The Hidradenitis Suppurativa ‘Omics Database (HS-OmicsDB)

To the Editor:

Hidradenitis suppurativa (HS) is a chronic inflammatory skin disease that has among the greatest patient-reported impact on quality of life of any skin condition (Matusiak et al., 2010, Onderdijk et al., 2013). Patients with HS develop exceptionally painful abscesses and lesions. The worldwide prevalence of HS is estimated to be about 0.4%, and the condition trends with ethnic background and family history, pointing to a key role for inheritance (Jfri et al., 2021, Sachdeva et al., 2021).

Recent studies point toward HS as an autoinflammatory keratinization disease, and a recent explosion of transcriptomic studies have identified numerous inflammatory pathways and cell-types involved (van Straalen et al., 2022). The growing list of cellular and molecular mediators in HS implies pathogenetic heterogeneity defined by a penchant for a specific immune dysregulation that may be influenced by an individual’s genetic background and environment. In fact, recent data from our lab has shown that the peripheral blood immunomes of HS patients are heterogeneous (Submitted manuscript). To date, no resources exist to reconcile this heterogeneity or elucidate the functional differences between upregulated cytokines in HS. Single-cell technologies enable resolution of cell-specific disease features and have proven fruitful in uncovering novel aspects of autoinflammatory diseases. Furthermore, meta-analyses of transcriptomic data has been proved fruitful in defining unique features of other inflammatory skin diseases (Martinez et al., 2022).

Here we present HS ‘omics database (HS-OmicsDB, https://shiny.hfhs.org/hsomicsdb/), an initiative to integrate the growing body of ‘omics data’ collected from studies of HS. The goal is to provide a tool for the HS research community to enhance investigations into the etiology of HS specific disease features. Our database compiles publicly available bulk-RNA sequencing (Gudjonsson et al., 2020, Navrazhina et al., 2022) and single cell RNA sequencing data (Gudjonsson et al., 2020, Mariottoni et al., 2021) deposited on the gene expression omnibus (GEO). Only data with deposited in fastq from high-throughput bulk-RNA sequencing experiments were included to minimize variation from datasets generated by distinct technologies. In the first iteration of HS-OmicsDB, we only include biologic naïve patients. Thus our bulk RNAseq includes samples from 5 GEO datasets (GSE154773 (Gudjonsson et al., 2020), GSE155176 (Lowe et al., 2020), GSE151243 (Unpublished), GSE189266 (Navrazhina et al., 2022), GSE213761 (Unpublished)). Only skin biopsy samples from untreated HS patients were included. Users can explore genes of interest to determine tissue-specific and cell-specific gene expression.

We aggregated bulk-RNAseq samples and aligned to GRCh38 using HISAT2 (Kim et al., 2019) with default parameters and batch-correction was performed with ComBatseq (Zhang et al., 2020) (Supplemental Methods). Principle component analysis shows minimal to no batch-effect (Figure S1A–B). Raw counts were quantified using featurecounts from the subread linux package (Liao et al., 2019) (Figure 1A). 73 lesional (L), 28 perilesional (PL), 7 0nonlesional (NL) and 19 healthy control (HC) samples are included in version 1 of our database. We incorporated two built in analyses: 1) expression of a gene of interest can be compared between L, PL, NL and HC skin; and 2) gene-correlation analysis, where all genes that significantly correlate with a gene of interest are identified by a Bonferroni-adjusted p-value (FDR) < 0.05 and can be exported for utilization with gene enrichment tools (e.g. ingenuity pathway analysis (IPA)).

Figure 1. Bulk transcriptomic meta-analysis using HS-OmicsDB.

(A) HSOmicsDB workflow for the aggregation of bulk and single-cell transcriptomic data. (B) Tissue-based expression of IL36A, IL17A, and IL6 in lesional (L), perilesional (PL), non-lesional (NL) and healthy control (HC) skin. (C) Heatmaps of genes that are significantly correlated with the designated cytokine. Samples are ordered based on the expression of the designated cytokine. (D) Unique and shared genes from gene-correlation analysis. (E) Pairwise comparison of pathways enriched from significantly correlated unique genes. *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. Natural Killer (NK); Interleukin (IL)

