Single-cell multiome of the human retina and deep learning nominate causal variants in complex eye diseases

Linking Iris Cis-Regulatory Variants to Primary Angle-Closure Glaucoma Via Clinical Imaging and Multiomics

Purpose

To elucidate the genetic basis of primary angle-closure glaucoma (PACG) by identifying pathogenic tissue and critical tissue-specific variants.

Methods

The correlations among PACG susceptibility, axial length (AL), and anterior chamber depth (ACD) were evaluated using meta-analyses. Propensity score matching was utilized on 2161 participants from the Handan Eye Study to determine the risk factors independent of ACD and AL for PACG. Subsequently, we employed the assay for transposase-accessible chromatin with sequencing (ATAC-seq) and allele-specific self-transcribing active regulatory region sequencing (STARR-seq) to screen 202 PACG genome-wide association study (GWAS) variants for chromatin accessibility and functional roles.

Results

The meta-analysis found that PACG susceptibility loci are not associated with ACD or AL. However, abnormal iris phenotypes emerged as significant independent risk factors for primary angle-closure disease (PACD), unrelated to ACD and AL. Substantial enrichment of PACG heritability was observed in the open chromatin regions of the human iris. Within the iris-relevant cellular context, 22 out of the 202 PACG GWAS variants could influence enhancer activity. Two variants in the iris open chromatin regions were implicated in the modulation of PLEKHA7 and C10orf53 expression. The downregulation of these two genes affects cytoskeletal organization.

Conclusions

Our findings underscore the importance of the iris in the pathogenesis of PACG and identified iris-specific, enhancer-modulating variants that may influence disease risk. Our approach also provides a generalizable framework for studying ocular diseases from the perspective of enhancer-modulating variants.

Single cell variant to enhancer to gene map for coronary artery disease

Although genome wide association studies (GWAS) in large populations have identified hundreds of variants associated with common diseases such as coronary artery disease (CAD), most disease-associated variants lie within non-coding regions of the genome, rendering it difficult to determine the downstream causal gene and cell type. Here, we performed paired single nucleus gene expression and chromatin accessibility profiling from 44 human coronary arteries. To link disease variants to molecular traits, we developed a meta-map of 88 samples and discovered 11,182 single-cell chromatin accessibility quantitative trait loci (caQTLs). Heritability enrichment analysis and disease variant mapping demonstrated that smooth muscle cells (SMCs) harbor the greatest genetic risk for CAD. To capture the continuum of SMC cell states in disease, we used dynamic single cell caQTL modeling for the first time in tissue to uncover QTLs whose effects are modified by cell state and expand our insight into genetic regulation of heterogenous cell populations. Notably, we identified a variant in the COL4A1/COL4A2 CAD GWAS locus which becomes a caQTL as SMCs de-differentiate by changing a transcription factor binding site for EGR1/2. To unbiasedly prioritize functional candidate genes, we built a genome-wide single cell variant to enhancer to gene (scV2E2G) map for human CAD to link disease variants to causal genes in cell types. Using this approach, we found several hundred genes predicted to be linked to disease variants in different cell types. Next, we performed genome-wide Hi-C in 16 human coronary arteries to build tissue specific maps of chromatin conformation and link disease variants to integrated chromatin hubs and distal target genes. Using this approach, we show that rs4887091 within the ADAMTS7 CAD GWAS locus modulates function of a super chromatin interactome through a change in a CTCF binding site. Finally, we used CRISPR interference to validate a distal gene, AMOTL2, liked to a CAD GWAS locus. Collectively we provide a disease-agnostic framework to translate human genetic findings to identify pathologic cell states and genes driving disease, producing a comprehensive scV2E2G map with genetic and tissue level convergence for future mechanistic and therapeutic studies.

SMARCA5-mediated chromatin remodeling is required for germinal center formation

The establishment of long-lasting immunity against pathogens is facilitated by the germinal center (GC) reaction, during which B cells increase their antibody affinity and differentiate into antibody-secreting cells (ASC) and memory cells. These events involve modifications in chromatin packaging that orchestrate the profound restructuring of gene expression networks that determine cell fate. While several chromatin remodelers were implicated in lymphocyte functions, less is known about SMARCA5. Here, using ribosomal pull-down for analyzing translated genes in GC B cells, coupled with functional experiments in mice, we identified SMARCA5 as a key chromatin remodeler in B cells. While the naive B cell compartment remained unaffected following conditional depletion of Smarca5, effective proliferation during B cell activation, immunoglobulin class switching, and as a result GC formation and ASC differentiation were impaired. Single-cell multiomic sequencing analyses revealed that SMARCA5 is crucial for facilitating the transcriptional modifications and genomic accessibility of genes that support B cell activation and differentiation. These findings offer novel insights into the functions of SMARCA5, which can be targeted in various human pathologies.

