Browse the corpus

Nonoperative Management of Mismatch Repair-Deficient Tumors. BACKGROUND: Among patients with mismatch repair-deficient (dMMR), locally advanced rectal cancer, neoadjuvant checkpoint blockade eliminated the need for surgery in a high proportion of patients. Whether this approach can be extended to all early-stage dMMR solid tumors, regardless of tumor site, is unknown. METHODS: We conducted a phase 2 study in which patients with stage I, II, or III dMMR solid tumors that were amenable to curative-intent surgery were treated with neoadjuvant dostarlimab, a programmed cell death 1 (PD-1) blocking agent, for 6 months. The response to treatment was assessed in two cohorts: patients in cohort 1 had dMMR, locally advanced rectal cancer, and patients in cohort 2 had dMMR nonrectal solid tumors. Patients with a clinical complete response could elect to proceed with nonoperative management; those with residual disease were to undergo resection. In this analysis, the primary end point, assessed in cohort 1, was a sustained clinical complete response at 12 months. Recurrence-free survival and safety were evaluated. RESULTS: A total of 117 patients were included in the analysis. In cohort 1, all 49 patients who completed treatment had a clinical complete response and elected to proceed with nonoperative management. A total of 37 patients had a sustained clinical complete response at 12 months, a finding that met the criterion for efficacy. In cohort 2, a total of 35 of 54 patients who completed treatment had a clinical complete response, and 33 elected to proceed with nonoperative management. Among the 103 patients who completed treatment across both cohorts, 84 had a clinical complete response, and 82 did not undergo surgery. Among the 117 total patients, recurrence-free survival at 2 years was 92% (95% confidence interval, 86 to 99); the median follow-up for recurrence was 20.0 months (range, 0 to 60.8). The majority of patients (95%) had reversible, grade 1 or 2 adverse events (60%) or had no adverse events (35%). The option for curative resection was not compromised during or after treatment in any of the patients. CONCLUSIONS: Among patients with early-stage dMMR solid tumors that were amenable to curative-intent surgery, neoadjuvant PD-1 blockade led to organ preservation in a high proportion of patients. (Funded by Swim Across America and others; ClinicalTrials.gov number, NCT04165772.).

Complete surgical resection is the predominant curative option for early-stage solid tumors.1 Non-operative management has been successfully implemented for several cancer types, including anal, bladder, rectal, and head and neck cancers, when radiation is combined with chemotherapy.2–7 Additionally, for small lung tumors, stereotactic radiation therapy can achieve outcomes comparable to surgery.8 MMRd tumors are highly sensitive to immune checkpoint blockade, and in the metastatic setting the substantial clinical benefit across tumor types resulted in the first tumor agnostic approval by the FDA.9–12 In the neoadjuvant setting, elimination of the primary tumor has been reported in colorectal cancers and in rectal cancer specifically, early studies suggest that surgery can be averted when a clinical complete response (cCR) is achieved.13–16 The complete elimination of MMRd primary tumors with PD-1 blockade alone in rectal cancer raised the question of whether this approach could be extended beyond rectal cancer to a broader, tumor-agnostic setting—mirroring the benefit of checkpoint blockade across MMRd cancers in the metastatic setting. The potential impact is substantial since 2–10% of all early-stage solid tumors are mismatch repair deficient.9 Despite data supporting efficacy in metastatic disease, concern exists for the risk of potentially missing the ‘window of opportunity’ if tumors were to progress and become unresectable during the course of treatment with an immune checkpoint inhibitor.

0% of all early-stage solid tumors are mismatch repair deficient.9 Despite data supporting efficacy in metastatic disease, concern exists for the risk of potentially missing the ‘window of opportunity’ if tumors were to progress and become unresectable during the course of treatment with an immune checkpoint inhibitor. To test the premise of non-operative management for MMRd early-stage tumors, we expanded on our experience in rectal cancer and enrolled patients with early stage MMRd tumors eligible for curative intent therapy from any site. Patients were treated with 6-months of PD-1 blockade with dostarlimab and monitored closely for response, progression, or recurrence.

