Therapeutic targeting of mismatch repair-deficient cancers

Abstract

DNA mismatch repair (MMR) is one of many evolutionarily conserved processes that act as guardians of genomic integrity. MMR proteins recognize errors that occur during DNA replication and initiate countermeasures to rectify those mistakes. MMR deficiency (MMRd) therefore leads to a dramatic accumulation of mutations. The MMRd genomic signature is characterized by a high frequency of single-base substitutions as well as insertions and/or deletions that preferentially occur in short nucleotide repeat sequences known as microsatellites. This accumulation leads to a phenomenon termed microsatellite instability, which accordingly serves as a marker of underlying MMRd. MMRd is associated with hereditary cancer syndromes such as Lynch syndrome and constitutional MMRd as well as with sporadic tumour development across a variety of tissues. High baseline immune cell infiltration is a characteristic feature of MMRd/microsatellite instability-high tumours, as is the upregulation of immune checkpoints. Importantly, the molecular profile of MMRd tumours confers remarkable sensitivity to immune-checkpoint inhibitors (ICIs). Many patients with MMRd disease derive durable clinical benefit when treated with these agents regardless of the primary tumour site. Nevertheless, a substantial subset of these patients will fail to respond to ICI, and increasing research is focused on identifying the factors that confer resistance. In this Review, we begin by discussing the biological function of the MMR machinery as well as the genomic sequelae of MMRd before then examining the clinical implications of MMRd with a specific focus on cancer predisposition, diagnostic approaches, therapeutic strategies and potential mechanisms of resistance to ICIs.

Key points

• Immune-checkpoint inhibitors (ICIs) confer remarkably durable clinical benefit in many patients with DNA mismatch repair-deficient (MMRd) tumours.

• MMRd tumours are thought to be responsive to ICIs because they harbour many single-base substitutions and frameshift mutations, which, if expressed, have the potential to encode tumour-specific immunogenic neoantigens.

• Immune-mediated killing of MMRd cancer cells can be orchestrated by various effector cells, enabling MMRd tumours to respond to ICIs despite major histocompatibility complex (MHC) class I loss.

• Most patients with MMRd tumours derive benefit from ICIs, although a substantial number have primary resistance and many more develop acquired resistance.

• Many potential predictors of response and resistance to ICIs are under active investigation, but none are currently ready for clinical implementation.

• The accurate diagnosis of MMRd status is an important determinant of ICI response. This is best achieved through a multimodal approach that involves immunohistochemical analysis of mismatch repair protein expression and microsatellite profiling.

Introduction

The history of mismatch repair deficiency (MMRd) traces back over 100 years when, in 1895, pathologist Aldred Scott Warthin began to characterize the high prevalence of cancer in a family with an organ-specific tumour predisposition that would ultimately be recognized as Lynch syndrome¹. Our understanding of DNA mismatch repair (MMR) and the consequences of MMRd has grown considerably since then. This increase in knowledge reflects tremendous advances over the past six decades beginning with the identification of prokaryotic and eukaryotic MMR genes and culminating, more recently, with the discovery that MMRd is a tumour agnostic predictor of response to immune-checkpoint inhibitors (ICIs)^2,3,4.

The primary role of the MMR machinery is to recognize and repair single-base substitutions and insertions and/or deletions (indels) that spontaneously develop during DNA replication³. MMRd therefore leads to a substantial accumulation of mutations over time^5,6. These mutations preferentially occur in short repetitive sequences of DNA known as microsatellites and manifest as microsatellite instability (MSI) when indels create novel alleles of an altered length. Thus, MMRd cancers can be distinguished by their high tumour mutation burden (TMB) and high MSI (MSI-H) status^1,7 (Fig. 1). MMRd/MSI-H is observed across a variety of tumour types with the highest prevalence in endometrial, small bowel, gastric and colorectal cancers (CRCs)^8,9,10. This characteristic is associated with sporadic tumour development as well as with multiple hereditary cancer syndromes such as Lynch syndrome and constitutional MMRd (CMMRD)^1,11,12. In addition to the accumulation of mutations in coding regions, which results in a high number of immunogenic neoantigens, MMRd tumours are associated with a high level of baseline immune cell infiltration and the concomitant upregulation of immune checkpoints^7,13. These molecular features are believed to contribute to the exquisite sensitivity of MMRd/MSI-H tumours to ICIs. Nevertheless, many patients with such tumours do not respond to ICIs, and many who initially derive benefit from treatment develop acquired resistance^{7,14,15,16,17,18,19,20} (Box 1).

Fig. 1: Mechanisms of MMR loss and the corresponding genomic and immunological sequelae.

DNA mismatch repair deficiency (MMRd) arises from biallelic inactivation of one of the four mismatch repair (MMR) genes. Loss of MMR function can occur through three different mechanisms. Sporadic MMRd describes the somatic inactivation of both alleles of an MMR gene. Lynch syndrome is a hereditary cancer predisposition syndrome characterized by germline heterozygous inactivation of one allele of an MMR gene; MMRd develops following somatic loss of the remaining wild-type allele. Finally, constitutional MMRd arises from germline biallelic inactivation of an MMR gene. MMRd can be diagnosed using immunohistochemistry (IHC). MSH2 and MLH1 are obligatory binding partners that stabilize their respective heterodimers. MSH2 loss therefore manifests as loss of MSH2 and MSH6 on IHC, and MLH1 loss manifests as loss of both MLH1 and PMS2. By contrast, MSH6 and PMS2 mutations lead to isolated losses of the respective proteins. Although the timing of biallelic MMR inactivation remains unclear in sporadic MMRd cancers, data from patients with Lynch syndrome indicate that small adenomas can be MMR proficient. Larger adenomas are more likely to be MMRd but can still be microsatellite stable. More advanced adenomas and colorectal cancer can then become microsatellite instability high (MSI-H). Once MMRd, cells can accumulate a large number of mutations over the course of time and many divisions. These tend to be single-base substitutions and insertions and/or deletions (indels), which can occur throughout the genome although microsatellites are particularly susceptible owing to their propensity for strand slippage during DNA replication. Indels that cause frameshift mutations within coding-region microsatellites are more likely to have functionally important consequences (such as inactivation of tumour-suppressor genes). Moreover, frameshift mutations can give rise to foreign-appearing peptides that are likely to contain immunogenic neoepitopes, which are thought to elicit the robust antitumour immune responses seen in MMRd cancers. CMMRD, constitutional mismatch repair deficiency; TMB-H, high tumour mutational burden; TILs, tumour-infiltrating lymphocytes.

In this Review, we first describe the biology of the MMR system before next examining the genomic sequelae and oncogenic events associated with MMRd. We then explore the clinical implications of MMRd/MSI-H. After reviewing hereditary and sporadic MMRd syndromes, we turn to diagnostic considerations and therapeutic strategies. Our discussion of the latter initially focuses on the efficacy of ICIs and potential mechanisms of resistance. Finally, we consider additional therapeutic interventions such as preventative measures for patients with hereditary MMRd.

Box 1 Notable features of MMRd tumours

• In MMRd tumours, the tumour mutational burden (TMB) and degree of microsatellite instability (MSI) evolve over time. A high TMB and high MSI (MSI-H) are genotypes that develop in the context of mismatch repair deficiency (MMRd). MMRd tumours can harbour a large number of single-base substitutions and insertions and/or deletions (indels), which preferentially occur in microsatellite regions, although these accumulate over time. Nascent MMRd tumours are therefore characterized by a low TMB and microsatellite stability and such tumours can become high TMB and MSI-H after months to years of division^5,6,98. Collectively, this progressive accumulation of alterations can lead to heterogeneity in MMRd status, TMB and the degree of MSI across tumour types.

• MMRd tumours are typically highly inflamed but can have an immune exclusion histology. MMRd tumours are often enriched with tumour-infiltrating lymphocytes at baseline and characterized by high levels of interferon signalling pathway activation and by upregulation of immune-checkpoint proteins^13,85. In MMRd colorectal cancers, the immune effectors seem to be primarily localized at the tumour–stroma interface^305,348.

• PD-L1 expression is predominantly confined to immune cells. In MMRd tumours, PD-L1 is usually minimally expressed on cancer cells and instead primarily expressed on myeloid cells present within the tumour microenvironment^155,305,307.

• Sensitivity to immune-checkpoint inhibitors (ICIs) is tumour agnostic. Although sensitivity to ICIs can be observed across MMRd cancers, the magnitude of response can nevertheless vary depending on the primary tumour type. For example, the responses of patients with MMRd gliomas to ICIs are generally less robust than those of patients with other tumour types^4,14,18.

• MMRd tumours can respond to ICIs despite major histocompatibility complex (MHC) class I loss. A prevailing notion exists that MMRd tumours are susceptible to CD8⁺ T cell-mediated cytotoxicity. However, MMRd tumours with loss of MHC class I expression, for example, owing to biallelic B2M loss, can still respond to ICIs. In these cases, tumour cell killing seems to be mediated by other effectors such as CD4⁺ T cells and γδ T cells^298,299,300.

• Microsatellite alterations can vary by cancer type. Only a fraction of microsatellites are altered in MMRd tumours^8,9,10.The affected microsatellites can vary in location and in size (longer versus shorter indels) across MMRd cancer types. This variability is typically consistent within a typical tumour type. For example, in MMRd endometrial cancers the affected microsatellites differ from those affected in other MMRd cancers (such as colon cancer) and the alterations are often shorter indels^9,10,218,219. Importantly, these differences can lead to the misclassification of certain MMRd tumours as microsatellite stable or MSI indeterminate depending on the sensitivity of the assays for smaller allelic shifts^221,222.

• MMRd tumours have varying degrees of chromosomal instability. MMRd tumours were historically thought of as chromosomally stable, and MSI-H was viewed as mutually exclusive with chromosomal instability^112,113,114. Although MMRd tumours do indeed have lower levels of aneuploidy than their MMR-proficient counterparts, many have at least some degree of aneuploidy, and some harbour extensive chromosomal abnormalities^115,116,117. These variations might be related to MMR loss or reflect loss-of-function alterations affecting other target genes with a role in maintaining chromosome integrity^118,119.

Mechanisms and function of DNA MMR

DNA replication fidelity in eukaryotes

Spontaneous mutations are uncommon. They occur at an estimated rate of 10⁻⁹ to 10⁻¹⁰ per base pair per cell division^3,21. Multistage quality control is crucial to ensuring such high-fidelity DNA replication. DNA polymerases are inherently accurate and virtually always incorporate the correct nucleotides. The rate of incorrect base insertion ranges around 10⁻⁴ to 10⁻⁵ and is influenced by several factors such as the type of DNA polymerase, the target nucleotide base and the surrounding microenvironment^22,23. Nonetheless, errors inevitably occur and when they do the replication machinery slows, providing time for editing by the proofreading exonuclease activity of DNA polymerases. This activity further reduces the nucleotide misinsertion rate to ~10⁻⁷. Post-replication MMR functions to correct any mistakes left behind by the DNA replication complex, increasing fidelity by an additional 1,000-fold²⁴. MMR is thus a key guardian of genome stability and has appropriately been a subject of longstanding research interest dating back to as early as the 1960s when its genetic origins were first deciphered in prokaryotes²⁵.

MMR in E. coli

The concept of mismatched nucleotide repair was introduced by Robin Holliday in 1964 as a potential explanation for ‘gene conversion’ during homologous recombination in yeast². Subsequent research in Escherichia coli led to the realization that MMR activity corrects errors occurring during DNA replication in addition to mismatches that arise from genetic recombination^26,27. The MMR machinery of E. coli has since been thoroughly delineated and so far remains the best understood across all living organisms. Several mismatch correction systems exist in E. coli including the methyl-directed, very short patch, MutY-dependent and RecF pathways. These pathways can be distinguished by their substrate specificity and individual repair mechanisms, although considerable overlap exists in their protein constituents²⁸. Key members include the Siegel mutator (MutS), Salmonella LT7 mutator (MutL) and Hill mutator (MutH), which are so named because their genetic inactivation is associated with a substantial increase in spontaneous hypermutability²⁹. Accordingly, the mutator family of genes is now widely recognized to have a crucial role in preserving DNA replication fidelity across organisms³⁰.

The methyl-directed MMR system has historically been of great interest because it is a model of strand specificity. In E. coli, newly synthesized DNA is temporarily undermethylated at adenines within dGATC sequences. This transient hemimethylated state acts as a time-dependent signal that directs the repair machinery exclusively to the hypomethylated error-containing daughter strand as opposed to the methylated non-error-containing parent strand³¹. MutS homodimers surveil heteroduplex DNA intermediates in an adenosine diphosphate (ADP)-bound state and exchange ADP for adenosine triphosphate (ATP) upon recognizing most non-canonical base–base mismatch combinations as well as short indel loops^28,32,33,34. ATP binding induces a conformational change in MutS, which leads to the formation of a stable hydrolysis-independent mobile clamp that can freely diffuse along the DNA backbone³⁵. Importantly, mismatch dissociation allows for the loading of additional MutS-ATP sliding clamps that collectively work to orient the mismatch site to regional GATC sequences³³. These complexes also provide a platform for recruiting MutL homodimers to the mismatched DNA³⁶. MutL is itself a molecular switch that bridges MutS-dependent mismatch recognition to the ensuing excision and repair efforts^29,37. MutS–MutL–mismatch complexes then stimulate the latent endonuclease activity of MutH to incise the daughter strand at a nearby unmethylated dGATC site in an ATP-dependent manner^38,39,40. This incision creates an entry point for downstream effector proteins, which then remove the replication error from the nascent strand, correctly resynthesize DNA and, finally, ligate the nick to complete repair^40,41.