Single-Cell Multiomics Profiling Reveals Heterogeneity of Müller Cells in the Oxygen-Induced Retinopathy Model

Purpose

Retinal neovascularization poses heightened risks of vision loss and blindness. Despite its clinical significance, the molecular mechanisms underlying the pathogenesis of retinal neovascularization remain elusive. This study utilized single-cell multiomics profiling in an oxygen-induced retinopathy (OIR) model to comprehensively investigate the intricate molecular landscape of retinal neovascularization.

Methods

Mice were exposed to hyperoxia to induce the OIR model, and retinas were isolated for nucleus isolation. The cellular landscape of the single-nucleus suspensions was extensively characterized through single-cell multiomics sequencing. Single-cell data were integrated with genome-wide association study (GWAS) data to identify correlations between ocular cell types and diabetic retinopathy. Cell communication analysis among cells was conducted to unravel crucial ligand-receptor signals. Trajectory analysis and dynamic characterization of Müller cells were performed, followed by integration with human retinal data for pathway analysis.

Results

The multiomics dataset revealed six major ocular cell classes, with Müller cells/astrocytes showing significant associations with proliferative diabetic retinopathy (PDR). Cell communication analysis highlighted pathways that are associated with vascular proliferation and neurodevelopment, such as Vegfa-Vegfr2, Igf1-Igf1r, Nrxn3-Nlgn1, and Efna5-Epha4. Trajectory analysis identified a subset of Müller cells expressing genes linked to photoreceptor degeneration. Multiomics data integration further unveiled positively regulated genes in OIR Müller cells/astrocytes associated with axon development and neurotransmitter transmission.

Conclusions

This study significantly advances our understanding of the intricate cellular and molecular mechanisms underlying retinal neovascularization, emphasizing the pivotal role of Müller cells. The identified pathways provide valuable insights into potential therapeutic targets for PDR, offering promising directions for further research and clinical interventions.

Unraveling the developmental heterogeneity within the human retina to reconstruct the continuity of retinal ganglion cell maturation and stage-specific intrinsic and extrinsic factors

Tissue development is a complex spatiotemporal process with multiple interdependent components. Anatomical, histological, sequencing, and evolutional strategies can be used to profile and explain tissue development from different perspectives. The introduction of scRNAseq methods and the computational tools allows to deconvolute developmental heterogeneity and draw a decomposed uniform map. In this manuscript, we decomposed the development of a human retina with a focus on the retinal ganglion cells (RGC). To increase the temporal resolution of retinal cell classes maturation state we assumed the working hypothesis that that maturation of retinal ganglion cells is a continuous, non-discrete process. We have assembled the scRNAseq atlas of human fetal retina from fetal week 8 to week 27 and applied the computational methods to unravel maturation heterogeneity into a uniform maturation track. We align RGC transcriptomes in pseudotime to map RGC developmental fate trajectories against the broader timeline of retinal development. Through this analysis, we identified the continuous maturation track of RGC and described the cell-intrinsic (DEGs, maturation gene profiles, regulons, transcriptional motifs) and -extrinsic profiles (neurotrophic receptors across maturation, cell-cell interactions) of different RGC maturation states. We described the genes involved in the retina and RGC maturation, including de novo RGC maturation drivers. We demonstrate the application of the human fetal retina atlas as a reference tool, allowing automated annotation and universal embedding of scRNAseq data. Altogether, our findings deepen the current knowledge of the retina and RGC maturation by bringing in the maturation dimension for the cell class vs. state analysis. We show how the pseudotime application contributes to developmental-oriented analyses, allowing to order the cells by their maturation state. This approach not only improves the downstream computational analysis but also provides a true maturation track transcriptomics profile.

