Tumor-specific antigens are the key to activate T cells to recognize tumor as foreign, which is the first step of tumor-induced adaptive immune responses and immune-mediated tumor killing (15). These neoantigens, interestingly, are derived from somatic mutations and contain new epitopes, and subsequently lead to tumor immunogenicity. Preclinical and clinical studies have revealed that the response of neoantigen-specific effector T cell (Teff) paralleled tumor shrinkage (16–20).

With the improvement of sequencing techniques, it was found that nonsynonymous mutations can generate neoantigens that trigger cytotoxic responses against tumors (21, 22). Nonsynonymous mutation burden, rather than total mutation burden of exons, was demonstrated to be more closely associated with the clinical advantage of anti-PD-1 treatment, validating the importance of neoantigens in dictating response (23). Tumor mutation burden (TMB) is calculated as the total number of nonsynonymous mutations per DNA Megabase (Mb) (21, 24, 25). Low TMB, or low numbers of clonal neoantigens, presenting reduced tumor immunogenicity, is considered as a primary resistance marker to immunotherapy (15, 26).

Clinically, low TMB or neoantigen load has correlated with inferior response and poor PFS to monotherapy of anti−PD1/PD-L1 antibodies in NSCLC (25, 27–30). However, it fails to predict the clinical outcomes, in regard to overall survival (OS) and combination regimens (31, 32). The influence on the OS by subsequent treatments and the additional complexities to the study of immunotherapy resistance added by combinations may partly explain these controversial findings. Recently, a corrected TMB (cTMB) approach based on the adjustment of tumor purity was developed by Anagnostou and colleagues, which was identified on abundant tumor samples mined from The Cancer Genome Atlas (TCGA) and then confirmed in a patient cohort received ICIs therapy. This cTMB more accurately predicted the outcomes of immunotherapy, suggesting that the TMB in samples with low tumor purity was mistakenly underestimated, which was especially important for metastatic NSCLC, because the tumor purity of tissue samples obtained by bronchoscopy or puncture biopsy was often limited (33).

The dilemma of insufficient tissue sample for TMB assessment in a considerable number of patients with NSCLC has given rise to the employment of peripheral blood TMB (bTMB) as a substitute predictor of response or resistance to ICIs in NSCLC (34). In keeping with what was previously reported in tissue, low bTMB evaluated by different plasma sequencing assays was significantly correlated with poor survival or response to immunotherapy in several retrospective and prospective studies (35–37).

In addition to the TMB or the numbers of clonal neoantigens, increased neoantigen intratumor heterogeneity (ITH, defined as relative fraction of subclonal neoantigens) can also impair the sensitivity to ICIs by elevating the likelihood of selection of subclones with poor immunogenicity (25, 38). The considerable variation of neoantigen heterogeneity was demonstrated by McGranahan and colleagues in seven primary NSCLCs (25). On average, 44% of heterogeneous neoantigens were reported only in a subset of tumor regions. They conducted neoantigen and clonality analysis in lung cancer data from TCGA and then validated the approach in a cohort of NSCLC patients treated with ICIs. Compared with high TMB alone, the combination of high TMB with low ITH seems to have a stronger association with clinical benefit to ICIs in this population.

Aberrations in oncogenes and tumor suppressor genes can regulate immune response by amending cytokine profile and immune cell composition and thus render tumor cells resistant or sensitive to ICIs.

Generally, alterations in oncogenic driver genes are characterized as resistant markers to immunotherapy. Although epidermal growth factor receptor (EGFR) mutations and anaplastic lymphoma kinase (ALK) rearrangement tend to have high PD-L1 expression due to the activation of signaling pathways (39, 40), the low mutation or neoantigen load (41), along with the following mechanisms, impairs the immunotherapy sensitivity in this group of patients with lung cancer. First, EGFR mutations have the potential to shape an inert immune environment by up-modulating a series of immune suppressors including inhibitory immune checkpoints (e.g., PD-1 and CTLA-4), immunosuppressive cells (macrophages and regulatory T cells), and cytokines (like TGF-β, IL-6, and IL-10) (42, 43). It has been reported that activated EGFR cascade was associated with elevated T-cell exhaustion and reduced cytotoxic T lymphocytes (CTLs) in a lung adenocarcinoma model (40). Second, downstream pathways of EGFR mutation, such as MAPK, PI3K/AKT, and Janus kinase (JAK)/STAT pathway, negatively affect immunoregulation. Other oncogenic driver-genes that frequently have high PD-L1 expression in lung cancer include ROS1 rearrangements (44), MET exon 14 skipping mutations (45), and BRAF mutations (44, 46). In contrast, RET rearrangements (47) and HER2 mutations (44) have been reported recently to exhibit low PD-L1 expression. None of these oncogenotypes demonstrated favorable clinical responses to ICIs monotherapy except for BRAF mutations, either V600E or non-V600E.

