Sex disparity in acute myeloid leukaemia with FLT3 internal tandem duplication mutations: implications for prognosis

Incidence, molecular presentation and outcome of acute myeloid leukaemia (AML) are influenced by sex, but little attention has been directed at untangling sex‐related molecular and phenotypic differences between female and male patients. While increased incidence and poor risk are generally associated with a male phenotype, the poor prognostic FLT3 internal tandem duplication (FLT3‐ITD) mutation and co‐mutations with NPM1 and DNMT3A are overrepresented in female AML. Here, we have investigated the relationship between sex and FLT3‐ITD mutation status by comparing clinical data, mutational profiles, gene expression and ex vivo drug sensitivity in four cohorts: Beat AML, LAML‐TCGA and two independent HOVON/SAKK cohorts, comprising 1755 AML patients in total. We found prevalent sex‐associated molecular differences. Co‐occurrence of FLT3‐ITD, NPM1 and DNMT3A mutations was overrepresented in females, while males with FLT3‐ITDs were characterized by additional mutations in RNA splicing and epigenetic modifier genes. We observed diverging expression of multiple leukaemia‐associated genes as well as discrepant ex vivo drug responses, suggestive of discrete functional properties. Importantly, significant prognostication was observed only in female FLT3‐ITD‐mutated AML. Thus, we suggest optimization of FLT3‐ITD mutation status as a clinical tool in a sex‐adjusted manner and hypothesize that prognostication, prediction and development of therapeutic strategies in AML could be improved by including sex‐specific considerations.

Incidence, molecular presentation and outcome of acute myeloid leukaemia (AML) are influenced by sex, but little attention has been directed at untangling sex-related molecular and phenotypic differences between female and male patients. While increased incidence and poor risk are generally associated with a male phenotype, the poor prognostic FLT3 internal tandem duplication (FLT3-ITD) mutation and co-mutations with NPM1 and DNMT3A are overrepresented in female AML. Here, we have investigated the relationship between sex and FLT3-ITD mutation status by comparing clinical data, mutational profiles, gene expression and ex vivo drug sensitivity in four cohorts: Beat AML, LAML-TCGA and two independent HOVON/SAKK cohorts, comprising 1755 AML patients in total. We found prevalent sex-associated molecular differences. Co-occurrence of FLT3-ITD, NPM1 and DNMT3A mutations was overrepresented in females, while males with FLT3-ITDs were characterized by additional mutations in RNA splicing and epigenetic modifier genes. We observed diverging expression of multiple leukaemia-associated genes as well as discrepant ex vivo drug responses, suggestive of discrete functional properties. Importantly, significant prognostication was observed only in female FLT3-ITD-mutated AML. Thus, we suggest optimization of FLT3-ITD mutation status as a clinical tool in a sex-adjusted manner and hypothesize that prognostication, prediction and development of therapeutic strategies in AML could be improved by including sex-specific considerations.

Introduction
Sex influences regulatory mechanisms of the haematopoietic system as well as innate and adaptive immune responses [1,2]. Strong age-and sex-specific discrepancies in incidence of autoimmune conditions [3] and cancer [4], including haematological malignancies [5], suggest fundamental genetic and endocrine sex-related variability.
Androgens have been used to treat various bone marrow (BM) failure syndromes since the 1960s [6], and the presence of hormone receptors on haematopoietic cells, including malignant cell populations, has been recognized for decades [7]. Yet, little is known about the molecular mechanisms modulating haematopoiesis through sex hormone pathways or their contribution in haematopoietic malignancies. It has been indicated that sex hormone receptors significantly contribute in regulation of haematopoietic cell subsets, including stem and progenitor cells [8]. Sex disparity in acute myeloid leukaemia (AML) incidence is known, with a progressive male excess with increasing age [9]. It has also been shown that male AML patients have significantly inferior outcomes compared to females, both in adult and in paediatric AML [10]. Sex-specific mutational profiles in AML have also been described; FLT3-ITD, NPM1 and DNMT3A mutations are overrepresented in females [11,12], while mutations in RUNX1, ASXL1, SRSF2, STAG2, BCOR, U2AF1 and EZH2 are more prevalent in males [12]; and female overrepresentation among AML patients with co-occurrence of DNMT3A, NPM1 and FLT3-ITD mutations has also been reported [13].