Patients from Memorial Sloan Kettering, Hartford Heath Care and Miami Cancer institute were screened and referred to the study if found to be MMRd. Patients with newly diagnosed stage I, II and III solid tumors amenable to curative intent therapy with loss of expression of MLH1, MSH2, MSH6, or PMS2 by immunohistochemistry (IHC) were eligible. Additional eligibility criteria are included in the accompanying protocol, available at nejm.org. This study was designed to evaluate neoadjuvant dostarlimab (500mg intravenously every 3 weeks for 9 cycles) in two cohorts, (i) locally advanced MMRd rectal cancers and (ii) non-rectal MMRd solid tumors. Following assessment of response, patients with residual tumor would be treated with standard of care neoadjuvant therapy, and applicable surgery. Patients achieving a clinical complete response following neoadjuvant dostarlimab were offered non-operative management. The institutional review board at Memorial Sloan Kettering approved the protocol and subsequent amendments. All patients provided informed consent. The manuscript was written solely by the authors, who collected the data and attest to its integrity, accuracy, and adherence to the protocol. Patients underwent genomic sequencing of tumor and matched normal blood samples for somatic and germline genomic alterations17. All samples were collected under an IRB approved protocol (NCT01775072) and written consent. Relatedness between baseline and post-treatment tumors was evaluated using a mutational profile similarity (MPS) score18 (Supplemental Appendix).

and matched normal blood samples for somatic and germline genomic alterations17. All samples were collected under an IRB approved protocol (NCT01775072) and written consent. Relatedness between baseline and post-treatment tumors was evaluated using a mutational profile similarity (MPS) score18 (Supplemental Appendix). A tumor-informed approach was used to measure circulating tumor DNA (ctDNA). Somatic mutations were identified with exome sequencing of each patient’s tumor tissue and matched normal DNA to assess up to 50 tumor-specific mutations in cell-free DNA isolated from plasma (Supplemental Appendix). The first co-primary endpoint of overall response rate greater that 25% was met when 12 consecutive patients achieved complete clinical responses15.

A tumor-informed approach was used to measure circulating tumor DNA (ctDNA). Somatic mutations were identified with exome sequencing of each patient’s tumor tissue and matched normal DNA to assess up to 50 tumor-specific mutations in cell-free DNA isolated from plasma (Supplemental Appendix). The first co-primary endpoint of overall response rate greater that 25% was met when 12 consecutive patients achieved complete clinical responses15. In this manuscript, we present the second co-primary endpoint based on data from December 2019 through April 2025, which required a cCR of at least 12 months after completion of dostarlimab with or without chemoradiation in patients who did not undergo surgery or who proceeded to surgery and achieved a pCR. We implemented a Simon’s 2-stage minimax design to show an improvement in the sustained clinical complete response rate from the unacceptable rate of 27.5% to an acceptable rate of 50% with type I error of 0.05 and 80% power. In the first stage, we planned to enroll 15 patients. If 4 or fewer patients achieved a sustained cCR of 12 months or greater after completion of dostarlimab, the study will stop for futility. If 13 or more patients with a sustained cCR of at least 12 months after completion of dostarlimab, the study will stop for efficacy. As previously established19, a cCR was defined as no residual disease on a digital and endoscopic rectal exam and absence of residual tumor on MRI (Supplemental Appendix, Protocol section 14.0).

r more patients with a sustained cCR of at least 12 months after completion of dostarlimab, the study will stop for efficacy. As previously established19, a cCR was defined as no residual disease on a digital and endoscopic rectal exam and absence of residual tumor on MRI (Supplemental Appendix, Protocol section 14.0). The primary objective for this cohort was exploratory and estimated the complete response rate. (Supplemental Appendix, Protocol section 14.0). All statistical analyses were performed using R, version 4.3.2. (Supplemental Appendix)

Patients from Memorial Sloan Kettering, Hartford Heath Care and Miami Cancer institute were screened and referred to the study if found to be MMRd. Patients with newly diagnosed stage I, II and III solid tumors amenable to curative intent therapy with loss of expression of MLH1, MSH2, MSH6, or PMS2 by immunohistochemistry (IHC) were eligible. Additional eligibility criteria are included in the accompanying protocol, available at nejm.org.

This study was designed to evaluate neoadjuvant dostarlimab (500mg intravenously every 3 weeks for 9 cycles) in two cohorts, (i) locally advanced MMRd rectal cancers and (ii) non-rectal MMRd solid tumors. Following assessment of response, patients with residual tumor would be treated with standard of care neoadjuvant therapy, and applicable surgery. Patients achieving a clinical complete response following neoadjuvant dostarlimab were offered non-operative management.

