Resistance to anticancer therapies includes primary resistance, usually related to lack of target dependency or presence of additional targets, and secondary resistance, mostly driven by adaptation of the cancer cell to the selection pressure of treatment. Resistance to targeted therapy is frequently acquired, driven by on-target, bypass alterations, or cellular plasticity. Resistance to immunotherapy is often primary, orchestrated by sophisticated tumor–host–microenvironment interactions, but could also occur after initial efficacy, mostly when only partial responses are obtained. Here, we provide an overview of resistance to tumor and immune-targeted therapies and discuss challenges of overcoming resistance, and current and future directions of development.
A better and earlier identification of cancer-resistance mechanisms could avoid the use of ineffective drugs in patients not responding to therapy and provide the rationale for the administration of personalized drug associations. A clear description of the molecular interplayers is a prerequisite to the development of novel and dedicated anticancer drugs. Finally, the implementation of such cancer molecular and immunologic explorations in prospective clinical trials could de-risk the demonstration of more effective anticancer strategies in randomized registration trials, and bring us closer to the promise of cure.
The therapeutic success of tumor- and immune-targeted therapies has revolutionized the way we understand the biology of tumors and radically changed the therapeutic landscape of cancers. Spectacular tumor responses were seen with targeted therapy in oncogene-addicted cancers, and unprecedented long-term remissions in some patients treated with immunotherapy. Despite this immense progress, advanced cancer is ultimately lethal for most patients due to treatment resistance.
Cancer resistance is classified into two broad categories: primary resistance, with an early tumor progression, without prior tumor response, and secondary (acquired) resistance, which occurs after initial tumor responses (1–3). Although the two types of resistance can occur with both tumor and immune-targeted drugs, it is more common to see acquired resistance to tumor-targeted therapies and primary resistance to immune-targeted therapies, where there is an important proportion of patients without response to therapy in many cancers (3–6). For targeted agents and notably EGFR tyrosine kinase inhibitors (TKI), the Jackman criteria have been accepted by the medical community for defining acquired resistance (7). No such definition exists for immunotherapy, despite ongoing efforts from the Society for Immunotherapy of Cancer (SITC; ref. 8).
Finding predictive biomarkers of resistance has important implications for cancer care. It may avoid giving an ineffective treatment to patients whose tumors do not respond (9–11) or guide therapeutic choices to overcome resistance. In the case of targeted therapy, several biomarkers of resistance are validated and routinely used. For immunotherapy, the focus has been primarily to find biomarkers associated with a positive predictive value of response [PD-L1, microsatellite instability, tumor mutational burden (TMB), etc.]. Therefore, data are still sparse, and there is no approved, clinically validated resistance biomarker to guide treatment selection (although the absence of PD-L1 expression is de facto rendering some patients with cancer not eligible for anti–PD-L1 therapies).
Here, we provide an overview of the main concepts of tumor resistance to cancer cell–, stromal cell–, and immune cell–targeted therapies, applicable across tumor types. We provide examples of the clinical and preclinical evidence of tailoring treatment beyond progression, the challenges faced today, and potential directions of research.
Mechanisms of Resistance to Cancer Cell–Targeted Therapy
A targeted therapy may fail because of drug resistance or lack of drug target, or following inadequate drug exposure. Here we will focus on mechanisms of drug resistance when the target is expressed and validated. An overview of mechanisms of resistance to targeted therapy is reported in Fig. 1.
Primary Resistance to Targeted Therapy
Primary resistance is mostly related to the lack of target dependency or the presence of additional targets, in the context of tumor heterogeneity. Biomarkers of primary resistance to a certain drug are (i) insensitive variants of the target [e.g., EGFR insertions in exon 20 and the majority of EGFR inhibitors (EGFRi) in EGFR-mutated non–small cell lung cancer (NSCLC)]; (ii) aberrations of oncogenic pathways connecting to the target (e.g., RAS mutations and EGFR inhibition in metastatic colorectal cancer; refs. 9, 10); (iii) activating mutations downstream to the target (RB1 mutations and CDK4/6 inhibitors in estrogen receptor–positive/HER2-negative breast cancer; ref. 12); or (iv) activation of parallel oncogenic pathways (e.g., de novo ALK rearrangement and EGFR mutation in NSCLC and single-agent EGFR/ALKi; Fig. 1A). The latter is a very rare situation where it is critical to establish the dominant driver if single-agent inhibition is planned. In NSCLC, de novo ALK and EGFR aberrations were found to colocalize in the same tumor cell population and possibly in the same cellular clone in 1.3% of NSCLC cases (13). The dominant driver receptor seems to be more frequently the EGFR mutation, as patients tend to benefit more from EGFRi than from ALKi. The sensitivity to TKI appears to be related to the differential phosphorylation of EGFR and ALK, but the clinical validity and utility of testing phosphorylation levels of these proteins is currently unknown (13, 14). Another example is the co-occurrence of EGFR exon 19 deletions with nondisruptive TP53 exon 8 mutations, which have been associated with primary resistance to EGFRi in NSCLC (15).
Despite the identification of a targetable genomic alteration, drugs may be ineffective in the case of gene silencing, when the target lacks its protein expression. An analysis focused on 50 putative drivers comparing whole-exome tumor (somatic)/normal (germline) sequencing with whole-transcriptome sequencing in 1,417 primary tumors revealed that nearly 13% of the somatic single-nucleotide variants (SNV) were unexpectedly not transcribed as RNA. SNVs with high transcription rates included TP53, PIK3CA, and KRAS, whereas those with lower transcription rates included ALK, CSF1R, ERBB4, FLT3, GNAS, HNF1A, KDR, PDGFRA, RET, and SMO alterations. Interestingly, the higher the mutational load, the higher the number of silenced variants (16). Another example of gene silencing is posttranscriptional editing with RNA silencing by miRNA, RNA interference, and small interfering RNA (17).
Short-Term Adaptation to Targeted Therapies
The inhibition of feedback loops by targeted therapies can lead to unintended compensatory overactivation of upstream pathways. In PIK3CA-mutant estrogen receptor (ER)–positive breast cancer, increased ER activity occurs early upon PI3K inhibition and limits the antitumor activity of single-agent PI3K inhibitors (18). Also, mTOR and dual PI3K/mTOR inhibitors suppress mTORC1 and S6K, which normally send inhibitory signals via the insulin/IGFR and other receptor tyrosine kinases (RTK). Once these feedback loops are released, resistance may occur via upstream activation of PI3K, AKT, and ERK, which circumvents the antitumoral effects of PI3K/AKT/mTOR inhibitors. Treatment-induced hyperglycemia induces systemic hyperinsulinemia, which may activate PI3K signaling despite continued suppression with PI3K inhibitors. The ketogenic diet or SGLT2 inhibitors that lower blood glycemia were shown to inhibit the insulin feedback, decrease mTORC1 signaling in the tumor, and enhance the efficacy of PI3K inhibitors (19). Another example is anti-BRAF monotherapy that fails in BRAFV600E-mutated colorectal cancer due to feedback activation of EGFR, as opposed to melanoma and NSCLC (20). Similarly, the modest benefit of sotorasib in KRASG12C-mutated colorectal cancer as compared with NSCLC is explained by EGFR dependency and signaling rebound kinetics (21). MEK inhibitors also induce PI3K/AKT activation via EGFR, due to the cross-talk between the PI3K and RAS pathways (22), and HER2 TKIs induce HER3 upregulation as a compensatory mechanism of PI3K/AKT inhibition (23). The translational control of gene expression may also mediate resistance, exemplified by the formation of the eukaryotic translation initiation complex eIF4F, which regulates translation of many oncogenic pathways, including RAS–MAPK and PI3K–mTOR. The abnormal expression of eIF4F has been linked to resistance to HER2, BRAF, and MEK inhibitors (24, 25). Another example of short-term adaptation is nicely illustrated by a preclinical study investigating KRASG12C inhibitors. First-in-class KRASG12C inhibitors bind only to the inactive guanosine-5′-triphosphate state of the oncoprotein. Upon MAPK suppression by KRASG12C inhibition, some quiescent cells can produce new, active/drug-insensitive KRAS, in an EGFR–SHP2- and AURK-dependent manner (26).
Acquired Resistance to Targeted Therapy
The main mechanisms of acquired resistance include: (i) on-target resistance alterations; (ii) bypass alterations in the same pathway or connecting pathways; (iii) alterations that cause a phenotypic transformation of the tumor [epithelial-to-mesenchymal transition (EMT), small cell transformation of NSCLC, squamous transformation of adenocarcinomas]; and (iv) loss of target and target dependency (Fig. 1B; refs. 4, 6, 27–29).