Context-aware single-cell multiomics approach identifies cell-type-specific lung cancer susceptibility genes

Genome-wide association studies (GWAS) identified over fifty loci associated with lung cancer risk. However, underlying mechanisms and target genes are largely unknown, as most risk-associated variants might regulate gene expression in a context-specific manner. Here, we generate a barcode-shared transcriptome and chromatin accessibility map of 117,911 human lung cells from age/sex-matched ever- and never-smokers to profile context-specific gene regulation. Identified candidate cis-regulatory elements (cCREs) are largely cell type-specific, with 37% detected in one cell type. Colocalization of lung cancer candidate causal variants (CCVs) with these cCREs combined with transcription factor footprinting prioritize the variants for 68% of the GWAS loci. CCV-colocalization and trait relevance score indicate that epithelial and immune cell categories, including rare cell types, contribute to lung cancer susceptibility the most. A multi-level cCRE-gene linking system identifies candidate susceptibility genes from 57% of the loci, where most loci display cell-category-specific target genes, suggesting context-specific susceptibility gene function.

Current genomic deep learning models display decreased performance in cell type-specific accessible regions

BACKGROUND

A number of deep learning models have been developed to predict epigenetic features such as chromatin accessibility from DNA sequence. Model evaluations commonly report performance genome-wide; however, cis regulatory elements (CREs), which play critical roles in gene regulation, make up only a small fraction of the genome. Furthermore, cell type-specific CREs contain a large proportion of complex disease heritability.

RESULTS

We evaluate genomic deep learning models in chromatin accessibility regions with varying degrees of cell type specificity. We assess two modeling directions in the field: general purpose models trained across thousands of outputs (cell types and epigenetic marks) and models tailored to specific tissues and tasks. We find that the accuracy of genomic deep learning models, including two state-of-the-art general purpose models-Enformer and Sei-varies across the genome and is reduced in cell type-specific accessible regions. Using accessibility models trained on cell types from specific tissues, we find that increasing model capacity to learn cell type-specific regulatory syntax-through single-task learning or high capacity multi-task models-can improve performance in cell type-specific accessible regions. We also observe that improving reference sequence predictions does not consistently improve variant effect predictions, indicating that novel strategies are needed to improve performance on variants.

CONCLUSIONS

Our results provide a new perspective on the performance of genomic deep learning models, showing that performance varies across the genome and is particularly reduced in cell type-specific accessible regions. We also identify strategies to maximize performance in cell type-specific accessible regions.

Mutant mice with rod-specific VPS35 deletion exhibit retinal α-synuclein pathology-associated degeneration

Vacuolar protein sorting 35 (VPS35), the core component of the retromer complex which regulates endosomal trafficking, is genetically linked with Parkinson's disease (PD). Impaired vision is a common non-motor manifestation of PD. Here, we show mouse retinas with VPS35-deficient rods exhibit synapse loss and visual deficit, followed by progressive degeneration concomitant with the emergence of Lewy body-like inclusions and phospho-α-synuclein (P-αSyn) aggregation. Ultrastructural analyses reveal VPS35-deficient rods accumulate aggregates in late endosomes, deposited as lipofuscins bound to P-αSyn. Mechanistically, we uncover a protein network of VPS35 and its interaction with HSC70. VPS35 deficiency promotes sequestration of HSC70 and P-αSyn aggregation in late endosomes. Microglia which engulf lipofuscins and P-αSyn aggregates are activated, displaying autofluorescence, observed as bright dots in fundus imaging of live animals, coinciding with pathology onset and progression. The Rod∆Vps35 mouse line is a valuable tool for further mechanistic investigation of αSyn lesions and retinal degenerative diseases.

Current genomic deep learning models display decreased performance in cell type specific accessible regions

Background

A number of deep learning models have been developed to predict epigenetic features such as chromatin accessibility from DNA sequence. Model evaluations commonly report performance genome-wide; however, cis regulatory elements (CREs), which play critical roles in gene regulation, make up only a small fraction of the genome. Furthermore, cell type specific CREs contain a large proportion of complex disease heritability.

