10 April 2020
AACR: Scalable plasma proteomics in early-stage non-small cell lung cancer
AACR Virtual Meeting II – American Association for Cancer Research – June 2020
John E Blume1, William C Manning1, Gregory Troiano1, Asim Siddiqui1, Philip Ma1, Robert Langer2, Vivek Farias2, Omid C. Farokhzad1
1Seer, Redwood City, CA, United States
2Massachusetts Institute of Technology, Boston, MA, United States
Introduction: Though early detection of NSCLC greatly improves prognosis, we lack useful clinical tests. Genomics approaches utilizing cell-free DNA provide suitable specificity but moderate sensitivity for early cancer detection. Plasma proteins have the potential to deliver robust panels of biomarkers for early cancer detection that may be complimentary to genomics markers. Complex workflows, which enable deep and unbiased interrogation of plasma proteins that span 10 orders of magnitude, have made it impractical to efficiently perform robust studies, and consequently, comprehensive proteomic data vastly lags other “omics”.
Herein, we report a multi-NP Proteograph platform that rapidly, reproducibly, deeply, and scalably interrogates proteins from biofluids. In a study of 268 subjects, comparing on average 1779 plasma proteins of NSCLC subjects to healthy and pulmonary co-morbid controls, we identified classification panels comprising proteins with known and unknown roles in NSCLC, offering the promise of new biomarkers for early disease detection.
Methods: Subject plasma samples were grouped into NSCLC stages 1,2,3 (early), NSCLC stage 4 (late), or healthy and pulmonary co-morbid controls, for a randomly selected cohort of 288 age- and gender-matched subjects, and interrogated with a panel of NPs in an efficient automated work-flow. Peptides from NP-bound proteins underwent data-independent-acquisition mass spectrometry. Subject samples were also interrogated using conventional Agilent MARS-14 immunodepletion column, which has historically yielded limited clinical value, to determine differences in depth and types protein coverage achieved as compared with panel of NPs.
Results: On average 1,779 proteins were detected from each of the 268 subject samples vs. 413 from depleted plasma. The healthy vs early NSCLC random classification after depleted plasma protein removal achieved an average AUC of 0.90. Classification of healthy subjects to late NSCLC had an average AUC of 0.98. Comparison of the top features of the NSCLC classifiers to the co-morbid classifier indicated clinically significant differences. Among the former were proteins with both known and unknown roles in NSCLC (OpenTargets), underscoring the value of unbiased proteomic analysis.
Conclusions: We demonstrate the utility of the multi-NP Proteograph platform to deeply profile plasma proteins as novel biomarkers. The performance of the healthy vs. early NSCLC classifier confirms the potential of proteins in early disease detection. Our platform enables deep unbiased plasma protein biomarker profiling that matches genomics workflow throughput and suggests feasibility of parallel large-scale complementary studies of proteins and nucleic acids