Mutational cooperativity of RUNX1::RUNX1T1 isoform 9a and oncogenic NRAS in zebrafish myeloid leukaemia

Acute myeloid leukaemias (AML) are an aggressive and heterogeneous group of cancers characterised by the accumulation of immature myeloid cells, and impairment of normal haematopoiesis. In WHO diagnostic criteria, translocation t(8;21) (q22;q22) is an AML defining genetic abnormality (Khoury et al., 2022), accounting for 11‒32% of paediatric and 4‒10% of adult AMLs (Bolouri et al., 2018; Chang et al., 2000; Grimwade et al., 1998; Ustun et al., 2018). The core-binding factor subunit gene RUNX1 (aka AML1, CBFA2), a transcription factor essential for haematopoietic differentiation and myeloid maturation, is translocated from chromosome 21q22.12 to the transcriptional corepressor RUNX1T1 locus (aka ETO, CBFA2T1) on 8q21 (Bruford et al., 2021; Lutterbach et al., 1998; Okuda et al., 1996). The resultant RUNX1::RUNX1T1 (R::RT1) oncofusion gene encodes the DNA-binding domain of RUNX1 fused in-frame to RUNX1T1, thereby converting the RUNX1 transcriptional activator into a transcriptional repressor with dominant negative activity at RUNX1 DNA-binding sites (Gelmetti et al., 1998; Wang et al., 1998). The oncofusion protein impairs differentiation and apoptosis and may predispose to additional genetic and epigenetic alterations that lead to leukaemia (Al-Harbi et al., 2020). Although t(8;21) is considered a comparatively favourable prognostic factor, relapse is frequent and in a recent international cohort of ∼250 adult R::RT1 AMLs, median overall survival was only 31 months (Borthakur and Kantarjian, 2021; Ustun et al., 2018). In addition, treatment of AML can cause life-shortening and life-limiting side effects in survivors, including cardiotoxicity and secondary malignant neoplasia (Abrahão et al., 2021; Feijen et al., 2019; Turcotte et al., 2019). Thus, further research is required to understand the molecular pathogenesis of R::RT1 AMLs to improve outcomes (Al-Harbi et al., 2020).

Early efforts to develop murine R::RT1 AML models revealed that the knock-in R::RT1/+ heterozygous state is embryonic lethal due to disruption of foetal haematopoiesis (Okuda et al., 1998; Yergeau et al., 1997). To overcome this, alternative strategies were employed, such as inducible expression and murine bone marrow transplant (mBMT)-based methods. These typically expressed R::RT1 or its splice variant 9a, alone or with other gene mutations that frequently co-occur with t(8;21), such as NRAS^G12D or cKIT^N822K (Abdallah et al., 2021; Desai et al., 2020; Higuchi et al., 2002; Schessl et al., 2005; Yan et al., 2006; Yuan et al., 2001; Zhao et al., 2014; Zuber et al., 2009). The rationale to co-express mutations with R::RT1 is that 95% of human t(8;21) AMLs harbour other mutations or chromosomal abnormalities. While co-occurring mutations may be passenger mutations that accumulate within the haematopoietic stem and progenitor population (HSPC) with age (Welch et al., 2012), recurring patterns of co-mutation and mutational exclusivity infer that cooperating oncogenic mutations act on distinct biological pathways to drive AML leukaemogenesis (Bolouri et al., 2018; Christen et al., 2019; Duployez et al., 2016; Faber et al., 2016; Ley et al., 2013; Papaemmanuil et al., 2016; Patel et al., 2012). In children diagnosed with t(8;21) AMLs at 3 to 12 years of age, t(8;21) was detectable in newborn blood samples, consistent with the acquisition of additional mutations over a latency period (Wiemels et al., 2002). In addition to somatically acquired mutations, recent studies indicate that R::RT1 misappropriates developmentally-related factors to induce neoplastic transformation (Abdallah et al., 2021). Although animal models and clinical data demonstrate that R::RT1 alone is necessary but nearly always insufficient for transformation, the molecular and cellular consequences of mutational cooperation are not fully elucidated or exploited clinically.

Zebrafish provide a genetically amenable system for studying haematological diseases due to conserved haematopoietic genetic networks, fecundity, and scalability (Gore et al., 2018; Jagannathan-Bogdan and Zon, 2013; Potts and Bowman, 2017). Broad expression of human R::RT1 cDNA in embryonic zebrafish alters primitive haematopoiesis, biasing myeloid lineages towards granulocyte production with concomitant loss of erythrocyte fate, a trend also seen in patients and murine models (Yeh et al., 2008). However, the impact of R::RT1 on zebrafish definitive haematopoiesis, mutational cooperation, and leukaemogenesis beyond the larval stage has yet to be fully explored. We hypothesised that restricting R::RT1 expression to the putative cells of leukaemia origin, the HSPCs, would bypass early lethality and that its co-expression with a commonly co-occurring NRAS^G12D mutation would lead to oncogenic cooperation and the establishment of AML in adult zebrafish. To this end, we developed transgenic zebrafish coexpressing cDNAs encoding human R::RT1 splice variant 9a (9a) and human mutant NRAS^G12D (NRAS) in HSPCs. We find that expression of 9a with NRAS, but not expression of either oncogene alone, promotes a haematological phenotype consistent with AML. The marrow of AML 9a+NRAS zebrafish is hypercellular and enriched in immature myeloid and blast-like cells but deficient in erythrocytes. By contrast, 9a-only-expressing animals are healthy with no overt signs of abnormal haematopoiesis, while those expressing NRAS^G12D alone show signs of blood disease that are associated with expansion of mature myeloid and lymphoid cells and a reduction in erythrocytes. Significantly, our zebrafish transplantation data infer that only 9a+NRAS disease, and not NRAS-only disease, is transplantable and able to re-establish leukaemia in recipients. Taken together, our findings suggest that when co-expressed in the zebrafish marrow, 9a and NRAS mutational cooperation significantly enhances the establishment of self-renewal, an essential leukaemia hallmark. This transgenic model provides a genetically amenable and scalable system for exploring t(8;21) AML and oncogenic cooperation.