On-Target Mechanisms of Resistance
On-target mechanisms of resistance include secondary mutations, target amplifications, or alternative splicing of the protein kinase mRNA. Following TKIs, “second-site mutations” are often seen in the intracellular kinase domain of the protein. They reduce drug affinity for the target by weakening the chemical bonds between the kinase and the drug (e.g., EGFRC797S), induce conformational changes of the kinase (gatekeeper mutations, e.g., BCR–ABLT315I, EGFRT790M, ALKL1196M, ROS1L2026M, RETV804M/L, and TRKAF589L) or cause direct steric hindrance to drug binding (solvent-front mutations, e.g., EGFRG796, ALKG1202R, ROS1G2032R, ROS1D2033N, RETG810, TRKAG595R, TRKBG639R, TRKCG623R; xDFG mutations, for example, ROS1G2101A/C, TRKAG667S, or TRKCG696A; refs. 30–41). As a consequence, some drugs with similar structure are susceptible to cross-resistance. Second-site mutations have been preponderantly described with TKIs targeting membrane receptors (RTKi; EGFR, cKIT, FGFR, MET, NTRK1–3, and RET) or cytosolic fusion proteins (ALK, ROS1, and BCR–ABL), whereas only exceptionally in case of intracellular protein kinases derived from oncogenic mutations. A single case of a secondary BRAF mutation (BRAFL514V) has yet been reported (42). Interestingly, the ALK fusion variant seems to affect acquired resistance, with more ALK resistance mutations in patients with variant 3 than those with variant 1, particularly ALKG1202R (43). Treatment with mAbs may result in mutations of the extracellular domain of the protein, which impair antibody binding (e.g., EGFR extracellular domain variants after cetuximab and panitumumab; refs. 44, 45). Mutations affecting the ligand binding domain may occur after endocrine therapy, leading to constitutive receptor activation (e.g., ESR1-activating mutations in endocrine treatment–resistant breast cancer, or androgen receptor splice variants, such as AR-V7, in enzalutamide/abiraterone-resistant prostate cancer; refs. 46, 47).
Aberrantly spliced proteins can mediate resistance via enhanced dimerization (e.g., splice variants of BRAFV600E that occur in 13% to 30% of patients with melanoma failing BRAF inhibitors (BRAFi; ref. 48) or by constitutive protein activation in the absence of ligand (e.g., androgen receptor splice variants in prostate cancer; ref. 47).
Exemplified by EGFR amplification emerging after EGFR inhibitors, BRAF amplification after BRAFi, or androgen receptor amplification in hormone-resistant prostate cancer, target amplification may limit the effectiveness of the drug by exceeding its inhibition capacity (49–51). In addition to the amplified mutant allele, high-level gene amplification of the wild-type allele can also induce resistance (51, 52). Another process that drives resistance is oncogene amplification on extrachromosomal DNA (ecDNA). These circular structures lacking centromers undergo an un-equal segregation during mitosis, which causes increased copy-number amplification with enhanced oncogene expression. This results in aggressive tumor behavior, poor prognosis, and drug resistance, but the exact underpinning mechanisms remain largely unknown (53). In experimental models of glioblastoma, resistance to EGFR TKI occurred through the loss of extrachromosomal (ec) EGFRvIII DNA, whereas drug withdrawal was followed by reemergence of clonal EGFR on ecDNA. This unexpected twist illustrates how EGFRvIII levels are modulated by ecDNA according to the tumor's needs (54).
Loss or Activation of Mirror Protein.
An illustrative example of “mirror proteins” is PI3K and PTEN proteins, with their divergent role on PIP3 formation. In PI3K signaling, the transformation of PIP2 to PIP3 is a necessary step for signal transduction, which is promoted by PI3K and negatively regulated by PTEN. In PIK3CA-mutated breast tumors treated with PI3Kα inhibitors, resistance may emerge via a progressive loss of PTEN expression, which abrogates the pathway inhibition. Preclinical studies point out that concomitant use of PI3Kα and PI3K p110β blockade might reverse resistance (55).
Bypass Resistance Mechanisms
There are three major oncogenic signaling pathways that drive cell growth and proliferation: the PI3K/AKT/mTOR (PI3K pathway), RAS/RAF/ERK (MAPK pathway), and STAT/JAK pathways. Oncogenic drivers frequently signal through the same pathways: EGFR, IGFR1, FGRF2, and HER2 signal via the PI3K and MAPK pathways and ALK, ROS1, MET, and cKIT via all three oncogenic pathways, whereas oncogenic BRAF signals exclusively through the MAPK pathway. Under targeted therapy, bypass tracks could eventually lead to the abnormal activation of the downstream pathway or connecting signaling pathways. This will ensure a sustained abnormal signaling despite continuous target inhibition. Notably, the more potent target inhibitors are used, the more frequently bypass tracks are likely to develop (56). Also, despite their apparent diversity, bypass tracks follow similar patterns due to the common signaling of targets. In the case of RTK inhibition, one common event among drivers is the activation of parallel RTK. For instance, in NSCLC, EGFR inhibition may result in the emergence of MET, ERBB amplification, FGF2–FGFR1 loop mutations, IGF1R activation, and fusion events (28, 57, 58); ALK inhibition could lead to increased EGFR activation or KIT amplification (52, 56); MET exon 14 inhibition could generate EGFR, HER2 amplifications (59). Examples of downstream pathway reactivation include MAPK pathway activation via BRAF, NRAS, and KRAS alterations following RTK inhibition in NSCLC (60–63), MAP2K1/2 and MITF alterations upon BRAFi in melanoma (64–66), NF1-inactivating mutations associated with endocrine therapy resistance in ER+ breast cancer (67) or PI3K activation via PTEN loss in breast cancer, and PIK3CA or AKT alterations as acquired resistance to RTK inhibitors in NSCLC, to anti-EGFR mAbs in colorectal cancer, or to BRAFi in melanoma (51, 64, 68–70).
Among bypass tracks, acquired fusions were recently discovered at EGFRi failure in EGFR-mutated NSCLC (71). The most frequently reported are the RET fusion (46%), followed by ALK (26%), NTRK1 (16%), and FGFR3 (11%; refs. 71–73). They were identified at higher rates after third-generation EGFRi than with first- or second-generation EGFRi (16% vs. 3%, respectively), when assessed in tissue biopsies by whole-exome sequencing (WES)/RNA sequencing (RNA-seq; ref. 72). As opposed to the classic oncogenic drivers, they frequently harbor uncommon 5′ partners, such as STRN for the ALK fusion and CCDC6 and NCOA4 for the RET fusion (71, 74). Fusions have been reported in other tumors where EGFR inhibition is routinely used, such as colorectal and head and neck cancers (75–77). Acquired fusions are not limited to EGFR inhibition, as they have also been recently described in metastatic ER+ breast cancer as a putative resistance mechanism to endocrine therapy (78).
Phenotypic switching, or cell plasticity, is a tumor-escape mechanism that allows cells with the same genotype to acquire diverse phenotypes in response to adverse tumor microenvironment (TME), such as hypoxia, inflammation, or exposure to targeted therapy. In the latter, this enables cells to proliferate independently of initial oncogenic drivers and promotes tumor resistance. For instance, resistant drug-exposed cells have been shown to regain sensitivity to the same drug after drug holidays, which suggests that nongenomic mechanisms of resistance are involved (79, 80). Cells could lose their epithelial phenotype and acquire mesenchymal characteristics, a process called EMT. This is induced by epigenetic modifications, such as upregulation of the histone methyltransferase enhancer of zeste homologue 2 (EZH2) or the RE1-silencing transcription factor (REST). Transformed cells feature increased migration and invasive potential, being highly refractory to targeted therapy. Proposed mechanisms are decreased levels of proapoptotic proteins and increased drug efflux due to upregulated ABC-binding cassette transporters. EMT has been described at failure of EGFRi in EGFR-mutated NSCLC (81). Lineage plasticity was also shown in breast cancer failing fulvestrant, where ARID1A-inactivating mutations promoted a phenotypic switch from ER-dependent luminal cells to ER-independent basal-like/stem-like cells (82). Cell transdifferentiation, a phenotypic transformation from adenocarcinoma to squamous or neuroendocrine carcinoma, has also been reported in 3% to 14% of patients with EGFR-mutated NSCLC after EGFRi and in around 17% of prostate cancer failing abiraterone/enzalutamide. Neuroendocrine transformation is preceded by the concomitant inactivation of TP53 and RB1, but this is not enough to induce lung cell transformation. Additional factors are necessary, such as MYC, BCL2 overexpression, and AKT overactivation (83, 84).
Loss of Target or Target Dependencies
Repairing Tumor Vulnerabilities.
In tumors with DNA damage repair alterations, cells rely on PARP-mediated DNA repair to survive. In tumors with frameshift or nonsense mutations of BRCA1/2, PALB2, and RAD51C/D treated with PARP inhibitors (PARPi), reversion mutations could restore the original open reading frame and thus regain protein function (85, 86). Reversion mutations emerge as a mechanism of resistance to platinum agents as well, being more frequently found in platinum-refractory/resistant than in platinum-sensitive ovarian tumors. In patients with high-grade ovarian carcinoma, the detection of reversion BRCA mutations in the liquid biopsy before the start of rucaparib has been correlated with a lower progression-free survival (PFS; ref. 87). Another example of repairing tumor vulnerabilities is the loss of target. For instance, the loss of PARP1 function by mutations in the DNA binding domain of PARP or by increased PARylation of PARP, which prevent PARP trapping, drives resistance to PARPi (29). Also, primarily ER+ breast tumors may lose the estrogen receptor at relapse, which is predictive for poor response to subsequent endocrine therapy (88).
