Genomic predictors of response to PD-1 inhibition in children with germline DNA replication repair d

Cancers arising from germline DNA mismatch repair deficiency or polymerase proofreading deficiency (MMRD and PPD) in children harbour the highest mutational and microsatellite insertion–deletion (MS-indel) burden in humans. MMRD and PPD cancers are commonly lethal due to the inherent resistance to chemo-irradiation. Although immune checkpoint inhibitors (ICIs) have failed to benefit children in previous studies, we hypothesized that hypermutation caused by MMRD and PPD will improve outcomes following ICI treatment in these patients. Using an international consortium registry study, we report on the ICI treatment of 45 progressive or recurrent tumors from 38 patients. Durable objective responses were observed in most patients, culminating in a 3 year survival of 41.4%. High mutation burden predicted response for ultra-hypermutant cancers (>100 mutations per Mb) enriched for combined MMRD + PPD, while MS-indels predicted response in MMRD tumors with lower mutation burden (10–100 mutations per Mb). Furthermore, both mechanisms were associated with increased immune infiltration even in ‘immunologically cold’ tumors such as gliomas, contributing to the favorable response. Pseudo-progression (flare) was common and was associated with immune activation in the tumor microenvironment and systemically. Furthermore, patients with flare who continued ICI treatment achieved durable responses. This study demonstrates improved survival for patients with tumors not previously known to respond to ICI treatment, including central nervous system and synchronous cancers, and identifies the dual roles of mutation burden and MS-indels in predicting sustained response to immunotherapy.

Accurate DNA replication in eukaryotic cells is ensured by the DNA polymerases Pol δ and Pol ε, which control base incorporation and proofreading, and by the mismatch repair (MMR) system that undertakes post-replication surveillance1. Germline and somatic mutations in POLD1 and POLE (termed polymerase proofreading deficiency, PPD), or in the MMR genes (MLH1, MSH2, MSH6, PMS2; termed MMR deficiency, MMRD) result in DNA replication repair deficiency. This is a major driver of hypermutation and microsatellite instability (MSI) in several adult and pediatric cancers2,3. Both germline PPD4 and monoallelic germline pathogenic variants in the MMR genes (Lynch syndrome)5 lead to adult-onset gastrointestinal and genitourinary cancers. In contrast, biallelic loss of MMR function in the germline causes constitutional MMRD syndrome, a highly penetrant and aggressive cancer-predisposing condition. Affected individuals typically develop cancers at a young age, most commonly malignant gliomas, and gastrointestinal and hematological malignancies6. These cancers are frequently chemo-resistant and result in poor survival for affected patients. Indeed, these individuals rarely survive beyond early adulthood6. The burden is significant in areas of high consanguinity7, including many developing countries, and in indigenous populations.

Cancers with DNA replication repair deficiency are universally hypermutant due to the continuous acquisition of multiple somatic mutations. The tumor mutation burden (TMB) of these cancers is 100–1,000-fold higher than MMR-intact pediatric cancers8. Furthermore, many of these tumors acquire a secondary somatic mutation in POLD1/ POLEleading to combined MMR + PPD, characterised by ultra-hypermutation (>100 mutations per Mb)2. As a result, these cancers harbour the highest TMB among all human cancers8. Hypermutant cancers such as melanoma9 and lung cancer10, which are driven by ultraviolet light and smoking, respectively, respond to immune checkpoint inhibitors (ICIs) targeting programmed death 1 (PD-1) signalling. However, despite the dramatic anti-tumor effects reported in several hypermutant adult cancers, these responses are sustained in only a subset of patients11. Our understanding regarding the relative importance and variable cut-offs of TMB in determining the nature and duration of response to ICI is still evolving12,13,14. Recent studies have raised questions regarding the roles of TMB and PD-ligand 1 (PD-L1) expression as robust biomarkers of response to ICI12,13,14,15. In contrast, MMRD colorectal carcinomas are reported to be responsive to ICI due to excess MSI16,17,18, suggesting that genomic features such as TMB or MSI may both drive immune responses to ICI19,20,21, but may not be individually sufficient for durable responses across different cancer types.

