Regulation of PD-L1 Trafficking from Synthesis to Degradation

Programmed death-ligand 1 (PD-L1) was first identified in a homology screen for the immune costimulatory molecules B7-1 and B7-2. Originally named B7 homolog 1 (B7-H1), it was later rechristened as PD-L1 emphasizing its ability to bind programmed cell death protein 1 (PD-1) on activated T cells (1, 2). The PD-L1/PD-1 interaction leads to the recruitment of the protein tyrosine phosphatases such as SHP2, which play a role in the dephosphorylation of key signaling mediators of T-cell activation (3–7). As an immune checkpoint, the PD-L1/PD-1 axis limits the intensity and duration of an immune response and maintains self-tolerance (8). Its dysregulation can lead to a wide array of diseases, such as lupuslike syndrome, type 1 diabetes mellitus, and loss of fetomaternal tolerance in mice (9).

Expression of PD-L1 protein is upregulated in response to different inflammatory cues (10). Cancer cells can coopt upregulation of PD-L1 expression to evade an effective antitumor response (11). In recent years, the PD-L1/PD-1 checkpoint axis has emerged as a therapeutic target, with eight mAbs specific for either PD-1 or PD-L1 approved by the FDA for clinical use in several cancers (12). While this has expanded the treatment options available for patients, for unknown reasons, most patients do not respond to immunotherapy targeting PD-L1/PD-1 (13). To better predict response rates and improve the efficacy of targeted anti–PD-L1 therapy, it is critical to have a detailed understanding of the regulation of PD-L1 at the transcriptional, translational, and posttranslational levels.

In this review, we discuss the transport of PD-L1 to and from the plasma membrane as a posttranslational mechanism that acutely regulates PD-L1 plasma membrane density and activity as a PD-1 ligand. We begin by briefly summarizing the transcriptional and translational regulation of PD-L1, topics that have been extensively reviewed elsewhere (14–16). We then describe the cellular distribution of PD-L1, the endocytic processes that mediate this distribution and how these may be exploited for cancer therapy.

Several inflammatory cytokines, including IFNγ, IFNα, TNFα, and multiple interleukins increase CD274 (PDL1) transcription (17, 18). The main transcriptional factors that regulate CD274 expression include IRF1, NFkB, and STAT3 (14). In addition to inflammatory cytokines, CD274 transcription has also been described to be upregulated in response to various cell stressors. For instance, during hypoxia, HIF1α and hedgehog signaling can induce PD-L1 expression (19, 20). Chemotherapy and radiotherapy have also been associated with enhanced PD-L1 expression (21, 22). Apart from cell-extrinsic stimuli, cell-intrinsic changes driven by oncogene signaling can also increase CD274 transcription. Driver oncogenes including those encoding mutant EGFR, BRAF, and ALK can enhance STAT3-mediated CD274 transcription via changes in c-Jun or AKT signaling (23–26). Oncogenic MYC has been shown to increase CD274 mRNA levels through direct binding to the CD274 promoter in several cancer models (27). CD274 transcription can also be epigenetically regulated by methylation of the promoter region and associated histones (28–30). Intriguingly, there is very little evidence indicating oncogene-driven CD274 transcription leads to increased steady state PD-L1 protein levels. In both melanoma and prostate cancer, no association was found between the presence of oncogenic mutations and PD-L1 protein level (31, 32).

Posttranscriptionally, PDL1 mRNA is subject to mRNA stability control. A number of miRNAs, such as miR-200 in non–small cell lung cancer (NSCLC) and miR-424(322) in ovarian cancer, have been shown to directly target the three prime untranslated region (3′-UTR) of PDL1 and enhance its turnover (33, 34). Moreover, PDL1 transcripts with aberrant 3′-UTR structure are associated with higher stability and enhanced PD-L1 protein levels (35). Expression of PD-L1 protein may also be more directly induced by recruiting polysomes to PDL1 mRNA, as happens in response to the activation of the PI3K/Akt/mTOR/S6K1 pathway (36, 37).