STK11 gene inactivation either by mutational or non-mutational machinery is linked to an indolent immune microenvironment with lower Tumor-infiltrating lymphocyte (TILs; CD3+, CD4+, and CD8+ cells) and PD-L1 expression in spite of the existence of moderate to high TMB (48). Inactivated STK11 gene was recently reported to weaken the innate immune responses by epigenetic inhibition of stimulator of IFN genes (STING), suggesting epigenetic silencing is likely to mediate the promotion of T-cell exclusion by the loss of STK11 (49). In line with these findings, it has been observed in several studies that compared with the wild-type gene, STK11 mutation predicted poorer clinical outcomes of immunotherapy in advanced NSLCL (50, 51). The tumor suppressor TP53 mutation, a well-known negative prognostic factor in lung cancer, is found to be associated with increased PD-L1 expression and higher TMB in non-squamous NSCLC (30) and KRAS-mutated lung adenocarcinoma (51).

KRAS-mutated lung cancer presents distinct immune profiles, biology, and therapeutic vulnerabilities in different subsets classified by co-occurring genetic events (50). Generally, KRAS/TP53 co-mutation predicts sensitivity while KRAS/STK11 co-mutation predicts resistance to immunotherapy in NSCLC. Dong et al. identified TP53/KRAS co-mutated subclass exhibited the highest percentage of PD-L1+/CD8A+ and particular increased PD-L1 expression. They further confirmed a remarkable clinical benefit from pembrolizumab in this population (52). Co-mutation of STK11 was shown to cause the accrual of neutrophils with T-cell-suppressive effects, accompanied with an analogous elevation in the production of T-cell depletion biosignatures and tumor-promoting cytokines (50, 53). TIL numbers and the expression of PD-L1 were also decreased (53). Consistent with these preclinical predictions, patients with KRAS/STK11 co-mutation or single mutation of STK11 had poor response and survival compared with those with wild-type when treated with ICIs (51, 54).

The Kelch-like ECG-associated protein 1 (KEAP1) gene, regulating the cellular antioxidant and cytoprotective transcriptional programs, plays a key role in mediating immune evasion in NSCLC. Depletion of KEAP1 is associated with reduced leukocyte infiltration, increased PD-L1 expression and might also influence other immune cells such as NK cell recruitment and function (55, 56). Co-occurring KEAP1 and phosphatase and tensin homolog (PTEN) inactivation represent an immunologically “cold” tumor while concurrent mutations in KEAP1 and STK11 leads to absence of pro-cancerogenic M2 macrophages (57). However, there are conflicting data on the role of KEAP1 mutation and its co-mutation with STK11 in immunotherapy resistance in NSCLC. KEAP1/STK11 co-mutations were verified to correlate with resistance to ICIs in patients with NSCLC despite high TMB (58). Similarly, STK11, and/or KEAP1 genomic variations posited lack of clinical advantages from combination of immunotherapy with chemotherapy in patients with NSCLC (59). However, inconsistent results were reported recently that clinical benefit from pembrolizumab compared to chemotherapy was poorer in the patients with STK11 and KEAP1 mutation compared with those in wild type in Keynote 042 trial, but the response and survival to immunotherapy were not significantly different between mutant and wild subgroups (60).

The WNT/β-catenin pathway is an additional immunotherapy resistance mechanism. A negative relationship was demonstrated between the level of β-catenin and TILs, which was modulated by deficiency in the recruitment of CD103+ dendritic cells (DCs) essential to T-cell priming and reduced expression of the cytokine CCL4, suggesting WNT/β-catenin signaling pathway is likely to mediate ICIs resistance through T-cell exclusion (61).

Similarly, the MAPK/PTEN/PI3K signaling pathway has been identified to be involved in immunotherapy resistance. Loss of PTEN and the bioactivation of the phosphatidylinositol 3-kinase (PI3K) signaling pathway in tumors decrease the activity of CTLs through the recruitment of inhibitory cells to the microenvironment and Vascular endothelial growth factor (VEGF) expression (62, 63), so that they promote resistance to ICIs (63, 64). The association of PTEN deletions or PIK3CA/AKT mutations with increased PD-L1 expression and immune resistance was also found in glioma (65). It was shown in preclinical models that a PI3K-γ inhibitor decreased myeloid-derived suppressor cells (MDSCs) and improved response to ICIs (66).

