Proteomics: Molecular Subtypes Guide CTCL Drug Response
Advanced mycosis fungoides and Sezary syndrome get treated from a rotating menu of retinoids, interferon, methotrexate, chemotherapy and newer targeted agents, and no molecular test says who belongs on which arm. Tissue proteomics offers one way in. A group at Peking Union Medical College Hospital profiled 33 lesional biopsies from 31 patients with advanced cutaneous T-cell lymphoma (CTCL), pairing DIA mass spectrometry on laser-microdissected tumor regions with RNA-seq on a subset, then asked whether protein-level structure inside the lesion tracks with how hard the disease is to treat.
Key Takeaways
Tissue proteomics split 33 advanced CTCL lesions into three molecular subtypes: intracellular signaling, metabolic, and extracellular matrix remodeling.
Phospho-AKT staining ran higher in patients who responded to the PI3K-delta inhibitor linperlisib than in those who did not (p = 0.035, seven patients).
CTSB, GSTO1 and WDFY4 separated patients held stable on immunomodulators from those escalated to chemotherapy, each at an AUC above 0.8.
Microdissecting tumor-enriched regions first keeps keratinocyte and stromal protein from diluting the lymphoma signal.
Tissue Proteomics on Microdissected CTCL Lesions
A punch biopsy of lesional skin is a mixed bag: malignant T cells, keratinocytes, fibroblasts, and whatever reactive infiltrate came along. Laser capture microdissection was used to cut tumor-enriched regions out first. DIA analysis identified 5,407 proteins and kept 5,194 after filtering anything missing in over half the samples, with QC injections correlating at 1.00. Seventeen microdissected samples went to RNA-seq and 14 cleared quality control, yielding 20,340 quantified transcripts. Subtyping ran on the proteome alone; the transcriptomic layer checked whether the same biology showed up in mRNA.
Key Findings
Three protein-defined subtypes. Cluster 1 (16 samples) ran on EIF2, mTOR and JAK/STAT signaling, Cluster 2 (9 samples) on redox homeostasis and protein folding, Cluster 3 (8 samples) on matrix remodeling.
Subtype tracked clinical phenotype. Lesion morphology distributed unevenly across clusters (p = 0.001), as did TNMB stage (p = 0.015), yet progression-free survival did not differ (p = 0.94, 25 patients).
A response marker for PI3K-delta blockade. All seven patients given linperlisib plus chidamide fell in the intracellular signaling subtype, and responders showed higher p-AKT optical density (p = 0.035).
Treatment-intensity proteins. Contrasting 11 patients managed on immunomodulators with 18 needing chemotherapy or targeted agents gave 64 differential proteins; CTSB, GSTO1 and WDFY4 went to immunohistochemistry.
GOLGA1 and STIP1 flagged progression. Both separated the 12-patient progression group from the 13 who stayed stable or responded, with staining rates matching the mass spectrometry result.
The layers disagreed in places. Redox and glucose-metabolism features of the metabolic subtype reappeared in the transcriptome; the matrix-remodeling signature did not.
Figure 1
Figure 1. Study workflow. Lesional skin from patients with advanced mycosis fungoides or Sezary syndrome was biopsied, tumor-enriched regions were collected by laser capture microdissection, and the material went to DIA proteomic analysis and transcriptome sequencing. Later steps covered consensus clustering into molecular subtypes, functional annotation, target analysis, and immunohistochemical validation across response and prognosis subgroups. Adapted from Zhang et al. (2026), Frontiers in Oncology.
What the Subtypes Say About Therapy Selection
The PI3K-AKT-mTOR arm is the part with immediate teeth. An earlier trial from this group reported a 59.1% objective response rate for linperlisib with chidamide in relapsed or refractory CTCL, a good number attached to a bad problem: nobody knows in advance who the non-responders will be. Every treated patient here sat in one proteomic subtype, and p-AKT abundance split the two response groups, which points at a pre-treatment stain that could triage the decision.
How This Fits the CTCL Biomarker Picture
Genomic work on CTCL, including the landscape mapped by Choi and colleagues (2015), established the mutational drivers without producing therapy-selection tools. Protein profiling sits closer to the drug target, which is why a phospho-epitope rather than a mutation carried the predictive signal here, the same logic that ran through a breast cancer study where one protein predicted pyrotinib response. My reservation is size: 25 patients with follow-up, seven on the inhibitor, no external cohort. The subtypes are believable; the markers are hypotheses.
From One Biopsy to Two Omics Layers
Reading protein and transcript from the same microdissected region is what makes the concordance argument worth anything; split aliquots run in separate batches would have left the disagreements uninterpretable. Material is the binding constraint here, and that's the tradeoff the single-injection Omni-MS workflow behind our MS-based proteomics services at Dalton was built around, reading several molecular layers from one preparation instead of dividing a scarce sample. Groups scoping multi-omics biomarker discovery on clinical specimens tend to hit that depth-versus-volume question before they hit the statistics.
Frequently Asked Questions
What is proteomics used for in cancer biomarker discovery?
Proteomics measures the proteins actually present in a tissue or biofluid, a step closer to drug action than DNA or RNA. In oncology it defines molecular subtypes, finds predictors of treatment response, and prioritizes targets. Mass spectrometry does the discovery; immunohistochemistry or targeted assays do the confirming.
Can proteomics predict which CTCL patients respond to PI3K inhibitors?
This work suggests it may. Responders to a PI3K-delta inhibitor carried higher phospho-AKT in lesional tissue (p = 0.035), all inside one proteomic subtype. Seven patients is too few to act on clinically, so read it as a hypothesis rather than a companion diagnostic.
How much tissue does a laser capture microdissection proteomics study need?
DIA workflows quantify thousands of proteins from microdissected regions of a few thousand cells, well within a standard skin biopsy. Fresh-frozen tissue, as used here, gives the cleanest proteome analysis, and RNA integrity is the tighter constraint if you also want sequencing.
Conclusion
Advanced CTCL is not one molecular disease, and this three-subtype structure is solid enough to design a prospective study around. Single-cohort AUCs above 0.8 in fewer than 30 patients rarely hold their value on replication, so the individual markers stay provisional. For groups running trials in rare cutaneous lymphomas, the practical move is banking microdissected fresh-frozen tissue now, so proteomic stratification can be tested once response data matures.
Related Reading
See how we run these analyses in one lab: Dalton's multi-omics CRO services.
Citation
Zhang, S., Guo, Z., Liu, Z., Pang, Z., Qi, F., Sun, H., Sun, W., & Liu, J. (2026). Proteomics and transcriptomics reveal molecular subtypes and biomarkers of advanced cutaneous T-cell lymphoma. Frontiers in Oncology, 16, 1849456. https://doi.org/10.3389/fonc.2026.1849456
Note
This blog post summarizes findings from the above-cited research. Figures are adapted from the original publication. For full details, please refer to the source article.
By Seungjun Yeo, CEO at Dalton Bioanalytics. Specializing in multi-omics mass spectrometry for drug discovery and biomarker research.