Large-Scale Plasma Proteome Epitome Profiling is an Efficient Tool for the Discovery of Cancer Biomarkers

Abstract

Current proteomic technologies focus on the quantification of protein levels, while little effort is dedicated to the development of system approaches to simultaneously monitor proteome variability and abundance. Protein variants may display different immunogenic epitopes detectable by monoclonal antibodies. Epitope variability results from alternative splicing, posttranslational modifications, processing, degradation, and complex formation and possesses dynamically changing availability of interacting surface structures that frequently serve as reachable epitopes and often carry different functions. Thus, it is highly likely that the presence of some of the accessible epitopes correlates with function under physiological and pathological conditions. To enable the exploration of the impact of protein variation on the immunogenic epitome first, here, we present a robust and analytically validated PEP technology for characterizing immunogenic epitopes of the plasma. To this end, we prepared mAb libraries directed against the normalized human plasma proteome as a complex natural immunogen. Antibody producing hybridomas were selected and cloned. Monoclonal antibodies react with single epitopes, thus profiling with the libraries is expected to profile many epitopes which we define by the mimotopes, as we present here. Screening blood plasma samples from control subjects (n = 558) and cancer patients (n = 598) for merely 69 native epitopes displayed by 20 abundant plasma proteins resulted in distinct cancer-specific epitope panels that showed high accuracy (AUC 0.826-0.966) and specificity for lung, breast, and colon cancer. Deeper profiling (≈290 epitopes of approximately 100 proteins) showed unexpected granularity of the epitope-level expression data and detected neutral and lung cancer-associated epitopes of individual proteins. Biomarker epitope panels selected from a pool of 21 epitopes of 12 proteins were validated in independent clinical cohorts. The results demonstrate the value of PEP as a rich and thus far unexplored source of protein biomarkers with diagnostic potential.

Keywords: biomarker; cancer; epitope; lung cancer; plasma epitom profiling; protein variants; proteoform.

Graphical abstract

Fig. 1

Analytical validation of the QP69 and QP300 biochips. Scheme (A) shows the principle of the single-binder capture inhibition assay (sbCIA). The reproducibility of the QP69 and QP300 mAb arrays was tested, and the results were plotted as histograms showing the number of antibodies with the given CV (%). Arrows indicate the number of antibodies performing with >20% CV. Intra-assay (B–D) and inter-assay (E–G) evaluation of the RLUmax (B and E) values obtained without plasma competition and the RLU/RLUmax values (%) obtained with competition using 100-fold (C and F) and 1000-fold (D and G) diluted plasma on QP69 mAbs spotted onto the Randox biochip array. Inter-assay test results for the RLUmax (H and I) and RLU/RLUmax (%) (J and K) values obtained with competition using 300-fold diluted plasma for the QP300 mAbs. Lot-to-lot (L and M) variability was measured by comparing the averaged RLUmax values of the QP69 (L) and QP300 (M) mAbs. Inter-operator variability for the QP300 mAbs was also determined based on the measured RLUmax (N and O) and RLU/RLUmax % (P and Q) values obtained with competition using 300-fold diluted plasma.

Fig. 2

Mimotope heterogeneity and epitopes of human complement system components. SPR experiment (A) with BSI0442 mAb (factor H) and one of it’s cognate, biotinylated mimotope peptide (#5, immobilized). Binding inhibition curves obtained by competition with pooled human plasma (left) or natural CFH purified from plasma (right). Domain structure (B) of the CFH, CFHL1 and CFHR proteins, the N-terminal domains are shown in yellow. (C) The CFH concentration (µg/ml) of 37 samples was determined using selected BSI mAbs and a commercial CFH kit (Hycult) in a sandwich ELISA setup. Data for three pairwise comparisons in which seven individual samples were randomly chosen and colored (three red, four green). Binding test summary (D) of QP mAbs that recognize CFH. mAb binding was tested on complete (FH), N-terminal (FH1-4), and C-terminal (FH15–20) fragments of factor H protein and on factor H related proteins (CFHR-1, CFHR-4B, and CFHR-5).

