Targeting the immune checkpoint molecules CTLA-4, PD-1, and PD-L1 is revolutionizing clinical management for a variety of solid and hematological malignancies by generating durable anti-tumor immune responses (1, 2). The success of these drugs is nevertheless currently limited to a subset of patients whose tumors likely produce neoantigens and/or whose anti-tumor responses are constrained by immune inhibitory pathways. Immune stimulatory/co-stimulatory pathways are tightly regulated by feedback inhibitory pathways that limit normal immune responses to prevent excessive activity. Thus, molecules controlling these pathways are being actively investigated to identify new and potentially more refined targets for cancer immunotherapy.

The tumor immune microenvironment varies considerably between solid tumor types as well as within tumors of the same organ or tissue developing in different individuals. There is a general consensus that chronic inflammation promotes tumor progression with tumors at diagnosis ranging from a baseline inflamed phenotype to an immunological desert (3). In breast cancer (BC), a variety of studies have investigated the clinical value of tumor-infiltrating leukocytes [CD45⁺ cells that are >90% lymphocytes] (4) and immune gene signature expression (5–8) with significant findings obtained for HER2-positive (HER2⁺) and triple negative (TN) BC. Our recent studies show that tumor-infiltrating lymphocytes (TIL) density in a cohort of BC patients (all subtypes) whose tumors were analyzed within hours of surgery forms a continuum (9). Using a threshold defined by normal breast tissues, we identified 25% of tumors as TIL-negative (TIL^neg). TIL-positive tumors (TIL^pos; 75%), equally divided into TIL-intermediate (TIL^int) and TIL-high (TIL^hi), were defined by a second threshold set based on non-adjacent non-tumor tissue from the same BC patients. We further found a positive correlation between the extent of TIL, the presence of organized tertiary lymphoid structures (TLS), and PD-1⁺ T cell TIL and/or PD-L1⁺ immune cells (7, 9, 10). Our data reveal significant heterogeneity and complexity in the adaptive anti-tumor immune response, which suggests that BC stratification based on the balance of TIL subpopulations and/or TLS might be clinically relevant and identify patients most likely to respond to immune checkpoint inhibitors.

We previously reported that TLS are principally located in the stroma of up to 60% of BC (7, 9). TLS were initially identified in lung (11) and colorectal (12) cancer but have now been associated with positive clinical outcomes in an expanding number of solid tumors (13). TLS architecture is characterized by a T cell zone adjacent to a B cell follicle similar to secondary lymphoid organs (SLOs) and mucosal-associated lymphoid tissues (e.g., tonsils). An active germinal center (GC) is often detected in the B cell follicle, providing a site for B cell differentiation and maturation into antibody producing plasmacytes or memory B cells. Specialized CD4⁺ T cells named T follicular helper cells (Tfh) play a key role in GC formation and B cell differentiation/maturation. Our previous work identified PD-1^hiCD200^hi Tfh cells as an important component of BC TIL and linked their presence with tumor-associated TLS and good clinical outcomes, particularly in HER2⁺ and TNBC (7). Tfh cells have been identified in other solid tumors (14), including pancreatic adenocarcinoma where TLS with GC were characterized by PD-1⁺ TIL in 32/104 patients (15). Our latest study identified a subpopulation of CXCR5⁻PD-1^hiICOS^intCD4⁺ Tfh TIL (named TfhX13 cells) that produce the chemoattractant CXCL13 and thereby recruit CXCR5⁺ T and B cells to TLS (10). Interestingly, we also detected a FOXP3^hiCXCR5⁻PD-1^intICOS^hiCD4⁺ TIL subpopulation, which is primarily regulatory T cells (Tregs), whose presence is linked with TfhX13 TIL accumulation. Together, our data suggest that PD-1/PD-L1 expression in BC may reflect ongoing, active immune responses rather than (or in addition to) immunosuppression via this pathway.

PD-1 is an important immune checkpoint molecule, which together with its principal ligand PD-L1, is actively targeted in the clinic. Despite promising initial outcomes in melanoma, lung, and kidney cancer (16, 17), the clinical benefit of immune checkpoint inhibitors in BC has been restricted to a subset of patients and effective predictive biomarkers remain undefined (18). Clinically, PD-1 expression in BC has been associated with good (7, 19) or bad (20) outcomes. Similarly, PD-L1 expression was associated with a good (19, 21–23) or bad prognosis (24–26). No clinical significance has been attributed to expression of the alternate PD-1 ligand PD-L2 (27). The first immune checkpoint molecule targeted in the clinic, CTLA-4, was analyzed in BC with the presence of CTLA-4⁺, PD-1⁺, Helios⁺, FOXP3⁺, GITR⁺, and/or CD103⁺ Tregs suggestive of an immunosuppressive microenvironment (28, 29).

