RUNX1 mutation has no prognostic significance in paediatric AML: a retrospective study of the AML-BFM study group

In acute myeloid leukaemia (AML) RUNX1 mutation is characterised by certain clinicopathological features with poor prognosis and adverse risk by the European LeukemiaNet recommendation. Though initially considered as provisional category, the recent World Health Organisation (WHO) classification of 2022 removed RUNX1-mutated AML from the unique entity. However, the significance of RUNX1 mutation in paediatric AML remains unclear. We retrospectively analysed a German cohort of 488 paediatric patients with de novo AML, enroled in the AMLR12 or AMLR17 registry of the AML-BFM Study Group (Essen, Germany). A total of 23 paediatric AML patients (4.7%) harboured RUNX1 mutations, 18 of which (78%) had RUNX1 mutation at initial diagnosis. RUNX1 mutations were associated with older age, male gender, number of coexisting alterations and presence of FLT3-ITD but mutually exclusive of KRAS, KIT and NPM1 mutation. RUNX1 mutations did not prognostically impact overall or event-free survival. Response rates did not differ between patients with and without RUNX1 mutations. This comprehensive study, comprising the largest analysis of RUNX1 mutation in a paediatric cohort to date, reveals distinct but not unique clinicopathologic features, with no prognostic significance of RUNX1-mutated paediatric AML. These results broaden the perspective on the relevance of RUNX1 alterations in leukaemogenesis in AML.


INTRODUCTION
Acute myeloid leukaemia (AML) is a genetically heterogeneous disease. Analyses of genetic aberrations helped to unravel a complex genetic landscape and improved our knowledge of AML pathogenesis, leading to refined, risk-adapted treatment allocation [1][2][3][4]. In paediatric AML, optimising genetic and responsedependent risk-group stratification, allocating high-risk patients to receive allogeneic stem cell transplantation (SCT) in first complete remission (CR) allows survival rates of about 80% [4]. Notably, frequency of several prognostic biomarkers differs greatly between adult and paediatric AML, suggesting a different ontogeny in childhood AML [5][6][7]. Moreover, distinct agespecific genetic profiles point at a unique biology in paediatric AML and implicate the necessity for precise, age-tailored diagnostic and therapeutic approaches [7][8][9][10][11].
Intragenic mutations of the runt-related transcription factor-1 (RUNX1) have been associated with distinct clinical and genetic features as well as inferior prognosis in adult AML [12][13][14][15][16]. These results prompted the inclusion of "AML with mutated RUNX1" as new provisional entity of the World Health Organisation (WHO) classification as part of the revision in 2016 [17,18]. Due to lack of specificity, this was rejected by the latest 5 th edition of the WHO classification [19]. Considering the independent association with unfavourable risk, the European LeukemiaNet (ELN) recommendation classified RUNX1mutated AML into the adverse risk category [20,21].
Intragenic mutations of the RUNX1 gene were preliminary reported in MDS, secondary and therapy-related AML, radiation exposed myelodysplastic syndrome (MDS) and AML [26][27][28]. Inherited RUNX1 mutation were found to cause familial platelet disorder (FPD) with a propensity to MDS and AML [29][30][31]. To date, the impact of the RUNX1 mutation in paediatric AML has remained unclear. There are only a few controversial reports, with a limited data and number of patients on the RUNX1 mutation in paediatric AML [32][33][34][35].
This retrospective study elucidates the pathological and clinical implication of RUNX1 mutation in a substantial paediatric AML cohort. By targeted sequencing, we assess concomitant genetic changes and their impact in development of paediatric AML.

Study participants
A paediatric cohort of 488 patients (0-18 years), diagnosed with de-novo AML between 2015 and 2021 was included. The cohort represents a German paediatric AML population, excluding patients with acute promyelocytic leukaemia (FAB M3), Down syndrome (ML-DS), secondary and treatment-related AML. All included patients were enroled and treated either according to the AMLR12 registry (EudraCT number: 2013-000018-39) or AMLR17 registry (DRKS number: DRKS00013030) of the AML-BFM Study Group. The ethical committees and institutional review boards of the university hospital of Hannover and Essen approved both registries according to the declaration of Helsinki. Informed consent was obtained from patients and their legal guardians prior to treatment participation. Initial and follow-up diagnostics including morphological, immunophenotypic and molecular genetic testing were performed centrally by the German AML-BFM reference laboratory Essen according to standard procedures. Analysis for classical cytogenetics and fluorescence in situ hybridisation (FISH) were carried out by the Institute of Human Genetics in Hannover Medical School. The treatment recommendations of analysed registries included similar cytarabine-and daunorubicin-based induction chemotherapy [36]. Depending on the risk stratification intensive chemotherapy, maintenance therapy and allogeneic SCT were performed by indication.