Genetic instability caused by alterations in DNA replication and repair genes can augment immunogenicity via a high-TMB neoantigen load (67–69). Correspondingly, deficient DNA mismatch repair (dMMR) or high microsatellite instability (H-MSI) are suggested as sensitive predictors to ICI immunotherapy in many tumor types. Beyond high TMB, increased CD8+ TILs were also reported to be associated with alterations in mismatch repair genes (70), BRCA2 (71), and POLE (72) in different tumors. However, the role of these genes in immunoregulation in NSCLC remains to be elucidated.

The interferon-gamma (INF-γ) signaling cascade is a crucial component of immunotherapy and tends to serve a critical function in primary, adaptive, and acquired resistance to ICI treatment (73–75). IFN-γ is a critical cytokine secreted by activated T cells, natural killer (NK) T cells, in the cancer microenvironment, and it moderates the immune reaction via the downstream enzymes JAK 1/2 and the signal transducer and activators of transcription (STATs) (76). The INF-γ axis exerts both positive and negative impacts on antitumor immune reactions (77). On one hand, it activates an functional antitumor immune reactive via (1) intensifying antigen presentation by up-modulated secretion of MHC-I; (2) recruiting other immune cells by up-regulation of the expression of chemokines (CXCL9, CXCL10, and CXCL11) with effective chemoattractant impacts on T cells (78); and (3) exerting direct anti-proliferative and pro-apoptotic impacts on cancer cells (79). On the other hand, IFN-γ acts in a negative-feedback axis to elevate PD-L1 expression as well as other crucial immune inhibitory components, including IDO1, down-modulating the cytotoxic reaction and adaptive resistance to cancer cells (80, 81) (Figure 1A).

Additionally, copy-number alterations (CNAs) linked to DNA damage response and regulation of DNA editing/repair gene expression were shown to emanate from the malignant exposure to IFN-γ-secreting antigen-specific CTLs in vivo, implying that intensified genetic instability could be among the mechanisms through which CTLs and IFN-γ immunoedits cancers, changing their immune resistance due to genetic evolution (82).

Tumors neutralize the impact of IFN-γ by mutating or down-regulating the molecules involved in the IFN-γ signaling pathway, including IFN-γ receptor chains, regulatory factors, JAK1/2, and STATs upon continuous IFN-γ exposure (73, 83). Multiple studies have demonstrated that mutations of IFN-γ axis and consequent loss of JAK/STAT contribute to immune escape of tumor cells and by that leads to primary or acquired resistance to ICI therapy via incapacity of up-regulating the expression of PD-L1 and MHC-I (73, 78, 84). Any deficiencies in IFN-γ, JAK1/2, or STATs including gene mutations, loss of protein expression, negative regulator presence, or epigenetic silencing would prevent signaling in response to IFN-γ and thereby end up to the up-regulated tumor growth and apoptosis inhibition and down-regulated T-cell infiltration and expression of PD-L1 and MHC-I (74, 78, 85, 86). Correspondingly, genomic changes disturbing IFN-γ pathway genes, including the amplification of suppressor genes PIAS4 and SOCS1 and the deletion of IFNGR1, IFNGR2, IFIT1, IFIT2, IFIT3, IRF1, MTAP, and miR31, have been described as possible machinery of primary resistance to various ICI therapies (73). An IFN-γ-related mRNA profile that contains 10 genes (CCR5, CXCL9, CXCL10, CXCL11, GZMA, HLA-DRA, IDO1, IFNG, PRF1, and STAT1) was additionally identified to predict the response to anti-PD-1 therapy in melanoma (87).

Loss of neoantigens in the context of immune-mediated pressure is postulated to be another mechanism leading to resistance. In the concept of immunoediting, the constant interactions between tumor cells and the immune system trigger the production of subclones that do not express neoantigens, consequently conferring poor immunogenicity and resistance to ICIs (117). It was demonstrated by Anagnostou and colleagues that seven to eight putative neoantigens were lost in the recurrent NSCLC after ICI treatment, suggesting that immunoediting plays a role in acquired resistance to immunotherapy (118). T-cell-mediated neoantigen immunoediting can be induced by the dynamic interactions between T cells and tumor cells, consequently causing partial or total loss of neoantigen (119). Consistently, deficiency in genes that encode target tumor antigens was demonstrated to be associated with acquired resistance in a murine model treated with adoptive T-cell therapy (ACT) in melanoma (120). However, this relationship between acquired resistance and loss of target neoantigens was not observed in a single patient case who achieved a complete response to ACT in a separate study (121), suggesting that down-regulation/loss of neoantigens may occur during immunotherapy, but should be taken as a canonical mechanism of acquired immune resistance.