CTLA-4 expression on tumor cells was associated with a bad prognosis (30) while good outcomes were observed when CTLA-4⁺ TIL were present (19, 31, 32).

The remarkable responses of immunotherapeutic agents in some patients is driving early-phase clinical studies designed to broaden their efficacy and benefit by combining agents that target non-redundant inhibitory pathways. LAG3, which inhibits the activity of MHC-II⁺ antigen presenting cells (APCs), was evaluated in a retrospective cohort of TNBC with 18% positivity detected; however, its expression was not correlated with outcome despite a trend toward a better prognosis for LAG3-positivity (33). Another target of clinical interest is TIM3, a molecule that curbs effector T cell activation after binding to galectin-9 on immune cells as well as stromal and tumor cells (34, 35). Interestingly, TIM3⁺ Tfh cells from BC patient blood were shown to have reduced functionality (36). Numerous ongoing clinical trials are combining agents targeting the PD-1/PD-L1 pathway with those targeting CTLA-4, LAG3, or TIM3.

The genetic and immune heterogeneity of BC together with the restricted number of patients that currently derive benefit from immunotherapy exposes an urgent need to improve patient selection for immunotherapy. The purpose of this study was to evaluate expression of PD-1, PD-L1, PD-L2, CTLA-4, LAG3 and TIM3 in the primary BC microenvironment. A combination of flow cytometry (FACS) and immunohistochemistry (IHC) allowed us to quantify their expression in fresh tissue homogenates coupled with a determination of their spatial location in the tumor.

The work presented in this study is the first, to our knowledge, to characterize a panel (PD-1, PD-L1, PD-L2, CTLA-4, LAG3, and TIM3) of targetable inhibitory immune checkpoint molecules in primary BC. This analysis was undertaken to (1) correlate their expression in the BC molecular subtypes; (2) associate specific BC cellular subpopulations with clinicopathological data; (3) situate their location in the tumor microenvironment; and (4) correlate their presence with BC prognosis. Overall, immune checkpoint molecule expression was significantly higher in basal-like (~80% TNBC) and HER2-enriched (~80% HER2⁺) compared to the other BC molecular subtypes. PD-1 and CTLA-4 were uniquely expressed on T cell TIL, while LAG3, TIM3, PD-L1, and PD-L2 were detected at low levels on TIL in tumors with extensive infiltration and TLS. PD-L1, LAG3, and TIM3 were also expressed on stromal and other immune cells in the tumor microenvironment. PD-1, PD-L1, and CTLA-4 gene expression was significantly associated with improved clinical outcome in basal-like and HER2-enriched BC. These data are consistent with previous studies using immune gene signatures (5–7, 49) and/or evaluating the overall presence of TIL (50–52). Our observation that this panel of immune checkpoint molecules is predominantly expressed on TIL suggests that their presence and abundance parallels the TIL continuum we recently established for primary BC (9). These data also reveal a contrast between BC and solid tumor types where checkpoint molecules are frequently expressed on tumor cells, such as melanoma, renal, and lung cancer (16, 17). This study validates the clinical value of a preexisting active immune response at the tumor site, particularly in TN and HER2⁺ BC (4).