Other Mechanisms of Resistance
Upregulation of genes encoding P-glycoprotein efflux pumps could result in increased efflux of drugs outside the tumor cell, if drugs are substrates of the P-glycoprotein. Some examples are PARPi or ALKi (with the exception of alectinib; refs. 29, 89).
Transient adaptive mutability has recently been described in colorectal cancer cell lines. In response to EGFRi and BRAFi, cancer cells experienced a downregulation of mismatch-repair proteins, decreased homologous recombination (HR) proficiency, and increased oxidative stress production. As a result, cells transiently increased their mutational load as a stress-response mechanism. This process was reversed once the tumor regained its abnormal growth capacity (90). Currently, there is no clinical evidence for this process.
Aberrant epigenetic modifications to the genome occur independently of the DNA sequence and could mediate drug resistance in addition to mutational processes. Epigenetic changes have been incriminated as key players in the induction of dormant, quiescent cells, which are mostly nondividing cells. These cells, called persister cells, constitute a small fraction of the tumor bulk that acquire stem cell features and survive despite high drug exposure or unfavorable microenvironmental conditions. They display a high expression of the KDM5A gene, which encodes a histone demethylase, resulting in reduced H3K4 methylation. The knockdown of KDM5A was shown to reverse the resistance phenotype and diminished the development of drug-tolerant persisters and their expansion in cell cultures (79).
Secretion of Soluble Growth Factors
Cancer-associated stromal cells may promote drug resistance by secretion of soluble factors, such as hepatocyte growth factor (HGF; ref. 83). HGF has been shown to confer innate resistance to BRAFi in BRAF-mutant melanoma, colorectal cancer, or glioblastoma cell lines (91), and to promote resistance to alectinib in ALK-positive NSCLC (92). HGF was able to activate the MET receptor, even in the absence of MET amplifications or activating mutations. As opposed to the selective ALKi alectinib, this phenomenon was not observed with crizotinib, because of its dual inhibition of ALK and MET (92).
The apolipoprotein B mRNA-editing catalytic polypeptide-like (APOBEC) enzymes, responsible for DNA cytosine deamination, are an important source of mutagenesis that fuel cancer diversity and subclonal evolution (93–95). APOBEC3B family member has been found to be upregulated in more than half of all cancers, and the APOBEC mutational signature has been shown to be the most prevalent in cancer after aging signatures (96, 97). Moreover, abnormal APOBEC function can generate late driver mutations as subclonal events (94).
Standard and Investigational Drugs for Overcoming Resistance to Targeted Therapy
Overcoming Primary Resistance
Promising results from phase I/II studies are currently available in the case of EGFR exon 20 insertions, with favorable objective response rate (ORR) and duration of response with mobocertinib (TAK-788), a selective inhibitor of EGFR and HER2 exon 20 insertion mutations (98), and with amivantamab (JNJ-61186372), a bispecific anti-EGFR–cMET antibody (99). Bispecific antibodies reduce the expression of EGFR and MET on the cell surface by promoting their internalization with subsequent downregulation by lysosomal degradation. They also induce tumor cell apoptosis in a BIM- and caspase-dependent fashion and show antibody-dependent cell-mediated cytotoxicity, with natural killer (NK) cells as effectors (100).
To overcome short-term adaptation mechanisms to targeted therapies, there is a need for concomitant blockade of feedback loops, such as blocking ER in addition to PIK3CA in PI3KCA-mutated ER+ breast cancer (101) or EGFR in addition to BRAF and probably KRAS in BRAFV600E- and KRASG12C-mutated colorectal cancer, respectively (21, 102).
Overcoming On-Target Resistance
Second-site mutations should be treated with drugs that have a different structure or a different binding and that are not subject to the same resistance alteration as the failing drug. In the case of TKIs, therapeutic strategies include the switch between drugs of different generations that differ in size and binding affinity, the switch between type I and II ATP-competitive inhibitors, an alternance between allosteric inhibitors and ATP-competitive inhibitors (Supplementary Table S1; refs. 32–38, 59, 103–112), and, more recently, the use of bispecific antibodies (113).
For example, in EGFR-mutated NSCLC, the EGFRT790M-resistant mutation located in the ATP-binding cleft of the kinase domain is able to restore the ATP affinity of the kinase. As a result, direct ATP competitors, first- and second-generation EGFRi, become ineffective (30). Osimertinib, a third-generation EGFRi, covalently binds to the C797 residue of the ATP-binding site and bypasses the change of the T790 residue (114). Nevertheless, tertiary resistance mutations may emerge under osimertinib, mainly at the C797 binding site, and further prevent the covalent binding of osimertinib. Other third-generation EGFRi with similar structures (e.g., rociletinib, olmutinib, and narzatinib) are likely to show cross-resistance with osimertinib (81). There are currently no approved treatments for these tertiary mutations, but very encouraging results were reported with amivantamab (JNJ-61186372) in the phase I CHRYSALIS trial, where patients with NSCLC failing third-generation EGFR inhibitors achieved partial responses in 10 of 47 cases, including 4 with EGFRC797S mutation (113).
In NTRK fusion–positive tumors, the second-generation TRK inhibitors selitrectinib and repotrectinib are able to overcome resistance to first-generation TRK inhibitors (entrectinib and larotrectinib) induced by solvent-front substitutions (39, 108). In preclinical models, TRK xDFG mutations (TRKAG667 and TRKCG696) were shown to mediate resistance to both first- and second-generation TRK inhibitors (type I inhibitors that bind to the active conformation state of the kinase), by stabilizing the kinase in an inactive DFG-out conformation. However, type II inhibitors (cabozantinib, ponatinib, and foretinib) were able to overcome this type of resistance, as the inactive state of the kinase facilitates their access to the allosteric “back pocket” of the DFG motif (115).
Another strategy is inducing degradation of target proteins, exemplified by selective ER degraders (SERD) for ER downregulation, or proteolysis-targeting chimeras (PROTAC) that act in conjunction with the ubiquitin-proteasome system (116, 117). Although the technology of PROTACs is still maturing, this is highly promising for undruggable, resistant targets (117).
Overcoming Bypass Resistance
Bypass tracks warrant a dual blockade of the founder mutation and the acquired resistance alteration. This strategy has proved to be safe and effective in the phase I TATTON trial evaluating savolitinib, a MET inhibitor, in combination with osimertinib in EGFR-mutated NSCLC (118). In individual case reports, acquired fusions have also benefited from a dual blockade (refs. 56, 61, 63, 71, 73, 119–124; Supplementary Table S2). Also, many drug combinations have proved to reverse resistance in in vitro studies, but clinical validation is pending (Supplementary Table S2). The ongoing phase II ORCHARD study (NCT03944772) evaluates combination strategies for bypass tracks in patients with NSCLC failing first-line osimertinib. Bispecific antibodies are also promising treatment strategies. In preclinical studies on wild-type and mutant EGFR NSCLC with c-MET pathway activation, the bispecific antibody targeting EGFR and c-MET (JNJ-61186372) demonstrated antitumor activity through target inhibition, enhanced antibody-dependent cell-mediated cytotoxicity, and FcγRIIIa binding (125). A phase I clinical trial is currently ongoing (NCT02609776). The prevention of acquired resistance is another valuable strategy. In BRAFV600E-mutated melanoma or NSCLC, MEKi are combined with BRAFi to prevent MAPK pathway reactivation, which frequently occurs under single-agent BRAFi (64, 126). Interestingly, aside from increasing efficacy and preventing resistance, the combination of MEKi and BRAFi also decreases cutaneous toxicities observed with paradoxical MAPK activation under BRAFi monotherapy (127, 128).
Histologic transformation is treated with chemotherapy or chemoimmunotherapy. Potential novel therapies suggested by preclinical data are targeting factors involved in lineage plasticity, such as blocking epigenetic factors (e.g., EZH2 inhibitors), antiapoptotic proteins (e.g., navitoclax, a BCL2 inhibitor), or stem cell markers (e.g., SOX2 inhibitors; refs. 83, 84).
Overcoming resistance to PARPi is a novel field where only preclinical data are available so far. In tumors with restored HR or restored replication fork protection, the combination of PARPi with ATR inhibitors (ATRi) might overcome PARPi resistance by hindering ATR-dependent mechanisms of proper HR/replication fork functionality. Moreover, the association between PARPi and ATRi might resensitize cells with combined disruption of BRCA1 and TP53BP1 or REV7 to PARPi (85).
As multiple oncogenic pathways may use the same key regulators of mRNA translation, the inhibition of these regulators may overcome resistance emerging from the inhibition of their upstream pathways. An example is the case of eIF4E for RAS/RAF and PI3K pathways, where resistance to BRAF, MEK inhibitors, or combinations, either ERK-dependent or ERK-independent, results in increased eIF4E complex formation, with subsequent signal transduction. It is hypothesized that combinations of BRAF signaling with drugs targeting eIF4E complex formation might overcome most of the resistance mechanisms arising from BRAF/MEK inhibition (25).
