Neoadjuvant Chemotherapy of Ovarian Cancer Results in Three Patterns of Tumor-Infiltrating Lymphocyte Response with Distinct Implications for Immunotherapy

Medical oncology is being transformed by the introduction of new immune modulatory strategies, in particular antibodies that inhibit the CTLA-4 and PD-1 pathways (so-called immune checkpoint inhibitors; ref. 1). These agents are hypothesized to be most effective against tumors that harbor pre-existing T-cell responses, as evidenced by tumor-infiltrating CD8 T cells (CD8 TIL), PD-L1 positivity, and other signs of active immunity (1). However, a substantial proportion of cancers lack pre-existing TIL responses and hence are less likely to respond to T cell-based immunotherapies (2). This has sparked intense interest in finding ways to induce or augment TIL and thereby sensitize tumors to checkpoint blockade and related forms of immunotherapy.

High-grade serous carcinoma of the ovary (HGSC) provides a compelling illustration of the need, challenges, and opportunities related to immunotherapy. HGSC typically presents at an advanced stage and exhibits a high degree of intratumoral heterogeneity (3–6). Tumors harbor an intermediate mutation load but an extreme degree of copy-number variation (7). HGSC is highly sensitive to frontline treatment with platinum- and taxane-based chemotherapy; however, recurrence is common and ultimately fatal in the majority of cases. Survival rates for HGSC have improved only modestly over the past few decades (7). Despite these unfavorable features, a significant proportion of cases present with brisk TIL responses that are strongly associated with patient survival (8). Prognostically favorable lymphocyte subsets include CD8 T cells, CD4 T cells, memory B cells, and plasma cells (PC; refs. 9–14). These lymphocyte subsets appear to work cooperatively, as the highest patient survival rates are associated with tumors containing all four cell types (10). Prognostically favorable TIL are further demarcated by functional markers such as TIA-1, CD103, PD-1, and CD27 (11, 15–17). Thus, despite the aggressive biological and clinical features of HGSC overall, many patients mount spontaneous, multi-modal, prognostically favorable immune responses against their tumors.

In addition to effector TIL, immune infiltrates in HGSC contain multiple suppressive cell types and factors. Indeed, ovarian cancer was among the first malignancies in which the negative prognostic impact of FoxP3⁺ regulatory T cells (Treg) was demonstrated (13, 18). HGSC tumors are also frequently positive for the enzyme indoleamine-2,3-dioxygenase 1 (IDO-1), which inhibits T-cell responses by depleting local tryptophan and producing kynurenines (19). The PD-1/PD-L1 pathway is also active in HGSC. PD-L1 is frequently expressed by tumor-associated macrophages (TAM) and, in some cases, tumor cells (20). Moreover, the inhibitory receptor PD-1 is expressed by a significant proportion of CD4 and CD8 TIL (15). In general, these and other immune-suppressive factors are present in the same tumors that harbor effector TIL, resulting in an apparent immunological “stalemate.” Other tumors are devoid of effector or suppressor immune cells, instead appearing to be immunologically inert. Consistent with this, gene expression profiling of HGSC has revealed four molecular subtypes that range from immunologically active (C2/Immunoreactive) to intermediate (C1/Mesenchymal, C4/Differentiated) to inactive (C5/Proliferative; ref. 21, 22). Thus, to sensitize HGSC to immune modulation, some tumors will presumably require enhancement of pre-existing TIL responses, whereas others will require the induction of de novo TIL responses.

Although historically viewed as immunosuppressive, chemotherapy is now recognized to be immune stimulatory in some circumstances. Agents such as anthracyclines, oxaliplatin, cyclophosphamide, and taxanes can induce immunogenic cell death (ICD) in tumors, leading to enhanced presentation of tumor antigens to the immune system and augmented T-cell responses (23). The immunostimulatory activity of anthracyclines involves multiple mechanisms, which include autophagy, the endoplasmic reticulum (ER) stress response, calreticulin exposure on the cell membrane, release of ATP and high-mobility group box 1 (HMGBI) from dying cells, and type I interferon signaling (23, 24). Taxanes can induce ICD by inducing polyploidy in cancer cells, which in turn leads to ER stress (25). Paclitaxel can deplete myeloid-derived suppressor cells (26) and directly stimulate Toll-Like Receptor 4 (27). However, despite these important examples, many chemotherapies in current clinical use are unable to stimulate ICD, in large part due to an inability to induce ER stress and calreticulin exposure (28). Of direct relevance to HGSC, cisplatin fails to induce ICD in cancer cell lines (29) despite being a highly effective cytoreductive agent in this disease.

