Beyond PD-1/PD-L1 Inhibition: What the Future Holds for Breast Cancer Immunotherapy

Cancer immunotherapy has altered the management of human malignancies, improving outcomes in an expanding list of diseases. Breast cancer - presumably due to its perceived low immunogenicity - is a late addition to this list. Furthermore, most of the focus has been on the triple negative subtype because of its higher tumor mutational load and lymphocyte-enriched stroma, although emerging data show promise on the other breast cancer subtypes as well. To this point the clinical use of immunotherapy is limited to the inhibition of two immune checkpoints, Programmed Cell Death Protein 1 (PD-1) and Cytotoxic T-lymphocyte-associated Protein 4 (CTLA-4). Consistent with the complexity of the regulation of the tumor – host interactions and their lack of reliance on a single regulatory pathway, combinatory approaches have shown improved efficacy albeit at the cost of increased toxicity. Beyond those two checkpoints though, a large number of co-stimulatory or co-inhibitory molecules play major roles on tumor evasion from immunosurveillance. These molecules likely represent future targets of immunotherapy provided that the promise shown in early data is translated into improved patient survival in randomized trials. The biological role, prognostic and predictive implications regarding breast cancer and early clinical efforts on exploiting these immune-related therapeutic targets are herein reviewed.


Introduction
The recognition of the importance of the tumor -host interactions in the prognosis of cancer patients significantly predates the current era of cancer immunotherapy. The gradual deciphering of these complex interactions is summarized in the conceptual framework laid out by Hanahan and Weinberg [1], where immunoediting is suggested as a driving force guiding tumor progression. Exploiting these advances only in part, cancer treatment by the inhibition of negative regulators has revolutionized the management of multiple human malignancies, culminating with the award of the 2018 Nobel Prize in Physiology or Medicine -the first ever bestowed upon research related to an anticancer therapy [2].
Beyond its utility as a treatment target, immune response to cancer has also been a subject of research concerning its role as both a prognostic and predictive biomarker. As an example, higher tumor-infiltrating lymphocyte (TIL) counts and expression of immune function genes have been shown to predict better outcomes in most breast cancer (BC) subtypes and increased rates of pathologic complete remission (pCR) following the administration of neoadjuvant chemotherapy (NACT) for early BC (EBC) [3,4]. In metastatic BC (MBC), TIL enumeration has not proven to be as successful [5], since TIL counts have been shown to be lower in metastatic sites compared to the primary tumor [6].

LAG-3
Lymphocyte activation gene-3 (LAG-3) is a cluster of differentiation 4 (CD4) related negative regulator of immune response considered as a marker of T-cell exhaustion. It is expressed on both effector and regulatory T-cells, Natural Killer (NK)-cells, B-cells and dendritic cells (DC) [12][13][14][15][16]. Identified LAG-3 ligands are MHC (Major Histocompatibility Complex) class II molecules expressed on antigen presenting cells (APC), LSECTin and Galectin-3 [17,18]. LAG-3 is thought to inhibit the activity and expansion of effector T-cells and enhances the suppressive activity of T-regulatory lymphocytes (Tregs) [19][20][21][22]. Interplay between tumor cells and immune system components in the tumor microenvironment. Abbreviations for represented cells and immune-related markers are explained in the main text.
Published data on the role of LAG-3 in BC indicate that it is overexpressed in the tumor compared to the adjacent healthy breast tissue [23][24][25], while its overexpression has been associated with improved patient outcomes [26] (Table 1). Following promising pre-clinical results, LAG-3 inhibitors are currently being tested in early phase clinical trials including BC, as monotherapy or in combination with chemotherapy or anti-PD-1 therapy (Table 2). One phase I/II clinical trial testing IMP321 (Eftilagimod), a recombinant soluble LAG-3 Ig (Immunoglobulin) fusion protein, in combination with weekly paclitaxel as a first line treatment in 30 patients with MBC showed promising results, with a response rate of 50% [27].
TIM-3 seems to be upregulated both in BC samples compared to normal adjacent tissue and circulating lymphocytes, possibly through hypomethylation of its promoter [23,29] (Table 1). However, expression on immune cells has been reported to vary widely [29,100]. Burugu et al. evaluated TIM-3 IHC expression in 3992 BC samples of all subtypes and found that the TIM-3 intraepithelial TIL infiltration is associated with a better outcome [32]. TIM-3 polymorphisms might also play a role in the susceptibility to, and prognosis of BC [101][102][103].
Drugs targeting TIM-3 are currently being tested in early phase clinical trials including BC, alone or in combination with anti-PD1/PD-L1 check point inhibitors, with no published results yet (Table 2).