Proinflammatory cytokines are likely to contribute to immune escape by inducing loss of antigen expression, resulting in acquired resistance too. The process of Tumor necrosis factor-α (TNF-α)-induced epithelial-to-mesenchymal de-differentiation was shown to lead to a loss of neoantigens causing transformation to a tumor phenotype that is less immunogenic and can more readily evade immune surveillance in the ACT-treated mouse model in melanoma (122). Other TIL generated cytokines, such as IL-6 or TGF-β, are also shown to be involved in the induction of epithelial-to-mesenchymal transition in mouse models across numerous types of tumors, indicating that acquired resistance can be promoted by inflammation.

Defective neoantigen presentation serves a crucial function in ICI acquired resistance. The alterations in this process could happen in beta-2-microglobulin (B2M), transporters associated with antigen processing (TAP), or MHC itself (123, 124) (Figure 1A).

As part of the MHC class I (MHC-I), B2M is crucial during antigen presentation and its genetic deficiency, including loss of heterozygosity (LOH) and deletions or point mutations, was identified to be an important route for primary and acquired resistance to ICIs (125, 126). Other defects that would affect neoantigen presentation include T-cell receptor (TCR) binding domain mutations of MHC-I (127), loss of tapasin (a MHC-I antigen processing molecule), selective epigenetic silencing of the human leukocyte antigen (HLA) A3 antigen, loss of one HLA haplotype (128, 129), and LOH in HLA (130). Homozygosity in one or more of the three highly variable genes (HLA-A, HLA-B, and HLA-C) that encode MHC-I, which are likely to restrict neoantigen presentation to CTLs, was identified to have a significant association between resistance to ICI therapy in a large cohort of cancer patients (131). In contrast to anti-CTLA-4, the expression of MHC-II (but not MHC-I) proteins by tumor and the presence of IFN-γ-mediated gene signatures were found to be associated with the positive responses to anti-PD-1 therapy in melanoma (132).

Defective neoantigen presentation may be mediated by IFN-γ signaling pathway through JAK1/2 and the STATs, by down-regulating the expression of MHC-I (133). Actually, the IFN-γ pathway has both unfavorable and favorable impacts on antitumor immune responses and plays a key role in acquired and primary resistance to ICI therapy (as discussed above).

Tumor-infiltrating lymphocytes constitute a complex group of immune cells with distinct functions and different clinical impacts. Among them, tumor-specific CD8+ T cells can execute anti-cancer activities by killing tumor cells directly and has a strong prognostic effect in NSCLC (134, 135). CD4+ cells are composed of a group of lymphocytes (Tregs CD4+, Th1, Th2, and Th17) secreting diverse cytokine to activate and suppress CD8+ cells. Th1 secretes IFN-γ and IL2, while Th2 secretes IL-4, IL-5, IL-9, IL-10, IL-13, and IL-25 (136, 137). CD45RO+ T cells, also known as memory T lymphocytes, are another subclass of TILs. Regulator and memory T lymphocytes will be discussed in section “Suppressive tumor microenvironment.”

Low CD8+ TIL density was correlated with impaired efficacy and survival in NSCLC patients treated with ICIs (138), suggesting that immunotherapy resistance was mediated by low TILs but was then positively modulated by PD-L1. TILs can be assessed by immunohistochemistry or standard hematoxylin and eosin (H&E) staining; however, no consensus has been reached hitherto in the various scoring models using H&E staining in NSCLC (139–142). A radiomic fingerprint of CD8+ TIL derived via computerized tomography was developed recently and showed promising efficacy in predicting response to ICI therapies but requires further validation (143).