Apart from the above theoretical considerations, relatively little is known about the effects of chemotherapy on anti-tumor immunity in HGSC patients. Polcher and colleagues (30) reported that neoadjuvant chemotherapy (NACT; primary chemotherapy prior to surgery) with carboplatin and docetaxel was associated with increased CD4⁺, CD8⁺, and Granzyme B⁺ TIL in post-NACT tumors, whereas FoxP3⁺ TIL remained unchanged. Peng and colleagues (31) reported that CD8⁺ TIL increased in ovarian cancer patients undergoing NACT, and Wouters and colleagues (16) found a similar trend. Finally, Bohm and colleagues (32) recently reported that, in patients who responded well to NACT, there was a significant decline in stromal FoxP3⁺ cells and an increase in Th1/cytolytic T cell gene signatures. Collectively, the above studies indicate that chemotherapy can enhance CD8/Th1/cytolytic TIL and may reduce Tregs. However, it remains unclear whether these changes result in complete, prognostically favorable TIL patterns involving T cells, B cells, and PCs (10) and with accompanying effects on major immunosuppressive pathways. To address these issues, we performed a comprehensive assessment of the effects of NACT on immune infiltrates in HGSC. Our findings reveal, for the first time, the emergence of three distinct immunologic patterns during chemotherapy, each of which carries unique implications for immunotherapy.

Additional information is provided in Supplementary Materials.

All tumor specimens and clinical information were obtained with informed consent (or a formal waiver of consent) with approval by the Research Ethics Boards of the BC Cancer Agency, University of British Columbia, and University Health Network. Three retrospective cohorts of HGSC cases were evaluated (Table 1). Cohort A consisted of 26 cases for which matched pre- and post-NACT tumor samples were available from exploratory laparotomy and interval debulking, respectively (Supplementary Table S3). Where possible, we analyzed two different tumor samples from the same patient and averaged the results. Most pre-NACT tumor samples were derived from omentum (76.9%), and the remainder were from pelvic sites (e.g., ovary or fallopian tube; 34.6 %) or other extra-pelvic sites (e.g., uterus, colon, abdomen, or cul-de-sac; 19.2%). Similarly, post-NACT samples were derived from omentum (61.5%), pelvic sites (53.8%), or other extra-pelvic sites (11.5%). For 18 cases, we were able to compare matched pre- and post-NACT tumor samples from the same anatomical region (extra-pelvic). Patients received a mean of 4 cycles of NACT with carboplatin and paclitaxel prior to interval debulking, followed by additional cycles of chemotherapy. Cohorts B and C consisted of 64 HGSC cases who underwent NACT similar to Cohort A but from whom only post-NACT tumor samples (also obtained at the time of interval debulking) were available due to clinical or logistical barriers to obtaining pre-NACT samples for research purposes. The assembly of cohorts A-C is summarized in Supplementary Fig. S1, and corresponding clinicopathologic information is provided in Table 1. 

Single- and multi-color immunohistochemistry (IHC) were performed using antibodies against multiple immune cell and functional markers (Supplementary Tables S1 and S2). Slides were scanned using the Aperio (Leica Biosystems, Germany), Pannoramic MIDI (3DHistech, Budapest, Hungary), or Vectra (PerkinElmer, Waltham, MA, USA) imaging systems. Slides were subjected to manual or automated scoring to quantify TIL and related markers. Details of all statistical analyses are provided in Supplementary Materials. P values of ≤0.05 were considered significant.

