Tumor-informed circulating tumor DNA stratifies recurrence risk and survival in anal squamous cell carcinoma

Abstract

Patients with anal squamous cell carcinoma (ASCC) who fail chemoradiation (CRT) have poor outcomes, underscoring the need for biomarkers to guide risk stratification. In a real-world two-center cohort of 84 adults with non-metastatic ASCC treated with curative-intent CRT, we prospectively evaluate a tumor-informed circulating tumor DNA (ctDNA) assay (Signatera^TM, Natera). Here we show that across 647 plasma specimens, ctDNA is positive at pre-treatment in 79% (61/77), including 89% (24/27) with stage III disease. End-of-treatment ctDNA positivity identifies patients with inferior one-year outcomes: 63% overall survival, 44% progression-free survival, and 39% locoregional failure. Conversely, patients who were ctDNA-negative at baseline or who cleared ctDNA during-treatment have 100% locoregional failure-free survival. During surveillance, ctDNA re-emergence precedes clinical or radiographic relapse in every case. These findings support the consideration of ctDNA as a dynamic, treatment-responsive biomarker warranting prospective validation for risk-adapted surveillance and adjuvant therapy in ASCC.

Prediction of tumour recurrence in anal squamous cell carcinoma can be challenging. Here, the authors use a tumour-informed ctDNA assay to show concordance with relapse in patients following failed definitive-intent chemoradiation.

Introduction

Anal squamous cell carcinoma (ASCC) is a human papillomavirus (HPV)–associated malignancy for which definitive-intent (chemo)radiation (CRT) remains the standard treatment for non-metastatic disease^1–3. Although CRT often achieves long-term control and sphincter preservation, up to 6 months of serial evaluations may be required to assess treatment response⁴. This prolonged interval may delay the detection of treatment failure, limit the window for effective salvage therapy, and contribute to patient anxiety and uncertainty in management. For patients with persistent or recurrent locoregional disease, salvage abdominoperineal resection (APR) remains the primary treatment, but it is associated with poor disease-specific and recurrence-free survival^5,6. These limitations underscore the need for early, reliable biomarkers to identify high-risk patients during or shortly after CRT, when timely therapeutic intervention may still improve outcomes.

Circulating tumor DNA (ctDNA) has emerged as a promising, minimally invasive blood-based biomarker that enables dynamic monitoring of disease burden through the detection of tumor-specific mutations in plasma^7–11. Prior studies in ASCC have predominantly focused on HPV-directed ctDNA assays^12–20, but these approaches may lack sensitivity in patients with HPV-negative (~5–10% of patients with ASCC)^21,22. Tumor-informed ctDNA assays, customized to each patient’s unique mutational profile, may offer a more sensitive and broadly applicable alternative, although their utility in ASCC has not been well established yet^23,24.

In this multicenter, real-world cohort study, we evaluated a commercially available, personalized, tumor-informed ctDNA assay for its prognostic and surveillance utility in patients with ASCC undergoing definitive-intent CRT. We hypothesized that ctDNA kinetics at pre-treatment, during-treatment, and end-of-treatment would correlate with clinical outcomes and, if validated, could inform risk-adapted, personalized treatment strategies.

Results

Patient, tumor, and treatment characteristics

Of 88 consecutive patients with localized ASCC who underwent definitive-intent dose-painted intensity-modulated radiation therapy (DP-IMRT), 84 patients (95%) had sufficient tumor tissue from diagnostic biopsies for ctDNA testing and were included in the final cohort (Fig. 1A). Whole-exome sequencing (WES) identified key mutations across this population (Fig. 1B). Demographic and treatment characteristics are summarized in Table 1 and Supplementary Table 1. The median follow-up among survivors, measured from diagnosis, was 18 months (interquartile range [IQR], 11–26 months).

Fig. 1. Patient, tumor, cfDNA and ctDNA characteristics.

A Representative overview of the cohort. B Oncoplot highlighting key genetic mutations as determined by whole exome sequencing. C Plot of cell-free DNA (cfDNA) concentrations obtained from plasma specimens across all time points (N = 647) relative to CRT course. Red= pre-treatment cfDNA specimens, green= duringtreatment cfDNA specimens, teal= end-of-treatment cfDNA specimens, purple= surveillance cfDNA specimens. D The detectability of ctDNA relative to variations in cfDNA concentration throughout treatment and surveillance. Blue= ctDNA negative. Red= ctDNA positive. ASCC anal squamous cell carcinoma, CRT chemoradiation, TMB tumor mutational burden, MSI microsatellite instability, MSS microsatellite stable, HPV human papillomavirus, HIV human immunodeficiency virus, N number.

