Synthetic Lethality in Cancer Therapeutics: The Next Generation

Abstract

Synthetic lethality (SL) provides a conceptual framework for tackling targets that are not classically ‘druggable,’ including loss-of-function mutations in tumor suppressor genes required for carcinogenesis. Recent technological advances have led to an inflection point in our understanding of genetic interaction networks and ability to identify a wide array of novel SL drug targets. Here, we review concepts and lessons emerging from first-generation trials aimed at testing SL drugs, discuss how the nature of the targeted lesion can influence therapeutic outcomes, and highlight the need to develop clinical biomarkers distinct from those based on the paradigms developed to target activated oncogenes.

INTRODUCTION

The implementation of the genetic concept of synthetic lethality (SL) in cancer therapy holds great promise but remains in its infancy (1,2). Recent technological advances in high-throughput forward genetic and chemogenetic screening in human cells have led to a rapid increase in our understanding of genetic interaction networks and ability to identify novel SL drug targets. Successive waves of focused and large-scale CRISPR-based screens are producing a profound increase in the number of candidate SL targets (3–11), many of which are currently in drug development pipelines. As such targets reach the later stages of preclinical validation and therapeutic development, clinical trials aimed at testing this new generation of SL therapeutics are poised to undergo a dramatic increase in number and scope within coming years.

Here, we review concepts and lessons emerging from first-generation clinical trials aimed at testing SL drugs, including inhibitors of the poly (ADP-ribose) polymerase (PARP) in the treatment of BRCA1/2-deficient tumors, an archetypal example of SL anticancer therapy, and highlight how the nature of the targeted lesion and its genetic context can influence efficacy and durability of response. Furthermore, we discuss the need to depart from existing paradigms for biomarker development, which were primarily developed for drugs targeting activated oncogenes, and emphasize additional factors likely to influence the success of future efforts to apply SL in cancer therapy.

The concept of SL can be traced back to initial observations made almost 100 years ago by the American geneticist Calvin Bridges, who noted that when crossing fruit flies (Drosophila melanogaster), inheritance of specific mutant gene pairs were lethal in combination, despite inheritance of either mutant gene alone resulting in perfectly normal viability. Further contemporary development of knockout libraries and systematic genetic interaction studies in yeast model systems led to refinement of SL, as defined by cellular or organismal lethality caused by combined alterations of gene pairs that are otherwise individually viable (Fig 1a). With respect to cancer therapy, the definition of SL has since been expanded to encompass pharmacologic inhibition of one gene product with inactivation of the other in cancer cells (12). If one considers chemical inhibition of a gene product as akin to genetic loss of function (LOF), drugs acting through an SL mechanism can be thought to target genetic alterations through the modulation of the product of a second unaltered gene. It should be noted, however, that the term SL has also been employed in the cancer therapeutics arena for the targeting of gain-of-function (GOF) alterations in cancer cells that are not amenable to direct targeting (e.g. targeting of mutant KRAS (1), or amplification of CCNE1 (13)).

Figure 1. Synthetic lethality: definition and approaches for the identification of synthetic lethal interactions.

A) Synthetic lethality is defined by cellular or organismal lethality caused by combined alterations of gene pairs that are otherwise individually viable. A commonly employed and therapeutically relevant definition encompasses pharmacologic inhibition of one gene product with genetic inactivation of the other. B) Identification of SL drug targets has been facilitated by high-throughput genetic and chemogenetic screening in human cancer cell lines. The use of isogenic models in forward CRISPR-based screens minimizes potential confounding by co-occurring genetic alterations and facilitates attribution of a cellular phenotype to a specific SL pair. Chemogenetic screens appear to be particularly efficacious in developing new patient-selection hypotheses for compounds with known mechanism-of-action (9,113–116) or to uncover potential mechanisms of resistance (117,118). C) Overview of the Cancer Dependency Map (DepMaP) a large-scale multi-institution functional genomics project aimed at creating a comprehensive database of potential novel drug targets and biomarkers across cancer types. A recent integrated analysis of CRISPR-based screens from the Cancer Dep Map effort identified >1000 candidate genetic dependencies across 786 cell lines representing 42 cancer types (119). Abbreviations: KO, knocked out; WT, wild-type; RPPA, reverse phase protein array; FDR, false discovery rate; GDSC, Genomics of Drug Sensitivity in Cancer; PRISM, Profiling Relative Inhibition Simultaneously in Mixtures; CTRP, Cancer Therapeutics Response Portal.

