Clinical utility and tissue concordance of circulating tumor DNA in pancreatic ductal adenocarcinoma

Abstract

Background

The utility of circulating tumor DNA (ctDNA) in addressing challenges of molecular tissue profiling and complementing next-generation sequencing is undefined in pancreas ductal adenocarcinoma. The objective of this study was to assess ctDNA detection rates by stage, disease burden, and metastasis patterns; compare overall survival between ctDNA-positive and ctDNA-negative patients cases; and determine concordance between ctDNA and matched-tissue biopsies.

Methods

Patients with pancreas ductal adenocarcinoma who had undergone Next Generation Sequencing by the MSK-ACCESS (Memorial Sloan Kettering – Analysis of Circulating cfDNA to Evaluate Somatic Status) ctDNA assay between 2019 and 2022 were included. Clinical and survival data were abstracted from a prospectively maintained clinical database.

Results

A total of 414 patients with pancreas ductal adenocarcinoma: 28% stage I-II, 21% stage III, 51% stage IV. ctDNA detection was highest among patients with advanced disease: 75% stage IV, 38% stage III, 34% stage I-II disease. For stage IV, ctDNA was more frequently detected in patients with at least 2 organs involved vs with less than 2 organs involved (76% vs 38%, P = .025). Higher rates of ctDNA detection were observed in patients with liver metastases vs without (82% vs 52%, P < .001). In the untreated stage IV cohort (n = 120), median overall survival was 10 months for those with detectable ctDNA (95% CI = 6.9 to 14 months) vs 19 months (95% CI = 13 months to not reached) for those with undetectable ctDNA (P = .1). Concordance between ctDNA and matched tissue next-generation sequencing was lower in untreated stage I-III disease, but high for untreated stage IV pancreas ductal adenocarcinoma, including a critical success index of 93.1% of KRAS variants.

Conclusion

ctDNA is a promising tool in the detection of somatic variants in pancreas ductal adenocarcinoma. Concordance between ctDNA and tissue is high for patients with untreated metastatic disease, notably for detection of KRAS variants.

Introduction

Pancreas ductal adenocarcinoma is a highly recalcitrant malignancy, with a poor prognosis but an incrementally improving 5-year survival rate of 13%.¹ Genetic testing is playing an increasingly important role in improving outcomes, and somatic and germline testing for patients with pancreas ductal adenocarcinoma are endorsed by national and international guidelines.^2,3 In small, mostly retrospective nonrandomized series, patients treated with molecularly matched targeted therapeutics have improved outcomes, particularly patients with homologous recombination deficiency variants such as BRCA1 and 2.^4-9 Given the rapid expanse¹⁰ of targeted therapeutics, identifying actionable variants is of increasing therapeutic relevance in pancreas ductal adenocarcinoma, particularly as we enter the era of KRAS-directed approaches.^11-15

However, there are notable challenges with molecular profiling specific to pancreas ductal adenocarcinoma.¹⁶ Next-generation sequencing is unable to detect low-frequency variants below 10%-20% tumor cellularity, which presents a unique problem in pancreas ductal adenocarcinoma given the desmoplastic, necrotic, and inflammatory features of tumor stroma—often with a dearth of epithelial malignant cells.^14,17 Furthermore, acquiring tumor tissue from patients with pancreas ductal adenocarcinoma can be technically challenging, particularly for patients with locally advanced tumors or with low tumor burden in sites not easily accessible for biopsy.^18,19 This challenge of obtaining viable tissue for next-generation sequencing is also notable in posttreatment biopsies, limiting insights into genomic resistance mechanisms.^20,21

The molecular profiling of circulating tumor DNA (ctDNA) captures genomic variants identified within the patient’s tumor(s)²² and may assist in overcoming some of the aforementioned challenges. Previous studies in other tumor types have demonstrated high concordance between ctDNA and tumor tissue in identifying relevant genomic variants.²³ Furthermore, the median time to ctDNA sequencing results is generally faster than tissue-based sequencing, which can guide timely treatment selection.^24,25 ctDNA molecular profiling and serial ctDNA data demonstrate the potential to identify targetable variants, detect minimal residual disease (MRD) or emerging recurrence, provide insight into tumor response dynamics, and help characterize resistance mechanisms.^26-34