VISTA
V-domain Ig suppressor of T cell activation (VISTA) is a negative regulator of the T-cell immune activity functioning both as a ligand and receptor [104]. It has been shown to be expressed by CD4+ and CD8+ T-cells, Tregs, DC, NK-cells, monocytes, macrophages and granulocytes [105,106], as well as tumor cells [107][108][109]. VISTA exerts its immunosuppressive function by decreasing the T-cell production of effector cytokines, diminishing T-cell proliferation and increasing conversion to Tregs [106]. To our knowledge, VISTA's expression and prognostic impact in BC has never been assessed, although a phase 1 clinical trial which enrolls TNBC patients and tests an oral inhibitor of PD-L1, PD-L2 and VISTA is currently ongoing (Table 2).

TIGIT
T-cell immunoreceptor with Ig and ITIM domains (TIGIT) is a co-inhibitory molecule expressed on effector, memory and regulatory T-cells, follicular helper (Tfh) and NK-cells [110,111]. It competes with CD223 to bind its two identified ligands, CD155 and CD112, expressed on APC, fibroblasts, endothelial, epithelial cells and also on a variety of cancer cells, including BC [112]. TIGIT has different ways of exerting its immunosuppressive action: Direct inhibition of NK-cell function [113], direct inhibition of T-cell activation, proliferation and cytotoxicity by attenuating TCR-driven (T-cell receptor) activation signals [114] and indirect inhibition of T-cells by promoting the maturation of immunoregulatory DCs [111]. It also promotes the Tregs function by being a direct target to FoxP3 (Forkhead box P3) and inducing an enhanced suppressive function [115,116].
TIGIT expression in BC has only been assessed at the transcriptomic level, with most studies showing overexpression [23,31,33,117] (Table 1). In one study, overexpression was correlated with improved patient survival in TNBC [33], leading to the development of antibodies targeting TIGIT in combination with PD-1 blockade (Table 2).

GITR
Glucocorticoid-induced TNFR-related protein (GITR) is a co-stimulatory member of the tumor necrosis factor (TNF) receptor superfamily expressed constitutively on all T-cells [118,119]. It is also expressed on NK-cells, eosinophils, basophils, macrophages and B-cells [120]. Its activating ligand is the GITR ligand (GITRL), expressed on APC and endothelial cells [121,122]. Upon binding, GITR exerts an immunostimulatory activity by directly enhancing T-cell proliferation and effector functions [123,124]. It also indirectly enhances the effector T-cell function by decreasing the intratumoral Treg numbers and suppressive function [125,126]. By avoiding activation-induced cell death, it also promotes an increase in memory T-cells [127].
Cari et al. assessed GITR mRNA expression in 3169 BC patients of all subtypes and found an overexpression in 42% of the cases [31]. Other studies demonstrated that expression is increased in both infiltrating [34] and circulating Tregs of BC patients [35,37]. Interestingly, GITR seems to also be overexpressed in CD4+ T-cells in BC-infiltrated lymph nodes [36] (Table 1).
BMS-986156, a GITR agonistic monoclonal antibody, in combination with nivolumab has demonstrated an acceptable safety profile and promising antitumor activity in advanced solid tumors [82]. Other agonist molecules targeting GITR are currently being tested in early phase clinical trials (Table 2).

B7-H3
B7 homolog 3 (B7-H3) is a member of the B7 family of immunomodulatory ligands. It is not spontaneously expressed in peripheral blood mononuclear cells but can be induced upon stimulation in APC, T-cells and NK-cells [128]. It is widely expressed in healthy solid organs and several malignancies, including BC [129]. Interestingly, it is also expressed by tumor-associated endothelial cells [45]. Although B7-H3 was initially seen as a co-stimulatory molecule, which increases CD4+ and CD8+ proliferation and enhances T cell cytotoxicity [129,130], the majority of recent studies highlight its co-inhibitory role. Indeed, it appears to downregulate T-cell proliferation and cytokine production [131], Th1 and Th2-mediated immune reactions [132] and inhibit NK cells activity [133]. Moreover, B7-H3 seems to influence cancer progression beyond its immunoregulatory role, by promoting migration, invasion and angiogenesis [134,135].
B7-H3 expression in BC has been extensively studied and demonstrated to confer worse prognosis [41,42] (Table 1). As a result, two antagonist drugs -a monoclonal antibody (enoblituzumab) and a dual-affinity re-targeting (DART ® ) protein (MGD009) -are currently under evaluation in early phase clinical trials including BC (Table 2).

