors. F. Potassium channel subunits G. Transcription factors. H. Adhesion/recognition molecules. In panels **D–H**, only genes expressed in > 20% of cells in at least one BC cluster are shown.

**Figure 7. Comparison of Drop-seq with Smart-seq2**
(A) Bulk RNA-seq expression levels of 15,063 genes tightly correlate across two biological replicates (~10,000 cells each) processed using the Smart-seq2 method. (B) Gene expression levels averaged across 229 single cells (3 biological replicates) tightly correlate with the expression levels in the bulk libraries. (C) Single-cell averaged expression levels of *Vsx2*-GFP cells (log(Transcripts-per-million+ 1) units) correlate between Smart-seq2 and Drop-seq datasets (D) Sensitivity of transcript detection in single cell libraries as a function of Smart-seq2 bulk expression levels. Curves show results for Smart-seq2 (3 replicates), Drop-seq (6 replicates) and deep-sequenced Drop-seq and downsampled Smart-seq2 data. (E) Clustering and tSNE visualization of Smart-seq2 single-cell data. Each cell is labeled on the tSNE map by its random forest (RF) assigned cell type. The RF model assigned one of 18 possible types including 14 BC types (1A-8/9), RBC, Müller glia (MG), Amacrine cells (A), rod photoreceptors (R), cone photoreceptors (C) or unknown (N). (F) Top 30 differentially expressed (DE) genes in each BC type computed using a *post hoc* test on the Smart-seq2 data based on the RF-assigned labels. BC types with fewer than 3 cells in the data were excluded. Black bars on the right mark genes that were common among the top 30 DE genes for the corresponding Drop-seq clusters (Table S3). (G) tSNE visualization of *Kcng4*-GFP Smart-seq2 data (309 single cells). Each cell is represented on the tSNE map by its RF-assigned class label. (**H–I**) Violin plots showing expression of known and novel BC5A-D markers (identified in Drop-seq) in the BC5A, BC5D and BC7 clusters. See also Figure S7 and Table S5.

See this image and copyright information in PMC

Cited by

Increased Connexin36 Phosphorylation in AII Amacrine Cell Coupling of the Mouse Myopic Retina.
Banerjee S, Wang Q, Zhao F, Tang G, So C, Tse D, To CH, Feng Y, Zhou X, Pan F. Banerjee S, et al. Front Cell Neurosci. 2020 Jun 1;14:124. doi: 10.3389/fncel.2020.00124. eCollection 2020. Front Cell Neurosci. 2020. PMID: 32547367 Free PMC article.
Evolutionary conservation of VSX2 super-enhancer modules in retinal development.
Honnell V, Sweeney S, Norrie J, Parks M, Ramirez C, Jannu AJ, Xu B, Teubner B, Lee AY, Bell C, Dyer MA. Honnell V, et al. Development. 2024 Jul 1;151(13):dev202435. doi: 10.1242/dev.202435. Epub 2024 Jul 12. Development. 2024. PMID: 38994775 Free PMC article.
Pinpointing Cell Identity in Time and Space.
Savulescu AF, Jacobs C, Negishi Y, Davignon L, Mhlanga MM. Savulescu AF, et al. Front Mol Biosci. 2020 Aug 14;7:209. doi: 10.3389/fmolb.2020.00209. eCollection 2020. Front Mol Biosci. 2020. PMID: 32923457 Free PMC article.
Comprehensive single-cell atlas of the mouse retina.
Li J, Choi J, Cheng X, Ma J, Pema S, Sanes JR, Mardon G, Frankfort BJ, Tran NM, Li Y, Chen R. Li J, et al. bioRxiv [Preprint]. 2024 Jan 28:2024.01.24.577060. doi: 10.1101/2024.01.24.577060. bioRxiv. 2024. Update in: iScience. 2024 May 08;27(6):109916. doi: 10.1016/j.isci.2024.109916 PMID: 38328114 Free PMC article. Updated. Preprint.
Center-surround interactions underlie bipolar cell motion sensitivity in the mouse retina.
Strauss S, Korympidou MM, Ran Y, Franke K, Schubert T, Baden T, Berens P, Euler T, Vlasits AL. Strauss S, et al. Nat Commun. 2022 Sep 26;13(1):5574. doi: 10.1038/s41467-022-32762-7. Nat Commun. 2022. PMID: 36163124 Free PMC article.

