Molecular Biomarkers of Primary and Acquired Resistance to T‐Cell‐Mediated Immunotherapy in Cancer: Landscape, Clinical Implications, and Future Directions

Immunotherapy has transformed cancer treatment, but some patients do not benefit from these therapies. This article describes candidate biomarkers that have been shown to predict resistance to T‐cell‐mediated immunotherapy in the clinical setting. The practical application of these markers and potential targets for overcoming resistance are discussed.

Abstract

The emergence of immunotherapy has revolutionized cancer treatment in recent years. Inhibitors of immune checkpoints, including antibodies against cytotoxic T‐lymphocyte‐associated protein 4, programmed cell death protein 1, and programmed death ligand 1, have demonstrated notable efficacy in certain advanced cancers. Unfortunately, many patients do not benefit from these therapies and either exhibit primary resistance to treatment or develop acquired mechanisms of resistance after initially responding to therapy. Here, we review the genomic and immune traits that may promote resistance to T‐cell‐mediated immunotherapy, with a focus on identifying potential biomarkers that could eventually be used in the clinical setting to guide treatment selection. We summarize the clinical evidence for these markers and discuss how current understanding of resistance mechanisms can inform future studies and aid clinical decision‐making in order to derive maximum benefit from immunotherapy.

Implications for Practice.

Immunotherapy has rapidly progressed as a treatment modality for multiple cancers, but it is still unclear which patients are likely to benefit from these therapies. Studies of resistance mechanisms have only recently started to identify biomarkers that can help predict patient outcomes. This review summarizes the available clinical data in regard to immunotherapy resistance, with a focus on molecular biomarkers that may be useful in guiding clinical decision‐making. It discusses possible applications of these biomarkers and highlights opportunities for further clinical discovery.

Introduction

Cancer treatment has been transformed in recent years by advances in immunotherapy, in particular the advent of inhibitors targeting cytotoxic T‐lymphocyte‐associated protein 4 (CTLA‐4), programmed cell death protein 1 (PD‐1), and programmed death ligand 1 (PD‐L1) [1], [2]. These agents block immune checkpoint pathways, thereby activating a tumor‐specific T‐cell immune response [3]. The efficacy of immunotherapy was initially demonstrated in patients with advanced melanoma who were treated with ipilimumab, an anti‐CTLA‐4 antibody [4], [5], [6]. Inhibition of PD‐1 with the antibodies nivolumab and pembrolizumab was subsequently found to be effective in a variety of malignancies, including melanoma [7], [8], [9], non‐small cell lung cancer (NSCLC) [10], [11], [12], Hodgkin's lymphoma [13], head and neck cancer [14], renal cell carcinoma [15], gastric cancer [16], and hepatocellular carcinoma [17]. More recently, antibodies against PD‐L1, which binds PD‐1, have shown efficacy in urothelial cancer [18], NSCLC [19], and Merkel cell carcinoma [20].

Despite these encouraging results, response rates vary widely across tumor types, and the majority of patients either do not respond to immunotherapy or subsequently exhibit disease progression [21]. These outcomes suggest several categories of resistance to immunotherapy (Fig. 1). In temporal terms, tumors can demonstrate primary resistance, in which patients do not exhibit a significant initial response to treatment, or develop acquired resistance, in which patients respond for some time but their disease subsequently progresses. This classification scheme is readily apparent from a clinical perspective, can be useful for algorithmic approaches to decision‐making, and thus will frame the ensuing discussion.

Figure 1.

Classifications for resistance biomarkers. Biomarkers of resistance to immunotherapy can be categorized via several relevant frameworks. A temporal division between primary and acquired resistance correlates with the real‐time observations of clinicians. The mechanism underlying each biomarker, in contrast, can be spatially sorted as being intrinsic to tumor cells or extrinsic in the microenvironment or systemic circulation. Finally, an immunological perspective describes resistance as mechanistically occurring because of failures in the cancer‐immunity cycle. An immune‐desert tumor is unable to elicit a strong immune response, an immune‐excluded tumor prevents immune cell infiltration despite sufficient immunogenicity, and an inflamed tumor suppresses immune actions despite ample immune cell intrusion.

Abbreviations: IFN, interferon; JAK, Janus kinase; MHC I, major histocompatibility complex class I; MMR, mismatch repair; PD‐L1, programmed cell death ligand 1; TIL, tumor‐infiltrating lymphocyte.