ICOS
Inducible T cell co-stimulator (ICOS) is a specific T-cell molecule of the B7-binding CD28 family, expressed on activated T-cells after TCR engagement and enhanced by CD28 co-stimulation [136,137]. Its only ligand is ICOS-L, mainly expressed on APC [138][139][140] but also on endothelial and lung epithelial cells [141,142]. Although typically seen as an immune co-stimulatory pathway, notably through promoting cell proliferation/differentiation, enhancing Th1/Th2 function and facilitating T-dependent B-cell activation [136,137,143], ICOS/ICOS-L interaction might also have an immunosuppressive role through the accumulation of Tregs and secretion of IL-10 [46,144].
In a study by Faget et al., BC patients overexpressing ICOS had a significantly worse survival in the univariate but not multivariate analysis [46], while certain ICOS gene polymorphisms have also been associated with increased BC susceptibility in Chinese populations [145,146] (Table 1). Ongoing trials of agents targeting ICOS are shown in Table 2.
Two studies using gene-expression datasets demonstrated that 4-1BB is overexpressed in BC and is associated with better prognosis [31,47] (Table 1).
Monoclonal agonist antibodies are currently being tested in early phase clinical trials including BC (Table 2). Two early-phase studies (NCT00351325 and NCT00309023) raised concerns due to two hepatotoxicity-related deaths, though not replicated in a follow-up phase 1 study [158].

CD27 and CD70
CD27 and its only ligand CD70, are members of the TNF receptor and ligand superfamily that interact exclusively with each other. CD27 expression on T-cells is tightly regulated, with upregulation upon activation after the TCR stimulation followed by downregulation once the effector T-cell differentiation is acquired [159]. CD27 is also expressed on B-cells (germinal center and memory B-cells) and NK-cells [160][161][162]. CD70 expression on immune cells is also tightly regulated and is present on activated T-cells, stimulated B-cells, mature DC and NK-cells [163][164][165][166]. Interestingly, CD70 has also been found to be expressed in various hematological, sarcoma and carcinoma cells including BC [167]. The CD27-CD70 pathway exerts its co-stimulatory activity in great part through CD27 interaction with TNF receptor associated factors (TRAF), resulting in the activation of transcription factors of MAPK (Mitogen-activated Protein Kinase) and NFκB (Nuclear Factor kappa-light-chain-enhancer of activated B-cells) family. This leads to the expansion and survival of activated T cells [168][169][170][171][172][173]; differentiation to memory and effector T-cells [173][174][175]; activation of NK-cells [176,177]; and differentiation plus activation of B-cells [178][179][180].
CD70 protein expression in BC was assessed in two studies with contrasting results [49,50] (Table 1). Of interest, Liu et al. demonstrated that a high CD70 expression was correlated with worse lung metastasis-free survival, but not with other metastatic sites following relapse of EBC. In addition, gene expression studies showed that CD70 was overexpressed in basal-like compared to Luminal A cancers and that overexpression after NACT was associated with a better outcome [51,181].
Two antibodies, ARGX-110 targeting CD70 and CDX-1127 (Varlilumab) targeting CD27 are currently in early phase clinical trials. In addition, a trial is testing the safety and activity of administering peripheral blood lymphocytes transduced with a CD70-binding Chimeric Antigen Receptor (CAR) to patients with CD70-expressing cancers (Table 2).
Several studies have assessed OX40 expression in BC, showing an expression varying from 15.5% to 85% of cases (Table 1). Interestingly, Xie et al. reported expression on cancer cells while all the other studies reported expression on TILs [52]. Consequently, a number of agonistic monoclonal antibodies targeting OX40 and a mRNA encoding OX40L (injected intra-tumorally) are currently being tested in early phase clinical trials including BC, alone or in combination with other immunotherapies. (Table 2) 2.11. BTLA [209][210][211][212]. BTLA has also been described as a potential receptor for B7-H4 in BC [213]. BTLA exerts its T-cell inhibitory action upon binding HVEM, leading to a decreased T-cell proliferation and cytokine production with a predominant effect on CD4+ cells [214][215][216][217][218][219]. Data concerning its action on B-cell function is scarce but it appears to negatively regulate B-cell activation [220]. Interestingly, BTLA and PD-1 seem to be co-expressed on CD8+ T-cells.