Genetic analyses
Primary bone marrow or peripheral blood samples from paediatric patients were subjected to molecular genetic screening at the time of diagnosis by next generation sequencing (NGS) using the TruSight® Myeloid Panel (Illumina, San Diego, California, USA). Using the standard NGS technology, the mutational screening included 54 different leukaemia-associated genes (Supplementary Table S2b). Sequencing was performed on the Illumina MiSeq™DX System (Illumina, San Diego, California, USA). Analysis was conducted using the SOPHIA DDM TM software (Sophia Genetics, Switzerland). RUNX1 mutations were confirmed by PCR and subsequent Sanger sequencing.
To check for germline mutations, paired remission samples of RUNX1mutated AML patients, harbouring a VAF above 30 were analysed by PCR to detect the respective RUNX1 mutation.
RNA sequencing was analysed using the TruSight® RNA Fusion Panel (Illumina, San Diego, CA, USA) according to manufacturer's recommendations. Classical karyotyping was performed following the International System for Human Cytogenetic (International System of Cytogenetic Nomenclature; ISCN 1995-2013).

Statistical analysis
Patient characteristics between groups were compared using Pearson chisquare or Fisher's exact test for categorical variables and Mann-Whitney U test for continuous variables. A p-value of less than 0.05 was considered to indicate statistical significance. Survival analysis included Overall survival (OS), defined as the time from diagnosis to death or last follow-up, Eventfree survival (EFS), defined as the time from diagnosis to relapse, secondary malignancy, death, or the time of last follow-up. Kaplan-Meier curves were calculated for OS and EFS. The log-rank test was obtained to compare survival differences. To identify prognostic variables, Cox-models were used for univariate and multivariate regression analyses. Results were reported as hazard risk ratios with 95% confidence intervals (CI) indicated. Following complexity and dimensionality reduction to a reasonable, recommended number of 50 principal components (PCs), the unsupervised clustered data were visualised using t-distributed stochastic neighbourhood embedding (t-SNE) according to recent recommendations [37] using Seaborn and Plotly Express Python visualisation libraries based on matplotlib (Python 3.9.13, Jupyter Notebook 6.4.12). Statistical analysis was conducted using SPSS software version 28.0 (Chicago, IL, USA) and R software version 4.0.2 (http://www.Rproject.org). Illustrations were generated using Graphpad Prism™ 8.3, R software version 4.0.2, Python 3.9.13 and DOG 2.0 [38].

Characteristics of RUNX1 mutations
In 23 patients (4.7%), 27 RUNX1 mutations (OMIM No. 151385) were detected of which 24 mutations were distinct from each other (Fig. 1, Supplementary Table S1a, b). Within the RUNX1mutated cohort, 18 out of 23 patients presented at least one RUNX1 mutation at initial diagnosis (78%). Among eight patientmatched paired diagnosis and relapse samples, RUNX1 mutations were detectable only at relapse in five cases (Supplementary Tables S1a and S6). The same RUNX1 mutation was present in two patients at both time points. In one case, RUNX1 mutation was undetectable at relapse. Comparing the paired samples in this case, two additional mutations were lost from initial diagnosis to relapse. Regarding morphologic, immunophenotypic and cytogenetic criteria the early relapse resembled the initial diagnosis, indicating that the loss of the RUNX1 mutation occurred during the process of clonal evolution (Supplementary Tables S6). Considering all 27 detected RUNX1 coding sequence alterations, the majority were characterised as frameshift (48.1%), followed by nine missense (33.3%) and two different deletion-insertions (delins) in five cases (18.5%). Four patients were found to have two different RUNX1 mutations, three of which shared the same combinatorial pattern (p.A208_M209delinsDR, p.R206C), including the sample with RUNX1 mutation only at diagnosis. Among the five relapsed samples harbouring RUNX1 mutation only at relapse, three frameshifts, two missenses, and the repeatedly occurring mutational combination were found. Considering the distribution on the RUNX1 gene ( Fig. 1), one mutation was allocated to exon 2, six mutations to exon 3, one to exon 4, seven to exon 5.1, (including p.A208_M209delinsDR, p.R206C), one to exon 5.2 and five to exon 6. Mutations were assigned to different functional groups, whilst the largest proportion of eleven mutations (40.7%) was clustered in the runt homology domain (RHD), two mutations (7.4%) were in the transactivation domain (TAD), four (14.8%) in the repression domain (RD) and eight (29.6%) mutations were located between the TAD and RD. One frameshift mutation was localised at C-terminal to the RD. In the RHD all three different kinds of mutations were discovered, whereas in the TAD and RD only frameshift mutation occurred. To exclude germline mutations and proof the somatic origins, RUNX1-mutated AML with a variant allele frequency (VAF) > 30% were selected for germline-testing. Since one of which never reached remission, samples at remission were available for nine patients. The suspected RUNX1 mutation was not detectable in any of the remission samples, confirming the somatic origin.
