HLA-G/ILTs Targeted Solid Cancer Immunotherapy: Opportunities and Challenges

Immune checkpoint inhibitors (ICIs) have become a promising immunotherapy for cancers. Human leukocyte antigen-G (HLA-G), a neoantigen, its biological functions and clinical relevance have been extensively investigated in malignancies, and early clinical trials with “anti-HLA-G strategy” are being launched for advance solid cancer immunotherapy. The mechanism of HLA-G as a new ICI is that HLA-G can bind immune cell bearing inhibitory receptors, the immunoglobulin-like transcript (ILT)-2 and ILT-4. HLA-G/ILT-2/-4 (HLA-G/ILTs) signaling can drive comprehensive immune suppression, promote tumor growth and disease progression. Though clinical benefits could be expected with application of HLA-G antibodies to blockade the HLA-G/ILTs signaling in solid cancer immunotherapy, major challenges with the diversity of HLA-G isoforms, HLA-G/ILTs binding specificity, intra- and inter-tumor heterogeneity of HLA-G, lack of isoform-specific antibodies and validated assay protocols, which could dramatically affect the clinical efficacy. Clinical benefits of HLA-G-targeted solid cancer immunotherapy may be fluctuated or even premature unless major challenges are addressed.


INTRODUCTION
Immune checkpoint inhibitors have become a promising immunotherapy for cancers, but durable clinical benefits are limited for existing agents (1). It's an exciting news released in Cancer Discovery "Gilead Buys into Tizona's Anti-HLA-G Strategy" that an early clinical trial with human leukocyte antigen-G (HLA-G) inhibitor TTX-080 is being launched for advance solid cancer patients (NCT04485013) (2).
Engagement of HLA-G/ILTs can induce either unresponsive or tolerogenic state of a wide range of immune effector cells. The underlying mechanisms for HLA-G as an immune checkpoint is that HLA-G can either directly bind tyrosine-based inhibitory motifs (ITIMs) containing immune inhibitory receptors, the immunoglobulin-like transcript (ILT)-2/CD85j/LILRB1 and ILT-4/CD85d/LILRB2, which is expressed on various immune competent cells such as T lymphocytes, natural killer cells (NK), dendritic cells (DC) (62), or indirectly by intercellular transfer by the process of trogocytosis and exosomes to drive a comprehensive immune suppression (63). Immune suppression induced by the HLA-G/ILTs signaling pathway includes inhibition of cytotoxicity (64), inflammatory cytokine production (65), chemotaxis and proliferation of T cells and NK cells (66,67), inhibition antibody production of B cells and maturation of antigen presenting cells (68,69), and dampen the anti-tumor functions of invariant natural killer T (iNKT) cells and tumor-infiltrating CD8 + PD-1 − ILT-2 + T cells (70,71). Also, HLA-G/ILTs engagement can also induce expansion of myeloid derived suppressive cells (MDSCs) and generation of regulatory T cells (72,73). In addition to immune suppressive functions, HLA-G/ILTs can promote intratumor vascular remodeling by enhancing vascular endothelial growth factor-C (VEGF-C) expression, and increase tumor metastasis by inducing cancer promoting factor matrix metalloproteinases (MMPs) expression (74,75) (Figure 1). Consequently, both innate and adaptive antitumor immune responses are impaired, thus favoring tumor cell immune evasion and disease progression. In this scenario, restoring antitumor functions of ILT-bearing immune cells with HLA-G inhibitors such as TTX-080 sounds reasonable.
As HLA-G expression is specifically induced in most types of solid cancer cells, clinical benefits of HLA-G inhibitors could be expected for cancer immunotherapy. However, challenges such as multiple HLA-G isoforms with distinct extracellular domains, different binding sites between HLA-G and ILT-2/ILT-4 interaction, intratumor or intertumor heterogeneity of HLA-G expression, and isoform-specific antibody and validated assay protocol lacking, remain tremendous hurdle in terms of HLA-G/ ILTs antibody-based solid cancer immunotherapy.

