Alterations of cohesin complex genes in acute myeloid leukemia: differential co-mutations, clinical presentation and impact on outcome

Functional perturbations of the cohesin complex with subsequent changes in chromatin structure and replication are reported in a multitude of cancers including acute myeloid leukemia (AML). Mutations of its STAG2 subunit may predict unfavorable risk as recognized by the 2022 European Leukemia Net recommendations, but the underlying evidence is limited by small sample sizes and conflicting observations regarding clinical outcomes, as well as scarce information on other cohesion complex subunits. We retrospectively analyzed data from a multi-center cohort of 1615 intensively treated AML patients and identified distinct co-mutational patters for mutations of STAG2, which were associated with normal karyotypes (NK) and concomitant mutations in IDH2, RUNX1, BCOR, ASXL1, and SRSF2. Mutated RAD21 was associated with NK, mutated EZH2, KRAS, CBL, and NPM1. Patients harboring mutated STAG2 were older and presented with decreased white blood cell, bone marrow and peripheral blood blast counts. Overall, neither mutated STAG2, RAD21, SMC1A nor SMC3 displayed any significant, independent effect on clinical outcomes defined as complete remission, event-free, relapse-free or overall survival. However, we found almost complete mutual exclusivity of genetic alterations of individual cohesin subunits. This mutual exclusivity may be the basis for therapeutic strategies via synthetic lethality in cohesin mutated AML.


INTRODUCTION
Acute myeloid leukemia (AML) is a genetically complex disease. The recently revised WHO classification acknowledges a variety of genetically defined alterations which constitute distinct disease entities [1]. Correspondingly, our understanding of myeloid neoplasms moves away from somewhat arbitrary numerical counts of bone marrow blasts and toward an appreciation of genetic drivers of disease as is acknowledged in the revised International Consensus Criteria [2]. On this basis, the recently revised European Leukemia Net (ELN) recommendations broaden the spectrum of clinically relevant genetic alterations with respect to individual patient risk warranting treatment that is adjusted to individual low-, intermediate-, and high-risk molecular alterations and cytogenetics [3]. In these updated definitions, mutations of the cohesin subunit SA-2 (STAG2) are recognized as a defining alteration of AML with myelodysplasia-related gene mutations (in absence of other defining alterations) irrespective of prior presence of myelodysplastic neoplasms [3]. Further, mutated STAG2 is defined as a prognostic marker of high-risk (if not cooccurring with favorable risk AML subtypes) incentivizing intensive treatment and, potentially, allogeneic hematopoietic stem cell transplantation (HCT) [3].
While initially inactivating mutations of the cohesin complex were thought to promote carcinogenesis via aberrant segregation of sister chromatids and subsequent aneuploidy, especially recent findings of altered cohesin subunits in commonly euploid myeloid malignancies (with the exception of myeloid leukemias associated with Down Syndrome [18]) hint at more complex mechanisms of pathogenesis [4]. For instance, the finding that cohesin-CCCTFbinding-factor sites are frequently altered in cancer cells underlines the cohesins' function in three-dimensional chromosome organization as a key component of carcinogenesis in a variety of neoplasms [25][26][27]. Further, inactivation of cohesin subunits may result in a complete collapse of topologically-associating-domain (TAD) structure [28][29][30]. Additionally, mutated cohesin subunits appear to play a role in stemness and differentiation in hematopoietic stem cells (HSC). Inactivation of STAG2, RAD21, SMC1A, and SMC3 was found to promote stem cell self-renewal in human and mouse HSCs in vitro and subunit-specific knockout mice were found to bear changes in erythroid and myeloid differentiation mimicking myeloproliferative disorders similar to early human leukemogenesis [31][32][33]. This results in a proliferation advantage hinting at a key function of the cohesin complex in regulating cellular differentiation [31][32][33].
Taken together, these findings suggest a multi-facetted role of the cohesin complex and its individual subunits in human carcinogenesis. The impact of individually altered cohesin subunits on patient outcome in AML is unclear as previous studies have suggested unfavorable [17], favorable [19] as well as no prognostic impact [20]. Therefore, we aimed to identify distinct co-mutational patters for mutations of STAG2 and other proteins of the cohesion complex that help to predict clinical outcomes in a large multicentric cohort of adult patients with AML.