Genetic mutations detected by NGS were assigned to different functional groups according to the functional categorisation of the mutated genes [2] (Supplementary Table S2c). The largest proportion was attributed to activated signalling, which made up 41% among RUNX1-mutated and 50% among wt-cases.
Mutations affiliated to epigenetic regulator occurred more frequently together with RUNX1 mutation (22 vs 10%). Alterations of transcription factors, excluding RUNX1, were the second most common aberrations among wt-cases, showing comparable frequency in RUNX1-mutated and wt-cohort (13 vs. 15%). Aberrations of tumour suppressor genes were evenly distribution between RUNX1-mutated and wt-cohorts (16 vs. 12%). Together, there was no clear association between RUNX1 mutations and any of the other genetic groups. Accordingly, RUNX1-mutated samples projected scattered without showing definable, separate clustering in the t-SNE plot, which contrasts with the well-known distinct entity of CBF in AML (Supplementary Fig. S1).
Impact of RUNX1 mutations on response, outcome and survival Response to therapy was comparable between RUNX1-mutated and wt-cases (94 vs. 84% achieved CR, p = 0.33). None of RUNX1mutated, but 20 of wt-cases suffered early death (4%) (Supplementary Table S3). The proportion of patients undergoing SCT in first CR was evenly distributed between the RUNX1-mutated, and wt-cohorts (23 vs. 22%). Likewise, number of patients transplanted as salvage therapy were almost equal in both groups (Supplementary Table S3).
Univariable regression analysis revealed no strong prognostic impact of RUNX1 mutation on OS (HR 0.43, p = 0.16) or EFS (HR 0.78, p = 0.55; Supplementary Table S4a). Multivariable models were created following adjustment for age, risk groups, FLT3-ITD, SCT and additional mutations (Supplementary Table S4b). RUNX1 mutation was not predictive to experience any event (HR 1.4, p = 0.39) nor of death (HR 4.2 p = 0.17). Since RUNX1 mutation was significantly associated with FLT3-ITD, increased patient age and number of mutations, subgroup analyses were performed to separately assess the relative prognostic relevance of RUNX1 mutation within these subgroups (Supplementary Table S5a Table S5b). Together, in our paediatric cohort, RUNX1 mutation showed no prognostic significance.

DISCUSSION
AML with RUNX1 mutation was considered as adverse-risk AML by the ELN and new provisional entity in the 2016 WHO classification given its characteristic clinicopathologic features and inferior outcome in adult AML [13,[16][17][18]20]. In 2022, the preliminary stand-alone entity was removed due to lack of specificity [19]. In paediatric AML, large-scale sequencing studies suggested a distinct genetic landscape with clinical significance [7,8,11]. However, impact of RUNX1-mutated AML in paediatrics has largely remained obscure.
This retrospective study evaluates RUNX1 mutations in the largest paediatric cohort to date. Prevalence of the studied RUNX1-mutated paediatric AML cases is relatively low (3.7%). In comparison, in adult AML, a frequency of~5-13% has been reported, with highest prevalence when including elderly patients (>60 years) [13,16,39]. Accordingly, reported prevalence of RUNX1 mutation was lower in younger patients and an association with patients age has been shown before [12][13][14]33]. Likewise, in the paediatric cohort studied here, RUNX1 mutation was found more frequently in older children. Consistent with previous findings our results confirm a positive correlation to male gender in paediatrics. Unlike previous studies [13,15,33], we could not detect a correlation with minimally differentiated AML (FAB M0). In AML with M0 phenotype, biallelic RUNX1 mutations have been frequently described, where in case of heterozygous RUNX1 mutation, a deletion on the other allele is suggested to result in a functional inactivation of RUNX1 [31,40]. However, this could not be confirmed in the current paediatric cohort, since there was only one heterozygous, biallelic RUNX1 mutation in a FAB M4 sample, without evidence of further biallelic mutation.