MMR in humans

The extensive body of work characterizing MMR in E. coli provided a prototype for understanding eukaryotic MMR (Fig. 2). MMR is a highly evolutionarily conserved process, and several human homologues of bacterial MutS and MutL have been identified. These genes include the instrumental MutS homologue 2 (MSH2), MSH3 and MSH6. Key MutL homologues are MutL homologue 1 (MLH1), MLH3 and postmeiotic segregation increased 2 (PMS2). In contrast with bacterial MutS and MutL, which operate as homodimers, their human counterparts act as distinct heterodimers, MutSα (MSH2–MSH6), MutSβ (MSH2–MSH3) and MutLα (MLH1–PMS2) are viewed as far and away the most important. These three complexes execute most MMR. A fourth, MutLγ, is formed by dimerization of MLH1–MLH3 but seems to have a lesser role in indel repair^42,43,44.

Fig. 2: Timeline of the major discoveries and milestones in MMRd/MSI research.

This timeline has been adapted and updated from that provided by Lynch et al.¹. CMMRD, constitutional mismatch repair deficiency; CRC, colorectal cancer; E. coli, Escherichia coli; FDA, The US Food and Drug Administration; ICG-HNPCC, International Collaborative Group on Hereditary Non-Polyposis Cancer; mCRC, metastatic colorectal cancer; MMR, mismatch repair; MMRd, DNA mismatch repair deficiency; MSI, microsatellite instability; MSI-H, microsatellite instability high; PFS, progression-free survival.

MutS/L complexes function as molecular switches that together bridge mismatch recognition with subsequent nucleotide correction. The principle downstream effectors include, at a minimum, replication factor C (RFC), proliferating cell nuclear antigen (PCNA), exonuclease 1 (Exo1), replication protein A (RPA) and DNA polymerases, which collectively facilitate strand-directed excision and repair. The signal that dictates strand specificity is different in eukaryotes, as they do not have analogues to the hemimethylated-GATC sequences found in E. coli. Instead, pre-existing nicks and gaps associated with leading and lagging strand synthesis are presumed to direct eukaryotic MMR machinery to the nascent DNA^45,46.

MMR in humans begins with substrate recognition by either MutSα or MutSβ (ref. ⁴⁰). MutSα recognizes the most common replication errors: single-base mispairings (including all eight possible base–base mismatches) and small insertion or deletion loops of one to approximately three extrahelical residues. MutSβ has a greater affinity for larger insertion or deletion loops that typically range in size from two to around ten extrahelical residues^41,47,48. As would be expected given their respective functions, MutSα is more abundant than MutSβ and is probably the primary sensor of DNA damage in humans^48,49. Our ensuing discussion therefore predominantly focuses on MutSα.

At a minimum, MutSα must perform three basic functions: ‘scanning’ (also known as damage detection), damage binding and activation of downstream effectors. The complex is first tasked with recognizing a substrate of genomic error amidst a vast landscape of correctly synthesized DNA. The crystal structure of MutSα, which has been well-characterized in humans, provides insights into this mechanism⁵⁰. MSH2 and MSH6 each contain five domains that are conserved from MutS⁵¹. These are classified as the mismatch-binding (domain 1), connector (domain 2), lever (domain 3), clamp (domain 4) and ATPase (domain 5) domains. MSH6 possesses an additional N-terminal disordered domain^32,34. Although conserved, the five corresponding domains within MSH2 and MSH6 differ in length and nucleotide sequence, which gives rise to pseudosymmetric tertiary structures. These homologues dimerize to form an asymmetric θ-like complex, with domains 1–5 of each protein juxtaposed against one another. The long axis of the oval is formed by the C-terminal ATPase domains at one end, lever domains in the middle and clamp domains at the other end. The mismatch-binding and connector domains constitute the bar that divides the oval into its two distinct channels. The lower channel, which is located distal to the ATPase domains, encircles and bends DNA in search of mismatches, whereas the upper channel remains empty^50,52,53. However, only the MSH6 mismatch-binding domain directly engages lesions. The MSH2 counterpart is bent away from and makes minimal contact with the DNA scaffold⁵⁰. Similarly, the MSH3 subunit of MutSβ is responsible for recognizing mismatched DNA. The mismatch-binding specificity of MSH6 is conferred by an evolutionarily conserved Phe–X–Glu motif^54,55,56. In this structure, the aromatic ring of the phenylalanine residue recognizes stereochemical distortions at mismatch sites, and the glutamate residue further engages substrates through hydrogen bonding as well as other interactions⁵⁷. By contrast, MSH3 engages the sugar-phosphate backbone of indel loops via a tyrosine–lysine pair within its mismatch-binding domain⁵⁸.

After recognition of a mismatch, MutSα must then activate downstream excision and repair proteins. Signal transduction is achieved through adenine nucleotide binding and hydrolysis^59,60. MSH2 and MSH6 both contain Walker ATP-binding/hydrolysis motifs that are essential for MMR but non-equivalent in their affinity for different adenine nucleotides as well as in their hydrolytic potential^40,61. In the scanning state, the MSH2–MSH6 heterodimer rapidly hydrolyses ATP and therefore predominantly exists in an ADP-bound conformation⁶². Mismatch recognition stabilizes the complex in an ATP-bound state⁶³. This process induces a conformational change that opens the middle bar of the θ-like complex to create a single large channel that allows MutSα to diffuse freely along the DNA backbone in a hydrolysis-independent manner⁶⁴. The open channel also provides a binding platform for MutLα, which interfaces with MutSα through the N-terminal domain of MLH1 (refs. ^53,65,66).

MutLα is a latent ATP-dependent endonuclease formed by the dimerization of MLH1 and PMS2 at their respective C-terminal domains. In each protein, the C terminus is connected to the N terminus by a lengthy and flexible linker arm⁵³. The N-terminal domains of both proteins confer both ATPase and DNA-binding activity. The endonuclease site of MutLα is a functionally conserved metal-binding DQHA(X)₂E(X)₄E motif located at the C terminus of PMS2 (ref. ⁶⁷). In the absence of adenine nucleotides, the linker arms are extended. Sequential ATP-binding condenses these arms and draws the C-terminal and N-terminal domains closer together, thus bringing the endonuclease and DNA binding sites into closer proximity to one another to facilitate strand-directed incisions⁶⁸. In addition to geometric reconfiguration, MutLα endonuclease activation also depends on the sliding clamp PCNA, which is recruited to the DNA helix by the clamp loader RFC. PCNA then interfaces with the C-terminal domain of PMS2 to activate the endonuclease and direct incisions in the daughter strand^69,70,71.

Mismatch recognition by MutSα or MutSβ and the recruitment of MutLα constitute early shared steps required for MMR. Subsequent excision and repair efforts occur through one of several pathways that can broadly be distinguished on the basis of whether Exo1 is involved. Exo1 is itself a 5′-to-3′ double-stranded (ds) DNA exonuclease that participates in multiple DNA repair processes in addition to MMR^72,73. Genetic studies have shown that abrogating Exo1 activity causes only a weak mutator phenotype, which is consistent with the existence of both Exo1-dependent and Exo1-independent MMR pathways^74,75. In Exo1-dependent MMR, the enzyme enters DNA at strand breaks and then proceeds to excise DNA in a 5′-to-3′ hydrolytic reaction stimulated by either MutSα or MutSβ (refs. ^67,76,77,78). MutLα-directed incisions are not required to initiate this process because Exo1 can enter DNA at pre-existing strand breaks. However, MutLα does help suppress Exo1 activity after mismatch removal is complete and has a more crucial role in Exo1-independent MMR, which is thought to involve processive nicking of the nascent strand by MutLα to either excise the DNA surrounding the mismatch or establish a substrate for strand displacement synthesis^73,77. After Exo1-dependent and Exo1-independent excision, the nascent strand is resynthesized by DNA polymerases and the remaining nicks are sealed by DNA ligase activity to complete MMR.

Genomic sequelae of MMRd

MMRd tumours have a unique genomic profile. The exomes of MMRd cancer cells contain hundreds to thousands of somatic mutations, a vastly greater burden than is seen in cancer cells with proficient MMR (MMRp)⁷. This mutational profile includes a large number of single-nucleotide base substitutions and indels, as would be expected given the substrate-specific roles of the MMR machinery. The most noteworthy and clinically relevant feature of such cells is the accumulation of indels within short repetitive sequences of DNA known as microsatellites, which yield novel alleles of an altered length compared with that of the counterpart normal alleles in non-malignant cells. This phenomenon is referred to as microsatellite instability. Tumours can be characterized as MSI-H, MSI indeterminate (MSI-I) or microsatellite stable (MSS) depending on the number of microsatellite markers that are MSI.

High tumour mutation burden

The genomes of MMRd cancer cells contain a remarkably high number of somatic mutations⁷. TMB provides a measure of the total number of non-synonymous somatic coding mutations per megabase of DNA profiled⁷⁹. Across MMRd tumours, the TMB is typically around 40 mutations per megabase (mut/Mb) and ranges between 20 and 60 mut/Mb in most patients^{80,81,82,83,84,85}. For those with CMMRD, which arises owing to germline biallelic loss of MMR function, TMB can be massive and often surpasses 100–200 mut/Mb especially in the context of co-occurring polymerase proofreading deficiencies⁸⁶. Of note, the high TMB state does not occur immediately but rather develops throughout the growth and progression of preneoplastic to neoplastic MMRd lesions. In vitro, nascent MMRd tumours become high TMB after months of cell division^5,6. The dynamics and timing of mutation accumulation in vivo remain unclear given that certain MMRd lesions could be subject to extensive immunoediting⁸⁷.

Microsatellites and microsatellite instability

Microsatellites are tandemly repeated short base-pair motifs that can be found throughout the genome⁸⁸. Most microsatellites are composed of one to three nucleotide repeats, although the repeated unit can be up to six nucleotides in length⁸⁹. Microsatellite lengths can change naturally over time and microsatellites have emerged as useful genetic markers owing to their propensity for polymorphism⁹⁰. Changes in microsatellite length arise through strand slippage during DNA replication. Slippage occurs when DNA polymerases stutter or pause within a repeat and then dissociate from the nascent strand^91,92. Susceptibility to slippage depends on the composition of the repeat motif as well as on its length and the overall length of the microsatellite tract, with relative mutation rates higher in longer overall sequences composed of many shorter repeat units (that is, lengthy homopolymers such as polyA or polyC repeats)^93,94,95,96. Polymerase dissociation enables the terminal end of the nascent strand to temporarily separate from the template strand. Misalignment of direct repeats when the two strands are reannealed can lead to a daughter strand with a different microsatellite length to that of its parent. Unless such additions and deletions are corrected, they become permanently fixed in the genome during the subsequent round of replication^89,97.

Accordingly, MMRd cells are characterized by the accumulation of indels within microsatellites and, in turn, the development of MSI. Functional data from experimental models of MMRd show that this process is not immediate but rather takes months to years to occur^5,6,98. Indel signatures from MMRd tumours suggest that MSI events are primarily observed in homopolymers as opposed to doublet or other repeat units and that such events are predominantly deletions rather than insertions. Several groups have now surveilled the global microsatellite landscape across thousands of MSI-H tumours^8,9,10. The human genome is estimate to include around 20 million microsatellite loci and of these, approximately 200,000–300,000 are located within exons. Most MSI-H tumours harbour hundreds of exonic MSI events and tens to hundreds of thousands of genomic MSI events, with a specific predisposition for the accumulation of instability within certain motifs, such as homopolymers. Although variations exist both between and within tumour types, a pattern of certain coding-region microsatellite loci exhibiting recurrent susceptibility to frameshift mutations can be seen. The development of repeat tract instability within these select regions raises the possibility that these regions confer a proliferative and/or survival advantage.

Oncogenic events associated with MMRd

Although the connection between MMRd and MSI is clearly delineated, how defective MMR specifically leads to tumorigenesis remains a major unresolved question. MMRd tumours are characterized by extensive genomic instability and, in some cases, by widespread epigenetic aberrations. These tumours are known to harbour many different types of oncogenically relevant alterations including loss-of-function mutations in tumour-suppressor genes, activating mutations in oncogenic drivers, somatic copy-number alterations and CpG-island promoter hypermethylation. The prevalence of these events varies across MMRd cancers and, in some instances, depends on the mechanism of MMR loss. However, the cancer-related role of the events is often unclear as is their timing with respect to biallelic MMR loss, and whether these are clonal aberrations that drive carcinogenesis or merely passenger events that occur in a hypermutable landscape remains to be determined.

Target gene loss-of-function mutations as potential drivers of oncogenesis

MMRd tumours frequently harbour loss-of-function mutations in tumour-suppressor genes and these often result from protein truncating frameshifts within coding-region microsatellites. The first putative driver mutations were identified in the gene encoding TGFβ receptor 2 (TGFBR2), which contains short tandem repeats at the beginning of the coding sequence. The data from three studies published in 1995 collectively demonstrate that frameshift mutations within these regions lead to a loss of protein function and are common in MSI-H colorectal and gastric cancers but not endometrial cancers^99,100,101. This asymmetric distribution suggests that heterogeneous mechanisms of tumorigenesis are involved, and PTEN has since been identified as a potentially more relevant target gene among MSI-H endometrial tumours¹⁰². Cancer-type specificity is also seen at various other microsatellite loci^9,10. Recurrently mutated genes with a potential role in oncogenesis in MSI-H CRC include members of the Wnt–β-catenin signalling pathway such as RNF43 and APC^10,103. Genomic data also implicate AXIN1 and AXIN2 in this regard⁹⁸. Frameshift mutations in RNF43 and APC have been shown to be mutually exclusive, and an absence of pathway redundancy supports the functional importance of these events¹⁰³. Moreover, this observation reinforces the notion that a central process (namely, coding-region MSI) might promote oncogenesis through myriad ways. Interestingly, three of the four major MMR genes (MSH2, MSH6 and PMS2) as well as MSH3 contain coding-region microsatellites and can themselves be subject to inactivation by frameshift mutations. Cumulative loss of function of multiple MMR genes might accelerate mutagenesis and further destabilize MMRd tumour cells^104,105.