Thus, tumors can be described as three main immune organization profiles (hot, altered, and cold) as per the presence of TILs and correlated proinflammatory cytokines (144). The “cold” immune tumor is characterized as absence of TIL within and at the edges of tumor, manifesting resistance to immunotherapy either due to absent immune stimulation (as with low neoantigen cancer’s poor antigen presentation) or because of failed T-cell priming (as with intrinsic insensitivity to T-cell killing). The “altered” immune tumor is characterized as low TIL within the tumor (“immunosuppressed”) or high TIL at the edges of the tumor (“excluded”), whereas “hot” is high degree of TIL (144). Recently, intratumorally geospatial heterogeneity of TIL was revealed in NSCLC. Tumor subclones from “cold” immune regions were related to mutation space more closely and diversifying more recently compared with those from “hot” immune regions. Higher risk of recurrence was observed in tumors with more than one “cold” immune region (145).

Reduced proliferation and inadequate diversification of T cells possibly contribute to ICI resistance. Impeded priming of naive T cells by blocked DCs recruitment was demonstrated in melanoma to be correlated to the lack of TILs and ICIs resistance (146). The function of DCs can be potentially influenced by the cytokines in the TME through (1) impaired migratory capacity as well as decreased synthesis of costimulatory components (CD86/80) by TGF-β (147, 148); (2) prevented DCs maturation by IL-6-gp130-STAT3 axis; and (3) inhibited activity by Indoleamine 2,3-dioxygenase 1 (IDO, will be discussed in Section “Suppressive tumor microenvironment”). IFN-α signaling pathway is important to the priming of T cells by DCs. It was found that TME with remarkable insufficient IFN-α-producing DCs naturally led to lessened antitumor T-cell priming and thus resistance to ICIs (149, 150). Activated IFN-α stimulated production of the chemokine CXCL10 to recruit TILs to tumor beds and in turn initiate spontaneous antitumor T-cell response (149–151). Preliminary trials combining IFN-α 2b therapy with anti-CTLA-4 inhibitors have indicated clinical activity, which could be caused by diminished populations of MDSC (152, 153). Combinations of other ICIs and IFN-α 2b are currently investigated (154).

Immune resistance also occurs if the tumors evolve the ability to prevent infiltration even if tumor-specific Teffs are formed. Mechanisms that lead to impaired T-cell infiltration involve components in the epigenetic silencing of immune cells (155) and the modification of secreted chemokines (156, 157). Transcriptional program that is associated with T-cell exclusion and thereby predictive resistance to anti-PD-1 therapy was identified in melanoma (158). Stromal cells surrounding tumors within TME can develop the capacity to obstruct effector T-cell entry, and the TGF-β cascade appears to serve a crucial role in promoting T-cell exclusion features in peritumoral fibroblasts (123, 159).

T-cell receptor clonality is emerging as a new biomarker to predict the resistance and immune-related adverse events to ICIs therapy. Since baseline CD8+ T-cell density was found to overlap between respondents and non-respondents to ICI therapy (160–162), it led to the speculation that a constrict TCR arsenal possessed by the baseline T cells concentrated on the antitumor immune reaction and is associated with response to ICI therapies. T-cell clones can be identified by detecting TCR rearrangements constituting genes in the variable (V)-diversity (D)-joining (J) region, which generate the antigen-specific complementarity-determining region 3 (CDR3). The responsivity of TCRs generated by TILs determines their potential to interplay with tumor antigens that are presented on APCs. Thus, the assessment of T-cell clonality divulges the extent of T-cell expansions caused by tumor antigens and contributes to explore the mechanisms underlying T-cell toleration to tumor antigens.

A lower baseline clonal T-cell arsenal has been shown to be linked to worse clinical benefits to ICIs and survival in cancer patients (162, 163). Besides, a remarkable increase in T-cell clones was reported in responders during anti-PD-1 therapy compared to non-responders, implying a cancer-specific reaction to immunotherapy for these patients. Moreover, baseline TCR clonality did not strongly associate with TIL density, implying that low-TIL density tumors could still respond to anti-PD-1 treatment if TIL has a narrow TCR clonality specific to the tumor antigen (164). Inconsistently, it was recently found that T-cell clonality had a positive relationship with T-cell density, PD-L1 expression, and TMB, and a negative relationship with EGFR mutation in NSCLC (165). A corresponding relationship was found between the number of TCR sequences and the number of nonsynonymous mutations, spatial heterogeneity in expanded TCR repertoire, and spatial mutational heterogeneity within tumors in NSCLC, respectively. This intratumorally spatial heterogeneity of TCR repertoire maps the neoantigen landscape, sculptured by focal antigen processing defects or HLA loss (166). Thereby, further investigations to identify the role of TCR clonality in immunotherapy are required.