The mutational spectrum revealed a higher proportion of frameshift mutations and a clustering of missense mutations at the N'-terminus, especially in RHD. A similar distribution of RUNX1 mutation according to different functional domains and clustered mutations at the N´-terminus was previously reported in adult AML, whilst Greif et al. detect a greater proportion of missense mutation in cytogenetically normal AML [14][15][16]. Unexpectedly, in three patients we found a recurrent combination of the same missense and delins mutational pattern, which were not described before. These mutations had a low VAF (<10) and were only detectable at diagnosis or relapse in paired patient-matched samples, suggesting high mutational instability. Prevalence and implication of germline mutations in AML samples with RUNX1 mutation has been discussed controversially [41,42]. In our study, there was no evidence of germline mutations, which may not seem surprising given the reported wide age range for AML onset in FPD [30,31]. Interestingly, in the current cohort, a child with Down Syndrome related AML (ML-DS) who was excluded a-priori from the analysis, harboured the same RUNX1 mutation in remission, suggesting a germline mutation (data not shown). To our knowledge, except for the increased dosing effect of RUNX1 in ML-DS, no increased prevalence and no specific mechanism of interaction between altered RUNX1 and GATA1 has been described [31,[43][44][45].
Regarding cytogenetics, we could not affirm an inverse correlation of RUNX1 mutation and the recurrent CBF-defining abnormalities t(8;21) and inv(16)/t (16;16). Consistent with reports in adult AML, RUNX1 mutation weakly associated with the unbalanced abnormality −7/7q [13,16], which has also been described as myelodysplasia related aberration. However, none of the analysed paediatric AML patients had a known history of MDS and further cytogenetic alterations related to myelodysplasia, including −5/5q, were almost absent in the evaluated cohort. In contrast to previous reports, RUNX1 mutation did not correlate with trisomy +8 [16,39,46], trisomy +13 was even absent in the analysed RUNX1-mutated paediatric cohort [13,15,47]. Distribution of normal and complex karyotype was largely balanced, which in part, agrees with results of a small Japanese paediatric AML cohort, where 6 out of 11 RUNX1-mutated patients had normal karyotype, whilst +8 and complex KT were present in one patient [33]. Ultimately, RUNX1-mutated paediatric AML did not show a distinct cytogenetic pattern, particularly with respect to the overlap with prognostically favourable subgroup CBF-AML, arguing against RUNX1 mutation as a distinct AML type in paediatrics.
RUNX1-mutated AML completely lacks KMT2Ar, indicating that RUNX1 mutation most probably do not contribute to leukaemogenesis in KMT2Ar AML in paediatrics. In fact, in AML mouse models harbouring KMT2Ar, RUNX1 activity mediated leukaemic cell growth and survival [48,49]. Similarly, RUNX1 was determined as transcriptional target in KMT2A::AFF1, previously known as MLL::AF4, AML by Wilkinson et al. [49], suggesting that RUNX1 activity cooperates with KMT2Ar to promote AML development. Considering the determined exclusivity of RUNX1 mutation and KMT2Ar, this could also be true for paediatric AML.
Given that RUNX1 mutations alone are highly unlikely to be sufficient to cause complete leukaemic transformation [50], investigation of cooperating mutations is required to further elucidate leukaemic mechanisms in RUNX1-mutated AML. Increased number of co-occurring mutations indicates a more complex genetic evolution of RUNX1-mutated AML, requiring a broad range of additional mutations.
FLT3-ITD was the most common co-occurring mutation and significantly associated with RUNX1 mutation in paediatric AML, which may suggest cooperativity between these two mutations. An early report indicated a link between RUNX1 mutation and FLT3-ITD contributing to leukaemogenesis in AML M0, in which RUNX1 mutations predominated [51]. In comparison, FLT3-ITD frequently occurred in RUNX1-mutated adult AML, albeit evenly distributed and without correlation with RUNX1 mutation [15,39], supporting the notion of a cooperation between RUNX1 mutation and FLT3-ITD, which may be specific to paediatric AML [52]. RUNX1 mutations were almost exclusive of alterations involved in other activated-signalling pathways, namely NRAS, KRAS and KIT, which in contrast, were relatively common in the wt-cohort. These results may point at a reassortment in the functional group of activatedsignalling, shifted towards FLT3-ITD, supporting the suggestion, that RUNX1 mutation preferably contributes to leukaemogenesis by cooperating with the activating FLT3-ITD, instead of other activating pathways in paediatric AML. Despite the fact that FLT3-ITD has long been known to associate with poor prognosis [53,54], in the present study, co-occurrence of FLT3-ITD with RUNX1 mutation did not affect outcomes in paediatric AML, which is consistent with results in adult AML [15].
