Cell-type-specific biomarkers of systemic sclerosis disease severity capture cell-intrinsic and cell-extrinsic circuits

Introduction:: Fibrotic skin disease is a sine qua non of diffuse cutaneous systemic sclerosis (dcSSc), the disease subset more commonly associated with most severe complications, including scleroderma renal crisis, joint contractures, and interstitial lung disease, as well as disease related mortality. Disease markers correlating with skin disease severity, as assessed by the modified Rodnan skin score (MRSS) have provided insight into disease pathogenesis and also biomarkers for clinical trials. In previous studies, examining gene expression in RNA extracted from entire biopsies (bulk RNA), we have shown that markers of fibroblasts, myofibroblasts, and macrophages correlate with the MRSS(1-3). These studies suggested that macrophage and fibroblast cell populations are particularly important in promoting disease. In this study, we sought to use scRNA-seq coupled to multivariate machine learning approaches to delve deeper into cell-type-specific changes that characterize patients across the disease severity spectrum. These analyses provide a deeper understanding of cell-type-specific biomarkers of disease severity, beyond correlates that stratify disease vs healthy. These comparisons also hone in on granular cell-intrinsic as well as cell-extrinsic factors and associated interactions linked to disease severity.


Materials and Methods:: We obtained single-cell RNA sequencing (scRNA-seq) data for 24 patients dcSSc. These patients have been previously described(9, 10). ScRNA-seq data from 24 SSc patients all using the same V2 chemistry was processed using a standard analytic pipeline comprising alignment via CellRanger and post-processing (dimensionality reduction and clustering) using Seurat. We used a systematic set of filters to remove a) unhealthy cells, b) dropouts and c) doublets. We then conceptually extended a predictive machine learning approach that we had previously used to identify correlates of natural and vaccine-induced immunity in a wide range of infectious disease contexts including HIV, malaria and tuberculosis. Our approach used a two-step feature selection strategy that takes into account the high dimensionality of the dataset and helps avoid overfitting via the use of stringent L1 regularization. Our goal was to determine a minimal number of cell-type specific biomarkers that best predict SSc severity as quantified by the MRSS. While our cohort included healthy controls, they were not used in the identification of biomarkers of SSc severity. This is because we were well powered to identify biomarkers of SSc severity, but not adequately powered with this cohort to identify differences between healthy and SSc. Further, correlates that can discriminate between SSc and healthy are already well characterized. Our study focuses on the more difficult and underexplored problem of identifying robust multivariate cell-type-specific signatures of SSc severity.


Results, Conclusions, and Discussions:: We found that the uncovered cell-type-specific predictive biomarkers of MRSS included previously implicated genes in fibroblast and myeloid cell subsets (e.g., SFPR2+ fibroblasts and monocytes), as well as novel gene biomarkers of MRSS, especially in keratinocytes. Correlation network analyses unveiled novel cross-talk between immune pathways as well as implicated keratinocytes in addition to fibroblast and myeloid cells as key cell-types involved in SSc pathogenesis. We also compared the predictive power of models built on transcript abundances from individual cell subsets as opposed to models built using transcript abundances across all cells and found that keratinocytes and fibroblasts were the most predictive (better than myeloid cells or any other cell subsets). Intriguingly, although the keratinocyte and fibroblast models are predictive of SSc severity, they do not perform as well as the overall model. Our results suggest that while individual cell-type-specific biomarkers are informative of SSc severity, a pan cell-type analysis can identify more robust biomarkers. These findings re-affirm the need to perform pan-cell-type analyses as both cell-extrinsic and cell-intrinsic biomarkers are associated with SSc severity. We then validated the uncovered association of key gene expression and protein markers in keratinocytes, KRT6A and S100A8, with SSc skin disease severity. Our global systems analyses uncover previous uncharacterized cell-intrinsic and cell-extrinsic signaling co-expression networks underlying SSc severity that involve keratinocytes, myeloid cells and fibroblasts.


Acknowledgements (Optional): :

References (Optional): :