Intratumor heterogeneity: the hidden barrier to immunotherapy against MSI tumors from the perspective of IFN-γ signaling and tumor-infiltrating lymphocytes

ICB is one of the most promising anti-tumor immunotherapies to this day. The two most promising targets are CTLA-4 and the interaction of PD-1 and PD-L1. Upregulation of these immune checkpoints is an adaptive resistance associated with poor prognosis [111] and actually represents a strong pre-existing anti-tumor response, based on which ICB is applied to re-invigorate the immune response [57, 112, 113]. dMMR/MSI has been found to promote ICB efficacy in multiple tumor types, including glioblastoma multiforme [114], urothelial tract cancer [115], melanoma [57, 97], endometrial cancer [59], non-small cell lung cancer (NSCLC) [57, 112] gastric cancer [116] (Table 4).

It is believed that oligoclonal expansion of the TIL repertoire is a symbol of low TCR affinity and T cell exhaustion [117], while an appropriate level of TIL heterogeneity may be the foundation of ICB and ACT [118]. In this scenario, ICB could rejuvenate the TCR repertoire extensively rather than focusing only on several T cell epitopes, resulting in more T cells responding to ubiquitous neoantigens, enhancing overall immune competence in the anti-tumor response and leading to most clinically significant responses [119, 120]. Additionally, CD4 + T cells that stimulate and suppress the immunity of CD8 + T cells coexist in the tumor microenvironment [121]. While Tregs are regarded as suppressive regulators in tumor immunology and a biomarker of poor prognosis [122], they still possess specific reactivity against tumor antigens, facilitating CTLA-4 therapy [123]. Although PD-1 indicates negative regulatory function and exhaustion of peripheral T cells induced by the PD-1 signaling pathway and may contribute to the decreased diversity of T cell repertoire [124, 125], CD8 + T cells may function efficiently after PD-1 immunotherapy [126, 127]. Therefore, even though TILs are considered an immunosuppressive phenotype, they possess substantial capacity to induce a cytotoxic effect against tumor cells and their potential proliferation [121].

Among the various cytokines, IFN-γ is the main factor that induces upregulation of PD-L1 [128]. JAK1/2–STAT1/2/3–IRF1 pathway is the most important signaling cascade that is involved [129]. When IFN-γ binds to its receptors interferon-gamma receptor 1/2 (IFNGR 1/2), it increases the level of IFN-stimulated noncoding RNA 1 (INCR1)—a major regulator of IFN-γ signaling by modulating post-transcriptional JAK expression [130]. The subsequent activation of JAK1/2 leads to phosphorylation and dimerization of the downstream signal transducers and activators of transcription (STATs). Then the downstream transcription factors IRFs bind to their response elements IRF-1 response elements 1/2 (IRE1/2) in the upstream 5′-flanking region of the PD-L1 gene promoter [131] and induce PD-L1 upregulation (Fig. 3). A positive correlation between IRFs and PD-L1 mRNA expression was found in hepatocellular carcinoma (HCC) [131]. Similar to PD-L1, the expression of CTLA-4 in human melanoma cells is also regulated by IFN-γ through the JAK1/2-STAT1-IRF1 pathway [132]. CTLA-4 induces antiproliferation of T cells, Tregs activation and upregulation of IDO [133], playing a negative role in anti-tumor immune response. Therefore, anti-CTLA-4 therapy is utilized to increase the ratio of effector T cells to Tregs [87], and, in turn, upregulate IFN-γ production. Higher expression of PD-L1 and IDO predicts a superior response to PD-1 blockade and CTLA-4 blockade (ipilimumab), respectively [57, 134, 135], emphasizing the role of IFN-γ-induced IDO in immune checkpoint blockade therapy. Additionally, IFN-γ can induce MHCII expression, which is correlated with multiple important prognostic pathways and better overall survival rate [58]. In melanoma, MHCII expression is a predictor for anti-PD-1 and anti-PD-L1 response [136]. Altogether, high expression of IFN-γ signaling indicates long-term benefits from ICB [89, 116, 137]. In line with the relationship between PD-L1, CTLA-4, IDO and immunotherapy discussed herein, targeting LAG-3 strongly stimulates CD8 + T cell infiltration and IFN-γ secretion [138, 139], suggesting the possibility of an alternative immunotherapy. Interestingly, blockade of a single immune checkpoint could lead to upregulation of others [140]. For example, inhibition of LAG-3 improves the efficacy of PD-1 blockade in several mouse cancer models [141‐144], indicating the better efficacy of combinatorial ICB.

