rasa1-related arteriovenous malformation is driven by aberrant venous signalling

The vascular tree relies on an orderly branched structure of progressively sized vessels to effectively transport nutrients and oxygen to cells. Vascular malformations such as arteriovenous malformations (AVMs), angiomas, haemangiomas, aneurysms and vascular tumours are the result of altered developmental vascular signalling that disrupts the tree-like structure of the vascular system. Capillary malformation-arteriovenous malformation (OMIM: 608354; CM-AVM1) is caused by mutations in RASA1, a Ras GTPase activating protein (Eerola et al., 2003). RASA1 is clearly important for vascular development across species, as loss of Rasa1 in mice and rasa1 knockdown in zebrafish leads to disordered vasculature (Henkemeyer et al., 1995; Kawasaki et al., 2014; Lubeck et al., 2014).

The most prominent presentation of human CM-AVM is capillary malformation (CM), which appears in ∼95% of patients (Duran et al., 2018; Lapinski et al., 2017; Revencu et al., 2008). CMs are cutaneous beds of permanently dilated capillaries that appear as a purple-red port-wine ‘stain’ on the skin. About one-third of patients also have an AVM that directly shunts blood between arterial and venous systems, by-passing normally interceding capillary beds. AVMs are fragile, prone to rupture and difficult to treat. The localized nature of these vascular malformations appears to be the result of a secondary somatic mutation that is permissive of lesion formation. Although it is ubiquitously expressed, RASA1 is necessary in endothelial cells for vascular homeostasis (Henkemeyer et al., 1995; Lapinski et al., 2012; Lubeck et al., 2014). The Ras-GAP domain of RASA1 inhibits Ras activation. Loss of the inhibitory function of RASA1 leads to overactivation of Ras signalling pathways.

The signalling pathway upstream of Rasa1 has become clearer through genetic analysis. In zebrafish, rasa1a morpholino knockdown has similar vascular defects to knockdown of the EphB4 kinase (ephb4a morphants), including vessel enlargement in the caudal venous plexus (CVP), lack of caudal blood flow and overabundance of intersegmental veins (ISVs) at the expense of intersegmental arteries (ISAs) (Kawasaki et al., 2014). Loss of EphB4 in mice leads to vascular malformations and, in humans, genetic changes in EPHB4 cause a strikingly similar disease called CM-AVM2 (OMIM: 618196), suggesting that EPHB4 and RASA1 act in the same genetic pathway (Amyere et al., 2017; Chen et al., 2022; Gerety et al., 1999; Gerety and Anderson, 2002). RASA1 binds EPHB4 in vitro in cultured cells (Kawasaki et al., 2014). Whether Rasa1 AVMs arise from defects in vein determination or formation is unknown.

Downstream of RASA1, RAS signalling can activate two downstream pathways, either MEK/ERK or PI3K/AKT/mTORC, both of which are key in arteriovenous specification; each has evidence of being overactivated in different forms of AVMs (Queisser et al., 2021). Inhibition of MEK1/2 reverses haemorrhage and oedema in Rasa1 mutant mice (Chen et al., 2019), while zebrafish rasa1a morphant embryos show fewer venous intersegmental vessels after inhibition of PI3K/mTORC (Kawasaki et al., 2014). However, it remains unclear which pathways downstream of RASA1 specifically drive AVM formation, as neither mouse nor zebrafish morphant models were studied with respect to AVM formation (Henkemeyer et al., 1995; Lubeck et al., 2014). Mice also show lymphatic valve defects resulting from cell death after Rasa1 inactivation at an adult stage (Lapinski et al., 2017), and cutaneous haemorrhage, oedema and endothelial cell death when inactivated conditionally at E12.5-E14.5, but not an AVM phenotype. Although zebrafish rasa1a morphants show an AVM, previous phenotypic analysis focused on a lack of caudal flow and intersegmental vessel defects.

Here, we have used a zebrafish model of Rasa1 CM-AVM to understand the real-time development of AVMs, and their effect on blood flow and signalling. rasa1 genetic mutants develop an AVM in the CVP. The AVMs progressively enlarge, disrupting blood flow and filling with stagnant blood; the mutation is lethal by 10 days post fertilization (dpf). We observe that both blood flow velocity and pulsatility are affected by the cavernous malformation, resulting in slower flow in the AVM and a substantial drop in pulsatility from the dorsal aorta to the caudal vein. Correspondingly, expression of the flow-responsive transcription factor klf2a is diminished, suggesting that changes in flow velocity and pattern might contribute to the progression of vessel malformations through changes in mechanosensory signalling. However, modulating flow does not rescue AVM formation, indicating that AVM initiation is flow-independent. At the cellular level, an increase in intraluminal pillars in rasa1 mutants in indicative of incomplete intussusceptive angiogenesis. Vein diameter and vein cell size are increased, suggesting abnormal endothelial cell behaviour. Molecularly, we see preferential activation of pERK in the vein of rasa1 mutants and rescue of blood vessel patterning with the inhibition of MEK/ERK signalling. Together our data point to a crucial function for Rasa1 in normal blood vessel patterning of the venous endothelium.

There are two RASA1 orthologs in zebrafish. We used CRISPR-Cas9 to create rasa1a^ca35 and rasa1b^ca59 mutants (Fig. S1). Single rasa1a or rasa1b mutants are homozygous viable. rasa1a mutants have a mild phenotype (Fig. S2) with small ectopic shunts between the dorsal aorta (DA) and caudal venous plexus (CVP) at 30 h post-fertilization (hpf) that completely resolve by 48 hpf. rasa1b single mutants have no observable phenotype. As a result of the mild single mutant phenotypes, rasa1a^−/−;rasa1b^−/− double mutants (hereafter, rasa1^−/− or rasa1 mutant) were used to characterize vascular phenotypes.