However, the underlying mechanisms of resistance can be fit into several other helpful frameworks. For instance, a resistance factor can be broadly categorized as being either intrinsic or extrinsic to tumor cells. The former encompasses internal characteristics, such as cancer‐specific genetic adaptations, whereas the latter acts to suppress immune function systemically or in the tumor microenvironment [22]. Another relevant schema links immunotherapy resistance to breakdowns in the cancer‐immunity cycle [23]. In this context, resistance occurs because tumors (a) fail to elicit an immune response (“immune desert”), (b) prevent infiltration of immune cells (“immune excluded”), or (c) suppress immune function despite adequate immune presence (“inflamed”).

Finally, resistance mechanisms can be discussed in relation to predictive biomarkers, which importantly have the potential ability to guide treatment decisions. Patients who are unlikely to respond based on these biomarkers might be directed toward alternative therapies and protected from avoidable toxicities. Identification of resistance biomarkers is thus crucial to the fully effective use of immunotherapy.

In this review, we describe candidate biomarkers that have been shown in the clinical setting to predict resistance to T‐cell‐mediated immunotherapy (Fig. 2). Furthermore, we discuss the practical application of these markers and survey potential targets for overcoming resistance.

Figure 2.

Biomarkers of resistance to immunotherapy. Tumor cells can evade T‐cell attack after immunotherapy via primary or acquired mechanisms of resistance. Potential biomarkers of primary resistance include the following: negative PD‐L1 expression (A); low neoantigen or mutation load (B); MMR proficiency leading to decreased neoantigens (C); low levels of TILs in the tumor parenchyma (D); increased frequency of circulating MDSCs (E); decreased levels of IDO (F); deleterious mutations in the IFN‐γ pathway, including in the IFNGR (G); loss of function mutations in JAK 1/2 (H); amplification of MDM2 (I); and increased VEGF signaling (J). Acquired resistance biomarkers include the following: loss of MHC I molecules, leading to inability of T cells to recognize neoantigens via the TCR (K); acquired mutations in JAK 1/2 (L); loss of immunodominant neoantigens (M); and upregulation of suppressors such as TIM‐3 or LAG‐3 (N). Biomarkers can be based on decreased or loss of function (red Xs) or may be upregulated during resistance (red upward arrows).

Abbreviations: APC, antigen‐presenting cell; CTLA‐4, cytotoxic T‐lymphocyte‐associated protein 4; IDO, indoleamine 2,3‐dioxygenase; IFN‐γ, interferon‐γ; IFNGR, interferon‐γ receptor; JAK 1/2, Janus kinase 1 and 2; LAG‐3, lymphocyte‐activation gene‐3; MDM2, murine double minute 2; MDSC, myeloid‐derived suppressor cell; MHC I, major histocompatibility class I; MMR, mismatch repair; PD‐1, 2programmed cell death protein 1; PD‐L1, programmed cell death ligand 1; STING, stimulator of interferon genes; TCR, T‐cell receptor; TIL, tumor‐infiltrating lymphocyte; TIM‐3, T‐cell immunoglobulin mucin‐3; VEGF, vascular endothelial growth factor.

Primary Resistance to Immunotherapy

PD‐L1 Expression

Beginning with the earliest clinical trials of PD‐1 inhibitors, high tumor levels of PD‐L1 have been shown to be associated with improved clinical response [7], [24]. Subsequent trials have repeatedly linked PD‐L1 positivity with favorable outcomes, including in NSCLC [25] and melanoma [26], [27]. As a result, PD‐L1 immunohistochemistry (IHC) assays have obtained regulatory approval both as companion diagnostics that are mandatory for certain treatment indications and as complementary tests deemed likely to predict patient response [28].

Although the general relationship between PD‐L1 and clinical response has been consistent, the degree of association has been markedly variable. Of concern, as many as 20% of PD‐L1‐negative tumors exhibited response to anti‐PD‐1 treatment in certain cohorts.

The basis for this predictive quality was first explained by studies that correlated PD‐L1 positivity with increased numbers of tumor‐infiltrating lymphocytes (TILs) [29]. Notably, these TILs were often found in close proximity to PD‐L1‐expressing cells. Based on this data, Taube et al. proposed a mechanism termed “adaptive immune resistance,” in which tumor cells adapt to immune attack by suppressing the action of cytotoxic T cells, specifically, in this case, by upregulating PD‐L1 [29]. Increased PD‐L1 levels cause a dampening of the immune response via the regulatory effects of PD‐1 pathway activation on these TILs. In keeping with the adaptive immune resistance hypothesis, analysis of tumors in the Cancer Genome Atlas demonstrated that high cytolytic immune activity can lead to PD‐L1 amplification [30].