Mechanisms of Resistance to Antiangiogenic Targeted Therapies
Resistance to antiangiogenic drugs is still poorly understood. Important interplayers are both the tumor and the TME, with treatment-induced hypoxia influencing both compartments. Tumor cells adapt to VEGF blockade and hypoxia by redundancy in angiogenic signaling and activation of compensatory signaling pathways of angiogenesis (Fig. 2; ref. 129). This explains why antiangiogenic drugs targeting different tyrosine kinases are effective even in sequential use (130). Hypoxia promotes the recruitment of cancer-associated fibroblasts (CAF) or M2–tumor-associated macrophages (M2-TAM), which can drive angiogenesis by activation of endothelial cells and their bone marrow–derived precursors (131). Stromal cell–mediated resistance to anti-VEGF therapy includes an increase of T helper (TH) type 17 tumor infiltration with release of IL17. In murine models, this resulted in higher G-CSF secretion by CAFs and G-CSF–dependent myeloid-derived suppressor cell (MDSC) activation and recruitment in the TME. Consequently, there was a secretion of G-CSF–inducible angiogenic factors independent of VEGF. Pharmacologic inhibition of TH17 rendered treatment with VEGF antibodies effective, suggesting that immunomodulation might have a role in overcoming resistance of antiangiogenic drugs (132). Hypoxia also drives cancer cells to undergo metabolic changes to face nutrient starvation and poor oxygenation. It may even promote the acquisition of stem-like properties, with increased VEGF production, enhanced migratory properties, and aggressiveness, thus limiting treatment efficacy (133–135). Secondary resistance with anti-VEGF treatment was associated with increased expression of apelin (APLN) in preclinical xenograft models of ovarian cancer. Patients with ovarian cancer and high APLN expression had a significantly shorter disease-free survival than those with a low APLN expression (136). APLN binds to APLN receptor and activates MAPK signaling, thus resulting in endothelial progenitor cell proliferation and new vessel formation (137). Also, in vitro experiments showed that resistance to VEGF inhibitors may be mediated by IL6/STAT3 activation and overcome by IL6 blockade (138). Bevacizumab resistance has been associated in two cases with a focal amplification on chromosome 13q12.2, reported in 8.7% of patients with metastatic colorectal cancer. This is explained by the amplification of POLR1D1, which acts as a potential driver and upregulates VEGFA, an important promoter of angiogenesis (139). In the case of TKI, patients with renal cancer refractory to sunitinib were found to have increased IL8 expression. This was confirmed in xenograft models, where sunitinib-resistant tumors had high IL8 tumor secretion, whereas IL8 blockade resensitized tumors to sunitinib (140).
Sustained tumor angiogenesis may occur via VEGF-independent pathways, including normal vessel co-option or vasculogenic mimicry. In glioblastoma, hypoxia induced by bevacizumab may rapidly trigger adaptive mechanisms and drug resistance via vasculogenic mimicry, a process where tumor cells acquire endothelial-like properties and develop vessel-like structures (141). In response to antiangiogenic drugs, primary resistance occurs most likely in tumors with alternative neovascularization, whereas in tumors with high VEGF dependency, such as in clear cell renal cell carcinoma, resistance is often acquired (142).
Mechanisms of Resistance to Immune-Targeted Therapies
Approved immune-targeted therapies for cancer include antagonistic mAbs targeting immune checkpoints such as CTLA4, PD-1, and PD-L1 with the aim of modulating the antitumor T-cell immune response. The mechanisms of resistance of such T-cell immune modulatory drugs can, to some extent, speak to the mechanisms of resistance to bispecific T-cell engagers and chimeric antigen receptor T (CAR-T) cells approved for CD19+ B-cell leukemias and lymphomas. Resistance to immune-checkpoint targeted immunotherapies [immune checkpoint blockade (ICB)] has been reported to be driven by genomic and nongenomic mechanisms, where the tumor–host–microenvironment relationship has an essential role. Defects or alterations in the processes orchestrating this relationship could be divided into three broad categories, which relate to the cancer cell–intrinsic biology, the impact of this cancer cell on the phenotype of the TME, and the biology of the host (refs. 2, 3, 5, 143; Fig. 3).
Cancer Cell–Intrinsic Mechanisms of Resistance
The proof that neoantigen presentation drives immunotherapy sensitivity lies in cancers with secondary resistance to immunotherapy, which lose their initial antigen presentation (Fig. 3A). Following adoptive T-cell transfer therapy, it was shown that melanoma may undergo dedifferentiation with loss of the melanocyte differentiation antigen (144). Of note, neoantigen loss has been described even in early-stage, untreated melanoma and NSCLC, caused by promoter hypermethylation of genes coding neoantigens, or by copy-number loss or chromosomal deletions of truncal alterations (145–147).
Defective Neoantigen Presentation
Disrupted antigen presentation may affect ICB efficacy. For instance, a comparison between squamous lung carcinoma and adenocarcinoma with similar predicted neoantigens showed that squamous lung carcinoma is more prone to immune escape by downregulating HLA class I genes (148). Other examples are loss of function (LOF) of β2-microglobulin, another component of MHC class I; dysregulation of proteins involved in the proteolytic degradation of antigens (PSMB5); of transporters that pump the antigenic fragments across the endoplasmic reticulum (TAP1, TAP2, and TAPBP); or disruption of IFN–JAK–STAT signaling, which normally promotes MHC I expression on the cell surface (refs. 147, 149, 150; Fig. 3B and C). Interestingly, only B2M LOF mutations were found to be significantly enriched in nonresponders versus responders in two independent cohorts, in contrast to mutations in IFNGR1, JAK1, JAK2, STAT2, and TAP1/2 found in both nonresponders and responders (150). Nevertheless, B2M mutations were reported in 24% of microsatellite instability–high (MSI-H) colorectal cancers, where they did not preclude response to ICB (151). B2M LOF or downregulation has also been shown to cause acquired resistance to ICB in melanoma, NSCLC, and MSI-H colorectal cancer (152–154).
Neoantigen Intratumoral Heterogeneity
T-cell responses are effectively elicited by clonal neoantigens, as opposed to subclonal neoantigens and intratumoral heterogeneity (ITH; ref. 148). This may explain why TMB, which does not distinguish between clonal and subclonal neoantigens, is only imperfectly correlated with ICB responses. Notably, patients lacking a durable benefit under anti–PD-1 were found to have higher ITH scores than patients with durable clinical benefits. The acquisition of subclonal neoantigens is induced by abnormal APOBEC activity (mutational signature 2) and favored by chemotherapy and radiation. This might contribute to immune escape by the outgrowth of T cells that act against only a limited tumor cell fraction (94, 148, 155).
IFN Signaling Defects
Copy-number alterations and mutations within the IFNγ pathway have been identified in 75% of melanomas with primary resistance to anti-CTLA4 inhibitors (156, 157). Less frequently, LOF of JAK1 and JAK2 has also been shown to abrogate IFNγ signaling and IFN-induced inhibition of growth, described in both primary and acquired resistance to ICB (152, 156, 158, 159). Deleterious JAK1 and JAK2 mutations are enriched in tumors with MSI, notably in endometrial cancer (35% of immune-naïve patients had JAK1 mutations; refs. 158, 160, 161). Interestingly, JAK2 and type I IFN genes are frequently codeleted with CDKN2A, due to their relative proximity on chromosome 9p, suggesting that CDKN2A loss may be a biomarker indirectly associated with immunotherapy resistance (162, 163). Patel and colleagues identified 13 IFNγ-induced genes and three TNFα-induced genes that could foster resistance to ICB (149). Also, the LOF of the APLNR gene, which normally encodes a rhodopsin-like receptor with roles in blood pressure regulation, has recently been shown to reduce immune response by IFNγ modulation via JAK1 (149). IFN secretion might also be impaired when STING and/or cGAS expression is silenced through epigenetic hypermethylation processes (164).
Oncogenic Signaling with T-cell Exclusion
Tumors lacking CD8+ T cells are termed “cold tumors” and have been associated with a lack of response to ICB. Such tumors do not display an inflammatory, chemokine signature and generally lack negative regulators, such as PD-L1, regulatory T cells (Treg), or indoleamine-2,3-dioxygenase (IDO; ref. 165). The lack of effector T-cell infiltration could be mediated by specific oncogenic signals, such as tumor-intrinsic active β-catenin signaling, loss of PTEN, STK11/LKB1, or KEAP1 aberrations, or because of IFN signaling defects (JAK1/2 and B2M LOF mutations; refs. 166, 167). Active β-catenin signaling was shown to promote T-cell exclusion by failure of T-cell priming, with resistance to anti–PD-L1/anti-CTLA4 mAb therapy in a melanoma mouse model (166), as well as resistance to T-cell adoptive transfer and vaccination (168). The activators of the β-catenin pathway could be activating mutations of CTNNB1, overexpression of specific WNT ligands or Frizzled receptors, or inactivating mutations of pathway inhibitors, such as the AXIN1 gene (Fig. 3D; ref. 166). Biallelic loss of PTEN was reported to mediate resistance to ICB (169, 170), possibly by inducing an increased expression of immunosuppressive cytokines and the recruitment of suppressive immune cells via VEGFR (171–173). STK11/LKB1 and KEAP1-mutant or -deficient NSCLC has been associated with impaired IFN signaling (167). Patients with NSCLC presenting STK11/KEAP1 aberrations had inferior PFS with combined chemoimmunotherapy compared with those without, and there has been no observed benefit from the addition of ICB to chemotherapy (174). The prognostic effect of STK11–KEAP1 was further confirmed in a large real-world data cohort of patients with advanced NSCLC, but not its predictive value (175). Notably, various KEAP1-driven comutations (STK11, SMARCA4, and PBRM1) have been associated with a lack of response to ICB despite a high TMB in lung adenocarcinoma (176). In KRAS-mutant NSCLC cell lines with STK11/LKB1 loss, STING expression was markedly silenced by hyperactivation of DNMT1 and EZH2, thus facilitating immune escape and explaining why such tumors are often refractory to ICB (177).