Prior studies have supported the use of ctDNA for identifying actionable variants in pancreas ductal adenocarcinoma.³⁵ However, there is a limited number of large datasets to inform the sensitivity and feasibility of ctDNA use in pancreas ductal adenocarcinoma and a scarcity of studies evaluating the concordance of tissue and ctDNA next-generation sequencing (Table S1). The objective of this study was to evaluate the role of ctDNA in a large cohort of patients with pancreas ductal adenocarcinoma, with a specific focus on defining ctDNA and somatic tissue concordance, describing ctDNA correlations with disease status and prognosis, and understanding the utility of ctDNA for treatment selection, particularly in respect to KRAS.³⁶

Methods

Clinical and genomic data collection

Institutional databases were queried with institutional review board approval (12–245, 19–242) to identify patients with pancreas ductal adenocarcinoma who had at least 1 ctDNA sample taken at Memorial Sloan Kettering (MSK) between August 2019 and July 2022. MSK-ACCESS (MSK-Analysis of Circulating CfDNA to Evaluate Somatic Status), a New York State–approved hybridization and deep-sequencing 129-gene assay was used for ctDNA analysis.³⁷ Detection of ctDNA (assigned “ctDNA detected”) was defined as the identification of at least 1 variant, copy number, or structural variant by MSK-ACCESS. Matched tissue ctDNA samples were defined as samples taken within 1 month with no therapy in between. Demographic, clinical, pathologic, and outcomes data were collected from medical record review.

Variant interpretation

MSK-IMPACT (Memorial Sloan Kettering – Integrated Mutation Profiling of Actionable Cancer Targets) was used to detect somatic variants in 468 (version 6) or 505 (version 7, latest) genes.³⁸ Formalin-fixed, paraffin-embedded neoplastic sample was compared with the patient’s matched buffy coat to allow proper filtering of germline variants. To pass filtering, a variant must have had a minimum depth of 20, alternate allele count of 8, and minimum variant allele fraction of 2% for a hotspot or 5% for a nonhotspot variant. Analysis of cell-free DNA was performed using MSK-ACCESS (version 1), a deep-sequencing next-generation sequencing assay designed to detect somatic variants in 129 genes.³⁷ To maintain a matched design similar to that of MSK-IMPACT, DNA extracted from buffy coat helped filter out germline events. MSK-ACCESS employs a de novo threshold and a tumor-informed criteria to call variants. For interpretation without support seen in a prior tissue or plasma sample, 3 duplex fragments are required for a hotspot and 5 duplex fragments are required for a nonhotspot variant. If there is prior support for an interpretation, the threshold is lowered to a single duplex fragment or 2 simplex fragments.

Within our cohort of 131 matched pairs of tissue and ctDNA testing, further genotyping was also performed on ctDNA samples using GetBaseCountsMultiSample v.1.2.5 (https://github.com/mskcc/GetBaseCountsMultiSample) to identify subthreshold empirical calls (simplex AD < 2).

Biostatistical analyses

Categorical data were summarized using descriptive statistics. Mean variant allele fraction was calculated as the sum of sequence reads of a variant called in MSK-ACCESS divided by the sum of the overall coverage at that locus. Associations between ctDNA positivity and the different disease subgroups (eg, burden of disease, sites of disease) were assessed using the Fisher exact test. Differences in biomarkers (eg, carcinoembryonic antigen [CEA], carbohydrate antigen 19-9, and alkaline phosphatase) between ctDNA detected vs ctDNA not detected patients were examined using the Wilcoxon rank sum test. Overall agreement between the gene variants identified in ctDNA compared with those identified in tissue was calculated as (true positive + true negatives)/(true or false positives + true or false negatives). In addition to the overall agreement, we reported the critical success index defined as true positives/(true positives + false negatives + false positives) that focused on mutation-positive patients and removed true negative patients, because in the presence of rarely mutated genes, agreement may be overestimated. Overall survival was calculated from time of ctDNA blood draw until date of last follow-up or death and estimated using the Kaplan–Meier methods. Statistical analyses were performed using R version 4.3.2 (R Foundation for Statistical computing, Vienna, Austria). All P values were 2-sided, and P values less than .05 were considered to indicate statistical significance.