Genetic Interaction Networks Underpin the Molecular Basis for SL Therapeutics

Initial efforts to develop personalized targeted anticancer therapies largely focused on the inhibition of proteins encoded by GOF alterations in driver oncogenes. In contrast, restoring the function of proteins encoded by inactivated tumor suppressor genes has proven to be markedly more challenging, providing a rationale to search for genes displaying SL with tumor suppressor genes. The prospective identification of targets that can lead to development of SL-based drugs has been enabled by large-scale tumor sequencing efforts that have now catalogued the somatic genetic alterations of cancer, technological advances in forward genetic and chemogenetic screens in human cells (Fig 1b, (14–16)), as well as large-scale cooperative efforts to map genetic dependencies across cancer cell types (Fig 1c). In particular, the emergence of CRISPR-based tools and the increasingly varied ways in which they can facilitate functional genomics has greatly accelerated the rate and robustness of SL target discovery (1).

Several large-scale functional genomic screening initiatives aimed at identifying genetic dependencies across hundreds of cancer cell lines have identified hundreds of potentially novel drug targets (3,17). These approaches have taken advantage of large-scale cancer cell line profiling to interrogate genetic interaction networks across intentionally diverse genomic contexts and uncover genetic essentialities conferred by specific genetic alterations (15). Notably, prior analyses of context-specific genetic dependencies identified in pan-cancer cell line LOF genomic screens have shown them to be highly enriched for SL interactions (18,19). Specific efforts include Project Achilles (Broad Institute, (20)), Project DRIVE (Novartis (18)), and Project Score (Sanger Institute (3,21)), each of which has contributed data to the Cancer Dependency Map (DepMap) (22), a collaborative project aimed at creating a comprehensive database of potential novel drug targets and biomarkers (Fig 1c). The resources generated by such large-scale approaches for target discovery are enabling the characterization of genetic interaction networks and the identification of potentially druggable SL cancer targets such as WRN in microsatellite-unstable tumors (11,23).

Despite the size of the cancer cell line panels used in large-scale genomic screening initiatives, this approach nevertheless has some limitations. For example, available cell panels often lack a corresponding normal ‘control’ cell line or have poor representation of certain genetic alterations or of certain tumor types. The identification of SL interactions in cell line panel screens therefore relies on comparing the growth of cells with a given (and represented) genetic alteration with those that do not. The attribution of a cellular phenotype to a specific SL pair can also be difficult to establish when statistical power is low due to underrepresentation of a genetic alteration in the cell line panel (15). Isogenic models, which vary only in the genetic alteration of interest, provide an attractive alternative as it is much easier to infer synthetic lethality relationships from screens in isogenic cell line pairs, and they can be used to probe any genetic alteration of interest (24). Isogenic models are of course not devoid of limitations; engineering specific genetic alterations outside of the typical cellular contexts in which they occur may lead to phenotypes that diverge from the human tumors they seek to model. Furthermore, these models do not take into account the potential impact of the chronology of the acquisition of specific genetic alterations on the dependency of a cell on a specific molecular pathway (25); hence, the biological impact of the genetic alteration edited into a cell may not be entirely reflective of its naturally occurring counterpart. As no model system is perfect, the combined use of sequenced cell line panels, isogenic cell line pairs, and animal models provide complementary and orthogonal validation of candidate genetic interactions.