Most cancers, including hypermutant adult brain tumors are considered ‘immunologically cold’ and are unresponsive to ICI22. Importantly, ICIs did not result in significant responses in multiple large pediatric clinical trials and is considered ineffective in the management of solid tumors in childhood and adolescence15,23,24,25. Additionally, for all solid tumors receiving immunotherapy, the distinction between true tumor progression and an inflammatory pseudo-progression is a major challenge, and a barrier to effective therapy26.

Despite the lack of response to ICIs observed in children in previous studies, we hypothesized that cancers originating from germline DNA replication repair deficiency may benefit from ICIs due to their excess mutational load27. Furthermore, we postulated that cancers driven by MMRD-only, PPD, or combined MMRD + PPD will respectively exert their own unique mutational spectrum, driving local and systemic immune reactions, which would help shed light on the mechanisms of both response and pseudo-progression following ICI.

To address these hypotheses, we conducted a large, observational, registry-based study, leveraging systematically collected data gathered both retrospectively and prospectively through the International Replication Repair Deficiency Consortium (IRRDC)2,6,8,27. This enabled us to evaluate real-world outcomes and predictors of response to anti-PD-1 therapy in children with cancers driven by germline DNA replication repair deficiency. Uniquely, this also provided us the opportunity to investigate the efficacy of ICI in individuals with synchronous malignancies who are otherwise excluded from conventional clinical trials.

Thirty-eight patients who developed 45 cancers were treated with PD-1 inhibitors and followed by the IRRDC study group between May 2015 and March 2019. The PD-1 inhibitor used was either nivolumab (n = 34, 75%) or pembrolizumab (n = 11, 25%) (Methods). All patients had germline RRD, diagnosed as constitutional MMRD (n = 28, 74%), Lynch (n = 8, 21%), or PPD (n = 2, 5%) syndromes (Supplementary Table 1). Median age at treatment was 12.1 years (range: 3.1–28.1) for patients with constitutional MMRD, and 15.7 years (range: 8.5–43.4) for those with Lynch syndrome (p = 0.07). Seven cancer types were included and classified into 3 major groups: central nervous system (CNS) tumors (n = 31, 69%; disseminated: 2, 6%), non-CNS solid tumors (n = 11, 24%; disseminated: 7, 64%), and haematological malignancies (n = 3, 7%) (Fig. 1a). The majority (n = 43, 93%) of cancers were progressive/recurrent after failure of first-line therapy. Three patients with gastrointestinal cancers received ICI directly following surgery; two had synchronous CNS tumors, and one who had metastatic disease. The data cut-off for outcomes was October 2019.

a, Distribution of tumor types across 38 patients who developed 45 tumors. b, Waterfall plot of all radiological responses in non-haematological malignancies. Values show the best fractional change in the 2 dimensions from baseline measurements as per RANO and RECIST criteria (Methods). Arrows point to representative T2-weighted FLAIR and T1-weighted contrast-enhanced MRI sequences in two patients showing flare and partial responses.

Full size image

Responses and/or stable disease were observed in 25/45 (55.5%) tumors, with most of the responses (n = 20; 80%) being sustained at a median follow-up of 1.87 years. Central radiological review (RANO and RECIST criteria; Methods)28,29 revealed complete response in 6 (17%), partial response in 9 (25%), stable disease in 7 (19%), and progressive disease in 14 (39%) (Fig. 1b). Among the 7 patients with synchronous malignancies, responses in both tumors were seen in one patient, and at least in one tumor in four patients (Fig. 2a). The three patients with haematological malignancies (leukemia, T-non-Hodgkin lymphoma) progressed at a median time of 4.5 months after starting ICI therapy. Responses were significantly different among the three types of tumors (P = 0.0041), with non-CNS tumors having the highest response (100%), followed by CNS tumors (64%), with hematological tumors having the lowest response (0%).

a, Swimmer plot by patient and tumor type. b, Kaplan-Meier (KM) estimates of overall survival for all patients (c) KM estimates of overall survival as per tumor type. Median survival for CNS tumors was 21.6 months. Median survival was not reached for non-CNS solid tumors. Two-sided log-rank test p-value is shown. d, KM estimates of progression free and overall survival for CNS tumors continuing ICI therapy. Note: prolonged median survival at 24 months (estimated 3 year OS = 49.1%) despite initial radiological progression at a median of 9.9 months (estimated 3 year PFS = 32%). Two-sided log-rank test p-value is shown.

