Immunological and clinicopathological features predict HER2-positive breast cancer prognosis in the neoadjuvant NeoALTTO and CALGB 40601 randomized trials

Abstract

The identification of prognostic markers in patients receiving neoadjuvant therapy is crucial for treatment optimization in HER2-positive breast cancer, with the immune microenvironment being a key factor. Here, we investigate the complexity of B and T cell receptor (BCR and TCR) repertoires in the context of two phase III trials, NeoALTTO and CALGB 40601, evaluating neoadjuvant paclitaxel with trastuzumab and/or lapatinib in women with HER2-positive breast cancer. BCR features, particularly the number of reads and clones, evenness and Gini index, are heterogeneous according to hormone receptor status and PAM50 subtypes. Moreover, BCR measures describing clonal expansion, namely evenness and Gini index, are independent prognostic factors. We present a model developed in NeoALTTO and validated in CALGB 40601 that can predict event-free survival (EFS) by integrating hormone receptor and clinical nodal status, breast pathological complete response (pCR), stromal tumor-infiltrating lymphocyte levels (%) and BCR repertoire evenness. A prognostic score derived from the model and including those variables, HER2-EveNT, allows the identification of patients with 5-year EFS > 90%, and, in those not achieving pCR, of a subgroup of immune-enriched tumors with an excellent outcome despite residual disease.

Fig. 1. Heterogeneity of BCR measures according to hormone receptor status and PAM50 subtypes.

a Comparisons of BCR normalized number of reads (“N reads”; represented on a log scale), number of clones (“N clones”), evenness and Gini index in HR- and HR+ HER2-positive breast cancer in NeoALTTO (N = 254) and CALGB 40601 (N = 264). Two-sided P values at the bottom of the panels are from Wilcoxon rank sum test, and FDRs obtained adjusting P values using Benjamini & Hochberg method. See also Supplementary Fig. 7 for the other BCR measures and Supplementary data 8 reporting P values and FDRs. b Comparisons of BCR normalized number of reads (“N reads”; represented on a log scale), number of clones (“N clones”), evenness and Gini index in PAM50 subtypes in HER2-positive breast cancer in NeoALTTO (N = 254) and CALGB 40601 (N = 264). Two-sided P values at the bottom of the panels are from Kruskal-Wallis test, while Wilcoxon rank sum test was used when comparing each group against each one of the others. FDRs were then obtained adjusting P values using Benjamini & Hochberg method. See also Supplementary Fig. 10 for the other BCR measures, Supplementary data 10, 11 reporting P values and FDRs. Source data are available. For Wilcoxon tests, FDRs < 0.05 are shown. In the panels: * = FDR < 0.05 and 0.01; ** = FDR < 0.01 and ≥ 0.001; *** = FDR < 0.001. The number of reads is normalized by the total number of reads mapping to the transcriptome in each sample, and multiplied by 1000. In boxplots, the boxes are defined by the upper and lower quartile; the median is shown as a bold colored horizontal line; whiskers extend to the most extreme data point which is no more than 1.5 times the interquartile range from the box. BCR B cell receptor, FDR false discovery rate, HER2-E HER2-Enriched, HR hormone receptor, LumA luminal A, LumB luminal B.

Fig. 2. Association of BCR and TCR measures with EFS in the NeoALTTO and CALGB 40601 cohorts.

a Forest plot for EFS in the NeoALTTO cohort, univariable analysis. b Forest plot for EFS in the CALGB 40601 cohort, univariable analysis. c Forest plot for EFS in the NeoALTTO cohort, correcting for clinicopathological parameters (age, hormone receptor status, tumor size, nodal status, PAM50 subtypes, and treatment arm). d Forest plot for EFS in the CALGB 40601 cohort, correcting for clinicopathological parameters (age, hormone receptor status, tumor size, nodal status, PAM50 subtypes, and treatment arm). For univariable analysis, P values are from likelihood ratio test. When correcting for clinicopathological characteristics, P values were obtained with an ANOVA on nested Cox models. P values are two-sided. Non-significant values (FDR > 0.05) are shown in dark gray, significant values are shown in red (HR > 1) and blue (HR < 1). Circles indicate HR, and error bars the 95% confidence interval (95% CI). Analyses were performed including patients with available data. In NeoALTTO, N = 254 for all BCR/TCR metrics, except TCR CDR3 length, TCR Gini, TCR Gini-Simpson, TCR top clone (N = 253) and TCR evenness, TCR second top clone (N = 251). In CALGB 40601, for all BCR/TCR measures N = 264 in univariable, N = 248 in multivariable. 95% CI 95% confidence interval, BCR B cell receptor, CDR3 complementarity-determining region 3, EFS event-free survival, FDR false discovery rate, HR hazard ratio, N reads number of normalized reads, N clones number of clones, TCR T cell receptor.