The continued development of functional genomic technologies has leveraged the flexibility and scalability of CRISPR-based forward genetics to precisely edit DNA or RNA and to modify sequence-specific regulatory processes (26). Engineering of CRISPR-Cas proteins or guide RNAs (gRNAs) has allowed for the development of Cas variants capable of recruiting additional enzymatic activities to specific genomic sequences in scalable fashion, and led to CRISPR-Cas platforms for base editing, direct RNA editing, epigenetic modification, and inhibition (CRISPRi) or activation (CRISPRa) of transcription, among other functions (27). This increased flexibility has enabled a diverse array of methods for dissection of genetic interaction networks and SL target discovery. For example, the use of orthologous CRISPR-Cas nucleases and gRNAs has been shown to facilitate ‘one-by-all’ genetic interaction screens in which a uniform ‘anchor’ gRNA is coupled with a variable ‘query’ gRNA to systematically identify alterations that are SL with the ‘anchor’ of interest (28,29). This approach can be particularly useful when the ‘anchor’ genetic alteration cannot be modeled using single cell-derived isogenic clones due to cellular essentiality (29). Although originally developed to study regulatory elements and long non-coding RNAs (lncRNAs) through generation of kilobase-long deletions (30,31), the development of dual- or multi-gRNA libraries has additionally facilitated multiplexing of ‘query’ gRNAs to uncover genetic interactions that would otherwise be obscured due to functional redundancy between gene paralogs (32,33). Moreover, the integration of tools for both GOF (CRISPRa) and LOF (CRISPRi or CRISPR-mediated knockout) perturbations has facilitated the development of combined bidirectional screens for the concurrent identification of genetic interactions involving LOF and/or GOF alterations (28,34,35).

Emerging data have also begun to demonstrate the promise of CRISPR-based perturbations when combined with multi-omic readouts, as exemplified by Kelly et al. (36), who demonstrated the promise of coupling broad discovery screens with more focused orthogonal validation screens. By combining protein-protein interaction mapping (using affinity purification/mass spectrometry) with a focused custom CRISPR knockout screen, the authors dissected functional interactions within the RAS pathway, and in doing so, identified novel genetic dependencies mediated by both direct protein-protein interactions and indirect functional genetic interactions (36). The development of single-cell technologies to assay the transcriptome of individual cells while simultaneously retrieving guide RNA sequences present in each cell has also allowed investigators to link CRISPR-mediated perturbations with transcriptional phenotypes (37,38). Additional recent work has highlighted the use of 3D culture systems in genetic interaction screens in improving the recapitulation of in vivo phenotypes as compared to traditional 2D screen, as illustrated by the their greater accuracy in the assessment of growth advantage resulting from the knocking-out of tumor suppressor genes (39). As evidenced by these studies, the development of appropriate workflows can leverage the flexibility and scalability of CRISPR-based forward genetics to systematically connect an increasing variety of perturbations with specific phenotypes in 2D, 3D and in vivo model systems (40,41).

The effectiveness of CRISPR-based approaches to SL target discovery has led to burgeoning interest in the adaptation of such technology to screen for non-cell autonomous therapeutic vulnerabilities. High-throughput in vivo CRISPR screens in primary CD8+ T-cells have been shown to allow for the discovery of negative regulators of anti-tumor T-cell immunity, as evidenced by their ability to recover known immunotherapy targets such as PD-1 and Tim-3 (42). Conversely, in vivo application of CRISPR-based screening tools may also prove to be useful in identifying cancer cell targets that stimulate host immunity and/or synergize with checkpoint blockade; in addition to recovering PD-L1, a recent in vivo genetic screening approach was used to identify unanticipated candidate genes with potential roles in mediating immune evasion (43). The continued development of such approaches has the potential to expand existing definitions of SL to encompass inter-cell genetic interactions and non-cell autonomous mechanisms.

Biomarkers for SL: When the Presence of a Mutation May Not Be Enough

The development of poly (ADP-ribose) polymerase (PARP) inhibitors as anticancer therapy for BRCA1/2-deficient tumors has provided an archetypal example of efforts to harness SL for therapeutic gain (44). Highlighting the substantial promise of SL therapeutics, recent data have demonstrated dramatic improvements in overall survival afforded by the use of PARP inhibition for BRCA1/2-deficient ovarian cancer (45). At the same time, PARP inhibitors have produced limited effects on survival in late-stage clinical trials of patients with metastatic breast, prostate, and pancreatic cancers (46–48). The explanation of such disparate and seemingly conflicting results, in combination with first principles gleaned from key preclinical studies, has begun to reveal factors potentially critical to the success of future efforts to apply SL in the clinic (Fig 2).