Cohort A consisted of 26 HGSC cases for which sufficient viable tumor epithelium from both pre- and post-NACT samples was available to quantify TIL densities (Table 1; Supplementary Fig. S1). The mean number of NACT cycles was 4. The majority of patients were deemed good responders (88.5%) and deemed optimally cytoreduced at the time of interval debulking surgery (57.7%; Supplementary Table S3). Matched pre- and post-NACT tumor samples were analyzed by IHC for a broad panel of immune markers. In pre-NACT samples, CD3⁺ TIL were found in all cases but at widely varying densities (1.9–61.6 cells/100 tumor cells; Fig. 1A). Similar results were seen for both CD4⁺ TIL (0.1–18.9 cells/100 tumor cells) and CD8⁺ TIL (0.4–29.3 cells/100 tumor cells; Fig. 1B and C). In contrast, CD20⁺ TIL were found in fewer cases (77%) and at much lower densities (0–1.6 cells/100 tumor cells; Fig. 1D). Similarly, CD79a⁺ CD138⁺ PCs were present in the stroma of 75% of tumors (Supplementary Fig. S2). Finally, cells expressing CD1a, CD56, or CD68 were present in 65% (0–10.0 cells/100 tumor cells), 60% (0–3.4 cells/100 tumor cells), and 100% (0.7–37.4 cells/100 tumor cells) of pre-NACT tumors respectively.

Following NACT, the intraepithelial density of CD3⁺ T cells increased from a median of 7.0 to 18.6 cells/100 tumor cells (P = 0.013; Fig. 1; Supplementary Table S4). Significant increases were also seen for CD8⁺ TIL (from a median of 6.7 to 13.5 cells/100 tumor cells, P = 0.006) and CD20⁺ TIL (from a median of 0.2 to 1.0 cells/100 tumor cells, P = 0.006; Fig. 1; Supplementary Table S4). There was also a trend toward increased density of CD4⁺ T cells (from a median of 6.4 to 9.8 cells/100 tumor cells, P = 0.060; Fig. 1B; Supplementary Table S4). By contrast, there was no significant change in the stromal density of PCs (Supplementary Fig. S2) or the intraepithelial median density of cells expressing CD1a, CD56, or CD68 (Supplementary Table S4). Thus, NACT was associated with a selective increase in the density of intraepithelial T cells (primarily the CD8 subset) and CD20⁺ B cells.

To assess the functional profile of TIL, we first analyzed TIA-1, a marker of cytolytic granules (33–35). Similar to CD8⁺ TIL, the intraepithelial density of TIA-1⁺ cells increased significantly between pre- and post-NACT samples (from a median of 2.2 to 3.3 cells/100 tumor cells, P = 0.010; Fig. 1E; Supplementary Table S4). To investigate the cell types expressing TIA-1, we performed three-color IHC using antibodies to CD3, CD8, and TIA-1 (Fig. 1F). TIA-1 was found predominantly in CD3⁺ CD8⁺ T cells, and the median density of these cells increased significantly in post-NACT samples (P = 0.010; Fig. 1F). CD3⁻ CD8⁻ TIA-1⁺ T cells, most of which were likely NK cells, were the second most abundant subset, followed by small numbers of CD3⁺ CD8⁻ TIA-1⁺ T cells; however, neither of these cell types increased significantly after NACT (P = 0.93 and P = 0.93 respectively; data not shown). Likewise, we did not observe significant changes in the median density of Granzyme B⁺ cells (Supplementary Table S4). We also evaluated CD103, a marker of intraepithelial TIL (17) and found that CD103⁺ TIL trended toward increased density following NACT (P = 0.058; Supplementary Table S4).

Finally, we evaluated expression of MHC class I and II by tumor cells. 91% and 78% of pre-NACT samples were positive for MHC class I and II, respectively (data not shown). After NACT, there was a modest but significant increase in the intensity of MHC class II (P = 0.039) but not MHC class I (Supplementary Table S4).