PD-L1 is a single-pass transmembrane protein with a 31 amino acid C-terminal cytoplasmic domain and a 220 amino acid ectodomain composed of IgV- and IgC-like domains (Fig. 1A; ref. 38). Because it is the PD-L1 IgV-like domain that binds PD-1 on T cells, PD-L1 must be properly translocated to the plasma membrane to engage PD-1 and exert its immunosuppressive effects (Fig. 1B; refs. 39–41). As a result, PD-L1 cellular compartmentalization is closely linked with its checkpoint activity.

IHC staining of human tissue, using mAbs that recognize the PD-L1 extracellular domain, indicate a predominantly plasma membrane localization in malignant epithelial cells as well as a variety of stromal cells across multiple tumor types (42). It is difficult to evaluate the cellular distributions in IHC studies of fixed-formalin paraffin-embedded (FFPE) tissue because the processing required for antigen retrieval can alter staining patterns (43, 44).

Studies of cultured cells, which are not subject to the same technical constraints as FFPE studies, have demonstrated PD-L1 at the plasma membrane and in intracellular compartments of several cancer cell lines (45–47). Part of the intracellular pool of PD-L1 is newly synthesized protein in the endoplasmic reticulum (ER) and Golgi that is being modified into the mature protein along the secretory pathway (48, 49). In addition to the biosynthetic-secretory pathway, a perinuclear staining pattern in some of these cells suggests PD-L1 may also be present in the endosomal recycling compartment (45, 47, 50). This is supported by studies that show PD-L1 colocalization with markers of the endosomal recycling compartment, such as the transferrin receptor and/or Rab11 as well as analysis of the PD-L1 proximal proteome using APEX2 proximity mapping (47, 50–53).

PD-L1 is primarily N-glycosylated, through a process that begins in the ER with the transfer of oligosaccharides to specific asparagine residues (N35, N195, N200, and N219) of the PD-L1 extracellular domain in the ER lumen (54–56). The core glycan is then remodeled and further processed in the ER and Golgi apparatus where it acquires complex glycan modifications before PD-L1 is translocated to the plasma membrane. At the plasma membrane, PD-L1 is anchored by the electrostatic interaction between the positively charged amino acids of its cytoplasmic domain and the phospholipids of the plasma membrane (57). Aberrant glycosylation during PD-L1 maturation can lead to PD-L1 accumulation in the ER, and reduced translocation to the cell surface. Abnormally glycosylated PD-L1 can be removed through ER-associated degradation (ERAD; ref. 49).

At the cell surface, PD-L1 is constantly endocytosed and recycled (Fig. 2; refs. 47, 51). Recent APEX2 proximity mapping of the PD-L1 proximal proteome supports constitutive traffic of PD-L1 through the compartments of the general endosomal recycling system (47). This provides a mechanism for cells to rapidly, reversibly, and precisely regulate expression of PD-L1 on the plasma membrane (58). PD-L1 endocytosis appears to occur constitutively and does not require ligand binding. There is some evidence showing that PD-L1 is internalized through a dynamin- and clathrin-dependent process (59–61).