BTLA (B and T Lymphocyte Attenuator) is an inhibitory Ig-domain-containing glycoprotein receptor of the CD28 superfamily expressed on activated T-cells, B-cells, Tfh cells, macrophages, DC, NKT-cells and NK-cells [204-208]. Its only proven ligand is HVEM (Herpes Virus Enter Mediator), a member of the TNF receptor family, expressed on CD4+ and CD8+ T-cells (strongly on resting T cells, downregulated upon activation), naïve and memory but not activated B-cells, monocytes, DC, solid organs, tumor-associated endothelial cells or on various cancer cells including BC
Data concerning BTLA expression in BC is scarce (Table 1). Although it seems to be overexpressed at the transcriptomic level, especially in TNBC where it was also associated with improved survival [57], protein expression appeared to be limited in another study [58]. To our knowledge, no clinical trials for therapeutic targeting of BTLA are currently ongoing.

TLR9
Toll-like receptors (TLRs) are type I transmembrane glycoproteins of the pattern recognition receptors (PRR). They play a key role in immunity by allowing immune cells to recognize non-self or altered-self molecular patterns, activating the innate immune response and coordinating the innate and adaptive immune responses. The most studied member in BC is the intracellular receptor TLR9.
TLR9 is a DNA receptor that migrates from the endoplasmic reticulum to the endosomal/lysosomal compartment when DNA enters the cell [221,222]. When activated by DNA recognition, TLR9 initiates a signaling cascade [222,223], leading to the activation of various transcription factors like NF-κB and AP-1 (Activator protein 1) [224], thus promoting the transcription of genes that are important for inflammatory and immune responses [225,226]. In addition, it promotes adaptive immunity by enhancing DC maturation and producing a favorable cytokine/chemokine milieu that results in the activation of Th1 and CD8 cytotoxic T lymphocytes as well as by promoting B-cell proliferation [227,228].
TLR9 expression and its prognostic role in BC has been reported by several studies with conflicting results [60,64] (Table 1). Nevertheless, it appears that TLR9 is expressed at higher levels in estrogen receptor (ER) negative and high-grade tumors. Regarding the prognostic significance of TLR9 expression, three studies associated high expression with a better outcome [60,61,64], while two other studies reported worse survival [63,66]. Of interest, Karki et al. demonstrated that BC patients have decreased serum levels of TLR9 compared to patients with benign lesions and healthy controls, proposing it as a potential diagnostic biomarker [229]. Moreover, several but not all studies have shown an association between TLR9 gene polymorphisms and BC susceptibility [230][231][232][233].
Therapeutic targeting of TLR9 has proven to be efficient in pre-clinical models of various cancers including BC and many drugs are currently being tested in several cancer types, some of them even reaching phase III (NCT03445533) ( Table 2).