Results

We evaluate genomic deep learning models in chromatin accessibility regions with varying degrees of cell type specificity. We assess two modeling directions in the field: general purpose models trained across thousands of outputs (cell types and epigenetic marks), and models tailored to specific tissues and tasks. We find that the accuracy of genomic deep learning models, including two state-of-the-art general purpose models - Enformer and Sei - varies across the genome and is reduced in cell type specific accessible regions. Using accessibility models trained on cell types from specific tissues, we find that increasing model capacity to learn cell type specific regulatory syntax - through single-task learning or high capacity multi-task models - can improve performance in cell type specific accessible regions. We also observe that improving reference sequence predictions does not consistently improve variant effect predictions, indicating that novel strategies are needed to improve performance on variants.

Conclusions

Our results provide a new perspective on the performance of genomic deep learning models, showing that performance varies across the genome and is particularly reduced in cell type specific accessible regions. We also identify strategies to maximize performance in cell type specific accessible regions.

Integration of spatial and single-cell data across modalities with weakly linked features

Although single-cell and spatial sequencing methods enable simultaneous measurement of more than one biological modality, no technology can capture all modalities within the same cell. For current data integration methods, the feasibility of cross-modal integration relies on the existence of highly correlated, a priori 'linked' features. We describe matching X-modality via fuzzy smoothed embedding (MaxFuse), a cross-modal data integration method that, through iterative coembedding, data smoothing and cell matching, uses all information in each modality to obtain high-quality integration even when features are weakly linked. MaxFuse is modality-agnostic and demonstrates high robustness and accuracy in the weak linkage scenario, achieving 20~70% relative improvement over existing methods under key evaluation metrics on benchmarking datasets. A prototypical example of weak linkage is the integration of spatial proteomic data with single-cell sequencing data. On two example analyses of this type, MaxFuse enabled the spatial consolidation of proteomic, transcriptomic and epigenomic information at single-cell resolution on the same tissue section.

Single-cell analyses reveal transient retinal progenitor cells in the ciliary margin of developing human retina

The emergence of retinal progenitor cells and differentiation to various retinal cell types represent fundamental processes during retinal development. Herein, we provide a comprehensive single cell characterisation of transcriptional and chromatin accessibility changes that underline retinal progenitor cell specification and differentiation over the course of human retinal development up to midgestation. Our lineage trajectory data demonstrate the presence of early retinal progenitors, which transit to late, and further to transient neurogenic progenitors, that give rise to all the retinal neurons. Combining single cell RNA-Seq with spatial transcriptomics of early eye samples, we demonstrate the transient presence of early retinal progenitors in the ciliary margin zone with decreasing occurrence from 8 post-conception week of human development. In retinal progenitor cells, we identified a significant enrichment for transcriptional enhanced associate domain transcription factor binding motifs, which when inhibited led to loss of cycling progenitors and retinal identity in pluripotent stem cell derived organoids.

Cone Synaptic function is modulated by the leucine rich repeat (LRR) adhesion molecule LRFN2

Daylight vision is mediated by cone photoreceptors in vertebrates, which synapse with bipolar cells (BCs) and horizontal (HCs) cells. This cone synapse is functionally and anatomically complex, connecting to 8 types of depolarizing BCs (DBCs) and 5 types of hyperpolarizing BCs (HBCs) in mice. The dendrites of DBCs and HCs cells make invaginating ribbon synapses with the cone axon terminal, while HBCs form flat synapses with the cone pedicles. The molecular architecture that underpins this organization is relatively poorly understood. To identify new proteins involved in synapse formation and function we used an unbiased proteomic approach and identified LRFN2 (leucine-rich repeat and fibronectin III domain-containing 2) as a component of the DBC signaling complex. LRFN2 is selectively expressed at cone terminals and co-localizes with PNA, and other DBC signalplex members. In LRFN2 deficient mice, the synaptic markers: LRIT3, ELFN2, mGluR6, TRPM1 and GPR179 are properly localized. Similarly, LRFN2 expression and localization is not dependent on these synaptic proteins. In the absence of LRFN2 the cone-mediated photopic electroretinogram b-wave amplitude is reduced at the brightest flash intensities. These data demonstrate that LRFN2 absence compromises normal synaptic transmission between cones and cone DBCs.Significance Statement Signaling between cone photoreceptors and the downstream bipolar cells is critical to normal vision. Cones synapse with 13 different types of bipolar cells forming an invaginating ribbon synapses with 8 types, and flat synapses with 5 types, to form one of the most complex synapses in the brain. In this report a new protein, LRFN2 (leucine-rich repeat and fibronectin III domain-containing 2), was identified that is expressed at the cone synapse. Using Lrfn2 knockout mice we show LRFN2 is required for the normal cone signaling.

