To investigate changes in 3D genome architecture, we performed Hi-C on primary monocytes isolated from four AH patients (all males) and 4 age-matched healthy controls (3 males and 1 female, Table 1, Supplemental Table 1). Cryo-preserved PBMCs were thawed and CD14+ and CD16+ monocytes were isolated by negative selection. Hi-C was then performed using the Arima kit, and all 8 libraries were sequenced to an average depth of about 570 million reads per sample (Supplemental Table 2). Using Juicer, the reads were aligned to the genome and chromatin conformation was determined at a resolution of 100kb. As expected, most of the chromatin interactions observed were local and intra-chromosomal (Figure 1, Supplemental Figure 1, Supplemental Table 2). As a result, we focused on intra-chromosomal analyses and not inter-chromosomal interactions.

Using HiCRep²³, we measured the correlation coefficient of the chromatin interactions for each chromosome between every sample to assess in an unbiased manner how much variation there is in genome architecture between different patients. For most comparisons, the correlation coefficient between all individuals was very high (>90%), indicative of the high conservation of 3D genome architecture in monocytes regardless of disease status. But for most chromosomes, higher correlations were observed between healthy controls or between AH samples, and not between the two groups (Figure 2), indicative of a change in genome architecture caused by disease. Notable exceptions include chromosomes 4, 11, 16, 19 and the X chromosome. For the X chromosome, one female healthy control sample (HC2) differed significantly from the 7 males, and the 3 HC males were more similar to each other than the 4 AH patients (Supplemental Figure 2). Chromosomes 4, 11, and 19 mostly showed higher similarity by disease, except the female healthy control was more similar to the AH patients. Chromosome 16 did not segregate by disease. These data indicate that while much of the 3D genome architecture of monocytes does not change with disease, there likely exist chromosome specific regions that differ significantly.

Hi-C data reveal regions of the genome in close proximity via the contact frequency between two loci. We can measure disease-specific changes in 3D genome architecture by measuring statistically significant differences in these frequencies. Using multiHiCcompare²⁴, we measured the contact frequency between every pair of 100kb genome windows (region-region pairs) and calculated statistically significant changes between HC and AH. We observed significant change in contact frequency throughout the entire length of each chromosome (Figure 3). To ensure the method was reliable, we randomized the samples 5 times prior to statistical testing and observed almost no significant changes between regions in any of the tests (Supplemental Table 3).

For most chromosomes, the changes in contact frequency caused by disease occurred in region-region pairs that are adjacent to each other, which corresponds to the fact that most Hi-C interactions are between local genomic regions (Supplemental Figure 3)^25,26. For example, interaction data from Chromosome 2 is primarily between adjacent regions and diminishes by distance, and similarly changes in contact frequency caused by disease are also mostly local (Supplemental Figure 4). Chromosome 1 is an exception because it can be separated into 3 major megadomains, and changes in genome structure in chromosome 1 were restricted to these regions, with almost nothing differing between each megadomain (Supplemental Figure 5). Some chromosomes did have long-distance structural changes between chromosome arms, including chromosomes 11, 12, 14, 16, 17, 19, and 20. The telomeric regions of Chromosome 14 had extensive increases in contact frequency with the body of the chromosome in AH. The telomeres of Chromosome 14 contain the T-cell receptor alpha (TRA) locus on one arm and the immunoglobulin heavy chain (IGH) locus on the other.

While there were differences in contact frequency throughout the genome in AH, there were a number of notable “hotspots” that contained a high density of the most significantly altered regions (Figure 4A). To identify these hotspots, we focused only on regions of the genome with exceptionally high statistically significant changes in contact frequency (padj<0.0000001). From this list of we identified 22 “hotspots” regions of the genome with a high frequency of these changes caused by disease (Table 2, Supplemental Table 4). The size of these hotspots varied significantly from 21MB down to less than 1MB. For most hotspots, the differences in contact frequency occurred entirely within the region. Many of the hotspots entirely increased or decreased contact frequency in AH while some contained separate areas that increased or decreased, further indicative of a localized rearrangement in genome structure, with some areas moving closer and others moving further away in disease.