BRAF ^V600E mutations and CIMP

Most cases of sporadic MMRd arise from biallelic hypermethylation of the MLH1 promoter, which is thought to occur in the context of widespread promoter CpG-island methylation (CIMP)^12,106,107. However, the functional implications of such diffuse epigenetic changes remain poorly understood, as does the aetiology. Approximately 50% of MLH1-hypermethylated CRCs harbour BRAF^V600E mutations¹⁰. One possibility is that BRAF^V600E constitutively activates DNA methyltransferases, although this hypothesis would not account for the detection of CIMP in the many MLH1-methylated CRCs that lack BRAF alterations¹⁰⁸. Moreover, BRAF^V600E mutations are uncommon in MMRd endometrial cancers despite most of these tumours arising from sporadic MLH1 methylation¹⁰⁹. Overall, BRAF^V600E is an established pan-cancer driver alteration, although the timing of these mutations with respect to CIMP and MMR loss in CRC remains to be further defined.

CIN in MMRd tumours

Chromosomal instability (CIN) is a hallmark of cancer observed nearly universally across solid tumours^110,111. CIN can manifest as aneuploidy through duplications or deletions of whole or large portions of a chromosome and can also result in DNA rearrangements as well as more focal copy-number alterations. Most cancers have at least some degree of CIN irrespective of the primary tumour location or histology, and this observation also applies for MMRd disease. However, MMRd tumours were traditionally viewed as chromosomally stable with minimal karyotypic abnormalities, and MSI-H was considered to be mutually exclusive with the presence of CIN^112,113,114. Whereas most MMRd tumours are indeed diploid, the degree of aneuploidy in such tumours is not universally low. In fact, a subset of MMRd colorectal and endometrial cancers have high levels of aneuploidy, although this typically manifests as focal gains and losses in specific chromosome regions as opposed to whole-chromosome aberrations^115,116. A subset of MSI-H tumours also harbour gene rearrangements¹¹⁷.

The aetiology of aneuploidy in MMRd tumours remains unclear. The MMR proteins themselves have a role in chromosome maintenance, for example, by recognizing and correcting mismatches during dsDNA break repair through DNA recombination. The fidelity of recombination might therefore be compromised in the setting of MMRd and lead to increased CIN¹¹⁸. Loss-of-function mutations in MSI-H tumours might also affect genes involved in chromosome maintenance¹¹⁹. Ultimately, the full spectrum and mechanism of CIN across MSI-H tumours remain to be fully defined as do the clinical implications. Tumour aneuploidy is usually associated with poor outcomes and has been associated with worse survival outcomes after treatment with ICIs, although whether this effect on response is specifically seen among patients with MSI-H tumours has yet to be determined¹²⁰.

Coding-region MSI as a source of immunogenic neoantigens

Coding-region microsatellite indels that alter the open-reading frame can give rise to foreign-appearing frameshift peptides (FSPs). MSI-H cancers are characterized by a high number of such frameshift events, and because these predominantly stem from deletions in areas of mononucleotide repeats and affect many of the same genes, the resultant pool of FSPs includes a large set of shared neoantigens^121,122. These alterations come with several important immunological considerations. Translational frameshifts that generate peptides comprising long stretches of novel amino acids have the potential to generate multiple immunogenic neoepitopes, which is in contrast with non-synonymous missense mutations that can only alter a single amino acid. In this regard, frameshift-derived neoantigens have been shown to be markedly distinct from missense-derived neoantigens and viral epitopes and able to elicit T cell responses¹²¹. Multiple other groups have reported similar findings on the immunogenicity of FSPs. In one study, for example, a highly immunogenic neoepitope resulting from a frameshift mutation in TGFBR2 was identified and found to be recognized by T cells from several patients with MSI-H colon cancer¹²³. In other studies, investigators observed that both tumour-infiltrating and peripheral T cells from patients with MSI-H CRC are able to recognize MSI-associated FSPs¹²⁴. Taken together, these data provide a molecular framework for understanding the pronounced lymphocyte infiltration typically seen on histopathological analysis of MSI-H tumours. Interestingly, peripheral T cells from individuals with Lynch syndrome who had not yet developed cancer also have FSP-specific reactivity, indicating a potential role of immunoediting in curtailing tumour development in this population¹²⁴. Finally, an inverse correlation between the predicted immunogenicity of a frameshift mutation and its frequency in colorectal and endometrial cancers has been reported, which further supports the notion that highly immunogenic FSPs are counterselected through immune surveillance of emerging cancer cells¹²².

Epidemiology and aetiology of MMRd

Epidemiology and clinicopathological characteristics

Altogether, approximately 2–4% of solid tumours are MMRd/MSI-H^{8,125,126,127,128} (Fig. 3). Most of these (~80–85%) arise in the context of sporadic loss of MMR function. The remainder originate from hereditary MMRd¹²⁸. The inherited MMRd syndromes include Lynch syndrome, its phenotypic variant Muir–Torre syndrome (MTS) and CMMRD. Both Lynch syndrome and MTS are caused by germline heterozygous loss of one of the MMR genes, whereas germline biallelic loss leads to CMMRD, a devastating paediatric tumour predisposition syndrome.

Fig. 3: Prevalence of MMRd across solid tumours.

The percentages of mismatch repair-deficient (MMRd) tumours among stage I–III and stage IV tumours across cancer types, as well as among select cancer types, are shown. The data are from consecutively run clinical tests performed by Caris Life Sciences. CUP, cancer of unknown primary; GBM, glioblastoma multiforme; GEJ, gastroesophageal junction; MMRd, DNA mismatch repair deficiency; NSCLC, non-small-cell lung cancer; SCC, squamous cell carcinoma.

MSI-H is most prevalent in endometrial cancer (accounting for ~25% of cases) followed by small bowel cancer (~15%), colon cancer (~10%) and gastric cancer (~10%). This characteristic is less common in rectal adenocarcinoma (~3–5%), and even lower frequencies are observed in other malignancies such as cervical, prostate and non-melanoma skin cancers^{8,125,126,127,128,129}. Only 2–3% of all endometrial cancers are associated with underlying Lynch syndrome. Similarly, only ~2–3% of all CRCs and ~1% of all gastric cancers are associated with Lynch syndrome^{8,127,128,130,131,132}.

MSI-H is generally more prevalent at earlier disease stages. For example, the prevalence of MMRd/MSI-H in early stage colon cancer has been reported to be as high as 15–20% but is only 4–5% among patients with metastatic disease^133,134,135. MMRd/MSI-H is similarly more prevalent in early stage gastric cancer¹³⁶. Patients with early stage MMRd tumours also seem to have a better prognosis than their MMRp counterparts. The prevailing notion is that the higher prevalence and the improved prognosis of the MMRd/MSI-H phenotype in patients with early stage disease reflects a more favourable tumour immune microenvironment characterized by high levels of infiltration of antitumour immune effector cells^{134,136,137,138,139,140,141,142,143}.

Interestingly, the prevalence of MSI-H varies between histological subtypes within the same cancer type, suggesting that MMRd cancers preferentially arise from lineage-specific cells. Among endometrial cancers, for example, MMRd/MSI-H occurs almost exclusively in endometrioid adenocarcinomas¹¹⁴. Intestinal-type gastric carcinomas are more frequently MSI-H than their mixed-type or diffuse-type counterparts¹⁴⁴. The prevalence of MSI-H in ovarian cancers is predominantly restricted to the subset of carcinomas that arise from endometriosis; the majority of these are endometrioid ovarian carcinomas, although MSI-H also occurs to a lesser extent in mixed and clear cell carcinomas^145,146.

MMRd tumours can also have other distinct clinicopathological features. For example, MMRd colon cancer predominantly arises in the proximal (right) colon rather than the distal (left) colon¹⁴⁷. MMRd CRCs are thought to be more resistant to 5-fluorouracil-based therapy than their MMRp counterparts^{142,148,149,150}. MMRd tumours also tend to be mucinous, poorly differentiated and enriched with tumour-infiltrating lymphocytes (TILs)^{13,151,152,153,154,155,156}. Interestingly, TIL density varies greatly among MSI-H tumours. As many as ~30% of MSI-H CRC lesions are characterized by low or intermediate levels of TIL infiltration. The data from several studies suggest that the number of frameshift mutations is correlated with TIL density, and low levels of TIL infiltration have been associated with inferior outcomes^{13,157,158,159}. Thus, although many features are shared across MSI-H tumours, there exists, at the same time, tremendous intertumour and intratumour heterogeneity. The mechanisms underlying MMR loss are increasingly recognized to have important phenotypic and clinical implications.

Lynch syndrome

Lynch syndrome is a hereditary cancer predisposition syndrome caused by germline monoallelic MMR gene alterations (Box 2). The primary aetiology is autosomal-dominant inheritance of a heterozygous pathogenic germline mutation in one of the four key MMR genes (MSH2, MSH6, MLH1 or PMS2)¹. Thus far, no compelling data linking germline alterations in either MSH3 or MLH3 with Lynch syndrome are available. In a small subset of individuals (~1–3%), large terminal deletions in the epithelial cell adhesion molecule (EPCAM) gene lead to transcriptional readthrough and subsequent epigenetic silencing of its structurally intact neighbour MSH2 (refs. ^160,161,162). Rarely (<1% of cases), Lynch syndrome can arise from a heritable epigenetic defect known as constitutional MLH1 epimutation, involving monoallelic hypermethylation of the MLH1 promoter and concomitant allele-specific loss of expression throughout somatic tissue¹⁶³.

Lynch syndrome is one of the commonest hereditary cancer predisposition syndromes. In total, approximately 1 in 300 people are estimated to carry a mutation in one of the major MMR genes (~1/700 for PMS2, ~1/750 for MSH6, ~1/2,000 for MLH1 and ~1/3,000 for MSH2)¹⁶⁴. Thousands of distinct germline MMR gene mutations have now been characterized, and approximately 40% of these occur in MLH1, ~35% MSH2, ~20% MSH6 and ~5% PMS2 (ref. ¹⁶⁵). Pathogenic mutations almost exclusively cause protein truncations or total loss-of-function as opposed to partial loss¹⁶⁶. Most of these are nonsense and frameshift mutations or large deletions, although other mutations such as non-synonymous base substitutions and splice variants can also occur. The variable penetrance across MMR genes reflects their relative functional importance. MLH1 and MSH2 encode proteins that are essential for all heterodimer complex formation whereas their counterparts are partially redundant (MSH6 with MSH3 and PMS2 with MLH3).

Patients with Lynch syndrome develop cancer following somatic loss of the remaining wild-type allele^{167,168,169,170}. CRC is the commonest index tumour in individuals with Lynch syndrome, followed by endometrial cancer, although an increased risk of myriad other malignancies exists including cancers of the stomach, small bowel, pancreas, urothelium, prostate and central nervous system (CNS)¹⁷¹. The clinical manifestations of Lynch syndrome vary greatly between patients, and a substantial portion of this phenotypic diversity can be accounted for by the affected MMR gene. MSH2 and MLH1 mutation carriers have what is described as a classic Lynch syndrome phenotype. These individuals have a higher lifetime risk of developing any Lynch-syndrome-associated cancer and typically develop cancer at a younger age than carriers of germline MSH6 or PMS2 variants. Carriers of MSH6 and PMS2 mutations generally present in a more atypical manner in that they have a comparatively lower lifetime risk of Lynch-syndrome-associated cancer and develop cancers at, on average, older ages¹⁶¹.

Established tumour predisposition spectra exist for each of the four MMR genes. For example, carriers of pathogenic MSH2 and MLH1 mutations have a similar risk of CRC and endometrial cancer but MSH2 mutation carriers have a higher risk of urothelial and prostate tumours than MLH1 mutation carriers¹⁷². MSH2 is the MMR gene most strongly associated with MTS, the rare phenotypic variant of Lynch syndrome characterized by a predilection for sebaceous neoplasms¹⁷³. Individuals with EPCAM deletions have a similar risk of CRC to that of MSH2 mutation carriers because they have MSH2 loss in colonic epithelial cells, in which EPCAM is expressed. For the same reason, this group tends to have a lower risk of endometrial cancer than that of MSH2 mutation carriers unless the EPCAM deletion extends to MSH2 (refs. ^162,174). Germline MSH6 mutations are strongly associated with endometrial cancer¹⁷⁵. PMS2 mutation carriers have a modest risk of CRC and endometrial cancer and a lower risk of other Lynch-syndrome-associated cancers compared with carriers of other MMR gene mutations¹⁷⁶.

The affected MMR gene also seems to have implications for the tumour genomic landscape. For example, Lynch-syndrome-associated tumours have varying degrees of MSI, which seems to be influenced at least in part by the affected MMR gene. In one study, as many as 36% of Lynch-syndrome-associated tumours were found to be MSS¹²⁸. These MSS tumours tended to be non-colorectal and non-endometrial cancers, and most (~78%) had germline mutations in the lower penetrance MMR genes (PMS2 and MSH6). By contrast, the majority (~71%) of MSI-H Lynch-syndrome-associated tumours arose in the context of germline MLH1, MSH2 or EPCAM mutations. MMRd tumours with MLH1 and/or PMS2 losses have been shown to have a lower TMB than those derived from the loss of MSH2 and/or MSH6 (ref. ¹⁷⁷). These observations are consistent with experimental data showing that cancer cells deficient in MLH1 need 1 year to accumulate enough mutations to become immunogenic whereas MSH2-deficient cells need only 3 months to acquire an immunosensitive phenotype^5,6.