Controlling donor and newborn neuron migration and maturation in the eye through microenvironment engineering

Ongoing cell therapy trials have demonstrated the need for precision control of donor cell behavior within the recipient tissue. We present a methodology to guide stem cell-derived and endogenously regenerated neurons by engineering the microenvironment. Being an "approachable part of the brain," the eye provides a unique opportunity to study neuron fate and function within the central nervous system. Here, we focused on retinal ganglion cells (RGCs)-the neurons in the retina are irreversibly lost in glaucoma and other optic neuropathies but can potentially be replaced through transplantation or reprogramming. One of the significant barriers to successful RGC integration into the existing mature retinal circuitry is cell migration toward their natural position in the retina. Our in silico analysis of the single-cell transcriptome of the developing human retina identified six receptor-ligand candidates, which were tested in functional in vitro assays for their ability to guide human stem cell-derived RGCs. We used our lead molecule, SDF1, to engineer an artificial gradient in the retina, which led to a 2.7-fold increase in donor RGC migration into the ganglion cell layer (GCL) and a 3.3-fold increase in the displacement of newborn RGCs out of the inner nuclear layer. Only donor RGCs that migrated into the GCL were found to express mature RGC markers, indicating the importance of proper structure integration. Together, these results describe an "in silico-in vitro-in vivo" framework for identifying, selecting, and applying soluble ligands to control donor cell function after transplantation.

Unique activities of two overlapping PAX6 retinal enhancers

Enhancers play a critical role in development by precisely modulating spatial, temporal, and cell type-specific gene expression. Sequence variants in enhancers have been implicated in diseases; however, establishing the functional consequences of these variants is challenging because of a lack of understanding of precise cell types and developmental stages where the enhancers are normally active. PAX6 is the master regulator of eye development, with a regulatory landscape containing multiple enhancers driving the expression in the eye. Whether these enhancers perform additive, redundant or distinct functions is unknown. Here, we describe the precise cell types and regulatory activity of two PAX6 retinal enhancers, HS5 and NRE. Using a unique combination of live imaging and single-cell RNA sequencing in dual enhancer-reporter zebrafish embryos, we uncover differences in the spatiotemporal activity of these enhancers. Our results show that although overlapping, these enhancers have distinct activities in different cell types and therefore likely nonredundant functions. This work demonstrates that unique cell type-specific activities can be uncovered for apparently similar enhancers when investigated at high resolution in vivo.

Gene regulatory network inference in the era of single-cell multi-omics

The interplay between chromatin, transcription factors and genes generates complex regulatory circuits that can be represented as gene regulatory networks (GRNs). The study of GRNs is useful to understand how cellular identity is established, maintained and disrupted in disease. GRNs can be inferred from experimental data - historically, bulk omics data - and/or from the literature. The advent of single-cell multi-omics technologies has led to the development of novel computational methods that leverage genomic, transcriptomic and chromatin accessibility information to infer GRNs at an unprecedented resolution. Here, we review the key principles of inferring GRNs that encompass transcription factor-gene interactions from transcriptomics and chromatin accessibility data. We focus on the comparison and classification of methods that use single-cell multimodal data. We highlight challenges in GRN inference, in particular with respect to benchmarking, and potential further developments using additional data modalities.

Context-aware single-cell multiome approach identified cell-type specific lung cancer susceptibility genes

Genome-wide association studies (GWAS) identified over fifty loci associated with lung cancer risk. However, the genetic mechanisms and target genes underlying these loci are largely unknown, as most risk-associated-variants might regulate gene expression in a context-specific manner. Here, we generated a barcode-shared transcriptome and chromatin accessibility map of 117,911 human lung cells from age/sex-matched ever- and never-smokers to profile context-specific gene regulation. Accessible chromatin peak detection identified cell-type-specific candidate cis-regulatory elements (cCREs) from each lung cell type. Colocalization of lung cancer candidate causal variants (CCVs) with these cCREs prioritized the variants for 68% of the GWAS loci, a subset of which was also supported by transcription factor abundance and footprinting. cCRE colocalization and single-cell based trait relevance score nominated epithelial and immune cells as the main cell groups contributing to lung cancer susceptibility. Notably, cCREs of rare proliferating epithelial cell types, such as AT2-proliferating (0.13%) and basal cells (1.8%), overlapped with CCVs, including those in TERT. A multi-level cCRE-gene linking system identified candidate susceptibility genes from 57% of lung cancer loci, including those not detected in tissue- or cell-line-based approaches. cCRE-gene linkage uncovered that adjacent genes expressed in different cell types are correlated with distinct subsets of coinherited CCVs, including JAML and MPZL3 at the 11q23.3 locus. Our data revealed the cell types and contexts where the lung cancer susceptibility genes are functional.