The institutional review board at Memorial Sloan Kettering approved the protocol and subsequent amendments. All patients provided informed consent. The manuscript was written solely by the authors, who collected the data and attest to its integrity, accuracy, and adherence to the protocol.

Patients underwent genomic sequencing of tumor and matched normal blood samples for somatic and germline genomic alterations17. All samples were collected under an IRB approved protocol (NCT01775072) and written consent. Relatedness between baseline and post-treatment tumors was evaluated using a mutational profile similarity (MPS) score18 (Supplemental Appendix). A tumor-informed approach was used to measure circulating tumor DNA (ctDNA). Somatic mutations were identified with exome sequencing of each patient’s tumor tissue and matched normal DNA to assess up to 50 tumor-specific mutations in cell-free DNA isolated from plasma (Supplemental Appendix).

The first co-primary endpoint of overall response rate greater that 25% was met when 12 consecutive patients achieved complete clinical responses15. In this manuscript, we present the second co-primary endpoint based on data from December 2019 through April 2025, which required a cCR of at least 12 months after completion of dostarlimab with or without chemoradiation in patients who did not undergo surgery or who proceeded to surgery and achieved a pCR. We implemented a Simon’s 2-stage minimax design to show an improvement in the sustained clinical complete response rate from the unacceptable rate of 27.5% to an acceptable rate of 50% with type I error of 0.05 and 80% power. In the first stage, we planned to enroll 15 patients. If 4 or fewer patients achieved a sustained cCR of 12 months or greater after completion of dostarlimab, the study will stop for futility. If 13 or more patients with a sustained cCR of at least 12 months after completion of dostarlimab, the study will stop for efficacy. As previously established19, a cCR was defined as no residual disease on a digital and endoscopic rectal exam and absence of residual tumor on MRI (Supplemental Appendix, Protocol section 14.0). The primary objective for this cohort was exploratory and estimated the complete response rate. (Supplemental Appendix, Protocol section 14.0). All statistical analyses were performed using R, version 4.3.2. (Supplemental Appendix)

The first co-primary endpoint of overall response rate greater that 25% was met when 12 consecutive patients achieved complete clinical responses15. In this manuscript, we present the second co-primary endpoint based on data from December 2019 through April 2025, which required a cCR of at least 12 months after completion of dostarlimab with or without chemoradiation in patients who did not undergo surgery or who proceeded to surgery and achieved a pCR. We implemented a Simon’s 2-stage minimax design to show an improvement in the sustained clinical complete response rate from the unacceptable rate of 27.5% to an acceptable rate of 50% with type I error of 0.05 and 80% power. In the first stage, we planned to enroll 15 patients. If 4 or fewer patients achieved a sustained cCR of 12 months or greater after completion of dostarlimab, the study will stop for futility. If 13 or more patients with a sustained cCR of at least 12 months after completion of dostarlimab, the study will stop for efficacy. As previously established19, a cCR was defined as no residual disease on a digital and endoscopic rectal exam and absence of residual tumor on MRI (Supplemental Appendix, Protocol section 14.0).

The primary objective for this cohort was exploratory and estimated the complete response rate. (Supplemental Appendix, Protocol section 14.0). All statistical analyses were performed using R, version 4.3.2. (Supplemental Appendix)

124 patients were enrolled. Seven patients were excluded from our analysis for the following reasons: three participants were found to have metastases or second malignancies before treatment and withdrew from the study; two participants opted to discontinue therapy after one dose of dostarlimab; one participant was lost to follow up; and one participant stopped treatment after 4 doses. (Figure 1A, Supplemental Appendix). Cohort 1 was comprised of only MMRd rectal cancers. Cohort 2 included MMRd non-rectal solid tumors, including esophagogastric, colon, hepatobiliary, genitourinary, and gynecologic tumors (Table 1; Table S1, Figure S1). On the date of data lock, 103 patients had completed therapy, 49 in cohort 1 and 54 in cohort 2, and 14 remain on therapy. (Figure 1A, Table 1, Table S1) Across both cohorts most patients had radiographically evident lymph nodes (64%). All tumors had loss of expression of mismatch repair proteins with loss of both MLH1 and PMS2 being the most common. Germline pathogenic variants in a mismatch repair gene were noted in 44% of cases. (Table 1, Table S1) No patients progressed or became unresectable while on therapy. In Cohort 1 (rectal cancer), 49 patients completed treatment. All 49 achieved a cCR, and all 49 opted for non-operative management. (Figure 1B) In Cohort 2 (non-rectal solid tumors), 54 patients completed treatment. Thirty-five (65%) of these 54 achieved a clinical complete response, and 33 (61%) also elected nonoperative management. (Figure 1B)