Figure 2. Conceptual framework for optimized prioritization of synthetic lethal therapeutic approaches.

Germane to the successful translation of synthetic lethal interactions into cancer treatments is the consideration of characteristics of the gene altered in the cancer cells including whether its loss is mono- or bi-allelic or biologically sufficient to cause a phenotype in the context of a synthetic lethal interaction, the prevalence of its loss in a given cancer type or across cancers and whether the loss of the gene is essential for tumor development and/or maintenance. The features of the target gene (i.e. gene to be inhibited therapeutically) also need to be considered, included its expression in the cell lineage and/or cancer type of interest and the toxicity impact of its inhibition. The characteristics of the synthetic lethal interaction itself also need to be considered, including the effect size (magnitude of the therapeutic index in preclinical models) and penetrance. Synthetic lethal interactions may be limited to a specific genetic context or tissue type/lineage; assessing the penetrance of the genetic interaction across varying model systems and cell lineages can inform the robustness of the therapeutic window.

A very important variable to consider for successful application of SL-based therapies is the zygosity of the target genetic alteration, as exemplified by prior experience with PARP inhibitors. Whilst GOF mutations affecting proto-oncogenes typically act dominantly, LOF alterations affecting tumor suppressor genes tend to be recessive, requiring biallelic inactivation for promotion of tumorigenesis. Homozygous or biallelic inactivation of a tumor suppressor most commonly involves a deleterious mutation of one allele followed by inactivation of the wild-type (WT) allele via loss of heterozygosity (LOH), a second somatic mutation or structural rearrangement; a homozygous deletion (i.e., physical genetic loss of both alleles) or epigenetic silencing. As the majority of tumor suppressors require biallelic LOF to promote tumorigenesis and therapeutically relevant phenotypes (49), the optimal development of SL therapies targeting such genes will likely require biomarkers that accurately assess LOH and zygosity. For example, the vast majority of trials examining PARP inhibition in genotypically selected patients have not included prospective assessment of biallelic gene inactivation (46,48,50–55), resulting in the likely inclusion of patients with monoallelic mutations who are less likely to respond. In a retrospective ad hoc analysis of one such trial (TRITON2 phase II study of rucaparib in metastatic prostate cancer), the rate of PSA response among patients with mono- vs biallelic BRCA1/2 inactivation was found to be 11% vs 81%, respectively (56).

This finding is consistent with multiple other lines of evidence suggesting that the presence of a single wild-type BRCA1/2 allele appears to be sufficient for most therapeutically-relevant phenotypes, including DSB repair. Although haploinsufficiency has been described for a subset of BRCA1- and BRCA2-related functions—namely suppression of aneuploidy and replication stress (57,58)—evidence to support a lack of functional HR deficiency in tumors with heterozygous inactivation of BRCA1 or BRCA2 includes preclinical functional data derived from both in vitro (59,60) and in vivo models (60–63). The presence of a single functional allele also appears to be sufficient for proficient DSB repair in human patients, as evidenced by i) the observation that HR-deficient genomic scarring is almost exclusively limited to tumors harboring biallelic BRCA1/2 inactivation (64–67), ii) PARP inhibitor and platinum resistance in tumors harboring heterozygous reversion mutations (68–70), and iii) the aforementioned TRITON2 analysis indicating a markedly poor response rate among patients with mono-allelic BRCA1/2 inactivation. (55). Notably, the above concept is not solely pertinent to HR-deficiency. Further examples of SL therapies thought to specifically target biallelic genetic inactivation include AURKA/B inhibition in RB1-mutant tumors (71,72), WRN inhibition in tumors with microsatellite instability induced by biallelic inactivation of mismatch repair genes (11,73), and PRMT5/MAT2A inhibition in tumors with biallelic collateral loss of MTAP (74).