Despite the demographic differences in survival and somatic mutation profiles, sex-specific considerations are currently not made in therapeutic assessment or clinical risk stratification. Among the mutations with reported sex-dependent discrepancies are internal tandem duplications (ITDs) of the FMS-like tyrosine kinase 3 (FLT3). FLT3-ITD is present in approximately 25% of the AML cases and is a negative prognostic marker that is integrated in standard risk stratification guidelines in AML as well as guiding FLT3-targeted therapy [11,14]. Yet, the association between sex and FLT3-ITD mutations has not been explored in depth. Here, we present results from genomic, functional and time-to-event-analyses of four well-annotated cohorts, including the Beat AML cohort [15], the LAML-TCGA cohort [16] and two independent HOVON/SAKK cohorts, in sum comprising 1755 AML patients. Cohort compositions with regard to sex and FLT3 mutation status are described in Table 1, and comparative analyses of sex differences related to clinical parameters are presented in Tables S1-S3.

Patient sample selection
We analysed four independent patient cohorts: Beat AML, LAML-TCGA, HOVON 1 and HOVON 2. The Beat AML and LAML-TCGA cohorts were used for descriptive analyses of somatic variant profiles. The Beat AML cohort was further investigated by differential gene expression (DGE) analyses and exploration of ex vivo drug response profiles. All four cohorts were included for time-to-event analyses. Data from Beat AML [15] and LAML-TCGA [16] are publicly available. All patients in the two HOVON cohorts provided written informed consent at trial inclusion.
The Beat AML sample cohort comprises 672 specimens from 562 individuals. We restricted our analysis to samples identified in the All.variants.csv file, downloaded from http://www.vizome.org/aml/geneset/ on the 01.11.2018. This included 608 samples from 519 individuals. At inclusion, 15 individuals had two samples acquired from different tissues. The sample with the lowest number of recurrent mutations was discarded. We filtered the remaining samples based on the variable 'dxAtSpecimenAcquisition' and retained samples annotated as 'Acute myeloid leukaemia (AML) and related precursor neoplasms', resulting in a total of 571 samples. For descriptive analysis of somatic variants, we included only the first sample when serial samples were present, resulting in 498 samples from 498 individuals. For DGE and drug response, we analysed 390 samples from 360 individuals and 359 samples from 322 individuals, respectively, including only samples with sample IDs overlapping with the exome sequencing data set (Fig. S1A). For outcome assessment, we restricted the analysis to diagnostic samples, denoted 'Initial Acute Leukemia Diagnosis', where survival data was present, resulting in 303 individuals (Fig. S1B). Notably, we constructed an extended FLT3-ITD annotation based on the combination of the consensus FLT3-ITD variable from the clinical summary table (Table S5-Clinical Summary) and Pindel call of FLT3-ITDs, as reported in the All.variants.csv file.
The HOVON 1 and HOVON 2 cohorts comprise (non-APL) newly diagnosed AML patients aged 15-80 years treated on various study protocols of the Dutch-Belgian Hemato-Oncology Cooperative Group (HOVON) and the Leukaemia Group of the Swiss Group for Clinical Cancer Research (SAKK) during the period 1987 to 2013. The sample selection analysed here was restricted to treatment na€ ıve non-APL AML patients with known FLT3-ITD status. This includes patients treated in the protocols HO04, HO04a [17], HO29 [18,19], HO42 [20,21], HO43 [22] and HO102 [23]. The studies are previously published, and sampling and data acquisition were performed as previously described [24,25]. Patients included in HO102 are referred to as HOVON2, while the remaining patients comprise HOVON1. Detailed information on the individual trials is available at http://www. hovon.nl. LAML-TCGA is a strongly selected sample cohort, composed to cover the major cytomorphologic and cytogenetic groups recognized in AML prior to 2013 [16]. The cohort comprises 200 de novo AML patients, including 92 females and 108 males ranging from 18 to 88 years. Due to its selective and unrepresentative composition, we have mainly used this cohort for integrated survival analyses. Single cohort comparative analyses of sex differences related to clinical parameters were not performed. The TCGA-AML data analysis is based exclusively on variables as presented in the file 'SuppTable01.xlsx', downloaded from https:// gdc.cancer.gov/node/876. For survival analysis, the variable 'OS months 3.31.12' is used.