Full size image

Of note, 12 (27%) tumors exhibited early radiological findings of edema and enhancement, suggestive of peri-tumoral inflammation or tumor progression (Figs. 1b and 2a). This phenomenon occurred at a median of 34 days (range 7–74) from treatment initiation and was termed tumor ‘flare.’30 These patients presented with acute clinical deterioration with headache, bone, or abdominal pain, depending on the location of their tumors. Eight patients (6 with CNS and 2 with non-CNS solid tumors) stopped therapy and died. Importantly, 4 patients (3 with CNS and 1 with non-CNS solid tumor) who continued to receive ICI with adequate supportive care subsequently demonstrated objective responses. Since this phenomenon suggested pseudo-progression, these tumors were studied in more detail.

Estimated 3 year overall survival (OS) was 41.4% (95% CI: 38.5–44.2) (Fig. 2b), with 18 (47%) patients being alive at the time of last follow-up (Fig. 2a). This is noteworthy considering the refractory nature of their cancers. Analysis by cancer type revealed that non-CNS solid tumors had a significantly better survival compared to CNS tumors (Fig. 2c and Supplementary Fig. 1d; P = 0.01). Nevertheless, the OS of 39.3%, (95% CI: 36.3–42.3) and PFS of 26.9% (95% CI; 23.2, 30.6) for patients with recurrent/progressive CNS tumors is a dramatic improvement compared to their historically rapidly fatal outcomes (Fig. 2c,d and Supplementary Fig. 1b)6,7,27. All patients with non-CNS solid tumors continuing ICI had durable responses and are alive at a median follow-up of 2.6 years (range: 0.38–3.5). Remarkably, 13 patients with CNS tumors who experienced radiological progression on initial ICI therapy had prolonged survival (median, 9.6 months; range, 1.5–27 months) (Fig. 2a,d and Supplementary Table 1). Plausible explanation for the late and continued responses to immunotherapy is the obligatory mutation accumulation in these cancers8, which result in novel immunogenic neoantigens and responses. Clinical variables such as age, gender, ethnicity, prior treatment, or choice of ICI agent were not associated with outcome (Supplementary Fig. 2).

Molecular determinants of response to immunotherapy

To better understand the molecular determinants of response to ICI, biopsy specimens and blood samples were collected before and during therapy from the patients for centralized analysis (Methods). Whole exome analysis of tumors (n = 39, Fig. 3) revealed high variability in the number of single nucleotide variants (SNV), including ultra-hypermutation (median, 233.8 mutations per Mb; range, 3.4–912), which was associated with tumor genotype. MMRD-only cancers (n = 16) had significantly fewer SNVs (median, 15.8 mutations/Mb) than MMRD + PPD cancers (n = 23, median 391.4 mutations per Mb; P < 0.0001; Fig. 4b). This was associated with germline status, given that 21 (67.7%) cancers in constitutional MMRD patients harboured MMRD + PPD (median, 398.98 mutations per Mb), while all cancers in individuals with Lynch syndrome lacked somatic PPD (median, 21.76 mutations per Mb; P = 0.03). Both cancers originating from germline PPD, one colorectal carcinoma (P13; ICI.29) and one glioblastoma (P17; ICI.33) had acquired somatic MMRD resulting in ultra-hypermutation (Figs. 2a and 3).

The tumors are arranged in descending order by total single nucleotide variants (SNVs) per megabase using semi-logarithmic scale. Tumor status for mismatch repair deficiency (MMRD) was determined based on MMR gene mutation status by whole exome sequencing (WES), or genetic testing of the patient germline. Tumor polymerase-proofreading deficiency (PPD) status was ascertained by presence of known POLE/POLD1 driver mutations and/or COSMIC signature status for PPD (COSMIC signature 10). The relative weight of each mutational signature contributing to each specific tumor sample analysis as reported using the DeconstructSigs package (Methods) has been color-coded, with the values provided below the onco-plot.