Fig. 3

Epitome profiling of high- and medium-abundance plasma proteins reveals a rich and granular source of cancer biomarkers. Minimal representational redundancy of epitopes detected by the QP69 and QP300 biochips. Histogram of Pearson correlation coefficients derived from pairwise comparison of the data of individual mAbs/epitopes for QP69 (A) and QP300 (B). The epitope profiles do not cluster with protein IDs. The computed pairwise Pearson correlation matrix was first subjected to column-wise unsupervised clustering, and the rows were then ordered as a function of the cognate protein ID of the QP69 and QP300 mAb library (C: QP69 data set, D: QP300 data set using cohorts NKTH and BD). Normalized signal intensity data obtained with LC and control sample sets on QP69 biochip (69 BSI mAb variables) and QP300 (280 BSI mAb variables) along with seven tumor marker data. Samples were randomly divided into training (70%) and test set (30%). Nonlinear SVM models were built on training set incorporating only BSI variables (blue), only tumor markers (brown) or BSI variables, and tumor markers together (green). ROC analyses performed on test sets QP69 (E) and QP300 (F).

Fig. 4

Plasma proteome epitopes associate with cancer.A, Abundant plasma proteins as detected by the QP69 biochip (individual and composite epitopes), showing an association with lung (Bsi0221, Bsi0782, and Bsi0142), colon (Bsi0097, Bsi0221, Bsi0239, and Bsi0182), and breast (Bsi0670, Bsi0268, and Bsi0182) cancers (NKTH cohort). B, Cancer-specific epitopes are revealed by the profiling of pooled human plasma samples prepared from samples of lung, breast, and colon cancer patients with both the QP69 and QP300 biochips. Venn diagram shows the number of unique and shared mAbs that discriminate between, in the case of breast (13) and colon (20) cancer, the disease plasma pools versus the apparently healthy control sample pool, and in the case of lung cancer, the 22 mAbs selected by epitope profiling data involving lung cancer plasma samples versus COPD control samples. C, Immunogenic epitopes of a representative set of the LC-associated biomarker proteins C9, C4BP, α2HSGP, and CFH show heterogeneity with respect to association with LC. A fraction of the mAbs that recognize the same protein shows highly positive (e.g., C9-specific mAb Bsi0639) ROC-AUC values, another fraction highly negative (e.g., C9-specific mAb Bsi0686) values, and a third, neutral (e.g., C9-specific mAb Bsi0449) values (please note that in order for the better visualization of disparate, positive, negative or neutral representational changes of specific epitopes, we chose to change the conventional ROC display where absolute values are shown). D, LC prediction, the performance of the QPLC21 epitomic panel in predicting LC determined on a 554 LC and 602 control plasma sample of the BD cohort.

Fig. 5

Impact of confounding factors on the performance of selected mathematical models. Performance of three selected mathematical models: M48 (left column), M61e (middle column), and M63e (right column) (A, B, and C, color code: LC – red, control – green). The ROC curve in red shows the entire LC population versus all control comparisons in each subfigure. Insets show the spread of the model performance score as boxplots. Subgrouping of the data by COPD (D, E, and F, color code: non COPD – blue, COPD – grey), gender (G, H, and I, color code: female – yellow, male – blue), age (J, K, and L, color code: 40–59 years – light blue, 60–79 – yellow), body mass index (BMI, M, N, and O, color code: BMI-A < 20 kg/m2 – light blue, 20 kg/m2 < BMI-B < 30 kg/m2 – purple, 30 kg/m2 < BMI-C – yellow).

Fig. 6

Impact of confounding factors on the performance of selected mathematical models (continued). Tobacco consumption in pack-years (PY, A, B, and C, color code: 0–10 PY – yellow, 10–20 PY – grey, >20 PY – light blue, non-smokers – purple, LC – red, control – green), LC stage (D, E, and F, color code: I–IIIA – purple, IIIB – light blue, IV – yellow), and LC histology (G, H, and I, color code: squamous-cell carcinoma (SQC) – yellow, adenocarcinoma – blue, small-cell carcinoma – grey). The pie chart (J) shows the fraction of in situ squamous cell carcinoma (light green), squamous-cell carcinoma (yellow), adenocarcinoma (blue), adenosquamous cell carcinoma (purple), and large cell carcinoma (light blue) of non-small-cell LC samples. The outer ring of the chart shows the staging of samples as a hue of the corresponding color (BD cohort).