Combined GWAS and single cell transcriptomics uncover the underlying genes and cell types in disorders of gut-brain interaction

Abstract

Disorders of gut-brain interaction (DGBIs), formerly known as functional gastrointestinal disorders, are extremely common and historically difficult to manage. This is largely because their cellular and molecular mechanisms have remained poorly understood and understudied. One approach to unravel the molecular underpinnings of complex disorders such as DGBIs is performing genome wide association studies (GWASs). However, due to the heterogenous and non-specific nature of GI symptoms, it has been difficult to accurately classify cases and controls. Thus, to perform reliable studies, we need to access large patient populations which has been difficult to date. Here, we leveraged the UK Biobank (UKBB) database, containing genetic and medical record data of over half a million individuals, to perform GWAS for five DGBI categories: functional chest pain, functional diarrhea, functional dyspepsia, functional dysphagia, and functional fecal incontinence. By applying strict inclusion and exclusion criteria, we resolved patient populations and identified genes significantly associated with each condition. Leveraging multiple human single-cell RNA-sequencing datasets, we found that the disease associated genes were highly expressed in enteric neurons, which innervate and control GI functions. Further expression and association testing-based analyses revealed specific enteric neuron subtypes consistently linked with each DGBI. Furthermore, protein-protein interaction analysis of each of the disease associated genes revealed protein networks specific to each DGBI, including hedgehog signaling for functional chest pain and neuronal function and neurotransmission for functional diarrhea and functional dyspepsia. Finally, through retrospective medical record analysis we found that drugs that inhibit these networks are associated with an increased disease risk, including serine/threonine kinase 32B drugs for functional chest pain, solute carrier organic anion transporter family member 4C1, mitogen-activated protein kinase 6, and dual serine/threonine and tyrosine protein kinase drugs for functional dyspepsia, and serotonin transporter drugs for functional diarrhea. This study presents a robust strategy for uncovering the tissues, cell types, and genes involved in DGBIs, presenting novel predictions of the mechanisms underlying these historically intractable and poorly understood diseases.

Figure 1:. Cohort characteristics of functional chest pain, functional diarrhea, functional dyspepsia, functional dysphagia, and functional fecal incontinence in the UK Biobank

A) Schematic of the UKBB GWAS pipeline. B) An overview of cases shared by different disease groups. C) Female to male odds ratios across disease case groups and each corresponding control group. D) Venn diagram illustrating the distribution of cases among different groups. E) Comorbidity analysis results showing the relative distance between disease groups.

Figure 2:. Genes associated with DGBIs map to relevant tissues and functional pathways in the GI

A) Top DGBI variants identified by GWAS and their closest gene. B) Schematic illustration of three categories of DGBI associated gene lists. 1. Genes mapped to SNP hits in our GWAS. 2. Genes filtered in via z-score threshold from gene-based association testing and 3. previously published GWAS. C) Venn diagram illustrating the overlap between DGBI associated gene categories. D) Stacked bar plot representing the composition of DGBI associated genes in each category. E) List of genes/transcripts in at least two out of three DGBI associated gene categories. Cat. 1: mapped genes to significant SNPs; Cat. 2: filtered from gene-based association testing; Cat. 3: from previously published GWAS.

Figure 3:. DGBI associated transcripts are enriched in enteric neurons

A) Summary of gene sets significantly enriched in the DGBI associated genes. The bar graphs highlight nervous system, muscle, and GI related pathways and the pie charts show the proportion of significantly enriched gene sets relating to these categories. B) Schematic illustration of the analysis design to evaluate expression of DGBI molecular features in GI cell types. C) Heatmaps of the average module scores of DGBI associated molecular features in human fetal (top) and adult (bottom) GI cell types using significantly associated transcripts from gene-based testing (z-score ± 3.5) and significantly associated genes from SNP-based testing (mapped genes). D) Heatmap of the gene-set enrichment analysis of DGBI associated genes among differentially expressed genes between clusters of human fetal and adult GI cell types. The values are calculated as the average of association z-scores for each cluster. E) UMAPs of the human fetal and adult GI cell types re-clustered based on their DGBI module scores. F) Heatmap of the average module scores of DGBI associated genes in the human fetal and adult GI cell types re-clustered based on their DGBI module scores. G) Stacked bar plot of the representation of human fetal and adult GI cell types in the clusters of human fetal and adult GI cell types re-clustered based on their DGBI module scores.

Figure 4:. Expression and gene set association testing-based methods identify enteric neuron subtypes highly associated with each DGBI

A) Schematic illustration of the analysis method to identify enteric neuron subtypes associated with DGBIs. B) Schematic illustration of two- and one-step approaches to identifying neurotransmitter and neuropeptide subtypes in single cell transcriptomics datasets. C) Scatter plot showing the relationship between individual enteric neuron subtypes and their combined DGBI score based on module score analysis and DE association gene-set enrichment analysis. Distances are calculated via multidimensional scaling and disease scores are the average score of two methods. (Galanin: GALAN, Endothelin.3: ET-3, Nitrergic: NITRG, Dopamanergic: DOPAM, Glutamatergic: GLUMT, Urocortin: UROCR, Somatostatin: SSTTN, Chromogranin.A: CHRG-A, Calcitonin.2: CTN-2, Neurexophilin.2: NRXPH-2, Proenkephalin: PRNKPH, Neuromedin: NMDIN, CART. Prepropeptide: CARTPP, Substance. P: SUB-P, Neuropeptide. Y: NP-Y, Cholecystokinin: CCK, Cholinergic: CHOLN, Secretogranin.3: SGRN-3, Serotonergic: SRTNR, GABAergic: GABA, Cerebellin.2: CERB-2, Catecholaminergic: CATAMN, Calcitonin: CALC) D) List of the neuronal subtypes with high disease association scores in both analysis approaches and their corresponding diseases.

Figure 5:. Protein-protein interaction analysis identifies proteins and protein pathways linked with DGBI risk

A) STRING protein-protein interaction networks formed from the DGBI associated genes. Protein nodes specific to an enteric neuron subtype are highlighted. The minimum required interaction score was set to 0.4, reflecting medium confidence interactions. B) Schematic illustration of the analysis pipeline to examine links between medications that target the network protein nodes and DGBIs in clinical records. C) Number of cases and controls UCSF extracted from the de-identified clinical database. D) Multivariate logistic regression analysis of the odds of DGBIs in patients treated with network-targeting drugs as well as commonly prescribed drugs for each DGBI as positive controls and multivitamin as negative control.

Figure 6:

Schematic illustration of the study framework