Data set and definitions
We retrospectively analyzed a cohort of 1615 adult AML patients that were treated in previously reported multicenter trials (AML96(ref. [ [38]. All studies were approved by the Institutional Review Board of the Technical University Dresden (EK 98032010). Complete remission (CR) and survival times including eventfree (EFS), relapse-free (RFS), and overall survival (OS) were defined according to ELN2022 criteria [3]. Patients were retrospectively re-stratified into ELN2022 risk groups [3]. Since patients from earlier clinical trials were only re-stratified according to ELN2022 criteria, study accrual was not influence based on ELN risk. A summary of individual study protocols is provided in table S1. AML was defined as de novo when no prior malignancy and no prior treatment with chemo-and/or radiotherapy was reported. AML was defined as secondary (sAML) when prior myeloid neoplasms were reported, and therapy-associated (tAML) when prior exposure to chemo-and/or radiotherapy was reported.

Molecular analysis and cytogenetic analysis
Pre-treatment peripheral blood or bone marrow aspirates were screened for genetic alterations using next-generation sequencing (NGS) with the TruSight Myeloid Sequencing Panel (Illumina, San Diego, CA, USA) covering 54 genes (table S2) that are associated with myeloid neoplasms including full coding exons for SMC1A, RAD21, and STAG2 and relevant exons (10, 13, 19, 23, 25, and 28) for SMC3 according to the manufacturer's recommendations as previously reported [39,40]. DNA was extracted using the DNeasy blood and tissue kit (Qiagen, Hilden, Germany) and quantified with the NanoDrop spectrophotometer. Pooled samples were sequenced paired-end (150 bp PE) on a NextSeq NGS-instrument (Illumina). Sequence data alignment of demultiplexed FastQ files, variant calling and filtering was performed with the Sequence Pilot software package (JSI medical systems GmbH, Ettenheim, Germany) with default settings and a 5% variant allele frequency (VAF) mutation calling cut-off. Human genome build HG19 was used as reference genome for mapping algorithms. Dichotomization of dominant and subclonal (or secondary) mutations was performed by comparing VAFs of detected mutations with VAFs of co-mutated driver variants. For resolution of putative subclonal mutations a minimum difference of 10% VAF was applied. For cytogenetic analysis, standard techniques for chromosome banding and fluorescencein-situ-hybridization (FISH) were used.

Statistical analysis
Statistical analysis was performed using STATA BE 17.0 (Stata Corp, College Station, TX, USA). All tests were carried out as two-sided tests. Statistical significance was determined using a significance level α of 0.05. Fisher's exact test was used to compare categorical variables. Normality was assessed using the Shapiro-Wilk test. If the assumption of normality was met, continuous variables between two groups were analyzed using the two-sided unpaired t-test. If the assumption of normality was violated, continuous variables between two groups were analyzed using the Wilcoxon rank sum test. With regard to outcome variables, patients were analyzed on a complete case basis. Univariate analysis was carried out using logistic regression to obtain odds ratios (OR). Time-to-event analysis was performed using Cox-proportional hazard models to obtain hazard ratios (HR) as well as the Kaplan-Meier-method and the log-rank-test. For survival times, OR and HR, 95%-confidence-intervals (95%-CI) are reported. Multivariable models were adjusted for ELN2022 categories and age. In the case of AML with mutated STAG2, additional adjustments were performed in multivariable analysis for frequently co-mutated genes with an established impact on patient outcome according to ELN2022 definitions [3]. Median follow-up time was calculated using the reverse Kaplan-Meier method [41].
AML with mutated RAD21 shows a distinct co-mutational pattern while RAD21 mutational status does not influence outcome The second most common alteration was RAD21 which was detected in 51 patients (3.2%), again mostly being nonsense (n = 35, 68.6%) rather than missense mutations (n = 16, 31.4%, Fig. 2A). Further, alterations of RAD21 were mostly dominant (n = 36, 70.6%). With respect to baseline patient characteristics, patients harboring RAD21 mutations showed significantly increased LDH (median 705.0 U/l vs. 440.2 U/l, p < 0.001) upon initial diagnosis.   AML with either mutated SMC3 or SMC1A does not differ from SMC3-or SMC1A-wild-type AML regarding clinical presentation, co-mutations, and outcome Twenty-five patients (1.5%) harbored alterations in SMC1A, while mutated SMC3 was found in 20 patients (1.2%). Alterations in both SMC1A and SMC3 were only detected as missense mutations (Fig. 3A, Fig. 4A) and the majority was found in the dominant clone (SMC1A: 60.0%, SMC3: 55.0%). There were neither differences in baseline clinical characteristics between patients with SMC1Amutated vs. SMC1A-wild-type AML (Table S12) nor between patients with SMC3-mutated vs. SMC3-wild-type AML (Table S13). With respect to co-mutations, patients harboring mutated SMC1A showed significantly increased rates of t(8;21) (20.0% vs. 3.5%, p = 0.002) while no other associations were found (Fig. 3B, Table  S14). Patients harboring SMC3 mutations showed significantly increased co-mutations of NPM1 (65.0% vs. 30.8%, p = 0.003) while no difference between mutated or wild-type SMC3 was found for other alterations (Fig. 4B, Table S15). CR rate did not differ neither for patients with SMC1A mutations nor patients with SMC3 mutations when compared to wildtype patients. With regard to survival times, again no difference was found both for patients with SMC1A-mutated vs. SMC1A-wildtype AML (Fig. 3C-E) as well as patients with SMC3 mutations when compared to their wildtype counterparts ( Fig. 4C-E). Further analysis with respect to clonality (dominant vs. subclonal) of the specific mutations did not show any differences for CR rates, EFS, RFS, or OS both for AML with mutated SMC1A and SMC3.

