e13573 Background: The protein Kaiso was originally identified as a transcription factor and member of the BTB/POZ, a subfamily of zinc finger proteins that interacts with p120 catenin-binding proteins. Recently, we have shown that the relative abundance of both Kaiso -nuclear and -cytoplasmic as determined by quantitative immune-histochemistry (IHC) are each independent predictors of breast cancer (BC) survival preferentially in women of African ancestry (PMID: 33526872). In this study, we combine quantitative IHC with gene expression data to develop and identify surrogate signatures, and pathways associated with the abundance of kaiso -nuclear, -cytoplasm, and LC3A/B in BC patients. We found 1) Kaiso, and LC3A/B derived signatures predict response to therapy and survival, 2) Reveal novel functional and predictive linkages between Kaiso’s subcellular distribution, LC3A/B, and TIL% in comparison to published and commercially available BC biomarkers. Methods: We used a machine Learning approach to assess a cohort of racially diverse 555 BC patients who underwent surgery for their primary BC in Greenville, NC and develop proteomics-based genomics (PbG) signatures. Statistical models were developed to predict the treatment response (pathological complete response, pCR) and distant recurrence-free survival (DRFS) with the help of those PbG signatures and performance was assessed by Receiver Operating Characteristic. The cross-validated models were developed on a pooled dataset of (N= 845) samples (primarily taxane and anthracycline based) for BC (PMID: 22508827). Again, models were validated on neoadjuvant BC chemotherapy cohort (N=415) including racial disparities. Results: We found that PbG biomarkers are associated with gene regulatory pathways linked to replication, cellular stress, and numerous immunological pathways based on their scoring in GSEA profiling. Our present meta-analysis indicates that gene expression signatures and profiles generated from both Kaiso and LC3A/B subcellular localization provide added prognostic value in predicting pCR or recurrence. Moreover, the utility and accuracy of the model were validated using a second independent data set. We are currently exploring differential modeling outcomes that incorporate patient race and genetic ancestry stratification. Conclusions: The use of surrogate markers of IHC-based protein expression through the generation of gene modules introduces a new class of predictive biomarkers that add predictive value for use in clinical trials to guide treatment decisions with respect to therapeutic response and relapse free survival. Validation of this approach in an independent neoadjuvant BC chemotherapy clinical data set supports further exploration of both these gene expression protein surrogates and the protein biomarkers themselves through further retrospective evaluation and future prospective clinical trials.
Read full abstract