Serum Proteomics Reveals Distinct Subtypes Associated with Treatment Response in Idiopathic Multicentric Castleman Disease

Human herpesvirus-8(HHV-8)-negative/idiopathic multicentric Castleman disease (iMCD) is a rare and poorly understood disorder diagnosed in ~1,000 individuals in the USA each year. It involves polyclonal lymphoproliferation, constitutional symptoms, systemic inflammation and an uncontrollable cytokine storm resulting in life-threatening multi-organ failure. Diagnosis and treatment can be difficult due to limited etiological understanding and heterogeneous presentation - clinical, laboratory, and histopathological abnormalities overlap with infectious, autoimmune and oncological diseases. iMCD symptoms and disease progression are largely believed to be driven by interleukin-6 (IL-6). However, approximately 66% of iMCD patients did not respond to anti-IL-6 therapy, siltuximab, the only FDA-approved iMCD therapy, in its phase II study (NCT01024036). Few treatment options exist for anti-IL-6 refractory patients because alternative driver cytokines and signaling pathways are not known. Herein we report the largest study to-date of iMCD serum proteomes with correlative anti-IL6 response data from 92 iMCD patients in disease flare (n=75 of which were collected as part of NCT01024036), in order to: (1) molecularly define iMCD, (2) identify predictors of response to anti-IL6 therapy, and (3) gain insights into the pathogenesis of iMCD. Proteomes of HHV8-positive MCD (n=20), Hodgkin lymphoma (n=20), rheumatoid arthritis (n=20) and healthy individuals (n=44) were also analyzed. Of the ~1,300 analytes measured using SomaLogic SOMAscan, 1,178 passed QC and were included in analyses. Each analyte was log₂ transformed and capped at the 2.5^th and 97.5^th percentiles. Clinical and laboratory data collected at the time of sample draw were used to calculate disease activity following a modified CHAP scale: C-reactive protein, hemoglobin and albumin; missing performance status. Response to siltuximab was determined in NCT01024036. Data analysis was performed using the Medidata Rave Omics machine learning platform and R v3.4.4. Clustering of baseline proteomic data for iMCD patients identified six clusters that ranged in size from seven to 27 subjects. No associations with race, site, sex, age, or batch were found. Analytes identified among the strongest differentiators include cytokines, chemokines and inflammatory molecules. Interestingly, the largest cluster was associated with response to siltuximab (p<0.05; 65% (11/17) vs 19% (5/27) in all others), higher disease activity (p<0.01), and higher IL-6 levels (p<0.01). Analysis of data for the entire study population separated HHV-8-positive MCD, Hodgkin lymphoma, and rheumatoid arthritis into distinct clusters. Of note, iMCD patients did not form a single or unique cluster, reinforcing the heterogeneity of the disease, and a subset of iMCD patients demonstrated similar, but not overlapping, proteomic profiles to those of Hodgkin lymphoma. These results indicate that previously undiscovered proteomically-distinct iMCD subtypes or disease states exist, which can be used to inform treatment options. Given that iMCD may have a sudden and severe onset, predictive markers of anti-IL6 therapy response are critical for timely administration of the correct treatment. Additionally, overlapping proteomic profiles of a subset of iMCD patients with Hodgkin lymphoma provide etiological insights. More broadly, this study represents the first use of high-quality serum proteomics data to study a rare non-malignant lymphoproliferative disorder and to assist with molecularly defining a heterogeneous disease, developing candidate diagnostic biomarkers, predicting response or failure to therapy, and identifying novel candidate therapeutic targets.