DISCUSSION
In our retrospective multi-center cohort study in 1615 intensively treated AML patients, we were able to ascertain distinct patterns of changes in cohesin complex genes, identify their association with other recurrent genetic alterations and clinical presentation as well as confirm their mutual exclusivity which may serve as a target for therapeutic approaches. We confirm that mutations in the genes of the cohesin complex are recurrent genetic events in AML with a reported frequency between 5.9-13.0%(ref. [15,19,20,23,24,42]) which is in line with our cohort where 11.4% of patients harbored an alteration of cohesin complex genes.
In accordance with previous studies(ref. [15,19,20,23,24,42]), we found these mutations to be mutually exclusive with the exception of one patient bearing both mutated STAG2 and SMC3. The inactivation of more than one subunit of the   For detailed information on co-mutations and results of individual significance tests, see Tab. S10. Survival analysis using the Kaplan-Meier method and logrank test for event-free (C), relapse-free (D) and overall survival (E) differentiating between RAD21 -mutated and RAD21 -wildtype AML.
cohesin complex may result in structural collapse of topologically associated domains which may explain the mutual exclusivity of these gene alterations [4]. STAG2 may form the exception as it has a functional homologue in STAG1 that can potentially compensate for its malfunction in three-dimensional genome organization [29]. In the early days of research into the role of the cohesin complex in carcinogenesis, it has been hypothesized that its malfunction may lead to aberrant segregation of sister chromatids and consequently aneuploidy as a major driver in neoplastic transformation [8]. However, especially studies investigating mutations of the cohesin complex in myeloid neoplasms have refuted this claim since these alterations are commonly found in AML with euploid karyotypes [20,24,42]. Correspondingly, the rate of patients with normal karyotypes in our cohort was significantly increased while the rate of complex aberrant karyotypes was significantly decreased for patients with cohesinmutated AML. These findings suggest alternate contributions of cohesin in carcinogenesis rather than mere aneuploidy and chromosomal instability. Alterations of cohesin subunit genes have both been described as early and late events in leukemogenesis [17,20,24,31,32,43] suggesting a passenger rather than a driver function. However, in our cohort, the majority of cohesion complex mutations were detected in dominant clonal constellations, pointing at a potential role as an early event during AML initiation. Likewise, cohesin plays an important role in regulating the stemness and pluripotency of stem cells [31][32][33]. Thus, an interplay of alterations of cohesin genes with other genetic events in driver genes such as NPM1 likely promotes malignant transformation. Mutated NPM1 has been associated with alterations of cohesin complex genes [19,20]. An interaction of cohesin proteins with NPM1 could be mediated by CCCTC-binding factor -a transcription factor that regulates tumor suppressor lociwhich has been shown to bind and interact with both [44,45], potentially contributing to their role in stem cell self-renewal [46,47]. In our cohort, we found mutated SMC3 and RAD21 to be associated with mutated NPM1 while NPM1 mutations were less frequently associated with mutated STAG2. In comparison to their wildtype counterparts, patients with mutated STAG2 more frequently also had mutated IDH2, TET2, BCOR, ASXL1, SRSF2, and ZRSR2. Further, patients with alterations of RAD21 showed increased rates of co-occurring mutations in EZH2, KRAS, and CBL besides NPM1. An association of cohesin mutations with mutated TET2, ASXL1, BCOR, and EZH2 has previously been reported [24,42], however, it is important to note that different subunits of the cohesin complex show different co-mutational patterns.
The prognostic impact of cohesin mutations in AML has been unclear as studies are not only limited but also report conflicting results. Tsai et al. [19] report increased OS and disease-free survival for patients with cohesin mutations, which was confirmed by multivariable analysis in a cohort of 391 patients with de novo AML. . For detailed information on co-mutations and results of individual significance tests, see Tab. S13. Survival analysis using the Kaplan-Meier method and logrank test for event-free (C), relapse-free (D) and overall survival (E) differentiating between SMC1A -mutated and SMC1A -wildtype AML.