Table 1.

Patient, tumor, and treatment characteristics, including sub-stratifications for pre-treatment ctDNA status

Patient and tumor characteristics	Total (N = 84)	ctDNA Positive (N = 61)	ctDNA Negative (N = 16)	ctDNA Unknown (N = 7)
Sex, N(%)
Female	58 (69%)	41 (67%)	12 (75%)	5 (71%)
Male	26 (31%)	20 (33%)	4 (25%)	2 (29%)
Age (Years) at CRT Start (median, IQR)	65 (59–73)	65 (60–72)	60 (56–73)	70 (56–80)
KPS score (median, IQR)	90 (80–90)	90 (80–90)	90 (90–90)	90 (80–90)
Unknown	2	1	0	1
HIV status, N(%)
Positive	9 (11%)	6 (10%)	2 (13%)	1 (14%)
Negative	49 (58%)	35 (57%)	10 (63%)	4 (57%)
FUnknown	26 (31%)	20 (33%)	4 (25%)	2 (29%)
Smoking history, N(%)
Yes	32 (38%)	21 (34%)	8 (50%)	3 (43%)
No	52 (62%)	40 (66%)	8 (50%)	4 (57%)
T stage, N(%)
T1/2	52 (62%)	36 (59%)	14 (88%)	2 (29%)
T3/4	28 (33%)	21 (34%)	2 (13%)	5 (71%)
Tx	4 (5%)	4 (7%)	0 (0%)	0 (0%)
N stage, N(%)
N0	26 (31%)	14 (23%)	10 (63%)	2 (29%)
N1	42 (50%)	35 (57%)	3 (19%)	4 (57%)
Nx	16 (19%)	12 (20%)	3 (19%)	1 (14%)
AJCC stage, N(%)
I	10 (12%)	3 (5%)	7 (44%)	0 (0%)
II	42 (50%)	34 (56%)	6 (38%)	2 (29%)
III	32 (38%)	24 (39%)	3 (19%)	5 (71%)
HPV status/p16 status, N(%)
Positive	48 (57%)	33 (54%)	10 (63%)	5 (71%)
Negative	7 (8%)	7 (11%)	0 (0%)	0 (0%)
Unknown	29 (35%)	21 (34%)	6 (38%)	2 (29%)
Treatment characteristics
Radiation dose, Gy (median, IQR)	54 (50–58)	54 (50.4–58)	50 (50–54)	54 (52–56)
Radiation fractions, N (median, IQR)	27 (27–29)	27 (27–29)	27 (25–27)	27 (27–29)
Radiation duration, days (median, IQR)	38 (36–41)	38 (37–41)	36 (35–37)	37 (36–44)
Concurrent chemotherapy, N(%)
5-Fluorouracil/Mitomycin	16 (19%)	12 (20%)	2 (13%)	2 (29%)
Capecitabine/Mitomycin	61 (73%)	43 (70%)	13 (81%)	5 (71%)
Capecitabine	6 (7%)	5 (8%)	1 (6%)	0 (0%)
None	1 (1%)	1 (2%)	0 (0%)	0 (0%)
Follow-up of survivors from diagnosis, months (median, IQR)	18 (11–26)	17 (11–26)	18 (11–23)	18 (18–28)

CRT chemoradiation, KPS Karnofsky Performance Status, HIV human immunodeficiency virus, HPV human papillomavirus; AJCC American Joint Committee on Cancer, N number, Gy Gray.

ctDNA detection and temporal dynamics

Across all timepoints, 647 plasma specimens were analyzed for ctDNA testing using a personalized, tumor-informed multiplex polymerase chain reaction-next-generation sequencing (mPCR-NGS) assay (Signatera^TM) (Fig. 1C). Cell-free DNA (cfDNA) concentrations increased during CRT, but ctDNA detection remained unaffected by cfDNA concentration (Fig. 1D). At pre-treatment, 79% (61 of 77) of patients were ctDNA-positive, including 89% (24 of 27) with stage III disease. Pre-treatment ctDNA positivity was significantly associated with T3/T4 tumors (odds ratio [OR], 4.08; 95% confidence interval [CI], 1.01–27.6; P = 0.049) and node-positive disease (OR, 8.33; 95% CI, 2.19–41.5; P = 0.001) (Supplementary Table 2). As treatment progressed, ctDNA levels generally declined, though levels remained higher at each timepoint among patients with more advanced clinical stage, compared to stage I (Supplementary Fig. 1 and Supplementary Table 3). Median ctDNA levels at pre-treatment, during-treatment, and end-of-treatment were 1.02 MTM/mL (interquartile range (IQR), 0.17–25.43), 0 MTM/mL (IQR, 0–0.46), and 0 MTM/mL (IQR, 0–0). Of the end-of-treatment samples, 6 were drawn in the final week of CRT (−7 to −1 days), 7 on the final day of CRT (0 day), and 60 after CRT completion (+1 to +42 days) (Fig. 1C).