We and others have previously shown that the density of CD3⁺, CD8⁺, CD20⁺, and TIA-1⁺ TIL in pre-treatment tumor specimens is strongly associated with survival (11, 13, 14). We hypothesized that post-NACT specimens might show an even stronger prognostic effect, given that the state of tumor immunity after treatment should have an important influence on the likelihood of disease recurrence. As cohort A (n = 26) was too small for outcomes analyses, this issue was addressed using an expanded cohort that included 64 additional cases (cohorts B and C) for which only post-NACT samples were available. The expanded cohort (n = 90) had similar clinicopathologic characteristics as cohort A (Table 1). A Cox proportional hazards analysis with multivariable comparison revealed CD20⁺ B cells were significantly associated with survival (P = 0.033; Table 2). Unexpectedly, we found no significant association between patient survival and the density of CD3⁺ T cells (P = 0.66), CD8⁺ T cells (P = 0.94), or TIA-1⁺ cells (P = 0.37; Table 2). Likewise, the ratio of CD8⁺ to FoxP3⁺ TIL was not associated with survival (P = 0.30, log-rank test; data not shown). Thus, with the exception of CD20⁺ B cells, post-NACT TIL patterns lacked prognostic significance.

To explain the limited association between post-NACT TIL patterns and survival, we hypothesized that NACT might also induce immunosuppressive cells and related factors. Therefore, we evaluated matched pre- and post-NACT samples from cohort A for the markers IDO-1, FoxP3, PD-1 and PD-L1. IDO-1 was expressed by tumor epithelium and/or infiltrating cells in 100% of pre-NACT samples and showed no significant change in frequency or intensity following NACT (Fig. 2A and data not shown). Similarly, intraepithelial FoxP3⁺ cells were found in 100% of pre-NACT samples (0.1–4.6/100 tumor cells) and showed no significant change after NACT (Supplementary Table S4). Intraepithelial PD-1⁺ TIL were present in 88% of pre-NACT samples (0–7.5/100 tumor cells) and, in contrast to IDO-1 and FoxP3, showed a marked increase after NACT (from a median of 1.2 to 1.8 cells/100 tumor cells, P = 0.012; Figs. 2B and 3A; Supplementary Table S4). To identify the cell types expressing PD-1, we performed four-color IHC to detect PD-1, CD8, FoxP3 and PD-L1 (Fig. 2C). Based on the first three markers, we identified three major PD-1⁺ TIL subsets: PD-1⁺ CD8⁻, PD-1⁺ CD8⁺ and PD-1⁺ FoxP3⁺ TIL (Fig. 2C); the latter subset likely represents Tregs, which we previously found to co-express FoxP3 and high levels of PD-1 (Supplementary Fig. S3; ref. 9). When considered individually, none of these three TIL subsets increased significantly following NACT (Fig. 3B); therefore, the overall increase in PD-1⁺ TIL might reflect smaller increases in all three subsets.

PD-L1 was expressed in 80% of untreated tumors and showed a patchy staining pattern, consistent with our prior report (20). Using two-color IHC, we found that PD-L1⁺ cells were a mixture of CD68⁺ macrophages and CD68⁻ cells, with the latter cells comprising a mixture of other infiltrating immune cells and tumor cells (Fig. 2D). Similar to IDO-1⁺ cells and Tregs, PD-L1⁺ cells showed no significant increase following NACT (Fig. 3C). This was true for both the CD68⁺ PD-L1⁺ and CD68⁻PD-L1⁺ subsets.

Thus, despite the observed increases in T cells and B cells following NACT, the density of immunosuppressive cell types remained stable. Furthermore, when assessed as single markers in post-NACT samples from the combined cohorts A-C, neither PD-1⁺ nor FoxP3⁺ cells showed prognostic significance (Table 2).

To investigate whether the observed increases in TIL after NACT reflected the enhancement of pre-existing responses or the emergence of de novo responses, we performed unsupervised hierarchical clustering of the pre-NACT data for the six IHC markers that were used in the survival analysis (i.e., intraepithelial CD3, CD8, TIA-1, CD20, FoxP3 and PD-1; Fig. 4A). This revealed three subgroups of patients. The first subgroup was positive for all or many of the immune markers, resembling typical immunoreactive tumors. The second subgroup showed low/intermediate levels of immune markers. The third subgroup was negative for all or most immune markers, resembling typical immunologically inert tumors. Examination of the corresponding post-NACT samples revealed three response patterns. In general, cases that were initially positive for immune markers showed even higher expression after NACT (pattern 1). Cases that initially had low/intermediate levels of immune markers generally became high after NACT (pattern 2), often to a similar extent as the first subgroup. Indeed, when unsupervised clustering was performed using the data from post-NACT samples, cases from patterns 1 and 2 clustered together, indicating they had equivalent TIL patterns (Supplementary Fig. S4A). Finally, cases that were initially negative for immune markers remained low/negative after NACT (pattern 3; Fig. 4A). The post-NACT patterns seen in the 26-case cohort were representative of the larger 90-case cohort in which survival analyses were performed (Supplementary Fig. S4B).