Our analysis of TIL in freshly prepared tissues together with their corresponding FFPE blocks produced some new and interesting findings. We found no differences in the proportion of PD-1⁺ TIL within the CD4⁺ or CD8⁺ compartments in the 56 primary BC analyzed. We further show that PD-1 expression is heterogeneous on both CD4⁺ and CD8⁺ T cell TIL with PD-1^hi, PD-1^int, and PD-1^neg subpopulations distinguishable by FACS, particularly in tumors with extensive TIL and TLS. Heterogeneity in the degree of PD-1 expression between CD4⁺ T cell subpopulations could reflect CD4⁺ TIL with different functional profiles (i.e., PD-1^hiCD4⁺ = Tfh; PD-1^intCD4⁺ = Treg, PD-1^negCD4⁺ = naïve T cells, etc.). We recently identified PD-1^hiICOS^intCD4⁺ TIL as a subpopulation containing effector helper (Th) cells and specialized CXCL13-producing Tfh cells, called TfhX13, which are thought to be important mediators of TLS formation and immune cell recruitment (10). We also showed that PD-1^intICOS^hiCD4⁺ TIL contain a preponderance of FOXP3^hi Tregs. Our observation that PD-1^hiICOS^int and PD-1^intICOS^hi CD4⁺ TIL are linearly correlated in 90% of BC, suggests that effector Th plus TfhX13 TIL frequently expand in parallel with activated Treg TIL (10), thereby generating anti-tumor immunity and memory that is regulated by immune feedback mechanisms. Following this rationale, an imbalance favoring regulatory cells would promote an immunosuppressive state that essentially disables active immune responses. In a recent study of claudin-low BC (a TNBC subset), investigators found that a substantial proportion of TIL in these aggressive tumors are Tregs. They further demonstrated that significant Treg depletion together with checkpoint inhibition was necessary for effective anti-tumor immunity (53). The role of Tregs in BC remains controversial with both positive and negative effects on outcome reported, potentially because their ratio to effectors is an important influence on immune activity and clinical outcome. This idea is supported by a study showing that higher ratios of FOXP3⁺CD4⁺ to CD8⁺ T cells in biopsies from BC patients with ductal carcinoma in situ predict their relapse (54). These investigators also observed that the same high ratio in adjacent normal tissue was predictive of poor outcomes.

PD-1^hiCD8⁺ and PD-1^intCD8⁺ TIL were detected in all of the TIL^int and TIL^hi primary BC we analyzed although their proportions varied. While a functional role for these CD8⁺ T cell subpopulations is currently unknown, CD8⁺PD-1^int TIL are thought to be more responsive to PD-1 or PD-L1 blocking agents (55). A separate study found that high PD-1 and PD-L1 expression in pretreatment biopsies was significantly correlated with higher TIL and higher rates of pathological complete response after neoadjuvant treatment (44). Further investigation is needed to determine whether the balance between PD-1 (PD-1^hi, PD-1^int or PD-1^lo) CD4⁺ and CD8⁺ TIL subpopulations is linked with clinical responses to checkpoint inhibitors targeting the PD-1/PD-L1 pathway.

The PD-1^hi or PD-1^int TIL subpopulations we identified by FACS (10) (Figure S1 in Supplementary Material) were also detected on dual PD-1/PD-L1 IHC-stained sections (Figure 2C). Interestingly, the PD-1^int TIL are often observed adjacent to areas containing PD-L1⁺ cells similar to SLO tissues. We further found that PD-1^hi TIL are principally located in the GC of a TLS while the PD-1^int TIL are outside the GC but primarily within a TLS or TIL aggregate. Taken altogether, these data suggest that PD-1 expression reflects an activated T cell presence in the tumor microenvironment; however, there are inconsistent findings between tumor types. In lung tumors with extensive CD8⁺ TIL, two groups were identified: (1) low immune checkpoint molecule expression, mature DCs located in TLS, and a good prognosis or (2) high checkpoint expression associated with a lack of mature DCs and an increased risk of relapse (56). In pancreatic cancer, B cells were localized in the tumor bed and TLS, with the latter alone predicting longer survival (15). Interestingly, a B cell TLS presence was associated with a GC gene signature, correlated with CD8⁺ TIL and a positive prognosis in these patients. Future studies using multi-plex IHC should help to precisely establish critical location-based interactions and balances between checkpoint molecules on effector and regulatory TIL that drive immune activation or suppression and help predict clinical responses.

