Imaging tumour heterogeneity of the consequences of a PKCα-substrate interaction in breast cancer patients

Abstract

Breast cancer heterogeneity demands that prognostic models must be biologically driven and recent clinical evidence indicates that future prognostic signatures need evaluation in the context of early compared with late metastatic risk prediction. In pre-clinical studies, we and others have shown that various protein-protein interactions, pertaining to the actin microfilament-associated proteins, ezrin and cofilin, mediate breast cancer cell migration, a prerequisite for cancer metastasis. Moreover, as a direct substrate for protein kinase Cα, ezrin has been shown to be a determinant of cancer metastasis for a variety of tumour types, besides breast cancer; and has been described as a pivotal regulator of metastasis by linking the plasma membrane to the actin cytoskeleton. In the present article, we demonstrate that our tissue imaging-derived parameters that pertain to or are a consequence of the PKC-ezrin interaction can be used for breast cancer prognostication, with inter-cohort reproducibility. The application of fluorescence lifetime imaging microscopy (FLIM) in formalin-fixed paraffin-embedded patient samples to probe protein proximity within the typically <10 nm range to address the oncological challenge of tumour heterogeneity, is discussed.

Figure 1. Imaging PKCα–ezrin interaction in FFPE samples

Representative images of breast cancer tissue stained with anti-ezrin IgG [labelled with Cy2 (A, C and D) or Alexa Fluor^® 546 (B)] and anti-PKCα IgG [labelled with Cy3 (A, C and D) or Cy5 (B)]. (A and B) FRET/FLIM images show interaction between proteins (decrease in lifetime, indicated by red pixels in the pseudocolour tumour map). (C and D) Utilization of images for AIS algorithm to generate imaging parameters shown to the right of the images.

Figure 2. Imaging activation status of ezrin and cofilin in FFPE samples

Representative images of breast cancer tissue stained with anti-ezrin IgG–Cy2 and anti-phospho-ERM IgG–Cy3 (A); and with anti-cofilin IgG–Cy2 and anti-phospho-cofilin IgG–Cy3 (B). Pseudocolour maps show higher co-localization intensities in one sample (upper panel) and lower in another sample (low panel).

Figure 3. Utilization of imaging parameters for clinical outcome prediction model

(A) SVM classification of ezrin–PKCα FRET positivity (1990s cohort, n = 134 patient samples). Prediction accuracy (FRET > 0 or < 0) is shown for a decreasing number of input covariates (AIS and manual scores). Empty points represent the prediction accuracies achieved for training sets; solid points for validation sets. Error bars represent the S.D. across 100 cross-validation iterations. Cross-validation (CV; training:validation ratio, 2:1; 100 iterations) was performed with balanced outcome classes (FRET > 0 or < 0) by randomly selecting an equal number of samples from each class. Ranking of variables was performed by sequentially removing the input variable with the lowest weight when averaged over CV iterations. (B) Kaplan–Meier curve for cofilin/phospho-cofilin co-localization for the 1980s/1990s samples with reported distant metastasis (‘relapse subgroup’). (C–E) Kaplan–Meier curves for ezrin–phospho-ERM co-localization. (C) 1980s/1990s samples with reported distant metastasis (‘relapse subgroup’) showing the upper quartile of values compared with all other samples; (D) all 1980s samples with available co-localization data (shown as upper quartile compared with lower quartile); (E) all 1990s samples with available co-localization data (shown as upper quartile compared with lower quartile). (F) Complexity-optimized Bayesian proportional hazards regression model showing the predictive accuracy for 7-year distant metastasis-free survival for the 1980s cohort. Predictive accuracy is shown for training (empty points) and validation sets (filled points) (CV; training:validation ratio, 1:1; 400 iterations). Models with more than six covariates show a decline in predictive accuracy for validation sets, indicating over-fitting. Models with up to six covariates show a predictive accuracy among validation sets of up to ~70 %. (Horizontal broken line indicates the predictive accuracy expected if all samples are assigned to one class.) (G) Predictive accuracy for 3-year distant metastasis-free survival for 1990s samples with reported metastatic relapse, displayed as for (F). (H and I) Predictive accuracy (training and validation sets, displayed as in F) using up to six covariates from the consensus set derived from analyses in (A and F) and shown for (H) 7-year distant metastasis-free survival in the 1980s group, and (I) 3-year distant metastasis-free survival among the 1990s relapse subgroup.