Efficacy of targeting a ubiquitous tumor antigen in adoptive cell therapy has been demonstrated [145]. Specific TCR-transduced T cells are clinically effective in treating patients with metastatic synovial sarcoma [7], while exploiting TILs to recognize multiple tumor neoantigens is effective in single-patient studies on several tumors [70]. Targeting several tumor antigens is an ideal scenario, which circumvents tumor escape mechanisms such as tumor heterogeneity and constructs a focused TIL repertoire against tumor cells [146].

However, the bottleneck of ACT is unable to address T cell migration and abnormal function at tumor sites. A recent study showed that PD-1 expression on transferred T cells could be induced by tumor environment [147], indicating that downregulation of immunosuppressive factors and silencing IFN-γ signaling to weaken PD-1-PD-L1 interactions may help improve potency. INCR1 knockdown cells are more susceptible to cytotoxic T cell-mediated death compared to controlled cells [130]. However, PD-1 blockade could improve therapeutic efficacy of ACT by enhancing T cell proliferation of T cells and upregulating IFN-γ [147, 148]. Importantly, functional IFN-γ signaling could induce chemokine (C-X-C motif) ligand 10 (CXCL10) to recruit more activated T cells and trigger a positive feedback loop [147] (Fig. 3). In addition, PD-1 blockade could increase the activation and proliferation of CAR-T cells in vitro and regress tumor growth in vivo through enhancing their anti-tumor effect and reducing myeloid-derived suppressor cells at tumor sites [149]. Noteworthy, a recent study also revealed that recurrent melanoma after ACT treatment exhibited high expression of IFN-γ signaling (PD-1, PD-L1, CTLA-4, though the picture was heterogeneous), which provided tractable targets for salvage immunotherapy, and indeed allowed for effective ICB [150]. As mentioned, IFN-γ plays an intricate role in ACT. ACT treatment outcomes are different when combined with other therapies due to the heterogeneity of IFN-γ signaling.

MMR-deficient subclones progress to manifest dMMR/MSI cancer lesions despite strong immunogenicity and immune surveillance due to upregulation of immune checkpoints and mutations favoring immune evasion. ICB remarkably benefits outcomes of dMMR/MSI tumors; in non-responders, combined with other immune-supportive approaches, it is expected to turn “cold” tumors into “hot” ones and improve the response rate. dMMR/MSI triggers frequent generation of frameshift mutations and gives rise to highly immunogenic frameshift-derived peptides (FSP), which contain multiple immunologically relevant neoepitopes [151]. These neoantigens are tumor-specific and shared by most MSI tumors [152]. A vaccine based on these neoantigens could be designed to prevent outgrowth of undetected dMMR/MSI subclones in pMMR tumors. A clinical Phase I/IIa trial found three commonly mutated FSPs (derived from genes AIM2, HT001 and TAF1B (NCT01461148), of which 98.5% of all MSI CRCs harbor at least one mutation [152]. Theoretically, immune response directed against FSPs can be induced in the majority of MSI CRCs, and the study results confirmed that this FSP vaccination was well tolerated and consistently induced immune responses [153]. The latest research analyzed 320 MSI tumors and selected 209 FSPs to generate a vaccine referred to as Nous-209. The vaccine induced IFN-γ + FSP-specific T cells in vaccinated mice and exhibited strong immunogenicity [154]. Its safety, tolerability and immunogenicity are currently under clinical evaluation in mCRC, gastric and gastro-esophageal cancer patients in combination with Pembrolizumab (NCT04041310) (Table 5). Vaccination with frameshift-derived neoantigen-loaded DC is also under investigation in MSI CRCs and persons who are known to harbor germline MMR gene mutation but without diseases yet (NCT01885702). Despite the therapeutic implications for MSI tumors, this trial could also explore the preventive significance of FSP vaccine for people with MMR mutations. Of note, a vaccine targeting these FSP antigens could broadly eliminate dMMR/MSI tumor cells despite the ITH and rapid tumor evolution, since these mutations are driver events at early stage of tumorigenesis [155, 156] (Fig. 3). Moreover, an IDO-derived peptide vaccine activates IDO-specific T cells which recognize and kill both tumor cells and immunosuppressive dendritic cells in vitro, significantly improving overall survival in III/IV NSCLC patients [157]. As combination therapy may have a synergistic effect due to distinct mechanisms of action, clinical trials are also underway to combine IDO and PD-L1 peptide vaccine with PD-1 blockade to treat metastatic melanoma (NCT03047928). Vaccines based on other upregulated antigens in dMMR/MSI tumors warrant further investigation.