In contrast, Thol et al. [20] found no impact of cohesin mutations on CR rate, RFS, and OS in a cohort of 389 intensively treated AML patients. In MDS, Thota et al. [24] reported decreased OS for patients with cohesin mutations, especially in STAG2-mutated MDS for patients who survived beyond 12 months. Commonly, previous studies were limited in sample size, often ranging between 300 and 600 patients. In our comparatively large cohort of 1615 intensively treated AML patients, we did not find a significant impact of any gene alterations of cohesin subunits on CR rate, EFS, RFS, or OS. The recently revised ELN2022 recommendations [3] introduce mutated STAG2 as a prognostic marker of adverse risk (if no markers of favorable risk are co-occuring). Multivariable models adjusted for mutation status of BCOR, ASXL1, and RUNX1, which were more prevalent in STAG2-mutated AML patients, demonstrated no independent impact of mutated STAG2 on patient outcome while these co-mutations had varying individual prognostic impact. Several reports agree that STAG2 mutations are associated with sAML, and thus, as a part of corresponding compound attributes they are associated with the overall adverse impact of sAML on outcome [17,43]. However, these mutations contribute only a minor part of this compound attributes. According to our observations such an adverse effect on outcome cannot be verified for the presence of STAG2 mutations per se. While the cohesin complex undoubtably plays a role in leukemogenesis, given the ambiguity of existing reports on cohesin's (and STAG2's) role in AML prognostication [19,20,24] caution may be warranted with respect to determining patient risk and ultimately treatment allocation. Nevertheless, it should be acknowledged that our study is limited by the fact that results are only available for intensively treated patients. The extent to which the reported results are also transferable to patients which receive less intensive regimens or targeted therapy remains to be evaluated.
While the prognostic impact of cohesin alterations in AML remains elusive, their co-mutational pattern with respect to mutual exclusivity may make them a viable option for targeted therapy. Mutually exclusive gene alterations may be utilized therapeutically via synthetic lethality [48]. If the alteration of one mutated gene provides a cancerous cell with a survival advantage as long as a second gene remains unaltered, the alteration or inhibition of the second gene or its gene product may confer apoptosis specifically in cells carrying the initial alteration [49,50]. Synthetic lethality via inhibition of mediators of replication fork stability such as poly ADP-ribose polymerase (PARP) has been demonstrated in BRCA-mutated breast, ovarian, pancreatic, and prostate cancer [51]. The functional homologues STAG1 and 2 offer the possibility for a synthetically lethal therapeutic strategy via PARP inhibition. In glioblastoma cells, Bailey et al. [52] have demonstrated that mutated STAG2 significantly increases the sensitivity to PARP inhibition. Further, Black et al. [53] found STAG2-deficient leukemic cells to bear a significantly higher  Fig. 4 Distribution, co-mutational spectrum and survival analysis for AML with mutated SMC3. Graphic representation of the domain structure of SMC3 and positions of SMC3 mutations in 20 AML patients (A). Mutations of SMC3 are categorized by function (missense=blue, termination=red), clonal rank (dominant=red/subclonal=blue), number of mutations, and variant allele fraction, and associated co-mutations (B). For detailed information on co-mutations and results of individual significance tests, see Tab. S14. Survival analysis using the Kaplan-Meier method and logrank test for event-free (C), relapse-free (D) and overall survival (E) differentiating between SMC3-mutated and SMC3wildtype AML.
susceptibility to treatment with talazoparib. Currently, a phase 1 study is ongoing investigating the safety and efficacy of talazoparib for cohesin-mutated AML and MDS with excessive blasts (NCT03974217) [54].
In summary, we report distinct co-mutational and clinical patterns for mutated STAG2 and RAD21 in a large sample of AML patients while mutated SMC3 and SMC1A lacked such patterns. However, no cohesin subunit-including mutated STAG2 that was recently added to the ELN2022 criteria as a marker of adverse risk-showed any impact on patient outcome regarding the achievement of CR, EFS, RFS, or OS. While we did not find a prognostic impact of cohesin alterations in AML, their mutual exclusivity may make them a potential target for therapeutic approaches based on synthetic lethality.

DATA AVAILABILITY
The datasets generated during and analyzed during the current study are available in the Kaggle repository, https://doi.org/10.34740/KAGGLE/DSV/4816451.