Although the general relationship between PD‐L1 and clinical response has been consistent, the degree of association has been markedly variable. Of concern, as many as 20% of PD‐L1‐negative tumors exhibited response to anti‐PD‐1 treatment in certain cohorts [31]. This inconsistency may represent true predictive uncertainty, or it may be related to issues with PD‐L1 assays themselves. A comparison of four PD‐L1 IHC assays found that changing the assay system led to a different PD‐L1 classification in 37% of cases [32]. One limitation of PD‐L1 staining is the arbitrary nature of cutoffs for “positive expression,” which are deceptive because PD‐L1 levels exist on a continuum [33]. Additionally, intratumor heterogeneity in PD‐L1 expression suggests that PD‐L1 assays are vulnerable to sampling variation [34]. This problem is particularly relevant to smaller‐sized biopsy samples, although one survey found over 90% concordance in PD‐L1 expression between needle biopsy and surgical resection specimens [35].

PD‐L1 assays have nevertheless been well validated as integral to patient evaluation prior to initiation of immunotherapy. Based on a phase II trial of patients with NSCLC [11], the U.S. Food and Drug Administration (FDA) has approved the PD‐L1 IHC 22C3 pharmDx as a companion diagnostic test for use with pembrolizumab [28]. Another assay using a different antibody, the PD‐L1 IHC 28‐8 pharmDx, has been approved as complementary to nivolumab in NSCLC [10] and melanoma [36]. In addition, the PD‐L1 SP142 assay has been approved for complementary use with the anti‐PD‐L1 antibody atezolizumab in urothelial carcinoma [18] and NSCLC [19], and the PD‐L1 SP263 assay has been approved as complementary to the anti‐PD‐L1 antibody durvalumab [37]. Although the utility of these assays remains somewhat controversial, they overall represent an important step forward in predicting patient outcomes using biomarkers.

Low Mutational Burden and Mismatch Repair Status

Specific targeting of tumors by T cells requires the presence of tumor‐specific antigens that are capable of eliciting an immune response [38]. Of particular interest are neoantigens, which are formed by somatic mutations and contain new epitopes that are recognized as foreign by the immune system [39]. Neoantigens have been postulated to serve as tumor‐ and even individual‐specific targets for T cells [40], whose antitumor activity can then be enhanced by immune checkpoint blockade [39], [41]. Consistent with this hypothesis, poor response to anti‐CTLA‐4 therapy has been correlated with both decreased tumor mutational burden and low neoantigen load [42], [43]. Similar associations between mutational burden and clinical benefit have also been reported with anti‐PD‐1 [44] and anti‐PD‐L1 [18] therapies.

Unfortunately, low mutational load remains a flawed biomarker. Although correlated with response, mutational load was unable to completely predict clinical benefit for anti‐CTLA‐4 therapy in melanoma [42]. Other studies have shown no difference in antigen density between tumors with and without evidence of tumor inflammation [45], [46]. This limitation may be explained in part by variable spatial patterns of neoantigen expression. Clonal neoantigens that are universally expressed within a tumor have been shown to confer greater treatment efficacy, whereas increased intratumor neoantigen heterogeneity is associated with poor response [47]. A more refined assessment of tumor neoantigen load may thus be necessary before it can be utilized clinically to predict immunotherapy response.

Tumor mutation rates are strongly influenced by DNA repair mechanisms [48], [49]. In colorectal cancer, defects in mismatch repair (MMR) can lead to an over 100‐fold increase in mutational burden [50]. MMR status can be assessed by measures of microsatellite instability (MSI), with high MSI indicating MMR deficiency [51]. The relevance of these findings was demonstrated by a phase II trial of pembrolizumab, which reported considerably decreased clinical benefit, including an objective response in 0 of 18 patients, for colorectal cancers with MMR proficiency (or low MSI) [52]. Studies in other solid tumors have similarly demonstrated links between MMR phenotype and clinical response to immune checkpoint inhibitors [52], [53]. A prospective cohort of 86 patients with MMR‐deficient tumors achieved objective radiographic response to PD‐1 blockade in an impressive 53% of cases [54]. Based on the above data, the FDA has now approved pembrolizumab for any MMR‐deficient solid tumors that have progressed on prior treatments. This landmark indication represents the unique development of a “pan‐tumor” biomarker applicable to all solid tumor types.