Certain clonal driver alterations associate with an increased immunosuppressive TME and low CD8+ T cell/Treg ratios in the tumor (156). In a retrospective study of 551 patients with oncogene-addicted NSCLC receiving ICB monotherapy, some patients experienced tumor regression, but generally the efficacy was lower as compared with the KRAS group. Notably, patients with ALK-positive NSCLC did not obtain any objective response (178). Immunotherapy responses seem to differ between EGFR mutation subtypes when compared with EGFR wild-type patients, with lower ORR and overall survival for EGFR exon 19 deletion and similar outcomes for EGFRL858R substitution. A possible explanation is the higher TMB observed among EGFRL858R-mutated tumors as compared with EGFR exon 19–mutated tumors (179). Also, oncogenic signaling, notably the RAS–MAPK and PI3K–mTOR pathways, stimulates the eIF4F eukaryotic translation initiation complex, which regulates the translation of STAT1 mRNA and mediates the induction of PD-L1 expression (180).
Tumors with increased levels of chromosomal instability activate the cGAS–STING pathway through chronic release of cytosolic DNA, but they may escape the immune system by suppressing downstream type I IFN signaling and instead upregulating the alternative NF-κB pathway. This leads to paradoxical protumorigenic effects and the recruitment of immunosuppressive cells (164, 181). A similar immune evasion may occur in HR-deficient tumors, where high levels of DNA damage may activate the ATM–TRAF6 alternative STING pathway, resulting in IL6 and TGFβ production, with M2-TAM and Treg recruitment (182, 183).
An interesting biomarker is polybromo-1 (PBRM1) gene LOF mutations, clinically validated as a predictor of ICB response in clear cell renal cell carcinoma in an independent cohort of patients treated with nivolumab within a randomized clinical trial (184). Nevertheless, in a retrospective analysis of 2,764 patients with NSCLC who received ICB as a second or later line of therapy, patients with PBRM1 mutations (n = 84 patients, 3%) had worse survival than patients with wild-type PBRM1, despite associating with a significantly higher TMB (185).
Resistance to Lysis of Metastatic Tumor Cells
In a prostate cancer model, both early and metastatic tumors induced cytotoxic T cells after irradiation, but only primary cancer cells were preponderantly eliminated by immunotherapy, whereas metastatic cells were resistant to cytotoxic T cell–induced cell lysis. Despite IFN-induced MHC I expression on metastatic cells, these cells maintained resistance to lysis (186).
Stromal Mechanisms of Resistance
Location of Metastasis
In patients with melanoma, the presence of visceral metastasis was shown to be associated with primary resistance to anti–PD-L1, as opposed to lung, lymph node, and soft-tissue metastases (187). In patients with both melanoma and NSCLC, patients with liver metastases had worse clinical outcomes with ICB, compared with patients without liver metastases (188, 189). Liver metastases exhibit an immunosuppressive TME, and it has been suggested that they may even promote systemic immunosuppression (188). In melanoma, patients with liver metastases had higher levels of eotaxin 2, interferon gamma-induced protein 10 (IP10) and IL8, MMP8 and HIF1α, as compared with patients without liver metastases (188). Brain metastases also display an immunosuppressive TME, with low tumor-infiltrating lymphocytes (TIL), increased M2-TAM and neutrophils, inhibition of dendritic cell maturation, TH1, and leukocyte extravasation signaling pathways, as well as a significant reduction in the proinflammatory cell adhesion molecule vascular cell adhesion protein 1 (VCAM1), responsible for leukocyte adhesion to inflammatory sites (190–192).
Expression of Multiple Inhibitory Regulators
In addition to PD-1 and CTLA4 expression, infiltrating CD8+ T cells may have increased expression of T-cell immunoglobulin mucin-3 (TIM3), lymphocyte-activation gene 3 (LAG3), and B and T lymphocyte attenuator (BTLA; Fig. 3E). These regulators are also expressed by other immune cells, such as Tregs or cells of innate immunity (193). The upregulation of coinhibitory molecules has been described as a mechanism of acquired resistance to single-agent anti–PD-1/PD-L1, causing an impairment of T cells to respond to polyclonal activation. It has been shown that multiple inhibitory checkpoints were acquired gradually, with PD-1 as an early event and LAG3/BTLA expression as a late event. This mechanism could drive resistance to ICB even in PD-1 high-expressing tumors (194, 195).
Exhaustion of Cytotoxic T Cells
The chronic exposure of cytotoxic T cells to high antigen load might severely impair their inflammatory and cytotoxic function, from a gradual loss of effector function to an exhausted phenotype (194). This exhausted phenotype can present abnormal DNA methylation (196), high PD-1, and LAG3 expression (148).
Tregs have important roles in preventing autoimmune reactions by secretion of immunosuppressive cytokines, such as TGFβ, IL10, and IL35. When present in the TME, they restrict T-cell responses (197). TGFβ may also be released by cancer cells or stromal fibroblasts (198). It has been found to be increased in the bone marrow from bone metastasis compared with the normal bone, which might explain the low benefit of ICB in metastatic castration-resistant prostate cancer (mCRPC) with bone-predominant disease. Moreover, in mice with bone mCRPC, the association of anti-CTLA4 and anti-TGFβ increased TH1 cells and decreased Tregs in the tumor-infiltrating T-cell population. The combination of drugs resulted in an improved tumor response and overall survival of mice in comparison with either drug alone (199). Innate immune cells, MDSCs, and M2-TAM have also emerged as major regulators of immune responses in cancer, and they have been shown to negatively affect the activity of ICB, adoptive T-cell therapy, or dendritic cell vaccination (Fig. 3F; ref. 3).
Under aerobic conditions, tumor cells rapidly process glucose into lactic acid, a process called the Warburg effect. This results in increased release of H+ ions into the TME. The abnormal tumor vasculature is unable to effectively clear H+ ions, which promotes an acidic tumor milieu. Consequently, TILs and cytokine production are inhibited, while the accumulation of immune-suppressive cells is favored, thus leading to immune resistance (193, 200). Another mechanism is the production of the tryptophan-metabolizing IDO by tumor cells in response to inflammation. IDO has immunosuppressive functions that limit effector T-cell activity and engage mechanisms of immune tolerance (165).
In the TRACERx and LATTICe-A cohorts of primary lung adenocarcinoma, the presence of immune cold regions within a tumor was shown to be an independent predictor of cancer relapse, even in the case of an average increased immune infiltration of the tumor. The comparison between cold and hot regions reflected different cancer subclones, possibly due to immunoediting (201). An increased heterogeneity was found in the immune microenvironment of different regions of the same tumor in nearly one third of early NSCLC tumors (145).
Characteristics of the Host
Age and Sex
Growing evidence points out that younger and female patients respond less to immunotherapy than older and male patients. It has been shown that younger and female patients are prone to stronger immunoediting early in their disease evolution, with depletion of potentially immunogenic mutant peptides and MHC presentation of less immunogenic, more invisible peptides to the immune system (202).
Specific gut bacteria are associated with ICB response and/or toxicity. For instance, in patients with melanoma treated with ICB, baseline gut microbiota enriched with Faecalibacterium and other Firmicutes was associated with enhanced antitumor responses (203, 204). Also, a molecular mimicry was recently observed between tumor MHC class I antigens and an enterococcal prophage, which correlated with long-term responses under anti–PD-1 therapy in renal and lung cancers (205). The composition of gut microbiota may affect immune responses of distant lesions, especially via the blood circulation of microbial metabolites produced in the colon through bacterial fermentation of dietary fibers. High blood levels of short-chain fatty acids, such as butyrate and propionate, have been shown to associate with CTLA4 blockade resistance and with an increased proportion of Tregs. In patients treated with ipilimumab, high blood levels of butyrate limited the accumulation of memory and ICOS+ CD4+ T cells, as well as IL2 impregnation (206).