Results

Cohort demographics

We identified 414 patients between 2019 and 2022 with pancreas ductal adenocarcinoma and at least 1 ctDNA sample. At time of ctDNA testing, 203 (49%) had stage I-III pancreas ductal adenocarcinoma (including patients with resected or unresected tumors) and 211 (51%) had stage IV disease (Figure 1). The cohort was 50% female (n = 208) and 78% White (n = 324); 32% (n = 133) of patients were former smokers, and 5% (n = 22) were current smokers. Detailed demographics and baseline tumor characteristics are further summarized in Table 1.

Figure 1.

CONSORT diagram. Abbreviation: MSK-ACCESS = Memorial Sloan Kettering – analysis of circulating cfDNA to evaluate somatic status.

Table 1.

Cohort demographics summary

Characteristics	Whole cohort, No. (%) (n = 414)	Stage I-II, No. (%) (n = 115)	Stage III, No. (%) (n = 88)	Stage IV, No. (%) (n = 211)
Male	206 (50)	54 (47)	42 (48)	110 (52)
Race
Asian	44 (11)	10 (9.3)	12 (14)	22 (11)
Black	20 (5.1)	4 (3.7)	6 (7.1)	10 (5.0)
Other	4 (1.0)	2 (1.9)	0 (0)	2 (1.0)
Unknown	22	8	3	11
White	324 (83)	91 (85)	67 (79)	166 (83)
History of smoking
Never smoker	257 (62)	71 (62)	56 (64)	130 (62)
Current smoker	22 (5.3)	8 (7.0)	3 (3.4)	11 (5.2)
Former smoker	133 (32)	35 (31)	29 (33)	69 (33)
Unknown	2	1	0	1
Personal history of cancer	92 (22)	27 (23)	12 (14)	53 (25)
History of pancreatitis	23 (5.6)	13 (11)	5 (5.7)	5 (2.4)
History of diabetes	127 (31)	35 (30)	19 (22)	73 (35)
First-degree relative with pancreas cancer	34 (8.3)	14 (12)	4 (4.5)	16 (7.6)
Unknown	2	2	0	0
Location of pancreas cancer
Head	220 (53)	75 (65)	56 (64)	89 (42)
Body	105 (25)	22 (19)	26 (30)	57 (27)
Tail	96 (23)	19 (17)	6 (6.8)	71 (34)
Treatment status at time of circulating tumor DNA
Treated	227 (55)	85 (74)	50 (57)	92 (43)
Treatment naïve	187 (44)	30 (26)	38 (43)	120 (57)
Circulating tumor DNA status
Detected	184 (44)	76 (66)	55 (63)	53 (25)
Not detected	230 (56)	39 (44)	33 (38)	158 (75)

Untreated stage IV cohort (n = 120)

Patients without any prior chemotherapy, radiation, or surgery in any disease setting were considered untreated. The frequency of detectable ctDNA increased with the number of organ systems involved in the untreated stage IV cohort (Figure 2). For patients with untreated stage IV and 2-3 sites of disease, 88 (87%) had detectable ctDNA. For patients with 4 or more sites of disease (n = 18), 100% had detectable ctDNA. Patients with liver-only metastases (n = 43) had a 91% ctDNA detection rate. Although limited by small sample numbers, those with lung-only metastases (n = 4) had a ctDNA detection rate of 100%, and those with peritoneal-only metastases (n = 9) had a ctDNA detection rate of 82% (Figure 2). Statistically significantly higher carbohydrate antigen 19-9, CEA, and alkaline phosphatase levels were observed in patients with detectable ctDNA (Table S2). Among patients with detectable ctDNA before commencing therapy (n = 115), the median overall survival was shorter: 10 months (95% CI = 6.9 to 14 months) compared with 19 months (95% CI = 13 months to not reached) for those who had undetectable ctDNA (n = 15), although not meeting statistical significance potentially owing to the small number of patients without detectable ctDNA (P = .10; Figure 3, A).

Figure 2.

Rates of ctDNA detection. A) Rates of ctDNA detection stratified by number of sites of organ involvement (full cohort: n = 414). B) Rates of ctDNA detection stratified by number of sites of organ involvement (stage IV untreated cohort: n = 120). C) Rates of ctDNA detection stratified by organs of involvement (full cohort: n = 414). D) Rates of ctDNA detection stratified by organs of involvement (stage IV untreated cohort: n = 120). Abbreviation: ctDNA = circulating tumor DNA.

Figure 3.

A) Overall survival for untreated stage IV group by ctDNA status (n = 120). B) Overall survival for untreated locally advanced unresectable group by ctDNA status (n = 38). C) Overall survival for resection group by ctDNA status. Abbreviation: ctDNA = circulating tumor DNA.