Lack of Tumor‐Infiltrating Lymphocytes

A long‐observed element of an effective antitumor immune response is the accumulation of TILs in the tumor parenchyma [55]. TILs represent a complex set of immune cells and include the tumor‐specific CD8+ T cells that are potentiated by immune checkpoint inhibitors [56], [57]. In patients with melanoma treated with anti‐CTLA‐4 therapy, tumor regression was associated with corresponding increases in CD8+ T‐cell infiltration [58]. A phase II trial of ipilimumab in advanced melanoma again observed a significant relationship between TILs and clinical activity [59].

The predictive value of TILs in immunotherapy was demonstrated in a cohort of patients with metastatic melanoma treated with anti‐PD‐1 inhibitors. Decreased expression of PD‐1, PD‐L1, and CD8 within the tumor margins were all correlated with poor response [60]. Similar conclusions were obtained in a study of anti‐PD‐L1 therapy against multiple cancer types [27]. These studies suggested that the effects of immunotherapy were mediated by T cells that had already infiltrated tumors but were then negatively regulated by PD‐L1. In mouse models, anti‐PD‐1 therapy was effective even after blocking T‐cell exit from lymphatic tissue, further supporting the idea that pre‐existing TILs are responsible for the clinical response seen with immunotherapy [61].

Tumors can thus be described as “inflamed” or “noninflamed” based on the presence of TILs and related proinflammatory cytokines [56]. A non‐T‐cell‐inflamed tumor can become immunologically “cold” either because of lack of immune activation (as with neoantigen‐poor cancers) or as a result of elements in the microenvironment that exclude T cells from the tumor interior [23], [62]. These factors potentially represent additional biomarkers of resistance. For example, gain‐of‐function mutations in beta‐catenin have been reported to cause decreased T‐cell infiltration by downregulating chemokines such as CCL4 [63]. Of note, these mutations accounted for less than half of non‐T‐cell‐inflamed tumors in this study, indicating the likely presence of multiple complex pathways that promote T‐cell exclusion.

Activation of Immune Regulatory Pathways

Even with adequate TIL presence (i.e., an inflamed tumor state), antitumor activity can be blunted by regulatory elements that suppress immune cells. These inhibitory molecules include of course CTLA‐4 and PD‐1 themselves, which can be expressed not only in effector T cells but also in other immune cells [64]. CTLA‐4 is constitutively expressed on regulatory T cells (T_regs), which diminish the immune response [64]. Depletion of these T_regs has been correlated with greater clinical benefit in patients treated with ipilimumab [65], [66], and therefore changes in T_reg levels can serve as an on‐treatment predictor of response. Factors that influence T_reg activity, meanwhile, might represent markers of resistance. For example, soluble CD25, an interleukin‐2 receptor whose binding has been hypothesized to stimulate T_reg proliferation, was reported as a negative correlate to overall survival with CTLA‐4 blockade [67].

Several other cell types have also been described as exerting an immunosuppressive effect [68]. Myeloid‐derived suppressor cells (MDSCs) negatively regulate immune activity in cancer, and increased circulating MDSCs have been identified as a poor prognostic factor [69], [70]. Tumors resistant to immune checkpoint inhibitors in mouse models became susceptible to therapy when treated with drugs that reduced MDSCs [71]. Among patients treated with ipilimumab for melanoma, higher frequencies of MDSCs were associated with poor outcomes [72].

Thus, multiple regulatory pathways act independently from PD‐1 and CTLA‐4 to abet tumor resistance. One biomarker that has raised considerable interest from a therapeutic standpoint is indoleamine 2,3‐dioxygenase (IDO) [61]. This enzyme suppresses T‐cell function by catabolizing the essential amino acid tryptophan [73] and is notably activated in dendritic cells after engagement with CTLA‐4 [74]. Histological studies from a phase II trial of ipilimumab in melanoma noted decreased baseline expression of IDO in poorly responding tumors [59]. Because IDO is upregulated by CTLA‐4 during adaptive immune resistance, a low IDO level might signify a lack of suppressed TILs available to be reactivated by immunotherapy.

Mutations in Janus Kinase 1 and 2 and Interferon‐γ Pathway

Tumor cells can also exploit genetic alterations in order to resist antiproliferative signaling by immune cells. The interferon‐γ (IFN‐γ) signaling pathway has been recognized as a critical component of immunotherapy. During adaptive immune resistance, tumors will upregulate PD‐L1 expression in direct response to IFN‐γ production by TILs [29]. IFN‐γ signaling also mediates many of the antitumor actions of immune cells. Increased expression of IFN‐γ‐inducible genes, including chemokines that promote T‐cell infiltration and activation, has been observed after anti‐CTLA‐4 therapy [75]. Meanwhile, defective mutations in the IFN‐γ pathway were found in 9 of 12 melanoma tumors resistant to ipilimumab, and increased copy‐number alterations in IFN‐γ was associated with poor response to ipilimumab [76]. These data together imply that reduced IFN‐γ signaling can lead to primary resistance in tumors.