Once internalized, a large portion of PD-L1 is recycled back in about 15 minutes, with about 70% of cycling PD-L1 found on the cell surface at equilibrium (45, 47, 51). Blocking recycling with primaquine rapidly depletes PD-L1 on the plasma membrane. PD-L1 recycling can be regulated by CKLF-like MARVEL transmembrane domain containing protein 6 (CMTM6), which is primarily located at the plasma membrane and in recycling endosomes, where it interacts with PD-L1 (51, 62). In CMTM6 knockout cells, PD-L1 is rapidly depleted from the cell surface compared with wild-type cells. However, in the presence of lysosomal inhibitors, the loss of surface PD-L1 is relatively small, suggesting that rerouting to the lysosome rather than slowed-down recycling is the main cause of reduced PD-L1 at the plasma membrane in the absence of CMTM6. The balance between recycling and lysosomal degradation is also central to the regulation of PD-L1 trafficking by trafficking protein particle complex (TRAPPC4) and Huntington-interacting protein 1-related (HIP1R; Fig. 2). TRAPPC4, like CMTM6, promotes PD-L1 recycling away from the lysosome (52). HIP1R, on the other hand, favors the degradative pathway and its ablation leads to increased distribution of PD-L1 from Rab7⁺ late endosomes to Rab11⁺ recycling endosomes (53). Distinct from this recycling–degradation modality, the small GTPase Arf6 has been shown to exert its effect solely on PD-L1 recycling (45). In platelet-derived growth factor–treated cells, ablating Arf6 or its downstream effector AMAP1 leads to reduced PD-L1 recycling and, consequently, lower cell surface levels of PD-L1 without changes in total protein.

In addition to the endomembrane system, PD-L1 has been found in the sera of patients and the supernatant of tumor cell lines and immune cells (63–65). Mechanisms by which PD-L1 could be secreted from cells include exosomal secretion, expression of truncated variants as well as proteolytic cleavage and shedding.

PD-L1⁺ exosomes are formed through an endosomal sorting complex required for transport (ESCRT)–dependent process, and PD-L1–containing exosomal fractions colocalize with ESCRT0, a complex required for the initial recognition of cargo destined for multivesicular bodies (MVB; Fig. 2; refs. 66, 67). Once PD-L1 is localized to the limiting membrane of MVBs, it is sorted into the intraluminal vesicles of MVBs through the action of the ESCRT accessory protein, ALG-2–interacting protein X (ALIX; ref. 68). The final fusion and release of PD-L1⁺ exosomes require the activity of Rab27A, a protein that is essential for the docking of MVBs at the plasma membrane (67, 68). PD-L1 on exosomes can inhibit T-cell activation in vitro and in the draining lymph node, promoting tumor progression in murine models of prostate cancer and melanoma (66, 67). Although PD-L1 has also been detected on microvesicles, the regulation of this process has not been investigated (66).

Distinct from its exosomal counterpart, soluble PD-L1 does not require membrane integration to be secreted (Fig. 2). Some of the soluble variants of PD-L1 result from alternative splicing that excludes the transmembrane domain encoded by exon 5. One such variant, identified in several cancers as well as normal placenta and myeloid cells, results from a read-through into intron 4 coupled with an in-frame stop codon and an alternative polyadenylation signal (69, 70). This variant appears to multimerize and lacks the ability to inhibit T cells both in vitro and in vivo. Instead, it was shown to act as a PD-1 antagonist, blocking the suppressive activity of cell-bound or exosome-bound PD-L1 (71). Other secreted splice variants, detected in melanoma cell lines and patient plasma, appear to result from aberrant splicing that leads to a premature stop codon well before exon 5 (72). In certain patients with NSCLC, two unique secreted PD-L1 variants that are associated with resistance to anti–PD-L1 therapy have been described previously (73). Both isoforms lack the transmembrane domain, and while one variant is devoid of the intracellular domain, the other maintains portions of the cytoplasmic domain (coded by exon 7) due to in-frame skipping of exons 5 and 6.

It is unlikely that all soluble PD-L1 variants arise due to alternative splicing. Several immune modulators can be released from the cell surface through the action of extracellular enzymes, and there is some evidence that PD-L1 may also be similarly shed (74). In vitro studies have shown that PD-L1 can be removed from the cell surface by matrix metallopeptidase 9 (MMP-9) and MMP-13 (75, 76). In addition, MMP inhibition results in reduced levels of soluble PD-L1 in cells engineered to overexpress PD-L1 (63). It is not clear whether these MMPs catalyze controlled proteolysis events that can result in a physiologically relevant PD-L1 fragment. In one study, the addition of MMP-13 to purified PD-L1 resulted in a reduction in full-size PD-L1 levels without the appearance of any discernable lower molecular weight variants, suggesting breakdown to nonfunctional peptides (77). On the other hand, in certain cancer cell lines, the cell surface proteases, ADAM metallopeptidase domain 10 (ADAM-10) and ADAM-17 have been shown to cleave PD-L1 at a yet unknown site(s) freeing a soluble 37 kDa PD-L1 fragment and an unstable, 18 kDa intracellular fragment. The soluble fragment appears to retain its ability to bind PD-1 and suppress CD8⁺ T-cell activity in vitro (78, 79).

