TIL counts
At baseline, cases with pCR had significantly higher TIL counts than cases with RD (p = 0.0129, Fig. 1a). TIL counts remained significantly associated with pCR (p = 0.0243), after adjustment for ER status and treatment arm.
Among patients with pCR, TIL counts were significantly lower in post-treatment samples compared to pre-treatment (p = 0.00063, Fig. 1b). Among RD cases, there was no significant difference in pre- and post-treatment TIL counts (p = 0.1263, Fig. 1c). However, a substantial subset of cases had numerically decreased post-treatment TIL count (n = 20) while a in a few cases TIL count increased (n = 8), but there was no difference in disease-free or overall survival by decrease or increase in TIL count in the residual cancer.
PD-L1 protein expression in tumor and stromal cells
Tumor cell PD-L1 protein expression did not change significantly after treatment, in either response group (Fig. 2a, b) and there was no association between baseline PD-L1 expression and response to therapy (Fig. 2c). Stromal PD-L1 expression was also similar between pre- and post-treatment samples (Fig. 2d, e) and between response groups (Fig. 2f).
We also compared PD-L1 expression by immunohistochemistry and by mRNA measurements. Samples showing positive PD-L1 protein expression, in either stromal or cancer cells, had significantly higher PD-L1 (CD274) mRNA expressions than samples showing no PD-L1 protein expression (Fig. 3).
PD-L1 percent positivity either on tumor (r2 = 0.23, p = 0.04) or stromal cells (r2 = 0.38, p = 0.0003) correlated weakly but significantly with TIL percent count (Additional file 4: Figure S4 A, B). PD-L1 positive cases, either on tumor or stromal cells, also showed significantly higher TIL counts compared to PD-L1 IHC negative cancers (Additional file 4: Figure S4 C, D).
Baseline immune gene expression and response to therapy
In ER status and treatment arm adjusted logistic regression analysis, higher expression of 24 immune genes were significantly associated with pCR (Fig. 4a). These included several activated cytotoxic T cell markers such as granzyme, granulysin, CD7 and chemoattractant cytokines CCL21 and CCL19. Expression of IL7R, that promote V(D)J T cell receptor and immunoglobulin recombination during lymphocyte maturation was also higher in cases with pCR.
A different and substantially larger number of immune genes (n = 63) were significantly associated with RD (Fig. 4a). Many of these genes were related to fibroblast and stromal functions (COL4A5, COL5A1, COL6A3, COL10A1, TIMP3, CDH-2, − 11, − 13, CHAD, VACN), some representing therapeutic targets including VEGFB, TGFB3, PDGFB/PDGFRB, FGFR1 and IGF1R. VEGF expression was not associated with response to bevacizumab in this study. The low affinity Fc gamma receptor (FCGR2A/CD32), the cytokine receptor CX3CR1 and IL11RA were also higher in cancers with RD. CXCR1 and IL11R represent potential therapeutic targets since they promote macrophage survival and induce apoptosis resistance in cancer cells [14] and also stimulate cancer cell survival and promote angiogenesis and metastasis formation [15], respectively. High expression of the putative cancer and hematopoietic stem cell marker THY1/CD90 and WNT11 and CTNNB1 also suggest potential novel therapeutic strategies to increase pCR rates.
At the metagene level, most immune cell types and immune functions showed similar or higher baseline expression in cases with pCR compared to RD except a mast cell signature and stroma and myeloid inflammatory cell signatures that were higher in patients with RD (Fig. 5a, b). None of the 26 previously reported prognostic or immunotherapy predictive signatures showed significant association with pCR after adjustment for ER status and treatment arm.
Age was not predictive of pCR in this study (Additional file 5: Figure S2A). MKI67 (Ki67) mRNA expression levels were also similar between cases with RD and pCR at baseline (p = 0.107, Additional file 5: Figure S2B) and in the post-treatment samples (p = 0.2823, Additional file 5: Figure S2C). Because most cases were stage III invasive ductal carcinomas (N = 2 inflammatory breast cancer), we could not correlate clinical stage or histology with pCR.
Changes in immune gene expression after chemotherapy
At the individual gene level, most immune genes had lower expression in post-treatment samples in both response groups. However, in cases with pCR, cellular stress and hypoxia associated genes (DUSP1, EGR1, CPA3, GAS1) and TNFSF12 showed increased expression post-treatment (Fig. 4b). In cases with RD, DUSP1 and EGR1 also showed increased expression post-treatment as well as IL6, ATF3, CD36, CXCL2, CD69, NGFR, KLF2, THBD, DAB2 and C7 (Fig. 4c). These genes are involved with tissue repair and inflammation after injury. Among known therapeutic targets in clinical development, CTLA4, PD-L1 and IDO1 expression remained unchanged in residual tissues.
At the metagene level, in both response cohorts, all immune cell metagene expressions decreased, except the NK cell metagene that remained the same. The mast cell metagene increased in post-treatment tissues of patients with pCR (Fig. 5c, d). Metagenes representing immune-related functions also decreased in the post-treatment tissues in both response groups except IL10 signaling, NOS2, ARG1 and APM and MMR loss that remained unchanged in both response cohorts (Fig. 5e, f).
We also examined the expression of 26 previously published prognostic and immunotherapy response predictive immune gene signatures. Most of these showed lower expression in post-treatment samples in both response groups (Fig. 6a, b). The decrease was statistically significant for the Interferon, MHC1, STAT1 and Treg signatures in both cohorts. Only the IL8/VEGF gene signature showed significantly higher expression in post-treatment samples in both response groups in (pCR cohort: p = 0.038, in RD cohort: p = 0.017). The activated-CD4 gene signature also showed significantly higher expression in post-treatment samples of patients with RD (p = 0.026).
We have also compared the IL8/VEGF immune gene signature and TIL counts between the treatment arms with or without bevacizumab and found that the IL8/VEGF signature showed significantly higher expression in post-treatment samples in both treatment arms whereas TIL counts were significantly lower in post-treatment samples in both treatment arms (Additional file 6: Figure S3). No treatment-arm specific changes could be detected.