In Cohort 1 (rectal cancer), 49 patients completed treatment. All 49 achieved a cCR, and all 49 opted for non-operative management. (Figure 1B) In Cohort 2 (non-rectal solid tumors), 54 patients completed treatment. Thirty-five (65%) of these 54 achieved a clinical complete response, and 33 (61%) also elected nonoperative management. (Figure 1B) Two patients (one with gastric cancer and one with urothelial cancer) who achieved a cCR, elected to undergo surgical resection. In both cases, no cancer was found in the resection specimen. (Figure 1A, Figure 1B, Table S1-2) Combining cohorts, among 103 patients who completed treatment, 84 (82%, 95% CI: 72%−88%) achieved a clinical complete response and 82 (80%, 70%−87%) also proceeded with non-operative management. (Figure 1B) In Cohort 1 (rectal cancer) no incomplete responses were observed. (Figure 1B, Figure 2A) In Cohort 2 (non-rectal solid tumors), 19 of 54 patients (35%) had incomplete clinical response. Sixteen of these patients underwent surgical resection of their primary tumors and 3 patients deferred surgery. At baseline, all tumors were found to be MMRd and MSI-high, except for one tumor that had subclonal MLH1/PMS2 loss and was found to be microsatellite stable (MSS). (Table S2) In the 16 patients who underwent surgery, evidence of tumor regression was noted in 14 (87%) of the resected tumors; 1 did not have evaluable tissue; 1 had achieved a pathologic complete response and three had complete pathologic response in the primary tumor but residual tumor present in the lymph nodes. (Supp Table 2).

In the 16 patients who underwent surgery, evidence of tumor regression was noted in 14 (87%) of the resected tumors; 1 did not have evaluable tissue; 1 had achieved a pathologic complete response and three had complete pathologic response in the primary tumor but residual tumor present in the lymph nodes. (Supp Table 2). We were able to compare the sequences of the pre and the remnant post-treatment tumors in the incomplete responders with genomic data and found a high likelihood of genomic relatedness (99.6%: 95% CI 97.43–99.97, n=17)18 in all but two patients (51 and 90). In these two cases the remnant tumor was likely a new primary as there was a low likelihood of genomic relatedness (mutational profile similarity score < 0.4%18). Furthermore, the remnant tumors in these two cases were also not MMRd or MSI-high. (Table S2; Figure S2). Other than the two unrelated tumors, no meaningful differences were noted between pre and remnant post-treatment tumors in median TMB (43 vs. 34 mutations per MB, Wilcoxon signed rank p=0.43, n=14), mismatch repair protein loss or median microsatellite instability (MSI) sensor score (15 versus 16, Wilcoxon signed rank p=0.47, n=12) (Table S2). Across both cohorts, (n=117) and the two year recurrence free survival was 92% (86–99), (Figure 2A and Figure S3). In Cohort 1, the two year recurrence-free survival was 96% (90–100) (n=50) and median follow-up for recurrence was 30.2 months (Range 5.8–60.8 months). To date, 37 patients have sustained a cCR for at least 12-months from completion of treatment, which satisfies the second co-primary endpoint.