An initial FACS study of PD-L1 expression on TIL and epithelial cells examined four tumors with 2/4 defined as positive (24). The present study analyzed both PD-1 ligands, PD-L1 and PD-L2, on fresh tissues from a substantially larger cohort of primary BC patients. Both studies found a low prevalence of PD-L1 positive TIL. Our previous study using IHC to examine full tumor sections found that PD-L1 expression is detected predominantly in TNBC and positively correlated with the extent of TIL and a peri-tumoral TLS presence (9). In addition to its expression on TIL and tumor cells, PD-L1 is also found on other immune cells, CD68⁺ tumor-associated macrophages and myeloid cells, such as follicular dendritic cells resident in TLS. Our results from the METABRIC gene expression analysis reveal that PD-L1 expression is associated with a better OS in basal-like BC, which is likely linked with the good prognostic value of higher TIL in this subtype. PD-L1 expression in primary BC has been widely investigated at both the RNA (57, 58) and protein (9, 24, 57, 59–63) levels. The study by Ali et al. (57) is noteworthy because it examined PD-L1 expression in 3,916 BC, finding 6% PD-L1⁺ immune cells (235/3,916) and 1.7% PD-L1⁺ tumor cells (66/3,916) by IHC. These percentages are lower than other studies presumably because IHC was performed on tissue arrays derived from archival samples, where there is an increased risk of PD-L1 degradation. Overall, the relationship between PD-L1 expression and BC prognosis remains an open question, with some studies correlating it with improved survival (57–59, 62, 64) and others not (24, 62, 63). These inconsistencies should be resolved once a standardized IHC assay (65) becomes more widely employed.

These observations reveal that PD-L1 expression is infrequent in BC with the highest prevalence observed in TNBC. Consequently, TNBC is the target of most current BC immunotherapy trials with anti-PD-1 and anti-PD-L1 agents demonstrating the best clinical benefits (although limited to a subset of patients) when given in the advanced and neoadjuvant settings (18). Tumor responses were highest in TNBC for first-line treatment of metastatic disease or when CD8⁺ TIL were present or if PD-L1 positivity was detected. At ESMO 2017, Loi et al. presented data that stromal TIL, assessed on hematoxylin and eosin-stained metastatic tissues (including metastatic LNs), were higher in treatment responders, particularly if anti-PD-1 (pembrolizumab) was given in the first line (66). Their analyses included two metastatic TNBC cohorts: one pre-treated for metastatic disease and not selected based on PD-L1 expression and the other untreated for metastatic disease and PD-L1 positive (KEYNOTE-086; monotherapy with anti-PD-1). Interestingly, stromal TIL were independent predictors of response to anti-PD-1 while PD-L1 positivity did not add any predictive value.

Examination of the other PD-1 ligand revealed that while PD-L2⁺ TIL were seldom detected in BC, their presence was associated with PD-1^hiCD8⁺, PD-L1⁺, and LAG3⁺ TIL. Expression of PD-L2 in tumors is rare and modest, principally concentrated in the B cell follicle similar to SLOs, where it has been previously shown to be typically expressed on activated APC (DCs and macrophages) (67). In contrast to tonsils, where PD-L2 expression is intense and scattered throughout the follicle, the few positive breast tumors we observed had weaker staining with only a few positive cells, like LNs and spleen. Previous studies found PD-L2 negativity in 49% of BC (N = 192) (27) or 20% of TNBC and HER2⁺ BC (N = 25) (68). The first study based positivity on nuclear and cytoplasmic staining while the second study was based on nuclear staining alone potentially explaining their differences in frequency. We scored positivity as membrane staining based on our analysis of SLOs, with normal breast and other healthy tissues consistently negative. Based on our data, PD-L2 does not appear to be a promising target for BC immunotherapy; however, additional tumors need to be analyzed to clarify discrepancies between studies.

CTLA-4 expression, principally on CD4⁺ T cells, was heterogeneous among our BC subtypes with CTLA-4^hiCD4⁺ TIL most frequently identified in TNBC (71%) and HER2⁺ (60%) BC. This potentially is a direct reflection of Treg abundance, an idea supported by our recent data showing PD-1^intiCTLA-4⁺FOXP3⁺CD4⁺ TIL are associated with extensive infiltration and TLS in BC (10). These data are supported by several studies, including an analysis of >100 treatment naïve BC patients where Treg abundance in the tumor microenvironment was characterized by high levels of surface CCR8, CTLA-4, and PD-1 (69). Another study similarly found that FOXP3⁺Helios⁺CTLA-4^hiPD-1^hiCD4⁺ TIL characterize Tregs infiltrating BC (28). Overall, current findings indicate that dual checkpoint inhibition targeting CTLA-4 and PD-1 potentially has important synergistic effects due to their distinct regulatory activities (70).