Using these analyses, we compared three inflammatory cytokines: IL36A, IL17A, and IL6. We found all three cytokines were upregulated in HS-L over at least one of the other tissue types (Figure 1B). Some HC skin samples had high expression of IL6. Correlation analysis found that IL36A, IL17A, and IL6 had 456, 859, and 418 significantly correlated genes, respectively (FDR < 0.05, |r| > 0.5) (Figure 1C). Comparing these gene lists 119 genes were found shared between these three gene lists (Figure 1D). IL6 and IL17 shared 87 genes, IL17A and IL36A shared 211 genes. IL6 and IL36 shared only 12 genes, highlighting divergent signaling pathways affected by these cytokines. Using IPA to enrich unique significantly correlated genes, we found that these cytokines may regulate distinct biological processes and signaling pathways. Activin inhibin signaling was more enriched in IL6 related genes over IL36 and fibrosis pathway was more enriched in IL6 over IL17 related genes. Compared to IL17, IL36 related genes were enriched for interferon signaling. IL17 related genes showed a strong enrichment for NK cell signaling and phagosome formation. This highlights the feasibility of HS-OmicsDB as a platform for dissecting potential downstream functions of specific cytokines (or any genes of interest) in HS studies. Meta-analyses such as this could support ongoing investigations unraveling the role of various effector molecules, cytokines, and chemokines in the pathogenesis of HS.

We also incorporated single-cell RNA sequencing data from three publicly available datasets (GSE154775. GSE175990, and GSE155850) (Gudjonsson et al., 2020, Lowe et al., 2020, Mariottoni et al., 2021) that assayed whole skin biopsies from patients with HS. Samples were aligned to GRCh38 using CellRanger v6.1.1. Finally, cells were combined, clustered, and annotated using a combination of manual annotation and label transferring from the human skin atlas via the Harmony and Seurat packages (Korsunsky et al., 2019, Reynolds et al., 2021, Stuart et al., 2019) (Figure 2A–B; Figure S1C). Genes of interest are simultaneously plotted on the UMAP and violin plots alongside bulk-RNA sequencing to provide insight into expression of genes at a cell-specific level.

Figure 2. Meta-analysis of aggregated single-cell data from HS identifies dysregulated interactions potentially involved in the formation of dermal tunnels.

(A) Annotated uniform manifold approximation and projection (UMAP) plot of single-cell data derived from whole HS skin. (B) Heatmap of normalized gene expression highlighting marker genes and gene clusters for UMAP annotations. (C) Interaction frequency from CellphoneDB output from healthy control samples from the Human Skin Atlas. (D) Interaction frequency from CellphoneDB output for HSOmicsDBV.1. (E) Top five ligand-receptor interactions where Basal-stem KCs are the receiver for healthy control samples from the human skin atlas. (F) Top five ligand-receptor interactions where Basal-stem KCs are the receiver for healthy control samples from HS-OmicsDB.

As an example, we employed CellphoneDB (Efremova et al., 2020) to compare ligand-receptor (L-R) interactions between HS and HC cells (from the human skin atlas) (Reynolds et al., 2021). This analysis highlighted significant dysregulation of cell-cell communications in HS skin (Figure 2C & D). Notably in HS we noticed a marked increase in the interactions between basal-stem keratinocytes (sKCs; Krt5⁺ Krt15⁺ Krt10⁻) with endothelial cells (ECs), fibroblasts, macrophages (Macs) and Langerhans cells (LCs). Further analyzing the signals received by sKCs in this dataset showed enrichment of distinct L-R pairs between HC and HS. Macrophage migration inhibitory factor (MIF) derived from ECs, Macs, and LCs are among the top 5 L-R interactions in HS and absent in HC (Figure 2E & 2F). Previous studies have found MIF is increased in psoriasiform dermatitis, where it was found to be critical for KC proliferation (Bezdek et al., 2018). In the context of HS, the increased interactions between sKCs, ECs, fibroblasts, Macs, and LCs may be indicative of interactions present in dermal tunnels. Additionally, there could be a role for MIF-signaling on sKCs that may be involved in the formation of tunnels.

In the wave of thriving HS research, this initiative could serve as a valuable resource for future investigations. HS-OmicsDB will prove to be a valuable validation and hypothesis generating tool in the study of HS. Future versions will incorporate the growing number of ongoing HS studies employing multiple ‘omics technologies (e.g. proteomics, lipidomics, mass cytometry, spatial sequencing etc.) to further enhance research in the HS community. We hope to build this resource to provide a ‘Rosetta stone’ of the complex cutaneous dysregulation that occurs in HS by integrating data from multiple ‘omics technologies, that is easy to be use and flexible in its application. Such a tool will help build an integrative view of the dynamic process that occurs throughout the progression of HS.

Abbreviations:

HS

Hidradenitis Suppurativa

HC

Healthy Control

L-R

Ligand – Receptor

IPA

Ingenuity Pathway Analysis

L

Lesion

PL

Perilesion

NL

Nonlesion

IL

Interleukin

KC

Keratinocyte

EC

Endothelial Cells

LC

Langerhans Cells

Macs

Macrophages

Data Availability

Data presented herein is freely accessible at https://shiny.hfhs.org/hsomicsdb/. For application of aggregated data with third-party packages the .RDS files are available upon request.