The establishment of leukaemia involves competition between mutated and healthy HSPCs (Glait-Santar et al., 2015; Loberg et al., 2019), ultimately leading to multiple AML subclones (Chopra and Bohlander, 2019; Jan et al., 2012). To model mutational cooperation in leukaemogenesis, we explored the inherent cellular mosaicism of F0 transgenics to study biologically distinct animals (Blainey et al., 2014; Ung et al., 2015). In R::RT1 AML patients, transcripts encoding full-length R::RT1 fusion protein and a C-terminal truncated variant, 9a, have been detected at varied ratios (Yan et al., 2006). There is no consensus regarding the prognostic significance of either isoform or their relative abundance (Agrawal et al., 2020; Jiao et al., 2009; Ommen et al., 2010). However, mouse modelling studies suggest that the 9a isoform may have greater oncogenic potency than full-length RUNX1::RUNXT1 (Yan et al., 2006). For this reason, we preferentially used the 9a variant in our expression constructs. We generated four transgenes: 9a-P2A-GFPNRAS^G12D (9a+NRAS), 9a-P2A-GFP (9a), GFPNRAS^G12D (NRAS), and GFP (Fig. 1A; Fig. S1; Table S1; see also, the Materials and Methods section). To achieve HSPC-restricted expression, transgenes were placed under the control of the conserved Runx1+23/24 enhancer element fused to the mouse β-globin minimal promoter (collectively abbreviated as R1⁺²³) (Nottingham et al., 2007; Tamplin et al., 2015). Constructs were injected into one-cell-stage wild-type embryos, and resultant transgenic animals were either used for embryonic whole-mount RNA in situ hybridization studies (WISH) or allowed to continue developing where they were monitored daily for signs of disease for up to 16 months.

HSPC generation and development are highly conserved between zebrafish and humans (Jagannathan-Bogdan and Zon, 2013; Mahony and Bertrand, 2019). In zebrafish the HSPCs emerge from the haemogenic endothelium of the ventral dorsal aorta wall (aorta-gonad mesonephros equivalent) at ∼26 hpf, and travel to the caudal haematopoietic tissue (CHT, foetal liver equivalent) where they undergo expansion before migrating to seed the kidney marrow. R1⁺²³ -driven expression of GFP in HSPCs can be detected as early as 32 hpf, in the CHT (Tamplin et al., 2015). The CHT HSPCs are a heterogeneous population consisting of a mixture of stem-like and lineage-committed HSPC subgroups (Xia et al., 2021). Fully differentiated cells associated with the CHT (macrophages, monocytes and erythrocytes) derive from previous haematopoietic waves. To assess whether HSPC-targeted expression of 9a and/or NRAS impacts CHT HSPC populations and differentiation, we examined expression of seven key haematopoietic genes at 72 hpf (3 dpf), using WISH (Fig. 1B–E; Fig. S2). These genes included stem cell marker, cmyb, and lineage restriction effectors that are known targets of R::RT1 in AML, such as cepba and pu.1 (myeloid factors) and gata1a (erythroid) (see Table S2 for probe information) (Choi et al., 2006; Stengel et al., 2021). Signal intensity in the CHT was quantified by measuring pixel intensity while blinded to genotype (Fig. 1B–E; see Materials and Methods for details). Expression of stem cell marker cmyb was not modified by 9a alone, but cmyb expression increased in NRAS and 9a+NRAS embryos. Expression of myeloid marker pu.1 was not modified by 9a alone but was decreased in NRAS and 9a+NRAS. The effects on cepba were less pronounced, with no significant differences in expression compared to GFP controls (Fig. 1B–E). Expression of gata1a and scl were not modified, but NRAS alone did lead to a reduction in expression of the globin encoding hbbe1.1. Expression of lyz was not modified (Fig. S2). Taken together, these results suggest that 9a and NRAS have distinct effects on HSPC proliferation and lineage restriction with the most pronounced effect being observed with NRAS increasing expression of the stem cell marker cmyb and decreasing expression of a myeloid marker pu.1.

In adult transgenic populations we observed that survival of the GFP control cohort over a 16-month time frame was not significantly different from that of the wild-type background used to generate the F0 transgenic animals (n=13 deaths/203, 6%; Fig. 2A). Survival of the 9a-only F0 cohort during the 16-month period was comparable to that of the GFP F0 population (n=12 deaths /132, 9%). Similarly, survival of F0 s expressing full-length R::RT1 alone was also comparable to GFP controls (n=13 deaths/51, 25%; Fig. S3) suggesting that expression of this oncofusion in zebrafish does not overtly impact viability, regardless of isoform. By contrast, F0 animals expressing oncogenic NRAS alone showed significantly reduced survival (n=48 deaths/131, 37%; P=0.0024 cf. GFP) beginning at ∼3 months, displaying signs consistent with haematopoietic disease, that included reduced activity, bleeding from the gills, and rapid breathing (compare Fig. 2B–D). Although some features were potentially non-specific, oncogene expression was restricted to the HSPC compartment, and phenotypes are consistent with murine leukaemia models where oncogenic NRAS expression leads to symptomatic anaemia and thrombocytopaenia (Li et al., 2011; Wang et al., 2013, 2011). Like NRAS F0s, 9a+NRAS F0 animals also showed significantly reduced survival from ∼3 months onwards, again associated with reduced activity, bleeding, and rapid breathing (n=202 deaths/385, 52%; P=0.0001 cf. GFP; Fig. 2D). Survival of F0 9a+NRAS fish was significantly lower than that of NRAS F0s (P=0.009 cf. NRAS) with 50% of 9a+NRAS F0 succumbing by 16 months compared to 40% of NRAS animals. Therefore, 9a+NRAS oncogene co-expression produced a more penetrant phenotype than expression of NRAS or 9a alone. To investigate this, we further characterised the haematopoietic system of these animals.

