Abstract 3612: High-throughput Mass Spectrometry Identifies Peptide Signature Clusters Predictive of Heart Failure Outcome: the Bucindolol Evaluation Survival Trial (BEST) Proteomic Substudy
Background: Single protein biomarkers have been predictive of clinical outcome in patients with congestive heart failure (CHF). We hypothesized that “shotgun” analysis of peptide signatures in the low molecular weight domain (2–20 kDa) using high-throughput, surface-enhanced laser desorption-time of flight (SELDI-TOF) mass spectrometry (MS) could identify biomarkers with enhanced predictive value in patients with chronic CHF.
Methods: We performed SELDI-TOF MS on 592 legacy samples collected for the Genetics Substudy of the Bucindolol Evaluation Survival Trial (BEST) which included subjects with NYHA Class III-IV CHFand left ventricular ejection fraction ≤ 35%. Thawed samples were run in duplicate and spectral analysis of mass/charge ratio amplitude peaks was performed. Spectra were assigned to bins of increasing size according to mass/charge ratio to reduce known effects of instrument drift. Spectral bins were compared to clinical events as adjudicated by the BEST Endpoints Committee which included a CHFoutcome (pump failure death, sudden death preceded by worsening CHF, hospitalization for CHF) and sudden death (SD) with or without preceding CHF. Analysis was performed on duplicate samples using 3 machine learning algorithms (n=6 results/subject) and significant bins were determined using a majority voting method with tie scores considered indeterminate.
Results: Peptide signature clusters chosen for maximum predictive accuracy were found for both CHF and SD outcomes (n=5 and n=4 signatures/ outcome, respectively). CHF and SD cluster ROC curve area for post-collection 60 day event rates were 0.718 and 0.680 vs. 0.667 and 0.626 for N-terminal proBNP (median cutpoint ≥ 632 pg/ml ), respectively. Log rank test analysis demonstrated improved predictive performance for the CHF and SD peptide signature clusters versus N-terminal proBNP (p ≤ 0.0001 and ≤ 0.05, respectively). In a separate validation set of 365 BEST samples, classification was possible in 93% and 87% of subjects for CHF and SD outcomes.
Conclusion: These preliminary data suggest that high-throughput MS-derived peptide signature analysis may be of value in CHF biomarker discovery. Further validation is required in a prospective sample.