The CVP of the zebrafish tail is homologous to mammalian vascular plexi where a surplus vessels develop initially that are gradually refined into an efficient vascular network. Patterning of vessels in the CVP does not follow a strict pattern in contrast to other highly studied beds such as the intersegmental vessels in zebrafish. We used confocal microscopy to characterize vessel structure at two key stages of development. 30 hpf is a crucial point in CVP development as the caudal vein (CV) undergoes angiogenic sprouting between 24 hpf and 30 hpf. By 30 hpf, the CVP has expanded ventrally and blood circulation within the vessel bed is robust. Between 30 hpf and 48 hpf, pruning and remodelling of the CVP occurs. To detect the AVMs, we measured the diameter of the largest vein in the CVP. rasa1 mutants show vein enlargement (defined as ≥1.5× average largest vein in wild type; Fig. 1A) as early as 30 hpf (wild type, 43.5 µm; rasa1^−/−, 73.5 µm; 86% penetrance, P=0.0018, Fig. 1D-G,N,P). At 48 hpf, the cavernous AVM is consistently located at the posterior of the tail plexus, connecting the rostral DA to the rostral CVP, subsuming many venous capillaries normally present in the CVP. The AVMs are obvious at 48 hpf (largest vein in wild type is 22.3 µm versus 108.4 µm in rasa1, P<0.0001, Fig. 1H-N,P; 69% penetrance). rasa1 mutant embryos develop severe oedema by 5 dpf and lethality by 10 dpf (Fig. S1). The DA upstream of the lesion does not differ in size between wild type and mutants at either time point, suggesting the AVM is a localized defect (Fig. 1A-M,O). At 30 hpf, wild-type DA is 17.5 µm versus 17.4 µm for rasa1^−/− (P>0.99). By 48 hpf, wild-type DA is 21.4 µm versus 21.4 µm for rasa1 (P>0.99). Vein enlargement is pronounced in mutants when compared with wild type when rendered in Simpleware (Fig. 1D,G,J,M). No AVMs were detected in the other vascular beds, including the brain; however, the lumenization of the central arteries appears to be impaired in rasa1 mutants (Fig. S3).

We next determined how the abnormal vascular architecture leads to altered blood flow patterns as the role of blood flow in AVM development and progression cannot be easily studied in other models. rasa1 mutant heart rate is not changed at either 30 hpf or 48 hpf, suggesting the heart output is normal (30 hpf, P=0.17; 48 hpf, P=0.13, Fig. S4). Using high-speed video imaging, we quantified blood flow at 30 hpf and 48 hpf (Movies 1-4). Mean velocities are calculated in the DA and the CV proximal to the flow return (Fig. 2A, Fig. S3). Heatmaps of representative single embryos (Fig. 2B-E) and averaged from multiple embryos (Fig. 2F-I) illustrate consistent flow changes in rasa1-dependent malformations. We focused on three areas of the vasculature: (1) the DA, which is not expected to differ between mutants and wild types as it is upstream of the malformation; (2) the caudal vein and (3) the point where the DA ‘turns’ 180° into the CVP, which we named the ‘flow return’ (the site where the malformation develops in mutants).

Blood flow rates vary according to the size and location of the vessel and developmental stage. In 30 hpf wild types, there is typically high velocity flow in the DA (573.1 µm/s), slower flow in the return (483.9 µm/s) and fast flow in the CV (593.2 µm/s, Fig. 2B,F,N, Fig. S3). Thus, in wild types the flow velocity in the DA is similar to the CV (P=0.62) at this stage. In rasa1 mutants, the average velocity in the DA is 591.1 µm/s, while CV is significantly faster at 731.7 µm (P=0.0062, paired t-test, Fig. 2D,H,N, Fig. S3). Mutants have a lower average flow speed at the flow return relative to the DA and CV (522.3 µm/s), but did not significantly differ from wild type (P=0.90). Thus, although average flow velocity in the DA and return is similar between wild types and mutants, there is faster flow in the mutant CV (DA: P>0.99, CV: P=0.034).

At 48 hpf, wild-type DA and CV have similar average velocities (DA, 568.2 µm/s; CV, 588.2 µm/s; P=0.73; Fig. 2C,G,O, Fig. S3). In rasa1 mutants at 48 hpf, the AVM is further enlarged. Despite this, there are no significant differences between DA or CV velocities (P=0.085; DA velocity, 536.8 µm/s; CV, 616.4 µm/s; Fig. 2E,I,O, Fig. S3). The key difference between mutants and wild types at this stage is the flow return velocities (436.3 µm/s in wild type versus 203.3 µm/s for rasa1, P=0.038, at position 4) likely because of the malformation. There are no significant changes in DA or CV velocities between wild types and mutants (DA, P=0.98; CV, P=0.98).

Owing to the pulsatile nature of blood flow, average velocities do not capture the complexity of flow changes that occurs through the cardiac cycle. At 30 hpf and 48 hpf, the minimum velocities during diastole in wild types are not significantly different in the DA (30 hpf, 74.4 µm/s; 48 hpf, 109.1 µm/s) or CV (30 hpf, 156.3 µm/s; P=0.42; 48 hpf, 189.7 µm/s; P=0.075; Fig. 2J,K). In contrast, at 30 hpf in rasa1 mutants, there is a drastic difference in minimum velocities between the two vessels, with the DA minimum velocity at −35.9 µm/s versus the CV minimum velocity at 309.0 µm/s (P<0.0001; Fig. 2J). A negative number indicates backflow in the DA. At 48 hpf, the same trend is observed, with the minimum velocity in the DA of mutants (−28.6 µm/s) being significantly lower than the CV (277.2 µm/s, P<0.0001; Fig. 2K). When comparing rasa1 mutants with wild types, the minimum DA velocity is significantly lower in mutants at both timepoints (30 hpf, P=0.026; 48 hpf, P<0.0001) but is higher in the mutant ventral vein at both timepoints (30 hpf, P=0.028; 48 hpf, P=0.045; Fig. 2J,K).