Ethics
All clinical trials were approved by local ethics committees and performed in accordance with the Declaration of Helsinki.

Statistics
For comparison of continuous variables, we applied the nonparametric Wilcoxon rank sum test/Mann-Whitney test. The Fisher exact test was applied for comparison of categorical data. For DGE analysis, we log-transformed the CPM matrix provided in ' Table S9-Gene Counts CPM' by the formula: CPM (log2+0.1). Analyses were performed in accordance with the Linear Models for Microarray Data pipeline in the BIOCONDUCTOR R package (LIMMA version 3.38.3) [26]. For exploration of the drug sensitivity data from the Beat AML cohort, we analysed the area under the curve (AUC) values provided in 'Table S10-Drug Responses', applying categories from 'Table S11-Drug Families'. Time-to-event analyses were performed by generation of Kaplan-Meier survival curves and compared for differences using the log-rank test. For continuous variables, impact on outcome was evaluated by univariate Cox models followed by multivariate

Comparative genomic architecture
To provide a context for sex-related variant patterns, we compared the sex-related distribution of somatic variants annotated in the Beat AML cohort. We found that the number of somatic variants did not differ significantly across sexes, although male individuals tended to have higher numbers (Fig. S2A). Comparing the number of highly recurrently mutated genes (mutated in > 2% of the cohort), no significant differences were observed (Fig. S2B). There were no sex-related differences within the FLT3-ITD-mutated subgroup. We  (Table S4B) (Table S4C). This pattern is largely attributable to the excess of male samples annotated as 'transformed' (from a prior haematological malignancy). Excluding these cases, only the relationship between male sex and SRSF2  (Table S4D).
For comparison, the sex-related distribution of somatic variants in the LAML-TCGA cohort is presented in Table S5. Due to the selective composition of this cohort, integrated analyses of somatic variants were not performed.
The 28 recurrently mutated genes in the Beat AML cohort were subsequently categorized by gene product functional properties: DNA methylation, activating signalling, transcriptional regulation, tumour suppressor function, epigenetic modification, cohesion complex regulation and RNA splicing. We found that 14% of female vs 31% of male individuals had at least one mutation in one of the four RNA splicing genes:  (Table S6; Fig. S4).
We note that none of the sex discrepancies in somatic mutations of single genes reported as statistically significant here were significant by adjusted P-value. However, many of our findings, including the association of FLT3-ITD, NPM1 and DNMT3A mutations in females and RUNX1, ASXL1, SRSF2, STAG2, U2AF1 and EZH2 mutations in males, have been reported by others [11][12][13]. Furthermore, when categorizing individual genes by functional gene class, the findings are significant by adjusted P-value. This significance is driven by the same individual genes significant by nonadjusted P-value. Therefore, findings on the single gene level are also reported here.
We further explored the distribution of mutations of the frequently FLT3-ITD co-mutated genes DNMT3A and NPM1. As this information was available for all four cohorts, integrated analysis was performed. Mutation status of all three genes was known in 1624 cases. We found that significantly more females than males had a mutation in at least one of these three genes (F: 401/748, M: 397/876, P = 0.001). The combination of FLT3-ITD and NPM1 mutation without DNMT3A mutation (F: 56/748, M: 40/876, P = 0.015, adj P = 0.04) and the combination of FLT3-ITD, NPM1 and DNMT3A mutation (F: 70/74, M: 43/876, P = 0.0006, adj P = 0.004) were both significantly associated with females (Table S7; Fig. S5).