The Adenosine Pathway in Breast Cancer
The adenosine pathway is an important peripheral control mechanism for regulating the immune response in order to prevent over-activation and tissue damage. As with other immunoregulatory pathways, cancer cells are capable of hijacking it in order to promote tumor escape. Important components of this pathway are the adenosine receptor A2a (A2aR), through which the extracellular adenosine can activate its intracellular signaling pathway and the ectonucleotidases CD39 and CD73, which participate in extracellular adenosine production by dephosphorylating ATP.
A2aR is a G-protein-coupled receptor expressed on T and NKT-cells, B-cells, monocytes, macrophages, DC, NK-cells, mast cells, eosinophils and platelets [234]. CD73 is a cell-surface enzyme that can also be found as an enzymatically active soluble form. It is widely expressed on immune cells including B-cells, CD8+ and CD4+ T-cells, Tregs, neutrophils, MDSC, monocytes, macrophages, DC and NK-cells [235]. It is also expressed on a wide range of epithelial cells, endothelial cells and cancer cells including BC [235][236][237]. CD39, another cell-surface enzyme which produces adenosine, is also expressed on a variety of immune cells [238][239][240]. It is also expressed on platelets, endothelial cells and cancer cells including lung, melanoma, pancreatic and lymphoma cells [241][242][243]. Like CD73, a soluble catalytically active form of CD39 exists [244] The adenosine pathway exerts an immunosuppressive action by inhibiting effector T-cell activation [245], proliferation, cytokine production and cytotoxicity as well as promoting their immunosuppressive cytokine production [246,247]. In addition, it promotes Tregs formation [246], inhibits NK-cell antitumor activity [248], NKT-cell production of cytokines [249], macrophage proliferation [250] and DC maturation [251]. It has also been shown to increase the expression of other immune checkpoints [252].
CD73 expression on BC cells ranges from 9 to 84% of the cases and is generally associated with worse outcome, although one study reported contrasting results [68] (Table 1). In addition, CD39 is overexpressed both in TILs and circulating T cells of BC patients when compared to healthy controls, but its prognostic value has not been studied.
Numerous pre-clinical studies have demonstrated the efficacy of targeting the adenosine pathway in BC models, leading to the development of A2aR oral inhibitors and antibodies targeting CD73, currently in early phase clinical trials (Table 2). CD39 targeting therapies are currently under pre-clinical development but to our knowledge none have yet reached clinical trials.

Tumor-Associated Macrophages and Related Markers
Tumor-associated macrophages (TAMs) represent a major and heterogeneous distinct immune cell subpopulation in the tumor microenvironment (TME). In many tumor types, including BC, TAMs play a key role in tumor progression, angiogenesis, immune evasion and metastasis [253]. They also interact with other cell types through the secretion of various cytokines which in turn can modify the balance between tumor, stromal, endothelial and immune cells. According to the markers expressed on their cell surface as well as the factors they secrete, TAMs can be divided into two subtypes: a) the classically activated M1-like macrophages which have pro-inflammatory, anti-tumoral properties mainly through the secretion of TNF-a (Tissue Necrosis Factor alpha), IL-1, IL-2, IL-6, IL-12; and b) the selectively activated M2-like macrophages with anti-inflammatory, pro-tumoral phenotype mainly through TGF-β (Transforming growth factor beta), IL-4, IL-10 and IL-13 [254]. In terms of prognosis, TAMs were associated with worse overall survival in many solid tumors according to a large meta-analysis [255]. In BC in particular, a meta-analysis of sixteen studies revealed that a high TAM density was associated with worse overall survival (Hazard Ratio [HR]=1.50; 95% Confidence Intervals [CI] 1.20-1.88) and disease-free survival (HR=2.22; 95% CI 1.71-2.89) [256]. Overall, therapeutic strategies against TAMs are based on two major approaches: a) targeting TAM recruitment and activation, and b) reprogramming macrophage polarization towards an anti-tumoral phenotype. The first approach includes the elimination of TAM and monocyte accrual to the tumor site through the inhibition of mainly CSF-1/CSF-1R (Colony Stimulating Factor 1/ Colony Stimulating Factor 1 Receptor) and CCL2/CCR2 (C-C Motif Chemokine Ligand 2/ C-C Motif Chemokine Receptor 2) signaling axes. The second approach relies on the fact that TAMs are mostly of the M2-like phenotype and thus, stimulating the properties of the M1-like phenotype could be an effective treatment option to restore anti-tumoral activity. Such potential treatments for the macrophage polarization shift include CD40-agonists and/or TLR7 agonists. Whether the aforementioned therapeutic agents can be combined with other therapies which can target angiogenesis, increase phagocytic activity or enhance anti-tumor immunity is currently under investigation [257,258]. Moreover, recognition and targeting of other pro-tumoral chemokines and cytokines [259] or novel targets could broaden the therapeutic spectrum in cancer immunotherapy.

CSF-1/CSF-1R
TAM recruitment is highly controlled by the interaction of the glycoprotein CSF-1 with its receptor CSF-1R, a member of type III receptor tyrosine kinase family. Binding of CSF-1 to CSF-1R leads to activation, recruitment and proliferation of TAMs [260]. CSF-1R is normally expressed in various cell types but its expression in BC cells has been correlated to worse prognosis [261][262][263][264] (Table 3). Therapeutic targeting of this axis is under active investigation (Table 4).

