Mapping disease regulatory circuits at cell-type resolution from single-cell multiomics data

Abstract

Resolving chromatin-remodeling-linked gene expression changes at cell-type resolution is important for understanding disease states. Here we describe MAGICAL (Multiome Accessibility Gene Integration Calling and Looping), a hierarchical Bayesian approach that leverages paired single-cell RNA sequencing and single-cell transposase-accessible chromatin sequencing from different conditions to map disease-associated transcription factors, chromatin sites, and genes as regulatory circuits. By simultaneously modeling signal variation across cells and conditions in both omics data types, MAGICAL achieved high accuracy on circuit inference. We applied MAGICAL to study Staphylococcus aureus sepsis from peripheral blood mononuclear single-cell data that we generated from subjects with bloodstream infection and uninfected controls. MAGICAL identified sepsis-associated regulatory circuits predominantly in CD14 monocytes, known to be activated by bacterial sepsis. We addressed the challenging problem of distinguishing host regulatory circuit responses to methicillin-resistant and methicillin-susceptible S. aureus infections. Although differential expression analysis failed to show predictive value, MAGICAL identified epigenetic circuit biomarkers that distinguished methicillin-resistant from methicillin-susceptible S. aureus infections.

Fig. 1 |. Overview of MAGICAL for mapping disease-associated regulatory circuits from scRNA-seq and scATAC-seq data.

a, Disease-modulated regulatory circutis. In the 3D genome, the altered gene expression in cells between disease and control conditions can be attributed to the chromatin accessibility changes of proximal and distal chromatin sites regulated by TFs. b, MAGICAL framework. To identify disease-associated regulatory circuits in a selected cell type (including ATAC assay cells and RNA assay cells from samples being compared), MAGICAL selects DAS as candidate chromatin sites (peaks) and DEG as candidate genes. Then, the filtered ATAC data and RNA data of DAS and DEG are used as input to a hierarchical Bayesian framework pre-embedded with the prior TF motifs and TAD boundaries. The chromatin activity A is modeled as a linear combination of TF–peak binding confidence B and the hidden TF activity T, with data noise contamination N_A. The gene expression R is modeled as a linear combination of B, T, and peak–gene looping confidence L, with data noise contamination N_R. MAGICAL estimates the posterior probabilities P(B|A,T), P(T|A,B), and P(L|R,B,T) by iteratively sampling variables B, T, and L to optimize against the data noise N_A and N_R in both modalities. Finally, regulatory circuits with high posterior probabilities of B and L (for example, a high confidence circuit with inferred interactions between TF1, Site2, and Gene1) are selected. c, Results validation. We evaluate the accuracy and cell-type specificity of the inferred peak–gene looping interactions by checking their enrichment with cell-type-matched chromatin interactions in Hi-C experiments. For the identified TFs, chromatin sites, and genes in circuits, we checked the accuracy of each using independent ChIP-seq, scATAC-seq, and scRNA-seq data. Finally, as a demonstration of the utility of MAGICAL, we used the circuit target genes as features to predict disease states.

Fig. 2 |. Validation of COVID-19-associated circuit chromatin sites and genes.

a, We applied MAGICAL to a COVID-19 PBMC single-cell multiomics dataset and identified circuits for the clinical mild and severe groups. We validated the MAGICAL-selected circuit sites and genes using newly generated and independent COVID-19 single-cell datasets. b, UMAPs of a newly generated independent scATAC-seq dataset including 16,000 cells from six people with COVID-19 and 9,000 cells from three controls showed chromatin accessibility changes in CD8 TEM, CD14 Mono, and NK cell types. c,d, The precision of MAGICAL-selected circuit sites is significantly higher than that of the original DAS, the nearest DAS to DEG, or all DAS in the same TAD with DEG. e,f, The precision of circuit genes are significantly higher than that of DEG. c,e, For mild COVID-19, MAGICAL identified 645 sites in CD8 TEM, 599 sites in CD14 Mono, and 148 sites in NK, regulating 153 genes, 183 genes, and 60 genes, respectively. d,f, For severe COVID-19, MAGICAL identified 78 sites, 202 sites, and 62 sites in the three cell types, regulating 25 genes, 81 genes, and 26 genes, respectively. c–f, Precision is defined as the proportion of the selected sites and genes to be differentially accessible and differentially expressed in the same cell type between infection and control conditions in independent datasets. Results are presented as bar plots where the heights represent the precision and the error bars represent the 95% confidence interval. Significance is evaluated using a two-sided Fisher’s exact test and P values between bars are shown.

Fig. 3 |. MAGICAL accurately identified distal regulatory chromatin sites and epi-driven genes associated with S. aureus infection.

a, We collected PBMC samples from 10 subjects infected with MRSA, 11 with MSSA, and 23 uninfected control subjects and generated sample-paired scRNA-seq and scATAC-seq data using separate assays. b, UMAP of integrated scRNA-seq data with 18 PBMC cell subtypes. c, UMAP of integrated scATAC-seq data with 13 PBMC cell subtypes. Under-represented subtypes including cDC1, CD4 TEM, CD8 CTL, pDC, and Plasmablast (representing less than 5% of cells in the scRNA-seq data in total), were not recovered from the scATAC-seq data. d, The number of MAGICAL-identified regulatory circuits in contrast analysis for each cell type. e, The number of shared and specific circuits between cell types. f, Enrichment of circuit peak–gene interactions in each cell type with cell-type-specific pcHi-C interactions. g–i, We specifically analyzed MAGICAL-identified regulatory circuits for CD14 Mono. g, TF motif enrichment analysis in circuit sites showed that AP-1 proteins are mostly significantly enriched at chromatin regions with increased accessibility in the infection condition. The log₂(FC) is calculated for each TF by dividing the number of binding sites with increased chromatin activity in the infection condition by the number of sites with decreased activity. h, In total, 633 circuit sites were identified by MAGICAL. Compared with all accessible chromatin sites, an increased proportion of circuit sites were in the range of 15 kb to 25 kb relative to gene TSS. In the curve, the center points represent the FC between the proportions of circuit sites and background sites at each location. The upper and lower points represent the 95% confidence interval. i, The circuit genes were significantly enriched with experimentally confirmed epigenetically driven genes (epi-genes) in monocytes. All significance was assessed using adjusted P values from a one-sided hypergeometric test.

Fig. 4 |. MAGICAL-identified circuit genes robustly predict S. aureus infection and bacteria antibody sensitivity.

a, Circuit genes in common to MRSA and MSSA infections achieved a near-perfect classification of S. aureus infected and uninfected samples in multiple independent datasets (one adult dataset and two pediatric datasets). b, Circuit genes that differed between MRSA and MSSA showed predictive value of antibiotic sensitivity in independent patient samples (three pediatric datasets).