In adult zebrafish, the site of haematopoiesis is the kidney marrow, a tissue considered functionally analogous to the bone marrow of vertebrates (Jagannathan-Bogdan and Zon, 2013; Mahony and Bertrand, 2019). Like vertebrate bone marrow, the zebrafish kidney marrow harbours the HSPCs, their descendant precursor cells (immature, incompletely differentiated terminal blood cells), and mature blood cells yet to be released into the peripheral blood. To determine whether alterations in haematopoiesis were responsible for 9a+NRAS or NRAS F0s sickness and mortality, we performed marrow cell flow cytometry, morphological analysis of marrow smears, RT-PCR for confirmation of transgene expression (Fig. S4), and histological analyses of whole animal tissue sections in GFP controls and oncogene transgenics.

Forward versus side scatter (FSC/SSC)-based cell flow cytometry resolves cells of the whole kidney marrow (WKM) into four major groups: progenitors/precursors (defined by the P gate), myelomonocytes (the M gate, which includes granulocytes, monocytes, basophils, and eosinophils), lymphocytes (L) and erythrocytes (E). The L gate harbours lymphocytes predominantly (Traver et al., 2003) (see Fig. S5A–C for cell morphology-flow gate correlation and Fig. S6 for WT versus GFP flow comparison). While R1⁺²³ GFP-marked HSPCs straddle the L and P gates, sorting of GFP-positive cells confirms that the majority of cells have a progenitor-like morphology (Fig. S5D) (Henninger et al., 2017; Tamplin et al., 2015). In flow analysis, 9a F0 marrows were not significantly different from that of age-matched GFP controls in the relative proportions of gated populations, consistent with the observation that 9a F0 viability was not significantly impacted by oncofusion expression (Fig. 2A; Fig. 3A–D; compare Fig. 3E and Fig. 3F). In contrast, sick F0 NRAS marrow had a significantly higher fraction of myelomonocytes and lymphocytes but fewer erythrocytes compared to 9a animals and GFP controls (Fig. S3A,C,D and flow profile example Fig. 3G), while the fraction of progenitor/precursors was not significantly different in these genotypes (Fig. 3B). This pattern (high M, high L and low E) was evident in 85% of F0 NRAS animals. An exception to this was a single NRAS F0 with a high proportion of progenitor/precursors (49%) and a low percentage of erythrocytes (18%) compared to GFP or 9a genotypes. These phenotypes are reminiscent of the impact of NRAS^G12D in murine systems where, depending on gene copy number, NRAS^G12D promotes a form of myeloproliferative disease (MPD; characterised by lymphoproliferation with mature myeloid cell expansion and erythrocyte hypoplasia) or acute monocytic myeloid leukaemia (characterised by expansion of immature cells) (Li et al., 2011; Wang et al., 2011; Wang et al., 2015). In our zebrafish model, HSPC-targeted expression of oncogenic NRAS promotes a trend towards MPD in most F0s. Such disturbances in marrow haematopoiesis, including marked erythrocytopaenia, may contribute to reduced survival of NRAS F0s animals.

In contrast to 9a and NRAS F0 animals, the 9a+NRAS cohort produced more variable flow profiles. As seen among NRAS F0s, sick 9a+NRAS animals had a significantly lower proportion of erythrocytes compared to 9a and GFP controls, consistent with previous studies showing that R::RT1, like oncogenic NRAS, inhibits erythrocyte differentiation (Choi et al., 2006; Fenske et al., 2004; Schwieger et al., 2002; Tonks et al., 2003; Yeh et al., 2008). Lymphocytes were also significantly higher in 9a+NRAS F0s compared to controls, possibly due to the action of NRAS (Fig. 3C). However, 9a+NRAS and NRAS cohort flow profiles differed in two important respects. First, unlike NRAS F0s, myelomonocytes (M) were not significantly higher than GFP in 9a+NRAS animals, consistent with the possibility that either 9a activity imposes a myeloid differentiation block or that NRAS expression levels are insufficient to drive myeloproliferation. Second, 9a+NRAS F0s showed a significantly higher fraction of cells in the progenitors/precursors (P) gate compared to NRAS (9a+NRAS P gate median=22.7% cf. NRAS median=17.6%, P=0.02; cf. 9a median=12.7% and GFP median=11.1%, P=0.0001; Fig. 3A–D; Fig. 3H). Moreover, 27% of 9a+NRAS animals (n=18/66 animals analysed) had a pronounced ‘AML-like’ flow profile, namely, progenitor/precursor percentages of 35% or higher that was invariably associated with low erythrocyte percentages (i.e. E<36%, that is, below the NRAS first quartile for E).

A principal component analysis (PCA) of flow data provides a visual representation of flow cytometry profiles in relation to genotypes (Fig. S7). The 9a+NRAS cohort showed the greatest phenotypic heterogeneity, plotting as a continuum rather than a discrete cluster, with some animals aligning more with 9a or NRAS alone genotypes, while those with AML-like profiles plotted more distantly from these. While our NRAS cohort demonstrates that NRAS alone can induce an AML-like phenotype, its frequency (1 in 16) is substantially lower than when NRAS is co-expressed with 9a (1 in 4), even though NRAS is likely expressed at higher levels in NRAS animals than it is in 9a+NRAS animals, due to the second gene position effects in the latter (Liu et al., 2017) (Fig. 1A).