Across both cohorts, (n=117) and the two year recurrence free survival was 92% (86–99), (Figure 2A and Figure S3). In Cohort 1, the two year recurrence-free survival was 96% (90–100) (n=50) and median follow-up for recurrence was 30.2 months (Range 5.8–60.8 months). To date, 37 patients have sustained a cCR for at least 12-months from completion of treatment, which satisfies the second co-primary endpoint. In Cohort 2, the two year recurrence-free survival was 85% (70–100) (n=67), and the median follow-up for recurrence was 14.9 months (Range 0–32.7 months). Only 5 patients across both cohorts developed disease recurrence. One patient had tumor regrowth at the rectal primary site and 4 recured solely in lymph nodes. (Table S1) None of these cases required resection of the primary tumor. One patient had resection of a lymph node and currently has no evidence of disease. Four patients restarted PD-1 blockade and 3 are currently disease-free. Original and recurrent tumors were consistently MMRd with high genomic relatedness and no meaningful differences in TMB or MSI between evaluable pre and recurrent tumors. (Table S1; Figure S4). Among this group the median disease-free interval from second recurrence was 6.6 months (3.1 – 14.9 months, n=4). (Supplemental appendix). No deaths have occurred in either cohort. Across both cohorts, the median follow-up for survival from the start of treatment was 21.5 months (Range 0–60.8 months, 95%CI 19.4–23.6, n=117).

Only 5 patients across both cohorts developed disease recurrence. One patient had tumor regrowth at the rectal primary site and 4 recured solely in lymph nodes. (Table S1) None of these cases required resection of the primary tumor. One patient had resection of a lymph node and currently has no evidence of disease. Four patients restarted PD-1 blockade and 3 are currently disease-free. Original and recurrent tumors were consistently MMRd with high genomic relatedness and no meaningful differences in TMB or MSI between evaluable pre and recurrent tumors. (Table S1; Figure S4). Among this group the median disease-free interval from second recurrence was 6.6 months (3.1 – 14.9 months, n=4). (Supplemental appendix). No deaths have occurred in either cohort. Across both cohorts, the median follow-up for survival from the start of treatment was 21.5 months (Range 0–60.8 months, 95%CI 19.4–23.6, n=117). In Cohort 1, all but one patient completed all 9 cycles (6 months). One patient chose to stop therapy after 8 cycles because of an unrelated asthma exacerbation. (Table S1) In Cohort 2 (non-rectal solid tumors), 49 of the 54 patients completed all planned cycles. Three stopped before achieving a complete response due to concerns for lack of response. Two stopped after achieving a cCR due to treatment-related adverse events, per their preference. (Table S1)

In Cohort 1, all but one patient completed all 9 cycles (6 months). One patient chose to stop therapy after 8 cycles because of an unrelated asthma exacerbation. (Table S1) In Cohort 2 (non-rectal solid tumors), 49 of the 54 patients completed all planned cycles. Three stopped before achieving a complete response due to concerns for lack of response. Two stopped after achieving a cCR due to treatment-related adverse events, per their preference. (Table S1) The most common grade 1 or 2 adverse events included fatigue (23%), pruritis (20%), and rash/dermatitis (20%). Four patients developed grade 3 events, which included diabetes (1%), lung infection (1%), encephalitis (1%) and one (1%) neutropenia, and one patient developed a grade 4 neutropenia (1%). (Table 2) Various assessment modalities were compared to measure response kinetics in patients who achieved a cCR without subsequent surgery. In combined cohorts, the median time to composite complete response was 6.2 months (95% CI 6.2–6.3 months). (Figure S5) Median time to complete response with imaging was 6.1 months (95% CI 6.1–6.2 months). (Figure S5). Endoscopies when available showed a median time to complete response of 6.1 months (95% CI 6.0–6.2 months). (Figure S5) Tumor biopsies were evaluated in both cohorts and the median time to a negative biopsy was 1.5 months (95% CI 1.4–2.7 months). (Figure S5) Circulating tumor DNA was evaluated in both cohorts. 95% (70/74) patients tested were positive at baseline. For evaluable patients also with cCR (n=52), the median time to ctDNA clearance was 1.4 months (95% CI 1.4–1.5 months). (Figure S5)

Tumor biopsies were evaluated in both cohorts and the median time to a negative biopsy was 1.5 months (95% CI 1.4–2.7 months). (Figure S5) Circulating tumor DNA was evaluated in both cohorts. 95% (70/74) patients tested were positive at baseline. For evaluable patients also with cCR (n=52), the median time to ctDNA clearance was 1.4 months (95% CI 1.4–1.5 months). (Figure S5) A total of 343 biopsies were obtained during the study. Comparison of detectable ctDNA levels, a surrogate for the presence of tumor, and biopsy showed substantial concordance (Kappa coefficient of 0.76, 95% CI 0.68 to 0.83) with the 307 (89.50%) observed agreements. Kinetics of ctDNA varied based on response to therapy. Patients whose ctDNA became undetectable during therapy and remained negative at the end of treatment achieved a cCR, whereas those patients whose ctDNA remained detectable during treatment and at the end of therapy had evidence of incomplete responses. In patients who developed recurrence, ctDNA remained detectable during treatment and was slower to clear. (Figure 3)