Compensatory up-regulation of numerous alternate immune checkpoint receptors during ICI therapy as a result of the activation of diverse cellular signals and IFN-γ signaling pathway were observed across multiple studies and have been characterized to be linked to ICI adaptive resistance in NSCLC (84, 167, 168). The expression of CD8+ T cells harboring receptors showed serious flaws in proliferation, migration, and cytokine secretion, indicating their immunosuppressive capacity. In addition, progressive T-cell exhaustion was found in the tumors with highly expressive or co-expressive receptors and different receptor displayed different exhausted phenotype (167).

Among these receptors, lymphocyte activation gene 3 (LAG-3) has great potentiality in cancer immunotherapy because co-expression of PD-1 and LAG-3 was often found in T-cell-depleted immune microenvironment, and PD-1 inhibitors combined with LAG-3 blockades showed strong synergic antitumor responses in preliminary models (169). LAG-3 is a co-inhibitory receptor extensively expressed in TILs in various tumors and serves a crucial function in mediating immune escape by suppressing T-cell antitumor functions. It exerts immunosuppression via binding to MHC-II molecules and other ligands such as galectin-3 and fibrinogen-like protein 1 (FGL1) (170). Thus, blocking LAG-3 can restore antitumor immunity and the combined LAG-3 inhibitors therapy may accordingly overcome immunotherapy resistance. In addition, the expression of these ligands, on the basis of the receptor–ligand interactions, may serve as important biomarkers to predict the efficacy of LAG-3 blockades in lung cancer (167).

The other alternative immune checkpoint receptors, e.g., T-cell immunoglobulin and mucin-3 (TIM-3), V-domain immunoglobulin-containing suppressor of T-cell activation (VISTA), B and T lymphocyte attenuator (BTLA; also referred to as CD272), T-cell immunoreceptor tyrosine-based inhibition motif domain (TIGIT), and sialic acid-binding Ig-like lectin 9 (SIGLEC9), have been discovered (144). Thus, these alternate immune checkpoints are likely to be combined with existing ICI therapy to conquer the resistance. Increased efficacy of PD-1 inhibitors combined with anti-TIM-3 or anti-LAG-3 regimens has been observed in either pre-clinical models or phase I clinical trials (171, 172). Currently, numerous clinical trials evaluating the therapeutic impact of alternate immune checkpoint blockade applied on its own or in combination with PD-1/PD-L1 inhibitors in multiple malignances are ongoing.

T-cell exhaustion is another factor involved in the primary and acquired resistance to ICI therapy (Figure 1C). Exhausted T cells exhibit impaired activity with progressive LOF and antigen persistence compared with Teffs and can be induced by the PD-1/PD-L1 interactions (173). Chronic exposure to cognate antigen triggers increased expression of PD-1, which results in the accumulation of T-cell exhaustion and thus T-cell dysfunction (174). The presence of PD-1 high expression can either exist prior to PD-1 inhibitors, which is associated with primary resistance partially depending on tumor-associated regulatory T cells (Tregs), or develop after the anti-PD-1 therapy, which leads to acquired resistance by severe T-cell exhaustion. In contrast, studies showed that the exhausted T cells with PD-1 low to intermediate phenotype retain the capacity to be reinvigorated by ICIs (158, 175). Epigenetic alterations were found to be associated with T-cell exhaustion too recently. Exhausted T cell displayed a unique chromatin landscape, which alters the transcriptional state, limits its effect function, and determines its capacity to be reprogrammed after therapeutic intervention (176–178).

The formation of memory T cells is crucial to the avoidance of tumor relapse and therapy resistance following drug withdrawal, especially in the long-lasting duration of responses to ICI therapy. Research evidence shows that patients with resistance to anti-PD-1 treatment have fewer tumor-correlated memory T cells relative to sensitive patients (179). Memory T cells remain dormant until antigen re-challenge (180, 181) and if precursor memory T cells are exhausted under chronic antigen exposure, it will lead to memory T-cell deletion and lack of formation (173, 177).

Acquired resistance can be mediated by the alteration from cytotoxic activity to inactivity phenotype of antitumor T cells during TCR-engineered ACT. The original highly cytolytic profile when administrated to patients, which showed strong efficacy initially, was reported to change to a phenotype with impaired cytotoxic functions and Th2-related cytokine release when tumor relapses within months (182, 183).

The TME is a complex net consisting of a variety of immune and stromal cells, cytokines, extracellular matrix, and vasculature, which affect response to immunotherapy. Immune-suppressive cells, including Tregs, MDSCs, M2 macrophages, along with inhibitory cytokines in the TME, can contribute to the inhibition to immune responses (136, 184) (Figure 1C).

