Combination of SELDI-TOF-MS and data mining provides early-stage response prediction for rectal tumors undergoing multimodal neoadjuvant therapy

Abstract

Objective: We investigated whether proteomic analysis of the low molecular weight region of the serum proteome could predict histologic response of locally advanced rectal cancer to neoadjuvant radiochemotherapy (RCT).

Summary background data: Proteomic analysis of serum is emerging as a powerful new modality in cancer, in terms of both screening and monitoring response to treatment. No study has yet assessed its ability to predict and monitor the response of rectal cancer to RCT.

Methods: Sequential serum samples from 20 patients undergoing RCT were prospectively collected. Time points sampled were as follows: pretreatment, 24/48 hours, 1 week, 2 weeks, 3 weeks, 5 weeks (last day of RCT), and presurgery. Response to treatment was measured using a 5-point tumor regression grade (TRG) based on the degree of residual tumor to fibrosis. All serum samples were analyzed in duplicate using surface-enhanced laser desorption/ionization-time of flight mass spectrometry (SELDI-TOF-MS). Support vector machine (SVM) analysis of spectra was used to generate a predictive algorithm for each time point based on proteins that were maximally differentially expressed between good and poor responders. This algorithm was then tested using leave-one-out cross validation.

Results: In total, 230 spectra were generated representing all available time points from 9 good responders (TRG 1+2) and 11 poor responders (TRG 3-5). SVM analysis indicated that changes within the serum proteome at the 24/48 hours time point into treatment provided optimal classification accuracy. In more detail, a cohort of 14 protein peaks were identified that collectively differentiated between good and poor responders, with 87.5% sensitivity and 80% specificity.

Conclusions: Serum proteomic analysis may represent an early response predictor in multimodal treatment regimens of rectal cancer. These data suggest that this novel, minimally invasive modality may be a useful adjunct in the multimodal management of rectal cancer, and in the design of future clinical trials.

Figures

FIGURE 1. Mandard tumor regression grade used to classify response in this study.

FIGURE 2. Representative portion of replicate patient spectra showing a positive marker (m/z 4159) and a negative marker (m/z 4188) of good response. Intensity, normalized to total ion current, is given on the vertical axis.

FIGURE 3. Time-course analysis of peak at m/z 4159 demonstrating marked treatment-induced alterations in expression level. Intensity, normalized to total ion current, is given on the vertical axis.

FIGURE 4. Time-course analysis of peak at m/z 3451 demonstrating >3-fold up-regulation during RCT, followed by return to pretreatment levels 6 weeks after stopping RCT. Intensity, normalized to total ion current, is given on the vertical axis.

FIGURE 5. This shows the ability of 14 “key” peaks to identify good and poor responders at the 24- and 48-hour time point. Vertical columns represent individual patient response profiles; rows represent individual predictive peaks. Shading represents mean-normalized expression values of the signal for a peak across the time points assessed. Black boxes indicate a higher than mean signal for that peak and vice versa. The predictive strength of each peak, derived from the negative natural log of its P value, is also indicated.