Rapid isolation of intact retinal astrocytes: a novel approach

Astrocytes are a major category of glial support cell in the central nervous system and play a variety of essential roles in both health and disease. As our understanding of the diverse functions of these cells improves, the extent of heterogeneity between astrocyte populations has emerged as a key area of research. Retinal astrocytes, which form the direct cellular environment of retinal ganglion cells somas and axons, undergo a reactive response in both human glaucoma and animal models of the disease, yet their contributions to its pathology and progression remain relatively unknown. This gap in knowledge is largely a function of inadequate isolation techniques, driven in part by the sparseness of these cells and their similarities with the more abundant retinal Müller cells. Here, we present a novel method of isolating retinal astrocytes and enriching their RNA, tested in both normal and ocular hypertensive mice, a common model of experimental glaucoma. Our approach combines a novel enzyme assisted microdissection of retinal astrocytes with selective ribosome immunoprecipitation using the Ribotag method. Our microdissection method is rapid and preserves astrocyte morphology, resulting in a brief post-mortem interval and minimizing loss of RNA from distal regions of these cells. Both microdissection and Ribotag immunoprecipitation require a minimum of specialized equipment or reagents, and by using them in conjunction we are able to achieve > 100-fold enrichment of astrocyte RNA.

Polygenic regression uncovers trait-relevant cellular contexts through pathway activation transformation of single-cell RNA sequencing data

Advances in single-cell RNA sequencing (scRNA-seq) techniques have accelerated functional interpretation of disease-associated variants discovered from genome-wide association studies (GWASs). However, identification of trait-relevant cell populations is often impeded by inherent technical noise and high sparsity in scRNA-seq data. Here, we developed scPagwas, a computational approach that uncovers trait-relevant cellular context by integrating pathway activation transformation of scRNA-seq data and GWAS summary statistics. scPagwas effectively prioritizes trait-relevant genes, which facilitates identification of trait-relevant cell types/populations with high accuracy in extensive simulated and real datasets. Cellular-level association results identified a novel subpopulation of naive CD8+ T cells related to COVID-19 severity and oligodendrocyte progenitor cell and microglia subsets with critical pathways by which genetic variants influence Alzheimer's disease. Overall, our approach provides new insights for the discovery of trait-relevant cell types and improves the mechanistic understanding of disease variants from a pathway perspective.

IL-33 regulates Müller cell-mediated retinal inflammation and neurodegeneration in diabetic retinopathy

Diabetic retinopathy (DR) is characterised by dysfunction of the retinal neurovascular unit, leading to visual impairment and blindness. Müller cells are key components of the retinal neurovascular unit and diabetes has a detrimental impact on these glial cells, triggering progressive neurovascular pathology of DR. Amongst many factors expressed by Müller cells, interleukin-33 (IL-33) has an established immunomodulatory role, and we investigated the role of endogenous IL-33 in DR. The expression of IL-33 in Müller cells increased during diabetes. Wild-type and Il33-/- mice developed equivalent levels of hyperglycaemia and weight loss following streptozotocin-induced diabetes. Electroretinogram a- and b-wave amplitudes, neuroretina thickness, and the numbers of cone photoreceptors and ganglion cells were significantly reduced in Il33-/- diabetic mice compared with those in wild-type counterparts. The Il33-/- diabetic retina also exhibited microglial activation, sustained gliosis, and upregulation of pro-inflammatory cytokines and neurotrophins. Primary Müller cells from Il33-/- mice expressed significantly lower levels of neurotransmitter-related genes (Glul and Slc1a3) and neurotrophin genes (Cntf, Lif, Igf1 and Ngf) under high-glucose conditions. Our results suggest that deletion of IL-33 promotes inflammation and neurodegeneration in DR, and that this cytokine is critical for regulation of glutamate metabolism, neurotransmitter recycling and neurotrophin secretion by Müller cells.