Full size image

Analysis of COSMIC signatures, which reflect the imprints of the underlying mutational processes31, revealed that signature 6 was enriched in tumors from germline Lynch syndrome patients (P = 0.003), whereas signatures 10 and 14 were not detected in these tumors. Furthermore, signature 6 was enriched in MMRD-only cancers (P = 0.04), and signatures 10 and 14 were enriched in MMRD + PPD cancers (P < 0.002). Last, signatures 10 and 14 were individually associated with response to ICI (P = 0.03). Importantly, signature 11, which is commonly detected in treatment-related hypermutant adult gliomas that do not respond to ICI32, was detected only in a single patient (P27) with synchronous leukemia and glioblastoma, who had been previously treated with alkylating agents for medulloblastoma. These findings highlight the unique diagnostic and prognostic roles of mutational signatures in replication repair-deficient cancers31.

As questions exist regarding the contribution of tumor-intrinsic characteristics such as mutation load in terms of SNVs, indels, and the microsatellite specific indels to ICI response, we sought to determine whether independent roles exist for each of these genomic features and their underlying driver mechanisms. High SNV/Mb (≥median: 275.38/Mb) was significantly associated with both response and survival, demonstrating that extreme mutation burdens were relevant even for a cohort of uniformly hypermutant cancers (P = 0.005, Fig. 4a and Extended Data Fig. 1a–d). Both response and survival were also associated with a higher tumor neoantigen load (Extended Data Fig. 1e). Enrichment of clonal mutations33 predicted response (Extended Data Fig. 2). Remarkably, response and survival were predicted by replication repair deficiency status. Patients whose cancers were MMRD + PPD had higher mutation burden (P < 0.0001), and these tumors were enriched among the responders (n = 17/21) as compared to non-responders (n = 3/10) (P < 0.001) (Fig. 4b).

a, Response and overall survival (OS) by single nucleotide variants (SNVs) per Mb. For survival analysis, median SNV burden was used. b, SNVs as a function of MMRD (blue) and MMRD + PPD (orange) status (left), and response association with both SNV and replication-repair deficiency status. c, Response and overall survival (OS) by microsatellite indels (MS-indels). For survival analysis, median MS-indel values were used. d, Response and overall survival by total MS-indel count for MMRD + PPD and MMRD-only cancers separately. e, Kaplan-Meier (KM) estimates using combined SNVs/Mb and MS-indel in all replication repair-deficiency cancers. For all box-plots for responders and non-responders, data are represented as median + /- interquartile range. For statistical significance in comparing responders and non-responders, the Wilcoxon-Mann-Whitney test was used. Survival analysis was performed using the Kaplan-Meier method, and the log-rank test was used to compare groups. All p values are 2-sided. (MMRD: mismatch repair deficiency; PPD: polymerase proofreading deficiency; MS-indel: microsatellite insertion/ deletion).

Full size image

Recent work has shown that the degree of genome-wide MS-indels correlate with response to ICI in MMRD tumors3,34,35. Indeed, across our entire cohort, total MS-indels, calculated by MS-mutect (Methods)36, were predictive of tumor response (P = 0.02) and patient survival (p = 0.006; Fig. 4c). This was as a result of frameshift indels in coding microsatellites and neoantigens generated by these MS-indels, which strongly predicted outcome (Extended Data Fig. 3).

Given that no significant correlation between MS-indels and SNVs were observed (Supplementary Fig. 3), we hypothesized that while both SNVs and MS-indels are independently immunogenic across all replication repair-deficient cancers, MS-indels may be particularly important for determining outcomes for MMRD-only cancers which harbour relatively lower SNVs. To test this, we analyzed MMRD-only and MMRD + PPD tumors separately (Fig. 4d and Supplementary Fig. 4). Total MS-indels were higher in responders than in non-responders and were significantly associated with survival in MMRD-only cancers (P = 0.003) and not in MMRD + PPD cancers (Fig. 4d). In contrast, high SNVs and total indels failed to predict response in MMRD-only cancers (P > 0.1, Extended Data Fig. 4cand Supplementary Fig. 4b). Combining the prediction models for both these types of replication repair deficiency cancers (MMRD-only and MMRD + PPD) revealed that high SNVs and total MS-indels together strongly predicted improved survival (Fig. 4e and Supplementary Fig. 3d–g; P = 0.0024).

