PD-L1 is the ligand of PD-1 and is related to immunosuppression. Under normal circumstances, the immune system reacts to foreign antigens in the lymph nodes or spleen by promoting activation of antigen-specific cytotoxic T cells (such as CD8⁺ T cells). PD-1 combined with PD-L1 can transmit inhibitory signals and reduce the proliferation of CD8⁺ T cells in lymph nodes, which leads to immune escape of tumor cells. The PD-1/PD-L1 inhibitors interrupt binding of PD-L1 to PD-1, and in this way, tumor cells cannot transmit inhibitory signals to T cells, and T cells recognize and destroy cancer cells. About 20% of TNBC cells express PD-L1 (15, 60). Several studies have explored the predictive value of PD-L1 for immunotherapy in TNBC (16, 19, 22, 39–41). However, inconsistent results have been shown in different studies (23, 40, 42, 43).

Some studies have provided evidence about the predictive value of PD-L1 for efficacy of ICIs in TNBC [ Table 2 (18–21–23, 29–34, 37)]. PD-L1 has been shown to predict the efficacy of PD-1/PD-L1 inhibitors in mTNBC, whether in monotherapy or combination therapy (22, 39). Atezolizumab-treated patients with advanced TNBC in the PD-L1-positive population had a higher objective response rate (ORR) compared with the PD-L1-negative population (22.2% vs 2.6%) (39). Similarly, the JAVELIN study reported that mTNBC patients with higher PD-L1 expression had better efficacy of atezolizumab (22). The IMpassion 130 and KEYNOTE-355 studies indicated that PD-L1 positivity was related to better efficacy of PD-1/PD-L1 inhibitors in mTNBC (16, 17, 19). These studies have suggested that PD-L1 expression can identify patients who will benefit from ICIs.

Conversely, other clinical trials have not supported PD-L1 as a predictor of the efficacy of ICIs (23, 42). The KEYNOTE-522 and IMpassion 031 studies in early TNBC patients showed that, irrespective of positive or negative PD-L1 expression, PD-1/PD-L1 inhibitors combined with chemotherapy had a higher pathological complete remission (pCR) than placebo combined with chemotherapy (23, 42). Notably, the patients in these studies had early rather than advanced TNBC. The different results between early and advanced TNBC patients suggest that PD-is not an ideal biomarker and its predictive value varies according to individual immune function and/or disease setting. However, the potential mechanism underlying these results remains unclear.

The potential prognostic value of PD-L1 in TNBC remains contentious. Some studies have provided evidence that PD-L1-positive may be associated with better prognosis (40, 41, 61, 62). A meta−analysis reported that PD-L1-positive on tumor cells was related to poor prognosis, whereas PD-L1-positive on TILs was associated with better survival (41). Li et al. found that PD-L1 expression on TILs suggested better disease-free survival (DFS) in TNBC (61). However, Barrett et al. found that PD-L1-positive on tumor cells was associated with prolonged OS in patients with TNBC (40). Similarly, Botti et al. showed that PD-L1-positive on tumor cells was associated with better DFS in TNBC (62). However, other studies questioned the potential prognostic value of PD-L1 in TNBC (63, 64). A meta-analysis exploring the relationship between PD-L1 and prognosis in TNBC found no significant association between PD-L1 expression and OS (64). Thus, the prognostic value of PD-L1 in TNBC remains unclear and further studies are required.