Although these data highlight the importance of LOH assessment in targeting genes that require biallelic inactivation for loss of function, it is important to note that a subset of tumor suppressors are known to be haploinsufficient and can produce targetable phenotypes with loss of only a single allele (75). The complexity of genotype-phenotype relationships is further illustrated by the observation that some genes may be haploinsufficient for only a subset of related phenotypes (41), whereas others (e.g. TP53) can harbor alleles with either LOF or dominant negative activity, depending on the specific mutation (50). With this in mind, we suggest that, apart from the context of SL therapies that target haploinsufficient genes and genetic alterations producing a dominant-negative phenotype, a robust assessment of biallelic gene inactivation as a biomarker will be necessary for the successful development of many, if not most, SL-based therapeutic strategies.

Acquired resistance is another phenomenon that affects PARPi efficacy in a major way. Among multiple resistance mechanisms that have been identified to date, the most frequently detected mechanism involves the restoration of BRCA1/2 function via secondary reversion mutations or intragenic deletions that restore the open reading frame. Under the selection pressure imposed by PARP inhibition, reversion mutations that restore functional homologous recombination (HR) provide tumor cells with a marked fitness advantage upon PARP inhibitor treatment, which in turn leads to acquired resistance and tumor progression. Reversion mutations have similarly been identified in patients with prior exposure to platinum-based chemotherapy, suggesting convergent resistance mechanisms for therapies that target defective HR (69,76). A recent study reported by Pettitt et al. demonstrated the existence of reversion ‘hotspots’ and ‘deserts’ in specific regions encoding BRCA2, suggesting that the likelihood of reversion is affected by the position of the original mutation (77). Interestingly, as reversion mutations that do not restore the same codon as the original mutation frequently result in short stretches of frameshifted neopeptides, the authors of this study postulated that some revertant alleles encode neoantigens that may increase immunogenicity and vulnerability to immune checkpoint blockade, although this possibility is yet to be tested experimentally (77). Of note, homozygous deletion, at variance with combined mutation and LOH, prevents the possibility of secondary mutations that restore the reading frame (unless only a portion of the gene is deleted). Unsurprisingly, long-term responders to PARP inhibition appear to be enriched in homozygous deletions or rearrangements in BRCA1/2 that prevent the clonal outgrowth of resistant clones harboring somatic reversion mutations (78,79). Deletions occurring at tumor suppressor loci can also extend beyond the bona fide driver genes, which may allow identification of genetic or pharmacological sensitivities caused by gene loss that is collateral to the actual cancer-driving event (80,81).

A recent study examining the therapeutic relevance of mutations in BRCA1/2 has suggested that tumor lineage may also be a relevant factor in influencing the penetrance of HR-deficient phenotypes (82). Examination of genomic scarring and responses to PARP inhibition revealed a significant interaction between cancer type and the phenotypic penetrance of deleterious mutations in BRCA1/2 (82). Although tumors with mutant BRCA1/2 arising from the breast, ovary, prostate, or pancreas (i.e., BRCA-associated cancer types) and non-BRCA associated cancer types both displayed higher levels of genomic signatures associated with HR deficiency, in a limited number of patients only BRCA-associated cancer types appeared to derive clinical benefit from PARP inhibition (82). Similar lack of efficacy for PARP inhibition was observed in a phase II randomized trial of rucaparib for genetically unselected patients with recurrent or metastatic bladder cancer (83). As other studies have demonstrated evidence to suggest that biallelic HR gene inactivation is penetrant in non-BRCA associated cancer types (5,61,62,65,84), these data have led to controversy surrounding the role of lineage in BRCA1/2-associated phenotypes (82,85). Nevertheless, a parsimonious interpretation of the lack of clinical benefit seen with PARP inhibition in non-BRCA associated cancer is that such results are confounded by higher rates of LOH in cancers originating from BRCA-associated cancer types than in other malignancies, given that BRCA-associated tumors are more than twice as likely to display biallelic inactivation than non-BRCA-associated cancers (90% and 44% for BRCA-associated tissue types and non-BRCA associated cancer types, respectively) (86).