Untreated stage III cohort (n = 38)

Of the 88 patients with stage III disease, 38 (43%) had ctDNA testing in the treatment-naïve setting. ctDNA was detected in only 13 (34%) with untreated stage III disease (compared with 88% [105 of 120] with untreated stage IV disease). There was no statistically significant difference in overall survival between those with detectable ctDNA (n = 13) vs those without detectable ctDNA (n = 75) (15 months, 95% CI = 15 months to not reached, vs 22 months, 95% CI = 22 months to not reached; P = 04) (Figure 3, B).

Postresection cohort (n = 29)

Of the 115 patients with stage I-II disease, 29 had a ctDNA sample taken after surgery and prior to commencement of adjuvant therapy. For the 5 patients with detectable ctDNA prior to initiation of adjuvant chemotherapy, the median overall survival was 30 months (95% CI = 33 months to not reached), and in the group without detectable ctDNA before commencing adjuvant chemotherapy (n = 24), median overall survival was not reached (95% CI = 28 to not reached, P = .4) (Figure 3, C). We again acknowledge the small sample sizes in this group and recommend that these results be interpreted with caution.

Rates of ctDNA detection: entire cohort

In the entire cohort (n = 414), 230 (56%) had at least 1 detectable variant on MSK-ACCESS, which we have designated “ctDNA detected.” For patients with stage I-II disease, 39 (34%) had detectable ctDNA; for stage III, 33 (38%); and for stage IV, 158 (75%). Detection of variants increased with the number of organ systems involved at time of ctDNA sampling. In the stage IV cohort of those with 2-3 sites of disease (n = 170), 124 (73%) had detectable ctDNA; for those with 4 sites of disease (n = 22), 20 (91%) had detectable ctDNA; and for those with 5 or more sites of disease (n = 11), 100% had detectable ctDNA (Figure 2). Among all patients with stage IV disease (including patients with newly diagnosed, untreated stage IV pancreas ductal adenocarcinoma), the presence of 2 or more organs of disease involvement at the time of ctDNA sample acquisition was more likely to yield a detectable ctDNA sample than those with less than 2 organs of involvement (76% vs 38%, P = .025; Table 2).

Table 2.

Association between ctDNA detected vs not detected based on disease burden, sites of organ involvement, and serum biomarkers in patients with stage IV disease

Characteristics	Whole cohort (n = 211)	ctDNA not detected (n = 53)	ctDNA detected (n = 158)	P
Number of organs involved				.025
<2	8	5	3
≥2	203	48	155
Liver metastases				<.001
No liver metastases	52	25 (48%)	27 (52%)
Presence of liver metastases	159	28 (18%)	131 (82%)
Lung metastases				.13
No lung metastases	163	45 (28%)	118 (72%)
Presence of lung metastases	48	8 (17%)	40 (83%)
Peritoneal metastases				.2
No peritoneal metastases	148	33 (22%)	115 (78%)
Presence of peritoneal metastases	63	20 (32%)	43 (68%)
Alkaline phosphatase at time of ctDNA collection
Median (range)	122 (4-1242)	95 (41-337)	143 (4-1242)	<.001
Unknown	29	11	18
Carbohydrate antigen 19-9 at time of ctDNA collection, U/mL				<.001
Median (range)	896 (0-142 745)	128 (1-12 253)	2286 (0-142 745)
Unknown	16	6	10
Carcinoembryonic antigen at time of ctDNA collection, ng/mL				<.001
Median (range)	8 (1-2020)	5 (1-51)	10 (1-2020)
Unknown	50	14	36

Abbreviation: ctDNA = circulating tumor DNA.