Considering their size, some hotspots contained hundreds of genes, including many genes associated with immune responses and have been associated with AH (Table 2, Supplemental Tables 5 and 6). We performed pathway analysis on all of the genes in the hotspots and in regions with very significant changes in structure (all genes in Supplemental Table 4) and found that these regions are enriched for innate immune signaling (Figure 4B). Many of the hotspots contain large families of genes that are clustered together in the genome. For example, Chromosome 4 (∼67-85M) contains the CXC-chemokines, Chromosome 5 (∼140-145M) contains the protocadherins, and Chromosome 12 (∼65-69M) contains IFNG, IL-22, and IL-26. One hotspot on Chromosome 19 (∼34-56M) contains multiple large families of genes that are involved in diverse aspects of immune regulation including killer-cell immunoglobulin-like receptors (KIRs), leukocyte immunoglobulin-like receptors (LILRs), sialic-acid-binding immunoglobulin-like lectins (Siglecs), C5 complement receptors, and NOD-like family of receptors (NLRPs). Additionally, hotspots also contained other notable immune genes like NFKB1 (Chromosome 4), IFNGR1 (Chromosome 6), Integrin Beta (ITGB1) (Chromosome 1), and NFATC2 (Chromosome 20).

Because hotspots represent regions of the genome that have a large number of structural changes in disease, and they contain important genes involved in innate immunity, these regions were examined with finer detail. Within the hotspots, TAD structure, such as the loss or expansion of a TAD, was unchanged. Instead, most changes in genome architecture were due to the interaction profile within individual TADs and the relationship between the TADs and the surrounding genome.

CXC-chemokines are upregulated in AH, and we previously found increased co-regulation of these genes in response to LPS in AH patients, using single cell data of monocytes²⁰. Zooming into the hotspot on chromosome 4 containing the CXC-chemokine cassette, we see that all of the CXC-chemokine genes exist within a single TAD (Figure 5). Interestingly, while the borders of the TAD did not change, the contact frequency within that TAD increased in AH. Zooming out, we observed an alternating pattern of interactions between this TAD and the neighboring TADs.

We previously found that monocytes in AH have increased expression of CLRs, which are a family of PRRs that sense a wide diversity of PAMPs and DAMPs. Many of these CLRs are found within the NK gene receptor complex on chromosome 12, and in particular, four of these CLRs (Mincle, Dectin-2, Dectin-3 and DCIR) are in close proximity and have highly co-regulated expression in AH. Similar to the CXC-chemokines, TAD structure in this hotspot did not change, but rather there were increased interactions within the individual TADs, increased connections between TADs within the entire NK-gene receptor complex (∼7M-10M), and reduced interaction immediately outside of it (Figure 6). Another hotspot on chromosome 12 at approximately 65-69M, which contains the genes for IL-22, IL-26, and IFNG, had the opposite structure, with fewer connections within the hotspot and more connections to the surrounding genomic landscape (Figure 7).

In AH, CC-chemokines are upregulated in monocytes, similar to CXC-chemokines and many other pro-inflammatory cytokines^20,27. In the genome, the CC-chemokine genes are in two loci separated by about 1.5MB. From the Hi-C data, each CC-chemokine locus is within an individual TAD (Figure 8). While this region of the genome was not a hotspot, there was significant changes in 3D genome architecture, particularly in terms of the relationship between the CC-chemokine loci and the surrounding area, which increased connectivity in disease.

We previously published single-cell RNA-seq data from PBMCs isolated from patients with severe AH and healthy controls, with and without ex vivo LPS challenge (100pg/ml for 24 hours). In response to LPS, both HC and AH monocytes express chemokines, though the expression is much higher in AH. Next, we measured how well correlated gene expression was in single cells, in order to determine if neighboring genes have coordinated expression (Figure 8). In HC, we observed highly co-regulated expression of the CCL-chemokines in one of the loci (CCL3, CCL4, CCL3L1, CCL4L2). In AH, CCL-chemokine expression was highly co-regulated in both loci, including a few more genes in the first locus (CCL23, CCL18, CCL3, CCL4, CCL3L1, CCL4L2) and genes in the second locus (CCL2, CCL7, CCL13). And importantly, expression between both loci were also coordinated, suggesting the change in 3D genome architecture allowed for coordinated expression of these genes.

Enrolled patients had confirmed diagnosis of AH by clinicians at the Cleveland Clinic based on medical history, physical examination, and laboratory results, according to the guidelines of the American College of Gastroenterology [https://gi.org/clinical-guidelines/] (Supplemental Table 1). Healthy controls were recruited from the Clinical Research Unit at the Cleveland Clinic.