LAG3 expression was associated with PD-L2⁺ and TIM3⁺ TIL in our cohort, which is in line with current thinking that LAG3 positivity reflects an ongoing, complex immune response regulated by several regulatory pathways (71). A recent study of two independent cohorts (N = 259 and N = 104) found 15% of early-stage TNBC expressed both PD-1 and LAG3 (33). In this study, PD-1⁺ and LAG3⁺ TIL were positively correlated with the extent of infiltration, particularly CD8⁺ TIL. The METABRIC dataset did not show any significant impact on survival in primary BC molecular subtypes; however, these survival data need confirmation from independent cohorts. A phase I trial where a LAG3 agonist was administered with paclitaxel detected increased activation of APC, NK, and CD8⁺ effector/memory cells and a 50% objective response rates in patients with advanced luminal BC (72). Currently, there are a number of trials evaluating anti-LAG3 agents in association with PD-1 or PD-L1 blocking agents in many advanced solid tumors including BC (71).

The data presented here show LAG3⁺ TIL are associated with TIM3⁺ TIL, and TIM3⁺CD4⁺ and TIM3⁺CD8⁺ TIL are marginally positively correlated with TLS. An earlier study of TIM3 in BC found higher expression on CD8⁺ TIL from primary tumors compared to their blood counterparts (73). Zhu et al. recently found TIM3 expression on Tfh cells (CXCR5⁺ICOS⁺ CD4⁺ T cells) from BC patient blood at the same frequency as healthy control blood; however, they identified higher numbers of TIM3⁺ Tfh cells in patient blood (36). These blood TIM3⁺ Tfh cells were less potent at stimulating naïve B cells in vitro, suggesting a possible dysfunctional profile in this subpopulation. They further showed that TIM3⁺ Tfh TIL were higher frequencies than in the patient’s blood. Our laboratory identified Tfh TIL as a hallmark of TLS with concentrations of PD-1^hi Tfh cells characterizing TLS with a GC (7, 10). Zhu et al. (36) were able to rescue Tfh cell functions (IL-21 and CXCL13 expression) by blocking TIM3, suggesting this as a potential mechanism for regulating Tfh-mediated activities in TLS. In another study, CD8⁺ TIL cultured with IL-15 expressed higher levels of TIM3⁺, while inhibiting TIM3 and IL-15 induced CD8⁺ TIL proliferation and IFNγ production (73). We show here that in addition to lymphocytes, TIM3 is expressed on cells from the myeloid lineage, such as macrophages and stromal cells, which could contribute to more widespread inhibition in the tumor microenvironment [reviewed in Ref. (74)]. Targeting TIM3 offers a promising new direction for immune checkpoint inhibition via synergistic interactions between inhibitory signals from the microenvironment and TIL.

Our data show that the extent of TIL and their organization in TLS are linked with expression of immune checkpoint molecules in BC, particularly in the TNBC and HER2⁺ subtypes, which are more frequently extensively infiltrated. Many TLS⁺ tumors had expression of one or more immune checkpoint molecules suggesting multiple anti-tumor immune responses at different stages of activity occur simultaneously in the tumor microenvironment. These responses are likely regulated by factors and cells in their immediate vicinity. Our observations in BC TLS paralleled SLOs (tonsils and LNs) suggesting that checkpoint pathways are influenced and functioning in response to normal immune control mechanisms. The positive correlation and association we found between immune checkpoint molecule expression and baseline TIL and TLS indicates that evaluation of these parameters in BC patients might identify tumors more likely to respond to immune modulating therapies.

This study was carried out in accordance with the recommendations of the Institut Jules Bordet and GZA Ziekenhuizen medical ethics committees with written informed consent from all subjects. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the Institut Jules Bordet and GZA Ziekenhuizen medical ethics committees.

CS and KW-G conceived the research and designed the experiments; CS performed the majority of experiments with critical support from SG, PS, AB, EM, GN, and HD; CS collected the clinical data; GE, AW, and DL performed the pathological evaluations; PR, JV, FR, and VD performed the statistical and public data analyses; LC, IV, and DL provided the patient samples; AA and MP-G. proposed important concepts; KW-G. supervised the research; CS and KW-G. analyzed and interpreted the data and wrote the manuscript; all authors provided advice on the final manuscript.

The present research was conducted in the absence of any commercial or financial relationships. Authors and their institutions did not receive any payment or services from a third party for any aspect of the submitted work. No financial relationships exist with entities that could have influenced or given the appearance of potentially influencing what we wrote in the submitted work.