124 patients were enrolled. Seven patients were excluded from our analysis for the following reasons: three participants were found to have metastases or second malignancies before treatment and withdrew from the study; two participants opted to discontinue therapy after one dose of dostarlimab; one participant was lost to follow up; and one participant stopped treatment after 4 doses. (Figure 1A, Supplemental Appendix). Cohort 1 was comprised of only MMRd rectal cancers. Cohort 2 included MMRd non-rectal solid tumors, including esophagogastric, colon, hepatobiliary, genitourinary, and gynecologic tumors (Table 1; Table S1, Figure S1). On the date of data lock, 103 patients had completed therapy, 49 in cohort 1 and 54 in cohort 2, and 14 remain on therapy. (Figure 1A, Table 1, Table S1) Across both cohorts most patients had radiographically evident lymph nodes (64%). All tumors had loss of expression of mismatch repair proteins with loss of both MLH1 and PMS2 being the most common. Germline pathogenic variants in a mismatch repair gene were noted in 44% of cases. (Table 1, Table S1)

No patients progressed or became unresectable while on therapy. In Cohort 1 (rectal cancer), 49 patients completed treatment. All 49 achieved a cCR, and all 49 opted for non-operative management. (Figure 1B) In Cohort 2 (non-rectal solid tumors), 54 patients completed treatment. Thirty-five (65%) of these 54 achieved a clinical complete response, and 33 (61%) also elected nonoperative management. (Figure 1B) Two patients (one with gastric cancer and one with urothelial cancer) who achieved a cCR, elected to undergo surgical resection. In both cases, no cancer was found in the resection specimen. (Figure 1A, Figure 1B, Table S1-2) Combining cohorts, among 103 patients who completed treatment, 84 (82%, 95% CI: 72%−88%) achieved a clinical complete response and 82 (80%, 70%−87%) also proceeded with non-operative management. (Figure 1B)

In Cohort 1 (rectal cancer) no incomplete responses were observed. (Figure 1B, Figure 2A) In Cohort 2 (non-rectal solid tumors), 19 of 54 patients (35%) had incomplete clinical response. Sixteen of these patients underwent surgical resection of their primary tumors and 3 patients deferred surgery. At baseline, all tumors were found to be MMRd and MSI-high, except for one tumor that had subclonal MLH1/PMS2 loss and was found to be microsatellite stable (MSS). (Table S2) In the 16 patients who underwent surgery, evidence of tumor regression was noted in 14 (87%) of the resected tumors; 1 did not have evaluable tissue; 1 had achieved a pathologic complete response and three had complete pathologic response in the primary tumor but residual tumor present in the lymph nodes. (Supp Table 2). We were able to compare the sequences of the pre and the remnant post-treatment tumors in the incomplete responders with genomic data and found a high likelihood of genomic relatedness (99.6%: 95% CI 97.43–99.97, n=17)18 in all but two patients (51 and 90). In these two cases the remnant tumor was likely a new primary as there was a low likelihood of genomic relatedness (mutational profile similarity score < 0.4%18). Furthermore, the remnant tumors in these two cases were also not MMRd or MSI-high. (Table S2; Figure S2).

99.97, n=17)18 in all but two patients (51 and 90). In these two cases the remnant tumor was likely a new primary as there was a low likelihood of genomic relatedness (mutational profile similarity score < 0.4%18). Furthermore, the remnant tumors in these two cases were also not MMRd or MSI-high. (Table S2; Figure S2). Other than the two unrelated tumors, no meaningful differences were noted between pre and remnant post-treatment tumors in median TMB (43 vs. 34 mutations per MB, Wilcoxon signed rank p=0.43, n=14), mismatch repair protein loss or median microsatellite instability (MSI) sensor score (15 versus 16, Wilcoxon signed rank p=0.47, n=12) (Table S2).