Integral plasma membrane proteins, such as PD-L1, are not directly accessible to the proteasome. Instead, cells rely on ubiquitination and membrane trafficking events to target these proteins from the plasma membrane to the MVB for lysosomal fusion and degradation. Ubiquitination involves the attachment of ubiquitin, a 76 amino acid peptide, to lysine residues of substrate proteins in a multistep process (80). The final step of ubiquitin attachment is facilitated by enzymes known as E3 ubiquitin ligases, which recognize the substrate protein and facilitate the transfer of ubiquitin. Conversely, deubiquitinating enzymes catalyze the removal of ubiquitin from target proteins. Ubiquitin itself can be ubiquitinated at its seven lysine residues, resulting in an ubiquitin chain. A protein may be tagged by a single ubiquitin molecule (monoubiquitination) or an ubiquitin chain (polyubiquitination). The type of ubiquitination will influence the fate of the modified protein. For instance, membrane proteins destined for lysosomal degradation are usually marked by multiple monoubiquitination or Lysine-63–linked polyubiquitination (81).

Monoubiquitinated PD-L1 can be destined for lysosomal degradation in an ESCRT-dependent process (50). Although the specific site of monoubiquitination remains to be identified, it likely occurs on the lysines of the PD-L1 cytoplasmic domain, as palmitoylation at the cytoplasmic residue C272 inhibits monoubiquitination (Fig. 3; refs. 50, 82). Likewise, CMTM6, which is predominantly present within the plasma membrane, protects PD-L1 from degradation presumably by masking the STUB1 E3 ubiquitin ligase recognition site in the transmembrane-intracellular domain of PD-L1 (Fig. 2; refs. 51, 62). Another CMTM family member, CMTM4, similarly inhibits PD-L1 degradation (62). CMTM4 activity, unlike that of CMTM6, can be downregulated by AMPK, which disrupts the PD-L1–CMTM4 interaction by phosphorylating PD-L1 (83). PD-L1 may also be targeted to the lysosome in a ubiquitin-independent manner: in a number of cancer cell lines, HIP1R, which can recruit ALIX through its dileucine motif, can directly bind PD-L1 and direct it to MVBs for lysosomal degradation (Fig. 2; ref. 53). What determines the fate of the MVB, fusion with the plasma membrane or fusion with lysosomes, is not understood.

Besides quality control, cells may target nascent proteins for proteasomal degradation as a way of regulating surface expression (81). This homoeostatic function is apparent in the fluctuation of PD-L1 protein level during cell-cycle progression observed in multiple human cancer cell lines: PD-L1 levels peak in M- and early G₁-phases, followed by a rapid reduction in late G₁- and S-phases (84). The mechanism involves cyclin D–CDK4, which is primarily active in the G₁–S-phase. Cyclin D–CDK4 phosphorylates and stabilizes speckle-type POZ protein (SPOP), which is a cullin-3–based E3 ligase substrate-recruiting adaptor protein. In turn, SPOP mediates PD-L1 polyubiquitination and degradation. This interaction is dependent on the PD-L1 cytoplasmic domain, deletion of which disrupts binding of PD-L1 to SPOP and renders PD-L1 resistant to SPOP-meditated polyubiquitination.