Treatment failure events

The clinical courses of the cohort, grouped by baseline stage and outcome, are shown in Fig. 2. Twelve patients (14%) experienced treatment failure (including death without progression): seven locoregional failures, four distant failures, and one death (Supplementary Fig. 2). Of the seven patients with locoregional failure, three had local-only failure, two had nodal-only failure, and two had combined local-nodal failure. Among patients with locoregional recurrence, three underwent salvage APR, two received nonoperative salvage with re-irradiation, and two were deemed medically ineligible for salvage therapy and were managed with best supportive care. Five deaths occurred in the post-CRT follow-up period. Four were attributed to disease progression, including three from distant metastatic disease and one from locoregional recurrence as the patient died from complications of the recurrent tumor. At the last available disease assessment prior to death, all four had positive ctDNA. One additional patient died without evidence of disease progression due to complications of COVID-19 and Clostridioides difficile (C. difficile) colitis.

Fig. 2. Swimmer’s plot representing the clinical courses of all 84 patients analyzed.

A Baseline Stage I patients without treatment failure (N = 10). B Baseline Stage II patients without treatment failure (N = 39). C Baseline Stage III without treatment failure (N = 23). D All patients who developed treatment failure (N = 12). Figure legend outlines ctDNA concentration (white circle outlined in black = ctDNA undetectable, black circle = ctDNA detectable with the size of the circle representing ctDNA concentration), CRT interval (orange rectangle), and clinical response endpoints (i.e., sigmoidoscopy (diamond) and PET/CT/MRI imaging (square)). Color of Clinical response endpoints (i.e., sigmoidoscopy (diamond) and PET/CT/MRI imaging (square)) show response (grey= not specified, orange= equivocal, green= negative, and red= positive).

Among the 12 patients with treatment failure or death, 11 (92%) had at least one high-risk feature (tumor >4 cm and/or nodal involvement). All patients completed their prescribed course of CRT without dose reductions or treatment breaks. In a competing-risks analysis, absence of mitomycin during CRT and HPV status were not associated with increased hazard of locoregional recurrence (Supplementary Table 4). In all cases, ctDNA positivity preceded clinical or radiographic evidence of recurrence (Supplementary Fig. 2). Notably, all patients who were ctDNA-negative at baseline remained recurrence-free during follow-up, suggesting this subgroup may represent particularly favorable biology.

Overall survival, progression-free survival, and locoregional failure

Pre-treatment ctDNA-negative patients had an estimated 1-year overall survival (OS) of 100% (95% CI, 100–100%), compared to 97% (95% CI, 92–100%) for those with ctDNA-positive status (Fig. 3A). Stratified by during-treatment ctDNA status, 1-year OS was 100% (95% CI, 100–100%) for ctDNA-negative versus 93% (95% CI, 84–100%) for ctDNA-positive patients (Fig. 3B). At end-of-treatment, ctDNA-negative patients had a 1-year OS of 96% (95% CI, 90–100%) versus 63% (95% CI, 32–100%) for ctDNA-positive patients (Fig. 3C).

Fig. 3. Key time-to-event analyses: overall survival (OS), progression-free survival (PFS), and locoregional failure (LRF), each stratified by ctDNA status (i.e., positive or negative) at various timepoints.

Kaplan–Meier estimates for OS stratified by ctDNA status at A pre-treatment, B during-treatment, and C end-of-treatment. Kaplan–Meier estimates for PFS stratified by ctDNA status at D pre-treatment, E during-treatment, and F end-of-treatment. P-values shown in A–F were calculated using two-sided log-rank tests and were not adjusted for multiple comparisons. Exact p-value for F: 3 × 10^-7. Cumulative incidence curves for LRF stratified by ctDNA status at G, pre-treatment, H during-treatment, and I end-of-treatment. P-values shown in G–I were calculated using two-sided Gray’s tests and were not adjusted for multiple comparisons. Exact p-value for I: 0.0003. Color of line indicates ctDNA status where red= ctDNA positive and blue= ctDNA negative. Source data is provided in the Source Data file.