We also examined maximum velocities during systole. In wild types, we find no significant difference in velocities at 30 hpf or 48 hpf (30 hpf, P=0.46; 48 hpf, P>0.99; Fig. 2L,M). At 30 hpf, the maximum velocities of rasa1 mutants did not differ between the two vessels (P=0.97; Fig. 2L). At 30 hpf, the maximum velocity is not significantly different in the DA or CV between wild type and mutants (DA, P=0.45; CV, P=0.089). By 48 hpf, rasa1 mutants have an increase in DA maximum velocity (1394 µm/s) over the CV (1052 µm/s, P=0.017; Fig. 2M). rasa1 DA maximum velocity is also significantly higher than wild type (1086 µm/s, P=0.027) with no significant difference in the CV (P=0.99).

Taken together, the AVM affects velocity extremes. At both timepoints, there is an increased minimum velocity in the CV of mutants and a decreased minimum velocity in DA. This suggests the malformation creates flow velocity patterns that are very different to what normal vessels would experience.

Pulsatility from discrete heart beats is reflected in the amplitude of change from maximum to minimum velocity (Fig. 3A,B and Eqn 1). We created representative single embryo pulsatility heatmaps (Fig. 3C-F) and averaged pulsatility heatmaps from seven embryos (Fig. 3G-J) to demonstrate variation in flow pulsatility across the DA and CVP. In wild types at 30 hpf, heatmaps reveal higher pulsatility in the DA (1.9) and lower pulsatility in the CVP (1.4, P=0.0006; Fig. 3C,G,K). By 48 hpf, flow pulsatility evens out across the two vessels in wild type, with no difference between the DA (1.7) and ventral vein (1.6, P=0.68; Fig. 3E,I,M). In contrast, 30 hpf rasa1 mutants have elevated pulsatility in the DA (2.2) relative to the ventral vein (1.2, P<0.0001; Fig. 3D,H,K). By 48 hpf, pulsatility in the DA of rasa1 mutants remains elevated relative to the ventral vein (DA, 2.7; CV, 1.3; P<0.0001; Fig. 3F,J,M). At both 30 hpf and 48 hpf, flow in the DA of mutants is more pulsatile in rasa1 mutants that in wild types (30 hpf: P=0.023, 48 hpf: P<0.0001) with no change in ventral vein pulsatility (30 hpf, P=0.23; 48 hpf, P=0.19; Fig. 3C-J,K,M). To resolve fine changes in pulsatility, we mapped pulsatility by positionally sampling three locations in the DA (positions 1-3), one location in the flow return (position 4) and three locations in the ventral vein (positions 5-7) (Fig. S4). At 30 hpf there are no positional differences in PI between wild types and mutants. However, in mutants there is an increase in PI at 48 hpf in the three DA positions (positions 1-3, p₁=0.040, p₂=0.0031, p₃=0.00014, Fig. S4), agreeing with the bulk flow data.

The pulsatility index (PI) drop is a measure of the differences between two vessels. We find that the AVM results in a significant PI drop from the DA to the ventral vein (Fig. 3A,B and Eqn 1). At 30 hpf, the average drop in PI from the DA to the VV in wild types is 22.7 compared with 43.2 in rasa1 mutants (P=0.0003). By 48 hpf, the drop in PI is reduced in wild types (8.5), but more drastically elevated in mutants (52.8, P<0.0001). This indicates that, as the malformation expands, pulsatility is more severely impacted.

As we detected large changes in velocity and pulsatility, it follows that mechanosensation by endothelial cells and regulation of downstream flow-responsive genes could be altered in rasa1 mutants. Expression of klf2a, a transcription factor that is downregulated by flow in the trunk, is altered at 36 hpf and 56 hpf. Wild-type embryos have strong klf2a staining in both the artery and vein of the trunk. rasa1 mutants have lower klf2a staining in the trunk, with expression ending more anteriorly (Fig. S5).

Hypothesizing that blood flow might contribute to lesion formation, we manipulated blood flow and shear stress. To increase flow, we used the non-selective β-adrenergic receptor agonist isoprenaline (10 µM) to increase heart rate over the 24-48 hpf developmental window when AVMs form. rasa1^−/− treated with isoprenaline had a higher heart rate (6.6% higher, P=0.0024; Fig. 4N), but were not rescued and had significantly larger vascular lesions than both wild-type and rasa1^−/− treated with vehicle (DMSO) (rasa1^−/−_iso, 158.6 µm; WT_DMSO, 20.5 µm; rasa1^−/−_DMSO, 95.2 µm; P<0.0001; Fig. 4A-D,G-I). Thus, increasing blood flow does not rescue lesion formation.

Decreasing blood flow in rasa1^−/− with 2 µM of the L-type calcium channel blocker nifedipine decreased heart rate in wild types by 11.2% (P=0.0074, unpaired t-test; Fig. 4O). However, lower flow did not rescue AVM formation in rasa1 mutants when compared with wild types (rasa1^−/−_Nif, 67.4 µm; WT_Nif, 22.5 µm; P<0.0001; Fig. 4J-M,P), or compared with mutants treated with DMSO (rasa1^−/−_DMSO, 75.7 µm; P=0.83; Fig. 4K,M,P).