PBMCs were isolated from human blood as previously described^20,30. Isolation of mononuclear cells was performed by density gradient centrifugation on Ficoll-Paque PLUS (GE Healthcare, Uppsala, Sweden). 1 mL of freshly collected Buffy Coat was mixed at a ratio of 1:5 (vol/vol) with phosphate buffered saline (PBS) at 37°C and divided and layered onto 8 mL Ficoll-Paque PLUS in two 15 mL conical centrifuge tubes. After centrifugation at 400 × g for 30 min at 20°C (no brake), buffy coat fractions were collected, pooled, resuspended in culture media (Roswell Park Memorial Institute (RPMI)-1640 supplemented with 100 μM Penicillin-streptomycin and 10% fetal bovine serum (FBS)), and centrifuged at 400 × g for 15 min at 20°C. The pellets were resuspended in 8 mL of culture media, counted, and again centrifuged at 400 × g for 8 min at 20°C. Cells were then cryo-preserved by resuspending in freezing media (50% culture media, 40% FBS, 10% dimethyl sulfoxide (DMSO)) at a concentration of 1.5 × 10⁶ cells/mL, and allowed to freeze slowly to −80°C in a styrofoam container. For long-term preservation, cells were stored in liquid nitrogen.

Cryopreserved PBMCs from 4 AH patients and 4 age-matched healthy controls were thawed following the 10x protocol for cryopreserved PBMCs. Monocytes were isolated by negative selection using the EasySep Human Monocyte Enrichment Kit without CD16 Depletion (StemCell Technologies, Cambridge, MA) according to factory instructions. Hi-C was performed with the Arima Genome-wide Hi-C kit (Carlsbad, CA) according to factory instructions. Libraries were pooled and sequenced with an Illumina Novaseq 6000.

Sequencing data for the Hi-C experiments were aligned to the genome (GRC38, release 93) using Juicer³⁵. From the output of Juicer, we summarized the quality of the Hi-C maps using the inter.txt files (Supplementary Table 2). Chromosomal correlations were calculated using HiCRep²³. Differential contact frequencies were measured using multiHiCcompare, at a resolution of 100kb and using the inter.hic files from Juicer²⁴. Hi-C data was visualized using Juicebox to view each of the inter.hic files³⁶. Regions with changes in contact frequency in disease were labeled using data from multiHiCcompare. To ensure these changes in contact frequency are due to disease and not random chance, we randomized the samples and recalculated differential contact frequency and found significantly fewer differences (Supplemental Table 3).

“Hotspots” were defined as regions of the genome with a high frequency of changes in contact frequency. To identify these regions, we took the list of all differential contacts from multiHiCcompare and filtered by a very stringent statistical cutoff of padj<0.0000001. Then we identified regions of the genome with a large number of these differential contacts, which identified 22 “hotspots” (Table 2, Supplemental Table 4). To identify the borders of these regions, we looked more liberally at the full list of differential contacts (padj<0.05). Gene enrichment for genes present in hotspots and other highly differential regions was performed using Metascape³⁷.

Sequencing data was aligned to the Human genome (GRC38, release 93) using cellranger (v3.0.2). All gene expression and clustering analyses were performed using Seurat (3.1.1) as previously described^20,38. Briefly, all samples were first normalized using SCTransform and then filtered to remove low quality cells (nFeature_RNA<4000, nFeature_RNA>200, percent.mt<20, which removes doublets, cells with low reads, and cells with high mitochondrial content))³⁹. All samples were combined using the PrepSCTIntegration and FindIntegrationAnchors functions to find common anchor genes in all samples for all cell types, and then integrated using the IntegrateData function, with all normalizations using the SCT transformed data^38,40. Clustering was performed using RunPCA and RunUMAP, and clusters were identified using FindNeighbors and FindClusters.

For correlation analyses, bigSCale2 was used to calculate correlations using the Z-score algorithm⁴¹. For these analyses, only monocytes were used (clusters labeled CD14_Monocyte1, CD14_Monocyte2, CD14_Monocyte3, and CD16_Monocyte). All gene clusters were determined by first ordering all annotated human genes by chromosome and start codon then finding the desired clusters and selecting an arbitrary set of genes surrounding it (genes that are not hypothesized to be involved) to ensure the entire cluster was obtained. Genes were then filtered for low expression using the criteria that bigSCale2 was unable to determine a correlation coefficient. Heatmaps of the correlation coefficients were made with only the top and bottom 5% of all correlation coefficients shown, to remove noise and isolate the most important correlations.