Another example of PD-L1 regulation by proteasomal degradation involves the GSK kinases, which lie downstream of the PI3K/AKT signaling axis (Fig. 3). In response to EGFR inhibition, GSKα is activated and phosphorylates PD-L1 on the cytoplasmic tail at positions Ser279/283 (85). This leads to the recruitment of the E3 ubiquitin ligase ARIH1, which ubiquitinates and targets PD-L1 for degradation. GSK3β, which shares 97% catalytic domain sequence identity with GSKα, also phosphorylates PD-L1, but it does so on the extracellular domain at residues T180 and S184 (54). βTRCP, a subunit of the SCF E3 ligase complex, is then recruited to PD-L1. It is important to note that to access these PD-L1 ectodomain residues, which would be positioned in the ER lumen, GSK3β and βTRCP must localize either in the lumen or luminal face of the ER membrane. However, there is little to no evidence showing such a localization (86). It is rather more likely that these interactions occur only after selection and initial translocation of PD-L1 into the cytosol, which would then allow access to the GSK3β and βTRCP. While the signal for substrate recognition is not clear, there is some evidence suggesting glycosylation status may play a role.

Glycosylation of PD-L1 at N192, N200, and N219 prevents phosphorylation by GSK3β, and as a result blocks βTRCP recruitment, and proteasomal degradation (54). The enzyme(s) responsible for PD-L1 glycosylation, in the context of βTRCP-mediated degradation, have not been investigated. Nonetheless, the N-glycosyltransferases STT3 and B3GNT3 have been shown to glycosylate PD-L1 and protect it from degradation in separate studies using different cell types (55, 87). Modulating glycosyltransferase activity can be a way of regulating PD-L1 levels posttranslationally. For instance, EGF-induced B3GNT3 upregulation and IL6-induced STT3 recruitment can result in reduced PD-L1 degradation (55, 88). NIMA-related kinase 2 (NEK2), which phosphorylates PD-L1 at T194/T210, similarly promotes PD-L1 glycosylation and stability (89). Direct changes to deubiquitinating enzymes such as CSN5 and USP22 can also affect PD-L1 stability (90, 91).

The potential of endocytic trafficking either as a therapeutic target or a means of drug delivery has been extensively studied in several diseases (92, 93). With respect to targeting PD-L1, current treatment options have focused on blocking its interaction with PD-1 using PD-L1–specific mAbs (94). Although it is known that certain therapeutic PD-L1–specific mAbs can be internalized after binding PD-L1, mAb binding does not appear to alter PD-L1 trafficking (47, 95). There is, however, considerable information regarding ways of therapeutically targeting PD-L1 trafficking in the preclinical setting. Most of these have focused on inducing PD-L1 degradation by either targeting it to the lysosome or interfering with its maturation in the ER.

STM108, an antibody that binds and targets glycosylated PD-L1 to the lysosome, inhibits tumor growth in a syngeneic mouse model of triple-negative breast cancer in which the cancer cells have been made to express human PD-L1 (55). Lysosome-targeting peptides that bind PD-L1 and lead to its degradation are also being studied. For instance, a rationally designed peptide called PD-LYSO targets PD-L1 to the lysosome, whereas a cell-penetrating peptide called PD-PALM blocks PD-L1 palmitoylation leading to increased degradation in the lysosome (50, 53). Small molecules have also been used to reroute PD-L1 to the lysosome. An inhibitor of Sigma1 induces PD-L1 degradation via autophagy in triple-negative breast and androgen-independent prostate cancer cell lines (96). SA-49, an alpperine derivate, decreased the protein level of PD-L1 in NSCLC cells and the Lewis lung tumor model (46). However, its effect appears to result from a general increase in lysosomal activity, rather than a targeted removal of PD-L1. A similarly nonexclusive approach has been used to induce and target misfolded PD-L1 to ERAD by using SAFit 1 and 2 to inhibit FKBP5, a cochaperone with protein folding activity (97). Likewise, resveratrol inhibits enzymes that process glycosylated PD-L1 and promotes ER retention. It is not fully clear whether ERAD removes this population of PD-L1 (48). Modulating the activity of kinases such as AMPK, GSKα/β, and NEK2 that regulate PD-L1 stability is also being explored as a way to regulate PD-L1 levels (98).