As there is a pool of stagnant blood in lesions, we assessed whether shear stress plays a role in AVM size or whether the accumulation of blood in the CVP worsens lesion size. We ablated erythrocytes before AVM formation by treating embryos at 24 hpf with 0.5 µg/ml phenylhydrazine (PTZ), a compound highly reactive with oxyhaemoglobin, to acutely ablate erythrocytes and decrease shear stress from 24 to 48 hpf. Visual inspection showed that erythrocytes were eliminated. rasa1^−/− treated with PTZ showed no significant change in AVM size in comparison with rasa1^−/− controls (rasa1^−/−_PTZ, 130.5 µm; P=0.29; Fig. 4A,B,E,F,G-I) or wild types (WT_PTZ, 20.8 µm; P<0.0001). There was no difference in vein size between wild types treated with DMSO or PTZ (P>0.99).

Together, our data demonstrate that although blood flow is strongly altered in the rasa1 AVM, affecting the upstream input and downstream output vessels, resulting in greater pulsatility and altered mechanosensory signalling, these flow changes do not cause rasa1 AVMs, as manipulation of blood flow and shear stress in rasa1 mutants did not prevent or rescue AVM development.

Differences in endothelial behaviour might underlie vascular malformation development and progression. For example, failed intussusceptive angiogenesis, angiogenic sprouting or migration of endothelial cells could result in vascular malformations. CVP remodelling relies on both intussusceptive angiogenesis and angiogenic sprouting. Intraluminal pillars are formed as vessels are split in intussusceptive angiogenesis; thus, we quantified pillar formation at 30 hpf and 48 hpf. At 30 hpf we saw no significant difference between wild types and rasa1 mutants in pillar numbers (wild type, 5.1 pillars; rasa1^−/−, 10.5 pillars; P=0.09; Fig. 5A-B′,E). However, at 48 hpf, rasa1 mutants had significantly more pillars than wild types (wild type, 4.6 pillars; rasa1^−/−, 19.0 pillars, P<0.0001; Fig. 5C-E), suggesting a block in intussusceptive angiogenesis.

To assess whether angiogenic sprouting was also impaired, we quantified the angiogenic migration distance of wild-type and rasa1 CVP endothelial cells by imaging the CVP at key developmental windows and CVP migration speed through timelapse imaging. Measuring the furthest extent of the CVP from the DA at different stages, we find no difference in CVP migration distance at 30 hpf (wild type, 98.0 µm; rasa1^−/−, 101.7 µm; P=0.89; Fig. S6). However, the CVP is significantly larger at 48 hpf in mutants than in wild types (wild type, 101 µm; rasa1^−/−, 128 µm; P<0.0001; Fig. S6), which might reflect the swelling of the AVM and not true migration. There is no change in mutant migration speed from timelapse imaging between 24 and 30 hpf, a critical time window in CVP development as the CV actively sprouts to form the CVP (wild type, 1.7 µm/h; rasa1^−/−, 0.99 µm/h; P=0.48; Fig. S6, Movies 5 and 6). There was no significant difference in filopodia numbers at the outgrowth front of the CVP in rasa1 mutants (P=0.82; wild type, 15.6; rasa1^−/−, 16.4; Fig. S6).

rasa1 is expressed ubiquitously so we tested whether AVMs could be found in another venous vascular plexus. We imaged the subintestinal venous plexus (SIVP), which develops slightly later and expands between 58 hpf and 76 hpf over the surface of yolk sac (Goi and Childs, 2016). We found no obvious vessel malformations or changes in SIVP migration distance at 58 hpf (P=0.45; Fig. S6) or at 76 hpf (P=0.20). No AVMs were observed in the animals in other locations, including the cerebral circulation. Thus, the CVP appears particularly sensitive to AVM development after loss of rasa1.

Overall, these results show that mutants have impaired CVP remodelling because of disrupted intussusceptive angiogenesis, the result of which is residual pillars in rasa1 mutants but no obvious impairment of angiogenic sprouting in rasa1 mutants.

AVMs could develop from an over-proliferation of endothelial cells or a collapse of a plexus into a singular vessel due to apoptosis. However, we find no change in total endothelial cell number between wild types and rasa1 mutants at either timepoint (30 hpf, P=0.99; 48 hpf, P>0.99; Fig. 6K). Proliferation, as detected by phospho-histone H3 (PHH3) staining shows no significant change at 30 hpf or 48 hpf in mutants versus controls (30 hpf, P=0.98; 48 hpf, P>0.99; Fig. 6A-D,I). We saw no change in apoptotic cells using immunostaining for cleaved caspase 3 (30 hpf, P>0.99; 48 hpf, P>0.99; Fig. 6E-H,J). Overall, this suggests that malformations in rasa1 mutants do not arise from differences of endothelial cell number.

Upregulation of MEK/ERK signalling was observed in vascular anomalies, including mouse and human RASA1 mutant cells. pERK immunostaining, a readout of active MEK/ERK signalling, reveals fewer pERK-positive nuclei in the DA of rasa1 mutants at 30 hpf (1.6 cells) versus wild types (4.3 cells, P=0.03; Fig. 7A,B). In contrast, we found a significant increase in pERK nuclei in the vein (wild type, 2.0 pERK positive nuclei; mutant, 5.1; P=0.0020). There is no change in the number of pERK-positive nuclei in the intersegmental arteries (P=0.52) sprouting from the DA.