To determine whether mutations and MS-indels drive responses within a more homogeneous cancer type, we interrogated both these genomic markers in CNS tumors. Both components independently contributed to the response and survival (Extended Data Fig. 5) and no differences in these genomic features were observed between CNS and non-CNS solid tumors (Supplementary Fig. 5a–d). Importantly, clonal mutations33 were particularly a strong predictor of response and survival for CNS tumors (Extended Data Fig. 2).

Collectively, these data suggest dual roles for SNVs and MS-indels in determining immunotherapy responses in replication repair-deficient cancers (Supplementary Fig. 3d,g), including CNS tumors (Supplementary Fig. 6).

Immune microenvironment and response to therapy

Next, we examined whether the tumor subgroups affect the tumor micro-environment and response to therapy. We tested multiple immune markers using immunohistochemistry for immune cell infiltration (CD3, CD4, CD8 and CD68) and checkpoint ligand (PD-L1) expression. All immune markers were scored independently by two pathologists with robust concordance (Methods; Supplementary Fig. 7).

PD-L1 expression was associated with both response and improved survival (P = 0.04, Fig. 5a). Overall, increased lymphocytic infiltration within the tumor microenvironment was associated with response (Extended Data Fig. 6). Specifically, high CD8-T cell infiltration predicted both response and improved OS (P = 0.0002, Fig. 5b). All non-CNS solid tumors including those with MMRD-only, harboured high MSI, exhibited high CD8+ T cell infiltration (Supplementary Fig. 5), and responded to ICI. This corroborates previous reports in which MMR-deficient gastrointestinal tumors had high CD8+ T cell infiltration37. High T cell infiltration was also observed in CNS tumors, which are traditionally considered an ‘immune-privileged’ site22 (Fig. 5b). CNS tumors with a high mutation burden and MMRD + PPD not only had increased CD8+ T cell infiltration, but also significantly higher expression of PD-L1 (Fig. 5d,e). In contrast, all tumors in our study had low to moderate CD68 expression (none > 30%). Furthermore, there was no association between CD68 and response or survival (Fig. 5c). This suggests that in the setting of ultra-hypermutation driven by combined MMRD + PPD and high genomic MS-indels, the increased activation of the immune microenvironment, associated with robust CD8+ T cell responses, can explain the remarkable responses seen even in the CNS tumors. Indeed, in these CNS tumors, high CD8+ T cell infiltration was associated with improved OS (P = 0.039; Extended Data Fig. 7b).

a, PD-L1 expression, response and survival in all replication repair-deficient cancers. Cut-off is ≥ 1% of cells (median; Methods). b, CD8 expression, response and survival for replication repair-deficient cancers. Cut-off is ≥ 3% of cells (median; Methods). c, CD68 expression, response and survival. Cut-off is ≥ 12.5% of cells (median; Methods). For (a), (b) and (c), the histology depicts glioblastoma at 20X magnification. d,e, Association of immune markers with SNV and replication repair-deficient status. f, Combined immune (PD-L1 and CD8 expression) and genomic (TMB and MS-indels) and overall survival in replication repair-deficient cancers. For all box-plots for responders and non-responders, data are represented as median + /- interquartile range. For statistical significance in comparing responders and non-responders, the Wilcoxon-Mann-Whitney test was used. Survival analysis was performed using the Kaplan-Meier method, and the log-rank test was used to compare groups. All p values are 2-sided.

Full size image

Overall, tumors exhibiting high expression of immune markers and a favorable genomic profile (high SNVs or MS-indels) had a 3 year OS of 87.8% (95% CI: 84–91.5) as compared to 33.2% (95% CI: 27.2–39.2) for tumors lacking these biomarkers (P = 0.005, Fig. 5f).