When developing vaccines, a suitable vehicle of transmission can greatly enhance the therapeutic effect. Nanoparticles have been the promising vehicle of vaccine. They are endowed with outstanding physiochemical properties, such as high tissue specificity, manageable surface chemistry and big specific surface area [158]. The nanoparticles can be the vehicle of certain bioactive substance such as PD-L1 inhibitory peptide [159], or be developed with certain features to cause damage to tumor cells [160]. A latest review summarizes two main mechanisms that contribute to the anti-tumor effects of immunotherapy based on nanotechnology: one is to elicit an efficient immune response against tumor during tumorigenesis, while the other is to turn the “cold” immune-suppressive tumor microenvironment into a “hot” immune activated [158].

When exploring treatments for tumor, components of TME such as macrophages, fibroblasts or even tumor vasculature and tumor-draining lymph nodes can be targets of nanoparticles [161]. A vaccine was designed to deliver antigenic microparticle, which transformed tumor infiltrated macrophages into a tumor-suppressive M1 phenotype, and activated strong host immune response against tumor [162]. To enhance the specificity of nanoparticles, particular conditions are used to stimulate the function of the materials. A type of supramolecular gold nanorods can be activated by the second near-infrared-window (NIR-II) light. The nanorods are designed to be the vehicle of CRISPR/Cas9, and they can disrupt PD-1 gene expression of the tumor cells and facilitate immunogenic cell death when irradiated by NIR-II laser [163]. Some other nanoparticles can be released from membrane when entering a microenvironment with specific pH. A short interfering RNA named siFGL1 delivered by nanoparticles with hybrid biomimetic membrane can efficiently silence the FGL1 gene, which is triggered by pH [164]. Whether employed independently or in combination with other immunotherapies as adjuvant, these nanomaterials can enhance immune responses and exhibit anti-tumor efficacy [160, 164].

Although more diversified intratumoral TCRs may be generated in the context of dMMR/MSI, they are not always associated with better clinical outcome [65, 66]. It has long been recognized that tumor progression is accompanied by an increase in tumor mutation load, and the inevitable generation of tumor neoantigens [165]. High ITH is connected to tumor progression and resistant to therapies in many cancer types [47]. Heterogeneity in tumor antigen and immune cells is also significant among melanoma metastases, which leads to different responses to immunotherapy [166]. Excessive expression of subclonal neoantigens may lead to the relatively low expression levels of neoantigens, and T cells may be unable to encounter and activate against those low-frequency neoantigens [167]. Moreover, TCR repertoire diversity is associated with inadequate expansion of TCR clones and deficient infiltration into tumors, which may result from the immunosuppressive state of T cells caused by T cell exhaustion, low TCR affinity, etc. [168, 169]. A higher degree of TCR ITH and consequent clonotypes with low frequencies were revealed in different kinds of tumors and were linked with unfavorable prognosis [65, 170, 171]. Besides, some TILs have lost their functions owing to other dysfunction during the process of immune response. For instance, the tumor antigen TILs previously recognized can be depleted following immunoediting [172, 173], and deprivation of the presenting MHC allele can disrupt antigen presentation [174, 175] (Fig. 4). Therefore, same as above, heterogeneity in the quality of T cell responses, instead of the quantity, may be a determinant factor in anti-tumor response [65].