Blood Immune Cells
Relative eosinophil count (REC) <1.5% was associated with a worse survival in patients with melanoma treated with pembrolizumab, as compared with REC ≥ 1.5% (187). Interestingly, lymphopenia per se is not associated with primary resistance to anti–PD-L1 but rather the proportion among other cells (187, 207), notably neutrophils, within a neutrophil-to-lymphocyte ratio (NLR; refs. 208–210). The negative impact of NLR seems to be limited to patients with stable disease (SD) and not to patients with progression or objective response as best response (211). Also, the proportion of neutrophils among total leukocytes minus neutrophils (dNLR), combined with lactate dehydrogenase (LDH) in the Lung Immune Prognostic Index (LIPI) score, has been associated with primary resistance to anti–PD-L1 (212).
IL6 has prognostic value, with high IL6 levels being correlated with a worse survival in patients with metastatic melanoma treated with ICB or chemotherapy (213). IL8 has been related to the level of infiltrating myeloperoxidase-positive (MPO+) and/or CD15+ monocytes and neutrophils in tumors (214). IL8-high patients, as defined by serum IL8 values above 23 pg/mL, represent around 25% to 30% of patients with advanced cancer. Interestingly, IL8 serum levels are independent from PD-L1 tumor expression by IHC and TMB.
Serum LDH value above normal threshold is a strong predictor of primary resistance to ICB (187, 209, 215, 216). LDH leads to accumulation of lactates in the TME, which leads to modifications of immune cells toward a tolerogenic phenotype (217) and negatively affects the type I IFN pathway by inhibition of retinoic acid–inducible gene 1 (RIG1; refs. 218, 219).
Of note, patterns of primary resistance differ among tumor types even for the same immunotherapy (Supplementary Table S3; refs. 145, 157, 158, 167, 174, 178, 185, 220, 221). Tumor types with a low likelihood of response to ICB, for instance, sarcoma or prostate cancer, are usually cold tumors with a low TMB. Responses may be exceptionally seen, for instance in sarcomas with B cell–rich tertiary lymphoid structures (221), or MSI-H or CDK12 biallelic alterations in prostate cancer (222, 223). In these cases, predictors of response should be searched. Conversely, in tumor types with a high likelihood of response to ICB, such as melanoma, Hodgkin lymphoma, and MSI-H tumors, predictors of resistance should be interrogated, and these usually consist of defects of antigen presentation and IFN signaling. Concerning acquired resistance, defects of antigen presentation, loss of antigen, IFN signaling defects, and overexpression of coinhibitory checkpoints seem to be the central players in progression, without evident differences among tumor types at the actual state of knowledge (144, 146, 147, 152, 153, 159, 170, 195).
Overcoming Resistance to Immunotherapy
Overcoming resistance to immunotherapy is guided by several main pillars: (i) increase tumor visibility to the immune system, (ii) enhance T-cell infiltration, (iii) remove barriers from T-cell intratumoral trafficking, (iv) enhance immune system function, (v) address genomic and epigenetic abnormalities, and (vi) adapt to the host (Fig. 4).
Increase Tumor Visibility to the Immune System
Making the tumor visible to the immune system is essential for T-cell activation and migration into the TME. One major strategy is the induction of “immunogenic cell death” by tumor cytotoxic therapies, including chemotherapy, irradiation, targeted therapies, epigenetic drugs, or oncolytic viruses or peptides (193, 224). Following cell death, tumor antigens, proinflammatory cytokines, damage-associated molecular patterns (DAMP), calreticulin, and ATP are released in the TME (Fig. 4A). This may favor antitumor immune responses by recruiting antigen-presenting cells (e.g., dendritic cells), and inducing their maturation (e.g., HLA-I, CD80/86 upregulation) and subsequent T-cell activation (e.g., upregulation of costimulatory checkpoints; refs. 224, 225). DNA lesions may activate the cGAS–STING pathway, which can increase the immunogenicity of the tumor by activating the type I IFN pathway (164). In addition, some chemotherapies can deplete some suppressive immune cells such as Tregs or MDSCs (193). Chemoimmunotherapy improved survival outcomes in comparison with chemotherapy alone in advanced lung cancer, regardless of PD-L1 expression (226). Interestingly, it seems to reduce the risk of early progression seen with ICB alone. It is unknown if a shorter administration of chemotherapy would be more effective when combined with ICB in order to prevent lymphopenia or the generation of subclonal antigens. Possibly, the next step will be the administration of antibody–drug conjugates (ADC) in combination with ICBs, as they would spare lymphocytes. Successful examples of combining antiangiogenic drugs with immunotherapy are especially seen in renal cell carcinoma (e.g., axitinib plus pembrolizumab, axitinib with avelumab), where in 90% of cases there is a deregulation of the VHL gene, which regulates HIF1α and HIF2α (227). Also, antiangiogenic drugs have been shown to modulate the expression of inhibitory checkpoints on CD8+ T cells and the levels of Treg-infiltrating tumors (228, 229). Radiotherapy has been shown to induce an abscopal effect, first described in a case report in 2012 (230), but this effect has not been reproduced in large studies. Concerns were raised on the effectiveness of single-site irradiation, as it fails to address intertumoral and immune heterogeneity (225, 231). New strategies of combining ICB with multiple site irradiation are being tested (225). It has been suggested that nodal irradiation should be avoided when used to enhance ICB effects, as it may compromise T-cell proliferation and activation (232). Furthermore, radiation-induced STING activation may cause resistance by recruiting MDSCs via the CCR2 pathway, for which a possible abrogating strategy might be the use of anti-CCR2 antibodies (233). Oncolytic viruses are used to preferentially infect cancer cells, where they trigger immune responses by viral replication, cancer cell release of DAMPs, pathogen-associated molecular patterns, and type I IFNs, with subsequent increase of TILs and even induction of abscopal effects (234–236). Oncolytic peptides may act by mitochondrial membrane permeabilization, which triggers a caspase-independent necrosis (237).
Enhance T-cell Infiltration
In addition to poor tumor immunogenicity, the lack of T-cell infiltration could be caused by inefficient T-cell priming or the lack of T-cell attraction. STING agonists, which activate the STING pathway, promote tumor secretion of type I IFN and proinflammatory cytokines (164). The role of chemokines on ICB to attract intratumor T cells is being investigated in early-phase clinical trials, as certain chemokines (e.g., CXCL13) may mediate the recruitment of T cells into the TME (238). The induction of tertiary lymphoid structure formation in tumors is being investigated using cytokines, chemokines, antibodies, antigen-presenting cells, or synthetic scaffolds (239). Other strategies include vaccines and adoptive cellular therapies (Fig. 4B). Engineered T cells with CAR or transgenic T-cell receptors (TCR) have shown modest activity in solid tumors so far, possibly explained by tumor heterogeneity, which limits the effectiveness of strategies targeting one single antigen, the lack of cancer-specific targets for CARs, and the difficulty in matching HLA antigen for TCRs (240–242). Promising results were seen in patients with neuroblastoma treated with CAR-T cells targeting disialoganglioside GD2, showing the potential of CAR-T cells in solid tumors, when the target is adequate (243).
Why Combination with ICB Is Important
Once non–T cell–inflamed tumors have T-cell infiltrates, IFNγ signaling will be activated, with subsequent expression of inhibitory mechanisms, such as PD-L1. For instance, in mCRPC, ipilimumab failed to induce durable responses in most patients despite being able to increase T-cell infiltration in the tumors. This was caused by the overexpression of compensatory inhibitory pathways, PD-L1 and VISTA (244). A similar escape mechanism was observed in patients treated with dendritic vaccines, which led to a significant increase in PD-L1 expression (245). This provides the rationale that bringing T cells into the tumor must be accompanied by a blockade of emerging inhibitory pathways.
Remove Barriers from Intratumoral T-cell Trafficking
Leaky blood vessels coupled with defective lymphatic drainage cause an elevated interstitial pressure in the TME, which may impair drug and immune cell entry. Moreover, TME cells exposed to hypoxia produce various cytokines with immunosuppressive effects (VEGF, TGFβ, and prostaglandin E), whereas tumor cells might acquire highly aggressive stemness features (193, 246). In addition to promoting immunogenic cell death, antiangiogenic drugs may normalize tumor vasculature and increase T-cell infiltration in the TME (Fig. 4C).
The spatial distance between T cells and tumor cells might be minimized by bispecific antibodies that bind T cells with tumor cells. For instance, an early success was seen with the bispecific antibody blinatumomab, approved in the treatment of B-cell leukemia, which acts by binding CD3 on T cells and CD19 on B cells (247).
Enhance Immune System Function
Enhancing immune functions focuses on promoting the infiltration and function of effector T cells and antigen-presenting cells on inhibiting immunosuppressive cells such as Tregs and MDSCs, or modifying cytokine production or cellular metabolism (Fig. 4D; ref. 193).