We subsequently compared the distribution of variant allele frequencies (VAF) of the genes mutated at least 10 times in the Beat AML cohort and found that VAF was significantly higher in male compared to female individuals in five genes. As expected, this included the X-linked genes STAG2, PHF6 and BCOR, but also the autosomal gene ASXL1. BCORL1, despite being X-bound, did not have significantly higher VAF in males. ZRSR2 was mutated exclusively in males (Fig. S6).  was not among the genes identified with significantly different VAF across the sexes. It has previously been shown that the allelic ratio (AR) of FLT3-ITD mutations has prognostic implications [28,29]. To investigate whether there were sex discrepancies in FLT3-ITD AR among these patients, we calculated the AR from the VAF (the approach is described in the figure legend of Fig. S7). We note that although there is no standardized approach to measure FLT3-ITD AR, it is commonly measured by DNA fragment analysis by capillary electrophoresis (CE) [30,31]. Here, we have used the VAF from next-generation sequencing (NGS) analyses to calculate AR, as it has previously been shown that there is high concordance between CE and NGS assays in measuring FLT3 mutational burden in AML patients [32]. As was the case for VAF, we did not find significant differences in AR between males and females in this cohort (Fig. S7).
To explore the functional relevance of genes identified as differentially expressed, we explored their interrelationship with disease outcome in the 220 individuals in the Beat AML cohort where survival and expression data were available. By univariate analysis, five of the 39 differentially expressed genes were identified as significantly associated with outcome (Table S9): NETO1, TMSB4Y, TTTY10, HMGA2 and FAT1. In multivariate analysis including these five transcripts, only NETO1 remained significant (Fig. S10). High expression of NETO1 significantly correlated with poor outcome (Fig. 2C). Stratifying the patients by FLT3-ITD status and sex, we found that high expression of NETO1 was significantly associated with poor prognosis in both FLT3-ITD-positive and FLT3-ITD-negative patients and in both sexes, although a stronger negative correlation was observed in the male subpopulation (Fig. S11). Analyses splitting the patients by both FLT3-ITD status and sex were not performed due to small effect sizes.

Drug sensitivity and resistance testing
To further explore the functional consequences of the sex-discrepant leukemic architecture, we assessed the variation in drug sensitivity profiles in the Beat AML cohort (Table S10). We focused on the 113 compounds tested in a minimum of 100 specimens, and samples overlapping with our previous analyses, resulting in 348 specimens from 311 individuals. 96/113 compounds were annotated by mechanism of action and categorized into one or more of 39 different groups (Fig. S12).
Comparing drug sensitivity across the 39 various groups between female and male individuals irrespective of ITD status, we identified a weak but significant relationship between RTK-RET inhibitors (1/39) and higher sensitivity in females. When comparing FLT3-ITD-positive cases, six inhibitor families differed significantly: PI3K-AKT-MTOR, PI3K-MTOR, RTK-ERBB, RTK-INSR-IGF1R, SYK and CAMK inhibitors all demonstrated higher sensitivity in male samples (Fig. S13). Comparing female and male FLT3-ITD-negative specimens, only MEK inhibitors were significantly different, and more potent in female samples.
When examining individual compounds, we identified five compounds with significantly sex-discrepant potency, irrespective of FLT3-ITD status: Females were more sensitive to AT7519, palbociclib and quizartinib (AC220), while males were more sensitive to cediranib (AZD2171) and pazopanib (GW786034). Further, we identified three compounds that were significantly more potent in female FLT3-ITD-negative samples, including AT7519, palbociclib and trametinib (GSK1120212). Importantly, we identified 14 To assess whether variation in gene expression correlated with variation in drug sensitivity, we further examined the pairwise relationships of the differentially expressed genes and the compounds with sexdiscrepant potency in FLT3-ITD-positive samples. We found that sensitivity to NVP-ADW742 correlated with increasing gene expression of 7/13 RNA transcripts identified as upregulated in female FLT3-ITDpositive AML. Conversely, increasing GLI2 expression correlated with reduced sensitivity to 4/14 drugs that were less potent in female FLT3-ITD-positive AML (Fig. 3B).