Much of what we understand about tumorigenesis in Lynch syndrome is derived from research involving the colorectum given that non-malignant, pre-malignant and malignant tissues in this region are all readily accessible for sampling via endoscopy. Colorectal adenomas in patients with Lynch syndrome can be MMRd, which indicates that biallelic loss-of-function is sometimes an early event in tumour progression^178,179. However, a subset of cancers occurs through non-MMRd-adenoma formation followed by secondary loss of MMR, and many other patients will directly develop invasive adenocarcinoma without evidence of precursor polyps^180,181. A certain degree of variance in the molecular mechanisms underlying tumour formation can be attributed to the underlying compromised MMR gene. For example, patients with constitutional MSH2 mutations have a higher incidence of somatic APC mutations, whereas those with germline MLH1 mutations more often have CTNNB1 mutations¹⁸². In the same study, a greater incidence of advanced adenomas was observed in MSH2 mutation carriers than in MLH1 mutation carriers despite a similar incidence of colorectal adenocarcinomas. Taken together, these findings suggest that haploinsufficiency in some MSH2 mutation carriers might lead to accelerated colorectal adenoma to carcinoma progression through the combination of MMRd and somatic APC mutations, whereas MLH1 mutation carriers might be more likely to immediately develop invasive adenocarcinoma owing to acquired CTNNB1 mutations¹⁸³. The latter scenario is proposed to follow a 2-in-1 hit mechanism. CTNNB1 and MLH1 are located within close proximity of one another on chromosome 3p22.1–p22.2. If cells with a pathogenic MLH1 variant acquire a monoallelic somatic CTNNB1 mutation, then a subsequent copy-number neutral loss of heterozygosity event could simultaneously lead to biallelic inactivation of MLH1 and biallelic activation of CTNNB1 (ref. ¹⁸⁴). Interestingly, data published in 2024 indicate that MLH1 haploinsufficiency is associated with a state of increased CIN that precedes the development of MSI and might therefore predispose to the initiating copy-number neutral loss of heterozygosity event¹⁸⁵. In fact, many patients with Lynch syndrome develop CRC despite receiving appropriate colonoscopic surveillance; a non-polypoid mechanism of tumorigenesis such as simultaneous second hits to MLH1 and CTNNB1 could account for the development of interval cancers among these individuals^186,187. These and other non-adenomatous lesions might arise within MMRd crypt foci, which are prevalent throughout the non-malignant intestinal mucosa of individuals with Lynch syndrome but difficult to detect using colonoscopy¹⁸⁸. MMRd crypt foci can be found in individuals with germline alterations affecting any of the MMR genes, and the remarkable abundance of these foci suggest that somatic second hits to the MMR machinery are a frequent occurrence in individual patients with Lynch syndrome. Yet, despite this abundance, the incidence of colorectal neoplasia remains comparatively low, implying that most of these foci do not progress to cancer. The molecular and pathological sequelae that follow biallelic MMR loss within intestinal crypts remains largely unknown. MMRd crypt foci have been shown to harbour MSI and therefore might be subject to immune surveillance^{188,189,190,191,192}.

Box 2 The history of Lynch syndrome

The history of Lynch syndrome can be traced back to nearly one hundred years before its genetic underpinnings were unveiled. In 1895, Aldred Scott Warthin, Chairman of the Department of Pathology at the University of Michigan, became interested in the family history of his seamstress who anguished over her many relatives that had died of cancer^1,349. Warthin called them ‘family G’ because they had emigrated from Germany. He published their pedigree in 1913 together with those of several other cancer-prone families³⁵⁰. The transmission of cancer in family G seemed to follow an autosomal-dominant pattern with, at the time of publication, 17 of the 48 descendants of the grandfather who died of cancer seemingly affected. Warthin presented a follow-up report on family G in the year 1925 (ref. ³⁵¹). By this point, the preponderance of gastrointestinal tract and uterine lesions was as noteworthy as the incidence of cancer itself and suggested an organ-specific effect. Warthin’s colleagues Hauser and Weller provided another update in 1936, several years after Warthin had passed away³⁵². Their findings confirmed the predilection for organ-specific cancers in family G across multiple generations and led them to hypothesize that the family possesses an inherited predisposition to cancer, which we now know to be Lynch syndrome owing to an MSH2 mutation³⁵³.

Despite this seminal work, the field remained relatively dormant for three decades until Henry T. Lynch reported on additional familial clusters of cancers with phenotypic similarity to those of family G^354,355,356. His work rejuvenated interest in the possibility of an underlying hereditary cancer predisposition syndrome and launched an era defined almost exclusively by the clinical characterization of Lynch syndrome, which lasted until its mechanistic aetiology was eventually delineated in the 1990s. Lynch reinvestigated family G in 1971, at which time he described their affliction as a cancer family syndrome³⁵⁷. C. Richard Boland published data on additional families with similar patterns of cancer predisposition but redesignated the condition Lynch syndrome in 1984 out of concern that cancer family syndrome implied a non-specific increased risk of cancer to all organs^358,359. For a period of time, the condition was also referred to as hereditary non-polyposis colon cancer (HNPCC) to distinguish it from familial adenomatous polyposis. However, this was ultimately deemed a misnomer given that patients with Lynch syndrome are prone to malignancies in a variety of extracolonic organs. For this reason, and owing to the risk of multiple adenomas, the term HNPCC is no longer a part of the accepted nomenclature^359,360.

Momentum accelerated when the International Collaborative Group on Hereditary Non-Polyposis Colorectal Cancer was conceived in 1989 as a means of coordinating the international study of Lynch syndrome. These efforts were spearheaded by Hans Vasen, Jukka-Pekka Mecklin, Meera Khan, Henry Lynch and others who worked together to identify and register families. The clinical diagnostic Amsterdam Criteria I were established at the organization’s first meeting in Amsterdam in the year 1990 (ref. ³⁶¹). The criteria were as follows: (1) at least three relatives with histologically confirmed non-familial adenomatous polyposis colorectal cancer (CRC), one of whom is a first-degree relative of the other two; (2) at least two successive generations affected; and (3) cancer diagnosed before the age of 50 years in at least one of the individuals. These initial criteria were felt to be too stringent because they focused exclusively on CRC and were amended in 1998 to improve sensitivity. The revised Amsterdam Criteria II included patients with cancers of the colorectum, endometrium, small bowel, ureter and renal pelvis³⁶². The International Collaborative Group on Hereditary Non-Polyposis Colorectal Cancer continued to meet annually until merging with the Leeds Castle Polyposis Group in 2001 to create the International Society for Gastrointestinal Hereditary Tumors (InSiGHT)³⁶³. The early efforts of this group to optimize clinical characterization, identify affected families, and coordinate international research ultimately facilitated the illuminative molecular studies that would delineate the genetic aetiology of Lynch syndrome.

Deficient MMR was formally linked to Lynch syndrome in 1993 and 1994, although the earliest clues of such a link came from observations made by microbial geneticists decades earlier. Research in bacteria and bacteriophage models in the 1970s showed that tandemly repeated DNA sequences are particularly susceptible to mutagenesis, and the high frequency of frameshift mutations in these regions was thought to result from slipped-strand mispairing^364,365. Subsequent studies found that Escherichia coli strains defective in methyl-directed MMR have a dramatic increase in the number of frameshift mutations present in sequences of reiterated base pairs^366,367. Years later, microsatellite instability (MSI) was found to occur in a subset of CRCs including among those associated with Lynch syndrome. Manuel Perucho evaluated paired samples of CRC and adjacent non-malignant tissue and found that some tumour samples contained deletions in simple repeat sequences. This genotype existed in only 12% of samples although cells harbouring these deletions were estimated to have as many as 100,000 such mutations^368,369. Stephen Thibodeau independently identified this phenotype in a subset of CRC lesions and found a direct correlation with proximal primary tumour location and better survival outcomes and an inverse correlation with loss of heterozygosity events³⁷⁰. This research by Thibodeau et al. appeared in Science on 7 May 1993, together with two complementary manuscripts put forth by an international consortium that included Lauri Aaltonen, Päivi Peltomäki, Bert Vogelstein and Albert de la Chapelle^371,372. The consortium used polymorphic microsatellite markers for genetic linkage analyses to identify potential susceptibility loci in Lynch syndrome families. In the first study, they reported a significant linkage with the dinucleotide microsatellite repeat marker D2S123 on chromosome 2p in two Lynch syndrome families (one from Canada and the other from New Zealand)³⁷¹. The second study, which examined additional kindreds, found that these familial cancers harbour MSI rather than loss of heterozygosity for chromosome 2 markers or an unduly large burden of KRAS, TP53 or APC mutations. These authors also observed that three of the families were not linked with the D2S123 marker, which raised the possibility that genetic heterogeneity underlies Lynch syndrome³⁷². Lindblom et al. linked a second marker with Lynch syndrome later in 1993; this one was located on chromosome 3p³⁷³. The pace of discovery rapidly accelerated from there. Driven by earlier observations that defective MMR leads to MSI in prokaryotes, several groups that included Bert Vogelstein, Nickolas Papadopoulos, Richard Fishel, Richard Kolodner and Ramon Parsons sought to determine whether the chromosome 2p and 3p loci were in fact the locations of human MMR analogues. In short sequence, MSH2 and MLH1 were localized to chromosomes 2p21 and 3p21–23, respectively, and deleterious mutations in both were found among families with Lynch syndrome^{374,375,376,377,378}. Germline mutations in PMS2 and later in MSH6 were also reported among individuals with Lynch syndrome^379,380. Defective MMR was shown to destabilize microsatellite regions in both humans and yeast, and conversely, tumour cells with a hypermutator phenotype were found to be MMR deficient^{97,381,382,383,384,385}. One century after Warthin first described Lynch syndrome, its genetic aetiology was finally brought to light.

CMMRD

CMMRD was first described in two publications published simultaneously in 1999; each reported MMRd in children conceived from consanguineous marriages involving kindreds with constitutional heterozygous MLH1 mutations^193,194. These children, who were found to possess homozygous germline MLH1 mutations, all developed haematological malignancies. Most of these children also had clinical features of neurofibromatosis type I such as café au lait macules. Hundreds of incidences of CMMRD have since been identified and characterized, thus yielding insights into this highly lethal autosomal-recessive condition.

PMS2 mutations account for approximately 60% of CMMRD cases, and another ~20% are the result of MSH6 mutations. The remaining ~20% are associated with MLH1 and MSH2 mutations^195,196,197. Similar to Lynch syndrome, CMMRD is phenotypically heterogeneous, and this heterogeneity can at least in part be attributed to the affected MMR gene. Individuals with CMMRD are at risk of a broad spectrum of malignancies including those of CNS, gastrointestinal and haematological origins. The most common tumours are malignant gliomas, colorectal adenocarcinomas, non-Hodgkin lymphomas and leukaemias¹⁹⁸. MLH1 or MSH2 mutation carriers seem to have a relatively higher risk of haematological cancers, whereas CNS tumours are more prevalent in PMS2 or MSH6 mutation carriers¹⁹⁹. Penetrance patterns are similar to those seen in Lynch syndrome in that biallelic MSH2 or MLH1 mutation carriers typically develop their first cancer at a younger age than MSH6 or PMS2 mutation carriers (median age of 4 years versus 9 years)¹⁹⁹. Most individuals with CMMRD are diagnosed with cancer within their first two decades of life²⁰⁰. Indeed, the cumulative incidence of cancer by 18 years of age is approximately 90%, and outcomes remain poor. Patients with MSH2 or MLH1 variants seem to have a worse prognosis than those with loss of MSH6 or PMS2 (ref. ¹¹).

The early onset and extremely aggressive nature of these cancers reflects a gene-dosage effect. Accordingly, CMMRD tumours are universally hypermutated. CMMRD-related CNS lesions seem to have the highest mutational load, which is probably owing to the presence of secondary polymerase mutations²⁰¹. An extraordinarily high TMB has been observed in biallelic MMRd brain tumours with early acquired polymerase proofreading defects (PPDs). These cancers, which were classified as ultrahypermutated, were initially unexpectedly found to be MSS¹¹⁶. Subsequent investigations showed that many tumours in children with CMMRD are misclassified as MSS when using standard PCR-based methods that evaluate small panels of mononucleotide repeats⁸⁶. Whole-exome sequencing not only confirmed that these tumours have a high indel burden but also enabled the identification of a distinct indel signature. Adult MMRd tumours preferentially develop multibase deletions within longer homopolymers such as polyA or polyC repeats, whereas tumours arising in the context of CMMRD that also harbour PPDs tend to accumulate single-base insertions. Standard PCR assays are not optimized for short indel detection and, in general, only reliably call MSI events that are ≥3 bases in length. These methods therefore might not be appropriate for the unique microsatellite landscape of tumours arising in the context of CMMRD. This limitation has important diagnostic implications given that complete ablation of replication repair is common in CMMRD tumours. This aspect might also be therapeutically relevant given that the neoantigens generated by single-base insertions are often highly immunogenic⁸⁶. From a screening standpoint, MSI can be detected in the non-neoplastic tissues of patients with CMMRD. This phenomenon is referred to as constitutional MSI²⁰². Notably, non-neoplastic tissues harbour the classic microsatellite deletion pattern arising from biallelic MMR loss in the absence of concurrent PPD⁸⁶.

Sporadic MMRd

Sporadic MMRd is most commonly caused by de novo biallelic hypermethylation of the MLH1 promoter^12,106,107. Tumours with sporadic losses of MSH2, MSH6 or PMS2 generally harbour double somatic pathogenic variants²⁰³. However, why hypermethylation preferentially affects MLH1 as opposed to the other MMR genes is not well understood. Biallelic methylation of MLH1 is thought to occur in the context of CIMP, which is characterized by diffuse hypermethylation throughout the CpG islands of promoter regions. Many sporadic MSI-H CRCs contain concomitant BRAF^V600E mutations. In fact, BRAF^V600E is present in approximately 50–60% of MLH1-promoter-methylated CRCs, but only in ~1% of Lynch-syndrome-associated CRCs^204,205,206. Thus, the coexistence of a BRAF^V600E mutation strongly implies MMRd of sporadic rather than germline origins²⁰⁵. However, the functional implications of this association have yet to be fully determined. One plausible explanation is that BRAF^V600E activates DNA methyltransferases, which in turn induce widespread promoter hypermethylation¹⁰⁸. Whether the presence of BRAF mutations confers worse outcomes after ICIs in patients with MMRd CRC remains to be determined^15,207. Furthermore, MSI-H tumours with concomitant BRAF alterations might be sensitive to targeted therapies after disease progression on ICIs²⁰⁸.