Boosting effector T cells can be achieved by concomitant blockade of inhibitory regulatory checkpoints or by activating stimulatory pathways (193–195). The combination of anti–PD-1/PD-L1 agents with anti-CTLA4 mAbs was the first proof of increased efficacy compared with monotherapy in several cancer types, albeit with the cost of increased toxicity (5). Ongoing clinical trials investigate the combination of anti–PD-1/PD-L1 or anti-CTLA4 with inhibitors of other immune-checkpoint regulators, such as LAG3 or TIGIT (5, 193). Recently, the phase II CITYSCAPE trial evaluating the anti-TIGIT mAb tiragolumab plus the PD-L1 inhibitor atezolizumab in patients with NSCLC with positive PD-L1 showed an improved ORR and PFS for the combination therapy, as compared with atezolizumab alone. Efficacy was higher in the high–PD-L1 subgroup (248). Phase III trials are ongoing (SKYSCRAPER-01 and SKYSCRAPER-02).
T-cell function can also be enhanced by removing immunosuppressive signals. In case of immunosuppressive TME, inhibitors of M2-TAM or ICB combinations with anti-TGFβ therapies may be used (156).
Both innate immunity and adaptive immunity may be stimulated by proinflammatory cytokines (249–251), as nicely illustrated in a phase I trial investigating an engineered IL2 receptor agonist (bempegaldesleukin) in combination with nivolumab in immunotherapy-naïve patients with advanced solid tumors. Longitudinal biopsies revealed increased T-cell infiltration and clonality in patients responding to therapy, with no expansion of Tregs in the tumor (249). In preclinical models, IL15 agonist or recombinant human IL2 used in combination with ICB was shown to be more effective than either monotherapy by stimulating both CD8+ T cells and NK cells (250, 251).
Innate immunity may be stimulated by intratumoral administration of Toll-like receptor 9 (TLR9) agonists (CpG olidonucleotides, e.g., SD-101; refs. 252, 253), double-stranded RNA viruses triggering TLR3 and RNA helicases (254), or by STING/cGAS activation (164). This might be a rational strategy for tumors that lose MHC, as MHC I–deficient cells are unable to present the antigen to effector T cells, but they are susceptible to NK cell–dependent cytotoxicity (255).
Address Genomic and Epigenetic Abnormalities
In mice bearing melanomas with PTEN loss, the addition of a PIK3β inhibitor to anti–PD-1 blockade was shown to increase intratumoral T-cell infiltration and to improve tumor control as compared with each treatment alone (171).
Abnormal IFN signaling is addressed by combinations of ICB with IFNγ or STING agonists (Fig. 4E; ref. 156). JAK inhibition was also tested in a phase I trial, but proved unsuccessful when combined with pembrolizumab, being associated with reduced peripheral T-cell activation and no significant changes in intratumoral Treg infiltration (256). Conversely, PD-1 blockade resistance caused by JAK1/2 and B2M inactivating mutations has been successfully reversed in murine models by TLR9 agonists and CD122-preferential IL2 pathway agonists, respectively (255).
The potential synergy between epigenetic drugs and ICB is also being investigated by addressing the epigenetic modulation of tumor neoantigens, chemokines, or inhibiting activation-induced cell death of T cells (193). For example, promising results were seen in the ENCORE-601 phase Ib/II trial in patients with melanoma, NSCLC, or mismatch repair–proficient (pMMR) colorectal cancer treated with pembrolizumab in association with a histone deacetylase inhibitor who obtained tumor responses even after prior immunotherapy failure (257).
Adapt to the Host
Younger and female patients may be better candidates for drug combinations, as opposed to older and male patients with higher chances of response to ICB (202). Manipulating microbiota is undergoing investigation for enhancing immune responses or managing immune-related adverse events (Fig. 4F; ref. 258). Studies investigating ICB combined with LDH inhibitors might be of interest for patients with high LDH levels, as experimental data showed antitumor effects of LDHA inhibitors by inducing reactive oxygen species and apoptosis (259).
Current Challenges with Tumor-and Immune-Targeted Therapies and Future Perspectives
Improving the Characterization of Resistance
Tumors may contain cells that follow different evolutionary tracks, including drug-sensitive cells, drug-tolerant persister cells, fully resistant cells, and partially sensitive or weakly resistant cells (79, 260). Tumor response is dictated by the tumor fraction harbored by these cell subtypes, which makes a deeper tumor characterization essential for anticipating resistance and improving treatment choices (Fig. 5A and B). We are currently limited in our ability to distinguish between the clonal expansion of preexistent resistant cells and acquired resistance (e.g., EGFRT790M might also be found in patients with treatment-naïve NSCLC; ref. 1). Treatment effectiveness may differ in these cases, with earlier treatment failure when resistant cells are present at baseline. To improve the characterization of tumor heterogeneity, single-cell analysis emerged as a technological breakthrough that may reveal clonal relationships with high precision (261, 262). However, the use of single-cell analysis in clinical practice is still limited by its cost and duration of sample processing.
Liquid biopsy, increasingly used in routine practice, is able to depict tumor heterogeneity and to reveal the spatial hierarchy and temporal distribution of cancer alterations. It effectively identifies resistance alterations (71, 75), and it may perform better than tissue in the identification of complex, polyclonal alterations (263, 264). Also, liquid biopsy may be used to monitor resistance during treatment or minimal residual disease after curative treatments. However, when circulating tumor DNA shedding is low (e.g., low tumor burden, isolated central nervous system progression, lung-only disease, or certain histologic subtypes such as urological cancers; refs. 265–267), more sensitive techniques are required (e.g., whole-genome sequencing of circulating free DNA; ref. 268).
In the case of immunotherapy, nongenomic mechanisms of resistance need to be addressed in conjunction with the genomic profile, where the main limitation is the lack of validated tools to provide a comprehensive view (2). An interesting approach is the investigation of genomic or transcriptomic signatures. The tumor–host–microenvironment interaction could leave specific genomic/transcriptomic scars, which could be identified by DNA-seq and RNA-seq. Examples include the T cell–inflamed TME signature, the transcriptional signatures of fibroblasts, endothelial cells, TGFβ signaling, and MDSCs and regulatory Tregs (156). The comparison of genomic/transcriptomic signatures between responsive and nonresponsive tumors might identify predictors of resistance. For instance, the innate anti–PD-1 resistance signature (IPRES) was found to be correlated with primary resistance to PD-1 blockade in patients with melanoma. IPRES displays enhanced expression of genes involved in mesenchymal transition, extracellular matrix remodeling, angiogenesis, cell adhesion, and wound healing. Notably, the same signature could occur in melanomas with activated MAPK signaling following treatment with TKIs, which could explain in some cases the cross-resistance with anti–PD-1 therapy (269). Another challenge is capturing the heterogeneity of TME immune infiltration, which might be addressed by single-cell analysis and radiomic signatures (e.g., the radiomic signature of CD8-infiltrating T cells; refs. 261, 262, 270).
Nongenomic mechanisms of resistance may also have an important role in tumors failing targeted therapies, especially in the absence of clear bona fide genomic alterations responsible for resistance. In addition to genomic analyses, the integration of a proteomic and epigenetic characterization may expand our knowledge and further improve treatment tailoring.
Defer Resistance and Treat Earlier
A growing tumor develops novel alterations and exhibits increasing genomic instability, clonal diversity, and tumor heterogeneity, which enhance drug resistance even in the absence of prior treatments (Fig. 5B; refs. 12, 181, 271). It is possible that an earlier drug administration might be more effective than a later administration. In favor of this hypothesis is the efficacy of tamoxifen in the adjuvant therapy of localized breast cancer, and osimertinib in the first-line setting of advanced EGFR-mutated NSCLC or even in the adjuvant setting, where it showed an impressive reduction in the risk of relapse (272–274). The lesser subclonal neoantigens and lower tumoral burden in earlier stages may also favor an earlier use of immunotherapy.
Novel-generation drug use in the up-front setting may prevent the emergence of on-target resistance mechanisms seen with older-generation drugs, as successfully exemplified in the FLAURA trial with osimertinib versus first-generation EGFRi in patients with treatment-naïve EGFR-mutated advanced NSCLC (273). However, in this trial and others, deferring resistance alterations is not the only factor responsible for improved clinical outcomes, as the increased intracranial activity of newer-generation drugs has an essential role in oncogene-addicted cancers with increased central nervous system (CNS) tropism (275–277). Thus, it would be interesting to investigate the role played by deferring resistance when comparing drugs that have similar CNS activity.
Other strategies to prevent resistance are the simultaneous inhibition of co-occurring drivers, proved to be effective and safe in the I-PREDICT trial (278), or the concomitant blockade of two proteins of the same oncogenic pathway for which cancer cells show a strong dependency (279). To maximize the expected outcomes, selected treatment should target the predominant cell populations within the tumor, prioritize clonal rather than subclonal target inhibition, and prioritize according to the highest level of actionability of the targets. In the case of sequential therapies, monitoring clonal and subclonal alterations may guide subsequent treatment based on the dominant resistant cell population (Fig. 5B).