Survival
Finally, we investigated the relationships between sex and the prognostic strength of FLT3-ITD mutation status on all cohorts combined as well as separately (Table S12a- Despite FLT3-ITD status being a recognized negative prognostic marker in AML, we found a significant association to poor outcome only in the HOVON 1 cohort (Fig. S15). When overall survival was analysed in the cohorts combined (n = 1560), we found as expected that FLT3-ITD mutation status was associated with significantly lower survival (Fig. 4A). However, when survival analysis for all cohorts combined was split by sex, the significant prognostic association of FLT3-ITD remained only in the female subpopulation (Fig. 4B). Analysing the cohorts independently, the same observation was apparent; FLT3-ITD was significantly associated with poor outcome only in female patients in the HOVON1 cohort, and the same trend was observed in the HOVON2, Beat AML and LAML-TCGA cohorts, although not significant (Figs S15 and S16; Table S12a-e).
A recent report suggests that the prognostic strength of FLT3-ITD is age-dependent [11]. Therefore, we performed age-adjusted survival analyses, dichotomizing the population using a cut-off of 60 years. Similarly, we found a significant association of FLT3-ITD mutations with poor overall survival in the younger population only (> 60 years) (Fig. 4C). However, when the analysis was split by sex, we found an association between younger age and poor overall survival in the female subpopulation only. In the male subpopulation, there was no significant difference in overall survival between FLT3-wt-and FLT3-ITD-mutated patients in the young (> 60 years) nor older (≥ 60 years) patients (Fig. 4D).

Discussion
In this work, we demonstrated sex disparity in somatic variant composition, gene expression and ex vivo drug response patterns in FLT3-ITD-mutated AML cases across four well-characterized AML cohorts. FLT3-ITD mutation status is integrated in standard risk stratification guidelines in AML. Yet, in the datasets we explored, FLT3-ITD mutation status separated the disease outcome only in female AML. This may indicate that the prognostic utility of FLT3-ITD mutation status is different across sexes. Of note, the prognostic value of FLT3-ITD mutations is reportedly influenced by the mutational burden. Several studies have shown that a high FLT3-ITD AR is linked to poor prognosis [28,29]. Furthermore, it has also been shown that AML patients with a high FLT3-ITD AR lacking NPM1 mutations have a worse prognosis [33]. In the latter study, an excess of males was reported in this subgroup, although no statistically significant sex disparity was observed. In our study, we did not identify FLT3-ITD VAF nor AR as significantly different across sexes, suggesting that a difference in the mutational burden of FLT3 is likely not the cause of the sex discrepancies observed in outcome. Another plausible explanation is related to discrepant distribution of age, disease presentation pattern and poor risk molecular features. We identified significant differences in the distribution of comutations within the FLT3-ITD-positive subgroup. While males more frequently present with somatic variants in epigenetic modifier genes and/or RNA splicing genes, co-mutations of FLT3-ITD, NPM1 and DNMT3A were overrepresented in females. The mutation pattern characterizing female patients in this subgroup is previously shown to be associated with adverse prognosis in FLT3-ITD-mutated AML [34]. Importantly, significant differences were also observed within the FLT3-ITD-negative subgroup, characterized by an abundance of WT1 mutations in female specimens contrasted by overrepresentation of mutations in RUNX1, SRSF2, U2AF1, ZRSR2 and EZH2 in the male subgroup. These differences are most pronounced in the Beat AML cohort, where the mutated genes dominating the male FLT3-ITD-negative subpopulation are associated with poor outcome and with myelodysplastic syndrome (MDS) and secondary AML [35]. Whether this relationship represents inclusion asymmetry or a natural distribution is unknown. A characterization of the population-based Swedish Acute Leukemia Registry, however, suggested that AML with an antecedent haematological disease is more frequent in males [36]. Similar to AML, there is female excess of MDS among younger individuals, and men with MDS have comparably inferior outcome [9]. Additionally, mutations in several genes overrepresented in males in the Beat AML cohort, including SRSF2 and U2AF1, are reportedly overrepresented in male MDS [37].