See all "Cited by" articles

References

1. Baas D, Bumsted KM, Martinez JA, Vaccarino FM, Wikler KC, Barnstable CJ. The subcellular localization of Otx2 is cell-type specific and developmentally regulated in the mouse retina. Brain Res Mol Brain Res. 2000;78:26–37. - PubMed
1. Berglund K, Schleich W, Krieger P, Loo LS, Wang D, Cant NB, Feng G, Augustine GJ, Kuner T. Imaging synaptic inhibition in transgenic mice expressing the chloride indicator, Clomeleon. Brain Cell Biol. 2006;35:207–228. - PMC - PubMed
1. Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E. Fast unfolding of communities in large networks. J Stat Mech-Theory E 2008
1. Borghuis BG, Looger LL, Tomita S, Demb JB. Kainate receptors mediate signaling in both transient and sustained OFF bipolar cell pathways in mouse retina. J Neurosci. 2014;34:6128–6139. - PMC - PubMed
1. Breiman L. Random forests. Mach Learn. 2001;45:5–32.

Publication types

Actions

MeSH terms

Actions
Actions
Actions
Actions
Actions
Actions
Actions
Actions
Actions
Actions
Actions
Actions
Actions
Actions
Actions
Actions

Substances

Actions

Grants and funding

LinkOut - more resources

Full Text Sources
Other Literature Sources
Molecular Biology Databases
- Mouse Genome Informatics (MGI)
- NIAID Data Ecosystem - Find datasets on Infectious and Immune-mediated Diseases
Research Materials
- Addgene Non-profit plasmid repository
Miscellaneous
- SciCrunch

[1] Baas D, Bumsted KM, Martinez JA, Vaccarino FM, Wikler KC, Barnstable CJ. The subcellular localization of Otx2 is cell-type specific and developmentally regulated in the mouse retina. Brain Res Mol Brain Res. 2000;78:26–37. - PubMed

[2] Baas D, Bumsted KM, Martinez JA, Vaccarino FM, Wikler KC, Barnstable CJ. The subcellular localization of Otx2 is cell-type specific and developmentally regulated in the mouse retina. Brain Res Mol Brain Res. 2000;78:26–37. - PubMed

[3] Berglund K, Schleich W, Krieger P, Loo LS, Wang D, Cant NB, Feng G, Augustine GJ, Kuner T. Imaging synaptic inhibition in transgenic mice expressing the chloride indicator, Clomeleon. Brain Cell Biol. 2006;35:207–228. - PMC - PubMed

[4] Berglund K, Schleich W, Krieger P, Loo LS, Wang D, Cant NB, Feng G, Augustine GJ, Kuner T. Imaging synaptic inhibition in transgenic mice expressing the chloride indicator, Clomeleon. Brain Cell Biol. 2006;35:207–228. - PMC - PubMed

[5] Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E. Fast unfolding of communities in large networks. J Stat Mech-Theory E 2008

[6] Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E. Fast unfolding of communities in large networks. J Stat Mech-Theory E 2008

[7] Borghuis BG, Looger LL, Tomita S, Demb JB. Kainate receptors mediate signaling in both transient and sustained OFF bipolar cell pathways in mouse retina. J Neurosci. 2014;34:6128–6139. - PMC - PubMed

[8] Borghuis BG, Looger LL, Tomita S, Demb JB. Kainate receptors mediate signaling in both transient and sustained OFF bipolar cell pathways in mouse retina. J Neurosci. 2014;34:6128–6139. - PMC - PubMed

[9] Breiman L. Random forests. Mach Learn. 2001;45:5–32.

[10] Breiman L. Random forests. Mach Learn. 2001;45:5–32.

Save citation to file

Email citation

Add to Collections

Add to My Bibliography

Your saved search

Create a file for external citation management software

Your RSS Feed

Comprehensive Classification of Retinal Bipolar Neurons by Single-Cell Transcriptomics

Affiliations

Comprehensive Classification of Retinal Bipolar Neurons by Single-Cell Transcriptomics

Authors

Affiliations

Abstract

Figures

Similar articles

Cited by

References

Publication types

MeSH terms

Substances

Grants and funding

LinkOut - more resources

Full Text Sources

Other Literature Sources

Molecular Biology Databases

Research Materials

Miscellaneous

Abstract

Figures

Similar articles

Cited by

References

Publication types

MeSH terms

Substances

Related information

Grants and funding

LinkOut - more resources

Full Text Sources

Other Literature Sources

Molecular Biology Databases

Research Materials

Miscellaneous