Background
Neoadjuvant chemotherapy is increasingly used to induce tumor shrinkage, allowing smaller surgical resection, eliminating clinically silent micrometastases, and providing prognostic information based on the extent of pathologic response. Pathologic complete response (pCR) predicts excellent survival while residual disease (RD) is associated with higher but variable risk of recurrence depending on the molecular subtype [
1‐
4]. Pretreatment immune infiltration in breast cancer predicts both for better prognosis, with or without adjuvant therapy, and also for greater sensitivity to chemotherapy reflected by the higher pCR rates in immune-rich cancers [
5,
6].
Several preclinical studies have suggested that cytotoxic agents partly exert their anti-tumor activity by induction of an anti-tumor immune response aimed at cells injured by chemotherapy. Injury from chemotherapy may lead to formation of new immunogenic epitopes, cytokine secretion, antigen cross-presentation, activation of dendritic cells, and induction of tumor-specific cytotoxic T cells. Recent studies on breast cancer patients have suggested that cytotoxic agents, including anthracyclines and taxanes, can induce a tumor-specific immune response, and that exposure to such drugs leads to accumulation of lymphocytes in the tumor bed [
7‐
9]. However, chemotherapy also has a direct cytotoxic effect on lymphocytes and could adversely impact the tumor immune microenvironment [
10,
11].
The simplest measure of immune activity in the tumor microenvironment is counting tumor infiltration lymphocytes (TILs). Many studies have shown that TILs in the tumor microenvironment are prognostic, particularly for ER-negative and highly proliferative ER-positive cancers [
12‐
14]. High TIL count is also associated with higher pCR rate after neoadjuvant chemotherapy [
9,
12,
13,
15,
16]. High TIL count in residual disease is also associated with better survival [
9‐
11].
The programmed cell death 1 receptor (PD-1) and its ligand PD-L1 are key immune regulatory molecules that play a pivotal role in switching off cytotoxic immune response as part of a complex immune checkpoint process [
17]. PD-L1 expression is present in a variety of cancers including those of the lung, melanoma, ovarian, colon, and breast [
14,
17‐
21]. PD-L1 is expressed by both tumor and stroma cells, and the tumor versus stromal expression frequency varies by cancer type. PD-L1 expression in the tumor microenvironment correlates strongly with the presence of TILs [
14,
19,
20,
22,
23]. Drugs that target the PD-1 or PD-L1 axis have demonstrated durable tumor responses in 10–40% of patients in clinical trials including metastatic melanoma, renal cell carcinoma, nonsmall cell lung cancer, bladder cancer, triple-negative breast cancer (TNBC), and several other solid and hematological malignancies [
24,
25].
The goal of this study was to assess changes in TIL count and PD-L1 expression in response to neoadjuvant chemotherapy for early-stage breast cancer. We reported previously on the prognostic and chemotherapy response predictive value of pretreatment TIL count, PD-L1 expression, and Ki-67 in the same patient cohort [
14‐
16,
20]. The current analysis includes examination of TILs and PD-L1 expression in the residual tumor and we assess changes between paired pre-treatment and post-treatment samples.
Discussion
Several studies have examined the prognostic and chemotherapy response predictive role of TILs in baseline, pretreatment biopsies of breast cancer [
12,
13,
16,
31‐
34], but few studies have examined change in TILs and immunological parameters during neoadjuvant therapy. In this study, we assessed changes in TIL count and PD-L1 expression in paired pre-neoadjuvant and post-neoadjuvant chemotherapy tissues and correlated residual cancer TIL counts and PD-L1 expression and change in these parameters with survival. We achieved a high interobserver concordance on TIL assessment which was based on the methodology of and recommendations by the Immuno-Oncology Biomarker Working Group [
26,
33].
The effects of neoadjuvant chemotherapy on TILs and immune gene signatures have been mostly studied in TNBC and HER2-positive tumors [
9‐
11,
13,
34,
35]. High TILs in residual cancer were associated with better RFS in two previous studies [
9,
34]. We did not observe this in our study which may be due to our smaller sample size. On the other hand, we did observe that an increase of TILs in the residual cancer was associated with better RFS, confirming a prognostic role of TILs. The importance of change of TIL counts following neoadjuvant chemotherapy was also evidenced in a parallel study by our group [
36] in specimens from the SWOG S0800 clinical trial [
28]. Therein, we observed a significant decrease of TILs following treatment. Higher TIL change correlated with higher rates of pCR. The cohorts from the two studies are not comparable, as SWOG S0800 patients were HER-2 negative, enriched in TNBC and lymphocytic predominant breast cancer (9%), and under more homogeneous treatment. Hence, in both studies the TIL count change is indicative of better outcome. Further TIL subtyping is warranted to elucidate the functional status of lymphocytes involved.
Moreover, we have shown that TIL counts in residual disease were higher in ER-negative cases, but we were not able to correlate the TIL changes with ER status (increase, decrease, or extent of change). This could be due to the small size of our cohort. Interestingly, in the TRYPHAENA study [
37] that compared TILs pre- and post- treatment with dual HER2 blockade, there was a significant overlap between ER-positive and ER-negative cases with residual disease. Similarly, there was a significant overlap between cases with low to moderate TIL infiltrate and achievement of pCR or not. Although this cohort is not directly comparable with ours, the results present some similarity in the difficulty to trace a clear trend between ER status and TIL change following treatment.