The IFN‐γ signaling pathway contains the enzymes Janus kinase (JAK) 1 and 2, which act downstream of IFN‐γ [29]. In melanoma cell lines, lack of PD‐L1 upregulation in response to IFN‐γ was traced to mutations in either IFN‐γ receptor 1 or JAK 1/2 [77]. Genetic analysis of nonresponders to anti‐PD‐1 therapy with high mutational loads revealed inactivating JAK 1/2 mutations in one case of melanoma and one case of MMR‐deficient colon cancer [78]. These mutations led to decreased signaling via IFN‐γ and resulted in negative PD‐L1 tumor expression. JAK inactivation thus may be useful as a resistance biomarker, and mutations in other immune signaling effectors should be explored as potential contributors to primary resistance.

Other Biomarkers of Primary Resistance

A plethora of other mechanisms have been implicated in immunotherapy resistance, although these all require further study. The growing acknowledgment of the influence of gut microbiota on the immune system has opened up new avenues of study. Resistance to ipilimumab was recently linked to the enrichment of Bacteriodes species in a cohort of 26 patients with metastatic melanoma [79], and oral administration of Bifidobacterium was synergistic with PD‐1 inhibition in mouse models [80]. Despite these encouraging results, the optimum microbiota composition remains in question. In contrast to the above study, an analysis of 25 patients treated with CTLA‐4 blockade for melanoma showed increased outgrowth of Bacteriodes species, and fecal transfer of these bacteria was actually associated with positive outcomes in mice [81]. Evidence is mounting in favor of the strong influence of the intestinal microbiome on immune function, but more clinical data are clearly needed in order to direct therapeutic options.

Other biomarkers that have evidence in the clinical setting include vascular endothelial growth factor (VEGF). Elevated levels of VEGF were associated with decreased overall survival in patients with melanoma treated with ipilimumab [82], and VEGF has been linked to both decreased T‐cell infiltration and immunosuppressive effects [83]. Loss of phosphatase and tensin homolog (PTEN) has emerged as another potential biomarker of resistance and was correlated with greater tumor reduction in patients with melanoma treated with anti‐PD‐1 therapy [84]. PTEN has been associated with greater PD‐L1 positivity on tumor cells [85], as have alterations in a number of oncogenic drivers [86], [87], [88], [89]. Mutations in another family of tumorigenesis drivers, the serpins (including the genes SERPINB3 and SERPINB4), were associated with increased survival after anti‐CTLA‐4 therapy for melanoma, although the exact mechanism remains unclear [90]. Tumor hypoxia has also raised interest as a resistance factor, and signals activated by hypoxic environments were associated with worse clinical outcomes in mice treated with PD‐1 blockade [91]. Metformin has been proposed as a method of reducing oxygen consumption and improving susceptibility to anti‐PD‐1 treatment and is being studied in clinical trials [92]. Lastly, one particularly concerning category of resistance involves patients who exhibit “hyper‐progression” after immunotherapy [93]. Amplifications of the E3 ubiquitin‐ligase protein murine double minute 2, which inhibits the p53 tumor suppressor, have been associated with these hyper‐progressor phenotypes [94], although the mechanism of such an effect remains unknown.

Acquired Resistance to Immunotherapy

Even with an excellent initial response to immunotherapy, patients are at risk of subsequently relapsing because of acquired resistance. This issue is drawing increased attention as clinicians gain longitudinal experience with immune checkpoint inhibitors.

Loss of Major Histocompatibility Complex Class I Expression

Some of the earliest examples of immunotherapy failure were attributed to acquired defects in major histocompatibility complex (MHC) class I molecules [95]. These patients were found to have tumor cells that specifically lacked functional beta‐2 microglobulin (B2M), a protein necessary for CD8‐mediated T‐cell recognition. Biopsies obtained prior to immunotherapy exhibited normal B2M levels, pointing toward an acquired escape mutation that allowed tumor cells to evade immune recognition. Other mechanisms of MHC class I loss have since been reported, including the downregulation of transporter associated with antigen processing 2 and low‐molecular‐weight protein 7 in MSI‐negative colorectal tumors [96]. Direct loss of the gene encoding the human leukocyte antigen (HLA)‐C*08:02 class I molecule was recently described in a colorectal tumor that progressed after initial response to T‐cell transfer therapy [97]. These T cells were HLA‐C*08:02‐restricted TILs, and thus this mutation directly allowed immune evasion by tumor cells.