CCR2/CCL2
The recruitment of circulating monocytes from the bone marrow into the TME is also mediated by the expression of the chemokine ligand CCL2. The binding to its receptor CCR2 leads to the differentiation of monocytes into TAMs and to the subsequent promotion of their pro-tumoral activity, tumor cell proliferation, angiogenesis and metastatic dissemination [265,266]. Expression of these chemo-attractants has been linked to worse prognosis in BC patients [267][268][269][270][271] (Table 3). Targeting this axis using CCR2 antagonists and anti-CCL2 antibodies is currently being explored in advanced solid malignancies, including BC (Table 4).

CD47 and SIRPa
Interaction between the two cell-surface immunoglobulin family members, CD47 and signal regulatory protein alpha (SIRPα), is crucial for the regulation of phagocytosis. CD47 is expressed on cancer cells while SIRPα is expressed on macrophages. Upon interaction, the anti-tumor immunity is diminished as CD47 represents a 'don't eat me' signal, thus impairing phagocytosis [296,297]. Through targeting this checkpoint axis using anti-CD47 antibodies, CD47-Fc and/or SIRPα-Fc fusion proteins, the macrophage phagocytic capacity can be restored (antibody-dependent cellular phagocytosis, ADCP) towards an effective immune response. The first reported efficacy results of the Hu5F9-G4 inhibitor combined with rituximab in non-Hodgkin's lymphoma are promising [298]. Possible synergistic effects of such treatments with anti-HER2 or anti-PD-L1/PD-1 antibodies are being tested in clinical trials (Table 4).

TLR7
TLR7 represents an intracellular receptor, member of the toll-like receptors transmembrane glycoprotein family. Its expression can enhance the DC function and can re-programme macrophages towards an anti-tumoral M1 phenotype [299,300]. Therefore, its activation using TLR7 agonists could provide effective anti-tumor responses. Indeed, the use of the topical TLR7-agonist imiquimod in combination with nab-paclitaxel led to the short-term regression of BC cutaneous metastases in early phase trials [301,302] (Table 4).

CD40
CD40 represents a co-stimulatory protein, member of the TNF receptor family and is an emerging target in cancer immunotherapy. CD40 is mostly expressed by APC and macrophages and binding of its ligand (CD40L) on T-cells results in T-cell activation [303]. Preclinical data of the CD40-agonist efficacy have been reported in BC and other tumor types, demonstrating the promotion of T-cell responses [304,305]. CD40 activation using agonistic monoclonal antibodies can also lead to the enhancement of macrophage tumoricidal and pro-inflammatory properties mainly through MHC-II upregulation [303]. Preliminary results indicate activity and durable immune responses [306] (Table 4).

Killer Immunoglobin Receptors (KIR)
NK-cells represent an immune cell subpopulation with an active role in effective antitumor immunity [307]. MHC class I specific Killer Immunoglobin Receptor (KIR) family members are mostly expressed on the surface of NK-cells. Some KIR -upon binding to their ligands HLA-B or HLA-C -can hinder NK cell activation [308], while others are associated with NK stimulatory properties and better prognosis for cancer patients [309,310]. Ongoing clinical trials are underway, testing antibodies against NK-inhibiting KIR family members in combination with other immune checkpoint inhibitors (Table 4).

CD94/NKG2A
NK group member 2A (NKG2A) represents a novel inhibitory receptor, which forms heterodimers with CD94, both belonging to the C-type lectin-like family and expressed mainly on the surface of NK-cells and also on CD8+ T-cells. Upon binding of the complex to its MHC class I (HLA-E) ligand, the anti-tumoral capacity of NK-cells can be hindered and an immunosuppressive phenotype through T-cell inactivation is established [308,311]. Recently, two preclinical studies in colorectal and head and neck carcinoma demonstrated that blockade of this receptor may be a new appealing immunotherapeutic target [312,313]. Expression of NKG2A has been described in BC patients [273], however no studies on therapeutic targeting are ongoing (Table 3).

NK-Cell Activating Receptors
NK-cells are activated through various receptors such as the natural cytotoxicity receptor (NCR) family (NCR1 or NKp46, NCR2 or NKp44, NCR3 or NKp30) and NK group member 2D (NKG2D). The latter recognizes several ligands including MHC class I polypeptide-related sequence (MICA/MICB) and UL16-binding proteins (ULBP1-6) and their interaction leads to enhanced cytolysis [314,315]. Expression of NKG2D ligands has been associated with improved survival in BC [274,316,317] (Table 3).