A multi-omics atlas of the human retina at single-cell resolution

Cell classes in the human retina are highly heterogeneous with their abundance varying by several orders of magnitude. Here, we generated and integrated a multi-omics single-cell atlas of the adult human retina, including more than 250,000 nuclei for single-nuclei RNA-seq and 137,000 nuclei for single-nuclei ATAC-seq. Cross-species comparison of the retina atlas among human, monkey, mice, and chicken revealed relatively conserved and non-conserved types. Interestingly, the overall cell heterogeneity in primate retina decreases compared with that of rodent and chicken retina. Through integrative analysis, we identified 35,000 distal cis-element-gene pairs, constructed transcription factor (TF)-target regulons for more than 200 TFs, and partitioned the TFs into distinct co-active modules. We also revealed the heterogeneity of the cis-element-gene relationships in different cell types, even from the same class. Taken together, we present a comprehensive single-cell multi-omics atlas of the human retina as a resource that enables systematic molecular characterization at individual cell-type resolution.

From function to translation: Decoding genetic susceptibility to human diseases via artificial intelligence

While genome-wide association studies (GWAS) have discovered thousands of disease-associated loci, molecular mechanisms for a considerable fraction of the loci remain to be explored. The logical next steps for post-GWAS are interpreting these genetic associations to understand disease etiology (GWAS functional studies) and translating this knowledge into clinical benefits for the patients (GWAS translational studies). Although various datasets and approaches using functional genomics have been developed to facilitate these studies, significant challenges remain due to data heterogeneity, multiplicity, and high dimensionality. To address these challenges, artificial intelligence (AI) technology has demonstrated considerable promise in decoding complex functional datasets and providing novel biological insights into GWAS findings. This perspective first describes the landmark progress driven by AI in interpreting and translating GWAS findings and then outlines specific challenges followed by actionable recommendations related to data availability, model optimization, and interpretation, as well as ethical concerns.

A systems biology approach uncovers novel disease mechanisms in age-related macular degeneration

Age-related macular degeneration (AMD) is a leading cause of blindness, affecting 200 million people worldwide. To identify genes that could be targeted for treatment, we created a molecular atlas at different stages of AMD. Our resource is comprised of RNA sequencing (RNA-seq) and DNA methylation microarrays from bulk macular retinal pigment epithelium (RPE)/choroid of clinically phenotyped normal and AMD donor eyes (n = 85), single-nucleus RNA-seq (164,399 cells), and single-nucleus assay for transposase-accessible chromatin (ATAC)-seq (125,822 cells) from the retina, RPE, and choroid of 6 AMD and 7 control donors. We identified 23 genome-wide significant loci differentially methylated in AMD, over 1,000 differentially expressed genes across different disease stages, and an AMD Müller state distinct from normal or gliosis. Chromatin accessibility peaks in genome-wide association study (GWAS) loci revealed putative causal genes for AMD, including HTRA1 and C6orf223. Our systems biology approach uncovered molecular mechanisms underlying AMD, including regulators of WNT signaling, FRZB and TLE2, as mechanistic players in disease.

Atlas of phosphoinositide signatures in the retina identifies heterogeneity between cell types

Phosphoinositides (PIPs) are a family of minor acidic phospholipids in the cell membrane. Phosphoinositide (PI) kinases and phosphatases can rapidly convert one PIP product into another resulting in the generation of seven distinct PIPs. The retina is a heterogeneous tissue composed of several cell types. In the mammalian genome, around 50 genes encode PI kinases and PI phosphatases; however, there are no studies describing the distribution of these enzymes in the various retinal cell types. Using translating ribosome affinity purification, we have identified the in vivo distribution of PI-converting enzymes from the rod, cone, retinal pigment epithelium (RPE), Müller glia, and retinal ganglion cells, generating a physiological atlas for PI-converting enzyme expression in the retina. The retinal neurons, rods, cones, and RGCs, are characterized by the enrichment of PI-converting enzymes, whereas the Müller glia and RPE are characterized by the depletion of these enzymes. We also found distinct differences between the expression of PI kinases and PI phosphatases in each retinal cell type. Since mutations in PI-converting enzymes are linked to human diseases including retinal diseases, the results of this study will provide a guide for what cell types are likely to be affected by retinal degenerative diseases brought on by changes in PI metabolism.