There is evidence to suggest that even with intact immune‐antigen recognition, tumors can avoid detection by reducing expression of these antigens. In one study, NSCLC tumors that progressed on immune checkpoint blockade were shown to have developed a loss of 6–18 neoantigens via chromosomal deletions [98]. These neoantigens were able to elicit T‐cell responses during in vitro assays, implying an integral role in facilitating the effects of immunotherapy. Similarly, another report described dedifferentiation and resulting antigen loss in melanoma cells as a method of acquiring resistance to adoptive T‐cell transfer [99]. These examples together reinforce a theme of tumors developing resistance by limiting the antigens available for recognition by immune cells.

IFN‐γ/JAK Pathway Mutations

As with primary resistance, tumor cells may attain genetic or molecular alterations that inhibit downstream effects of immune signaling. An important clinical report analyzed the molecular characteristics of four patients who responded to pembrolizumab and subsequently relapsed [100]. One patient had tumor cells that exhibited a B2M mutation leading to loss of MHC I expression, as described above. Notably, one patient had homozygous mutations in JAK 1 (Q503* nonsense mutation) and another had homozygous mutations in JAK 2 (F547 splice‐site mutation), both of which were predicted to lead to nonsense‐mediated decay or truncation of the protein prior to the active kinase domain. Tumors of both patients exhibited decreased downstream signaling in response to IFN‐γ. In conjunction with the previously described data on JAK 1/2 inactivation in primary resistance, these data establish IFN‐γ/JAK pathway aberrations as a route that tumors can exploit in order to avoid the effects of immunotherapy. Of concern, these mutations may exist at high frequency in tumors, with pretreatment melanoma biopsies demonstrating IFN‐γ pathway mutations in as many as 19% of samples [101].

Activation of Alternative Immunosuppressive Pathways

Tumors have also been shown to utilize alternative immune checkpoints that can take over immunosuppressive functions when CTLA‐4‐ or PD‐1‐dependent pathways are blocked. For example, T‐cell immunoglobulin mucin 3 (TIM‐3), which has been previously described as a marker of T‐cell exhaustion [102], was found to be upregulated in two patients who developed resistance to anti‐PD‐1 therapy [103]. In a mouse model of TIM‐3‐mediated resistance to anti‐PD‐1 therapy, administration of anti‐TIM‐3 antibodies resulted in enhanced therapeutic efficacy [103]. In addition, the combination anti‐PD‐1 and anti‐TIM‐3 therapy has shown synergistic effects in preclinical studies [104]. Other immune checkpoints might prove similarly useful in identifying resistance. For example, lymphocyte‐activation gene 3 (LAG‐3) has been shown to coexpress with PD‐1 as an immune suppressor [105], and anti‐LAG‐3 plus anti‐PD‐1 therapy resulted in enhanced tumor regression in mouse models [106].

Overall, mechanisms of acquired resistance remain poorly understood, given the paucity of clinical data on the subject. It can, however, be hypothesized that many of the mechanisms involved in primary resistance are able to develop later in the treatment course. This is especially true of genetic mutations, such as those in JAK 1/2. In another example, biallelic loss of PTEN was recently found in a patient with uterine leiomyosarcoma who acquired resistance to PD‐1 blockade [107]. As trials continue to come forth on the efficacy of immunotherapies, additional markers of acquired resistance should be identified. It is telling that in the above‐mentioned study by Zaretsky and colleagues, the authors were unable to identify an explanatory mutation in one of the four relapsed patients [100], exemplifying the shortcomings in our current understanding.

Discussion and Practical Application

T‐cell‐based immunotherapy is an exciting treatment modality that holds great promise but remains limited in part by our incomplete understanding of resistance. Given the vast genetic and molecular variability among and within tumors, it is not surprising that identifying definitive resistance biomarkers has been elusive. However, several candidates discussed in this review have shown promise and accumulated some clinical evidence as predicting primary (Table 1) and acquired (Table 2) resistance. Certainly, FDA‐approved biomarkers for PD‐L1 status and MMR proficiency should be taken into account in all applicable cases, despite lingering concerns over issues such as assay consistency. These flaws instead highlight the fallibility of single tests and underscore the need for comprehensive tumor analysis and patient assessment in order to optimize treatment decisions.

Table 1. Clinical evidence for biomarkers of primary resistance to immunotherapy.

Ranges in parentheses indicate 95% confidence interval.