The inconsistency of these studies suggests that the PD-L1 expression is affected by factors such as complex immune environment and different detection methods. First, expression of PD-L1 is regulated by various mechanisms including the signal transducer and activator of transcription 3 and nuclear factor-κB pathways (65). Additionally, the function of PD-L1 is influenced by ubiquitination, glycosylation, phosphorylation and methylation (65). Therefore, expression of PD-L1 may be altered over time and be induced by other therapies. Second, the different antibodies or detection methods may have affected the results of PD-L1 expression in many studies (66–68). Antibodies for the detection of PD-L1, such as 28-8, 22C3, SP263 and SP142, have been approved as companion/complementary diagnostics to nivolumab, pembrolizumab, durvalumab and atezolizumab, respectively (69). In clinical application, the major differences among the four antibodies are mainly stained cells and scoring criteria of PD-L1. The Blueprint Project showed that 28-8, 22C3 and SP263 mainly stained tumor cells, and the test results were similar. SP142 stained immune cells more prominently than the other antibodies did. Compared with tumor cell staining, immune cell staining was more heterogeneous (70, 71). At present, there are four scoring criteria of PD-L1: combined positive score (CPS), tumor proportion score, immune cell score and tumor cell score (72, 73). Because of the different scoring criteria, the definitions of PD-L1-positive tumors were different. For example, PD-L1-positive tumors in the IMpassion 130 study were defined as staining of any intensity in immune cells occupying ≥1% of the tumor area tested by SP142 (16). In the KEYNOTE-355 study, PD-L1-positive tumors were defined as CPS ≥1 and CPS ≥10, where CPS was the ratio of PD-L1-positive cells (tumor cells, lymphocytes and macrophages) to the total number of tumor cells tested by 22C3, multiplied by 100 (19). However, these four evaluation methods have not been comprehensively compared; therefore, which method can better reflect the expression level and predictive value of PD-L1 requires further study. Additionally, there were temporal and spatial differences in PD-L1 expression between primary and metastatic lesions (74). Expression of PD-L1 in the metastatic site was significantly lower compared with the primary site (75, 76). Factors such as the empirical judgment of pathologists, heterogeneity of PD-L1, and the effect of drugs might also interfere with PD-L1 expression (65, 70, 77–79).

In summary, the potential predictive and prognostic values of PD-L1 in TNBC remain controversial. Understanding of the tumor, microenvironment, and host factors that influence response to ICIs may contribute to identifying more reliable biomarkers (80). Accurate methods are needed to detect PD-L1 expression and guide precision medicine (81). At present, identifying patients who can benefit from ICIs partly relies on immunohistochemical assays used in clinical trials (82). However, it is difficult to detect the dynamic change in PD-L1 expression, and some factors can interfere with the results. Therefore, the determination of the optimal assay will require further rigorous studies. Scoring systems and thresholds for PD-L1 positivity lack standardization, and this may affect the judgment of PD-L1 positivity. Fortunately, this study area is rapidly developing and PD-L1 as a potential prognostic and predictive biomarker will be fully optimized for TNBC in the future.

CTLA-4 is one of the immunoglobulin superfamily and a signal receptor on the T-cell membrane (83). It is homologous to CD28 on the surface of T cells and competes with CD28 to bind to B7-1 (CD80) and B7-2 (CD86) on antigen-presenting cells, although it has a stronger affinity for B7-1 and B7-2 (84). When B7 binds to CD28, it initiates an activation signal, which is inhibited when CTLA-4 binds to B7 (85). Normally, CTLA-4 participates in negative immunoregulation. However, tumors can also participate in these immunoregulatory pathways by expressing CTLA-4, which decreases immune cell functions (86, 87). Some studies have found that the high levels of CTLA-4 correlate with better efficacy of anti-CTLA-4 therapies in melanoma (88, 89). However, there is a lack of data from clinical trials about its predictive value for ICIs in TNBC. In addition to the above, the potential prognostic value of CTLA-4 in BC has been reported (55, 56). Yu et al. analyzed tissue samples from 130 BC patients who underwent surgery. They found that more interstitial CTLA-4⁺ lymphocytes were related to longer DFS and OS, whereas more CTLA-4⁺ tumor cells were related to shorter DFS and OS (55). Lu et al. analyzed an RNA-sequencing dataset and found that BC patients with high CTLA-4 expression had a significantly elevated risk of death compared with those with low CTLA-4 expression (56).

Thus, CTLA-4 expression in BC may be a potential prognostic biomarker. However, whether these results can be applied to TNBC is worthy of further investigation. The potential predictive value of CTLA-4 for efficacy of ICIs in TNBC has not been clarified. Relevant research should be carried out in the future to explore the potential prognostic and predictive value of CTLA-4 in TNBC.

TILs are heterogeneous lymphocyte groups that exist in tumor nests and interstitial cells. They are dominated by different degrees of monocyte and lymphocyte infiltration. The percentage of TILs is higher in TNBC than in luminal type and HER2-enriched BC (46, 90). Some studies have reported that the quantity of TILs in TNBC has predictive value for efficacy of ICIs [ Table 3 (27, 31, 32, 35)]. In the KEYNOTE-086 study, Sherene et al. found that high ORR for mTNBC patients treated with pembrolizumab was associated with high level of TILs (27). Similar findings were reported in the KEYNOTE-173 study, where a high level of TILs was significantly related to better pCR or ORR for TNBC patients treated with pembrolizumab (44). Loi et al. found that stromal TILs ≥5% predicted the response to pembrolizumab monotherapy (45). The biomarker analyses of the GeparNuevo trial showed that higher level of stromal TILs was associated with pCR in the overall cohort but did not predict the efficacy of durvalumab (28). The increased level of intratumoral TILs from before to after treatment was predictive for pCR specifically in the durvalumab arm (28). An increase in TILs in early TNBC patients after neoadjuvant therapy was associated with improved DFS and OS (46, 47). A phase III trial reported an approximately 15% reduction in death and recurrence for every 10% increase in TILs (47). At present, several studies have demonstrated the predictive value of TILs, but there is a lack of high-quality evidence. Therefore, the predictive value of TILs for efficacy of ICIs in TNBC remains contentious.