Tumor flare is an immune reaction to therapy

To determine etiology of tumor flare, we first analyzed the genomic and immune markers of these tumors. Cancers developing flare had pre-treatment genomic and immune characteristics similar to the responders without flare (Extended Data Fig. 8a–d). We then compared the pre- and on-therapy tumors in two patients who had further surgical debulking at flare. Transcriptomic analysis and immune inference using single state deconvolution (Methods) revealed an increase in the overall immune cell expression at flare. Notably, transcriptome signaling revealed that activated CD8+ T cells were significantly increased in both samples following ICI (Fig. 6a,b and Supplementary Fig. 8). Using T cell receptor clonotype analysis (Fig. 6c,d and Supplementary Fig. 9), we observed a dramatic increase of T cell repertoire at flare as compared to their baseline. One sample demonstrated an increase in both clonality and diversity of the T cell population, with some original clones expanding during flare (Fig. 6c), whereas for the other sample, there was reduction in diversity but significantly heightened clonality in the T cell population, involving unique clonotypes harbouring complementary-determining regions (CDR3) not previously reported in public databases (Fig. 6d and Supplementary Fig. 9c). Additionally, increased CD8+ T cell infiltration and PD-L1 expression were observed in both tumors during flare when compared to their pre-treatment samples (Fig. 6e,f and Extended Data Fig. 9). These observations suggest a pre-existing (specific) immune response and further ICI-driven (non-specific and specific) tumor-directed immune expansion at flare.

a-d, Analysis of 2 patients who had tumor-debulking prior to therapy and at the time of flare. (a, b) Total immune cell content in pre-therapy and at flare. c,d, The corresponding CapTCR-sequencing and T cell receptor clonotype (TCR) analysis in these samples. Each box represents a specific TCR clonotype. Specific clonotypes which were shared between the baseline samples and at flare were tracked using the same colour between the baseline and flare samples. e,f, Immunohistochemistry for PD-L1 expression, and CD8-T cell infiltration in the pre-therapy sample and at flare, as shown in the representative 20X images from the tumor sample in patient-1 (P33). All immunohistochemistry was analysed by two independent pathologists. g, Representative flow cytometry plot showing activation of CD + T cell (TIGIT and 4-1BB) from the blood sample of a patient before treatment initiation and at flare. h, 4-1BB + CD8 + T cells in blood from responders without flare, non-responders and flare. For the box-plots, data are represented as median + /- interquartile range. For statistical significance in pairwise comparison of responders, non-responders and flare, the Wilcoxon-Mann-Whitney test was used. All p values are 2-sided.

Full size image

We next investigated whether this immune activation could be observed systemically (Fig. 6h, Extended Data Fig. 10 and Supplementary Fig. 10). We performed serial flow cytometry analysis of blood samples from multiple patients, prior to treatment initiation and within the first 90 days of initiation of ICI (Methods). Flare was associated with an expansion of peripheral CD8+ T cells expressing Ki67 (a marker of proliferation) and 4-1BB (a member of the TNF receptor family) (Extended Data Fig. 10a–d and Fig. 6g). Detection of Ki67+ CD8+ T cells has been previously reported to be associated with superior response to PD-1 blockade38. 4-1BB is well-known as a marker of T cell activation and has co-stimulatory activity for activated T cells39. Flare was associated with higher proportion of 4-1BB+ CD8+ T cells in the peripheral blood as compared to non-responders, as well as responders without flare (Fig. 6h). Studies have previously shown that 4-1BB+ CD8+ T cells correlate with response to PD-1 blockade40. This uniform increases in 4-1BB and Ki67+CD8+ T cells, was accompanied by an insignificant change in the 4-1BB+ CD4+ T cell population (Extended Data Fig. 10e,f) and no difference between responder, none-responders and flare for the latter (Supplementary Fig. 10a), supporting our hypothesis that CD8+ T cells expressing 4-1BB reflect the expansion of a tumor-specific response. Taken together, our data suggests that flare following PD-1 blockade is with a result of inflammatory response and proliferation of tumor-antigen reactive T cells.