Besides novel ways of targeting PD-L1, another approach has been to enhance the effectiveness of existing antibody treatment by modulating PD-L1 trafficking. Examples of this include the use of prochlorperazine (PCZ), a clinically available drug known to inhibit dynamin (59). By decreasing PD-L1 internalization and, consequently increasing surface expression, PCZ was shown to enhance the effect of the PD-L1–specific mAb avelumab in a model of colon cancer in mice. The drug tiplaxtinin, which inhibits PAI-1–mediated PD-L1 internalization, similarly synergizes with anti–PD-L1 therapy in a murine model of melanoma (60). It should be noted that both studies used anti–PD-L1 with an intact Fc region capable of inducing antibody-dependent cellular cytotoxicity (ADCC). Indeed, the improved efficacy of avelumab in the presence of PCZ is due to increased natural killer cell–mediated ADCC (99). It is not clear whether a similar approach would be useful for other PD-L1–specific antibodies such as durvalumab and atezolizumab, which lack a functional Fc region, and therefore do not induce ADCC. Combination of cyclin-dependent kinase (CDK) inhibitors with anti–PD-L1 therapy is also being extensively investigated. CDK 4/6 inhibitors such as palbociclib and abemaciclib have been shown to reduce tumor cell growth by cell-cycle arrest and positively impact T-cell activation, leading to increased memory T-cell phenotype and enhanced effector functions (84, 100, 101). However, CDK4/6 inhibition increases PD-L1 expression as previously discussed, therefore has limited efficacy alone (84). The increased PD-L1 can be offset by immune checkpoint inhibitor (ICI) therapy. Consistent with this, CDK inhibition has been shown to prime tumors for anti–PD-L1 therapy to markedly improve overall survival in several preclinical models (84, 100).

The clinical application of ICIs in treating cancer has vaulted the study of immune coinhibitors such as PD-L1 from the purview of basic immunology to the center stage of translational medicine. While there was, initially, great enthusiasm for ICIs directed against PD-L1, this has been tempered by low response rates and the associated suboptimal predictive value of PD-L1 expression. Consequently, there is a growing appreciation for exploring the basic biology of PD-L1 with the perspective that this may lead to a better understanding of the potential as well as the limitations of ICIs. Accordingly, a body of work focusing on mechanisms regulating PD-L1 expression at the transcript and protein level has emerged. Yet, the dynamic nature of an immune response from initiation to resolution makes it crucial to also investigate rapid regulatory mechanisms, such as changes in PD-L1 trafficking, which might play a role in modulating ICI activity.

As described in this review, a relatively small, but growing, number of studies have established PD-L1 as a constitutively recycling receptor that can be localized to different cellular compartments. What emerges from these findings is a model of posttranslational regulation centered around modulating total levels of PD-L1 through ubiquitin-mediated degradation in the lysosome and proteasome. A PD-L1 C-terminal degradation sequence appears to be crucial in determining the ubiquitination and, therefore, the half-life of PD-L1; CMTM4/6, glycosylation, and palmitoylation shield access to this putative degron(s), while cyclin D–CDK4 stabilizes the ubiquitin ligase that recognizes it. By controlling access to a native degradation sequence in such a manner, cells may rapidly upregulate or downregulate total PD-L1 levels (102). It is not clear whether the PD-L1 proteasomal degradation pathways that have been reported so far are distinct from the conventional quality control mechanism that operates in the ER. If PD-L1 can indeed be extracted from the plasma membrane or other intracellular compartments and targeted to the proteasome, it will be important to elucidate how this process unfolds, especially in relation to the conventional lysosomal pathway. Given that frequently used markers of PD-L1 proteasomal degradation such as K48-linked ubiquitination (103) and inhibition by the drug MG132 (104) can also affect MVB maturation and lysosomal degradation, it will be essential to use more refined methods to identify the degradative fate of PD-L1.

