Interaction between Molecular Subtypes and Stromal Immune Infiltration before and after Treatment in Breast Cancer Patients Treated with Neoadjuvant Chemotherapy

Breast cancer is the most frequently diagnosed cancer and the leading cause of cancer-related death in women. Neoadjuvant chemotherapy (NAC) is increasingly prescribed for patients with locally advanced breast cancer and provides opportunities for studying and monitoring the treatment sensitivity of tumors “in vivo.” A pathologic complete response (pCR) after NAC is a surrogate marker of good prognosis in triple-negative breast cancer (TNBC) and HER2-positive breast cancer, and is now used in FDA trials as a means of accelerating the approval of new drugs.

The role of tumor-infiltrating lymphocytes (TIL) in breast cancer has been studied over the last decade. Many studies have reported associations between high TIL levels at diagnosis and a better response to NAC (1–3), and a better prognosis in both neoadjuvant and adjuvant chemotherapy settings (4–7), particularly for TNBC and HER2-positive breast cancer. In 2015, an international consortium provided guidelines for the standardized evaluation of TILs in clinical practice (8), and their assessment is encouraged in routine practice, although the results of such evaluations currently have no impact on therapeutic strategy in clinical practice.

The analysis of residual tumor burden after systemic neoadjuvant treatment is an underexplored area that may help us to understand the mechanisms of resistance to specific treatments in breast cancer. However, only a few studies have investigated the variation of TIL levels in response to NAC. Furthermore, studies of the prognostic significance of postchemotherapy TILs have focused almost exclusively on TNBCs (9, 10).

The aim of this study was to report and compare the predictive and prognostic values of TIL levels (before and after NAC) as a function of breast cancer subtype, in a real-life cohort of 718 breast cancer patients treated with NAC.

We analyzed a cohort of 718 patients with nonmetastatic breast cancer treated with NAC with or without trastuzumab, followed by surgery, at the Institut Curie (Paris and Saint Cloud, France). The cohort and treatments have been described in detail elsewhere and are summarized in the Supplementary Material. This study was approved by the Institutional Review Board of Institut Curie and was conducted in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki. By the law, no informed consent from the patient was required in this observational study.

Breast cancer tumors were classified into subtypes [TNBC, HER2-positive, and luminal HER2-negative (referred to hereafter as “luminal”)] on the basis of IHC and FISH (see the Supplementary Material). In accordance with the guidelines used in France (11), cases were considered estrogen receptor (ER)–positive or progesterone receptor (PR)–positive if at least 10% of the tumor cells expressed ER/PR, and endocrine therapy was prescribed when this threshold was exceeded.

Pretreatment core needle biopsy specimens and the corresponding post-NAC surgical specimens were reviewed independently by two experts in breast diseases (M. Laé and D. De Croze).

Formalin-fixed paraffin-embedded tumor tissue samples were studied. TILs, residual cancer burden (RCB) indices, and pre- and post-NAC cellularity were reviewed simultaneously, specifically for the purposes of this study, between January 2015 and March 2017. In accordance with the recommendations of the international TILs Working Group (8), we checked for presence of a mononuclear cell infiltrate in the stroma on hematoxylin and eosin–stained sections without additional staining, after excluding areas around ductal carcinomas in situ, and tumor zones with necrosis and artifacts. Infiltrates were scored on a continuous scale, as the mean percentage of the stromal area occupied by mononuclear cells. After NAC, we assessed TIL levels within the borders of the residual tumor bed, as defined by the RCB index (12). Nothing is known about the clinical, biological, and prognostic significance of TILs in the area of regression in cases of pathologic response, but the TILs' international working group recently called for their evaluation for research purposes. In cases of pCR, the scar area was measured on macroscopic examination. The scar appeared as a white area in the breast parenchyma corresponding to the tumor bed modified by NAC. It was characterized by the presence of histiocytes, lymphocytes, macrophages, fibrosis, and elastosis. The whole fibro-inflammatory scar was evaluated on hematoxylin and eosin sections (size in mm and stromal TIL level evaluation; Supplementary Fig. S1). We determined the RCB index, as described by Symmans and colleagues (12), with the web-based calculator freely available via the Internet (www.mdanderson.org/breastcancer_RCB). Invasive tumor cellularity before and after NAC was determined as the percentage of the tumor area occupied by invasive cancer.

We defined pCR as the absence of invasive residual tumor from both the breast and axillary nodes (ypT0/is N0). Disease-free survival (DFS) was defined as the time from surgery to death, locoregional recurrence, or distant recurrence, and overall survival (OS) was defined as the time from surgery to death. For patients for whom none of these events were recorded, we censored data at the time of last known contact.