As our data indicate a venous-specific upregulation of pERK, we tested whether a rasa1-like cavernous lesion could be induced by activation of MEK signalling in the venous endothelium. For these experiments, we used a sensitized background of a rasa1a single mutant that does not develop AVM. We find that AVMs are induced by ectopic expression of activated map2k2a^S219D in venous vascular endothelium under the mrc1a promoter (mrc1a:map2k2a^S219D) in rasa1a mutants but not in wild type. Largest vein diameters measured 123.9 µm by 48 hpf in map2k2a-injected rasa1a mutants when compared with injected wild-type embryos (24.4 µm, P<0.0001; Fig. 7C,F,G). No change in vessel size was seen between uninjected wild-type embryos (24.7 µm) and uninjected rasa1^−/− fish (24.8 µm, P>0.99; Fig. 7C,D) or injected wild types (P>0.99; Fig. 7C,D,F), confirming that rasa1a mutants show no phenotype without an additional perturbation to the pathway.

Expression of constitutively active map2k2a^S219D using the pan-endothelial promoter kdrl (kdrl:map2k2a^S219D) also resulted in AVM formation in rasa1a^−/− mutants (rasa1a^−/−_inj, 128.2 µm; WT_inj, 36.1 µm; P<0.0001; Fig. S7). No change in vein diameter was observed between uninjected and injected wild types or uninjected rasa1a^−/− controls (wild type, 24.7 µm; P=0.97; WT_inj, 36.1 µm; P=0.93; rasa1a^−/−, 24.8 µm; P>0.99; Fig. S7).

Having identified that Rasa1 activates pERK, we also wanted to test the effect of genes upstream of Rasa1 on AVM formation. KRas leads to brain AVMs (Nikolaev et al., 2018), thus could be an important mediator in RASA1 CM-AVM signalling. Alternatively, RRSas is the most strongly inhibited Ras by RASA1 in vitro (Garrett et al., 1989; Li et al., 1997), but is not associated with vascular malformations. Endothelial expression of constitutively active human KRAS^G12V and zebrafish Rras^Q61R using kdrl:KRAS^G12V and kdrl:rras^Q61R resulted in AVMs when compared with injected wild types (rasa1a^−/−_inj.KRAS, 135.2 µm; WT_inj, 27.7 µm; P<0.0001; rasa1a^−/−_inj.rras, 116.5 µm; WT_inj, 22.5 µm; n=7; P<0.0001; Fig. S8).

As MEK/ERK signalling is implicated in artery specification, we used in situ hybridization to assess whether initial artery and vein specification are normal (Fig. S5). Both arterial gridlock (grl), efnb2a and venous ephb4a expression at 30 hpf appear identical in wild-type and rasa1 mutants, suggesting early arteriovenous specification is unaffected (Fig. S5). We also used an artery transgenic, Tg(flt1:citrine), and a venous transgenic, Tg(COUPTFII-965-e1b-GFP), to visualize whether expression of artery and vein markers was altered in the AVM. No change in flt1 and COUPTFII (nr2f2) expression was evident between wild-type and rasa1 AVMs at 38 hpf (Fig. S9).

To test whether MEK inhibition could prevent AVM formation, we applied 5 µM SL327 (a MEK1/MEK2 inhibitor) from 24 to 48 hpf. The largest veins of control embryos treated with DMSO showed enlarged AVMs (wild type, 18.9 µm; rasa1^−/−, 93.2 µm; P<0.0001; Fig. 8A,B,I). MEK inhibition reduced enlarged vessels in rasa1^−/− to a calibre that was indistinguishable from wild type (WT_SL327, 19.8 µm; rasa1^−/−_SL327, 33.7 µm; P=0.64; Fig. 8C,D,I). A second pathway reported to be activated in rasa1a morphants is PI3K/mTOR. However, inhibition of PI3K/mTOR signalling by 0.5 µM BEZ235 did not rescue AVM formation in rasa1^−/− mutants when compared with wild types (WT_BEZ, 24.2 µm; rasa1^−/−_BEZ, 76.2 µm; P<0.0001, Fig. 8G,H,J). Treating rasa1 mutants with 5 µM and 10 µM of BEZ235 did not rescue AVM formation but impaired ISV sprouting and anastomosis, as well as CVP outgrowth in both mutants and wild types (Fig. S10). These data show that MEK/ERK is involved in AVM formation whereas PI3K/mTOR activation is not. Taken together, our data suggest that directly increasing MEK signalling in the vein or whole endothelium, or increasing MEK through Ras mediators drives cavernous lesion formation, mimicking the rasa1^−/− phenotype. MEK signalling is therefore a key target in the vein downstream of rasa1.

Cell size changes might underlie the enlarged vein phenotype in rasa1 mutants. We used two methods to outline cell boundaries in vessels at 48 hpf. Using manual tracing of VE-Cadherin (Cdh5) antibody staining from confocal z-stacks we found that rasa1 mutant vein cells were significantly larger (WT_artery, 257 µm²; WT_vein, 278 µm²; rasa1^−/−_artery, 286 µm²; rasa1^−/−_vein, 413 µm²). Using an endothelially expressed transgenic reporter of cortical actin, Tg(fli1a:lifeact:GFP), we similarly found that vein cell size was significantly larger in rasa1 mutants (WT_artery, 282 µm²; WT_vein, 291 µm²; rasa1^−/−_artery, 319 µm²; rasa1^−/−_vein, 477 µm²). Both methods show that vein cell size is significantly increased in rasa1 mutants when compared with wild-type vein cells and in comparison with rasa1 arterial cells (for Cdh5, rasa1_vein versus WT_vein, P>0.0001; rasa1_vein versus rasa1_artery, P>0.0001; for fli:lifeact, rasa1_vein versus WT_vein, P>0.0001; rasa1_vein versus rasa1_artery, P>0.0001; Fig. 8K-N, Fig. S11). VE-Cadherin staining also shows that rasa1 mutant vein cell size increases are also present at 30 hpf before the AVM develops (rasa1^−/−_vein, 354 µm²; WT_vein, 269 µm²; P=0.02; Fig. S11). A limitation of this analysis is that measurements need to be made on flattened z-stacks as the antibody signal is too faint in single slices. In addition, measurements are unable to capture the curvature of vessels, meaning that actual cell areas will be larger. This is a systemic limitation in all measurements. Our data highlight that the enlargement of vein cells in rasa1 mutants is present as the lesion develops and is the result of increased MEK but not PI3K/mTOR signalling.