Besides regulating degradation, changes in protein trafficking can reversibly dictate protein function by translocating membrane proteins from intracellular stores to the plasma membrane. For instance, the immune checkpoint receptor, CTLA-4 can be translocated to the cell surface in response to T-cell receptor signaling at the immune synapse (105). PD-L1 has also been shown to accumulate at the immunological synapse formed by the interaction of PD-1⁺ T cells and activated dendritic cells (106).

It is plausible that PD-1/PD-L1 interactions immobilize PD-L1 at the plasma membrane and inhibit its endocytosis. This would then lead to a rapid accumulation at the plasma membrane, as exocytosis would still be maintained while PD-L1 molecules that would have otherwise been internalized and routed to degradative endolysosomal compartments would remain at the cell surface. In such a scenario, free PD-L1, owing to its constitutive endocytosis and lysosomal degradation, will not be abundantly available at the cell surface to be bound by either diagnostic or therapeutic antibodies. Yet, when bound by PD-1, PD-L1 would accumulate at the plasma membrane and exert its immune checkpoint function. This context-dependent PD-L1 surface expression may partially explain why there are low response rates to anti–PD-L1 therapy, and why certain tumors that are classified as PD-L1^– using diagnostic antibodies are found to respond to anti–PD-1 therapy (13, 107). This is supported by preclinical studies that show an enhanced response to anti–PD-L1 therapy when PD-L1 surface expression is increased through the inhibition of PD-L1 endocytosis and/or degradation (59, 60, 84).

Finally, there is emerging evidence supporting PD-L1 having cell-intrinsic functions in addition to its role as a transmembrane ligand for PD1 in regulation of epithelial–mesenchymal transition, cell migration, mutant EGFR turnover, extracellular vesicle formation, resistance to chemotherapeutic agents as well as IFN-mediated cell toxicity (47, 51, 108–111). These functions may be dependent on subcellular PD-L1 localization, and therefore trafficking of PD-L1 may contribute to modulation of PD-L1–dependent biology beyond controlling PD-L1 density in the plasma membrane. In addition, therapeutic anti–PD-L1 treatment of EGFR-mutant cells but not EGFR wild-type cells, phenocopies the effect of PD-L1 knockout on cell migration and extracellular vesicle secretion (47), an intriguing finding considering EGFR-mutant tumors respond poorly to ICI therapy. Further studies are required to determine the full impact of PD-L1 cell-intrinsic biology and how ICI therapy affects these functions.

Much of the work discussed in this review has investigated PD-L1 trafficking in tumor cell lines. While immune cells such as dendritic cells also express PD-L1, we know very little about how PD-L1 is trafficked in these cells. Elucidating the basic mechanisms of PD-L1 trafficking in immune cell models is essential as it can be a steppingstone to addressing physiologically pertinent questions such as how PD-L1 trafficking is regulated by immune receptors and ligands that are present in the immune microenvironment. The relation between PD-L1 localization and its emerging tumor cell–intrinsic functions is another area that awaits investigation. Addressing these questions—using cell biological, biochemical, and advanced imaging tools that have already been refined to study protein trafficking—may open the door not only to new paradigms of immune checkpoint regulation, but also to the development of novel therapeutic strategies.

N.K. Altorki reports grants from AstraZeneca and Jansen; personal fees from AstraZeneca and Regeneron outside the submitted work. T.E. McGraw reports grants from NCI during the conduct of the study; nonfinancial support from Medimmune/AstraZeneca, and grants from Johnson and Johnson outside the submitted work. No disclosures were reported by the other authors.

N.K. Altorki and T.E. MsGraw are supported in part by NCI UG3 CA244697. E.Y. Lemma is supported by a diversity supplement to CA244697.