Clinical diagnostic and laboratory testing for MMRd and MSI

Accurately determining MMR and MSI status at the time of cancer diagnosis is essential to: (1) identify patients with germline alterations who require tailored long-term management and monitoring for themselves as well as any affected relatives and (2) guide therapeutic decisions, including whether to administer ICIs. Two approaches to screening for MMRd currently exist. The first, immunohistochemistry (IHC), uses antibodies against the four MMR proteins to evaluate their expression in tumour cell nuclei. The second involves measuring microsatellite length to identify tract instability. PCR and next-generation sequencing (NGS) assays both enable reliable interrogation of MSI status; these can be performed on either tissue or liquid biopsy samples. The results of MMR IHC and MSI testing strategies are nearly always congruent; however, given that discordant results can occur and that each approach carries its own pitfalls, running the tests in parallel garners the most comprehensive picture of tumour MMR status and the microsatellite landscape.

Testing for MMRd with IHC

MMR IHC is the most cost-effective and readily available method of determining MMR status. This technique requires minimal quantities of tissue and offers the added benefit of helping to identify the specific affected MMR gene. Pathogenic MMR gene mutations and epimutations typically result in either loss or truncation of the corresponding protein or in epigenetic silencing of the gene, all of which manifest as absent staining on IHC. MMRd is therefore diagnosed when there is complete loss of nuclear expression of one or more MMR proteins in a biopsy sample or surgical specimen²⁰⁹.

Several considerations must be taken into account when interpreting MMR protein IHC. First, MSH2 and MLH1 encode obligatory partner proteins required for MMR heterodimer formation. The loss of either protein almost always leads to proteolytic degradation of their secondary partners. For this reason, MSH2 mutations typically manifests as loss of both MSH2 and MSH6 on IHC, and MLH1 mutations lead to loss of both MLH1 and PMS2. Conversely, mutations affecting the secondary MMR genes (MSH6 and PMS2) do not affect the stability of MSH2 and MLH1, which can continue to dimerize with MSH3 and MLH3, respectively. Thus, MSH6 mutations lead to isolated loss of MSH6 and likewise for PMS2. Second, although MMR protein loss is nearly always homogenous, tumour tissue slices can, rarely, have heterogeneous staining for these proteins owing to technical issues, subclonality or mosaicism²¹⁰. As an example, three of the MMR genes (MSH2, MSH6 and PMS2) contain coding-region microsatellites and are therefore potentially subject to inactivation in the setting of pre-existing MMRd, which can lead to heterogeneous MMR protein expression within the same sample²¹¹. Third, IHC staining can provide false-negative results in as many as ~5–10% of MSI-H tumours. This scenario can occur if, for example, the protein is expressed but dysfunctional owing to an inactivating missense mutation^212,213,214. Fourth, isolated PMS2 loss can occur in the context of MLH1-promoter methylation with heterogeneous MLH1 expression²¹⁵.

Testing for MSI with PCR and NGS

PCR-based testing and NGS can both be used to evaluate tumour DNA for MSI. PCR quantifies the nucleotide repeat burden within predefined panels of microsatellite markers. The currently preferred testing platform comprises five poly(A) microsatellites (BAT25, BAT26, NR21, NR22 and NR24). This pentaplex panel has high levels of sensitivity and specificity, and the repeats display limited degrees of polymorphism, which obviates the need to analyse matched non-malignant tissues^216,217. Newer commercially available panels seek to optimize MSI detection by including longer repeat tracts or more than five microsatellite markers. PCR assays share standard technical limitations including minimum tumour purity requirements. Moreover, most PCR-based testing strategies are calibrated to the microsatellite landscape of CRC, despite MMRd tumours often having organ-specific differences in their MSI profiles. For example, MMRd endometrial cancers tend to have fewer unstable microsatellites with smaller allelic shifts than MMRd CRCs^218,219. As such, the current PCR panels might underestimate the extent of MSI in endometrial and other non-colorectal tumours.

NGS is increasingly being used as a strategy for determining MSI status. NGS offers an important benefit in that it enables mapping of the length distribution of thousands of microsatellite loci. Cataloguing such a large number of microsatellites improves assay sensitivity and probably also helps to overcome the issues relating to heterogeneous MSI profiles across different tumour types²²⁰. More than 10,000 exonic microsatellites can be scanned using whole-exome sequencing; large NGS panels typically cover >1,000 exonic microsatellites with a focus on homopolymers containing more than five repeats. As an example, the MSIsensor algorithm reports the percentage of homopolymer microsatellites harbouring deletions²²⁰. Most tumours categorized as MSI-H have scores >10%. In scenarios when the score is indeterminate (3–10%), an analysis using additional methods such as MMRd IHC or a second bioinformatics approach to draw conclusions regarding tumour MSI status is often necessary^221,222. The primary disadvantages of NGS include the need for technical and bioinformatic expertise, as well as facility capabilities and the associated costs, which can result in longer turnaround times than both IHC and PCR assays.

Importantly, both PCR and NGS have been validated as methods of MSI detection in cell-free DNA and provide results with good levels of concordance with tissue-based tests^223,224,225. This performance provides a minimally invasive alternative MSI testing strategy and can be particularly relevant in patients without tissue available for analysis.

Cotesting strategies

MMR IHC, MSI PCR and MSI NGS all have high levels of sensitivity, specificity and concordance, particularly for gastrointestinal tract and endometrial malignancies^226,227. What then constitutes the best testing strategy? Performing immunostaining for MMR proteins in parallel with NGS is an effective and efficient approach to determining MMR status that minimizes the risk of clinically meaningful false negatives and provides a comprehensive profile of the tumour genome, which can guide treatment selection regardless of MSI status. In patients with tumours harbouring focal MMR losses or preserved MMR protein expression on IHC, NGS can be used to confirm the presence of MSI. However, paired sequencing data suggest that the degree of MSI can vary from one cancer type to another, and some MMRd tumours are MSS or MSI-I, which emphasizes the importance of including MMR IHC²²⁸. Assessments of TMB can help provide diagnostic clarity when assessing tumours suspected to be MMRd but classified as MSS or MSI-I.

Universal screening of MMR/MSI status

The importance of identifying MMRd/MSI-H disease is well-established given the potential therapeutic and germline genetic implications and is particularly pertinent for the three tumour types with the highest prevalence of MMRd: endometrial, colorectal and gastric cancers. To this end, the National Comprehensive Cancer Network guidelines were updated to recommend universal MMR testing of patients with CRC in the year 2017, endometrial cancer in the year 2018 and gastric cancer in the year 2022 (refs. ^229,230,231). Universal testing is both feasible and cost-effective, yet reports from the USA indicate that crude testing rates in patients with CRC increased from 22.7% in the year 2012 to only 71.5% in the year 2021 (refs. ^{232,233,234,235}). Testing rates were especially low for patients with stage IV disease (~66%) despite the known benefits of ICIs for those with MMRd/MSI-H metastatic CRC. Much of the variance in implementation of testing could be attributed to differences between hospitals, and discrepancies between academic and community centres have previously been reported. Taken together, these data highlight the need for institution-level policy implementation to ensure universal MMR testing for patients presenting with CRC^234,236.

Germline MMR testing

Dedicated germline sequencing is currently recommended for all patients with MMRd/MSI-H tumours arising from any MSH2, MSH6 or PMS2 loss or from MLH1 loss that is unrelated to hypermethylation²³⁷. This strategy should not only apply to patients with endometrial, colorectal or gastric cancers; it is important for those with other rarer MMRd/MSI-H tumours regardless of the primary cancer site¹²⁸. Germline sequencing can potentially be omitted for patients with MLH1-promoter-hypermethylated and/or BRAF-mutant MMRd CRC, given that these features are often mutually exclusive with germline MMR variants. However, some patients with germline MLH1 variants might present with concurrent hypermethylation of the second allele²³⁸. Germline testing might therefore be recommended for all patients with MMRd/MSI-H tumours going forward regardless of the MMR gene affected.

Therapeutic strategies for MMRd/MSI-H tumours

Over the past decade, tremendous advances have been made in the management of patients with MMRd/MSI-H cancers. This progress has built off the longstanding observations that MMRd/MSI-H tumours harbour a large number of coding-region microsatellite indels capable of generating FSPs, contain dense lymphocytic infiltrates suggestive of an immune response to tumour-specific neoepitopes and have upregulation of immune-checkpoint proteins indicative of tumour immune evasion. These characteristics all suggest potential susceptibility to ICIs. We now know that antibodies targeting PD-1, its ligand PD-L1, or CTLA4 can provide durable clinical benefit for a large portion of patients with metastatic MMRd/MSI-H cancers, largely irrespective of the anatomical location or histological subtype of the primary tumour. This effect was initially demonstrated in the second-line and later-line treatment settings and led to the first ever tumour-agnostic US Food and Drug Administration (FDA) approval of a cancer medication (Box 3). In subsequent studies, ICIs have been proven to be remarkably efficacious as first-line therapies for patients with metastatic disease and as neoadjuvant therapy across a variety of MMRd/MSI-H tumour types.

Box 3 Pembrolizumab for MMRd tumours: the first tumour-agnostic approval of an anticancer therapy

On 23 May 2017, the US Food and Drug Administration granted accelerated approval to pembrolizumab for the treatment of adult and paediatric patients with unresectable or metastatic mismatch repair-deficient/microsatellite instability-high (MMRd/MSI-H) tumours regardless of the tissue or site of origin. This decision was based on data aggregated from five multicentre, single-arm trials. The pooled cohort included 149 patients with MMRd/MSI-H tumours of 15 different primary histologies. The objective response rate was 39.6%, and the duration of response (DOR) was ≥6 months in 78% of patients²⁵⁴.

This was a historic moment for the field of oncology. This decision marked the first time that the Food and Drug Administration approved a treatment based on a biomarker rather than the primary disease site or underlying histology. In a larger sense, it reflected a major paradigm shift that aligns with our understanding of oncogenesis as a biological process that can involve tumour-agnostic molecular drivers and, in turn, shared therapeutic vulnerabilities. This pan-cancer approach might not always be appropriate, for example, when targeting certain alterations proves to be efficacious in some tumour histologies but ineffective in others. Nevertheless, the implications of this approval were enormous, and it has transformed the landscape of drug development for both common cancers as well as rare malignancies in which the traditional approaches to clinical trial design are often not feasible. Multiple other approvals of tumour-agnostic, biomarker-based treatments have since followed this decision, including those targeting a high TMB, RET fusions, NTRK fusions, HER2 and BRAF^V600E, in addition to MMRd/MSI-H, and more are certain to follow²⁵³.

ICIs for advanced-stage disease

CRC

The efficacy of ICIs in patients with MMRd CRC is by now well-established (Supplementary Table 1). However, early studies testing these agents in unselected patients with metastatic CRC (mCRC) demonstrated almost no clinical activity. The first-in-human, phase I, dose-escalation trial testing nivolumab enrolled a total of 39 patients with treatment-refractory metastatic solid tumours. Only 1 of the 14 patients with mCRC had a response²³⁹. Two years later, a separate phase I trial testing nivolumab demonstrated that 0 out of 19 individuals with mCRC derived benefit from this agent²⁴⁰. The sole responder across these two trial subgroups initiated nivolumab in July 2007 and received a total of five doses, with subsequent imaging investigations in 2008 showing a complete response (CR) and no evidence of disease recurrence as of 2011. This patient had MSI-H mCRC²⁴¹.

Bearing this patient in mind, it was hypothesized that MMRd tumours might be more responsive to ICIs than MMRp tumours. This hypothesis was formally tested in the proof-of-concept phase II KEYNOTE-016 trial, which evaluated pembrolizumab in individuals with MMRd mCRC (cohort A) and MMRp mCRC (cohort B). A third cohort included non-CRC MMRd tumours (cohort C). Pembrolizumab proved effective in patients with MMRd mCRC but not in those with MMRp mCRC. The objective response rates (ORRs) were 40% and 0%, respectively. Median progression-free survival (PFS) and overall survival (OS) were not reached for cohort A but were 2.2 and 5.0 months for cohort B, indicating that those with MMRd CRC who respond to pembrolizumab often glean long-lasting benefit⁷. An expansion analysis published 2 years later reaffirmed these findings⁴.

The multicohort phase II CheckMate 142 trial evaluated single-agent nivolumab as well as nivolumab-based combination regimens in patients with recurrent or metastatic CRC. Nivolumab monotherapy yielded durable responses in patients with MSI-H disease. At a median follow-up duration of 12.0 months, the ORR was 31.1% and the median DOR had not been reached²⁴². Previously treated patients with MSI-H disease received the combination of nivolumab plus ipilimumab in a separate cohort, and indirect comparisons suggest that this approach might improve outcomes compared with anti-PD-1 antibodies as monotherapy. The ORR was 65%, and median PFS, DOR and OS were not reached^243,244. KEYNOTE-164 provided further support for pembrolizumab in previously treated MMRd mCRC. The patients were assigned to cohorts based on whether they had received ≥2 prior treatment lines (cohort A) or ≥1 (cohort B). At a median follow-up of 31.3 months for cohort A and 24.2 months for cohort B, the ORR was 33% in each, and median DOR was not reached in either. Median PFS was 2.3 and 4.1 months, respectively, and median OS was 31.4 months versus not reached²⁴⁵.