IDO
Indoleamine 2,3 dioxygenase-1 (IDO1) is an enzyme mostly found in DC and an appealing target for cancer immunotherapy [318]. It plays an important role in metabolism-mediated immune regulation by catalyzing the conversion of amino acid tryptophan to kynurenine and thus impairing T-cell activation and promoting Treg expansion [319,320]. IDO expression in BC patients has been extensively studied, with varying positivity, from 14 to 100% of the cases [276,279]. Most of the studies describe a predominant expression by tumor cells with limited expression by stromal dendritic-like cells and occasional expression by myoepithelial cells. Although conflicting results have been reported, the majority of the studies show that the IDO expression is correlated to an advanced stage at diagnosis, high grade, ER negativity and worse outcome [277,278]. Recent findings from a phase I trial, indicate the activity and safety of targeting IDO in combination with anti-PD-L1 monoclonal antibody atezolizumab in various advanced solid tumors including BC [321].

Myeloid-Derived Suppressor Cells
MDSCs represent a heterogeneous population of immature myeloid cells including progenitor cells, immature DCs, macrophages and granulocytes. In humans, MDSCs are defined by the positive expression of CD33 and CD11b and negative or reduced expression of HLA-DR. MDSCs are further classified as monocytic or granulocytic MDSCs when CD14 or CD15 is expressed, respectively.
MDSCs play a major role in promoting an immunosuppressive microenvironnment through various mechanisms: Production of reactive oxygen and nitrogen species depleting TILs [322,323], impairment of lymphocyte-homing [324], promotion of other immunosuppressive cells such as Tregs and M2-macrophages [325,326], depletion of metabolites involved in the T cell function such as L-arginine and cysteine [327,328] PD-L1 expression [329] and adenosine production by upregulating the expression of ectonucleosidases CD39 and CD73 [330]. In addition to their immunosuppressive effect, MDSCs also promote tumor dissemination and metastasis by affecting epithelial-mesenchymal transition [331], degradation of extra-cellular matrix [332], stem cell formation [333], angiogenesis and formation of premetastatic niches [334,335].
Presence of MDSCs in BC patients has been studied both in peripheral blood and primary tumors. Patients with BC have elevated levels of circulating MDSCs compared to healthy donors or patients with benign lesions and the levels of MDSCs increase with tumor burden (i.e. clinical stage), making it a potential tool for BC diagnosis [336,337]. MDSCs are also present in the BC tumor microenvironment at significantly higher levels than the adjacent healthy breast tissue and one study found that TNBC seems to be more infiltrated than other BC subtypes [338][339][340]. Moreover, MDSCs represent a potential biomarker for predicting both survival and response to NACT, with higher levels of circulating of infiltrating MDSCs being associated with worse survival and pCR rates [340][341][342][343].
As a result, targeting MDSCs is a putative therapeutic tool for BC patients and different strategies have shown promising results in pre-clinical studies [344][345][346][347]. Briefly, current treatment strategies aim to modulate myelopoïesis by forcing differentiation into mature cells or inhibiting maturation from precursor cells, block MDSC accumulation in tumor sites and block MDSC immunosuppressive functions [348]. To our knowledge, only pre-clinical data of MDSC targeting in BC have been published but three early-phase clinical trials are currently ongoing (NCT03145012; NCT02922764; NCT02499328).