We stained marrow smears with May-Grünwald-Giemsa (MGG) to determine if oncogene expression impacted haematopoietic cell fates, as determined by morphological criteria (number of animals examined per genotype: GFP, 22; 9a, 31; NRAS, 11; 9a+NRAS, 34). Overall, the distribution of fates was highly consistent with flow cytometry patterns. The cellular composition of 9a smears resembled age-matched GFP animals (Fig. 4B,F,cf. A,E). Compared to 9a and GFP animals, NRAS F0 smears harboured a higher proportion of mature myeloid cells (monocytes, eosinophils, and granulocytes), mature lymphocytes and immature erythroid cells (Fig. 4C,G). Significantly, a subset of 9a+NRAS smears contained a high proportion of blast-like cells and immature myeloid cells, suggestive of AML disease (Fig. 4D,H). As noted above, R::RT1 AML is characterised by the accumulation of abnormal myeloid granulocytes exhibiting varied differentiation states (Arber et al., 2002; Jiang et al., 2020; Tonks et al., 2004; Westendorf et al., 1998). To determine if granulocytic cells were more abundant in 9a or 9a+NRAS F0s, we performed a cytochemical staining assay for myeloperoxidase (MPO) enzyme activity, a granulocytic fate marker (see Fig. 4I–L; Fig. S8). Although animals with the highest MPO percentages were associated with the 9a+NRAS genotype, statistical analysis revealed no significant difference compared to other genotypes, possibly owing to the high degree of phenotypic heterogeneity among 9a+NRAS animals.

To assess the impact of oncogene expression on the overall marrow architecture and cellular density, we fixed and sectioned animals of each genotype and stained tissue sections with Haematoxylin and Eosin (Fig. 4M–T; number of examined animals per genotype, n: GFP=2; 9a=5; NRAS=4; 9a+NRAS=12). The marrows of 9a, GFP and sick NRAS animals were qualitatively similar for cell density (Fig. 4M–O,Q–S), suggesting that neither NRAS- nor 9a expression alone induces overt cell proliferation. As observed in the flow analysis, 9a+NRAS F0s displayed greater phenotypic variability. While some were comparable to GFP, 9a and NRAS, 1 in 6 9a+NRAS F0 animals (i.e. n=2 of 12) showed significant marrow hypercellularity (Fig. 4P,T; Fig. S9A).

As an independent assessment of cell number, we resuspended the WKM cells from a non-overlapping set of animals (including those used in the transplant experiments described below) and counted total WKM cell number (Fig. S9B). While NRAS F0s WKM cell counts were higher than those of 9a and GFP, 9a+NRAS F0 counts were significantly greater (NRAS versus GFP, P=0.02; 9a+NRAS versus GFP, P=0.0005). Furthermore, 1 in 8 (n=4/37) 9a+NRAS F0s had counts exceeding the NRAS F0 maximum. This incidence of high 9a+NRAS WKM cell counts was consistent with the incidence of 9a+NRAS hypercellularity in sectioned animals (1 in 6) and AML-like flow profiles (1 in 4) in 9a+NRAS F0s.

Taken together flow cytometry and histological data suggest that NRAS alone and 9a+NRAS expression cause disturbances in marrow haematopoiesis that are sufficiently severe to account for the high rates of mortality in these cohorts. A major distinction between NRAS- and 9a+NRAS-associated pathologies is the progenitor/precursor pool expansion and marrow hypercellularity in 9a+NRAS animals. Overall, these data argue that when NRAS is co-expressed with 9a, the probability of neoplastic transformation is increased.

While the decreased survival, hypercellularity, precursor/progenitor cells expansion and erythrocytopaenia in 9a+NRAS animals represent evidence for an AML phenotype, a more definitive test is to demonstrate that the disease can be transplanted. Current models of cancer evolution posit that leukaemias harbour leukaemia stem cells (LSCs), which can establish leukaemia when transplanted into naïve healthy recipient animals (Chopra and Bohlander, 2019). To test whether either 9a+NRAS- and NRAS-induced blood diseases might harbour LSCs, we performed transplant experiments using the standard allograft procedure employed in published zebrafish AML model studies (Fang et al., 2021; Le et al., 2007; Wang et al., 2023; Xu et al., 2020; Zhao et al., 2018). Specifically, we isolated WKM cells from sick 9a+NRAS and NRAS donors and transplanted these into 5–10 wild-type recipient animals per donor (∼300,000 donor cells per recipient) (Fig. 5; Fig. S10). These wild-type recipients were of the same genetic stocks used to generate our F0 transgenics. Forty-eight hours prior to transplantation, recipients had been exposed to a sub-lethal dose of gamma-radiation to induce transient immunosuppression thereby enhancing the probability of donor cell engraftment (see Materials and Methods for details). The transient nature of this immunosuppression is demonstrated by the fact that irradiated, non-transplanted animals are viable, and their marrow flow profiles are comparable to non-transplanted animals at 60 days post-irradiation (Fig. S10). As controls, WKM cells from healthy age-matched GFP control animals were similarly transplanted.