In general, similar response patterns were observed when unsupervised hierarchical clustering was performed using all immune markers and cell types, including data from both the epithelial and stromal compartments (Fig. 4B). This expanded analysis revealed that stromal immune response patterns generally mirrored the epithelial patterns. Furthermore, effector markers (e.g., CD4, CD8, TIA-1, CD20, CD3, CD103) generally clustered separately from suppressor markers/subsets (e.g., FoxP3⁺ PD-1⁺, IDO-1, PD-L1⁺ CD68⁺) in the epithelial compartment. However, despite clustering separately, effector and suppressor markers tended to exhibit similar patterns of change after NACT (Fig. 4B).

We investigated NACT-associated changes in TIL and related immunologic factors in HGSC. NACT was associated with increased densities of CD3⁺ and CD8⁺ T cells, CD8⁺ TIA-1⁺ T cells, PD-1⁺ T cells, and CD20⁺ B cells. Other immune subsets were unchanged, including CD79a⁺ CD138⁺ PCs and CD68⁺ macrophages. Moreover, no changes were seen in any of the immunosuppressive cell types examined, including FoxP3⁺ PD-1⁺ cells (putative Tregs), IDO-1⁺ cells, and CD68⁺ PD-L1⁺ macrophages. Post-NACT TIL patterns (based on the markers CD3, CD8, CD20, FoxP3, and PD-1) showed a limited association with patient survival. Unsupervised clustering of IHC data revealed that the enhancement of TIL seen after NACT was generally restricted to cases that were positive for TIL at baseline. Thus, NACT improves the infiltration of tumors by T cells and B cells in many cases, but the resulting immune cell patterns lack several major features associated with robust antitumor immunity, retain major suppressive mechanisms, and confer limited prognostic benefit. Our findings raise questions as to whether platinum- and taxane-based chemotherapy will prove an effective means of sensitizing tumors to checkpoint blockade and other forms of immune modulation in HGSC.

This study has limitations that should be factored into the interpretation of results. First, an unavoidable caveat of NACT studies is that, at the time post-NACT samples are collected, patients have typically only received half the number of chemotherapy cycles they will ultimately receive. Thus, the post-NACT immune patterns we evaluated might undergo further changes during subsequent rounds of chemotherapy. This is a difficult issue to address, as surgery is rarely performed at the completion of chemotherapy or at the time of relapse in HGSC. Nevertheless, post-NACT samples provide a useful “snapshot” of antitumor immunity at a time point when checkpoint blockade or other immunotherapies might be applied. Second, our main analyses were restricted to cases for which matched pre- and post-NACT tumor samples were available (Supplementary Fig. S1). The requirement for pre-NACT samples (which applied to cohort A only) led to the exclusion of five NACT cases that otherwise met our inclusion criteria but had insufficient specimen and/or information. The requirement for post-NACT samples applied to all three cohorts. For cohorts A and B, this led to the exclusion of 18 cases, seven of which represented complete responses. Importantly, however, 88.5% of cases in cohort A were deemed good responders to NACT, and 57.7% were deemed optimally cytoreduced at the time of interval debulking surgery (Supplementary Table S3). Therefore, despite excluding complete responders, our cohort contained a reasonable sample of cases demonstrating a favorable clinical response.