Similarly, 1-year progression-free survival (PFS) was 100% (95% CI, 100–100%) for pre-treatment ctDNA-negative patients and 85% (95% CI, 76–96%) for ctDNA-positive patients (Fig. 3D). During-treatment ctDNA-negative patients had a 1-year PFS of 100% (95% CI, 100–100%), compared to 78% (95% CI, 62–98%) for ctDNA-positive patients (Fig. 3E). End-of-treatment ctDNA-negative patients had a 1-year PFS of 92% (95% CI, 84–100%) versus 44% (95% CI, 17–100%) for ctDNA-positive patients (Fig. 3F).

Locoregional failure (LRF) at 1 year was 0% (95% CI, undefined) among ctDNA-negative patients versus 11% (95% CI, 4–23%) among ctDNA-positive patients at pre-treatment (Fig. 3G). Pre-treatment quantitative ctDNA was not associated with LRF (HR 1.0, 95% CI 0.99–1.00). During treatment, 1-year LRF was 0% (95% CI, undefined) versus 19% (95% CI, 5–38%) for ctDNA-negative and ctDNA-positive groups, respectively (Fig. 3H). At the end-of-treatment, the 1-year LRF was 6% (95% CI, 2–16%) for ctDNA-negative versus 39% (95% CI, 2–79%) in ctDNA-positive patients (Fig. 3I).

When stratified by the timing of initial ctDNA clearance, patients who were ctDNA-negative at pre-treatment or who cleared ctDNA during-treatment had a 1-year OS of 100% (95% CI, 100–100%; Fig. 4A), 1-year PFS of 100% (95% CI, 100–100%; Fig. 4B), and 1-year LRF rate of 0% (95% CI, undefined; Fig. 4C). Among patients who cleared ctDNA at end-of-treatment, 1-year OS was 89% (95% CI, 75–100% Fig. 4A), PFS was 81% (95% CI, 65–100%; Fig. 4B), and LRF was 15% (95% CI, 4–35%; Fig. 4C). Patients who were ctDNA-positive at end-of-treatment had significantly worse outcomes with 1-year OS of 63% (95% CI, 32–100%), PFS of 44% (95% CI, 17–100%), and LRF of 39% (95% CI, 2–79%) (log-rank P = 0.01 for OS and P = 2.4 × 10⁻⁶ for PFS; Gray’s P = 0.001 for LRF). All patients who were ctDNA-negative at end-of-treatment achieved a clinical complete response.

Fig. 4. Key survival outcomes stratified by the time frame of achieving initial ctDNA clearance.

A Overall survival by time of initial ctDNA clearance. B Progression-free survival by time of initial ctDNA clearance. C Locoregional failure by time of initial ctDNA clearance. Exact p-value for B: 3 × 10⁻⁶. P-values in A, B were calculated using two-sided log-rank tests and P-value in C was calculated using a two-sided Gray’s test. P-values were not adjusted for multiple comparisons. Color of line indicates ctDNA response where blue= ctDNA negative at pre-treatment, green= ctDNA negative during-treatment, purple= ctDNA negative end-of-treatment, and black= ctDNA positive end-of-treatment. Source data is provided in the Source Data file.

ctDNA surveillance monitoring

Seven patients developed molecular recurrence during the surveillance period after initially achieving clearance (Fig. 5). All these patients subsequently experienced confirmed relapse, five locoregional and two distant, with ctDNA re-positivity consistently preceding radiographic or clinical evidence of recurrence by a median of 2.4 months (range 1–8.7 months). Because ctDNA re-positivity triggered off-schedule “enhanced” assessments, the 2.4-month median lead time in the surveillance subset likely underestimates the biological lead time relative to standard NCCN surveillance. Among all patients who were persistently ctDNA-negative during the surveillance period, no treatment failure events occurred.

Fig. 5. ctDNA profiles of patients who achieved ctDNA clearance later followed by molecular recurrence (N = 7).

Molecular recurrence preceded clinical or radiographic evidence of recurrence in all cases. Color of line and circles signify visit timepoint where blue = pre-treatment, red= during-treatment, purple= end-of-treatment, and green= surveillance. ctDNA status indicated shaded circle (ctDNA positive) or unshaded circle (ctDNA negative). Duration of CRT indicated by grey shading. Vertical lines indicate imaging timepoint and status where dotted lines indicate no evidence of disease whereas solid lines indicate disease recurrence. The color of the vertical lines indicates type of imaging performed (green= PET-CT, blue= MRI, and orange= CT imaging). CRT chemoradiation, ctDNA circulating tumor DNA. Source data is provided in the Source Data file.