eQTL Studies: from Bulk Tissues to Single Cells

An expression quantitative trait locus (eQTL) is a chromosomal region where genetic variants are associated with the expression levels of certain genes that can be both nearby or distant. The identifications of eQTLs for different tissues, cell types, and contexts have led to better understanding of the dynamic regulations of gene expressions and implications of functional genes and variants for complex traits and diseases. Although most eQTL studies to date have been performed on data collected from bulk tissues, recent studies have demonstrated the importance of cell-type-specific and context-dependent gene regulations in biological processes and disease mechanisms. In this review, we discuss statistical methods that have been developed to enable the detections of cell-type-specific and context-dependent eQTLs from bulk tissues, purified cell types, and single cells. We also discuss the limitations of the current methods and future research opportunities.

Whole Exome Sequencing Reveals Novel Candidate Genes in Familial Forms of Glaucomatous Neurodegeneration

Glaucoma is the largest cause of irreversible blindness with a multifactorial genetic etiology. This study explores novel genes and gene networks in familial forms of primary open angle glaucoma (POAG) and primary angle closure glaucoma (PACG) to identify rare mutations with high penetrance. Thirty-one samples from nine MYOC-negative families (five POAG and four PACG) underwent whole-exome sequencing and analysis. A set of prioritized genes and variations were screened in an independent validation cohort of 1536 samples and the whole-exome data from 20 sporadic patients. The expression profiles of the candidate genes were analyzed in 17 publicly available expression datasets from ocular tissues and single cells. Rare, deleterious SNVs in AQP5, SRFBP1, CDH6 and FOXM1 from POAG families and in ACACB, RGL3 and LAMA2 from PACG families were found exclusively in glaucoma cases. AQP5, SRFBP1 and CDH6 also revealed significant altered expression in glaucoma in expression datasets. Single-cell expression analysis revealed enrichment of identified candidate genes in retinal ganglion cells and corneal epithelial cells in POAG; whereas for PACG families, retinal ganglion cells and Schwalbe's Line showed enriched expression. Through an unbiased exome-wide search followed by validation, we identified novel candidate genes for familial cases of POAG and PACG. The SRFBP1 gene found in a POAG family is located within the GLC1M locus on Chr5q. Pathway analysis of candidate genes revealed enrichment of extracellular matrix organization in both POAG and PACG.

Impact of the Human Cell Atlas on medicine

Single-cell atlases promise to provide a 'missing link' between genes, diseases and therapies. By identifying the specific cell types, states, programs and contexts where disease-implicated genes act, we will understand the mechanisms of disease at the cellular and tissue levels and can use this understanding to develop powerful disease diagnostics; identify promising new drug targets; predict their efficacy, toxicity and resistance mechanisms; and empower new kinds of therapies, from cancer therapies to regenerative medicine. Here, we lay out a vision for the potential of cell atlases to impact the future of medicine, and describe how advances over the past decade have begun to realize this potential in common complex diseases, infectious diseases (including COVID-19), rare diseases and cancer.

Deep learning approaches for noncoding variant prioritization in neurodegenerative diseases

Determining how noncoding genetic variants contribute to neurodegenerative dementias is fundamental to understanding disease pathogenesis, improving patient prognostication, and developing new clinical treatments. Next generation sequencing technologies have produced vast amounts of genomic data on cell type-specific transcription factor binding, gene expression, and three-dimensional chromatin interactions, with the promise of providing key insights into the biological mechanisms underlying disease. However, this data is highly complex, making it challenging for researchers to interpret, assimilate, and dissect. To this end, deep learning has emerged as a powerful tool for genome analysis that can capture the intricate patterns and dependencies within these large datasets. In this review, we organize and discuss the many unique model architectures, development philosophies, and interpretation methods that have emerged in the last few years with a focus on using deep learning to predict the impact of genetic variants on disease pathogenesis. We highlight both broadly-applicable genomic deep learning methods that can be fine-tuned to disease-specific contexts as well as existing neurodegenerative disease research, with an emphasis on Alzheimer's-specific literature. We conclude with an overview of the future of the field at the intersection of neurodegeneration, genomics, and deep learning.