In the phase III KEYNOTE-177 trial, 307 patients with treatment-naive MMRd mCRC were randomly assigned to receive first-line pembrolizumab or standard-of-care chemotherapy. Pembrolizumab conferred an approximately twofold improvement in PFS (median 16.5 versus 8.2 months; hazard ratio (HR) 0.59, 95% confidence interval (CI) 0.45–0.79). The median DOR was not reached for pembrolizumab versus 10.6 months for chemotherapy, which provides further evidence of the durable antitumour activity of ICIs. The median OS was not reached for patients receiving pembrolizumab and 36.7 months for those receiving chemotherapy (P = 0.036). This difference did not achieve the prespecified α of 0.025, although this comparison might have been confounded by the high crossover rate (60%) from chemotherapy to on-study or off-study ICIs^15,207.

Despite these promising results, many patients with MMRd mCRC who receive first-line anti-PD-antibodies have disease progression as their best response. Growing data now suggest that outcomes improve with combination ICI regimens. The data from the multicohort CheckMate 142 trial showed that patients with MMRd mCRC derive high levels of benefit from first-line nivolumab plus low-dose ipilimumab. The ORR was 69% after a median of 29.0 months of follow-up, and the median PFS and OS were not reached after a follow-up duration of 24.2 months¹⁷. CheckMate 8HW is an ongoing phase III trial comparing nivolumab plus ipilimumab versus chemotherapy or nivolumab alone in patients with MMRd/MSI-H mCRC. Nivolumab plus ipilimumab was associated with significantly improved PFS compared with chemotherapy alone: 24-month PFS was 72% versus 14% (P < 0.001; difference in estimated restricted mean survival time 10.6 months, 95% CI 8.4–12.9) in patients with centrally confirmed MSI-H or MMRd disease¹⁹. A more-recent update indicates that nivolumab plus ipilimumab is associated with significantly improved PFS compared with nivolumab monotherapy across all treatment lines (HR 0.62, 95% CI 0.48–0.81) raising the possibility of this combination as standard-of-care first-line therapy for patients with MMRd mCRC²⁰.

Endometrial cancer

KEYNOTE-016 provided the earliest evidence in support of anti-PD-1 antibodies for patients with previously treated advanced-stage MMRd/MSI-H endometrial cancer. This trial involved patients with treatment-refractory non-CRC MMRd tumours. Both patients with MMRd/MSI-H endometrial cancer had a partial response to pembrolizumab⁷. An expansion analysis published in 2017 included 86 patients with 12 different cancer types, including 15 with endometrial cancer. Across the entire trial, objective responses were observed in 53% of patients, and those with endometrial cancer similarly had an ORR of 53% (ref. ⁴). The phase II KEYNOTE-158 trial further evaluated pembrolizumab in patients with previously treated advanced-stage non-CRC MMRd/MSI-H tumours. Patients with endometrial cancer again derived clear benefit from pembrolizumab, with an ORR of 57.1% in an initial analysis of data from 49 patients¹⁴. In a subsequent report that included more patients with endometrial cancer (90) with longer follow-up, the ORR was 48%, median PFS was 13.1 months, and the median DOR and OS were not reached²⁴⁶. The phase I GARNET trial included 143 patients with endometrial cancer who had disease progression on or after a platinum-containing regimen. The ORR was 45.5%, and the median DOR was not reached^18,247. In a single-arm phase II basket trial that included 35 patients with previously treated, advanced-stage MMRd or hypermutated endometrial and ovarian cancers who received nivolumab, the ORR was 58.8%, and the median DOR was not reached²⁴⁸.

Three randomized controlled trials have confirmed that adding an ICI to first-line chemotherapy improves outcomes for patients with MMRd endometrial cancer. The phase III NRG-GY018 trial randomly assigned patients with advanced-stage or recurrent endometrial cancer to receive carboplatin–paclitaxel plus pembrolizumab or placebo. The addition of pembrolizumab led to significantly improved PFS. This effect was particularly pronounced among the MMRd/MSI-H disease subgroup (composed of 225 patients); at the 12-month analysis in this subset, PFS was 74% for the pembrolizumab group and 38% for the placebo group (HR 0.30, 95% CI 0.19–0.48)²⁴⁹. These results were reaffirmed at an interim analysis²⁵⁰. The phase III RUBY trial randomly assigned 494 patients with primary advanced-stage or recurrent endometrial cancer to receive carboplatin, paclitaxel and either dostarlimab or placebo. PFS and OS at 24 months were both significantly improved in the dostarlimab group. Furthermore, benefit from PD-1 blockade was again most pronounced in the subset of patients with MMRd tumours; the 24-month PFS was 61.4% versus 15.7% (HR 0.28, 95% CI 0.16–0.50; P < 0.001), and the median OS not reached versus 31.4 months (HR 0.32, 95% CI 0.17–0.63; nominal P = 0.0002)^16,251. In the phase III DUO-E trial, patients with newly diagnosed advanced-stage or recurrent endometrial cancer were randomly assigned to receive carboplatin and paclitaxel plus either placebo, the anti-PD-L1 antibody durvalumab or durvalumab plus the PARP inhibitor olaparib. Treatment with durvalumab or durvalumab plus olaparib was associated with significantly improved PFS outcomes compared with the control treatment. The magnitude of benefit was again greatest for patients with MMRd disease (for durvalumab versus placebo, median PFS was not reached versus 7.0 months; HR 0.42, 95% CI 0.22–0.80)²⁵². Altogether, these data provide robust evidence supporting the use of first-line regimens that combine anti-PD-1 or anti-PD-L1 antibodies with chemotherapy.

ICIs across other MMRd solid tumours

Data from various studies demonstrate that ICIs confer durable clinical benefit in patients with MMRd/MSI-H solid tumours regardless of the anatomical origins or histological subtype (Supplementary Table 2). The first trial to do this was KEYNOTE-016, which tested pembrolizumab in patients with non-CRC MMRd cancers. The ORR was 71%, the median PFS was 5.4 months and the median OS was not reached⁷. The expansion analysis included 86 patients with 12 different non-CRC MMRd/MSI-H cancer types, and the ORR was 54% (ref. ⁴). These results were combined with data from four other single-arm trials testing pembrolizumab. The pooled cohort included 149 patients with MMRd/MSI-H cancers who collectively had an ORR of 39.6%. Most patients (78%) had a DOR ≥6 months. In light of these data, the FDA approved pembrolizumab for patients with treatment-refractory, unresectable or metastatic MMRd/MSI-H solid tumours in May 2017. This marked a historic moment as the first tumour-agnostic approval of any cancer drug and set a precedent to be followed by other future approvals^253,254.

The KEYNOTE-158 and GARNET trials provided further support for the tumour-agnostic use of anti-PD-1 or anti-PD-L1 antibody monotherapy in patients with MMRd/MSI-H cancers. KEYNOTE-158 tested pembrolizumab in 233 patients with one of 27 tumour types. These included endometrial cancer (as discussed above), gastric cancer, cholangiocarcinoma, pancreatic cancer, cancer of the small intestine, ovarian cancer and CNS cancers. At a median follow-up duration of 13.4 months, the ORR for the entire cohort was 34.3%. ORRs were 45.8% for gastric cancer, 42.1% for cancer of the small intestine, 40.9% for cholangiocarcinoma, 33.3% for ovarian cancer and 18.2% for pancreatic cancer. No responses were observed among patients with CNS cancers. As had been seen previously, those with a response often had durable benefit¹⁴. The GARNET trial, testing the anti-PD-1 antibody dostarlimab, also included patients with small intestine cancer (n = 23), gastric cancer (n = 22), pancreatic cancer (n = 12) and ovarian cancer (n = 7); the ORRs in these groups were 39.1%, 45.5%, 41.7% and 42.9%, respectively. The patients with a response again tended to have durable clinical benefit¹⁸.

ICIs for patients with CMMRD

Children and young adults with CMMRD have a markedly elevated risk of multiple cancers including CNS, gastrointestinaI and haematological malignancies and traditionally had a dismal prognosis owing to a dearth of effective therapies. After the efficacy of ICIs for MMRd solid tumours was initially established, a hypothesis emerged that patients with CMMRD-associated tumours would similarly benefit given their characteristic high mutation burden and neoantigen load. In 2016, investigators described the activity of nivolumab in two siblings with germline homozygous, biallelic PMS2 mutations and recurrent multifocal glioblastoma. Both patients had durable objective responses²⁰¹. Elsewhere, the activity of anti-PD-1 antibodies was investigated in 38 paediatric and young adult patients with 45 hypermutated tumours driven by germline DNA repair defects. This cohort included 28 patients with CMMRD, 8 with Lynch syndrome and 2 with PPDs. Their cancers, which were predominantly recurrent and/or treatment refractory, were classified as CNS tumours (n = 31), non-CNS solid tumours (n = 11) and haematological malignancies (n = 3). Across the entire cohort, 55.5% had either a response or stable disease, and most of these (80%) were sustained after a median follow-up duration of 1.9 years. In line with this observation, the estimated 3-year OS across the entire cohort was 41.4%. The patients with non-CNS solid tumours had the best outcomes (100% response rate) although the response rate of 64% and OS of 39.3% among patients with CNS tumours is encouraging given the historically dismal prognosis of this group. None of the three patients with haematological cancers had a response. Nonetheless, the consistent efficacy seen across solid tumours confirms that anti-PD-1 antibodies are a rational therapeutic approach for patients with CMMRD-associated non-haematological malignancies. Determining whether the high risk of second primary malignancies is ameliorated by anti-PD-1 antibodies in this population will be an interesting question. Future research should also focus on the added value of first-line ICI combination regimens²⁵⁵.

Duration of ICIs

ICIs are increasingly being used as first-line therapies for patients with advanced-stage or recurrent MMRd tumours, therefore, defining the optimal treatment duration for those who have a response has become an important question. To this end, data from 757 patients with MMRd/MSI-H mCRC who received ICIs were evaluated in a retrospective analysis. This analysis revealed no statistically significant difference in OS between patients who received ICIs for 2 years versus >2 years. In patients who had a CR, treatment discontinuation after 1 year was not associated with worse OS. These data suggest that a fixed treatment duration of up to 2 years might be appropriate for most patients; shorter durations of therapy can be considered for those with a CR²⁵⁶. Going forward, determining whether these findings translate to other MMRd cancer types will be important.

Neoadjuvant and adjuvant ICIs

Neoadjuvant ICIs

Given that patients with a variety of advanced-stage and metastatic MMRd tumours can respond to ICIs, considerable research interest exists in defining the role of these agents in patients with early stage disease (Supplementary Table 3). Accumulating data support the use of neoadjuvant ICIs for patients with MMRd/MSI-H cancers. In the phase II NICHE trial, 20 patients with resectable MMRd colon cancer received one dose of ipilimumab and two doses of nivolumab before proceeding to surgery. All patients had a pathological response, and 12 (60%) had a pathological CR (pCR)²⁵⁷. In the follow-up phase II NICHE-2 trial, investigators observed a pathological response in 99% of efficacy evaluable patients (109/111) and a pCR in 75/111 (68%)²⁵⁸. The phase II NICHE-3 trial demonstrated that 2 months of neoadjuvant PD-1 blockade combined with an anti-LAG3 antibody is similarly efficacious²⁵⁹. In the phase II NEONIPIGA trial, neoadjuvant nivolumab plus ipilimumab for 3 months followed by surgery and then adjuvant nivolumab was tested in patients with locally advanced, resectable MMRd/MSI-H gastric or gastroesophageal cancer. All 29 patients who underwent surgery had an R0 resection and 17 (59%) had a pCR²⁶⁰. In another phase II trial, a 6-month course of neoadjuvant pembrolizumab was found to be efficacious across 35 patients with resectable solid tumours including 27 with CRC. The pCR was 65% among 17 patients who underwent surgery and 79% for the subset with CRC. At a median follow-up duration of 9.5 months, only two of the patients who did not undergo surgery had disease progression, suggesting that ICIs alone might act as definitive therapy for some patients with locally advanced MMRd tumours²⁶¹. To this end, in a phase II trial, 12 patients with previously untreated stage II or III MMRd rectal adenocarcinoma received 6 months of neoadjuvant dostarlimab. This neoadjuvant therapy was to be followed by active surveillance for those with a clinical CR (cCR) versus standard chemoradiotherapy and surgery for those without a cCR. All 12 patients had a cCR to dostarlimab, allowing them to forego chemoradiotherapy and surgery in favour of active surveillance²⁶². These results were confirmed in an extended cohort in which all patients with MMRd rectal cancer (100%) had a cCR. In a separate cohort of patients with non-rectal MMRd solid tumours, 6 months of neoadjuvant dostarlimab led to a cCR rate of 65% without compromising the feasibility of surgical resection if needed²⁶³. The implications of these findings are enormous given the potential short-term and long-term morbidities avoided and the opportunity for organ preservation provided by this approach.

Multiple other similar studies continue to add to this body of literature, and the available data highlight the promise of neoadjuvant ICIs for patients with early stage MMRd cancers^{264,265,266,267,268,269,270,271,272,273,274,275,276,277,278,279}. Across these studies, the proportion of patients with pathological responses is remarkably high, and in many cases, the pCR rate is unprecedented. Going forward, elucidating why ICIs have greater clinical activity when used earlier and in the non-metastatic setting will be an important step. Improvements in response rates are not limited to patients with MMRd cancers but have also been observed in neoadjuvant trials involving patients with melanoma and to some extent non-small-cell lung cancer. Multiple possible reasons for this improved activity exist and are probably not necessarily specific to MMRd tumours, including the selection of cancer cells lacking immunogenic neoantigens and the activation of immune evasion signals, both of which can facilitate metastatic spread and impede responses to ICIs.