Discussion

This real-world, multicenter study demonstrates that a commercially available and clinically validated, personalized, tumor-informed mPCR-NGS ctDNA assay (Signatera^TM, Natera) is a sensitive tool for assessing ctDNA, a minimally-invasive blood-based biomarker in patients with ASCC treated with definitive-intent CRT. We report four key findings: (1) end-of-treatment ctDNA positivity identifies patients at the highest risk for recurrence and death; (2) early ctDNA clearance stratifies patients with excellent outcomes; (3) ctDNA re-emergence during surveillance reliably precedes clinical relapse; and (4) tumor-informed ctDNA testing integrates effectively into real-world workflows. This study demonstrates that ctDNA dynamics correlate with all three major clinical outcomes, including overall survival, progression-free survival, and locoregional failure, in ASCC, underscoring the broad prognostic value of this approach.

End-of-treatment ctDNA positivity was strongly associated with poor outcomes, including 1-year OS of 63%, PFS of 44%, and LRF rate of 39%. These data provide evidence of the variability in oncological endpoints as a function of ctDNA dynamics in patients with ASCC undergoing definitive-intent CRT. The end-of-treatment result provided the clearest prognostic separation, highlighting its value as a clinically informative and potentially actionable timepoint for future trials of adjuvant or intensified surveillance.

Conversely, patients who were ctDNA-negative at baseline and remained negative, as well as those who converted to negative during treatment, had uniformly excellent outcomes. In contrast, some patients who were ctDNA-negative at end-of-treatment later reconverted to ctDNA positivity during surveillance and subsequently relapsed. These findings suggest that earlier and sustained ctDNA negativity provides the strongest reassurance, whereas negativity at the end-of-treatment may be less reliable. Thus, serial monitoring across treatment phases provides important context beyond any single time point, supporting its use in future efforts to refine risk stratification and guide surveillance strategies.

Our data also inform surveillance paradigms. ctDNA positivity consistently preceded clinical and/or radiographic recurrence. Importantly, no relapses were missed by ctDNA testing. Because ctDNA re-positivity triggered off-schedule clinical evaluation and imaging, the observed lead time is likely conservative relative to standard NCCN surveillance intervals⁴, as expedited assessments likely led to a shorter interval between molecular and clinical/ radiographic detection. These results highlight ctDNA’s potential as an early signal of relapse, although whether initiating salvage therapy, such as (chemo)immunotherapy, in the setting of “molecular relapse” improves outcomes remains to be tested prospectively. Notably, all patients with negative baseline ctDNA remained recurrence-free, suggesting that this subgroup may represent favorable biology and could eventually warrant de-intensified surveillance if confirmed.

Importantly, the end-of-treatment window emerged as both clinically informative and actionable. While prior studies have associated HPV-ctDNA measured at 3 months post-CRT with recurrence¹⁶, few have demonstrated the utility of testing during or at the end-of-treatment (Supplementary Table 5)^{13–20,25–28}. In our cohort, end-of-treatment ctDNA positivity offered an earlier and more prognostic signal of residual disease. For example, 69% of patients in this study would have met eligibility for the EA2165 trial (NCT03233711), which evaluates adjuvant nivolumab in high-risk ASCC. Yet many of these patients did not recur, highlighting the potential role of ctDNA to refine risk stratification, enrich for patients most likely to benefit from adjuvant therapy, and spare others from unnecessary treatment and toxicity.

Comparison with viral-based assays further clarifies the complementary strengths of tumor-informed ctDNA assays. HPV-ctDNA (ctHPVDNA) tests such as TTMV-HPV DNA (NavDx®) demonstrate good performance in HPV-positive patients, with sensitivity of 83%, specificity of 98%, positive predictive value (PPV) of 96%, and negative predictive value (NPV) of 93%²⁷. While these assays are simple and tissue-independent, they are not applicable to HPV-negative disease, which represents ~5–10% of patients with ASCC⁶. In contrast, tumor-informed assays are individualized to each patient’s mutational profile, enabling use across HPV-positive and HPV-negative subgroups. In our cohort, outcomes were similar between HPV-positive and HPV-negative patients, with no statistically significant association between HPV status and locoregional recurrence when considering distant recurrence and death as competing events (Supplementary Table 4), underscoring the applicability of tumor-informed monitoring even in HPV-negative disease. As shown in oropharyngeal cancer, next-generation sequencing-based ctHPVDNA assays achieve higher sensitivity and longer lead times than ddPCR-based approaches²⁹, suggesting potential complementarity between viral and tumor-informed strategies.