Sex disparity in AML demography is well known, with a progressive male excess with increasing age [9]. As the male population is generally older, one could speculate whether older age is a contributing factor to the inferior outcome of male patients, within both the FLT3-wt-and FLT3-ITD-mutated subgroups. Specifically, age differences could potentially result in sex disparities of treatment (i.e. proportion of patients fit to receive intensive chemotherapy and/or allogeneic haematopoietic stem cell transplantation), which might be an important confounder for this study. In the HOVON cohorts, however, we found no significant differences between males and females in the proportion of transplanted patients (Table S2). Due to insufficient annotation across the cohorts, additional effects of treatment were unfortunately not evaluated in this study, and the impact of such therefore remains uncertain. Interestingly, it was recently shown that the prognostic impact of FLT3-ITD mutations is also agedependent, with poor overall survival observed in younger FLT3-ITD-mutated (< 60 years) AML patients, but not within the older population (60-74 years) [11]. We corroborate this observation in ageadjusted survival analyses in this study. Interestingly, when splitting the analysis by sex, this finding was significant only within the female subpopulation, further emphasizing the poor prognostic association between female sex and FLT3-ITD mutations.
The sex-specific age distribution characterizing AML demography suggests that cohort composition is an important confounder. The age composition of the Beat AML cohort resembles the reported demographic distribution of AML far better than the strongly selected LAML-TCGA cohort, in part explaining the lack of sex-biased mutations in this cohort. One could argue that age-matching and/or use of a populationbased sample selection is the optimal condition for comparison. What remains, however, is to identify the cell-intrinsic and cell-extrinsic mechanisms underlying the sex disparity in AML incidence and molecular presentation. We hypothesize a significant contribution of sex-specific leukaemia-host interactions related to disease development. Mutations are stochastic events, and there are thus no apparent reasons to suggest that male and female haematopoietic stem cells acquire discrete mutations. What is plausible, however, is that sex is an important contextual contributor in determining the translational effect in comparative fitness of novel gene variants. Interestingly, one of the very first papers characterizing the FLT3 receptor suggested FLT3-mediated regulatory function not only in the haematopoietic compartment but also in the gonads, placenta and brain [38]; tissues characterized by sexual dimorphism. This suggests that downstream effects of FLT3 signalling may be influenced by sex variation. This hypothesis has been substantiated by studies demonstrating functional relevance of FLT3 expression in germinal tissue [39,40] and by investigation of the oestrogen receptor in FLT3-positive haematopoietic cells, progenitor cells and mature cellular subsets like dendritic cells [8,41]. Together, these observations suggest inter-regulatory pathways between sex steroid receptors and FLT3.
The DGE analysis showed distinct expression of leukaemogenesis-associated genes, suggesting functionally relevant cell-intrinsic sex-dependent differences. Of the genes more highly expressed in female FLT3-ITDpositive AML, the hedgehog signalling mediator GLI2 has been associated with FLT3-ITD-mutated leukaemia [42], while HOXB-AS3 is reportedly upregulated in NPM1-mutated AML, and has been implied to contribute in regulation of AML cell-cycle progression [43]. NETO1, however, the only differentially expressed gene with prognostic impact in this cohort, is predominantly studied in the central nervous system where it is found to regulate kainate receptor signalling [44], and a role in haematopoiesis has to our knowledge not been investigated. The significance of this finding is therefore uncertain. Of the genes more highly expressed in male FLT3-ITD-positive AML, FAT1 was previously shown to be somatically mutated in FLT3-ITD-positive AML and significantly in combination with NPM1 and DNMT3A [45]. This study also showed that FAT1 exerted tumour suppressor activity specifically in FLT3-ITD-positive AML, perhaps providing a partial explanation for the lack of prognostic impact of FLT3-ITD in this subpopulation.
The sum of the ex vivo drug response patterns and clinical outcome analysis suggests that response to therapeutic intervention in FLT3-ITD-mutated AML may be influenced by sex. Interestingly, this is in line with observations from the phase III clinical trials RATIFY and QuANTUM-R, treating FLT3-ITDpositive de novo AML with FLT3-targeting inhibitors (midostaurin and quizartinib, respectively), both reporting significant survival benefit in the male subpopulation only [14,46]. Unfortunately, neither of these trials were powered to address potential sex differences in therapeutic response, and additional trials would be necessary to validate this observation. Our ex vivo drug sensitivity analyses did not, however, show a similar sex difference in the sensitivity to FLT3 inhibitors. It has previously been shown that FLT3-ITD dependency in ex vivo assays is contingent on the availability of growth factors, and thus, the response to FLT3-targeting drugs is permissive of the culture conditions [47]. This could provide a partial explanation for this discrepancy. However, the long incubation time is also an important confounder for ex vivo drug screens, and conclusions regarding the translational value of such data should be drawn with caution. Notably, sex-related ex vivo drug response patterns have also been reported in the study of the Cancer Cell Line Encyclopedia [48] and in clinical trials of the tyrosine kinase inhibitor sunitinib in other malignancies [49]. Thus, these observations may be of great importance for the field of precision haematooncology, as novel targeted therapies may benefit female and male individuals differently. Although men and women in an unstratified AML population have similar prospects, this picture may be significantly different within discrete molecularly defined strata.