Despite the small number of patients, which makes the accurate assessment of additional genetic associations more difficult, concurrent genetic aberrations occurred more frequently in epigenetic modifiers, with a trend towards BCORL1, EZH2, and KDM6A. A propensity for association was noted for spliceosome U2AF1, whereas other spliceosome-related genetic aberrations were absent in RUNX1-mutated paediatric AML. In contrast, RUNX1-mutated adult AML has been associated with mutations in the spliceosome and cohesion complex, in addition to epigenetic modifiers [13], consistent with those reported in high-risk MDS [55,56]. This further underlines the heterogeneity in biological properties between adults and paediatric AML.
Consistent with previous studies [12][13][14][15]57], mutations of the transcription factor NPM1 were completely absent in the RUNX1mutated paediatric cohort. NPM1 mutation is considered as founding genetic event in AML with certain biological and clinical features, warranting its own entity [52,58]. The exclusivity of NPM1 and RUNX1 mutation in AML might suggest that RUNX1 mutation shares a similar role with NPM1 mutations in leukaemia development. In two cases we observed that RUNX1 mutation overlaps with double mutated CEBPA in paediatrics, contradicting the notion that RUNX1-mutated AML represents a distinct entity. Thus, according to the results of this cohort, previously reported characteristic features, favouring the concept of considering RUNX1-mutated AML as a distinct entity [13,17], cannot be confirmed in paediatric AML since RUNX1 mutation lacks these mutational features and overlaps with AML defining mutational patterns.
In the recent ELN-recommendation, RUNX1 mutation was stated as a late event in leukaemia development [21], which contradicts earlier studies, considering RUNX1 mutations as early events in leukemognenesis [15,50]. The observed inconsistency of RUNX1 mutations in six out of eight paired diagnosis-relapse samples in this paediatric cohort, indicates instability, suggesting that RUNX1 mutation rather is a late genetic event contributing to advanced clonal evolution. Moreover, the detected high number of additional aberrations accompanying RUNX1 mutation and the observed overlap with the AML-causing events CBF-fusions or NPM1 mutation in the assessed paediatric patients, further support the concept that RUNX1 mutations are more likely to contribute to the late-stage leukaemia development.
In this study, RUNX1 mutation did not show a prognostic significance for outcomes and survival. These results contrast with previous data on RUNX1 mutation in AML, indicating that RUNX1 is a negative prognosticator [12][13][14]33]. Since some reports combined primary and secondary AML, Quesada et al. assessed the prognostic impact of RUNX1-mutated de novo AML according to the WHO definition and detected no differences in outcome [57], which is concordant with our findings in primary paediatric AML. In addition, Greif et al. proved negative prognostic significance in elderly (> 60 years) cytogenetically normal AML, whereas this could not be confirmed for younger patients [14,16], suggesting age-dependency of the poor prognostic impact of RUNX1 mutation. This observation could be strengthened by our results, where RUNX1 mutation showed no predictive relevance in paediatric AML. However, we did not observe a strong difference when separately assessing agedepending prognostic impact of RUNX1 mutation in paediatric patients. Since the survival was not affected by RUNX1 mutation in different risk groups, there seems to be no relevance for including RUNX1 mutations in the risk stratification of paediatric AML. In terms of indication for SCT, survival was comparable in patients with and without SCT in first CR, suggesting that the presence of RUNX1 mutation does not require modification in eligibility criteria in paediatric AML.
This analysis uncovers the impact of RUNX1 mutation in a well characterised paediatric AML cohort and provides new insights into its clinicopathological relevance. Although this study evaluates RUNX1 mutations in the largest paediatric cohort studied to date, analyses are still limited due to small sample sizes for some analysis-subgroups and the retrospective dataset, necessitating further validation to confirm these observations.
In conclusion, RUNX1 mutations did not impact outcomes in this paediatric AML cohort, suggesting a clinical impact that differs from that reported in adult AML. Importantly, RUNX1 mutations lacked clinical characteristics and genetic features, which would justify a separate WHO entity. Therefore, the previously stated provisional WHO entity 'RUNX1 mutated AML', is not appropriate in paediatric AML. The results of this study agree with the recently published fifth edition of the WHO classification which removed RUNX1 mutations as distinct entities in AML [19].