Several limitations warrant mention. The cohort size was modest, with relatively few events and limited follow-up. Some patients who cleared ctDNA with CRT ultimately developed ctDNA re-positivity followed by clinical recurrence, likely reflecting disease below the ctDNA detection threshold. This reinforces the importance of ongoing serial monitoring and highlights that treatment de-escalation or surveillance de-escalation should not be based on a single negative test. In this study, we did not evaluate patient-reported outcomes (PRO) related to anxiety or psychosocial impact of serial ctDNA monitoring. Although no participant requested discontinuation of testing or reported testing-related distress during clinical encounters, future studies incorporating standardized PRO instruments are warranted to quantify psychosocial implications of MRD surveillance. Subgroup analyzes (e.g., HPV-negative patients, non-mitomycin patients) were limited by the small sample size. These caveats highlight the importance of validation in larger, prospective studies.

Despite these limitations, this is one of the largest longitudinal ctDNA studies in ASCC to date, with real-time monitoring across the treatment continuum. By showing that ctDNA dynamics correlate with OS, PFS, and LRF, our findings support tumor-informed ctDNA as a hypothesis-generating biomarker that should inform the design of prospective trials. With broader validation, ctDNA-guided strategies may eventually enable risk-adapted treatment, earlier escalation for high-risk patients, and thoughtful de-intensification for patients with molecular clearance during treatment.

Methods

Study design and participants

The institutional review and privacy boards of Memorial Sloan Kettering Cancer Center (MSK; IRB 16-370) and the University of South Florida (USF; IRB 004774) approved this cohort study with a waiver of informed consent. Patient confidentiality was maintained as required by the Health Insurance Portability and Accountability Act (HIPAA). Beginning in March 2021, adult patients (≥18 years) with biopsy-confirmed, non-metastatic ASCC were prospectively offered longitudinal ctDNA monitoring, and all patients who provided informed consent to ctDNA testing were included. All patients were staged per the American Joint Committee on Cancer (AJCC), 8th edition³⁰ and received curative-intent (chemo)radiation. Data collection concluded on July 31, 2024. Data sharing was governed by a three-way agreement among MSK, USF, and Natera, Inc., to allow data to be shared freely across all participating institutions, with the final dataset maintained at MSK.

Personalized mPCR-NGS ctDNA assay

A clinically validated, personalized, tumor-informed 16-plex mPCR-NGS assay (Signatera^TM, Natera, Inc.) was used for the detection and quantification of ctDNA in blood samples³¹. Formalin-fixed, paraffin-embedded tumor tissue from surgical resection or biopsy samples and matched normal DNA extracted from a peripheral blood sample were processed for whole-exome sequencing to identify and track up to 16 patient-specific and tumor-specific somatic single-nucleotide variants (SNVs) in the associated patient’s plasma using a multiplex PCR-NGS approach. Cell-free DNA was extracted from patient plasma (median, 9.4 ml; range, 2.9–10.2 ml) at a given time point and was used to detect ctDNA. A plasma sample was classified as ctDNA-positive if two or more tumor-specific variants out of the 16 were detected above a predefined threshold. The predefined threshold is based on Natera’s proprietary variant calling method, with greater than 95% sensitivity at a mean variant allele frequency of 0.01% and 99.7% specificity³². ctDNA concentration was reported as mean tumor molecules per milliliter (MTM/mL) of plasma. Testing cadence was at the discretion of the treating physician, with results available in real time. Lead time was defined as the interval between the first ctDNA re-positivity after a documented negative result and the date of first clinical or radiographic evidence of recurrence.

Treatment and response assessment

All patients received curative-intent (chemo)radiation, typically with DP-IMRT. Radiation dose was adapted to tumor size: <2 cm, 50 Gy; 2–4 cm, 54 Gy; >4 cm, 58 Gy, all delivered in 2-Gy fractions⁶. Most patients received concurrent fluoropyrimidine-based chemotherapy, with or without mitomycin, unless clinically contraindicated.

