1. Mechanisms Underlying the Overexpression of ICPMLs on Tumor Cells

Overexpression of ICPMLs on tumor cells can be the result of different stimuli, either cell-autonomous stimuli or stimuli from the tumor microenvironment (TME). The mechanisms underlying the overexpression of ICPMLs on tumor cells have been most thoroughly investigated for PD-L1 and have been reviewed recently ^[1]. As regards tumor cell-autonomous stimuli, overexpression of PD-L1 can be the result of intrachromosomal or extrachromosomal events. Copy number alterations in chromosomal region 9p24.1 that encompasses the loci for PD-L1 and PD-L2, inversions, deletions, translocations, generation of chimeric fusion transcripts, and disruption or mutation of the 3′-untranslated region of the

PD-L1 gene are intrachromosomal events that can lead to PD-L1 overexpression ^[2][3][4]. Tumor cell-autonomous, extrachromosomal events are receptor-activating mutations or receptor overexpression ^[5], gain-of-function or loss-of-function mutations affecting intracellular signaling molecules ^[6][7], activation or overexpression of transcription factors (e.g., hypoxia-inducible factor-α, signal transducer and activator of transcription (STAT) 3, MYC) ^[8][9][10]. More recently, also epigenetic mechanisms have been reported to induce or contribute to the overexpression of tumor cell-associated PD-L1 ^[11][12]. Tumor cell-exogenous stimuli that can lead to the overexpression of PD-L1 are cytokines (e.g., interferon (IFN)-γ, tumor necrosis factor (TNF)-α) ^[13][14] and various other stimuli from the TME like hypoxia or pseudohypoxia ^[10][15], antitumor drugs (chemotherapeutics, targeted therapeutics) ^[16] or metabolites (e.g., lactate) ^[17]. While the mechanisms leading to the overexpression of other tumor cell-associated ICPMLs have been much less investigated, they appear to be similar to those for PD-L1. Thus, hypoxia or pseudohypoxia lead to the overexpression of B7-H4 ^[18], CD70 ^[19], CD47 ^[20]. Antitumor drugs lead to the overexpression of CD70 and B7-H3 ^[21][22]. Activation of the Ras-Raf-MEK-extracellular signal-regulated kinase pathway leads to overexpression of CD155 and CD137 ^[23][24], Hedgehog signaling to overexpression of CD155 ^[24], the Janus kinase 2-STAT3 pathway to overexpression of fibrinogen-like protein 1 (FGL1) ^[25]. While the stimuli that induce overexpression of ICPMLs on tumor cells appear to be similar, in some instances subtle differences in the intracellular signaling pathways regulating the expression of two different ICPMLs have been observed ^[26], suggesting that these differences may explain the different patterns of expression that have been observed between different tumor cell-associated ICPMLs (see

Section 3

2. The Consequences of the Expression of ICPMLs on the Biology of Tumor Cells

In addition to transmitting signals to other cells (mostly immune cells) upon engagement of their cognate receptors ^{[27][28][29][30]} tumor cell-associated ICPMLs also exert cell-autonomous functions. Thus, their expression is associated with changes whereby tumor cells acquire enhanced capabilities to migrate, invade and metastasize to distant organs, undergo faster growth and metabolic alterations ^[31][32], acquire tumor-initiating potential ^[20][21][33] as well as resistance to antitumor drugs and apoptosis ^[1]. Collectively, these changes, when they are accompanied by the expression of specific transcription factors and molecular modifications ^[34] are referred to as tumor cell epithelial-mesenchymal transition (EMT) ^[35][36]. Indeed, the causal relationship between ICPML expression on tumor cells and EMT has been shown in several instances with a variety of technical approaches (e.g., siRNA, CRISPR/Cas) ^{[37][38][39][40][41]}. Expression of ICPMLs on tumor cells can be both a consequence ^[42], as well as a cause of tumor cell EMT ^{[37][38][43][44][45]}, suggesting the existence of a positive feedback loop between the expression of ICPMLs and EMT ^[1]. Interestingly, tumor cell EMT can also have immunosuppressive effects ^[46] and it has recently been shown that loss of the epithelial marker E-cadherin, a hallmark of EMT, reduces responsiveness to ICIs in a mouse melanoma model ^[47].

As regards individual ICPMLs, the following have been reported to be associated with tumor cell EMT: PD-1 ^[48][49] PD-L1 ^{[48][49][50][51]}, PD-L2 ^[48][49], B7-H3 ^[49], B7-H4 ^[37][52], CTLA-4 ^[48][49], OX40 ^[48], CD47 ^[53][54][55], CD137 ligand ^[56], CD155 ^[38], FGL1 ^[40], T-cell immunoglobulin and mucin-domain containing-3 (TIM-3) ^[49][57] and B- and T-lymphocyte attenuator (BTLA, CD272) ^[49]. Other ICPMLs, while not having been formally associated with EMT (e.g., CD70, galectin-9), are expressed by tumor cells displaying EMT-related functionalities ^{[58][59][60][61]}.

The data discussed so far suggest the existence of a close association between expression of ICPMLs on tumor cells and EMT and raise the question as to whether this association is absolute. In fact, data show that the association of an ICPML (PD-L1) and EMT on tumor cells is not coincident ^[33]. Moreover, as already mentioned, tumor cell-associated PD-L1 expression can be induced by intrachromosomal events. In these cases, PD-L1 overexpression is independent of tumor cell EMT ^[2][4][62], but it cannot be excluded that it may contribute to the induction of tumor cell EMT. The observation that genomic amplification targeting

PD-L1

PD-L2 is enriched in triple-negative breast cancer (TNBC), a cancer type with a predominantly mesenchymal phenotype suggests that this may, indeed, be the case ^[63].