Tregs can inhibit Teff reactions by secreting certain inhibitory cytokines (IL-10, IL-35, and TGF-β) or by direct cell contact (185–187). The cytokine IL-10 influences antigen presentation by down-regulating the expression of MHC-II and co-stimulatory components on DCs, thus intercepting the Teff activation (187). The ratio of Teffs to Tregs was shown to be related to the responses to ICIs in mouse models, in that incapacity of either increasing Teffs or decreasing Tregs may cause resistance to immunotherapy (188, 189). Factors that affect Tregs activity, at the same time, are putative biomarkers of resistance. For instance, soluble CD25, an IL-2 receptor whose binding is assumed to stimulate Treg proliferation, was established as a negative predictor of OS for patients treated with anti-CTLA−4 (190). However, tumor-infiltrating Tregs might likely coexist with multiple immune cells, insinuating a potential immunoreactivity. It was reported that a high baseline expression of FoxP3+ Tregs in the tumor is positively associated with better survival in a retrospective study involving patients under the treatment of anti-CTLA-4 antibodies (161).

Myeloid-derived suppressor cells promote immune evasion and tumor growth and have emerged as critical modulators of immune responses in cancer. Studies have suggested the existence of MDSCs in TME correlates with reduced efficacy of immunotherapies, including ICIs therapy (191), ACT (192), and DC vaccination (193). Therefore, reprogramming or eradicating MDSCs might improve clinical response to immunotherapy.

Tumor-associated macrophages (TAMs) can be classified into M1 and M2 macrophages according to disparities in surface molecules, expression of transcription factors, cytokine profiles, and metabolism (194, 195). They promote antitumor immunity effects (mediated by M1) and pro-tumorigenic properties (mediated by M2) that modify the TME (196). The role of TAMs in mediating immunotherapeutic resistance in tumor has been discussed in several reports (197, 198). It was indicated to directly inhibit T-cell responses through PD-L1 in preclinical studies of liver (199) and ovarian cancer (200). The inhibitor of CSF-1R, a receptor for macrophage colony-stimulating growth factor, was investigated in mouse models of pancreatic cancer where it decreased the frequencies of TAMs, while increasing IFN production and delaying tumor progression (201, 202). Similarly, CSF-1R inhibitor was found to synergize ACT therapy in a melanoma model (203). These data suggest that CSF-1R inhibitor may overcome the resistance to immunotherapy.

Specific chemokines, such as CCL5, CCL7, and CXCL8, play an important role in the recruitment of Tregs and MDSCs to the TME, consequently boosting an immunosuppressive climate (204). Alternately, chemokines CXCL9 and CXCL10 recruit CTLs to the TME (205) and the epigenetic silencing of the genes encoding them can reduce TILs and consequently promote resistance to ICIs (205). Epigenetic modulators of these chemokine receptors relieved the suppression of these Th1-cell-type chemokines and increased TILs leading to improved therapeutic efficacy of PD-L1 inhibitor in a model for ovarian cancer (155).

The cytokine milieu is critical to the recruitment, activation, and proliferation of immune cell, performing both immune stimulatory and suppressive effects (206). Transforming growth factor-β (TGF-β) is a cytokine playing key roles in angiogenesis and immunosuppression by stimulating Tregs (207) and excluding T cell in peritumoral fibroblasts (123, 159). Up-regulated TGF-β signaling was correlated with poorly immunogenic tumors and restrained response to ICIs in a colorectal cancer model, indicating resistance to therapy (159). Consequently, enhanced antitumor response to ICIs was observed following application of TGF-β inhibitor either alone or in combination with anti-CTLA-4 or radiation therapy (208, 209). Bintrafusp alfa, a bifunctional fusion protein composed of the extracellular domain of TGF-β receptor II (a TGF-β “trap”) fused to a human immunoglobulin G1 antibody blocking PD-L1, demonstrated favorable efficacy in patients with advanced NSCLC. Ongoing phase III trial is expected to validate the efficacy of bintrafusp alfa vs. pembrolizumab in the first-line setting in advanced NSCLC (NCT03631706).

