Journal Publication

Serum three-protein score predicts immunotherapy response in NSCLC

A 72-patient proteomic analysis identified SERPINE2, DAZAP1 and MGAT4B as a pre-treatment blood signature that tracked checkpoint-inhibitor response and survival in non-small cell lung cancer.

A retrospective proteomic analysis of pre-treatment serum from 72 NSCLC patients identified a three-protein risk score that predicted response to immune checkpoint inhibitor therapy.

By AppliedXLJuly 17, 2026

Executive Summary

Researchers profiled pre-treatment blood serum from NSCLC patients receiving checkpoint inhibitor therapy to find a protein signature that could flag likely responders before treatment begins.
A composite score built from three serum proteins distinguished responders from non-responders with high accuracy, and separately tracked with how long patients survived and stayed progression-free.
Current biomarkers for immunotherapy response rely mostly on tumor tissue, which is invasive to obtain and does not always predict benefit, so a blood-based alternative addresses a real gap in how patients are selected for treatment.
The result comes from one cohort within a single retrospective study, so its accuracy in a broader, independently collected patient population remains to be shown.

The stake

Immune checkpoint inhibitors have extended survival for a portion of non-small cell lung cancer (NSCLC) patients, but many do not benefit, and the biomarkers used to identify likely responders are largely drawn from tumor tissue, which carries sampling limitations and requires an invasive biopsy. Researchers set out to test whether a minimally invasive blood-based signature could do the same job. "Predictive biomarkers based on tissue of origin of the tumor have their limitations, and thus there is a need for and minimally invasive predictive biomarkers," the study states. A

How it was done

The analysis was a retrospective extension of the TD-FOREKNOW trial, using deep proteomic profiling on pre-treatment serum from 72 NSCLC patients who received neoadjuvant immunotherapy plus chemotherapy. Mass spectrometry using data-independent acquisition quantified 1,802 serum proteins, of which 59 were differentially expressed between outcome groups. Univariate logistic regression followed by LASSO regression (a variable-selection statistical method) narrowed the panel to three candidates, SERPINE2, DAZAP1, and MGAT4B, whose baseline serum levels correlated with treatment response. The three-protein composite was then tested against both treatment response and survival outcomes in the same cohort. A

The result

The three-protein risk score predicted immunotherapy response with an area under the curve (AUC) of 0.946 (95% CI: 0.874-1.000), a level of discrimination close to perfect classification within this cohort. Patients classified as low-risk by the score had a lower hazard of progression, with a hazard ratio of 0.13 (95% CI: 0.04-0.46, P=0.002) for progression-free survival, and a hazard ratio of 0.14 (95% CI: 0.03-0.62, P=0.033) for overall survival, compared with high-risk patients. Both the response-prediction and the survival separation point in the same direction: patients whose baseline serum profile matched the low-risk pattern did better on treatment by two independent measures. A

What it changes

The study demonstrates that a pre-treatment blood draw, without a tumor biopsy, can carry a signal for who is likely to benefit from checkpoint inhibitor therapy in NSCLC. It adds serum proteomics to the set of liquid-biopsy approaches being tested as an alternative or complement to tissue-based predictive testing, and it identifies three specific candidate proteins, SERPINE2, DAZAP1, and MGAT4B, for further mechanistic and validation work. The authors describe the approach as demonstrating the prospect of human serum proteomics as a platform for biomarker discovery and clinical prognostic modeling. A

This analysis was produced using AI-assisted reporting systems, AppliedXL data, and official public records. These systems undergo editorial review, quality checks, and regular audits by human experts. Errors may still occur, as with any automated system. Always consult the linked primary sources. Read our AI Editorial Policy.