0667: Correlation Matrices Visualize Differential Degree of Cell and Pathway Heterogeneity in Skin of Cutaneous Lupus Erythematosus Treatment Subgroups

Felix Chin¹, Thomas Vazquez², Josh Dan³, DeAnna Diaz⁴, Grant Sprow⁵, Jay Patel⁶, Nilesh Kodali⁷, Rui Feng⁸ and Victoria Werth⁹, ¹University of Pennsylvania Perelman School of Medicine, Philadelphia, PA, ²FIU Wertheim College of Medicine, Virginia Beach, VA, ³Philadelphia VAMC, Philadelphia, PA, USA and Department of Dermatology, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, ⁴Philadelphia College of Medicine, Philadelphia, PA, ⁵Albert Einstein College of Medicine, Philadelphia, PA, ⁶Philadelphia VAMC, Philadelphia, PA, USA and Department of Dermatology, Perelman School of Medicine at the University of Pennsylvania, Philadelphia, PN, ⁷New Jersey Medical School, Coppell, TX, ⁸University of Pennsylvania, Philadelphia, ⁹University of Pennsylvania and Corporal Michael J. Crescenz VA Medical Center, Philadelphia, PA

Background/Purpose: First-line treatment for cutaneous lupus erythematosus involves the use of antimalarials. Treatment response is highly variable with some patients responding well to hydroxychloroquine (HCQ), some necessitating the addition of quinacrine (QC) and others refractory to treatment from both (NR). Here we compared the immune infiltrate composition of CLE skin between 3 different treatment response groups with imaging mass cytometry. We then used correlation matrices to highlight the differences in pathway heterogeneity between each of the groups. We also employed unsupervised learning to visualize the distinct cell-dominant subtypes in each of the groups.

Methods: Infiltrate composition of CLE skin biopsies were characterized using imaging mass cytometry, where we stained biopsies (12 HCQ, 11 QC, 20 NR) with two separate panels of 37-metal conjugated antibodies, which enabled the measurement of different cell types, cytokines and pathway proteins within the tissue sample. Correlation matrices of cytokines/pathways with superimposed hierarchical clustering were constructed for each of the three treatment groups to visualize correlation strength for all markers simultaneously. Cell measurements were scaled within each cell type and then subjected to K-means clustering. Results of cell clustering were visualized using horizontally scaled heat maps, which demonstrated cellular predominance within each patient's biopsy.

Results: Correlation matrices revealed a single large correlation cluster in HCQ, two medium clusters in QC and many clusters in NR. This indicates relatively homogenous cytokine/pathway activity in HCQ, two separate mechanisms in QC and significant heterogeneity in NR. Cell clustering heatmaps demonstrated heavy CD4/CD8 dominance in HCQ, two noteworthy QC subgroups, one being dominated by CD8 cells (61% vs. 7%, p=0.1) and the other by cDC (62% vs. 15%, p=0.1), and three noteworthy NR subgroups being CD8-dominant (48% vs. 9%, p< 0.00001), cDC-dominant (38% vs. 10%, p< 0.001) and CD14+CD16+ dominant (45% vs. 4%, p< 0.00001). Furthermore, CD4/CD8 dominance in HCQ appeared to be positively associated with general increased levels of pathways.

Conclusion:
HCQ responsive patients may share a common origin for pathway activity whereas QC patients potentially possess 2 separate mechanisms responsive to quinacrine therapy. NR patients showed substantial heterogeneity, which implies multiple possible origins of pathway activity thus supporting the need for personalized therapy. In terms of cells, tandem CD4/CD8 presence could potentially be used as a predictor for hydroxychloroquine-responsiveness. In QC and NR subgroups, most patients displayed single cell type predominance, most notably cDC and CD8, which suggests potential utility for personalized medicine in targeting single cell types.
Fig 1. Size of each dot is a reflection of correlation strength (red = positive). Correlations were grouped using hierarchical clustering. HCQ shows single large cluster indicating pathways coexist together homogeneously. QC shows two main clusters suggesting two potential mechanisms responsive to quinacrine. NR shows no clusters indicating significant heterogeneity.


Fig 1. K-means clustered heatmaps illustrating the different cell clusters that exist within the three treatment response groups. Each row represents a single patient. Greater CD4/CD8 activity appeared to correlate with general pathway activity in the HCQ group. QC and NR clusters contain similarities in terms of single-cell dominance, both showing CD8 and cDC-predominant subgroups..

Disclosures: F. Chin, None; T. Vazquez, None; J. Dan, None; D. Diaz, None; G. Sprow, None; J. Patel, None; N. Kodali, None; R. Feng, None; V. Werth, AbbVie/Abbott, Amgen, AstraZeneca, Bristol Myers Squibb (BMS), Celgene, Eli Lilly, Genentech, GlaxoSmithKline (GSK), Janssen, Merck/MSD, Gilead, Novartis, Pfizer, Rome Pharmaceuticals, Horizon Therapeutics, Regeneron, argenx, CSL Behring, AnaptysBio, Biogen, Corbus, EMD Serono, Galderma, Nektar, Octapharma, Principia, Resolve, Sanofi, Syntimmune, Viela.