Patients with liver metastases (including those with liver metastases only or liver and other sites of metastases) were observed to have detectable ctDNA more frequently compared with patients without liver metastases at the time of ctDNA sampling (82% vs 52%, P < .001) (Table 2 and Figure 2). Among patients with metastasis to the liver only (n = 81), 63 (78%) had detectable ctDNA. For patients with metastasis to the peritoneum only (n = 16), 10 (63%) had detectable ctDNA. For patients with metastasis to the lung only (n = 9), 5 (56%) had detectable ctDNA. For patients with liver metastasis and another metastatic site (n = 159), 131 (82%) had a detectable variant on ctDNA. Across all disease stages, higher carbohydrate antigen 19-9, CEA, and alkaline phosphatase levels were observed in patients with detectable ctDNA compared with those without detectable ctDNA (Table S2).

ctDNA and tissue concordance analyses and somatic mutational profile

For the purposes of concordance analyses, we focused on patients who were treatment-naïve at the time of ctDNA testing. Of the 68 patients with untreated stage I-III disease, 29 patients had a matched tissue next-generation sequencing and ctDNA biopsy. In this group, the critical index concordance was 39.3% for KRAS variants, 39.1% for TP53, 33.3% for CDKN2A, and 12.5% for SMAD4 (Figure 4, A). Of the 120 patients with untreated stage IV disease, 62 had a tissue biopsy for next-generation sequencing at the same time as ctDNA sample. In this group with matched samples, the critical success index between ctDNA and tissue was 93.1% for KRAS variants, 84.3% for TP53, 89.5% for CDKN2A, and 63.6% for SMAD4 (Figure 4, B).

Figure 4.

A) Agreement between somatic tissue and ctDNA biopsies for the untreated stage I-III cohort. B) Agreement between somatic tissue and ctDNA biopsies for untreated stage IV cohort. Abbreviation: ctDNA = circulating tumor DNA.

The mutational profile across all patients with detectable ctDNA (n = 230) is summarized in Figure 5, A. The most observed variants were in KRAS (68%), TP53 (53%), CDKN2A (12%), and SMAD4 (10%), including patients with early stage pancreas ductal adenocarcinoma, accounting for the lower rates per gene alteration than would be expected in tissue. The concordance between ctDNA and tissue-based next-generation sequencing for the 20 most common gene variants is represented in Figure 5, B (excluding patients without any gene variants identified).

Figure 5.

A) OncoPrint of the most frequently altered genes identified by ctDNA. B) Prevalence of genomic variants identified by ctDNA compared with matched tissue. Abbreviation: ctDNA = circulating tumor DNA.

KRAS variant detection and allele spectrum

In the entire cohort, 170 of 414 (41%) patients had a KRAS variant detected by ctDNA, including G12D (n = 80, 47%), G12V (n = 53, 315), and G12R (n = 25, 15%) (Figure S1, A). The KRAS variant allele fraction tended to increase with stage: for stage I pancreas ductal adenocarcinoma, the mean KRAS variant allele fraction was 0.038; for stage II, 0.024; for stage III, 0.066; and for stage IV, 0.091 (Figure S1, B). In 131 patients with matched ctDNA and tissue next-generation sequencing samples, the distribution of KRAS variants was numerically similar. In the somatic tissue samples, the distribution of KRAS variants was 46% G12D, 27% G12V, 17% G12R and in the ctDNA samples the distribution was 56% G12D, 25% G12V, and 13% G12R (Figure S1, C).

Discussion

Herein, we present a comprehensive analysis of a large cohort of patients with pancreas ductal adenocarcinoma and ctDNA testing, including detailed tissue-ctDNA concordance data, thorough annotation of KRAS status, and mature survival outcomes. Our data demonstrate that ctDNA is concordant with somatic tissue testing for patients with stage IV pancreas ductal adenocarcinoma.

In our cohort, ctDNA driver oncogene results were highly concordant with tissue next-generation sequencing results in patients with stage IV pancreas ductal adenocarcinoma. The presence of tumor next-generation sequencing variant knowledge lowers the threshold for variant calling on ctDNA, and thus the ctDNA in this setting acts as a tumor-informed instead of tumor-agnostic assay. Comparing with other tumor types, in the MIRROR breast cancer study, the concordance rate of all variants was 97.2%.³⁹ In a study of gastrointestinal cancers (including 25 pancreas ductal adenocarcinoma), concordance rates of 96% for KRAS amplification and for KRAS G12V were observed for ctDNA and tumor results.⁴⁰ For patients in our study with untreated stage IV disease and a matched tissue sample (n = 62), the critical index concordance (excluding true negatives) between ctDNA and tissue was 93.1% for KRAS variants. As we enter the era of KRAS-directed therapies, including allele-specific inhibitor therapies, expedited testing results for KRAS variants will inform therapeutic selection.^{11,  41,42} One patient in our cohort for whom tumor-based sequencing had not previously been feasible had a KRAS G12C variant identified by ctDNA at the time of disease progression, facilitating enrollment on a clinical trial of KRAS G12C inhibitor.