Tumor necrosis factor-α pathway is postulated to be another immune evasion machinery conferring resistance to PD1 blockade. The expression of TNFα in an inflamed TME positively correlates with the expression of PD-L1 and TIM-3, along with impaired accumulation and increased activation-induced death of CD8+ TILs in melanoma models treated with anti-PD1 therapy. Accordingly, inhibition of TNF-α prevents the expression of PD-L1 and TIM3 and hampers anti-PD1-induced TIL death (210). Therefore, this study offers a rationale for the combination of PD-1/PD-L1 inhibitors with TNFα blockade as a novel immunotherapeutic strategy to overcome resistance in lung cancer, and the phase I clinical trial testing the combination is ongoing (NCT03293784).

Vascular endothelial growth factor has been linked to both decreased T−cell infiltration and immunosuppressive effects in addition to promoting angiogenesis and thus is associated with resistance to ICIs (211). Multiple mechanisms are involved in the interaction of VEGF with antitumor immunity: (1) VEGF prevented the commitment of lymphoid progenitors, decreasing progression to the T-cell lineage (212); (2) VEGF signaling promotes the infiltration of Tregs through a selective endothelium and reduces trafficking and extravasation of CTLs into the TME (213); and (3) VEGF increases expression of inhibitory receptors, contributing to CTL exhaustion (214). Higher levels of VEGF were found in anti-PD-1-resistant patients than sensitive ones (160). Based on these findings and the synergy between angiogenesis blockade and ICI therapies observed in preliminary studies, multiple trials of combination therapy are underway, including bevacizumab and VEGFR-TKI with anti-PD-1 therapy.

Higher levels of interleukin 6 (IL-6) and interleukin 8 (IL-8) have been found recently to be linked to reduced responses and worse clinical outcomes to ICI therapies across multiple types of cancers (215–217). IL-6 is a proinflammatory cytokine generated by T cells and macrophages and is usually involved in the immunoregulation connected to the IFN-γ signaling pathway. It can reduce the expression of PD-L1 and MHC-I, and result in tumor escape and resistance to ICI therapy (218). IL-8 is a proinflammatory chemokine and a chemoattractant for myeloid leukocytes expressed in multiple cancers (219, 220). It potently regulates the chemotaxis of neutrophils (221, 222) and exerts direct pro-tumorigenic effects (223). High levels of IL-8 are regarded to be associated with more neutrophil and monocyte infiltration, defective T-cell functions, and impaired antigen presentation, which subsequently result in resistance to ICI therapy (216, 217).

Contributions from inflammatory processes could participate in quashing the desired impacts of ICIs (Figure 1C). Adenosine can be produced under the condition of hypoxia and ischemia caused by tumor inflammation. It was reported to inhibit the cytotoxic function and proliferation of T cells via the A2A receptor on T cells (224). CD73, which mediates the generation of adenosine through dephosphorylation of adenosine monophosphate, was also demonstrated to suppress immune function (225). CD73 overexpression promotes T-cell exhaustion and is linked to the resistance to ICIs (226, 227).

IDO1 expressed in myeloid cells and cancer cells is a rate-limiting enzyme that converts tryptophan to its immunosuppressive metabolite kynurenine. This enzyme can induce T-cell anergy and apoptosis by gathering kynurenines and consuming the essential amino acid tryptophan and prevent the T-cell clonal expansion (228). It is particularly activated in DCs after binding with CTLA−4 and can be unregulated by CTLA−4 during adaptive immune resistance (229). Reduced expression of IDO at baseline was noted to be associated with poor response to ipilimumab in a phase II study in melanoma (161). Low level of IDO is likely to manifest insufficiency of suppressed TILs feasible to be reactivated by immunotherapy. Correspondingly, IDO-knockout mice exhibited improved OS with ICI compared with wild-type mice, and ICI therapy combined with IDO inhibitors showed both increased numbers and functions of TILs in the TME in an experimental setting (230, 231). However, despite encouraging results observed in preclinical and early-phase clinical studies in different types of tumors, no difference was shown in pembrolizumab combined with IDO1-selective inhibitor vs with placebo in a phase III study in metastatic melanoma (232).

B7-H4 has been proposed as another resistance marker to ICIs recently due to its negative modulation of T cells. B7-H4 constitutes a type I transmembrane protein of the B7 immunoglobulin superfamily and is encoded by the V-set domain containing T-cell activation inhibitor 1 (VTCN 1) gene. It is induced by activated T lymphocytes and down-regulates T-cell function by inhibiting proliferation, cytotoxicity activity, and interleukin secretion, after binding with T cells (233–235). Positive B7-H4 protein expression in patients with advanced NSCLC treated with nivolumab was recently reported to have an enhanced risk of tumor progression and tumor-related death compared with negative expression (236).