Pre- and post-NAC TIL levels were analyzed as continuous variables, after performing linearity tests (see the complementary statistical methods section of the Supplementary Material). RCB index was assessed as a continuous variable in both univariate and multivariate analyses. All analyses were performed on the whole population and after stratification by breast cancer subtype. TIL levels and qualitative variables in classes were compared by ANOVA, with post hoc Tukey analysis when required, or in Mann–Whitney U or Kruskal–Wallis tests, where indicated. Absolute and relative changes in TIL levels were calculated as the difference between pre- and post-NAC TIL levels and as these levels divided by pre-NAC TIL levels, respectively. Changes in mean values were investigated in paired t tests. The classical statistical methods used to analyze univariate and multivariate associations with pCR (logistic regression models) and survival (Cox proportional hazard models) are described in the complementary statistical methods section of the Supplementary Material.

In total, 718 patients were included in the cohort [luminal (n = 223), TNBC (n = 320), and HER2-positive (n = 175); Supplementary Table S1). Mean pre-NAC TIL level was 24.2% (luminal: 16.2%; TNBC: 28.5%; HER2-positive: 26.5%; P < 0.001), and the distribution of TILs differed between breast cancer subtypes (Fig. 1A and B).

Pre-NAC TIL levels were significantly higher in tumors for which pCR was achieved than for tumors for which residual disease (RD) was detected, except in HER2-positive breast cancers (Supplementary Table S2, P_Interaction = 0.001). Pre-NAC TILs were significantly associated with pCR (all: OR, 1.03; confidence interval, 1.02–1.04; P < 0.001; Table 1]. However, after stratification by breast cancer subtype, this association was found to be significant only in TNBCs (luminal: OR, 1.03; CI, 1–1.06, P = 0.058; TNBC: OR, 1.03; CI, 1.02–1.04; P < 0.001; HER2-positive: OR, 1.01; CI, 0.99–1.03, P = 0.341; Supplementary Table S3). The association between TILs and pCR (Fig. 1C) was linear for all groups (Fig. 1D, E, and G) except TNBCs, for which it was best fitted by a cubic spline (P = 0.006; Fig. 1F). In univariate and multivariate analysis, pre-NAC TIL levels were significantly associated with pCR in the whole population and in the TNBC subtype.

Pre-NAC TIL levels were significantly associated with DFS in the whole population (HR, 0.988; CI, 0.979–0.998; P = 0.017; Table 2) and in the TNBC subgroup (HR, 0.982; CI, 0.971–0.993; P = 0.002), but not in the other subgroups (luminal: HR, 0.994; CI, 0.971–1.018; P = 0.641; HER2-positive: HR, 1.007; CI, 0.981–1.032; P = 0.611; Supplementary Table S4). Statistical tests revealed significant deviations from the assumption of linearity in the whole population, and in the luminal and TNBC subgroups, consistent with a nonlinear prognostic effect of TILs. No such deviation from linearity was observed in the HER2-positive population (Fig. 1H–K). In addition, the interaction test between pre-NAC TILs and chemotherapy regimen on DFS was significant (P_interaction = 0.05), suggesting that the positive impact of TILs on DFS was different according to the NAC used (Anthra-taxanes: HR, 0.993; 95% CI, 0.983–1.003; P = 0.18; Others: HR, 0.968; 95% CI, 0.944–0.994; P = 0.014).

After chemotherapy, TIL levels decreased in 61.6% of tumors (n = 441), did not change in 17.7% (n = 127), and increased in 20.7% (n = 148). Mean TIL levels were higher before than after NAC (all: 24.1% vs. 13.0%, P < 0.001; luminal: 16.0% vs. 11.2 %; TNBC: 28.5% vs. 15.4 %; HER2-positive: 26.5% vs. 10.9 %, P < 0.001; Fig. 2A). These results were similar according to NAC regimen (Supplementary Fig. S2).

Mean TIL variation differed according to pCR status (pCR: −25.2 vs. no pCR: −5.6, P < 0.001). TIL levels were more likely to increase or remain stable after NAC if pre-NAC TIL levels were low than if they were high (Fig. 2B–D). PCR status was strongly associated with the magnitude of TIL level decrease (Fig. 2F); however, the variation of TIL level was strongly inversely correlated with pre-NAC TIL levels (r = −0.80, P < 0.001) regardless of pCR status (Fig. 2E). Overall, these findings suggest a strong inverse correlation between pre-NAC TIL levels and the variation of TIL levels, both of which are also strongly associated with pCR (Supplementary Fig. S3). This was true irrespective of breast cancer subtypes and NAC regimen (Supplementary Figs. S4 and S5).