Our study reveals dramatic responses to PD-1 blockade and an associated improved survival for relapsed/refractory hypermutant cancers in children and young adults with germline DNA replication repair deficiency. Several insights can be derived from the sustained responses in different tumor types, and the contributions of SNVs, MS-indels and the microenvironment to both response and flare.

The unique responses to ICIs in children with germline replication repair deficiency are different from previous observations by several groups. First, our data starkly contrast with the lack of ICI efficacy in childhood cancers in general15,23,24,25, and specifically in progressive paediatric brain tumors41. The lack of response in paediatric cancers, which is independent of PD-L1 expression15 or systemic immune activation15,24, has been attributed to the low tumor mutation burden2,8,27, low expression of major histocompatibility complex42, and the predominance of macrophages in the tumor micro-environment43. Some of these causes may need to be re-examined in view of our data.

Second, more than half of replication repair-deficient pediatric CNS tumors had objective responses, resulting in a median survival of 2 years when ICI was continued. This is remarkable, as historically, rapid progression with a median post-relapse survival of merely 2.6 months has been reported in children with replication repair-deficient high-grade glioma7,27. Furthermore, these responses contrast with the poor response to ICIs in adult hypermutant gliomas32,44. Indeed, comparative mutational analysis in these two cohorts revealed significantly higher mutation burden in paediatric MMRD + PPD gliomas, and higher MS-indels in all germline replication repair-deficient gliomas, as compared to adult hypermutant gliomas (Supplementary Fig. 6). Moreover, ICI-responders among germline replication repair-deficient gliomas exhibited predominance of clonal mutations (Extended Data Fig. 2), which is different from treatment-related secondary MMR deficient adult gliomas, where late acquisition of sub-clonal mutations during tumorigenesis results in suboptimal responses to ICI32. Additional causes may include an immunosuppressed tumor microenvironment dominated by myeloid infiltrates45 which are frequent in adult gliomas as compared to replication repair-deficient gliomas, where myeloid infiltrates are less predominant (Fig. 5).

Third, all patients who continued ICI therapy for non-CNS solid tumors including disseminated cancers, responded, and are alive at a median follow-up of 2.6 years, with an estimated survival rate of 80%. Although still preliminary, these data are encouraging, as the survival in most recurrent childhood cancers is poor, especially for metastatic disease46,47. Lack of response to ICI in previous clinical trials in pediatric recurrent/refractory solid tumors probably reflects the lack of inclusion of DNA replication repair-deficient tumors in these studies15,23,24,25. As we observed responses in metastatic colorectal and genitourinary cancers, our results compare favourably with the otherwise grim survival for such tumors in both children48,49 and adults50,51 when using conventional therapies. Interestingly, our data compare favourably even with studies using ICI in MMRD cancers in older adults (median age: 60 years, versus 12.3 years in our study), in whom late failures were noted, resulting in 50–55% survival at 2 years17,18,52.

Collectively, the dramatic responses and favourable outcome observed in childhood replication repair-deficient cancers can be explained by several key biological features. First, the earlier onset8, combined with significantly higher mutations2, MS-indels3, and neoantigen burden drive CD8+ T cell activation, which is especially robust in children and adolescents when compared to older patients53. Second, the additional loss of the polymerase-proofreading mechanism confers genomic mutational signatures (both MS-signatures3 and COSMIC signatures31 10 and 14) which may play unique immunogenic role in determining response and survival. Third, given that MMRD + PPD cancers cannot repair errors during DNA replication, the exceptionally high rate of obligatory and continuous accumulation of mutations in these cancers8 probably confers ongoing immunogenicity, contributing to immune-surveillance leading to both the durable and the delayed responses observed in our cohort. This probably contributed to the difference between progression-free (post-ICI) and overall survival for patients who continued anti-PD1 treatment after a second progression while on ICI therapy (Fig. 2d and Supplementary Fig. 1b,c). The use of combinatorial therapies targeting additional immune-checkpoints and pathways can therefore be of further benefit for patients whose tumors escape immune surveillance following single-agent anti-PD1 blockade54.