The cellular origins and mechanisms by which AVMs arise are incompletely understood. Here, we provide five lines of evidence pointing towards a venous origin for rasa1 AVMs. We show that calibre and cell size are changed in the vein without any changes in the adjacent artery of rasa1 mutants. Furthermore, we show impaired remodelling in the venous plexus leading to a cavernous vein. Finally, we find pERK upregulation in mutant vein, and demonstrate that vein-specific overactivation of MEK in rasa1a mutants drives AVM formation. This evidence suggests that the rasa1 AVM phenotype is driven by changes in vein-specific signalling and cellular behaviours.

Loss of Rasa1 results in overactivation of Ras signalling; two pathways, PI3K or MEK/ERK, can act downstream of Ras. Both pathways are drivers of different types of AVMs in humans and animal models (Alsina-Sanchís et al., 2018; Chen et al., 2019; Fish et al., 2020; Iriarte et al., 2019; Kawasaki et al., 2014; Lubeck et al., 2014; Nikolaev et al., 2018; Ola et al., 2016). In zebrafish rasa1 mutants, we find that increased MEK/ERK signalling in the vein is involved in the initiation of rasa1 vascular malformations. MEK/ERK inhibition blocks AVM formation, and overactivation promotes AVM development. In contrast, PI3K inhibition has no effect on rasa1 mutants.

The zebrafish dorsal aorta and caudal venous plexus are directly connected, but molecularly distinct. Genetic establishment of the artery or vein occurs early in development, and key markers such as EphrinB2a and EphB4 are differentially expressed early in development (Hamada et al., 2003; Wang et al., 1998). Crosstalk between MEK/ERK and PI3K/AKT/mTORC pathways is important in the specification of arteries and veins (Deng et al., 2013; Hong et al., 2006; Ren et al., 2013; Wythe et al., 2013; Zimmermann and Moelling, 1999). Intriguingly, arteries and veins appear to be correctly specified in rasa1 mutants at 30 hpf, suggesting that the mutant phenotype does not result from perturbed initial artery-vein specification, although additional studies are needed to confirm whether later changes occur. Rasa1 interacts directly with EphB4 (Kawasaki et al., 2014). EphB4 is a tyrosine-kinase receptor that plays an important role in the separation of vein from artery through interaction with the arterial ligand EphrinB2 (Hamada et al., 2003; Wang et al., 1998). Thus, Rasa1 acts downstream of a gene that is crucial for vein identity. We note that, although Rasa1 and EphB4 can physically interact, this interaction appears to be dispensable for angiogenesis (Chen et al., 2022; Kawasaki et al., 2014), thus their interactions are complex.

Which Ras protein is inhibited in endothelial cells during development in vivo by Rasa1 is not known. In vitro, Rasa1 is most effective at suppressing activation of RRas over other Ras proteins (Sung et al., 2016). On the other hand, KRAS is associated with human brain AVM (Nikolaev et al., 2018), whereas there is no known role for RRAS in AVMs. Here, we show that constitutively active rras and kras were equally efficient at driving AVM in the zebrafish. Others have shown that constitutively active braf has a similar AVM phenotype to rasa1 mutants (Al-Olabi et al., 2018). This suggests that, irrespective of Ras protein, overactivation of downstream MEK/ERK signalling is permissive of vascular lesion formation. Our data support the hypothesis that inhibitors broadly targeting Ras and MEK signalling may be clinically relevant in treating CM-AVM, as they have the potential to dampen the pathological overactivation of ERK signalling.

At the cellular level, we find that both aberrant intussusceptive angiogenesis/remodelling, combined with localized increases in vein cell size drive cavernous AVM formation. The CVP is formed through sprouting angiogenesis and intussusceptive angiogenesis, where vessels are split by the formation and merging of intraluminal pillars; the latter process is partially controlled by blood flow (Karthik et al., 2018). We did not observe impairment of sprouting angiogenesis in rasa1 mutants (migration speed or filopodia). However, we saw increased intraluminal pillars in the CVP of rasa1 mutants, suggesting that disrupted intussusceptive angiogenesis contributes to AVM formation. Incomplete intussusceptive angiogenesis was also implicated in cavernous tail lesions of a zebrafish CCM model, which also showed dysregulation of klf2a expression (Li et al., 2021). There are both flow changes and klf2a expression changes in rasa1 mutants that may impair intussusceptive angiogenesis, although we note that the molecular mechanisms controlling intussusceptive angiogenesis are poorly understood.

Over the window of AVM formation, we did not observe changes in endothelial cell proliferation, cell death or endothelial cell number. The role of endothelial cell proliferation and cell death varies between vascular malformations, suggesting they are a feature of some but not all malformations. It is also possible that they are present during some stages of malformation development (initiation or progression) (Bravi et al., 2015; Chen et al., 2011, 2019; Fish et al., 2020; Malinverno et al., 2019; Nikolaev et al., 2018; Rochon et al., 2016; Roman et al., 2002; Sugden et al., 2017; Tual-Chalot et al., 2014).

