Abstract
Systemic mastocytosis (SM) is a rare clonal disorder marked by the pathological accumulation of mast cells, key effectors of inflammation. Despite the central role of inflammation in SM pathogenesis, the specific soluble mediators that correlate with clinical outcomes and disease burden remain poorly defined. High-throughput proteomic profiling offers an unbiased approach to quantify a wide range of circulating inflammatory markers. When combined with machine learning, this approach can uncover novel biomarker signatures that reflect the clinical and biological heterogeneity of SM.
We collected baseline (at the time of diagnosis) plasma samples from 82 SM patients seen at MD Anderson (2007-2024). Inflammatory protein profiling was performed using the NULISAseq™ Inflammation Panel 250, an NGS-based proximity-ligation assay with attomolar (10⁻¹⁸) sensitivity. Associations between protein levels and clinical features or outcomes were evaluated using Spearman correlation, Wilcoxon, and Kruskal–Wallis tests. Differential biomarkers between advanced and non-advanced SM were identified via Wilcoxon rank-sum tests with Benjamini–Hochberg correction. We then trained an XGBoost model on these biomarkers to classify disease severity, and assessed biomarker importance via SHAP values.
The cohort included 82 adult patients with WHO-confirmed SM (mean age 49.7; range 19.6–78.8; 37.8% male). WHO-defined subtypes were ISM (75.6%), SSM (2.4%), ASM (9.8%), and SM-AHN (12.2%). Advanced SM (ASM, SM-AHN, MCL) had higher age at diagnosis (60.3 vs. 46.8 years; p<0.001), LDH (345.8 vs. 257.0 IU/L; p=0.019), and uric acid (6.4 vs. 4.9 mg/dL; p=0.019). Non-significant trends were observed in alkaline phosphatase (167.6 vs. 96.7 U/L), hemoglobin (12.6 vs. 13.6 g/dL), and serum tryptase (117.8 vs. 72.6 ng/mL). Among patients with available data, bone involvement (osteolytic lesions or fractures) was present in 32.9% (26/79), hepatosplenomegaly in 29.8% (14/47), ascites in 9.0% (7/78), and weight loss in 23.1% (18/78); organ findings were based on physical exam or imaging.
Building on the plasma proteomic dataset generated from our 82 SM patients using the 250-protein NGS-based inflammatory panel of NULISA, we identified distinct inflammatory protein signatures (FDR < 0.08) associated with clinical outcomes in SM. Bone involvements were linked to CCL24, SPP1, CD80, and TNFSF8, all involved in osteoclast activity and bone turnover. Ascites was associated with IL-6, IL-10, SPP1, TIMP1, VCAM1, and LIF - proteins previously linked to ascites in cancer or inflammatory conditions - as well as novel signals including CALCA, SCG2, and OSMR. Hepatosplenomegaly was associated with IL1RL1, IL6ST, MMP3, and CCL14, alongside exploratory associations with CD93 and SLAMF1. Weight loss showed associations with CHI3L1, AREG, IL-2, and TNFSF11.
We then aimed to explore key biomarkers that stratify disease severity. We first identified 76 of 250 proteins as significantly different between advanced and non-advanced SM (FDR <0.05). An XGBoost-based machine learning classifier was trained on the top 15 key biomarkers (FDR < 0.005), and achieved 81.04% classification accuracy, outperforming the clinical-data-only model. The SHAP, a model-agnostic interpretation method, revealed that 6 inflammatory/immune-related proteins - IL1R1, IL1RL1, OSMR, SPP1, AREG, and CHI3L1 - contributed over 80% of predictive weight. IL1R1 and IL1RL1 (IL-1 receptor family) have previously been linked to severe inflammatory states and were elevated in advanced SM. OSMR, which forms the type II oncostatin M receptor complex with the IL-6 signal transducer, supports the established role of IL-6 in SM pathogenesis. The newly identified SPP1 (osteopontin) is involved in bone remodeling and cytokine regulation, while AREG and CHI3L1 are known to modulate inflammation and tissue repair.
By integrating high-throughput plasma proteomics with machine learning, we uncovered distinct baseline inflammatory protein signatures associated with clinical features and disease status in SM. Notably, SPP1, CHI3L1, and AREG were consistently linked to both clinical outcomes and disease severity, highlighting their potential as novel blood-based biomarkers. These proteins may represent previously unrecognized contributors to SM biology and offer new opportunities for diagnostic and prognostic refinement. All findings are being actively validated in a prospective cohort.