Treating Multiple, Highly Resistant Alterations
The treatment of complex, polyclonal aberrations is a major challenge. The sequential use of TKIs may result in a stepwise accumulation of mutations that highly increase drug resistance (56, 263, 280). Compound on-target alterations may induce resistance even to third-generation TKIs. Also, multiple bypass tracks or on-target and bypass alterations might co-occur in the same patient (59). In these cases, it is currently unknown if targets should be ranked for treatment prioritization, or if drug combinations should be used. With the increasing number of acquired alterations, multidrug resistance would probably need different strategies. One example is the administration of nonselective cytotoxic therapies with preferential delivery in the tumor, such as ADCs. The selective intratumoral accumulation of ADCs is based on their ability to target proteins overexpressed in the tumor (281). The novel HER2-targeted ADC trastuzumab deruxtecan recently showed increased efficacy in heavily pretreated HER2-expressing or HER2-mutant solid tumors (282). As tumor heterogeneity is likely to be increased in this stage, the use of ADCs that harbor a bystander effect is likely to have more efficacy.
Manipulating the Mutational Process Might Be More Important than Treating the Genomic Alteration That Mediates Resistance
If preventing resistance is more effective than treating resistance, targeting the mutational process that drives cancer progression might have a paramount importance. Blocking mutagenesis by “therapy by hypomutation” (e.g., A3B inhibition of the APOBEC complex) would be expected to prevent subclone diversification and thus limit the aggressiveness of tumors and drug resistance (94, 283). The therapeutic targeting of ecDNA would share a similar purpose and is worth further investigation (53). In advanced cancers, such strategies should be evaluated in conjunction with the blockade of tumor drivers, where it may increase the treatment duration on targeted treatments, while in early cancer stages potentially as maintenance therapy after local treatments. A reverse strategy is the use of “therapy by hypermutation,” by enhancing mutagenesis until cell destabilization and cell death. This could be useful in advanced, pretreated APOBEC-high tumors that already achieved hard-to-treat genomic alterations. Strategies may use agonists of A3, PARP, and ATR inhibitors (94, 283). A similar strategy may be used in tumors with high chromosomal instability, by using agents that further increase genomic instability (e.g., WEE1 inhibitors in conjunction with DNA-damaging agents).
Demonstrate the Missing Links between Oncogenic Stresses and Immune Suppression
Active research is needed to deepen the current understanding of immune exclusion and suppression associated with certain oncogenic pathways, as to be able to reverse this phenomenon and to integrate immunotherapy in the therapeutic armamentarium of oncogene-addicted tumors. In this domain, research is still in its infancy. Examples of putative players are the expression of CD73 in EGFR-mutated cancers that might predict ICB responses (284), the PTEN-loss PI3K pathway activation that induces PD-L1 expression and promotes resistance to T-cell lysis (285), or the INFα/β-induced p53 gene activation with subsequent cell apoptosis (286). Associations between ICB and epigenetic drugs should be used in a personalized fashion, guided by epigenetic modifications. Epigenetic alterations would need clinical studies to interrogate and validate a potential predictive role for ICB response, as certain mutations have been correlated with ICB response, especially the loss of SWI/SNF components (287, 288). Also, genomic ITH might determine different immune infiltration patterns. For instance, tumor subregions with copy-number gain in chromosome 7 were associated with unfavorable immune composition, with a lack of leukocyte infiltration and presence of activated neutrophils (289). The identification of the molecular-associated immune response in the tumor might be crucial for anticipating resistance patterns and personalizing treatment.
Personalizing Treatment Selection with Immunotherapy
There is an urgent need to personalize the administration of immunotherapy, as the blinded use of ICB and investigational immunotherapies is most likely to fail in a significant proportion of cases where the target is absent. The most illustrative example is the administration of PD-1/PD-L1 inhibitors in tumors with no T-cell infiltration. Another example is the combination of anti–PD-1 with anti-IDO that showed promising antitumor activity in phase I–II trials on biomarker-unselected patients, but the phase III trial showed no benefit of the combination in comparison with single-agent PD-1 inhibitor (290). Would the results be the same in selected patients with high IDO–expressing tumors? Also, strategies for enhancing the function of effector T cells are likely to work in “hot” tumors with T-cell infiltration, but are they effective in tumors featuring immune exclusion or minimal immune cell infiltrates? MHC I–deficient cells are unable to present antigen to effector T cells, but they are susceptible to NK cell–dependent cytotoxicity, which provides a rationale for NK-cell stimulation in such cases (255). Such tumors are also good candidates for CAR-T cell therapy, as CAR-T cells are able to recognize antigens without MHC, whereas TCR therapies could better address tumors with lower antigen densities within the TME (193). Similarly, investigational agents aiming to overcome acquired resistance to immunotherapy are unlikely to succeed without targeting the specific resistance mechanism. Also, as opposed to targeted therapies, advancements in immunotherapeutic drug discovery and research are slowed down by the intrinsic limitations of current preclinical models for testing, which need an intact human TME in order to be useful (291).
Finding better tumor-agnostic biomarkers to predict immunotherapy response is essential to find patients responding to therapy in tumor types not expected to benefit. Tertiary lymphoid structures, recently shown to highly correlate with immunotherapy responses even in tumors with low mutational burden, warrant further investigation in a pan-tumor setting (221). Resistance biomarkers need further validation and ranking, based on their magnitude of prediction. For example, resistance mechanisms found in nonresponding tumors that were highly expected to respond, such as MSI-high tumors, should have a high prediction value of resistance in other tumor types as well.
Patient “immunograms,” first proposed by Blank and colleagues (292), should be further developed to integrate the current state of knowledge (Fig. 6A) and guide treatment choice (Fig. 6B). Trials investigating a personalized approach with combination therapies based on biomarkers are ongoing (e.g., ADVISE, NCT03335540; PIONeeR, NCT03833440). The preliminary results of the PIONeeR trial suggest a predictive value for PD-L1 density on all cell types, PD-L1 tumor expression, infiltrating cytotoxic T cells, and immunosuppressive cell density. Of note, nonresponding patients with high PD-L1 tumors actually had low PD-L1 density on all cell types (tumor and stromal; ref. 293).
Better Define Acquired Resistance to Immunotherapy
An accepted set of criteria to define acquired resistance to immunotherapy is essential for future drug development. Many new immunotherapy compounds have failed because they enroll in clinical trials both patients with primary resistance and patients with acquired resistance to ICB. The lack of consensus for defining resistance means that patients with acquired resistance are defined by some as only patients with partial response/complete response versus others who also account for stable disease (SD) for a certain period. It is totally unclear if SD for 3 or 6 months truly represents a sensitive patient who then acquires resistance. In order to increase the likelihood of response to new immunotherapy molecules, we should probably focus more on patients who truly responded to ICB initially and then developed acquired resistance. Such patients may still have inflamed tumors that can be amenable to a new immunomodulation.
Aim to Cure: The Role of Surgery or Local Ablative Therapies in the Treatment of Metastatic Disease
In oligoprogressive tumors, local therapies are used, as they allow the pursuit of an effective treatment for the rest of the lesions and improve survival outcomes (294–296). They are also discussed in tumor boards in the case of dissociated response under ICB, but prospective studies are still pending to confirm benefit (297). The choice between surgery or local ablative therapies still needs refinement, especially in selected oligoprogressive patients with a long disease control under treatment. Surgery remains the most effective treatment for ITH, hypoxic tumors, and persister cells, and these might be the exact triggers behind progression. However, the selection of the ideal candidates is challenged by the difficulty of predicting the behavior of the rest of the lesions. Innovative methods predicting relapse, based on genomics and radiomics, would be desired. Also, radiotherapy could be modulated based on the tumor genomic and microenvironmental features, as well as its immune infiltration. Another direction of research is to investigate the role of a “consolidation” therapy for patients responding to treatment in a heterogeneous manner. The less responsive lesions might exhibit genomic/immune heterogeneity and be the culprit of subsequent local or distant progression. In the case of ICB, the addition of a local treatment of liver metastases might also be beneficial, especially when considering recent hypotheses of systemic immunosuppression induced by liver metastases, which might impair ICB efficacy.
Having all this in mind, it is clear that despite major progress in cancer care, drug resistance evolves dynamically and remains the most important barrier to achieving cure. A deeper understanding of the processes underlying resistance and more potent drugs and innovative strategies are urgently needed.
F. Andre reports grants from Novartis, Roche, Daiichi, Eli Lilly, Astra Zeneca, and Pfizer outside the submitted work. A. Marabelle reports grants, personal fees, non-financial support, and other from AstraZeneca, BMS, MSD, and Roche/Genentech and personal fees, non-financial support, and other from Pfizer, Merck Serono, and Novartis outside the submitted work. F. Barlesi reports personal fees from AstraZeneca, Bayer, Bristol-Myers Squibb, Boehringer-Ingelheim, Eli Lilly Oncology, F. Hoffmann-La Roche Ltd, Novartis, Merck, Mirati, MSD, Pierre Fabre, Pfizer, Seattle Genetics, and Takeda outside the submitted work. J.-C. Soria reports personal fees from Relay Therapeutics, and other from AstraZeneca, Daiichi Sankyo, Gritstone, and Hookipa Pharmaceuticals during the conduct of the study. No other disclosures were reported.
The figures are previously unpublished original works, created by Mihaela Aldea and Semih Dogan with BioRender.com, for the express purpose of publication in this Cancer Discovery review.