For 9a+NRAS, 12 donors were transplanted into ∼5 recipients each (producing n=72 recipients total). Of the 12 donors, marrow transplants from 9 induced signs of sickness in their recipients within 40 days post-transplantation (Fig. 5A). Moreover, 27% of these unwell 9a+NRAS recipients (3/11, representing 8 donors) had marrow flow cytometry profiles with AML-like features (P gate>35%, E gate <36%), inferring the presence of LSCs in the donor marrow (Fig. 5B–E,H). Thus, the frequency of sickness in 9a+NRAS recipients (60%) and penetrance of AML-like profiles among these (27%) is in striking agreement with the frequency of sickness and AML-like flow cytometry profiles among sick F0 9a+NRAS animals (∼50% and 27%, respectively). By contrast, all NRAS recipients (n=29, generated from 5 donors) and the GFP control recipients (n=20, generated from 4 donors) had normal flow profiles, remaining healthy, even after 60 days, despite evidence of engraftment in their post-mortem analysis (Fig. 5B–E,F,G). For GFP-cell transplanted recipients, donor cell engraftment was confirmed by detection of GFP-positive cells in control recipient flow; for NRAS cell recipients, engraftment was additionally confirmed by performing PCR for the NRAS transgene and transcript, at 60 days post transplant (Fig. S11). These data suggest that, although NRAS donor cells were present, their engraftment and proliferation were insufficient to establish disease or outcompete resident healthy cells. These data demonstrate that 9a+NRAS donor marrow, but not NRAS donor marrow, harbour LSCs that can re-establish aggressive leukaemic disease.

Although remission induction for RUNX1::RUNX1T1 AMLs is generally successful, overall survival and treatment toxicity remain unacceptably poor, underscoring the need for mechanistic studies to guide new therapeutic strategies. Recent WHO classifications of myeloid neoplasia have progressively emphasised the importance of AML defining genetic abnormalities such as RUNX1::RUNX1T1 and co-occurring molecular alterations, highlighted by removal of the requirement of over 20% blasts in marrow to define leukaemia when AML-defining genetic abnormalities are present (Khoury et al., 2022). This was based on preclinical and clinical data demonstrating that combinations of mutations drive leukaemogenesis and predict clinical behaviour and treatment response more precisely than morphological criteria alone (Duployez et al., 2016; Faber et al., 2016; Jahn et al., 2020; Papaemmanuil et al., 2016). Somatic NRAS mutations are frequent in R::RT1 AML and are linked to inferior prognosis. However, NRAS mutations are associated with treatment failure in multiple other molecular and cytogenetic AML subtypes. Thus, NRAS mutations in AMLs have broad prognostic and therapeutic significance (Amatangelo et al., 2017; Christen et al., 2019; Duployez et al., 2016; McMahon et al., 2019; Winter et al., 2014). These observations prompted us to establish a F0 adult zebrafish R::RT1 transgenic model of AML by co-expressing R::RT1 splice form 9a and NRAS^G12D in zebrafish HSPCs.

We demonstrate that expressing both R::RT1 and NRAS led to a marked decrease in survival associated with marrow hypercellularity, an increase in blast/progenitor cells and a reduction in mature erythrocytes. Although this phenotype was also observed with mutant NRAS alone, the frequency and penetrance of this phenotype was substantially higher with co-expression of NRAS and R::RT1 9a. Moreover, transplantation of F0 9a+NRAS marrow cells into sub-lethally irradiated recipients induced AML with a shorter latency than its parent donor (F0 donors: 9–12 months cf. recipients: <40 days). Transplantability is a hallmark of AML models and is considered the gold standard to prove that the disease observed is indeed leukaemia driven by LSCs (Chopra and Bohlander, 2019). As these leukaemic features were not observed in animals expressing either 9a or NRAS oncogenes alone, we hypothesise that establishment of AML stemness is mediated by the convergence of R::RT1- and NRAS-dependent activities on conserved haematopoietic regulatory networks. Future molecular analysis should reveal the identity of these targeted pathways and the mechanisms underlying their disruption.

The model's strengths include overcoming the insufficiency of R::RT1 alone in the zebrafish system and leveraging a highly specific promoter/enhancer combination to drive spatiotemporal specificity of R::RT1 expression (Abdallah et al., 2021; Nottingham et al., 2007; Tamplin et al., 2015). In contrast to early murine knock-in and larval zebrafish transgenic R::RT1 models, where R::RT1 was widely expressed outside HSPCs (Kalev-Zylinska et al., 2002; Okuda et al., 1998; Yeh et al., 2008; Yergeau et al., 1997), neither 9a nor full-length R::RT1 caused embryonic lethality in our system. We speculate this is because the R1⁺²³ enhancer limits expression of R::RT1 and 9a to HSPCs and delays expression onset until after the lethal period defined by Yeh et al. (2008), who found embryonic lethality was diminished if heat shock promotor activation occurred after 21 hpf. Consistent with this hypothesis, GFP in R1⁺²³:GFP zebrafish only becomes apparent after this sensitive period, from 32 hpf onwards (Tamplin et al., 2015). Enhancer tissue specificity also likely explains why we did not observe RAS-driven oncogenesis outside the haematopoietic compartment (Le et al., 2007).

The use of a Runx-derived enhancer to drive oncogene expression may better recapitulate natural R::RT1/9a expression, as modelling outcomes appear to be sensitive to the oncofusion balance of endogenous RUNX1 and which HSPC subpopulations are targeted (Abdallah et al., 2021; Agrawal et al., 2020; Nafria et al., 2020; Yan et al., 2023). Using F0 animals to achieve reproducible leukaemic transformation in adult fish mimics the mosaic and competitive nature of human cancer environments and potentially reduces selection bias inherent in stable transgenic animals (Avagyan et al., 2021; Callahan et al., 2018). Furthermore, measuring altered haematopoiesis by flow cytometry allowed us to survey a much greater number of WKM cells than possible using morphology alone, improving confidence in our observations.