Provided that IFN-γ signaling displays a degree of heterogeneity and its downregulation correlates with an acquired resistance phenotype, alterations of essential components within IFN-γ signaling pathways could modify therapeutic efficacy. Recent studies demonstrate that INCR1 is transcribed as an antisense RNA from the PD-L1/PD-L2 locus and knockdown of INCR1 decreases PD-L1 expression [130]. JAK1/2-deficient cells emerged under/after ICB in patients with advanced melanoma and obtained resistance to PD-L1 blockade, which may result from pre-existing heterogenous subclones or through an adaptive response [9, 176, 177]. JAK loss is possibly correlated with lack of T cell infiltration based on the findings that factors downstream of JAK1/2 controls chemokines with chemoattractant effect on T cells, such as CXCL9, CXCL10 and CXCL11 [113, 178]. Also, high expression of PD-L1 significantly correlates with an objective response to PD-L1 blockade compared to PD-L1 negative patients [112, 113]. Altogether, dysfunction of IFN-γ signaling leads to the lack of PD-L1 expression, resulting in off-target of PD-L1 blockade, and less T cell infiltration for an anti-tumor effect (Fig. 4). Consistent with what’s described above, an interesting study mixed IFN-γ-insensitive tumor cells of melanoma with wild type (WT) tumor cells to mimic ITH. IFN-γ-insensitive cells finally grow out in the context of anti-PD-L1 therapy as a result of (1) failure to activate positive immune response by IFN-γ (2) lack of PD-L1 upregulation as the treatment target (3) immunodepressive microenvironment because of PD-L1 provided by WT. Moreover, IFN-γ could push the tumor further toward the IFN-γ-insensitive cells [179].

In addition, the JAK mutation contributes to the primary resistance to anti-PD-1 therapy in patients with advanced melanoma and colon cancer despite having a high mutation load [59, 96, 180, 181]. In previous studies, copy number alterations (CNAs) and single-nucleotide variants (SNVs) of IFN-γ signaling including loss of IFNGR1/2, JAK1/2, IRF1, as well as amplification of important IFN-γ pathway inhibitors SOCS1 and PIAS4, were found in patients with metastatic melanoma resistant to anti-CTLA-4 therapy. In addition, CXCL10 is reduced compared to the IFN-γ responsive cells [177]. Moreover, the heterogeneity of MHC expression on tumor cells and its lack of coordination with IFN-γ signaling have a significant impact on ICB. In sum, expression of IFN-γ strongly correlates with the response to ICB [182] and has validated in several studies. Deficiency of IFN-γ signaling can weaken the effect of positive immunoregulation in multiple aspects, thereby reducing efficacy of ICB. Diverse subclones carrying heterogenous IFN-γ signaling within tumors have an impact on drug response and should be considered when selecting therapeutic regimens. Given that CTLA-4 blockade leads to increased production of IFN-γ and thereby upregulating PD-L1, combination with PD-L1 blockade could make a better clinical response; and combination with new immune-related targets needs to be studied unremittingly in the future.