HLA-G ISOFORMS MOLECULAR STRUCTURE
The HLA-G gene contains eight exons and seven introns. However, most full-length transcripts carry only seven exons because exon seven is usually spliced out. Due to a premature stop codon in E6, the HLA-G full-length protein has 338-amino acids, which is relatively shorter compared with classical HLA class I molecules. Among these exons, E1 generates the signal peptide, E2-E4 generate extracellular a1, a2, and a3 domains, respectively. E5 generates the transmembrane domain, and E6 generates the intracellular cytoplasmic tail of HLA-G (76).
Due to its primary transcript alternative splicing, diverse molecular structures of HLA-G have been observed. Seven HLA-G isoforms including four membrane-bound (HLA-G1-HLA-G4) and three soluble (HLA-G5-HLA-G7) monomers have been identified. With a premature stop codon in E6, membrane-bound (HLA-G1-HLA-G4) isoforms have a unique truncated cytoplasmic tail comparing to other classic HLA class I molecules. Soluble HLA-G5 and HLA-G6 isoforms are resulted from a stop codon in intron 4, and HLA-G7 are generated from a stop codon in intron 2, which prevents the translation of their transmembrane domain (77,78) (Figure 2).
Each HLA-G isoform has its unique extracellular structure. HLA-G1 is the only full-length isoform with extracellular a1, a2, and a3 domains; HLA-G2 has a1 and a3 domains; HLA-G3 has the only a1 domains; HLA-G4 has a1 and a2 domains. Similarly, HLA-G5 has the extracellular a1, a2, and a3 domains; HLA-G6 has a1 and a3 domains, and HLA-G7 has the only a1 domains. a1 and a2 domains form the peptide binding cleft, and a3 domain non-covalently bind to the light chain b 2 -microglobulin (b 2 m). Novel HLA-G isoforms such as lacking a transmembrane region and a1 domain have been predicted with RNAseq technology (79) (Figure 3). Moreover, higher molecular weight of HLA-G has been associated with post-translational modifications. Homo-and hetero-HLA-G dimers can be formed through intermolecular disulfide bonds with Cys 42 or Cys 147 in the extracellular a1 or a2 domain; others such as glycosylated, nitrated, and ubiquitinated HLA-G molecules have also been confirmed (80)(81)(82)(83).
An independent prognosis factor.
Associated with disease progression.
Low expression when no signal or discrete staining; high expression when moderate or intense staining.
Associated with disease stage.
Not associated with survival. 40 Both membrane-bound and cytoplasmic HLA-G expression were considered as positive.
Associated with cancer cell differentiation, lymph node metastasis.
Ewing sarcomas 47 (primary) Graded by low, intermediate or strong densities. Associated with tumor infiltrating T cells.
An independent prognosis factor.
Associated with advanced disease stage.
An independent prognosis factor.
Associated with advanced disease stage.
Negative ≤ 5% and positive >5%. Predominately expressed in adenocarcinoma. A recent study revealed that ILT-2/-4 extracellular D1D2 are responsible for the interaction with HLA-G binding, while D3D4 act as a scaffold (86). Also, ILT-2/-4 are more accessible to the HLA-G dimer binding than that of HLA-G monomer, leading to much stronger inhibitory signals. ILT-2 and ILT-4 extracellular D1D2 binds to the extracellular a3 domain of HLA-G, but structurally dependents. ILT-2 only binds to the HLA-G heavy chain associated with b 2 m, while ILT-4 can bind to both b 2 m free HLA-G heavy chain and HLA-G heavy chain with b 2 m. Moreover, residues Tyr 38 and Tyr 76 in ILT-2 are responsible for binding to HLA-G Phe 195 in the a3 domain, whereas residues Tyr 36 and Arg 38 in ILT-4 interact with the Phe 195 -Tyr 197 loop in the HLA-G a3 domain (87). The different binding sites between ILT-2/-4 and HLA-G could be an explanation for the higher affinity of ILT-4 than the affinity of ILT-2 when they interact with HLA-G (88). Based on the different extracellular structure of HLA-G isoforms and ILT binding characteristics, ILT2 can bind to the b 2 m associated HLA-G1 and HLA-G5 isoforms. However, ILT-4 can bind to both b 2 m associated or b 2 m free HLA-G isoforms, which include HLA-G1, -G2, -G4, -G5 and HLA-G6 isoforms (Figure 4).