In Cohort 1, all but one patient completed all 9 cycles (6 months). One patient chose to stop therapy after 8 cycles because of an unrelated asthma exacerbation. (Table S1) In Cohort 2 (non-rectal solid tumors), 49 of the 54 patients completed all planned cycles. Three stopped before achieving a complete response due to concerns for lack of response. Two stopped after achieving a cCR due to treatment-related adverse events, per their preference. (Table S1) The most common grade 1 or 2 adverse events included fatigue (23%), pruritis (20%), and rash/dermatitis (20%). Four patients developed grade 3 events, which included diabetes (1%), lung infection (1%), encephalitis (1%) and one (1%) neutropenia, and one patient developed a grade 4 neutropenia (1%). (Table 2)

Various assessment modalities were compared to measure response kinetics in patients who achieved a cCR without subsequent surgery. In combined cohorts, the median time to composite complete response was 6.2 months (95% CI 6.2–6.3 months). (Figure S5) Median time to complete response with imaging was 6.1 months (95% CI 6.1–6.2 months). (Figure S5). Endoscopies when available showed a median time to complete response of 6.1 months (95% CI 6.0–6.2 months). (Figure S5) Tumor biopsies were evaluated in both cohorts and the median time to a negative biopsy was 1.5 months (95% CI 1.4–2.7 months). (Figure S5) Circulating tumor DNA was evaluated in both cohorts. 95% (70/74) patients tested were positive at baseline. For evaluable patients also with cCR (n=52), the median time to ctDNA clearance was 1.4 months (95% CI 1.4–1.5 months). (Figure S5) A total of 343 biopsies were obtained during the study. Comparison of detectable ctDNA levels, a surrogate for the presence of tumor, and biopsy showed substantial concordance (Kappa coefficient of 0.76, 95% CI 0.68 to 0.83) with the 307 (89.50%) observed agreements. Kinetics of ctDNA varied based on response to therapy. Patients whose ctDNA became undetectable during therapy and remained negative at the end of treatment achieved a cCR, whereas those patients whose ctDNA remained detectable during treatment and at the end of therapy had evidence of incomplete responses. In patients who developed recurrence, ctDNA remained detectable during treatment and was slower to clear. (Figure 3)

These data from a small cohort of patients demonstrate that PD-1 blockade enables non-operative management for many Stage I, II and III MMRd tumors. As seen in the metastatic setting, the efficacy of checkpoint blockade appears to be relatively tumor-agnostic, driven primarily by the MMRd phenotype rather than the tumor’s site of origin. In the curative setting, six-months of neoadjuvant PD-1 blockade was associated with few adverse events. Of equal importance, the option for curative intent surgery was never compromised despite a prolonged course of treatment relative to other neoadjuvant studies. A major concern during neoadjuvant treatment is that the ‘window of opportunity’ for resection may lapse and tumors may grow and spread to the point of no longer being curative by surgery, which often limits the duration of treatment to a few cycles of therapy. To maximize clinical response, we treated with six-months of checkpoint blockade, which in prior studies appears to result in a higher complete clinical response rate than treatment for a shorter duration.20–22 All of the patients who achieved a clinical complete response have preserved their organs without additional therapy, In three cases of rectal cancer, patients were able to conceive and deliver healthy children which would not have been possible with standard treatment of rectal cancer23.

In the curative setting, six-months of neoadjuvant PD-1 blockade was associated with few adverse events. Of equal importance, the option for curative intent surgery was never compromised despite a prolonged course of treatment relative to other neoadjuvant studies. A major concern during neoadjuvant treatment is that the ‘window of opportunity’ for resection may lapse and tumors may grow and spread to the point of no longer being curative by surgery, which often limits the duration of treatment to a few cycles of therapy. To maximize clinical response, we treated with six-months of checkpoint blockade, which in prior studies appears to result in a higher complete clinical response rate than treatment for a shorter duration.20–22 All of the patients who achieved a clinical complete response have preserved their organs without additional therapy, In three cases of rectal cancer, patients were able to conceive and deliver healthy children which would not have been possible with standard treatment of rectal cancer23. The most common tumor types not to achieve a complete clinical response were prostate and gastroesophageal cancers. In the remaining cases, it is unclear why different sites of primary disease have varying frequency of complete clinical responses, but the contribution of the microbiome and other cellular components of tumor microenvironment, like myeloid cells, and underlying tumor heterogeneity likely contribute to these outcomes. It is also possible that a longer duration of therapy or combination immune checkpoint blockade would have led to complete responses in these tumors.22