Moreover, the lack of coincidence between ICPML expression and tumor cell EMT may also be the consequence of EMT plasticity, whereby tumor cells undergoing EMT cover a whole spectrum of phenotypes spanning from a fully epithelial to a fully mesenchymal one ^[64]. This suggests the possibility, for example, that an individual ICPML on tumor cells may be expressed at EMT initiation, when epithelial markers still predominate over mesenchymal markers. Moreover, heterogeneity of EMT marker expression is paralleled by the heterogeneity of ICPML expression ^[65]. Such heterogeneity applies both to individual ICPMLs, with ICPML-negative and -positive tumor cells coexisting within the same tumor ^[66], as well as to different ICPMLs showing non-overlapping or partially overlapping expression within the same tumor cell population. As regards the heterogeneous expression of different ICPMLs, it has been reported, for example, that a fraction of PD-L1-negative melanomas expressed high levels of CD155 and this was associated with a poor response to anti-PD-1/anti-CTLA4 therapy ^[67]. Moreover, expression of B7-H4 was prevalent among immune-cold TNBCs, and correlated inversely with that of PD-L1 ^[68][69]. In hepatocellular carcinoma tissues, FGL1 and PD-L1 had distinct distribution and relationships with each other ^[70]. Expression of Herpes virus entry mediator (HVEM) was found to be broader than that of PD-L1 on cells of melanoma metastases from 116 patients ^[71]. Moreover, in some situations, administration of an anti-ICPM antibody (anti-PD-1) has been shown to lead to the upregulation of an ICPML (TIM-3) ^[72].

3. Why Non-Depleting Antibodies Have Been Used against Inhibitory ICPMLs

In fact, ICI antibodies of IgG1 isotype are able, in addition to inhibit the interaction with the cognate ICPM, to induce cytotoxic or phagocytic effects (antibody-dependent cellular cytotoxicity (ADCC), antibody-dependent cellular phagocytosis (ADCP), complement-dependent cytotoxicity (CDC)) on cells expressing the targeted antigen. As to currently used antibodies, the anti-PD-L1 mAb atezolizumab has an aglycosylated Fc region devoid of effector functions, and the anti-PD-L1 mAb durvalumab is of IgG1 isotype with three mutations in the Fc domain resulting in greatly reduced ADCC and CDC ^[73]. A notable exception to this picture is the anti-PD-L1 mAb avelumab, which will be discussed later. The reason as to why several clinically approved ICIs have been selected to be devoid of effector functions is due to the fact that ICPMLs can be expressed not only on tumor cells, but also on cells of the innate and adaptive immune system and that their depletion through ADCC, CDC or ADCP might lead to undesired immunosuppressive effects. In fact, taking a closer look to ICPML-expressing immune cells, one may reach the conclusion that their depletion may not be necessarily of harm, because in many instances such cells have immunosuppressive effects. In the following, we will briefly discuss this knowledge which has been obtained mainly for PD-L1.

Tumor-associated dendritic cells (DCs) upregulate PD-L1 in response to T-cell derived inflammatory cytokines like IFN-γ ^[74], while M1 macrophages do so in response to another inflammatory cytokine, interleukin (IL)-1β ^[75]. PD-L1

⁺ DCs can lead to functional inactivation of T cells upon interaction with PD-1 ^[76]. Similarly, other PD-L1

⁺ antigen-presenting cells like macrophages can induce anergy in T cells upon interaction with PD-1 ^[77], explaining why expression of PD-L1 on immune cells, rather than tumor cells, has been found in some studies to correlate with a favorable response to anti-PD-1 therapy ^[78]. Additionally, B7-H4 is expressed on immunosuppressive tumor-associated macrophages (TAM) ^[79]. Moreover, upon PD-1/PD-L1 interaction, macrophages can produce increased levels of immunosuppressive cytokines like IL-10, but reduced levels of inflammatory cytokines like IL-6 ^[77][80]. Additionally, tumor-infiltrating T cells can express PD-L1 upon activation and this PD-L1 is important for T-cell survival ^[81]. Ligation of T cell-associated PD-L1 can have immunosuppressive effects by promoting M2 polarization of macrophages, reducing the production of inflammatory cytokines and inducing an anergic phenotype or apoptosis in T-cells ^[82][83]. PD-L1 expression has also been documented on non-tumor cells of the TME that may play tumor-promoting and immunosuppressive roles like cancer-associated fibroblasts (CAF) ^[84]. Eventually, mice lacking CD155 on both tumor-infiltrating myeloid cells as well as tumor cells showed greater reduction of tumor growth and metastasis compared to mice lacking CD155 only on tumor cells ^[85]. Importantly, the immunosuppressive effects of ICPMLs may be context-dependent as has been shown for PD-L1, with tumor-associated PD-L1 playing a predominantly immunosuppressive role in some tumor types, and PD-L1 expressed on tumor-associated immune cells playing a predominantly immunosuppressive role in other tumor types ^[86][87]).

So far, we have listed several downsides related to ICPML expression on immune cells. There are, however, some observations suggesting that the expression of PD-L1 on immune cells may contribute to antitumor effects of the immune response. Thus, some tumors were shown to induce expression of PD-L1 on natural killer (NK) cells and this led to enhanced NK-cell function. These PD-L1-positive NK cells could be activated with an anti-PD-L1 antibody to perform increased antitumor activity ^[88]. Depletion of PD-L1-expressing NK cells led to the suppression of this antitumor mechanism. Still another possibility to be considered is that depletion of ICPML

⁺ immunosuppressive cells triggers direct tumor-promoting effects of the immune system like those that may occur during hyperprogressive disease observed during ICI therapy ^[89][90][91].