In contrast, for patients with stage I-III pancreas ductal adenocarcinoma and a matched tissue sample (n = 29), lower rates of agreement between the variants observed by ctDNA and tissue next-generation sequencing were observed. The critical index concordance was 39.3% for KRAS variants. The lower concordance in stage I-III is likely due to lower tumor cellularity and known limitations of tissue acquisition for primary pancreas ductal adenocarcinoma and decreased tumor shedding, which is already low in pancreas ductal adenocarcinoma compared with other malignancies such as colorectal cancer.^43,44 Advanced tumors have been observed to shed more ctDNA compared with premalignant and earlier-staged cancers.^19,45 In addition to lower rates of tumor DNA shedding, other factors may also contribute to lower concordance in early stage tumors, such as differences in the degree of vascularization and its effect on ctDNA detection rates. Unfortunately, viable tissue acquisition is most challenging for patients with localized pancreas ductal adenocarcinoma, making ctDNA testing theoretically particularly valuable in this disease setting. Although ctDNA testing provides insight in some cases, improved sensitivity is needed, and somatic tissue next-generation sequencing in combination with ctDNA testing is optimal.⁴⁶

We observed that ctDNA positivity rates differed depending on the sites of metastases. Patients with liver metastases were observed to have detectable ctDNA more frequently compared with patients without liver metastases at the time of ctDNA sampling (82% vs 52%, P < .001; Figure 2; Table S2), although this observation was not made for lung or peritoneal metastases. For patients with peritoneal metastases only, detection may be lower because of the plasma-peritoneal barrier. In a study of 279 patients with various gastrointestinal cancers, Sullivan et al.⁴⁷ observed that peritoneal carcinomatosis was associated with statistically significantly lower ctDNA levels compared with visceral metastases. In colorectal cancer, peritoneal-only and lung-only metastases were associated with lower ctDNA detection.⁴⁸ When using ctDNA in patients without liver metastases, acknowledgment of the inherent limitations of testing in these circumstances is key.

Aside from the site of metastasis, rates of ctDNA detection also differed based on the number of organs of disease involvement. In patients with stage IV pancreas ductal adenocarcinoma, ctDNA was detected more often in those with 2 or more organs of disease involvement at the time of ctDNA sampling (76% vs 38%, P = .025). As detailed in Figure 2, 100% of patients with 4 or more sites of disease (n = 18) had detectable ctDNA, indicating that irrespective of organs of metastasis, the higher the burden of metastatic disease, the more likely a ctDNA biopsy is to detect a variant. This assertion is further supported by findings that statistically significantly higher median carbohydrate antigen 19-9 and CEA levels were noted in patients with stage IV disease and detectable ctDNA (Table 2), again reflecting a higher burden of disease.

To ascertain the prognostic value of ctDNA in our cohort, we focused on key groups where the role for ctDNA is likely to be most useful clinically: patients with untreated stage IV disease; patients with untreated stage III disease; and patients who underwent tumor resection. In patients with untreated stage IV disease, the median overall survival for those with detectable ctDNA was 10 months (95% CI = 6.9 to 14 months) compared with 19 months for patients with nondetectable ctDNA variants, although statistical significance was limited by the small number of patients with undetected ctDNA (95% CI = 13 months to not reached, P = .10). Detection of ctDNA also suggested inferior median overall survival in the surgically resected setting compared with patients with undetected ctDNA in this setting (30 months vs not reached, P = .4), although this estimate is limited by modest numbers, including only 5 patients in the ctDNA-detected group. The estimation of survival based on ctDNA detection was exploratory in nature and not the primary objective of this study, however the prognostic value of ctDNA among these groups is consistent with previous cohorts. Botrus et al.³⁵ observed that detectable ctDNA correlated with short overall survival in advanced pancreas ductal adenocarcinoma. Similar, Botta et al.³¹ observed that detectable ctDNA correlated with shorter disease-free survival in patients with resected pancreas ductal adenocarcinoma. In a large meta-analysis, Lee et al.⁴⁹ observed that detectable ctDNA was associated with shorter relapse-free survival and overall survival. Several other cohorts have suggested similar outcomes (Table S1). Of note, in our patients with untreated stage III pancreas ductal adenocarcinoma (n = 38), the median overall survival of patients with detectable ctDNA (n = 13) was not reached (95% CI = 15 months to not reached) compared with 22 months (95% CI = 14 months to not reached, P = .40) for patients without detectable ctDNA (n = 25). This trend is unexpected and is likely explained by limitations of small numbers.