We found a localized increase in venous cell size that could help drive the progressive enlargement of vascular malformations. Although mTOR signalling is typically associated with cell size increases, addition of a PI3K/mTOR dual inhibitor did not rescue rasa1 malformations. TOR-independent cell size regulation may occur through upregulated pERK, which inhibits p38 MAPK phosphorylation through activating MAPK phosphatase 1 activity. Blocking p38 MAPK activity is permissive of cell size enlargement (Liu et al., 2018; Tan et al., 2021). Thus, the overactivation of pERK may also drive the pathological increase in venous cell size in rasa1 mutants, contributing to AVM formation.

rasa1 mutant zebrafish allow detailed in vivo analysis of AVM formation (Fig. 9) because they develop stereotypical AVMs where the dorsal aorta (DA) connects to the caudal venous plexus (CVP). The AVM develops from the CVP and subsumes both vessels. In humans, vascular lesions arise from a secondary somatic effect (Cai et al., 2018; Lapinski et al., 2018; Macmurdo et al., 2016), whereas our model is a full genetic knockout. The single rasa1a mutant zebrafish, however, is a sensitized background that will allow further study of pathways implicated in AVM development. rasa1 mutant zebrafish do not survive to adulthood; similarly, no humans have been identified with homozygous loss of RASA1 function (pLI=1.0 in GnomAD v2.1.1). In zebrafish, the CVP may be particularly prone to developing vascular malformations due to its active remodelling. Alternatively, as the malformations develop at the intersection between the DA and the CVP, any perturbation of arteriovenous specification or maintenance could predispose the region to develop malformations.

The AVM in rasa1 mutants impacts flow velocity and pulsatility without impacting heart rate. Flow changes greatly alter the forces detected by endothelial cells (Lara et al., 2013; Shepherd et al., 2009), meaning that both changes in velocity and pulsatility can contribute to the progression and enlargement of a vascular lesion. Although rasa1 mutants show changes in expression of the flow-responsive transcription factor klf2a, increasing or decreasing flow did not alleviate vascular lesion size, nor did the ablation of red blood cells to decrease shear stress. This suggests that the AVMs are genetically initiated. As the rasa1 mutant is larval-lethal, altered flow patterns that change mechanosensing in the endothelium may promote lesion progression at a later timepoint, similar to other malformation models (Corti et al., 2011; Rasouli et al., 2018; Renz et al., 2015; Sugden et al., 2017; Zhou et al., 2015, 2016).

In summary, the genetic rasa1 mutant zebrafish model has allowed us to dissect the AVM component of the CM-AVM disorder, highlighting for the first time the role of ectopic venous MEK/ERK activation in rasa1 AVM development, failed vascular remodelling through incomplete intussusceptive angiogenesis and hypertrophy of venous endothelial cells. We have shown the development of cavernous AVMs in the tail plexus and illustrate that perturbed blood flow and pulsatility from vessel malformations do not drive AVM initiation but may contribute to the progression of the lesion, as downstream flow responsive signalling is altered. Together, our data argue that the vein is particularly susceptible to perturbation of rasa1.

All experimental procedures were approved by the University of Calgary's Animal Care Committee (Protocol AC17-0189). Danio rerio embryos were maintained at 28.5°C and in E3 medium (Westerfield, 1995).

Transgenic lines used include: Tg(kdrl:mCherry)^ci5 (Proulx et al., 2010), Tg(kdrl:GFP)^la116 (Choi et al., 2011), Tg(gata1a:dsRed)^sd2 (Traver et al., 2003), Tg(fli:nEGFP)^y7 (Roman et al., 2002), Tg(fli1a:lifeact:GFP)^ca31 [reinjected from a construct that was a kind gift from Dr Li-Kun Phng (Phng et al., 2013)], Tg(flt1:citrine)^ca83 [reinjected from a construct that was a kind gift from Dr Jeroen Bussmann (Bussmann et al., 2010)] and Tg(COUPTFII-965-e1b-GFP)^ca82 [reinjected from a construct that was a kind gift from Dr Sarah De Val (Neal et al., 2019)].

rasa1a^ca35 and rasa1b^ca59 mutants were generated using CRISPR-Cas9 mutagenesis (Gagnon et al., 2014). Briefly, a 20-mer target with T7 promoter and constant Cas rev oligo (IDT) was annealed to synthesize sgRNA through in vitro transcription with MAXIscript T7 Transcription kit (Ambion, AM1312). Embryos were injected at the one-cell stage with 1 μl (∼200 ng/μl) sgRNA, 1 μl 300 ng/μl nlsCas9 mRNA and 1 μl 10 μM stop codon cassette. Mutant alleles were cloned and sequenced from genomic DNA. All primers are listed in Table S1, reagents in Table S2 and sequences in Table S3.

Genomic DNA (gDNA) was extracted from whole embryos as described in ‘PCR Sample Preparation’ from ZIRC protocols. PCR was performed with primers listed in Table S1 and visualized on an agarose gel.

Embryos were dechorionated before drug treatments from 24 hpf until 48 hpf. Stock solutions were heated to 65°C for at least 20 min before dilution in PTU. All drug treatments were carried out in a 24-well plate, with ∼20 embryos per well (Table S2). SL327 (Sigma S4069) was used for MEK1/MEK2 inhibition at a concentration of 5 µM (Shin et al., 2016). The dual PI3K/mTOR inhibitor NVP-BEZ235 (Cayman Chemical Company 10565) was used at 0.5 µM (similar to Kawasaki et al., 2014), with control experiments at 5 µM and 10 µM (Sasore and Kennedy, 2014). To increase flow, we treated embryos with 10 µM isoprenaline hydrochloride (Sigma I5627); to slow flow, embryos were treated with 2 µM nifedipine (Sigma N7634), which was titrated to minimize pericardial oedema. Phenylhydrazine (Sigma P26252) was used to ablate blood cells at a concentration of 0.5 µg/ml. Controls were shared between drug treatments conducted on the same day.