Taken together, findings discussed in this section suggest that the convenience of enrolling patients for trials testing SL-based therapies merely on the basis of the presence of a pathogenic mutation affecting the target gene, in a way akin to how the enrollment of patients in studies testing small molecule inhibitors or antibodies targeting oncogenes activated by a given hotspot mutation, would be expected to result in modest therapeutic benefit. Patient populations defined on this basis would be heterogeneous and the probability of a targetable phenotype would likely vary according to the frequency of loss of the WT allele in a given lineage. The inability to assess the zygosity of LOF alterations is a clear limitation for many current biomarker platforms, including those approved as companion diagnostics for PARP inhibitors (87,88).

Additional promising approaches to identify patient populations likely to benefit from SL therapies include the use of functional or phenotypic biomarkers. Such measures may be particularly valuable in addressing the large number of variants in tumor genomes that are of undetermined functional significance, while additionally facilitating the identification of tumors that become therapeutically vulnerable through epigenetic mechanisms. In the well-studied example of HR-deficiency, phenotypic biomarkers include genomic ‘scars’ detected by sequencing or single nucleotide polymorphism array (i.e., large-scale state transitions, COSMIC signature 3, HRDetect) (64,66,89,90), and ex vivo assays of HR proficiency that rely on detection of damage-induced nuclear foci (i.e., RAD51 focus formation assay) (64,91). Although these approaches have shown promising results in specific scenarios, each has important trade-offs and limitations. For example, markers that quantify genomic ‘scarring’ have shown to be sensitive for the detection of tumors harboring biallelic HR gene inactivation (65,66), but have nevertheless been limited by modest positive-predictive value and the inability to provide a ‘real-time’ readout of phenotype, as the emergence of drug resistance does not remove the existing ‘scars’ of prior HR-deficiency. In contrast to genomic approaches, functional assays of HRD have the potential to provide a dynamic, real-time readout of DNA repair capacity. The most promising functional markers involve the measurement of downstream changes in protein localization that reflect the fidelity of multiple upstream components. For example, the accumulation of RAD51 into nuclear foci (a marker of intact HR) in response to DNA damage has long been considered a gold standard for evaluating HR function in preclinical models. We and others have demonstrated that assessment of RAD51 foci can provide a robust binary readout of HR proficiency versus deficiency (64,91) that accurately reflects both drug response and acquired resistance (92,93), highlighting its potential as a dynamic biomarker of HR integrity. Nevertheless, the clinical development and broader adoption of such assays has to date been hampered by the need for fresh tissue and ex vivo manipulation of cancer cells. The development of assays that obviate such logistical hurdles will be key to the clinical implementation of functional biomarkers aimed at providing a dynamic readout of therapeutic vulnerability.

LOF Alterations Required for Tumor Initiation Versus Maintenance

The application of evolutionary theory to cancer has provided a theoretical framework for understanding the molecular processes responsible for the malignant state and how such processes are constrained by selection (25). Moreover, evolutionary theory has provided important lessons relevant to the identification of robust SL interactions and strategies to pair exploitable target alterations with tractable drug targets. For example, the effect of PARP inhibition on the clonal evolution of BRCA1/2-deficient tumors suggests that loss of HR proficiency may be critical to promoting early stages of tumorigenesis, but dispensable for maintenance of the malignant phenotype (94). Reversion mutations that restore the function of BRCA1/2 do not appear to impair tumorigenesis at later stages of tumor evolution. Such observations indicate that BRCA1/2 inactivation causes genomic instability that promotes the acquisition of additional genetic alterations, rather than being essential for the transformed cell phenotype. This is in contrast to genetic alterations required for tumor maintenance that may conversely maximize the fitness penalty incurred by resistance mechanisms to SL anticancer therapies.