Mutations in JAK and STAT result in impaired IFN-γ signaling, lack of induced MHC class I expression, as well as inhibition of the WNT signaling pathway [11, 183]. A study investigating immune evasion in 1,211 CRC patients found that non-responsive dMMR/MSI patients frequently underwent immunoediting through upregulated WNT/β-catenin signaling and complete disruption of key genes in the antigen presentation pathway [7, 8]. High WNT signaling with mutations of β-catenin is inversely correlated with TIL independent of high TMB in melanoma and CRC, thereby reducing the efficacy of ICB [7, 184]. Other studies found that hyperactivation of WNT/β-catenin signaling suppressed effector T cells function by reducing IFN-γ [185] and led to defective migration of adoptive CD8 + T cells into tumors in melanoma [186]. This indicates that WNT signaling inhibitors may reverse immune evasion to facilitate immunotherapy. Approximately 30% of dMMR/MSI CRC display gene alterations of β2 microglobulin (β2M) in that the β2M gene harbors four coding microsatellites (cMS) [152]. β2M gene mutations lead to impaired MHC class I function, defective recognition and presentation of neoantigens which render the immune evasion from immunotherapy [176, 187, 188]. Altogether, mutations of IFN-γ signaling, WNT/β-catenin signaling and antigen presentation machinery, followed by resistance to T cell-induced death could all trace back to dMMR/MSI-induced heterogeneity (Table 6) (Fig. 4). Although high TMB is discussed as a positive predictor of immunotherapy, the quality of mutations to generate a robust T cell response may outweigh the quantity.

The optimal treatment is expected to target the trunk of all subclone mutations and subclonal driver events [19]. Therefore, it is indispensable to overcome the spatial heterogeneity and understand the full range of tumor tissues. The key step is accurate assessment, which is supported by a wealth of progressive studies [28, 197]. The conventional detection methods for dMMR/MSI are PCR and IHC. However, detection accuracy is limited by unfaithful Taq polymerase, limited panel numbers, the necessity for matched normal tissues and experience-dependent IHC [28]. Next-generation sequencing (NGS) allows for comprehensive investigations of multiple microsatellite loci simultaneously. MSI detected by PCR and 592-gene NGS was compared across 26 cancer types and a cutoff of ≥ 46 altered loci was found to classify samples as MSI [198], indicating that MSI-NGS is valid across cancer types and not limited by normal tissue acquisition. Additionally, tools based on NGS including mSing [199], MSIsensor [200], MSIplus [201] and MANTIS [202] have significantly improved sensitivity and specificity.

Several breakthroughs have been made with single-cell sequencing. Tumor cell diversity is analyzed by flow cytometry through a single-cell suspension which fully represents an intact tumor, providing the highest resolution to determine the true number of heterogenous subclones and characterize them without aggregating the information from multiple cells [203, 204]. Among all technologies, transcriptome analysis—single-cell RNA sequencing (scRNA-Seq) is the most advanced [203]. scRNA-Seq sheds light on the tumor immune microenvironment by showing the proportions of TILs. In mCRC samples, proportions of CD8 + T cells, Th1/2 cells and memory T cells were lower, and approximately 81.94% (118/144) of the genes related to WNT signaling were upregulated [205]. Patients with large B cell lymphoma who achieved complete response or remission showed improvement of memory T cells in scRNA-Seq of CAR-T cells [206]. Furthermore, scRNA-Seq identified TILs with high heterogeneity in Osteosarcoma (OS) and high expression of LAG-3 and TIGIT (T cell Immunoreceptor with Ig and ITIM domains) on CD8⁺ T cells, identifying new therapeutic targets for OS [207]. scRNA-Seq could also offer TCR sequence information and provides insight into TCR rearrangements at the single-cell level, unfolding dynamic responses to immunotherapy including vaccine and ICB [208]. TCR sequencing has been widely used and has helped probe into the dynamic combinations of T cell subsets and the spatial heterogeneity of TILs [84, 209, 210]. Single-cell sequencing has identified the heterogeneous expression of IFN-γ-related genes including MHCII in single cells, of which higher expression drives patients’ responsiveness to PD-1 blockade based on longitudinal scRNA-Seq [58, 211]. Enrichment of 227 IFN-γ-dependent transcripts including PD-L1 and IDO was also identified across multiple tumors and could be utilized to stratify immunotherapy response [212]. Mitra et al. found that single-cell analysis of a targeted transcriptome which predicted drug responses for individual cells was able to predict the response to a proteasome inhibitor when combined with machine learning in multiple myeloma [213]. Conceivably, it could also apply to immunotherapy based on correlative transcriptome signatures. Finally, simultaneous triple omics sequencing could reveal complex interplays within genetic, epigenetic and transcriptomic levels and provide the most complete maps of tumor cell subpopulations to guide treatment options [16].