Our results regarding T-cell subsets are largely in agreement with prior studies in HGSC. Similar to Polcher and colleagues (30), we observed an increase in CD8⁺ TIL (and TIA-1⁺ TIL in our study) with no change in FoxP3⁺ TIL (nor FoxP3⁺ PD-1⁺ in our study) after NACT. Likewise, three other studies reported significant increases (31) or upward trends (16, 32) in CD8⁺ TIL after NACT. Accordingly, Bohm and colleagues (32) observed a significant increase in T_h1-associated gene expression. In apparent contrast to the present and prior studies (30), Bohm and colleagues detected a significant decline in stromal FoxP3⁺ cells in a subgroup of 7 cases that exhibited “good” responses to NACT, a finding that was supported by flow cytometric analysis of CD25⁺ FoxP3⁺ T cells. However, an important difference is that, in the Bohm study, patients with a “good” response to NACT had insufficient tumor content in post-NACT samples to evaluate TIL in the epithelial compartment; instead, by necessity, many of their comparisons were limited to tumor stroma. We noted a slight decrease in stromal FoxP3⁺ cells after NACT, but this did not reach significance (Supplementary Table S4). Despite these differences, the four studies support the general conclusion that NACT leads to increased cytolytic versus regulatory T cells in HGSC. Similar results have been reported in breast cancer, where increased CD8⁺ and decreased FoxP3⁺ TIL have been observed following anthracycline-based NACT (36–38).

NACT was also associated with a significant increase in CD20⁺ TIL, which indeed was the only subset associated with survival after NACT (Table 2). However, the density of PCs did not change, suggesting incomplete enhancement of the B cell arm of antitumor immunity. Studies in other tumor sites have yielded conflicting results regarding the effects of chemotherapy on B/PC responses. In breast cancer, two studies have reported decreased CD20⁺ TIL after anthracycline-based NACT despite sustained or increased T-cell infiltrates (38, 39). Conversely, in a murine prostate cancer model, oxaliplatin chemotherapy led to marked infiltration of tumors by IgA⁺ PCs, which expressed multiple immunosuppressive molecules (including PD-L1) and inhibited T-cell responses (40). Although we did not functionally characterize the PCs in post-NACT tumors, we did not see increased PD-L1 expression (even among CD68⁻ cells), arguing against a major influx of PD-L1⁺ PCs. Furthermore, in contrast with prostate cancer, tumor-associated PCs in HGSC express IgG and are associated with higher CD8⁺ TIL and favorable prognosis, indicating they represent a distinct functional class (10).

Unexpectedly, none of the five T-cell markers (CD3, CD8, TIA-1, FoxP3, or PD-1) showed prognostic significance in post-NACT samples, despite being strong prognostic markers in prior studies of untreated tumors (11–13, 15). Similarly, Wouters and colleagues (16) reported that neither CD8⁺ nor CD27⁺ TIL had prognostic significance in post-NACT samples. This appears to contradict the report by Polcher and colleagues (30) that low FoxP3⁺ TIL (or a high Granzyme B:FoxP3 ratio) in post-NACT samples is associated with increased survival. However, their conclusion was based on only 29 HGSC cases, in contrast to the present (n = 90) or Wouters (n = 84) cohorts. Why might post-NACT TIL lack prognostic significance? First, patients who underwent a complete response during NACT were, by necessity, excluded from our analysis. Second, many patients undergo NACT because they have more extensive disease at diagnosis, which could override the prognostic significance of TIL. That said, TIL carry prognostic significance even in suboptimally debulked patients (14, 41). Third, patients with an exceptionally poor response to NACT might not have undergone interval debulking surgery. These possibilities could potentially be resolved by evaluating TIL patterns in matched pre-NACT samples; however, such samples were not available for the majority of cases in cohorts B and C. Thus, this issue awaits further study using larger collections of matched tumor samples.