Abbreviations: CNA, copy number alterations; EIA, enzyme immunoassay; IDO, indoleamine 2,3‐dioxygenase; IFN‐γ, interferon‐γ; IHC, immunohistochemistry; JAK 1/2, Janus kinase 1 and 2; MB, ; MDM2, murine double minute 2; MDSC, myeloid‐derived suppressor cell; MMR, mismatch repair; MSI, microsatellite instability; NSCLC, non‐small cell lung cancer; ORR, objective response rate; PBMC, peripheral blood mononuclear cell; PCR, polymerase chain reaction; PD‐1, programmed cell death protein 1; PD‐L1, programmed cell death ligand 1; PFS, progression‐free survival; Ref., reference; TIL, tumor‐infiltrating lymphocyte; TTF, time to treatment failure; VEGF, vascular endothelial growth factor; WES, whole exome sequencing.

Table 2. Clinical evidence for biomarkers of acquired resistance to immunotherapy.

Abbreviations: B2M, beta‐2 microglobulin; CTLA‐4, cytotoxic T‐lymphocyte‐associated protein 4; JAK 1/2, Janus kinase 1 and 2; MHC, major histocompatibility complex; NSCLC, non‐small cell lung cancer; PD‐1, programmed cell death protein 1; RT‐PCR, reverse transcription polymerase chain reaction; TIM‐3, T‐cell immunoglobulin mucin 3; WES, whole exome sequencing.

Multifactorial Resistance Biomarkers

Several investigators have proposed combining multiple biomarkers to arrive at a more accurate predictor of treatment resistance [108], [109]. For example, both inadequate immunogenicity and a noninflamed tumor milieu have accumulated substantial evidence as contributing to negative prognostics with immunotherapy. Neither of these measures is perfectly predictive of response [110], an issue that may be ameliorated by aggregate measures. In one analysis of melanoma samples, an immune signature termed innate anti‐PD‐1 resistance, or “IPRES”—consisting of resistance‐associated genes involved in mesenchymal transition [111], matrix remodeling [112], and angiogenesis [83]—was associated with improved response to PD‐1 inhibitors [113]. Another study found that increased copy number alterations and low mutation load were nonredundant predictors of poor response, again suggesting the benefit of a combinatorial biomarker [114].

Multifactorial assays have also been shown to be useful for longitudinal monitoring [115]. An immune signature utilized as an early on‐treatment marker was highly predictive of response to immunotherapy [115]. Ultimately, effective biomarker analysis may entail the monitoring of multiple genetic and molecular factors in a longitudinal manner, thereby providing comprehensive and dynamic information regarding response to treatment. This would unfortunately require sequential biopsies, a logistical challenge that could be circumvented by using markers in peripheral blood. For instance, a score using on‐treatment levels of exhausted‐phenotype T cells that had been “reinvigorated” was found to correlate with patient response to PD‐1 blockade [116]. A variety of other blood‐based immune cell markers can be measured to assess patient response [117], but MDSC level represents one of very few such biomarkers that can be predictive prior to treatment. Identification of peripheral blood biomarkers thus represents an area greatly in need of further investigation.

Loss of MHC class I expression represents a particularly problematic resistance mechanism, as it can negate the ability of the immune system to act on tumor cells.

Treatment Strategies in Response to Resistance

The search for therapies that circumvent immunotherapy resistance has been progressing in parallel to mechanistic inquiries. Multiple targets described in this review have shown promise both as biomarkers of resistance and as targets of combination therapy. The most visible example of combination therapy in this field involves the dual targeting of PD‐1 and CTLA‐4. A clinical trial of patients who progressed on anti‐CTLA‐4 therapy showed improved response to nivolumab as compared with chemotherapy [118], suggesting that primary resistance can be overcome by targeting related pathways. Subsequent studies in melanoma have in fact demonstrated a synergistic effect of anti‐PD‐1 and anti‐CTLA‐4 treatment [36], [119], [120]. Ongoing clinical trials of combination therapies include those targeting immunosuppressive factors, such as IDO [121]. Alternative checkpoint molecules like TIM‐3 have shown efficacy in preclinical combination studies and are pending clinical evaluation [122]. Emerging data on oncogenic driver mutations in immunotherapy have additionally justified several exploratory trials for dual targeting [22]. Forthcoming results could confirm these combination treatments as prime methods for combating resistance.