Our WISH analyses of transgenic larvae show that 9a and NRAS have distinct effects on early haematopoiesis. In the CHT at 3 dpf, 9a expression did not significantly increase stem cell or myeloid commitment. By contrast, NRAS F0 larvae showed increased cmyb HSPC abundance but reduced pu.1 expression. Decreased pu.1 could reflect impaired myeloid lineage commitment. Alternatively, NRAS-dependent pu.1 downregulation could facilitate HSPC expansion as PU.1 also constrains HSPC proliferation to prevent stem cell expansion and exhaustion (Staber et al., 2013). The effects of NRAS in our model are consistent with murine and human data showing NRAS^G12D drives HSPC proliferation, thereby diminishing self-renewal potential (Li et al., 2013; Wang et al., 2011). However, HSPC proliferation was not enhanced in 9a+NRAS larvae. Neither oncogene, singly nor in combination, altered HSPC erythroid lineage commitment in larvae. These observations suggest that the progenitor expansion and erythrocyte deficit seen in 9a+NRAS adults are emergent properties.

AML-like disease (defined here as having >35% progenitor/precursors and transplantable disease phenotype) was preferentially associated with 9a+NRAS coexpression. We speculate that oncogene cooperation conferred a competitive advantage within the context of the adult WKM niche, an area worthy of further investigation (Yamashita et al., 2020). While ∼27% of sick 9a+NRAS F0s developed AML-like pathology, the remainder displayed heterogeneous cell profiles. Such phenotypic variability is potentially because we are expressing two oncogenes and that there will be animal to animal variability in the levels of 9a and/or NRAS expression. It is also possible that additional oncogenic mutations were acquired during leukaemia development, given the mutator activities of R::RT1 and oncogenic NRAS, a question to be explored in future studies.

We could not demonstrate that our phenotype correlates with a high percentage of MPO-positive immature granulocytes. This may be because our oncogene combination can only confer a subset of AML features, and additional mutations are required. For example, murine models co-expressing NRAS^G12D with R::RT1 or splice variant 9a increased AML penetrance and reduced latency without altering the granulocytic character of the AMLs (Abdallah et al., 2021; Zuber et al., 2009). In contrast, expression of NRAS^G12D alone led to monocytic myeloid leukaemias (Li et al., 2011; Wang et al., 2011; Wang et al., 2015). The early expression of R::RT1 in our system could have contributed to our finding that MPO-positive cell frequency was not significantly enhanced in 9a+NRAS animals, as would be expected for R::RT1-driven AMLs. A recent study by Abdallah et al. (2021) demonstrated the impact of R::RT1 induction timing on AML penetrance and state of differentiation. Specifically, induction of R::RT1 in 3-day-old mice produced AMLs without differentiation (M1), whereas later induction at 2 weeks of age favoured M2 AML (i.e. AMLs with differentiation). Induction at even later time points (4, 8 or 16 weeks) resulted in a reduced incidence of M2 AMLs and an increasing trend towards myeloproliferative disease or no disease. The choice of R::RT1 driver may explain differences in AML penetrance and cooperating gene dependence between our model and Abdallah et al. (2021). Abdallah et al. targeted oncofusion expression to the murine HSPC population using an R1⁺²⁴ driver (the same Runx1 +23 element) but fused to a basal heat shock promoter (Ng et al., 2010). Consequently, R::RT1 expression in R1⁺²⁴-positive cells was ultimately controlled by the Rosa26 promoter after Cre^ER-mediated recombination. In contrast, in our model, R1⁺²³ directly drives R::RT1 expression from early embryonic life onwards. While HSPC-targeted expression of R::RT1 alone was sufficient to drive AML in the murine model, we observed that neither 9a- nor full-length R::RT1 alone produced an AML phenotype in zebrafish (Fig. S3). While species-related differences may have contributed to this outcome, we speculate that our model's inherent mosaicism, combined with the relative strength of the ultimate R::RT1/9a expression driver (Rosa26 promoter in the murine model, R1⁺²³ in ours), may be contributory. Single cell sequencing may offer methodologies to explore this important observation in the future.

Our finding that 9a+NRAS impacts myeloid and erythroid adult lineages similarly to R::RT1/9a expression in other models and in humans reinforces that these oncogenes work through conserved molecular targets in zebrafish, leading to impairment of normal haematopoiesis (Kalev-Zylinska et al., 2002; Yeh et al., 2008) and offers multiple avenues for future exploration. t(8;21) AMLs with mutations in kinase signalling, chromatin modifiers or cohesins have a higher risk of relapse (Christen et al., 2019; Duployez et al., 2016). Our model provides a sensitised genetic background for co-expressing additional cohesin complex mutations such as RAD21 to explore the underlying biology further (Jann and Tothova, 2021; Leeke et al., 2014). Different AML model systems have provided complementary insights into how oncogenic NRAS mutations promote AML development, revealing actionable drug targets and pathways. In an in vitro R::RT1 cord blood model, co-expression with NRAS^G12D increased replating potential and reduced apoptosis by increasing levels of anti-apoptotic factor BCL2 (Chou et al., 2011). The BCL-2 inhibitor venetoclax is being trialled in paediatric and adult AMLs. Similarly, in a murine model of inv(16) AML driven by rearrangement of CBFB (core-binding factor subunit beta), NRAS^G12D blocked cell death through upregulation of Bmi and the MEK pathway, the latter rendering this AML sensitive to MEK inhibitors (Xue et al., 2014). In contrast, in a KMT2A-AF9 murine model, NRAS^G12D enhanced KMT2A-AF9-mediated expression of Myb, a master regulator of transcriptional programs underlying self-renewal, through upregulation of the mTOR-AKT pathway (Sachs et al., 2014). These myeloid model systems helped elucidate the diverse mechanisms underlying oncogenic cooperation and identified potential predictive biomarkers that could guide patient care. Here, we describe a novel adult zebrafish model of a common AML subtype to further unravel the molecular and therapeutic implications of oncogene cooperation.

