Background: Prior gene expression profiling studies have identified markers that are predictive of chemotherapy response using pre-treatment biopsies. However, this approach does not account for chemotherapy-induced perturbation in signaling within tumors early in treatment that may predict response to therapy with superior sensitivity to baseline signatures. We hypothesized that measuring early changes in gene expression induced by neoadjuvant chemotherapy might yield improved markers of treatment efficacy and patient outcome.
Methods: Transcriptome data from 34K probe cDNA microarrays were assembled using serial biopsies obtained from 36 I-SPY 1 TRIAL patients before treatment (T1), and 24-72 hours after beginning neoadjuvant anthracycline-based chemotherapy (T2). Outcome parameters included residual cancer burden (RCB) after therapy and recurrence free survival (RFS). Gene expression changes occurring between T1 and T2 (T2-T1) were compared between known responders (RCB 0/1) and non-responders (RCB 2/3) using a permutation test based on the t-statistic; Cox proportional hazards modeling was used to identify early response genes associated with RFS. Due to the small sample sizes, we adopted a relaxed significance threshold for outcome associations (p-value<0.005 without multiple testing correction). Pathway analyses were performed with Ingenuity IPA software.
Results: 97 genes were found to be differentially altered upon comparing early gene expression changes (T2-T1) between responders (RCB 0/1) and non-responders (RCB 2/3). Ingenuity IPA software identified cell cycle as the top enriched pathway (p = 0.0008) among these differentially altered genes, with responders showing relative up-regulation of translation regulator EIF4EBP1 and cell cycle regulators CDKN2B and SMARCB1 after treatment. Survival analysis identified changes (T2-T1) in 293 genes as significantly associated with RFS; these genes were enriched in pathways including lipid antigen presentation by CD1, cell death and drug metabolism, and notch signaling. Surprisingly, only 4 genes were associated with both RCB and RFS, likely reflecting a prognostic signal from the many non-responding patients with favorable outcomes. Intrinsic subtype assignments were 75% concordant between time points T1 and T2, with basal classifications remaining stable (13/13) and some inter-conversion among LumA, LumB, Her2, and Normal (9/23). Interestingly, a larger number of genes were associated with chemotherapy response when one considered the change in expression (T2-T1), rather than absolute expression levels at T1 (17 genes) or at T2 (87 genes). Moreover, there was negligible overlap between these three response associated gene sets (T1, T2, and T2-T1 change), indicating that early expression changes may provide information beyond signatures obtained at static time points.
Conclusion: These analyses suggest that pre-treatment, early treatment and early changes provide non-redundant information on chemotherapy responsiveness and outcome. Early expression changes might be combined with data from pre-treatment biopsies to construct early predictors of non-response, an essential component within an adaptive treatment framework.