Taking advantage of the early studies on fetal-maternal immune tolerance, HLA-G-mediated immune inhibition has been well-acknowledged in the broader spectrum of health and disease situations. HLA-G expression favors the acceptance of
Associated with poor prognosis, disease recurrence or metastasis.
Lower HLA-G expression associated with longer survival.
Positive HLA-G expression was highly represented in patients with ovarian carcinoma recurrence. allograft organ transplantation, whereas it provides an additional strategy for cancer cells to escape from immune surveillance and clearance (89,90). HLA-G/ILTs engagement activates the phosphorylation of the ITIMs contained in the tyrosine residues, which present docking sites for Src homology 2 (SH2) protein tyrosine phosphates SHP-1 and SHP-2, thus initiating an inhibitory signaling cascade. Meanwhile, ITIMdependent recruitment of SHP1/SHP2 can markedly suppress the ITAM activated Syk/Src signal cascades in immune cell activation (91).

CHALLENGES IN HLA-G/ILTs TARGETED SOLID CANCER IMMUNOTHERAPY
One strategy successfully deployed by cancer cells for immune evasion is the impairment of the classical HLA class I and II antigens to hide infected cells from T cell recognition, while aberrant induction of HLA-G expression by cancer cells makes host anti-tumor immune system rather vulnerable (92). Though development of HLA-G/ILTs interaction targeted ICIs is promising for cancer immunotherapy, much real-world information on HLA-G/ILTs status is extremely necessary for both future basic and clinical investigations.

Lack of HLA-G Isoform-Specific Monoclonal Antibodies
The most fundamental task is to develop HLA-G molecule universal or distinct HLA-G isoform-specific mAbs. As high as 80% amino acid sequence identity sharing in the extracellular domain of all HLA class I antigens, only distinct feature of HLA-G is its molecular weight of 39 kDa (HLA-G1 isoform) which is less than 45 kDa of the other classical HLA class I antigens (93), and this is rather similar to other non-classical HLA class I antigens including HLA-E and HLA-F (https://www.uniprot.org/uniprot/P17693; https:// www.uniprot.org/uniprot/P13747; https://www.uniprot.org/ uniprot/P30511). CLUSTALO sequence alignment results showed that, among HLA-G, HLA-F, and HLA-E molecules, the full-length amino acid sequence are identity round 62%, 75.9% between HLA-G and HLA-F, 71.2% between HLA-G and HLA-E, and 40.1% between HLA-E and HLA-F (Supplementary Figure 1). Thus, the cross-reactivity of anti-HLA-G mAbs to other classical and nonclassical HLA class I antigens remains a huge task to be improved (94). Indeed, cross-reactivity of the most widely used such as mAb 4H84 for denatured HLA-G form has been observed to react with other HLA class I antigens (95,96). Moreover, due to HLA-G primary transcript alternative splicing, in addition to seven everidentified HLA-G isoforms and more novel isoforms can be expected, lack of HLA-G isoform-specific mAb prevents advances in characterizing their biological functions and clinical significance (97). To be noted, a panel of novel HLA-G isoforms predicted by deep transcriptome analysis have been reported. Among these isoforms, HLA-G1L have extra five amino acids (NKTPR) ahead  the beginning residue Methionine at the N-terminal ends, others such as isoforms lack a1 or both a1 and a2 extracellular domains, and novel soluble HLA-G isoforms with distinct C-terminal ends generated by skipping exons 5 and 6. Among anti-HLA-G antibodies, mAbs 4H84 and 5A6G7 are with known recognizing epitope in HLA-G heavy chain. mAb 4H84 generated by amino acid residues from 61 to 83 (in a1 domain HLA-G, which probes all denatured HLA-G isoforms (98). mAb 5A6G7 generated by a 22-mer C-terminal amino acid sequence in HLA-G5 and HLA-G6, which probes both native and denatured HLA-G5 and HLA-G6 isoforms (99,100). mAbs 4H84 and 5A6G7 are the most widely used antibodies for evaluating total HLA-G (mAbs 4H84 and MEM-G/1) or HLA-G5/6 isoforms (mAb 5A6G7) respectively in malignant lesions with immunohistochemistry. As a result, novel isoforms without a1 domain can't be detected by the mAb 4H84 (23,79). In this scenario, the interpretation of clinical significance of HLA-G in cancers seems rather premature unless more reliable specific anti-HLA-G mAbs are used ( Figure 5).