Mechanisms of acquired resistance have only recently begun to be elucidated and will become increasingly relevant to clinical practice. Loss of MHC class I expression represents a particularly problematic resistance mechanism, as it can negate the ability of the immune system to act on tumor cells. In more fortunate cases, the emergence of acquired resistance can be localized to a single metastasis or recurrent tumor. In a case series of 36 patients who developed acquired resistance, 15 patients continued to respond well to immunotherapy after treating a solitary resistant lesion with surgery and/or radiation [123]. In situations of more widespread resistance, chemotherapy may be indicated unless novel methods of reactivating MHC class I expression are discovered.

Genetic mutations in signaling cascades, such as those in the IFN‐γ/JAK pathway, at least offer the ability to intervene at other steps. Inactivating mutations in JAK 1 or JAK 2, for example, might be overcome by stimulating downstream or parallel pathways. The stimulator of interferon genes (STING) pathway, for example, has been shown to upregulate type I IFN production [124] and may additionally play a role downstream of JAK 1/2 [125]. Activation of STING has also been shown in mouse models to induce an inflammatory microenvironment and cause tumor regression [126]. Such methods of immune enhancement may be helpful in converting non‐T‐cell‐inflamed tumors into inflamed ones that are more conducive to immunotherapy [22]. A variety of other techniques have been proposed to promote this transition to an inflamed tumor, including epigenetic modifications that reduce T‐cell exhaustion [127] and the use of BRAF inhibition to improve tumor antigen recognition [128] and promote a more favorable microenvironment [129].

Final Recommendations

Although biomarker analysis remains in its infancy, it would still be appropriate to assess any patients being considered for treatment with immunotherapeutic agents (Fig. 3). Candidate patients should be evaluated for PD‐L1 status and MMR status; if sufficient resources exist, they can also be assessed for mutation load, TIL levels, IFN‐γ/JAK pathway mutations, and biomarkers relevant to any indications or trials for combination therapies. Re‐evaluation should occur periodically and with any evidence of acquired resistance. Ultimately, a tumor profile that meets criteria for multiple resistance biomarkers should be deemed unlikely to respond to simple immune checkpoint blockade, and alternative options should be discussed. Treatment regimens in such a situation should depend on whether resistance involves (a) deficits in tumor immunogenicity and inflammation requiring general immune enhancement, (b) coexisting molecular markers that can be targeted with combination therapy, or (c) compromise of immune pathways to an extent that would preclude immunotherapy as an appropriate treatment modality. If possible, patients who do not respond to treatment should be evaluated for causative mechanisms in order to inform future decisions and to further develop our understanding of immunotherapy resistance.

Figure 3.

Proposed evaluation of prospective candidates for immunotherapy. Candidate patients should undergo an initial evaluation for resistance biomarkers (A). A clinical profile consistent with a noninflamed tumor microenvironment may warrant enhancement of immune activity. Biomarkers that can be targeted would suggest benefit from combination therapies. Primary resistance might also be suggested by negative prognostic indicators during early on‐treatment (B). Patients whose disease progresses after initial response should receive workup for acquired resistance (C). These resistance mechanisms can similarly be addressed by enhancing the immune response or targeting immunosuppressive pathways with combination therapy. Finally, loss of MHC‐I may prohibit further use of immunotherapy and indicate alternative modalities.

Abbreviations: B2M, beta‐2 microglobulin; HLA, human leukocyte antigen; IDO, indoleamine 2,3‐dioxygenase; JAK, Janus kinase; LAG‐3, lymphocyte‐activation gene‐3; MDSCs, myeloid‐derived suppressor cells; MHC‐I, major histocompatibility class I; MSI, microsatellite instability; PD‐L1, programmed cell death ligand 1; STING, stimulator of interferon genes; TAP2, transporter associated with antigen processing 2; TIL, tumor‐infiltrating lymphocytes; TIM‐3, T‐cell immunoglobulin mucin‐3; T_reg, regulatory T cells; VEGF, vascular endothelial growth factor.

Conclusion

Cancer immunotherapy, although only recently implemented as a foundation of treatment, has already yielded exciting results and holds significant promise for the future. As mechanisms of resistance are more comprehensively revealed, clinicians may be able to more effectively treat patients based on their specific genetic and molecular characteristics. Finding biomarkers that can predict patient response to immunotherapy remains a primary objective for investigators, although clinicians should approach any single marker with caution, given the complexity of tumor biology. Nevertheless, the identification and proper use of resistance biomarkers should remain a focus of ongoing study in order to extract the greatest benefit from these treatments.

Author Contributions

Conception/design: Young Kwang Chae

Manuscript writing: Michael S. Oh

Final approval of manuscript: Young Kwang Chae, Michael S. Oh, Francis J. Giles

Disclosures

The authors indicated no financial relationships.