An additional observation worth discussion is the response in a tissue-agnostic manner in a patient with synchronous tumors exhibiting favourable genomic and immune biomarkers for response (P01). As cancer immune surveillance is different than the irreversible resistance which occurs upon progression after chemo-radiation approaches, these data support the exploration of neoadjuvant, maintenance, and combinatorial uses of ICI in these patients, to limit toxicities and improve effectiveness of first-line strategies55.

There is increasing evidence that responses to ICI cannot be predicted by single biomarker56. Our data confirm that this is true in replication repair-deficient cancers which are driven by dysfunction in both SNV and MS-indel repair. Although initially TMB was thought to be the sole contributor to immune response in hypermutant cancers, indels and MS-indels have been suggested to be the important drivers of response to ICI in MMR-deficient cancers3,35,57. We add a new dimension to this concept by revealing that in cancers driven by MMRD-only, TMB is relatively lower, and MS-indels drive response, while in MMRD + PPD cancers, the role of MS-indels is attenuated, and TMB is the major driver of response. The dual roles of both mutational mechanisms also affect the microenvironment, with upregulation of PD-L1 and infiltration of CD8+ T cells. Importantly, the combination of both genomic mechanisms and immune markers are powerful predictors of survival in replication repair deficient cancers and should be incorporated as combined biomarkers in future clinical trials. Additionally, the contribution of an immune microenvironment to the ICI response may be relevant in the context of similar findings reported in other subsets of paediatric cancers not known to be driven by replication repair deficiency58.

Paradoxically, this hyperactivation of the immune microenvironment can also be detrimental, since tumor flare, which indicates both specific (pre-existing) and non-specific (new) expansion of TCR clones, can be misinterpreted as tumor progression and lead to premature treatment abandonment. In this study, we observed flare to be quite common in germline replication repair-deficient cancers, which are highly immunogenic due to their extreme SNV and MS-indel burden. These flare responses are distinct from hypermutant adult glioblastomas originating as a result of previous chemo-radiotherapy, where true progression is reported more frequently than pseudo-progression32. The higher prevalence of flare (26%) in our study, and the prolonged survival in patients who continued therapy, further support this notion. Although vaccine-based approaches lead to a significant local immune response59, these may be driven by different mechanisms. Indeed the timing of flare was earlier (median: 34 days) as compared to that reported in studies in adult patients with glioblastoma treated using immune-directed approaches (median: 10 weeks; commonly <3 to up to 6 months)59. Although the observations of intra-tumoral inflammatory response at flare are limited and need to be ratified in larger cohorts, our data, which demonstrate similar genomic and immune profile among tumors which responded and those that developed flare, can explain the late responses observed in patients where therapy was continued. This highlights the importance of developing novel functional imaging techniques60 and minimally invasive biomarkers61 to better predict and diagnose tumor flare, and innovative strategies to modulate this aggressive immune response.

Despite the limitations of a registry study, this is the first description of a large cohort of children and young adults with previously fatal, recurrent/progressive germline DNA replication repair-deficient cancers, demonstrating impressive responses to PD-1 blockade. Importantly, this registry study was able to collect germline and tumor tissue from all patients, including longitudinal blood samples and relapsed tumor tissues whenever repeat surgeries were performed, which has historically not been performed in prospective paediatric studies including recent immunotherapy trials. The robustness and consistency of our results for response, survival and predictive biomarkers despite the heterogeneity of our cohort support a strong biologic rationale for use of ICI in this rare patient population. Although longer follow-up is required to determine whether immunotherapy can be a curative strategy for RRD cancers, the sustained responses and lack of late relapses in a significant number of patients in this cohort are encouraging. This study also sheds light on the complex interplay between the tumor genomic status, microenvironment, and the systemic immune response, especially in the context of extreme mutation and MS-indel burdens. Future trials should prospectively analyse the roles of germline versus somatic deficiency, and the components of the replication repair machinery, to identify patients who are likely to derive maximal benefit from anti-PD-1 immunotherapy. Last, our study highlights the impact of studying a genetic cancer syndrome to understand general cancer processes and deriving direct therapeutic implications for patients.