Implementing Combination Immunotherapy in the Clinic
Blockade of the PD-1/PD-L1 axis through the use of monoclonal antibodies as monotherapies has met with considerable success during the past decade. The central concept of immunotherapy with the inhibition of negative regulators of the immune response is the restoration of activity of exhausted cytotoxic T-lymphocytes. As evidenced by the observation of responses among patients lacking a local immune response (no PD-1/PD-L1 expression at the protein level, absence of TIL), a pre-existing immune response is not an absolute prerequisite needed for the elicitation of responses to treatment. Nevertheless, response rates and response duration following treatment with a monotherapy seem to be lower among those patients [349].
Intriguingly, the combined immune checkpoint blockade confers superior results compared to PD-1 blockade alone in this patient group. Data derived from the phase 3 CheckMate 067 trial indicate that double PD-1 and CTLA-4 blockade with nivolumab and ipilimumab improved both progression-free (HR=0.67; 95% CI were not reported) and overall survival (HR = 0.70) compared with nivolumab alone in patients with metastatic melanoma and PD-L1 expression lower than 1% [350]. Although this analysis is exploratory and the trial was not designed to perform this comparison, it provides support for immunotherapy combinations. The theoretical background seems intuitive. Mechanistically the two checkpoints function on different sites of immune activation: CTLA-4 carries out its function at the sites of priming whereas PD-1 is responsible for maintaining tolerance by dampening the activation of T-lymphocytes in the periphery [351]. It is unclear however whether the combinatory approach is successful thanks to an additive effect of the two inhibitors or if it results from the suppression of escape mechanisms. Similarly, it is conceivable that the inhibition of other negative regulators or agonistic activation of co-stimulatory molecules in combination with each other or with established immunotherapies can lead to further improvements in terms of patient outcomes. It is clear however that a mechanistic understanding of the biology of the candidate therapeutic targets and of the cross-talk that is activated upon inhibition is of paramount importance. Further underscoring the need for a deep understanding of the underlying biologic processes and the rational design of novel agents is the failure of the combination of the once promising IDO1 inhibitor epacadostat to improve outcomes in combination with pembrolizumab versus pembrolizumab alone in patients with metastatic melanoma [352].
While increased efficacy is the main goal, two barriers need to be overcome for successful integration of novel immunotherapies: Toxicity and financial cost. The clinical use of the checkpoint inhibition is associated with a risk for serious, potentially fatal immune-related adverse events (irAEs). Following this paradigm, the ability to inhibit multiple targets simultaneously may be limited by the adverse event profile of such combinations. It is important to note that while it is unclear whether the same molecular mechanisms that drive tumor rejection are to blame for the induction of irAEs, both retrospective [353] and limited prospective data [354] show a correlation between irAEs and better outcomes. This correlation has not been adequately studied if it also concerns combinatorial immunotherapy, which is associated with a higher risk for severe irAEs according to the aforementioned CheckMate 067 trial [350].
On the other hand, the revolution of cancer immunotherapy has brought to the limelight the associated financial costs. Published data indicate that the combination of nivolumab and ipilimumab, despite its efficacy, is not a cost-effective option [355]. How quickly and widely the combination will be adopted in light of the positive results from randomized trials on malignancies that can be readily treated with other options [356,357], remains to be seen. It is reasonable to assume that future combinations with novel agents will not differ in that respect. In addition, the evaluation of novel combinations will likely be plagued by the same problems that have affected PD-1/PD-L1 inhibitors: Inconclusive predictive biomarkers lacking analytical validity and clinical validity/utility, variety of companion diagnostics using different antibodies and cut-offs, trials reporting different results from different antibodies in the same clinical setting and overabundance of available options with no hints on their differential efficacy [7]. It is therefore imperative that future phase 3 trials will be based on robust preclinical and early clinical data.

Conclusions
A large number of co-stimulatory or co-inhibitory molecules regulating tumor evasion from immunosurveillance have been studied in BC ( Table 5). As reviewed here, there are solid pre-clinical data on the function of these factors and emerging data on their regulation and their role in the clinical setting. These molecules likely represent future targets of immunotherapy provided that the promise shown in early data is translated into improved patient survival in randomized trials. Table 5. Overview of immune-related markers' characteristics including origin of expression and their role in anti-tumor immunity.

Marker
Types of Cells Expressed Function on Anti-tumor Immunity While it seems counterintuitive that the development of the next generation of immunotherapy agents precedes the optimization of the currently available ones, early recognition of the most promising agents can hasten their implementation in clinical practice. As we previously characterized the emergence of the PD-1/PD-L1 inhibition as the "end of the beginning" of cancer immunotherapy [7], the exciting advances that are described in this review could very well represent the "beginning of the end" of non-selective cytotoxic chemotherapy.
Author Contributions: Conceptualization, I.Z., A.M. and T.F.; Literature review, S.C. and I.Z.; Writing-initial draft preparation, S.C. and I.Z.; Writing-review and editing, J.B., A.M. and T.F. All authors have read and approved the submitted version of the manuscript.