n of the microbiome and other cellular components of tumor microenvironment, like myeloid cells, and underlying tumor heterogeneity likely contribute to these outcomes. It is also possible that a longer duration of therapy or combination immune checkpoint blockade would have led to complete responses in these tumors.22 Of note, only 5 patients developed disease recurrence. In 4 of these cases, recurrence was restricted to lymph nodes that were treated with either resumption of PD-1 blockade (4 patients) or by resection the solitary recurrent lymph node. Anti-PD-1 rechallenge resulted in complete disease regression in every patient, suggesting that resistance had not developed to the pre-existing neoantigens or that new neoantigens emerged sufficient to trigger a de novo immunotherapeutic response.24–26 A significant challenge moving forward is to determine the best approach to monitor tumor responses. We employed tumor specific imaging, clinical exam and visualization by endoscopy, and biopsy. While GI tumors were amenable to most of these approaches, tumors in sites not accessible endoscopy relied completely on imaging. Monitoring response with imaging was successful in most cases. However, imaging is not as definitive as obtaining tissue.

c imaging, clinical exam and visualization by endoscopy, and biopsy. While GI tumors were amenable to most of these approaches, tumors in sites not accessible endoscopy relied completely on imaging. Monitoring response with imaging was successful in most cases. However, imaging is not as definitive as obtaining tissue. Analysis with ctDNA provided a unique window into the dynamic response to PD-1 blockade. This tumor informed approach, where mutations in the tumor are used to design probes to analyze circulating cell-free DNA was first described in mice and later in humans for the purposes of tracking tumor burden and to measure minimal residual disease.27,28 It is important to note that ctDNA detects active disease in real time, in part, because ctDNA has a short plasma half-life of less than 2 hours.28 In our study, we employed a novel assay, that achieves very high sensitivity and specificity by simultaneously measuring up to 50 tumor-defined mutations in cfDNA and by discarding sequencing-induced artifactual mutations by requiring detected genomic alterations to be present on the forward and reverse strands of a cfDNA fragment.29–31 Using this approach, more than 90% of our early-stage tumors were detected at baseline, and we were able to reliably view the therapeutic response in real time. All patients with a complete clinical response cleared ctDNA by the end of treatment, while those with incomplete responses or eventual recurrences were slow to clear ctDNA and levels remained elevated throughout treatment in many cases.

baseline, and we were able to reliably view the therapeutic response in real time. All patients with a complete clinical response cleared ctDNA by the end of treatment, while those with incomplete responses or eventual recurrences were slow to clear ctDNA and levels remained elevated throughout treatment in many cases. Furthermore, when we compared ctDNA levels with matched biopsy results the concordance was high, suggesting ctDNA was in fact a reliable ‘liquid biopsy’. In the future, this approach will be an important addition to response assessment, as novel therapies enter the neoadjuvant arena and will require methods to measure response, especially when tumors are in locations where direct visualization and biopsies are not feasible.

ctDNA was in fact a reliable ‘liquid biopsy’. In the future, this approach will be an important addition to response assessment, as novel therapies enter the neoadjuvant arena and will require methods to measure response, especially when tumors are in locations where direct visualization and biopsies are not feasible. While encouraging, these data will require a larger study to confirm the long-term benefit of this approach especially for non-rectal cohort with a median follow-up for recurrence of 14 months. Whether these studies are tumor agnostic or restricted to a single tumor type will be an important consideration as well as whether a randomized study is necessary to confirm these findings. In tumors where surgical resection is less morbid and has less impact on quality of life (e.g., colon cancer), a randomized study designed to measure overall survival is prudent. However, in tumors where the surgery itself is significantly life altering (e.g. esophagus, stomach, pancreas, bladder or rectum), randomizing to surgery may be less feasible and here a single arm experience especially in tumors with a high complete response rate should be sufficient to modify clinical practice. This study provides a foundation to alter the traditional therapeutic paradigm for early stage MMRd tumors and eliminate the need for surgery and other therapy for most patients.