All zebrafish (Danio rerio) strains were maintained under standard husbandry conditions and followed protocols approved by the Animal Ethics Committee of the University of Auckland (AEC22627). Wild-type (AB) zebrafish were obtained from the Zebrafish International Resource Centre (ZIRC).

Tol2-based expression vectors were used for transgenesis of zebrafish (Kwan et al., 2007). Middle entry vectors (pMEs) encoding 9a and 9a+NRAS were generated using a multi-cistronic base pME vector, pME MCS 2x P2A GFPNRAS. This base vector was generated by inserting a customized gene block (IDT) between the att sites of pENTR/D-TOPO (Invitrogen) (Fig. S1). This insert consists of the two copies of the P2A (porcine teschovirus-1 2A) ‘ribosome-skipping’ sequence, each proceeded by unique restriction enzyme sites for inserting GoIs. The P2A sequences are followed by an in-frame ORF encoding EGFP fused to human NRAS^G12D (GFPNRAS^G12D, abbreviated here to NRAS) and a termination codon. To generate pME 9a+NRAS (9a-P2A-GFPNRAS^G12D), a Smi1-Zra1 PCR fragment encoding 9a with a canonical Kozak sequence (Grzegorski et al., 2014) but no termination codon was generated using R::RT1 9a gene-specific primers (Table S1) and pMIG AE9a plasmid as the R::RT1 DNA template (Desai et al., 2020). The 9a PCR fragment was then cloned into the Smi1 and Zra1 site of pME 2 x P2A NRAS (Fig. S1). pME 9a (9a-P2A-GFP) was derived from pME 9a+NRAS by cutting with Xho1 and Pac1 to excise NRAS, followed by end-filling and re-ligation. To generate pME NRAS, a PCR fragment containing GFPNRAS^G12D with a canonical Kozak and a stop codon, was generated using gene-specific primers (Table S1) and pME 2x P2A NRAS as the template. This fragment was then cloned into pENTR/D-TOPO (Thermo Fisher Scientific), as per the manufacturer's instructions. The final pDEST expression constructs were generated by performing Gateway LR reactions (Life Technologies), as per the manufacturer's instructions, using 5′ entry vector p5E R1⁺²³ (a gift from Owen Tamplin and Leonard Zon, Addgene plasmid # 69602; http://n2t.net/addgene:69602; RRID:Addgene_69602), 3′ entry vector p3E polyA (zebrafish Tol2kit plasmid ID 191), destination vector pDEST Tol2 CG2 (ID 204; with transgenesis heart marker cmcl:EGFP) and one of the following middle entry vectors: pME GFP (ID 180), pME 9a+NRAS, pME NRAS, pME 9a or pME R::RT1-9a-NRAS. The presence of a P2A sequence between 9a and NRAS ORFs enabled the generation of discrete 9a and GFPNRAS translational products (Liu et al., 2017; Szymczak et al., 2004). In 9a+NRAS vectors, NRAS was strategically placed downstream of 9a and the P2A sequence, exploiting the second position effect observed with bi-cistronic constructs, whereby the gene in the second position is translated at a lower frequency than that in the first, thus serving to reduce NRAS expression levels and the likelihood of NRAS dominating any resultant phenotype (Liu et al., 2017). IRES-GFP-based expression constructs were generated by recombining p5E R1⁺²³, pME full-length R::RT1 or 9a, p3E-IRES-EGFPpA (ID 195) and pDEST Tol2 CG2 (ID 204).

Transgenic zebrafish were generated by micro-injection of pDEST expression vectors DNA (Fig. 1) into one-cell stage wild-type AB embryos (Suster et al., 2009). Wild-type embryos for injection were collected from naturally spawned wild-type zebrafish. Injections were performed using pre-pulled microinjection needles calibrated to inject at 1 nl/pulse. The injection cocktail consisted of pDEST expression plasmid DNA (final concentration of 25 ng/µl), transposase mRNA (25 ng/µl) and Phenol Red as a visible marker. At 24 h post injection, embryos were screened for GFP expression in the heart. Positive embryos were either used for WISH experiments or raised to adulthood using standard zebrafish husbandry procedures.

WISH was performed on 4% PFA-fixed embryos as previously described (Thisse and Thisse, 2008). RNA probes were generated by linearisation of vectors containing the relevant cDNA sequence (see Table S2 for probe sequence information). Digoxigenin-labelled antisense probes were synthesised using an RNA Labelling Kit (SP6/T7; Roche). Staining was revealed with NBT/BCIP or INT/BCIP substrate (Roche). Two technical replicates were performed for each probe (∼30 animals/genotype/probe/replicate). WISH embryos were imaged in 100% glycerol, using a Zeiss Axio Zoom microscope. CHT staining intensity was quantified as described previously (Dobrzycki et al., 2020) and scored blind of genotype. Specifically, all technical replicates were imaged under identical lighting and exposure conditions. For each animal a ‘corrected CHT signal’ was obtained (the raw CHT signal ROI − a background pixel intensity ROI of the same size). For each probe, the mean ‘corrected CHT signal’ for GFP was calculated and used to normalise the CHT values for all other genotypes of that technical replicate. These normalised ‘Signal Intensity’ values were plotted in Fig. 1 and Fig. S2.

Sick or well zebrafish adults were humanely euthanised by placing them in an ice bath. To isolated whole kidney marrow (WKM) cells, kidney marrows were dissected out then transferred to cold FACS buffer consisting of 0.9 x PBS (Gibco) and 5% FBS (Moregate Biotech). Marrow cells were manually dissociated by pipetting up and down then passing through a 40 µM cell strainer (Falcon).