After NAC, mean TIL levels were 13%, and differences were observed between breast cancer subtypes (TNBC: 15.4%; luminal: 11.3%; HER2-positive: 10.9%, P < 0.001; Fig. 3A and B).

Post-NAC TIL levels differed significantly between tumors with and without pCR (no pCR/pCR: 14.7% vs. 8.8 %, P < 0.001; Fig. 3C; Supplementary Table S5) except in luminal breast cancers (TNBC: 18.6% vs. 10.3%, P < 0.001; HER2-positive: 14.0% vs. 6.2%, P < 0.001; luminal: 11.4% vs. 7.8%, P = 0.27). Post-NAC TIL levels were associated with aggressive tumor characteristics in the HER2-positive population, but not in luminal tumors and TNBCs (Fig. 3D–E). Significant interactions were observed for the association between post-NAC TILs, breast cancer subtype, and post-NAC mitotic index (P_Interaction: 0.037), invasive tumor cellularity (P_Interaction < 0.001), and RCB class (P_Interaction = 0.05, Fig. 3F).

Post-NAC TIL levels were not associated with DFS in the whole population (HR, 1.01; 95% CI, 0.099–1.02; P = 0.325; Table 2), but a significant interaction with breast cancer subtype was observed (P_Interaction = 0.04). Post-NAC TILs had no impact on prognosis in the luminal subgroup (HR, 0.996; CI, 0.964–1.029; P = 0.79) or TNBC subtypes (HR, 0.998; CI, 0.983–1.013; P = 0.786), but had a significant adverse impact in HER2-positive breast cancers (HR, 1.04; 95% CI, 1.016–1.064; P = 0.001; Supplementary Table S4). No significant deviation from the assumption of linearity was observed. In the population with RD, an adverse impact of post-NAC TILs was observed for patients with HER2-positive disease (HR, 1.029; CI, 1.002–1.057; P = 0.034), whereas a trend toward a protective effect of high post-NAC TIL levels was observed for TNBC (HR, 0.984; CI, 0.966–1.003; P = 0.095).

After multivariate analysis, pre-NAC TIL levels, breast cancer subtype, RCB index, and post-NAC mitotic index were significantly associated with DFS (Table 2). In TNBCs, pre-NAC TIL levels were an independent predictor of better DFS (Supplementary Table S4), whereas post-NAC TIL levels were an independent predictor of impaired DFS in the HER2-positive subgroup. Neither pre-NAC nor post-NAC TIL levels were associated with DFS in the luminal subgroup.

Detailed analyses performed after stratification by RCB class are provided in Supplementary Tables S6 and S7. The association between post-NAC TILs and DFS was not significant in the RCB-0-I or RCB-II classes, whereas post-NAC TILs were associated with poor outcome in the RCB-III class (HR, 1.02; CI, 1.001–1.037; P = 0.036).

We investigated survival as a function of neoadjuvant trastuzumab use (n = 144, 82.3%) or nonuse (n = 31, 17.7%) in the HER2-positive population (Supplementary Table S8). Pre-NAC TIL levels were not associated with DFS in any of the two groups, and post-NAC TIL levels were significantly associated with impaired DFS only in the population treated with neoadjuvant trastuzumab (HR, 1.038; CI, 1.011–1.065; P = 0.005).

We analyzed the HER2-positive population according to ER status (ER positive, n = 98; ER negative, n = 77; Supplementary Table S9). Tumors from the ER⁻/HER2⁺ subgroup were of higher grade, and TIL levels were higher before chemotherapy than in the ER⁺/HER2⁺ subgroup (30.6% vs. 23.2%, P < 0.01). After chemotherapy, there was no difference in the TIL levels.

Pre-NAC TIL levels were neither associated with pCR nor DFS in any of the ER-positive or ER-negative subgroups. Post-NAC TILs were associated with impaired DFS in ER-positive population (HR, 1.04; 95% CI, 1.02–1.07; P < 0.01) but not in the ER-negative population (HR, 1.04; 95% CI, 0.99–1.09; P = 0.13). This difference might be explained by a lack of statistical power (P_{interaction with ER status} = NS).

We report here detailed analyses of associations between baseline and posttreatment immune infiltration levels in a large cohort of paired pre- and post-NAC breast cancer samples. Our findings extend existing knowledge in this field in several ways.