Fig. 3. Event-free survival outcomes based on the groups identified by the prognostic HER2-EveNT score in the NeoALTTO dataset.

a Kaplan–Meier plot showing EFS in the NeoALTTO population (N = 233) with information available for all variables included in the model (breast pCR, hormone receptor status, clinical nodal status, TILs, BCR evenness). b Kaplan–Meier plot showing EFS in the NeoALTTO subgroup with all variables in the model available and breast pCR (ypT0/is) at surgery (N = 82). c Kaplan–Meier plot showing EFS in the NeoALTTO subgroup with all variables in the model available and without pCR in the breast at surgery (N = 151). Patients are stratified according to low risk (good prognosis group, first tertile) and high risk (poor prognosis group, tertiles 2 and 3 combined), based on the HER2-EveNT score derived from the prognostic model. Tables show 5-year EFS rates and HRs with respective 95% CI. P values are from log-rank test, HR describes the risk of an event as defined by EFS in the good prognosis group compared to the one with poor prognosis. 95% CI 95% confidence interval, BCR B cell receptor, EFS event-free survival, HR hazard ratio, pCR pathological complete response, TILs tumor-infiltrating lymphocytes.

Fig. 4. Event-free survival outcomes based on the groups identified by the prognostic HER2-EveNT score in the CALGB 40601 independent validation dataset.

a Kaplan–Meier plot showing EFS in the CALGB 40601 population (N = 230) with information available for all variables included in the model (breast pCR, hormone receptor status, clinical nodal status, TILs, BCR evenness). b Kaplan–Meier plot showing EFS in the CALGB 40601 subgroup with all variables in the model available and breast pCR (ypT0/is) at surgery (N = 105). c Kaplan–Meier plot showing EFS in the CALGB 40601 subgroup with all variables in the model available and without pCR in the breast at surgery (N = 125). Patients are stratified according to low risk (good prognosis group) and high risk (poor prognosis), based on the HER2-EveNT score derived from the prognostic model. The cutoff to identify the prognostic groups is derived from NeoALTTO (patients with a score ≤ −1.3763 were assigned to the good prognosis group). Tables show 5-year EFS rates and HRs with respective 95% CI. P values are from log-rank test, HR describes the risk of an event as defined by EFS in the good prognosis group compared to the one with poor prognosis. 95% CI 95% confidence interval, BCR B cell receptor, EFS event-free survival, HR hazard ratio, pCR pathological complete response, TILs tumor-infiltrating lymphocytes.

Fig. 5. Heatmap of GSVA scores for hallmark gene sets and correlations with the HER2-EveNT score in the NeoALTTO and CALGB 40601 datasets.

a Heatmap showing the GSVA scores for 42 hallmark gene sets in the NeoALTTO HER2-EveNT cohort (N = 233). b Heatmap showing the GSVA scores for 42 hallmark gene sets in the CALGB 40601 HER2-EveNT cohort (N = 230). Annotations on the top section of the heatmap include the HER2-EveNT score, the prognostic groups, presence or absence of EFS events, pCR, hormone receptor and clinical nodal status, TIL levels (%), BCR evenness (values as used in the model), and PAM50 subtypes. Cutoff represents the value to divide the two prognostic groups (−1.3763). On the left side, Spearman correlations values between the GSVA scores and the prognostic score are shown if P < 0.05 (two-sided). The red line divides positive and negative correlations. See also Supplementary data 23 and Source data. Basal basal-like, BCR B cell receptor, EFS event-free survival, GSVA gene set variation analysis, Her2 HER2-Enriched, HR hormone receptor, LumA luminal A, LumB luminal B, Normal normal-like, pCR pathological complete response.