ctDNA kinetics can be useful for monitoring disease status, serving as a surrogate for proximate response evaluation. Previous studies have also reported the utility of ctDNA in assessing treatment response in pancreas ductal adenocarcinoma. Kruger et al.⁵⁰ observed that for patients with pancreas ductal adenocarcinoma and serial ctDNA samples (n = 24), increases in KRAS variant allele fraction indicated disease progression on gemcitabine-based regimens correctly with 83% sensitivity and 100% specificity. Similarly, Lapin et al.⁵¹ observed that patients with pancreas ductal adenocarcinoma and detectable ctDNA after 2 months of therapy had a statistically significantly shorter median progression-free survival compared with patients without (0 vs 3.4 months, P = .002). Although evaluation of ctDNA kinetics was not a key focus of our study, several patients did undergo serial ctDNA collection, which impacted management. One example is a patient with Lynch syndrome and a microsatellite instable pancreas ductal adenocarcinoma (germline pathogenic mutS homolog 2 [MSH2]) variant), where an initial ctDNA test demonstrated 14 variants prior to commencing immune checkpoint blockade. The patient had an excellent clinical, radiographic, and biochemical response to immunotherapy but with marked immune–related toxicity. Follow-up testing after 5 months of immune checkpoint blockade demonstrated no detectable ctDNA, which was sustained on serial sampling. ctDNA clearance in this scenario helped inform the decision to safely withhold further immune checkpoint blockade.

We acknowledge several limitations in our study design. First, this study was conducted at a single, tertiary medical center and is retrospective in nature. Second, we acknowledge the relative lack of diversity of our patient cohort with respect to demographics. Third, although patient numbers are relatively large, the cohort is heterogeneous in terms of disease stage, disease burden, and the timing of ctDNA sampling, meaning that the numbers of patients included in some subgroups are small.

At this time, the detailed clinical and genomic annotation required for the concordance analyses is not readily available in external cohorts. However, a commercial version of MSK-ACCESS is in use by SOPHiA Genetics at several institutions globally. Previously, our group demonstrated MSK-ACCESS to be comparable with a different ctDNA assay (Resolution Biosciences Platform), as was validated by the New York State Department of Health. Notably, the automated integration of previously available somatic tissue results from MSK-IMPACT with MSK-ACCESS enables a higher sensitivity.^37,52 Moreover, the use of matched white blood cell sequencing further enables a higher specificity for somatic and tumor-specific alterations and allows for the identification of clonal hematopoiesis as distinct from true ctDNA.³⁷ Although the size of the panels and performance characteristics of MSK-ACCESS are similar to some widely used commercially available panels, the authors do not have data available currently comparing MSK-ACCESS with any other commercial panels.

To summarize, our data indicate that ctDNA has a role in the management of patients with pancreas ductal adenocarcinoma, given the need for timely, accurate molecular profiling of tumors. Our data support the use of ctDNA in pancreas ductal adenocarcinoma as a biomarker for disease burden, monitoring disease status in patients, as an important prognosticator, and has utility in subsets of patients, for example patients with treatment-naïve stage IV disease, liver metastases, higher burden of disease, and elevated tumor markers. In addition, targeted molecular profiling of ctDNA is highly concordant with somatic tissue testing for patients with stage IV disease, in the detection of KRAS variants. Our results support the use of ctDNA as a complementary adjunct to tissue-based sequencing.