Treatment response was assessed 12–16 weeks after CRT using physical examination (including digital rectal and inguinal lymph node exams), anoproctoscopy, and cross-sectional imaging (MRI, CT, or PET-CT). Clinical complete response (cCR) was defined as no detectable tumor locally, regionally, or distantly. For patients with an incomplete but improving primary tumor response, short-interval re-evaluation (within 4–12 weeks) was permitted. Recurrence was defined as local, regional, or distant disease. Locoregional failure was defined as recurrence at the primary site or in pelvic/inguinal regional nodes, with ‘combined local–nodal failure’ reserved for concurrent recurrence at both sites. Salvage management for locoregional recurrence was categorized as operative salvage (APR), nonoperative salvage (re-irradiation and/or systemic therapy), or best supportive care for patients medically ineligible for salvage therapy.

Phlebotomy timeline

Peripheral blood was collected at physician-defined intervals throughout the treatment course (Supplementary Fig. 3). Pre-treatment samples were obtained at any point before treatment initiation or within 5 days after starting CRT. During-treatment samples were collected more than 5 days after CRT initiation but at least 7 days before its completion. End-of-treatment samples were drawn from 7 days before through 42 days after the end-of-CRT. Samples collected more than 42 days after CRT completion were categorized as surveillance timepoints. Patients who re-converted from negative to positive ctDNA status during surveillance underwent “enhanced surveillance”, including directed clinical examination (digital rectal examination and, when appropriate, flexible sigmoidoscopy/anoscopy) and expedited cross-sectional imaging (MR pelvis, CT chest/abdomen/pelvis, and/or PET-CT).

Study endpoints

The primary endpoint was OS, defined as the time from ctDNA assessment to death from any cause or to last follow-up. Secondary endpoints included progression-free survival and the cumulative incidence of locoregional failure. PFS was defined as the time from ctDNA assessment to the first occurrence of local or regional failure, distant metastasis, or death. LRF was defined as persistent, progressive, or recurrent disease in the primary tumor or regional lymph nodes, with death and distant recurrence treated as competing-risks.

Statistical analyses

Baseline characteristics were summarized using medians and interquartile ranges (IQR) for continuous variables, and frequencies with percentages for categorical variables. Univariate logistic regression was used to evaluate predictors of pre-treatment ctDNA positivity.

Kaplan–Meier estimates were used to evaluate OS and PFS from the date of end-of-treatment ctDNA assessment, stratifying patients by time of clearance. Additional timepoint-specific Kaplan–Meier analyzes were conducted for OS and PFS according to ctDNA status at each sampling point, measured from the corresponding ctDNA collection date. Survival differences were compared using log-rank tests. LRF was analyzed at the same timepoints using competing-risk regression, with cumulative incidence curves by Gray’s test.

Patients without a ctDNA result at a given time point were excluded from time point-specific survival analyzes. If multiple samples existed in a given phase (e.g., during-treatment), the most recent test was selected. In surveillance, if all ctDNA results were negative, the most recent was used; if any were ctDNA-positive, the first positive test was selected. Analyzes were performed in R (v4.4.0) using tidyverse (v2.0.0), gtsummary (v2.0.0), ggsurvfit (v1.1.0), ggalluvial (v0.12.5), and ggpubr (v0.6.0).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Authors P.B.R., A.B., J.A., C.E.K., M.G., J.J.S., and R.T. had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Concept and Design: P.B.R. Acquisition, analysis, or interpretation of data: P.B.R., A.B., J.A., C.E.K., Y.J.H., M.G., J.J.S., and R.T. Drafting of Manuscript: P.B.R., A.B., J.A., R.T. Critical Revision of the manuscript for intellectual content: P.B.R., A.B., J.A., C.E.K., A.A., E.K.L., Y.J.H., N.M., M.R., R.Y., D.A.R.O., J.J.C., M.Z., C.O., R.R., J.L., W.Z., V.N.A., S.S., M.M., S.R.H., N.S., A.J., M.C.L., A.W., V.W., L.C., E.P., D.R., N.H.S., P.B.P., M.R.W., A.C., J.M., J.G.A., C.H.C., M.H., J.J.S., R.T. Statistical Analysis: P.B.R., A.B., J.A., C.E.K., Y.J.H., M.G. Administrative, technical, or material support: P.B.R., M.R.W., P.B.P., M.G., J.G.A., J.J.S., R.T. Supervision: P.B.R.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data necessary to interpret the findings are presented in the article (Fig. 1 shows patient-level mutational profiles; Fig. 2 shows patient-level courses, outcomes, and ctDNA results for all patients; Fig. 5 shows ctDNA profiles, clinical and radiographic response assessments, and details on treatment failures for the seven patients with molecular recurrence; Supplementary Fig. 2 shows ctDNA profiles, clinical and radiographic response assessments, and details on treatment failures for all patients with treatment failure or death). Source data underlying Figs. 3, 4, and 5 alongside Supplementary Figs. 1 and 2 are provided with this paper in the Source Data file (see README tab for sheet map and variable definitions). De-identified clinical/ctDNA datasets (sample IDs, time-stamped results, MTM/mL) and underlying raw sequencing files are available under restricted access because public deposition is not permitted by IRB approvals, participant consent, HIPAA privacy regulations, and institutional/contractual obligations. Access can be obtained by emailing the corresponding author (P.B.R.; romessep@mskcc.org) with a brief research proposal, institutional affiliation, and IRB/ethics documentation (or a non-human subjects determination). Requests will be acknowledged and reviewed within 2–3 weeks; upon approval, a data transfer/use agreement with Memorial Sloan Kettering Cancer Center, coordinated as needed with the University of South Florida and Natera, will be executed and data shared via secure transfer for non-commercial verification/replication only, with no redistribution or secondary use, for the duration specified in the agreement. No external datasets were used in this study. Source data are provided with this paper.