Most of the above studies have focused on the predictive value of the level of TILs rather than TIL subsets for efficacy of ICIs in TNBC (27, 44–47). TIL subsets with different immune cell compositions represent different immune responses and prognosis (48, 91). On the one hand, TIL subsets can predict the efficacy of ICIs for TNBC. For example, mTNBC patients who received atezolizumab as monotherapy with intratumoral CD8⁺ T cells >1.35% prior to treatment presented trends toward higher ORR and longer OS (92). An exploratory analysis of the IMpassion 130 study reported that the percentage of CD8⁺ T cells (≥0.5%) was predictive for the efficacy of atezolizumab plus nab-paclitaxel in mTNBC (24). Similarly, Jiang et al. found that a high CD8 immunohistochemical score was associated with better efficacy of immunotherapy in the IM subtype of TNBC (93, 94). On the other hand, TIL subsets might be associated with worse prognosis in BC (50, 95). For example, a high enrichment score of immature DCs and eosinophils is associated with poor OS (95). Additionally, lymphocytes with positive expression of fork head box protein 3 in tumor tissues are significantly associated with poor prognosis in BC (50).

Taken together, the potential predictive value of TILs in TNBC needs further exploration, and TILs may have potential prognostic value in TNBC. The 17th St Gallen International Breast Cancer Consensus Conference has shown that TILs may serve as a prognostic biomarker in TNBC (96). The following factors may affect the predictive effect of TILs. First, the evaluation of TILs is still mainly dependent on pathologists, who may obtain different results. Because the composition of TILs is complex, it is a challenge for researchers to distinguish the functions of different cells and their predictive values. Restrictions among TILs and the influence of cytokines on their functions can also influence their predictive value. Fortunately, testing standards have been formed for TILs, and artificial intelligence has gradually been applied to case interpretation (97, 98). Therefore, TILs can be detected more objectively. In the future, the clinical application of TILs will have broad prospects in TNBC, but currently, we should not use TILs to select individual patients for ICIs in clinical practice.

Myelopoiesis is a tightly regulated process that is altered in cancer, leading to the expansion of immature myeloid cells, now called MDSCs (99). Tumor cells secrete interleukin (IL)-8 to recruit MDSCs into the tumor microenvironment; inhibit T-cell activation by consuming and limiting cysteine and other essential amino acids, such as cysteine, for T-cell activation; induce immunosuppression; and promote tumor progression (100–104). MDSCs can be divided into two major groups: polymorphonuclear and monocytic MDSCs (105). Some studies have shown that the subsets of MDSCs are associated with the efficacy of ICIs in NSCLC and melanoma (106–108). However, there is no evidence whether the MDSCs are related to the efficacy of ICIs in TNBC. Therefore, the predictive value of MDSCs for efficacy of ICIs in TNBC is not clear.

Other studies have shown that higher levels of MDSCs are associated with worse prognosis in patients with solid tumors such as advanced BC (57, 58, 109, 110). Furthermore, advanced BC patients with circulating MDSCs >3.17% at baseline had poorer median OS than patients with circulating MDSCs ≤3.17% (5.5 vs 19.32 months) (57). In support of the prognostic value of MDSCs, Bergenfelz et al. observed 54 patients with metastatic BC and found that higher MDSC count was associated with worse PFS and OS (58).

As mentioned above, MDSCs may have potential prognostic value in BC, although no similar study has focused on TNBC. The potential predictive value of MDSCs for efficacy of ICIs in BC has not yet been clarified. Some studies have reported that ICIs reduce the number of circulating MDSCs, which implies that ICIs might have an MDSC-inhibiting effect (111, 112). Therefore, the detection of circulating MDSCs may contribute to a better understanding of the predictive value of MDSCs for efficacy of ICIs in TNBC. In the future, MDSCs are worth further exploration, especially for the potential predictive value of ICIs and prognostic value in TNBC.