Several other questions persist. The potential role of neoadjuvant ICIs is clear for histologies in which upfront surgery can be highly morbid such as rectal, hepatobiliary and bladder cancers. However, the added value of this approach remains unclear for cancers that are more readily resectable with low morbidity, such as colon and endometrial cancers, which can have high cure rates with surgery alone. The optimal duration of neoadjuvant ICIs is another important question. A longer duration of therapy is associated with better pathological response rates, which is in line with the fact that antitumor immune responses often deepen over the course of months ^262,280 (Fig. 4). This association is especially pertinent when bearing in mind the goal of organ preservation. For organ-sparing strategies, defining the optimal restaging modality is imperative given that data from several studies indicate that the findings of radiological evaluations are poorly correlated with pathological response^257,261. Measurements of circulating tumour DNA (ctDNA) might prove helpful towards this end²⁸¹. Considering the robust activity of anti-PD-1 antibodies as monotherapy, yet another question exists as to whether ICI combination regimens augment response rates without worsening toxicities. Finally, early evidence suggests that the primary disease location of MMRd cancers has implications for outcomes. For example, patients with gastric cancers have lower response rates compared with those with rectal adenocarcinoma. However, the tumour intrinsic and extrinsic features that confer better sensitivity to neoadjuvant ICIs in patients with locally advanced disease remain incompletely understood.

Fig. 4: Response to neoadjuvant ICIs based on duration of therapy.

This plot presents the observed complete response (CR) rates from individual trials testing neoadjuvant immune-checkpoint inhibitors (ICIs) in patients with localized and locally advanced mismatch repair-deficient (MMRd) tumours according to the median duration of treatment. A CR includes both pathological and clinical CR (pCR and cCR, respectively). The plot is adapted from work by Rousseau et al.²⁸⁰. The data are reported by treatment category and cancer type, and further information is provided in Supplementary Table 3. Shown in blue is the estimated range of observed CRs across studies. CTLA4, anti-CTLA4 antibody; LAG3, anti-LAG3 antibody; PD-1, anti-PD-1 antibody; PD-L1, anti-PD-L1 antibody; COX2i, cyclooxygenase 2 inhibitor; CRC, colorectal cancer; GEJ, gastroesophageal junction; VEGFRi, vascular endothelial growth factor receptor inhibitor.

Adjuvant ICIs

Growing interest exists in the role of postoperative ICIs for patients with MMRd tumours who have undergone curative intent surgery. The phase III KEYNOTE-B21 is testing adjuvant pembrolizumab versus placebo in combination with chemotherapy with or without radiotherapy in patients with resected endometrial cancer. The addition of pembrolizumab improved disease-free survival among the subgroup of 281 patients with MMRd disease (HR 0.31, 95% CI 0.14–0.69) but not in the pMMR population²⁸². Several other trials evaluating the role of adjuvant ICIs in patients with MMRd tumours are ongoing. The phase III ATOMIC trial is comparing chemotherapy plus atezolizumab to chemotherapy alone in patients with stage III MMRd colon cancer. The addition of atezolizumab has been shown to significantly improve disease-free survival (HR 0.50, 95% CI 0.35–0.72)²⁸³. In a single-arm pilot trial, patients with MSI-H tumours who had detectable ctDNA after completing curative intent surgery and standard-of-care perioperative systemic therapy appeared to derive clinical benefit from 6 months of pembrolizumab²⁸⁴. Although these trials broaden our understanding of the activity of ICIs in patients with early stage MMRd cancer, given the opportunities offered by neoadjuvant ICIs including the potential for organ preservation or less-extensive surgery, the role of postoperative ICIs will need to be refined and might depend on biomarkers such as ctDNA.

Determinants of response and resistance to ICIs

Although ICIs confer long-term disease control for a subset of patients with MSI-H tumours, many individuals do not respond to treatment (primary resistance), and a smaller number will have disease progression after an initial period of clinical benefit (acquired resistance). As many as 25–40% of patients with metastatic MMRd disease will have primary resistance to anti-PD-1 antibodies as monotherapy^14,15,18,19. In patients with MMRd mCRC, this number is lower (~10–15%) for those who receive ICI combination regimens^17,20. A further ~20–30% of patients will initially benefit from treatment but develop acquired resistance to ICIs^{14,15,16,17,18,20}. Much remains to be learned about the mechanisms of primary and acquired resistance to ICIs in patients with MMRd disease. Here, we provide a summary of the molecular, immunological and clinical features associated with the efficacy of ICIs in patients with MMRd cancers.

Mechanism of MMR loss

A growing body of literature suggests that the mechanism of MMR loss has important clinical implications in patients with MMRd endometrial cancer. The data from several studies show that MLH1-promoter-hypermethylated tumours have distinct molecular and clinicopathological features compared with those with mutations in MMR genes. Endometrial tumours with MLH1-promoter hypermethylation have been reported to have a lower TMB and fewer TILs than those with germline and/or somatic mutations in MMR genes²⁸⁵. Other groups have similarly shown that patients with methylated MMRd endometrial cancer have a lower TMB and less immune cell infiltration compared with their counterparts with mutational MMRd disease^80,286. Taken together, these data suggest that MLH1-promoter-methylated MMRd tumours have a unique genetic and epigenetic landscape with reduced levels of immunogenicity. Although the precise mechanisms underlying this phenotype remain unclear, it might have prognostic and therapeutic relevance. Studies focusing on response to anti-PD-1 antibody monotherapy suggest that patients with methylated MMRd tumours have worse outcomes than those with mutational MMRd disease. In a phase II trial that compared the outcomes of patients with mutated versus methylated MMRd endometrial cancer receiving pembrolizumab, patients with methylated disease had worse 3-year PFS and OS (100% versus 30% and 100% versus 43%, respectively)²⁸⁷. In a translational analysis of samples from this trial, patients with epigenetic MMRd tumours who had a response to pembrolizumab had improved CD16⁺ natural-killer-cell-driven immunity, whereas samples from those with mutational MMRd who had a response were enriched with effector CD8⁺ T cells⁸⁰. However, data from two phase III trials that evaluated chemotherapy with versus without pembrolizumab or dostarlimab in patients with advanced-stage endometrial cancer did not demonstrate a statistically significant difference in PFS based on the mechanism of MMR loss^288,289. The difference in outcomes seen with ICIs alone might have been corrected for by the addition of chemotherapy for all patients in these trials. Overall, the underlying features that predispose to differential antitumour immune responses and dictate overall prognosis in patients with mutated versus methylated MMRd endometrial cancer have yet to be determined. One important question is whether and to what degree the global epigenetic landscape of MLH1-promoter-hypermethylated cancers affects tumour phenotype, including responsiveness to CD8⁺ T cell-mediated cytotoxicity.

Going forward, determining whether findings from endometrial cancer translate to other MMRd cancer types will be important. Emerging evidence from patients with CRC supports this possibility. For example, sporadic MMRd CRCs are characterized by a lower somatic mutational burden, lower neoantigen load and fewer TILs relative to their germline counterparts²⁹⁰. Another study showed that patients with CRC or endometrial cancers with co-loss of MLH1 and PMS2 have worse outcomes than those with co-loss of MSH2 and MSH6 (ref. ²⁹¹). Although these tumours were not classified by methylation status, we know that most MLH1 losses occur in the context of MLH1-promoter hypermethylation. Along the same lines, data from the NICHE-2 trial demonstrate that patients with Lynch-syndrome-associated CRC have a higher pCR rate compared with those with sporadic MMRd loss²⁵⁸.

Mutation burden, class and clonality

MMRd tumours were initially hypothesized to be more sensitive to ICIs than MMRp tumours because they harbour a large number of mutations and, in turn, many potential mutation-derived neoantigens. However, the available data paint a more nuanced picture. Although TMB correlates with the predicted number of mutation-associated neoantigens, this metric has limited predictive utility in the context of MMRd disease, and the optimal TMB threshold in MMRd tumours remains unclear^80,81,82,292. This observation most probably reflects that the mutation class carries more relevance to antitumour immunity than the overall mutation load. In this regard, frameshift mutations, which are coding-region indels that alter the translated open-reading frame and can therefore generate highly immunogenic FSPs, are thought to elicit stronger immune responses than the neoepitopes resulting from missense mutations¹²¹. Yet, although the number of frameshifts or indels has been shown to correlate with higher TIL levels and better responses to ICIs, frameshift mutational burden alone is an insufficient biomarker^4,5,293,294. In a similar vein, MSI score, which provides another measure of indel burden, is associated with a higher likelihood of a response to ICIs in some but not all studies^5,292. Classifying FSPs by their predicted major histocompatibility complex (MHC)-binding affinity and neoepitope immunogenicity might provide a more refined prediction of the likelihood of a response to ICIs. The clonal burden of these neoantigens might also dictate the extent of engagement of antitumour immunity. MMRd tumours have high levels of intratumoral genomic heterogeneity, and data suggest that clonal neoantigen burden correlates with benefit from ICIs^87,295. At the same time, higher baseline T cell receptor (TCR) diversity has been associated with a greater likelihood of benefit from ICIs, suggesting that the recognition of multiple neoantigens might be crucial for deeper antitumour responses⁸¹. Ultimately, however, these analyses have involved cohorts of limited size, and the interplay between clonal genomic architecture and T cell response in MMRd tumours requires further investigation.

MHC class I defects

Whether compromised antigen processing and presentation facilitates immune evasion and, in turn, impedes responses to ICIs in patients with MMRd disease has been a longstanding area of research interest. Many MSI-H tumours do indeed harbour alterations that impair MHC class I (MHC I)-mediated antigen presentation. For example, B2M, which encodes the shared β2-microglobulin subunit of MHC I molecules, contains multiple coding-region microsatellites, and approximately 25–30% of MSI-H cancers harbour biallelic loss-of-function mutations in B2M²⁹⁶. Other human leukocyte antigen (HLA) class I-related genes are also frequently mutated in MMRd tumours, such as the MHC I heavy chain-encoding genes HLA-A, HLA-B and HLA-C, as well as the transporters of antigen presentation TAP1 and TAP2, which have a role in loading peptides onto MHC I molecules²⁹⁷. Importantly, however, the presence of these mutations does not preclude clinical benefit from ICIs, and patients with MMRd tumours that lack MHC I expression can still glean durable clinical benefit from ICIs^298,299,300 (Fig. 5). This seemingly paradoxical observation suggests that MMRd cancers with MHC I defects are sensitive to immune effectors other than CD8⁺ T cells. Towards this end, translational data from two studies demonstrate that MHC I-deficient MMRd tumours are sensitive to ICI combination regimens owing to the antitumor activity of CD4⁺ and γδ T cells^298,299. The added value of combining ICIs targeting CTLA4 or LAG3 with those targeting PD-1 or PD-L1 in MMRd tumours with MHC I loss remains to be prospectively validated.

Fig. 5: Dynamics of the immune microenvironment in MMRd tumours.

The tumour microenvironment of DNA mismatch repair-deficient (MMRd) tumours is enriched at baseline with a variety of immune cells including CD8⁺ T cells, CD4⁺ T cells, natural killer (NK) cells, dendritic cells and myeloid cells. Classically, these leukocytes are present in an immune exclusion pattern featuring upregulation of immune-checkpoint proteins such as PD-1 and CTLA4. In these cancers, PD-L1 is primarily expressed on immune cells rather than on the tumour cells themselves. In this context, the antitumour immunity mediated by immune-checkpoint inhibitors (ICIs) is only partially dependent on whether the cancer cells are major histocompatibility complex (MHC) class I proficient or deficient. In MHC-I-proficient MMRd tumours, the response to ICIs seems to be mediated by CD8⁺ T cells, whereas in those that are MHC-I-deficient, this response is mediated by CD4⁺ T cells and γδ T cells. The data from patients treated with PD-1 blockade suggest that response is initially characterized by an increase in effector T cells and the production of interferon-γ (IFNγ). MMRd tumours that respond to ICIs are also characterized by B cell infiltration, and data from serial biopsies suggest that the presence of tertiary lymphoid structures (TLS) is a later development. HLA, human leukocyte antigen; TAMs, tumour-associated macrophages; TCR, T cell receptor; ?, potential immune effectors that remain to be determined.

Tumour immune microenvironment

MSI-H tumours have long been known to be enriched with TILs at baseline (Fig. 5). The prominent antitumour immune landscape of such tumours includes a diverse array of lymphocytes such as CD4⁺ T cells, CD8⁺ T cells, γδ T cells, NK cells and B cells^{13,158,159,257,301,302,303}. This typically high level of lymphocyte infiltration is thought to be elicited by immunogenic FSPs, although interestingly, immune infiltration is even seen in the polyps of individuals with Lynch syndrome irrespective of somatic mutation rate, neoantigen load and MMR status³⁰⁴. Nevertheless, limited evidence exists that the pretreatment abundance of intratumoural immune cells predicts benefit from ICIs. Several groups have described relatively higher amounts of cytotoxic lymphocytes at the invasive margins of MMRd CRCs compared with within the centre of the tumour, although whether spatial localization of the in situ immune response informs upon outcomes following treatment with ICIs remains to be determined^13,305.

MSI-H tumours are characterized by parallel upregulation of immune-checkpoint proteins including PD-1, PD-L1, CTLA4 and LAG3 (refs. ^13,155). PD-L1 expression by cancer cells and within the tumour microenvironment (TME) has been described in many cancer types as predictive of benefit from ICIs. However, evidence supporting a role of PD-L1 expression as a biomarker specifically in the context of MMRd disease remains scarce. Patients with CMMRD-related cancers and high levels of PD-L1 expression (defined as ≥1% PD-L1-positive tumour cells) had better responses to anti-PD-1 antibodies than those with PD-L1 levels below this threshold²⁵⁵. In another trial, cancer cell PD-L1 expression (based on the same cutoff) did not correlate with a response to ICIs³⁰⁶. Work by others suggests that MMRd tumours have limited or even absent cancer cell PD-L1 expression. Instead, PD-L1 expression seems to be primarily confined to infiltrating immune cells located within the TME^155,305,307. Data from patients with MMRd CRC demonstrated that PD-L1 is predominantly expressed on tumour-associated monocytes and macrophages, neutrophils and myeloid regulatory dendritic cells³⁰⁸. Therefore, the combined-positive score, which accounts for PD-L1 expression on tumour and immune cells, might have greater predictive value than the other available PD-L1 scores and has been associated with response to anti-PD-1 antibodies in patients with MMRd endometrial carcinoma³⁰⁹. Elevated PD-1 and CTLA4 expression on TILs within the MMRd TME might also be more informative than cancer cell PD-L1 expression^302,310.