Immediately prior to flow cytometry, 1 µl 1% Propidium Iodide (Sigma-Aldrich) was added to dissociated WKM cells in FACS buffer and cells were briefly vortexed to mix. Cells (100,000 events) were analysed for size (Forward Scatter, FSC) and granularity (Side Scatter, SSC) using a BD LSRII Flow Cytometer (Becton Dickinson Biosciences). Gates were drawn around the distinct populations of erythrocytes, myelomonocytes, lymphocytes and precursors/progenitors (Traver et al., 2003). After flow analysis, the remaining cells were either used for smears (applied to slide manually or using a cytospin machine), RNA isolation and/or cell transplantation. For cell sorting, gated populations were sorted into either 25 µl FACS buffer 1 mM EDTA (myelomonocytes, progenitor/precursors, lymphocytes, R1⁺²³:GFP-positive cells) or FACS buffer with 2% (v/v) heparin (erythrocytes). All flow cytometry data was acquired using BD FACSDiva software (v6.1.1) and analysed with FlowJo (10.8.1).

Sorted cells (1×10⁵) in FACS Buffer [1 mM EDTA or 2% (v/v) heparin] or unsorted WKM cells (3×10⁵) in FACS buffer, were spun onto slides at 120×g, 4°C using the Aerospray^® Hematology Pro (ELITech Group Inc.) and allowed to dry overnight before staining with May Grünwald-Giemsa stain (ELITech Group Inc.) applied using the cytospin machine.

WKM smears were air dried for 24 h before staining for myeloperoxidase. 3,3′-Diaminobenzidine (DAB, Sigma-Aldrich) was added to MPX Buffer (50 mM Tris-HCl pH 7.4) and mixed well prior to adding 3% H₂O₂. Slides were then flooded with cold buffered formalin acetone (5 mM Na₂HPO₄, 30 mM KH₂PO₄, 45% (v/v) acetone, 25% (v/v) 37% formalin) for 30 s before rinsing in distilled water. Slides were immersed in the DAB solution for 15 min at room temperature, then rinsed with distilled water and counterstained in Mayer's modified Haematoxylin Solution (Abcam). Following imaging, cells were counted using ImageJ (Fiji).

Adult zebrafish were euthanised on ice before placing in 4% (v/v) PFA (EMS) in the fridge from 4 days to 2 weeks. After this, all PFA was removed and replaced with 0.25% (w/v) EDTA to decalcify for 5–7 days. Post decalcification, animals were left in 75% EtOH for 3–7 days before being embedded and sectioned (10 µM thickness). Sagittal sections were then stained in Haematoxylin and Eosin.

To image live adult zebrafish, animals were anaesthetised in 4% (v/v) Tricaine (Sigma-Aldrich). Imaging was performed using a Leica MZ10F Microscope. Images were processed using ImageJ (Fiji). Sections and sorted or unsorted kidney marrow slides were imaged using the Zeiss Axio Imager M2 (ZEISS) and MetaMorph^® software (Molecular Devices, LLC.).

WKM cells were isolated and resuspended in FACS buffer as described for flow cytometry. A minimum of 500,000 cells were recovered by centrifugation and the pellet resuspended in Qiagen RLT buffer supplemented with 2-mercaptoethanol (Merck). RNA was isolated using Qiagen Rneasy Mini columns according to the manufacturer's instructions. The purified RNA was then Dnase-treated using ezDNase Enzyme (Invitrogen). A portion of Dnase-treated RNA was kept aside as a “No RT” control, while the remainder was used in first strand cDNA synthesis reactions with Superscript IV (Invitrogen), priming with random hexamers oligos (Invitrogen). PCR was performed using the KAPA2G Robust HotStart PCR Kit (Kapabiosystems) with 2 µl of cDNA synthesis reaction (+RT) or No RT template, per PCR reaction and the gene-specific primers for 9a, NRAS (this study) or ef1a (Oehlers et al., 2011) (Table S3). PCR products were run on Tris Acetate EDTA pH 8-buffered 10% polyacrylamide gels alongside a GeneRuler 50 bp DNA ladder (Thermo Fisher Scientific).

Healthy wild-type zebrafish were gamma irradiated with a sublethal dose of 18 Gy, 48 h prior to transplantation. Whole kidney marrows were isolated from prospective donor animals. WKM cell resuspension counts were determined by counting Trypan Blue-stained cells manually on a haemocytometer. Irradiated recipients (5–10 animals) were individually anaesthetised in 4% (v/v) Tricaine then injected intra-peritoneally with 3×10⁵ of WKM cells per recipient.

The data analysed corresponds to animals from 2 to 4 injected batches per genotype, generated over the course of 4 years. All statistical analysis was done using GraphPad Prism version 9 for the Mac (GraphPad Software). Statistical tests used in analysis were unpaired t-test, one-way analysis of variance (ANOVA) with Tukey's multiple comparison, Kruskal–Wallis or a Log-rank (Mantel-Cox) test.

We thank Stephen Edgar for assistance with flow cytometry and Satya Amirapu for tissue sectioning (BIRU Core Facility, FMHS, University of Auckland); Carolyn Bunkall (Lab Plus, Auckland District Health Board) for the myeloperoxidase staining protocol; Tanja Linnerz, Jonathan Astin and Christopher Hall for helpful suggestions regarding zebrafish husbandry, generation of transgenic lines and for sharing gateway plasmids and WISH reagents and expertise; James Hearn and Jacqueline Ross for microscopy advice; Yih Jian Sun, Shantanu Patke and Alhad Mahagaonkar for zebrafish technical support; Julia Horsfield and Alan Davidson for helpful comments on the manuscript.