Author contributions

Fergus Keane (Conceptualization, Data curation, Formal analysis, Writing—original draft, Writing—review & editing), Lily V. Saadat (Conceptualization, Data curation, Formal analysis, Writing—original draft, Writing—review & editing), Catherine A. O’Connor (Conceptualization, Data curation, Formal analysis, Writing—original draft, Writing—review & editing), Joanne F. Chou (Conceptualization, Data curation, Formal analysis, Methodology, Resources, Software, Supervision, Validation, Writing—review & editing), Anita S. Bowman (Conceptualization, Data curation, Formal analysis, Methodology, Software, Writing—review & editing), Fei Xu (Conceptualization, Data curation, Formal analysis, Software, Writing—review & editing), Fionnuala Crowley (Data curation, Formal analysis, Investigation, Writing—review & editing), Neha Debnath (Data curation, Formal analysis, Writing—review & editing), Joshua D. Schoenfeld (Conceptualization, Data curation, Methodology, Writing—review & editing), Anupriya Singhal (Data curation, Investigation, Writing—review & editing), Drew Moss (Data curation, Formal analysis, Investigation, Writing—original draft, Writing—review & editing), Darren Cowzer (Data curation, Formal analysis, Investigation, Writing—review & editing), Emily Harrold (Data curation, Formal analysis, Investigation, Writing—review & editing), Wungki Park (Conceptualization, Methodology, Supervision, Writing—review & editing), Anna M. Varghese (Conceptualization, Supervision, Writing—review & editing), Fiyinfolu Balogun (Conceptualization, Investigation, Methodology, Supervision, Writing—review & editing), Kenneth H. Yu (Conceptualization, Methodology, Supervision, Writing—review & editing), Alice Zervoudakis (Methodology, Supervision, Writing—review & editing), Marinela Capanu (Conceptualization, Formal analysis, Methodology, Resources, Software, Supervision, Writing—review & editing), Michael F. Berger (Conceptualization, Investigation, Methodology, Supervision, Writing—review & editing), Alice C. Wei (Conceptualization, Investigation, Methodology, Supervision, Writing—review & editing), and Angela Rose Brannon (Conceptualization, Formal analysis, Investigation, Methodology, Resources, Software, Supervision, Validation, Writing—review & editing), and Eileen M. O’Reilly (Conceptualization, Formal analysis, Investigation, Methodology, Resources, Supervision, Writing—original draft, Writing—review & editing).

Supplementary material

Supplementary material is available at JNCI: Journal of the National Cancer Institute online.

Funding

This study has been supported by the following funding: Cancer Center Support Grant/Core Grant P30 CA008748 NCI/NIH P50 CA257881-01A1.

Conflict of interest

A.B. reports the following relationships: IP for MSK-ACCESS licensed to SOPHiA Genetics. W.P. discloses the following relationships; grant and research support from Astellas, Merck, NIH/NCI, Parker Institute for Cancer Immunotherapy, and Break Through Cancer; consultancy: Astellas; research funding to MSK: Merck, Astellas, and Miracogen; funding: NIH Pancreas SPORE (1P50CA257881-01A1), NIH K12 CA184746 Paul Calabresi Career Development Award for Clinical Oncology, and MSK Parker Institute for Cancer Immunotherapy (PICI) Pilot Grant Award; Breakthrough Cancer—Conquering KRAS for Pancreatic Cancer, Society of MSKCC Research Grant Award—iBTC, TimIOs: Society of Immunotherapy for Cancer (SITC), Merck Investigator Studies Program: Investigator-Initiated Trial. A.M.V. discloses the following relationships: consulting/advisory role: Roche (immediate family member), Lilly (immediate family member), Astra Zeneca (immediate family member), PAIGE (immediate family member); research funding: Illumina (immediate family member), Lilly, Verastem, BioMed Valley Discoveries, Bristol-Myers Squibb, Silenseed. F.B. discloses a research/career development award from Winn-AACR supported by BMSF. K.H.Y. discloses serving in an advisory role for Ipsen and receives research funding from Halozyme, Bristol-Myers Squibb, and Ipsen. M.B. reports the following relationships: consulting fees (Eli Lilly, AstraZeneca, Paige.AI) and intellectual property rights (SOPHiA Genetics). A.R.B. discloses the following relationships: stock holding in J&J, intellectual property rights in SOPHiA Genetics. E.M.O. reports the following relationships: research funding to institution: Genentech/Roche, BioNTech, AstraZeneca, Arcus, Elicio, Parker Institute, Digestive Care, NIH Pancreas SPORE (1P50CA257881-01A1). Other: associate editor JCO, senior editor: AACR Cancer Research and Communications; SU2C, Breakthrough Cancer. Consulting/DSMB: Arcus, Alligator, Agenus, BioNTech, Ipsen, Merck, Novartis, Revolution Medicines, Syros, Tango, Leap Therapeutics, Astellas, BMS, Fibrogen, Merus, Ikena; AbbVie (spouse).

Data availability

The data underlying this study can be accessed via the following link: https://www.cbioportal.org/study/summary? id=pancreas_ctdna_msk_2025