The impact of sex on disease presentation and outcome is likely a composite picture involving multiple factors, including societal influence and socialization (e.g. risk behaviours, occupation, stress and inclination to seek health care). Most importantly, however, sex is associated with major physiological differences, the impact of which has been insufficiently explored in haematology and in medicine in general. The field of cardiovascular disease is a noteworthy exception, where sex disparity is now widely recognized and where sex considerations are becoming integrated into research, diagnostics and clinical disease management [50,51]. Similarly, we hypothesize that including sexspecific considerations in preclinical and clinical research in AML could advance the pathophysiological understanding, which could ultimately lead to more precise prognostication and improved therapeutic options for these patients.

Conclusions
Assessment of mutation status is becoming an integrated part of the diagnostic characterization of adult AML patients. This information is subsequently incorporated into risk stratification models and clinical decision-making. FLT3-ITD mutation status is well  as a biomarker in AML. However, important questions remain regarding its optimal application and utility. Our observations suggest that FLT3-ITD mutation status could be optimized as a clinical tool in a sex-adjusted manner. Furthermore, we suggest that sex-specific considerations should be considered in preclinical and clinical experimental design and biomarker analyses in AML. Sex should represent an independent stratification factor when randomizing to clinical trials, and be systematically included when analysing and reporting on clinical data in AML. We hypothesize that addressing sex-related regulation of molecularly defined subgroups of AML could advance pathophysiological understanding, perhaps ultimately revealing new therapeutic possibilities.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. Overview of the Beat AML sample selection analysed. Fig. S2. Distribution of somatic variants in the Beat AML sample selection. Fig. S3. Sex-specific distribution of somatic mutations in the Beat AML cohort. Fig. S4. Age-and sex distribution of samples with somatic mutations, presented by gene class. Fig. S5. Sex-specific distribution of comutations of FLT3-ITD, NPM1 and DNMT3A in the HOVON1, HOVON2, Beat AML and LAML-TCGA cohorts. Fig. S6. Sex-specific distribution of VAF of mutations detected in a minimum of 10 samples in the Beat AML sample selection. Fig. S7. Sex-specific distribution of FLT3-ITD allelic ratio. Fig. S8. Expression level of genes identified as differentially expressed between male and female FLT3-ITDpositive samples. identified as differentially expressed between male and female FLT3-ITD-positive samples. Fig. S10. Cox Proportional-Hazards model including the five genes where expression was identified as significantly correlated with outcome by univariate analysis. Fig. S11. Kaplan-Meier curves comparing patients with high and low expression of NETO1, split by FLT3-ITD mutation status and sex. Fig. S12. Overview of drugs and drug classes. Fig. S13. Comparison of drug sensitivity scores between FLT3-ITD-mutated male and female samples, presented by drug class. Fig. S14. Comparison of drug sensitivity scores of drugs identified with significantly different potency in male and female FLT3-ITD-mutated samples.  (D) Somatic mutations in the Beat AML sample cohort (not transformed). Table S5. Somatic mutations in the LAML-TCGA cohort (all). Table S6. Mutations in Beat AML cohort categorized by gene product function. Table S7. FLT3-ITD, DNMT3A and NPM1 comutations in the Beat AML, LAML-TCGA HOVON1 and HOVON2 cohorts. Table S8. Beat AML cohort -Differential gene expression analysis. Table S9. Univariate Cox regression analysis of differentially expressed genes in the Beat AML cohort. Table S10. Number of FLT3-ITD and FLT3-ITD wt male and female samples screened for each drug in the in drug screen. Table S11. P-values for comparison of drug sensitivity. Table S12.