First, our results confirm the widely reported association between pre-NAC TILs and pCR (1–3, 13–17), but we nevertheless observed (i) a nonlinear effect in TNBCs and (ii) a significant interaction with breast cancer subtype.

Nonlinear effects have been reported for the association of pCR and TIL levels in HER2-positive tumors (2, 18, 19), but linearity has never been investigated in detail for TNBCs in the neoadjuvant setting (no linearity test reported; refs. 1, 3, 9, 10, 15–17, 20–27). In addition, our data also revealed a nonlinear prognostic impact of TILs, differing by breast cancer and by NAC regimens.

Second, significant interactions with breast cancer subtype have been described only in the GeparSixto trial (3) so far. It is unclear why pre-NAC TILs were associated with pCR only in TNBC. Although the relationship we found here is almost constant in TNBC studies (1, 3, 17, 23), this effect seems less clear in HER2-positive breast cancer (2, 14, 20, 28). Several studies on HER2-positive breast cancer [NeoSPHERE (20), NeoALTTO (2), and GeparSepto (28)]—including ours—showed no association between pre-NAC TILs and pCR, whereas other did [GeparQuattro and GeparQuinto (14)]. Several hypotheses could explain such differences: (i) Differences in tumor biology; (ii) quantitative and qualitative differences in the immune infiltration and corresponding threshold values for defining high-TILs tumors (2, 18); (iii) the use, the type, and the interaction of TILs with anti–HER2-targeted therapies (2, 7, 29); (iv) the type and the sequences of NAC regimen, as interactions have been previously described between TILs, subtype, and chemotherapy regimen (3, 4, 5); and (v) and the difference in the percentages of ER-positive disease in the different HER2-positive breast cancer cohorts. Regarding luminal breast cancer, the number of patients whose tumor reached pCR was very low, and a lack of statistical power may partially explain why the association between pre-NAC TILs and pCR failed to reach statistical significance (P = 0.058).

Our results are in line with a recently published pooled analysis from German Breast Group (13) analyzing the relationships between TIL levels in baseline samples and oncologic outcomes in a large cohort of 3,771 patients receiving NAC. Denkert and colleagues (13) found that higher TIL levels were associated with a DFS benefit in HER2-positive and TNBC tumors, but with a poor OS in luminal HER2-negative breast cancers. These results underline the complexity of the relevance of TILs to each specific breast cancer subtype. We believe that both (i) TILs' subsetting and (ii) analyses of previous data taking into account chemotherapy regimen, type of targeted therapy, and breast cancer subtypes may help understanding these noticeable discrepancies.

Third, we demonstrated a decrease in mean TIL levels when comparing levels before and after chemotherapy. Only a few cohorts (9, 10, 15, 21, 24–27, 30–32) have reported pathologic TIL evaluations on paired matched samples, and all these previous cohorts were small (Supplementary Table S10). Two studies assessing lymphocyte density by computational pathology on large cohorts of patients from neoadjuvant trials [Neo-tAnGo (25) and ARTemis (27)] found that both pre-NAC immune infiltration and a decrease in immune infiltration were associated with pCR, and another study found that larger decreases in CD3 levels after treatment were associated with better DFS and OS (15). Due to the strong association between pre-NAC TILs, TILs changes, pCR status, post-NAC TILs, and DFS, their respective part regarding the association with prognostic remains unknown. Notably, TILs' changes might be an interesting parameter, as it was both strongly associated with pCR and DFS. As these data on TILs' variation are unprecedented on a large cohort of breast cancer patients in the literature, it calls for further validation of this endpoint on independent cohorts.

Fourth, regarding the immune infiltration after treatment, there was almost no post-NAC LPBCs (TILs ≥ 60%), highlighting the need for a revision of TIL level cutoff points after NAC. Post-NAC TIL levels were higher in tumors with RD than in areas of scarring in tumors displaying pCR. TIL levels have never been reported from pCR specimens, but recent guidelines (33) have suggested that these levels could be evaluated for research purposes. We found that TIL levels were extremely low in most, but not all, tumors scars. Our findings might suggest that, once the immune cells have eradicated the tumor, they would move into the periphery similar to responses to infection or other anomalies eliciting an immune response. Research on post-NAC TILs in pCR specimen could be of interest notably to analyze their association with the rare subgroup of patients experiencing relapse after their tumor reached pCR.