Furthermore, evaluation criteria for lesion HLA-G expression including staining protocols, cut-off levels, and cross-assay concordance are far from standardized, which could dramatically affect the definition of HLA-G expression and interpretation of its clinical significance, even with a same mAb to detect HLA-G within a certain type of cancer (62,96). In this context, a reference criterion should be recommended by international community for HLA-G staining to minimize discrepancies across studies is urgently warranted.
Also, given different binding specificity of HLA-G between ILT-2 and ILT-4 resulted from its extracellular structure, that ILT-2 binds HLA-G1 and HLA-G5 while ILT-4 binds HLA-G1, -G2, -G4, -G5, and HLA-G6 isoforms, the landscape of HLA-G isoforms and their degree of expression in cancers can tremendously affect the benefits of HLA-G/ILTs based cancer immunotherapy.

Heterogeneity of HLA-G Expression in Malignancies
Accumulating evidence solidify the concept that tumor heterogeneity, including inter-patient, inter-tumor, and intratumor heterogeneity, is the main cause for variable responses and clinical outcomes to anti-cancer therapy (101). Since the first report HLA-G expressed in tumors, the degree or proportion of HLA-G expression in thousands of malignant lesions among over thirty different types of cancers have been explored. Data revealed that HLA-G expression is restricted to malignant lesions, but not in adjacent non-tumorous tissues, and that neo-HLA-G expression is strongly related to metastasis, advanced disease stage, poor prognosis, and clinical outcome (89). However, inter-patient, inter-tumor, and intra-tumor heterogeneity of the HLA-G expression in each histopathological type of malignancies is also evident (102,103). In addition to the HLA-G heterogeneity caused by tumor cell itself, such as clonal growth with genetic alterations, epigenetic and posttranslational modifications, explanation of HLA-G expression could be biased due to usage of different current available anti-HLA-G monoclonal antibodies, and different assay protocols (62,96). Very recently, using the method of flow cytometry, our data showed that HLA-G expression in 157 epithelial cell adhesion molecule (EpCAM) positive-gated colorectal tumor lesions is with a median of 14.90% (range: 1.81~79.90%) (24).
For an example, the inter-patient proportion of HLA-G expression in cancers has been observed from 24 to 94.1% in breast cancers (12,14), and from 22.1 to 70.7% in colorectal cancers (19,20). Noteworthy, only few studies on the Similarly, a study by Swets et al. (16) showed that HLA-G expression was found in 29% of the primary colorectal cancer and 30% of the corresponding liver metastases. Finally, regarding the intra-tumor heterogeneity of the HLA-G, Rouas-Freiss et al. (103) reported that, among 19 clear cell renal-cell carcinoma lesions, the proportion of HLA-G expression varies dramatically on CA9+ clear cell renal-cell carcinoma cells in different zones in each sample, which could be ranged from negative to almost totally positive for HLA-G expression. Also with clear cell renalcell carcinoma lesions, a study by Tronik-Le Roux et al. (79) released that HLA-G isoform expression including HLA-G1, -G5, and HLA-G6 are extremely heterogeneous among distinct subcellular locations and zones within a same tumor. In a serial section study with colorectal and esophageal cancer lesions, our recent findings further revealed that intratumor heterogeneous expression of HLA-G is a very frequent phenomenon among different zones within a tumor (102).