Our study is the first to report on changes in PD-L1 expression after neoadjuvant chemotherapy. Preclinical studies suggested that PD-L1 expression might be stimulated by chemotherapy [
38]. However, in our study, only 17% of residual cancers were positive for PD-L1 expression using our AQUA method. The majority of cases were negative at baseline and remained negative after chemotherapy. In most of the initially PD-L1 positive tumors, PD-L1 expression decreased after chemotherapy. Therefore, overall, we observed a significant decrease in PD-L1 expression (as defined by continuous AQUA QIF scores) in residual disease compared to pretreatment biopsies. However, PD-L1 levels or changes in PD-L1 expression were not significantly associated with survival, but we cannot exclude the lack of power of our study to evaluate this. The neoadjuvant chemotherapy-induced changes in PD-L1 expression could provide a rationale for use of immune checkpoint inhibitors in the adjuvant and, most importantly, neoadjuvant setting. The results of the I-SPY2 [
39] study showed important improvement of pCR by combined PD-axis blockade and chemotherapy in the neoadjuvant setting in breast cancer.
We also need to note that most PD-L1 expression was stromal. Although we did not use a PD-L1 multiplexed assay to characterize the cell types expressing PD-L1, we can assume based on morphology criteria that most PD-L1-expressing cells were not TILs, but more compatible with macrophages or fibroblasts. This observation could also explain why we observe this “disconnection” in PD-L1 and TIL change following treatment.
The field of PD-L1 assessment is rapidly evolving and several companion and complementary diagnostic applications are FDA cleared. In the clinic, all of the assays are based on DAB-based chromogen visualization. As a result, they lack standardization, and are limited by the subjective nature of this technique. Recently we have investigated the concordance of QIF and chromogenic assays (scored by pathologists) as well as the concordance between pathologists [
40]. Pathologists were highly concordant for PD-L1 lung tumor scoring, but not for stromal/immune cell scoring. A similar concordance amongst pathologists for tumor PD-L1, but not immune cell PD-L1, was also described recently by the NCCN/BMS study on this topic [
41]. In the case of breast cancer, the relevance of these findings might be even more impactful, as the levels of PD-L1 expression are much lower than in the case of lung cancer (usually less than 30%) [
14,
20,
23] and, as we show here, PD-L1 is mainly stromal. A cutoff of 50%, like the one used in the 22c3 antibody chromogenic assay, would automatically exclude nearly all breast cancer cases from PD-1/PD-L1-based treatments. It is possible that other assays may be required to match PD-1 axis therapies to responders in this disease.
It is inherent to studies that compare baseline and residual breast cancer tissues after neoadjuvant chemotherapy that the tissue acquisition methods differ between pre- and post-treatment. The baseline tissues come from core needle biopsies, while the residual cancer tissues come from surgically resected tissue. This may introduce sampling bias; however, we previously examined intratumor heterogeneity in immunological parameters in primary breast cancers and found only modest biopsy site-to-site variation, and therefore the sampling bias may be less important in breast cancer than in other cancers [
18,
42]. Because the pre-treatment PD-L1 assessment was also performed earlier, different antibodies were used to quantify PD-L1 expression in the pre-treatment (E1L3N) and post-treatment (SP142) tissues. While it would have been optimal to use the same antibodies for each aspect of the study, the timing of different parts of the study made this impossible. However, while the assays in the clinic are clearly different, the antibodies within the assays have been shown to be essentially identical [
43]. The work by Gaule et al. from our group showed that, at optimal titration in optimal staining conditions, SP142 and E1L3N are nearly identical. Although the use of different antibodies is a limitation of this work, we believe our previous data make these results scientifically sound.
Another important limitation is the lack of specific guidelines in the assessment of TILs in residual disease. The definition of residual cancer burden, the inclusion of cases with pCR or not, and the evaluation of areas with invasive tumor only or in the “previous tumor bed” remain elusive. The guidelines of the Immuno-Oncology Biomarker Working Group have contributed in reproducibility of TIL evaluation among different studies, but they are not currently addressing these points. An updated version of TIL guidelines from the Immuno-Oncology Biomarker Working Group will include suggestions for TIL evaluation in the residual disease, later this year.
A more significant limitation of our study is that it is based on a small, single breast cancer cohort, with heterogeneity of breast cancer subtypes and non-uniform treatment administration. The tissues were collected retrospectively and included patients in different clinical stages and hormone receptor and HER2 status. Survival and disease status were also not available for all cases. The small sample size limited the statistical power to perform between subtype comparisons and adequately powered multivariate analysis. Larger, prospective studies, incorporating multi-institution cohorts, homogeneous breast cancer tumor subtypes, and treatment regimens, are required to validate and support the clinical relevance of our findings.
In summary, we observed a non-significant trend toward increased TIL counts in residual cancer tissues, whereas PD-L1 expression decreased. An increase in TIL count in residual cancer indicates more favorable RFS compared to no change or a drop in TIL counts.
Conclusions
We have shown in patients who did not achieve pCR following neoadjuvant chemotherapy that even minor changes in TIL counts can provide hints for a better outcome. Importantly, this immune-related effect of neoadjuvant chemotherapy was observed in tumors that at baseline had low to moderate TIL counts, providing hints to further explore and valorize the functional status of immune signatures for these patients with residual disease.
Moreover, by QIF we have observed a significant decrease of PD-L1 expression following neoadjuvant chemotherapy. This disconnection with TIL counts is intriguing and could, at least partially, be explained by expression of PD-L1 by stromal cells, other than TILs, such as tumor-associated macrophages. Moreover, immune checkpoint blockade in the neoadjuvant setting could further enhance the effects of the conventional neoadjuvant chemotherapy alone.
Taken together, our data suggest the importance of further exploration of the immune potential of residual tumors in larger studies in order to ameliorate patients’ outcome by combined personalized immunotherapies. Inclusion of immunotherapy regimens in the neoadjuvant setting could also potentiate the pCR rates through multiple immune-parameter modulation. Hence, comprehensive host, tumor-intrinsic and microenvironmental baseline biomarker assessment is critical to predict benefit from personalized immunotherapies.