The dynamics of immune cell infiltration after ICIs provides important clues on the mechanisms that govern antitumor immunity in the context of MMRd disease. MMRd CRC lesions responding to ICIs have increased CD8⁺ T cells and high IFNγ scores compared with baseline values^257,262. Similarly patients with MMRd gastric cancer responding to ICIs also have an expansion of CD8⁺ and γδ T cell populations whereas those without a response have increased levels of CD8⁺ T cell exhaustion⁸¹. The evolving antitumour immune response following ICIs has been best described in localized MMRd CRC²⁶². Analysis of serial biopsy samples shows that patients with a response are more likely to develop intratumoural tertiary lymphoid structures (TLS)^257,262. In complete responders, T cells and B cells depart the healed epithelium, which is later repopulated by resident CD8⁺ T cells and B cells^262,311. Thus, the establishment of TLS might have an important role in long-term locoregional disease control.

Finally, data from several studies have shown that an immune-excluded TME enriched with fibrotic stroma impedes responsiveness to ICIs. This effect has been reported in both MMRd gastric and CRCs^292,312. Fibrosis might in and of itself dampen antitumour immune activity, although fibrotic tumours also have higher expression of TGFβ, which can facilitate immune evasion and T cell exhaustion and thus further drive resistance to ICIs³¹³.

cGAS–STING signalling

The cyclic GMP–AMP synthase (cGAS) and stimulator of interferon genes (STING) signalling pathway has an important role in innate antiviral immunity, and a growing body of data suggest that this signalling pathway has a role in the anticancer immune response in MLH1-deficient MMRd tumour cells. cGAS is a cellular sensor that binds to aberrant dsDNA including viral DNA or, more importantly, self dsDNA that is located in the cytoplasm instead of the nucleus. The binding of cGAS to cytosolic dsDNA fragments activates its catalytic activity leading to the production of cyclic GMP–AMP (cGAMP). In turn, cGAMP binds to STING, which mediates downstream induction of type 1 IFNs and chemokines. Signalling downstream of IFN and chemokine receptors results in the activation of dendritic cells and, ultimately, supports the priming of CD8⁺ T cells^314,315,316.

MMRd tumours with MLH1 loss are particularly susceptible to high levels of cytosolic dsDNA. This reflects the role of the MLH1–PMS2 heterodimer MutLα in the regulation of downstream Exo1 exonuclease activity. In the setting of MLH1 loss, Exo1 hyperactivity induces perpetual DNA breaks, resulting in the concomitant leakage of damaged nuclear dsDNA into the cytosol³¹⁷. The data from preclinical models show that functional cGAS–STING signalling is imperative for antitumour immunity in MLH1-deficient MMRd tumours. In line with this observation, patients with MLH1-deficient MSI-H cancers and high levels of cGAS or STING signalling had better outcomes following treatment with anti-PD-1 antibodies than their counterparts with low levels of cGAS or STING activity³¹⁴.

Clinical determinants of response

Multiple studies have investigated the predictive and prognostic utility of pretreatment clinical features in patients with MMRd cancers receiving ICIs. Growing amounts of data demonstrate that patients with better performance status and those who receive treatment at an earlier rather than later juncture are more likely to derive durable benefit^15,318. This observation might reflect a variety of reasons including differences in tumour burden, resistance to apoptosis after prior chemotherapy and general immunosuppression in heavily pretreated patients and/or the loss of neoantigens after previous systemic therapies. The data from several studies suggest that both the primary and metastatic sites of disease might also have implications for response rates and overall prognosis. For example, MMRd cancers of the CNS seem less likely to respond to ICIs than other MMRd tumours³¹⁹. Patients with peritoneal metastases or a higher overall disease burden also tend to have worse outcomes^320,321. Whether these differences in prognosis arise from variances in the MMRd genotype across primary and metastatic tissues or simply reflect that relatively immunoprivileged sites such as the peritoneum, pancreas and brain tend to be less responsive to ICIs irrespective of MSI-H status remains to be determined.

In conclusion, although many insights into the determinants of ICI response and resistance in MMRd tumours have been reported, our biological understanding of these potential predictors remains incomplete, and the level of definitive evidence for any given candidate biomarker is typically low enough to preclude clinical implementation in their current form. Interestingly, the data from multiple studies have shown that poor responses to ICIs can be attributed to the misclassification of a tumour as MMRd/MSI-H when it is in fact MMRp/MSS^292,322. Thus, accurate diagnosis of MMR status is itself an important determinant of response.

Future treatment avenues for MMRd tumours

Targeting oncogenic drivers

MMRd cancers harbour many potential driver alterations, and although these might include potential therapeutic targets, the multitude of mutations confers a high risk of developing resistance to specific kinase inhibitors. Sporadic MMRd CRCs often harbour BRAF^V600E alterations that are theoretically targetable with BRAF inhibitors, although benefit from this approach has not been clearly shown in clinical trials^323,324. CRCs arising from sporadic MMRd are also enriched in rare oncogenic gene fusions, for example, involving NTRK1 and NTRK3 that might be targetable with the same inhibitors that have been highly successful in patients with MMRp solid tumours harbouring such fusions¹¹⁷.

Werner syndrome helicase

Werner syndrome helicase (WRN) is a multifunctional RecQ DNA helicase with a key role in multiple DNA repair pathways. One of the functions of this enzyme is to unwind DNA and resolve secondary structures such as forks, bubbles and displacement loops^325,326,327. These substrates are highly prevalent in MSI-H cells owing to widespread MSI. TA-dinucleotide repeats are particularly susceptible to large-scale expansions that can subsequently form stable non-B DNA structures, which are in turn dependent on WRN for repair³²⁸. In fact, WRN seems to be selectively essential for maintaining chromosomal integrity in MSI-H cells, and growing amounts of data support the existence of a synthetic lethal interaction between MSI-H status and WRN inactivation^329,330,331. Indeed, WRN inhibition in MMRd tumours has been shown to increase dsDNA breaks, downregulate G2/M checkpoint progression and increase the expression of proapoptotic genes^331,332. Allosteric WRN inhibitors have shown promising cytotoxic activity in preclinical models of MSI-H cancers^333,334. Interestingly, WRN inhibition was successful in these models regardless of previous exposure to ICIs, suggesting that this is a rational therapeutic strategy for patients with tumours that have already evaded the immune system^330,331,333. Several clinical trials evaluating WRN inhibitors are now ongoing, such as NCT05838768 and NCT06004245.

Preventative strategies for patients with hereditary MMRd syndromes

Screening and surveillance

Patients with heritable MMR alterations have an elevated risk of the early onset of various malignancies including colorectal, endometrial and gastric cancers. Comprehensive surveillance programmes are essential to optimize the early detection of cancer in this population and, in turn, improve their long-term prognosis. Detailed screening guidelines have been put forth by multiple groups, are largely consistent and, in some cases, are tailored to the specific MMR gene affected^{335,336,337,338}. High-quality colonoscopy is recommended for all pathogenic and probable pathogenic MMR variant carriers. CRC screening should generally begin by no later than 25–35 years of age depending on the MMR gene affected. The recommended repeat colonoscopy interval varies between guidelines and is based on the MMR gene variant. The level of evidence supporting surveillance strategies for other hereditary MMRd-associated cancers is not well-established, and specific screening interventions should be considered on an individual patient basis. Importantly, individuals with pathogenic and probable pathogenic MMR gene variants should establish care with a specialized practitioner to develop a personalized follow-up plan.

Prophylactic surgery

Prophylactic hysterectomy and bilateral salpingo-oophorectomy might reduce the risks of endometrial and ovarian cancer in women with Lynch syndrome and should therefore be considered for this group. The selection and timing of such procedures should be individualized and based on several factors including the completion of childbearing, a family history of cancer, the MMR gene affected and menopausal status^338,339. Prophylactic total colectomy is not recommended for patients with Lynch syndrome owing to the incomplete penetrance of the alterations, quality of life implications and the efficacy of routine minimally invasive endoscopic screening^336,338. In individuals with Lynch syndrome who have been diagnosed with a primary colon cancer or adenoma that cannot be endoscopically removed, extended surgical resection (subtotal or total colectomy) should be recommended owing to the increased risk of metachronous cancer among those undergoing segmental resection. However, less-extensive surgery can be considered for certain patients including those initially diagnosed with colon cancer at an older age^{336,337,338,340}.

Aspirin prophylaxis

Regular use of aspirin might reduce the risk of CRC in individuals with Lynch syndrome, although aspirin use does not seem to confer protection against other Lynch-syndrome-associated cancers. The prospective randomized CAPP2 trial initially found no significant clinical benefit from aspirin prophylaxis. However, long-term follow-up analyses showed that individuals with Lynch syndrome who received 600 mg of aspirin per day for 2 years are less likely to develop CRC than those who received placebo (for the intention-to-treat analysis, HR 0.65, 95% CI 0.43–0.97). However, benefit did not become apparent until approximately 5 years after treatment initiation. No significant difference in the development of extracolonic Lynch-syndrome-associated cancers emerged (for the intention-to-treat analysis, HR 0.94, 95% CI 0.59–1.50)³⁴¹. These data support a potential role for long-term aspirin use in reducing the risk of developing CRC in individuals with Lynch syndrome.

Vaccines

Vaccination against recurrent FSP neoantigens provides a promising strategy for the interception of developing cancers in individuals with Lynch syndrome and a possible treatment for those with manifest MSI-H disease. In a first-in-human phase I/II trial, investigators evaluated the safety and immunogenicity of a FSP neoantigen-based vaccine that included three recurrent FSPs: TAF1B, HT001 and AIM2 (ref. ³⁴²). The vaccine was well-tolerated, and all 19 individuals who were evaluable for immunological efficacy had an immune response to at least one of the peptides. The polyvalent viral-vectored vaccine Nous-209 is currently being evaluated as a cancer prevention strategy for individuals with Lynch syndrome without evidence of cancer (NCT05078866). Nous-209 encompasses 209 FSPs derived from 204 frameshift mutations that are shared across MMRd CRC, endometrial and gastric cancers³⁴³. This approach is also being tested as a therapeutic strategy in combination with anti-PD-1 antibodies for patients with MSI-H CRC, gastric and gastroesophageal cancers (NCT04041310). The preliminary data are promising; Nous-209 is well-tolerated, immunogenic and seems to have improved efficacy when combined with pembrolizumab^344,345,346.

Preventative ICIs

Immunoprevention of cancer with ICIs is a biologically plausible strategy for individuals with Lynch syndrome. Yet, in a cohort of 172 patients with Lynch-syndrome-associated cancers who previously received ICIs, 21 (12%) developed subsequent primary malignancies (SPMs), and the overall incidence of neoplasia per patient did not change from before to after ICIs³⁴⁷. The incidence of SPMs was also similar between patients receiving and not receiving ICIs, and the rate of adenoma development also did not change in the post-ICI period. Collectively, these data suggest that ICIs do not affect the development of pre-malignant lesions or cancers in patients with Lynch syndrome. Moreover, although the vast majority (92%) of SPMs were MMRd, these cancers seemed to be genetically ‘younger’ than their first primary malignancy counterparts, as indicated by lower levels of MSI and fewer somatic mutations. These findings suggest the existence of a period of time during MMRd tumour development during which neoantigen quantity and quality do not exceed the threshold required to activate an innate immune response. The lack of shared neoantigens between SPMs and first primary malignancies is equally noteworthy; the immunological memory established after ICIs might not be relevant to new clones with new neoantigens. Patients with Lynch syndrome-related cancers should therefore continue with cancer surveillance after treatment with ICIs. Importantly, after stratifying SPMs by cancer type, the incidence of non-index visceral cancers decreased after ICIs whereas that of skin cancers increased even in those who had not been previously diagnosed with MTS. Dermatological screening should therefore be considered for those with Lynch syndrome who receive ICIs.

Conclusions

Over the past decade, data from multiple clinical trials have shown that a subset of advanced-stage MMRd cancers are remarkably sensitive to ICIs regardless of the primary tumour site or histology. Growing evidence suggests that this treatment strategy is even more effective in the neoadjuvant setting. Considerable research interest exists in defining the efficacy of ICIs across other clinical scenarios in the context of MMRd cancer, including as an organ-sparing therapeutic approach for patients with locally advanced disease, as adjuvant or maintenance therapies for those with minimal residual disease and as a cancer prevention strategy for individuals with hereditary MMRd syndromes.

Despite the robust activity of ICIs in patients with MMRd tumours, many do not have a response to treatment. We are only beginning to understand the factors that drive primary resistance, and given that most studies have focused on comparing data from responders versus non-responders, much remains to be learned about mechanisms of acquired resistance. Predictors of ICI response or treatment failure with demonstrated utility in MMRp tumours include TMB, PD-L1 positivity and MHC I deficiency, although these have limited predictive value in patients with MMRd disease. This aspect in and of itself highlights the unique biology of MMRd tumours.

MMRd seems to be acquired early during oncogenesis and is followed by the progressive accumulation of mutations and neoantigens, which ultimately predispose to immune sensitivity. Going forward, deciphering the interplay between continuous genomic evolution and the timing and mechanisms of immune escape will be imperative. These efforts stand to inform the windows of opportunity for treatment with ICIs, to identify novel immune-modulating approaches for patients with ICI-resistant disease and to identify therapeutic vulnerabilities related to DNA repair itself.