Genetic alterations required for tumor maintenance were initially identified using in vitro and in vivo models of GOF tumorigenesis (95,96). In seminal studies, acute inactivation of transforming oncogenes, including MYC, KRAS, and ABL (among others) was found to result in differentiation, apoptosis, or cell cycle arrest, depending on tissue context (97–99). This reversal of malignant phenotype led to the concept of ‘oncogene addiction,’ which forms the basis for most targeted anticancer therapies. It has since been demonstrated that genetic alterations required for tumor maintenance are not limited to GOF alterations. Inactivation of APC, for example, is required for the maintenance of colorectal cancer, as evidenced by murine models with established colorectal tumors in which APC restoration drives rapid tumor cell differentiation and sustained regression without evidence of relapse (100). LOF alterations required for maintenance of tumor progression extend beyond those that inactivate APC; inducible restoration of WT p53 protein in established tumors caused by loss of p53 function has been shown to lead to tumor regression in most genetic contexts (101–103).

Growing evidence suggests that inactivation of ataxia telangiectasia mutated (ATM) as a tumor suppressor gene is required not only for initiation but also for maintenance of malignant phenotypes. In addition to its role in DNA damage signaling, ATM has a key gatekeeper role via regulation of p53 stabilization in response to cellular stress (104); inactivation of ATM appears to suppress p53 signaling to a degree that is sufficient for tumorigenesis, evidenced by the observation that LOF alterations in ATM and TP53 are mutually exclusive across cancer types (105,106). Furthermore, a recent study using a conditional murine model has shown that ATM reactivation in initially ATM-deficient B- and T-cell lymphoma results in sustained tumor regression (107). Consistent with this notion, early-phase clinical data examining synthetic lethal strategies based on ATR inhibition in ATM-deficient tumors have revealed long and durable responses in heavily pretreated cancer patients (108,109), a clinical observation distinct from those seen with PARP inhibition in BRCA1/2-mutated tumors. Albeit limited, these data are consistent with the notion that targeting LOF alterations required for tumor maintenance can enhance the likelihood of durable tumor control. We thus anticipate that the pattern of ATR inhibitor efficacy in patients with tumors with ATM LOF will be different from that observed with PARP inhibitors, reinforcing the biological differences in these two pairs.

CONCLUSIONS & FUTURE DIRECTIONS

Recent technological advancements have greatly accelerated the process of identifying tumor-specific genetic dependencies and candidate SL targets. Despite a marked increase in the rate of target discovery, relatively few SL drugs have been tested in the clinic, and the field remains largely in its infancy. As the number of novel SL drugs entering clinical trials continues to grow, we review concepts and lessons emerging from the first wave of clinical trials aimed at testing SL drugs, including inhibitors of the poly (ADP-ribose) polymerase (PARP) for BRCA1/2-deficient tumors, an archetypal example of SL anticancer therapy. The continued maturation of clinical data in this space has begun to provide important insights into how the nature of the targeted lesion and its genetic context can influence clinical outcomes, and highlights the need to develop clinical biomarkers distinct from existing platforms developed for the targeting of activated oncogenes. In particular, the development of companion diagnostics capable of assessing biallelic gene inactivation, and/or clinically relevant phenotypes, will be key to further progress in this arena. The next few years are likely to provide a roadmap for how the next generation of synthetic lethal interactions can be exploited therapeutically in the context of combinatorial immuno-oncology treatments. Emerging data examining how distinct elements of the DDR influence tumor immunity (110,111), and how SL interactions may result in immunogenic cell death (112), suggest the possibility of strong synergy between these two classes of therapeutics. In combination with ongoing efforts to expand the known spectrum of SL genetic interactions, a strong foundation is being laid for the development of effective SL therapeutics and novel combinatorial regimens exploiting SL interactions.

STATEMENT OF SIGNIFICANCE.

Synthetic lethality, as applied to cancer therapy, targets tumor-specific alterations not amenable to direct targeting, including selected oncogenes, such as those targeted by gene amplification or protein overexpression that results in inactivation of the same pathways controlled by tumor suppressors, or genes whose tumor-specific loss of function alterations result in greater vulnerability to perturbation of a second pathway. Identifying and targeting synthetic lethal interactions has been accelerated by high-throughput CRISPR technology, opening up a next generation of approaches with a greater opportunity for selective tumor cell killing that can translate into drug development and novel biomarkers for cancer clinical trials.