In cases of RD, TILs were associated with aggressive post-NAC patterns only in the HER2-positive subgroup. It remains unclear whether this difference reflects inherent differences between the three breast cancer subtypes, the use of neoadjuvant trastuzumab, or differences in the immune infiltration in RD. Two hypotheses can be drawn. On the one hand, TILs in specimen with RD could be active but may not have had sufficient time to completely eradicate the tumor. Our data do not support this hypothesis, because we found no correlation at all between time from biopsy to surgery and post-NAC TIL levels (Supplementary Fig. S6). On the other hand, the immune response may not recognize the tumor cells and post-NAC TILs could be unable to exert their antitumor function, possibly due to a surrounding immunosuppressive milieu.

Finally, our results suggest that pre- and post-NAC TIL levels may have different impacts on outcome. In TNBC, high pre-NAC TIL levels were an independent predictor of good prognosis, whereas in HER2-positive breast cancer, high post-NAC TIL levels were an independent predictor of poor outcome. We are currently characterizing the immune subpopulations, immune checkpoint, and immune checkpoint ligand expression in residual tumor specimens, in the hope that this will shed further light on the mechanisms underlying the observed differences in the prognostic impact of post NAC-TILs in the three breast cancer subtypes. Analyses of spatial and temporal dynamics, particularly to determine whether TIL location (intratumoral vs. stromal) has a differential effect on outcome, will also be of interest.

The strengths of this study include the large sample size and the availability of paired matched pre- and post-NAC samples for 716 patients. In addition, the patients and samples were derived from an institutional cancer center cohort, and therefore reflect real-life conditions more faithfully than analyses of results for randomized trials including only highly selected patients. The limitations of this study include the lack of data on TILs during NAC (i.e., on-treatment biopsies), which would have (i) provided insights into the mechanisms underlying immune response to chemotherapy and (ii) confirmed if pre-NAC TIL levels go straightforward to the levels observed after NAC, or if it is preceded by an initial increase. On-treatment data from the I-SPY trial suggest that the immune genes expression decreases as soon as 1 to 4 days after NAC beginning (34). In addition, the study was performed at a single center, making external validation necessary. Large integrative and collaborative analyses may make it possible to decipher the role of immune infiltration in breast cancer in more detail, particularly in cases of RD after NAC. We therefore provide our original data as an open-access resource for the medical and scientific community, for pooling with existing datasets (Supplementary Table S11).

Our results have several implications. First, they suggest that future studies should include interaction and linearity tests, to help determining and validating TIL thresholds values relevant to each breast cancer subtype, both in the pre- and in the post-NAC setting. Second, due to the multiplicity of interactions (breast cancer subtypes, NAC regimen, benefit from targeted therapy, RCB score), efforts should be paid in routinely score TILs both in the pre and the post-NAC setting, and share data within collaborative projects, as such complex associations may only be deciphered with a very large amount of patients and samples. Finally, the adverse outcome associated with high TIL levels after the completion of NAC in some subgroups (HER2-positive patients; RCB-III tumors) highlights the urgent need for second-line trials in the post-NAC setting. Immunotherapies may theoretically be of interest for the treatment of tumors with an immune infiltrate associated with a poor prognosis.

F. Reyal is an advisory board member/unpaid consultant for Agendia. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.-S. Hamy, H. Bonsang-Kitzis, J.-Y. Pierga, J.-G. Feron, M. Laé, F. Reyal

Development of methodology: A.-S. Hamy, H. Bonsang-Kitzis, F. Reyal

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.-S. Hamy, D. De Croze, L. Darrigues, E. Menet, A. Vincent-Salomon, F. Lerebours, J.-Y. Pierga, E. Brain, G.-T. Lam, M. Laé, F. Reyal

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.-S. Hamy, H. Bonsang-Kitzis, E. Laas, L. Topciu, J.-Y. Pierga, E. Brain, F. Reyal

Writing, review, and/or revision of the manuscript: A.-S. Hamy, H. Bonsang-Kitzis, E. Laas, L. Darrigues, E. Menet, F. Lerebours, J.-Y. Pierga, E. Brain, G. Benchimol, M. Laé, F. Reyal

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.-S. Hamy, E. Laas, A. Vincent-Salomon, G. Benchimol, G.-T. Lam, F. Reyal

Study supervision: A.-S. Hamy, E. Brain, G. Benchimol, M. Laé, F. Reyal

We thank Roche France for financial support for the construction of the Institut Curie neoadjuvant database (NEOREP). Funding was also obtained from the Site de Recherche Integrée en Cancérologie/Institut National du Cancer (grant no. INCa-DGOS-4654) and from the ARCS (Association d'aide à la recherche cancérologique de St Cloud). A.-S. Hamy was supported by an ITMO-INSERM-AVIESAN translational cancer research grant. We thank Dr. Bernard Asselain for his helpful advice and for validating the statistical methodology used in this study.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.