An additional possibility for the lack of prognostic significance of post-NACT intraepithelial T cells is that they might have low tumor specificity and/or functional competence. Notably, not all TIL are genuinely tumor-reactive; a significant number are seemingly irrelevant bystanders that are drawn non-specifically to the tumor microenvironment. It is possible that the increased TIL densities after NACT could reflect the selective killing of tumor cells over bystander lymphocytes, or increased infiltration of bystander cells in response to inflammatory signals triggered by chemotherapy. An indicator of bonafide tumor recognition by CD4 and CD8 T cells is the production of interferon-γ (IFN-γ), which in turn upregulates target genes such as MHC class I and II, IDO-1, and PD-L1, a phenomenon termed “adaptive resistance” (42). Indeed, in a murine ovarian cancer model, paclitaxel treatment induced the infiltration of tumors by CD8⁺ T cells and increased PD-L1 expression by tumor cells (31). However, in the present study, the increase in CD8⁺ TIL seen after NACT was not accompanied by increased expression of PD-L1, IDO-1 or MHC class I, and only modest upregulation of MHC class II was observed. Moreover, we did not see increased infiltration by Tregs, which can also be induced by CD8⁺ TIL (43). Thus, despite marked changes in multiple TIL subsets after NACT, we saw little evidence of a counter-response by the tumor microenvironment, raising questions about the tumor specificity and/or functional competence of these TIL. In future, it will be important to directly address this issue by performing tumor recognition experiments using viable TIL from interval debulking samples.

Hierarchical clustering of our IHC data revealed three general response patterns, each of which may require a different approach to immunotherapy. Pattern 1 represented cases that were positive for TIL at baseline and presumably represent the C2/Immunoreactive molecular subtype (21, 22). These cases generally showed a further increase in TIL after NACT. However, this was offset by the retention of multiple immunosuppressive markers in these tumors, including PD-L1, PD-1, FoxP3, and IDO-1. Given this assortment of suppressive factors before and after NACT, such cases might require multi-pronged immune modulation to unleash effector TIL activity. Pattern 2 included cases with low/intermediate TIL at baseline and may correspond to the C1/Mesenchymal and/or C4/Differentiated subtypes. They too exhibited an increase in TIL after NACT, achieving a pattern indistinguishable from pattern 1 (according to the markers used here). Since these tumors went from being TIL^low-negative to TIL-positive after NACT, they had the potential to switch from an unfavorable to favorable prognosis. However, given the limited prognostic significance of TIL in post-NACT samples, it appears this did not happen. Instead, we speculate that the incoming T cells in this subgroup may lack tumor specificity and/or be impeded by the various immunosuppressive mechanisms in the tumor microenvironment. It will be important to distinguish these possibilities to design effective immune interventions for this relatively large subgroup. Finally, pattern 3 comprised cases that had negligible TIL before or after NACT and likely represent the C5/Proliferative molecular subtype (21, 22). The mechanistic basis for this immunologically inert phenotype is unknown but could reflect lymphocyte infiltration barriers, hostile vasculature, immunosuppressive signaling pathways, unfavorable epigenetic patterns, or other factors that, with better understanding, could potentially be reversed (2, 8, 44, 45). Encouragingly, the three TIL response patterns reported here were detectable at baseline (Fig. 4), suggesting that, with further refinement and validation, it may be possible to use pre-treatment TIL patterns to predict the optimal immunotherapeutic intervention for individual patients.

No potential conflicts of interest were disclosed.

Conception and design: C.S. Lo, S. Sanii, B.H. Nelson

Development of methodology: C.S. Lo, D.R. Kroeger, K. Milne

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C.S. Lo, S. Sanii, D.R. Kroeger, K. Milne, K. Rahimi, P.A. Shaw

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C.S. Lo, D.R. Kroeger, A. Talhouk, D.S. Chiu, B.A. Clarke, B.H. Nelson

Writing, review, and/or revision of the manuscript: C.S. Lo, S. Sanii, D.R. Kroeger, K. Milne, A. Talhouk, D.S. Chiu, P.A. Shaw, B.A. Clarke, B.H. Nelson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C.S. Lo, K. Milne

Study supervision: B.H. Nelson

We thank Yuzhuo Wang, Ladan Fazli, and Alireza Moeen Rezakhanlou for assistance with image analysis; Peter Watson and John Webb for advice regarding TIL scoring; Blake Gilks, Jessica McAlpine, Dianne Miller, Cheng-Han Lee, Janice Kwon, Devi Dhillon, and Margaret Luk for assistance with patient cohorts; and our patients for donating biospecimens and clinical data.

Supported by Canadian Institutes of Health Research (Awards MOP-137133 and MOP-142436), U.S. Department of Defense (Award W81XWH-12-1-0604), and BC Cancer Foundation.

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.