AUTHORS
Sushil Selvarajan, M.D, D.M, Kuo-Kai Chin, MD2, Xin Wang, MD, MSCE3, Baher Krayem, MD4, Andriy Derkach, PhD5, Ali Kamali, PhD6, Gregory Goldgof, MD, PHD, MS7, Eytan M Stein I, MD8
AFFILIATIONS
1 Christian Medical College, Vellore, Vellore, Tamil Nadu, India, Memorial Sloan Kettering Cancer Center, New York, New York, United States
2 Memorial Sloan Kettering Cancer Center, New York, New York, United States
3 Memorial Sloan Kettering Cancer Center, New York, New York, United States
4 Memorial Sloan Kettering Cancer Center, New York, New York, United States, Rambam Health Care Campus, Haifa, Israel, Haifa, Israel
5 Memorial Sloan Kettering Cancer Center, New York, New York, United States
6 Memorial Sloan Kettering Cancer Center, New York, New York, United States
7 Memorial Sloan Kettering Cancer Center, New York, New York, United States
8 Memorial Sloan Kettering Cancer Center, New York, New York, United States
ABSTRACT
Background: The cohesin complex comprises SMC1A and SMC3, bridged by RAD21 and stabilized by STAG1 or STAG2, while PDS5B regulates cohesin dynamics. Pathogenic or likely pathogenic alterations in these genes mark presumed cohesin dysfunction, but whether they define a single AML entity or distinct molecular subsets remains unclear.
Methods: We retrospectively reviewed adults diagnosed with AML from January 2020 through March 2026 who had a pathogenic or likely pathogenic alterations involving STAG1, STAG2, RAD21, SMC1A, SMC3, or PDS5B. For this study, we refer to AML harboring such alterations as cohesin complex gene (CCG) dysfunctional AML. Clinical data were curated using Hematologic Multimodal Analytics Platform (HeMAP), a validated in-house Python pipeline applying sequential large language model prompts to extract structured variables from unstructured electronic health record data. The database was manually reviewed and reconciled by study investigators before analysis. Patients were grouped by CCG alteration detection at treatment-naive AML diagnosis versus alterations first detected at relapse. Associations were assessed by Fisher exact tests; OS by Kaplan-Meier and univariable and multivariable Cox models including age, AML ontogeny, treatment intensity, STAG2, NRAS, and TP53.
Results: Among 173 patients with CCG-dysfunctional AML, 128 had a CCG alteration detected on treatment-naive sequencing at AML diagnosis and 45 first had an alteration detected at relapse; the two groups were analyzed separately. In those with CCG alterations detected at diagnosis, the median age was 70.5 years, 91 patients (71.1%) had secondary and/or therapy-related AML and 90 (70.3%) had ELN 2022 adverse risk disease. CCG alterations involved STAG2 in 77 patients, RAD21 in 25, SMC1A in 20, SMC3 in 13, PDS5B in 9, and STAG1 in 1. Fifteen patients (11.7%) had alterations involving more than one CCG.
In the cohort of patients detected at diagnosis, STAG2 mutant AML was associated with a distinct clinical and co-mutational profile. Compared with non-STAG2 CCG dysfunctional AML, STAG2 mutated disease was more frequently secondary AML (64.9% vs 39.2%; P=.006) and was enriched for additional myelodysplasia-related mutations, including ASXL1 (58.4% vs 13.7%; P<.001) and SRSF2 (42.9% vs 13.7%; P<.001). In contrast, non-STAG2 disease was more frequently associated with NPM1 (27.5% vs 7.8%; P=.005) and FLT3-ITD mutations (13.7% vs 1.3%; P=.007).
First-line therapy was HMA/venetoclax in 59 patients and intensive chemotherapy in 50; 61 (47.7%) underwent allogeneic transplantation. After a median follow-up of 34.9 months, 54 patients in the at-diagnosis cohort had died. Median OS was 45.6 months, with estimated 1 and 2 year OS rates of 81.1% and 56.4%, respectively. NRAS and TP53 mutations were present in 27 (21.1%) and 13 patients (10.2%), respectively. On univariate analysis, older age, NRAS mutations, and TP53 mutations were associated with inferior OS. In multivariate analysis, NRAS (HR, 3.59; 95% CI, 1.91-6.75; P<.001) and TP53 (HR, 3.09; 95% CI, 1.34-7.13; P=.008) remained independently associated with inferior OS.
We separately analyzed the 45 patients in whom a CCG alteration was first detected at relapse and identified a distinct clinical and genomic profile. Compared with the at-diagnosis cohort, these patients more frequently had therapy-related AML (55.6% vs 33.6%; P=.013). Importantly, CCG alterations were less frequently STAG2 (33.3% vs 60.2%; P=.003) and more frequently SMC3 (28.9% vs 10.2%; P=.006). At AML diagnosis, they were also less likely to harbor ASXL1 (11.1% vs 40.6%; P<.001), SRSF2 (11.1% vs 31.3%; P=.010), or TET2 mutations (6.7% vs 25.0%; P=.009), indicating a distinct baseline genomic context in the group with CCG alterations first identified at relapse.
Conclusions: To our knowledge, this is the largest single-institution cohort of CCG dysfunctional AML reported to date and identifies distinct clinical and genomic subsets. STAG2 mutated disease was enriched for secondary AML, ASXL1, and SRSF2, whereas non-STAG2 disease more often harbored NPM1 and FLT3-ITD. NRAS and TP53 independently predicted inferior OS. Genomic lanscapes of CCG alterations detected at diagnosis and at relapse were associated with distinct co-mutational profiles, supporting context-specific heterogeneity within CCG dysfunctional AML.
DISCLOSURES
Sushil Selvarajan, M.D, D.M: No relevant conflicts of interest to declare.