Native HLA-G Isoform Expression in Cancer Lesions Needs Evaluated
As aforementioned, most currently available information of HLA-G expression in cancer lesions were evaluated by immunohistochemistry with the mAbs 4H84 and/or 5A6G7, which represents all a1 domain containing HLA-G or HLA-G5/6 isoform expression. Whether these data are consistent with the levels of tumor cell surface HLA-G expression remain elusive. Previous evidence showed that no correlation has been established between the degree of cancer lesion HLA-G5/6 expression evaluated with immunohistochemistry and peripheral soluble HLA-G levels (42). Given the fact that HLA-G/ILTs interaction is conformation dependent, tumor cell surface native or conformational HLA-G expression be evaluated with assays such as flow cytometry is necessary (105). To address this issue, more reliable and specific antinative or -conformational HLA-G mAbs are yet to be explored.

OTHER HLA-G RECEPTORS
In addition to the ILT-2 and ILT4, receptors including killer inhibitory receptor (KIR) 2DL4/CD158d, CD8, CD160 could also bind HLA-G (106). To be noted, NKG2A/CD94 has been recently reported which could bind to allelic specific products of HLA-G (107). KIR2DL4 is a member of the killer cell immunoglobulin (Ig)like receptor (KIR) family, but with an atypical feature owing to a D0 and D2 hybrid extracellular domain, a positively charged arginine residue in the transmembrane region and one ITIM domain in its cytoplasmatic tail. The charged arginine residue enables KIR2DL4 to associate with the Fc fragment receptor g (FcRg), which contains two cytoplasmic immunoreceptor tyrosine-based activation motifs (ITAMs). With this unique structure, KIR2DL4 is of both the activation (ITAM) and inhibitory (ITIM) signaling domains (108). The biological function of KIR2DL4 and HLA-G binding is to work as an FIGURE 5 | Epitopes in HLA-G recognized by mAbs 4H84 and 5A6G7. Among HLA-G antibodies, only mAbs 4H84 and 5A6G7 were generated with definite immunogen epitopes. mAb 4H84 generated by 61 st~8 3 rd amino acids (EEETRNTKAHAQTDRMNLQTLRG) in HLA-G a1 domain which recognizes denatured heavy chain of all seven identified HLA-G isoforms containing a1 domain. mAb 5A6G7 generated by 21-mer C-terminal amino acid (SKEGDGGIMSVRESRSLSEDL) in HLA-G5 and -G6 isoforms, which recognizes both native and denatured heavy chain of HLA-G5 and HLA-G6 isoforms. Novel HLA-G isoforms such as isoforms without a1 domain and transmembrane region, or with an extended 5'-region were predicted. However, no current antibody is available to detect. and represent mAb 4H84 and 5A6G7 recognizing site in HLA-G isoforms.
activating receptor, which induces strong pro-inflammatory cytokine and chemokine immune responses through the endosomal DNA-dependent protein kinase (DNA-PKcs) signaling pathway, but not the NK cell cytotoxicity (109). KIR2DL4 is predominately expressed in decidual NK cells and HLA-G/KIR2DL4 interaction plays critical roles in the regulation of maternal-fetal immune microenvironment, spiralartery remodeling and fetal growth (4).
The heterodimer NKG2A/CD94, a member of the C-type lectin-like receptor family, is a well-known immune inhibitory receptor for HLA-E binding. NKG2A/CD94 is mainly expressed on CD8+ T lymphocytes and subsets of NK cells (110). Additionally, NKG2A/CD94 has recently been found to be an HLA-G allelic product dependent receptor. Hò et al. (107) revealed that a remarkably higher binding affinity was observed for the HLA-G*01:04/NKG2A/CD94 interaction than those for the HLA-G*01:03/NKG2A/CD94 and HLA-G*01:01/NKG2A/ CD94 interactions. Moreover, no engagement was observed between the activating receptor NKG2C/CD94 and HLA-G. HLA-G allelic product dependent binding of NKG2A/CD94 could result from the single amino acid residue in the HLA-G*01:01 heavy chain differing from HLA-G*01:04 (p.110L > I) and HLA-G*01:03 (p.31T > S), and these different residues may affect the peptide repertoire and receptor recognition (111). Consequently, different biological function modulation could be expected for the different allelic HLA-G molecules on the immune cells expressing NKG2A/CD94.