The above discussion prompted us to quantify ITH and stratify patients by classifying potential responses to immunotherapy using combined biomarkers. Studies have classified immune status of tumors into several subtypes to support decision making and facilitate response prediction, based on TIL, IFN-γ signaling signatures and immune checkpoints expression [77, 214, 215]. Future studies should consider including multiple biomarkers to optimize this stratification method.

Due to the temporal heterogeneity during natural tumor progressing and therapeutic interventions, it is important to achieve real-time monitoring in a minimally invasive way and promptly adjust therapeutic regimens. Longitudinal analysis of tumor-derived genetic materials including CTCs and ctDNA extracted from patients’ blood has achieved promising progress across several types of solid tumors [216‐219]. These materials display all the alterations present in the tumor and the metastasis, which help eliminate false results caused by spatial heterogeneity. ctDNA analysis by liquid biopsy (blood test) is feasible and has been found to be sensitive and specific in various cancer types [220‐222]. Studies showed that ctDNA identified genomic profiling highly consistently with and beyond the findings of tissue biopsy [223‐228]. In 433 metastatic prostate cancer cases, dMMR identification using ctDNA was highly concordant with IHC and PCR of tumor tissue. Subclonal diversity and β-catenin activation were detected with sensitivity as well [229]. Detection of MSI using ctDNA with NGS in CRC was better than PCR and demonstrated high overall accuracy in pan-cancer [230]. Additionally, an initial peak following by a rapid decrease in ctDNA level indicates an early response for ACT, which in turn allows for early identification of those at risk of poor response and treatment optimization [206, 231]. Analysis of CTC also enables real-time monitoring and provides insight into the genomic profiling [232]. High expression of PD-L1 on CTC at baseline may be predictive to screen patients for PD-1/PD-L1 blockade and reduction of total CTC through longitudinal monitoring indicated a good response [233, 234]. Adjuvant PD-1/PD-L1 blockade deserves evaluation in patients whose PD-L1 ( +) CTCs are detected after curative treatment [235]. The number of CTCs significantly decreased after NK cell treatment in NSCLC and liver cancer, reflecting the therapeutic efficacy with decent sensitivity [236, 237]. Moreover, overexpression of β-catenin was detected in melanoma CTCs, but not in healthy donor and lacking in patients with complete response to ICB [238]. TMB measured from liquid biopsy was also found to be a predictive biomarker for atezolizumab (anti-PD-L1) in NSCLC, and able to identify patients who would benefit accurately and reproducibly [239]. In aggregate, liquid biopsy is a highly sensitive and informative method that can overcome ITH to identify low-frequency alterations and enable early detection of resistance or relapse.

Moreover, imaging techniques also allow for repeated response measurements during treatment, enabling visualization of ITH. Positron-emission tomography (PET) imaging with ⁸⁹Zr-atezolizumab (anti-PD-L1) in NSCLC, bladder and triple-negative breast cancer showed that tracer uptake was heterogenous and corresponded to PD-L1 and IFN-γ signaling levels at sites, appearing to be a strong predictor of atezolizumab response [240]. Radiolabeled [¹¹¹In] PD-L1-mAb and near-infrared dye conjugated NIR-PD-L1-mAb also demonstrably detected graded levels of PD-L1 expression with high specificity using SPECT/CT imaging [241, 242]. Transitioning these detective methods to combat ITH from the bench to bedside and evaluate and monitor patients’ potential benefits from immunotherapy is an enormous challenge that requires more clinical studies.