Code availability

Analyses were performed in R (v4.4.0) using openly available packages (tidyverse v2.0.0, gtsummary v2.0.0, ggsurvfit v1.1.0, ggalluvial v0.12.5, ggpubr v0.6.0). No novel algorithms were developed. Custom scripts used for data wrangling, survival analyses, and figure generation are available under restricted access; requests to the corresponding author (romessep@mskcc.org) will be reviewed within 2–3 weeks and, if approved, shared under a data transfer/use agreement for non-commercial verification only (no redistribution or secondary use).

Competing interests

Dr. Romesser provides compensated professional services and activities for EMD Serono, Faeth Therapeutics, HPV Alliance, and Natera Inc. He also offers uncompensated professional services and activities for 10x Genomics, XRad Therapeutics, and the HPV Alliance and Anal Cancer Foundation non-profit organizations. Dr. Smith received travel support for fellow education from Intuitive Surgical (August 2015). He also served as a clinical advisor for Guardant Health (March 2019). He provides compensated professional services and activities for Johnson and Johnson, GlaxoSmithKline, Foundation Medicine, UroGen, Regeneron, and Vaniam Group. Dr. Garcia-Aguilar owns equity and received honoraria from Intuitive Surgical. Dr. Weiser owns intellectual property rights and provides compensated professional services for UpToDate. Dr. Cercek owns equity in Haystack Oncology, Inc. She also provides compensated professional services for AbbVie, Agenus, Amen, Daiichi Sankyo, GlaxoSmithKline, Illumina, Janssen Oncology, Inc., Merck & Co. Inc., Regeneron Pharmaceuticals, Inc., Roche Diagnostics Asia Pacific Ltd., and Seagen. Dr. Crane owns equity in Oncternal Therapeutics and provides compensated professional services for Trisalus Life Sciences. Dr. Segal provides compensated professional services for Puretech Health, Regeneron Pharmaceuticals Inc., and Agenus Inc. Dr. Reyngold provides compensated professional services and activities for Elekta and offer uncompensated professional services for the National Comprehensive Cancer Network. Dr. Yaeger provides compensated professional services and activities for Lilly Oncology, Mirati Therapeutics, Revolution Medicines, and Merck. Dr. Wu provides compensated professional services and activities for CivaTech Oncology, Inc., MORE Health, Inc., Nanovi A/S, and Simphotek, Inc. Dr. Connell provides compensated professional services and activities for Intera Oncology, Inc. Dr. Rao owns equity in Abbott Laboratories, CVS Health Corp., GlaxoSmithKline, Merck & Co., Novartis Pharmaceuticals Corporation, Procter & Gamble, Sandoz, Inc., Sanofi-Aventis U.S. LLC, and UnitedHealth Group. Drs. Aushev, Sharma, Malhotra, Rivero-Hinojosa, Jurdi, and Liu are employees of Natera, Inc. and receive equity in the company. All other authors do not have any conflicts of interest to disclose. Role of Natera: Natera, Inc. generated the personalized ctDNA assays and performed bioinformatic variant calling; Natera staff assisted with generation of Figs. 1B, 2, and 5. All clinical endpoints and outcome analyses were performed independently at MSK by institutional statisticians; patient management and outcome adjudication were conducted independent of the company.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-69984-y.