Other receptors are the glycosylphosphatidylinositol-anchored receptor CD160, which is expressed on activated endothelial cells, CD16 + CD56 dim NK cells, and CD8 + T cells (112,113). CD160 signaling depends on adapter proteins, such as the phosphoinositide-3 kinase, to activate endothelial cell migration and angiogenesis, and immune cells, such as cytokine releasing and target cell lyses (112,114). Moreover, the cytotoxic T cell surface marker CD8 is reported to interact with the HLA-G protein, which can induce apoptosis of the CD8 + T cells and CD56 + CD8 + NKT cells through the Fas/FasL pathway (115).

CONCLUSIONS AND PERSPECTIVES
In most scenarios, HLA-G/ILTs interaction promotes cancer cells to escape immune surveillance and anti-tumor immunity (97,116). Interference or blockade of HLA-G/ILTs interaction can restore host anti-tumor immune responses, which providing a strong rationale and opportunities to develop HLA-G/ILTstargeted ICIs for solid cancer immunotherapy. Indeed, early clinical trials based on this immune checkpoint is being launched for different advanced solid cancer treatment 2 . However, many challenges remain to be addressed.
First, multiple identified HLA-G isoforms and more can be expected, which have distinct extracellular domain(s) for each of them. This information indicates that biological function of different isoforms is diverse. Being lack of isoform-specific antibodies, their tumor tissue expression characteristics and clinical significance is unclear. Currently available mAb 4H84, which probes all isoforms containing a1 domain, can't distinguish either a distinct isoform or combination of different isoforms expressed on tumor tissues. Novel isoforms without a1 domain can't be detected with any available mAbs which can be ignored in tumor tissues. These findings sharply compounded the clinical relevance of HLA-G in tumor patients (62). Second, being lack of international validated or recommended assay protocols, both performance and cut-off points are diverse and unconcordance in interpretation of HLA-G expression across studies is rather common (117). Third, being the landscape of HLA-G isoform expression in tumor tissues can't be specified, application of HLA-G/ILTs-targeted ICIs for cancer treatment can be aimless. Furthermore, HLA-G/ILTs engagement is isoform dependent, where HLA-G1 and HLA-G5 binds ILT-2, and HLA-G1, -G2, -G4, -G5, and HLA-G6 binds ILT-4 (118). Differential expression of HLA-G isoforms does exist in tumor cells whereas the underlying mechanisms are yet to be explored. In this scenario, patient individualized landscape and degree of HLA-G isoform expression should be defined before therapy due to which can tremendously affect clinical benefits of HLA-G-based solid cancer immunotherapy.
In summary, exploration of more reliable and HLA-G isoformspecific antibodies, implementation of international community validated HLA-G detection protocols, deeper insight of patient individualized landscape of HLA-G expression, and development of isoform matched both HLA-G and ILT-2/-4 blocking antibodies, are future directions warranted for the HLA-G/ILTs-targeted solid cancer immunotherapy. Despite these challenges, owing to HLA-G expression is restrict to malignant tissues, and HLA-G/ILTs signaling is involved in a broader spectrum of immune responses than CTLA-4/B7 and PD-1/PD-L1 does, the clinical effects of the immune checkpoint HLA-G/ILTs is optimistic.

AUTHOR CONTRIBUTIONS
Conceptualization, data curation, funding acquisition, writingoriginal draft, review and editing: W-HY and AL. Both authors contributed to the article and approved the submitted version.