In situ hybridization for ephb4a, ephrinb2a and klf2a were performed as described with modifications (Lauter et al., 2011). Pre-hybridization and probe hybridization were performed in 50% formamide hybridization buffer (50% formamide, 5×SSC, 5 mg/ml torula yeast RNA, 50 µg/ml heparin, 0.1% Tween-20 in water) with 5% dextran sulfate. Embryos were washed twice for 5 min with 50% formamide, 2× SSC and 0.1% Tween-20 at 60°C, washed for 15 min with 2× SSC at 60°C, washed twice for 30 min with 0.2× SSC at 60°C, and blocked with 10% non-specific sheep serum (NSS) in PBT for 1 h. All in situs were performed with anti-digoxigenin F_AB fragments conjugated with alkaline phosphatase in 10% NSS/PBT and probe detection was performed with NBT/BCIP were diluted in NTT [100 mM Tris (pH 9.5), 100 mM NaCl and 0.1% Tween 20 in water]. Once the reaction was finished, embryos were fixed for 15 min in 4% PFA and cleared in glycerol overnight before imaging. Data for klf2a and grl expression were blinded for analysis.

Antibodies and concentrations are listed in Table S2. Phospho-p44/42 MAPK (ERK1/2) (Thr2020/Tyr204) antibody (Cell Signalling, 9101) was performed as described (Randlett et al., 2015). For active caspase 3 antibody (BD Biosciences, 559565) and phospho-histone H3 (Ser10) antibody (Millipore Sigma, 06-570) staining, embryos were fixed in 4% paraformaldehyde overnight and washed in 100% methanol overnight. Embryos were rehydrated into PBT and permeabilized with acetone at −20°C for 25-30 min before blocking in 10% non-specific sheet serum (NSS) in PBT for 1-2 h. Embryos were incubated with a 1/250 dilution of primary antibody in 10% NSS overnight, followed by four washes with 1% NSS and secondary antibody staining (1/500 dilution) for at least 2 h before visualization using confocal microscopy. VE-Cadherin (Cdh5) staining was performed as described. The Cdh5 antibody was a kind gift from Dr H. Belting (Blum et al., 2008). Data for pERK and cell size were anonymized for analysis.

Zebrafish were mounted on glass-bottomed petri dishes (MatTek, P50G-0-30-F) using 0.8% low melt agarose (Invitrogen, 16520-050) dissolved in E3 fish medium. Confocal imaging used a Zeiss LSM700 or Zeiss AiryScan LSM880. All images were obtained with the 488 nm and 555 nm lasers, with a slice interval of 1-3 μm with a 20× (NA 0.8) objective unless otherwise specified. Embryos imaged to characterize vessel morphology were anesthetized with 0.004% tricaine methanesulfonate (Sigma, A5040). Timelapse stacks were collected on a Zeiss LSM700 at 15 min intervals from 24 to 30 hpf.

For modelling vessels with Simpleware ScanIP (Synopsys, 2018), high resolution confocal images were captured with using AiryScan Fast imaging on a Zeiss AiryScan LSM880 confocal microscope with an Apo 40xW (NA 1.1) objective using the Argon multiline laser for 488 nm excitation and the DPSS 561 nm laser for 555 nm excitation. Slice intervals for these images were 0.25 µm. A mask was created from the image and further refined using the Gaussian filter for smoothing, island removal for RBC artefact removal and flood fill for ensuring the model was contiguous.

Images were processed using ImageJ/Fiji. Vessel diameters were measured at positions where there were no other vessels directly connecting to the vessel being measured. The vessel diameter was measured using the external diameter of the endothelium for the dorsal aorta and the internal diameter of the endothelium for the caudal veins due to the complex nature of this vessel bed. DA and PCV were measured using the full z-stack by finding the slice with the largest diameter through scanning through the stack. The imaging was carried out stereotypically at the same A-P position in the embryo, allowing direct comparisons. This was the most accurate way to find the maximum diameter. Three measurements were obtained across the dorsal aorta and averaged to obtain an average dorsal aorta diameter. Vein enlargement was measured as the internal diameter of the largest vessel in the caudal venous plexus. Vessel enlargement was designated as a vessel that was 50% larger than the average wild-type vessel. CVP and SIVP outgrowth was measured perpendicularly from the ventral side of the DA to the ventral most aspect of the CVP/SIVP. Outgrowth speed was calculated from outgrowth measurements of the same embryos at two timepoints, divided by the time between imaging. Pillars were identified as holes in the vascular plexus with diameters ≤2.5 µm and counted in the posterior 200 µm of lumenized vasculature of the CVP.

Endothelial cell counts were performed on vasculature posterior of the yolk extension in Tg(fli:nEGFP^y7;kdrl:mCherry) embryos using the multi-point tool in ImageJ and looking through all slices. Cell size was measured by outlining either VE-Cadherin-stained membranes or fli1a:lifeact:GFP transgenic membranes using the polygon tool in ImageJ. Cells in which the entire membrane could be observed on the vessel surface (i.e. did not wrap around the vessel out of view) were used. Because not all cell outlines were clear in VE-Cadherin/fli:lifeact images, the number of measurements from cell size data does not reflect actual artery or vein cell counts.

Movies were taken of embryos mounted in low melt agarose without tricaine at 10× magnification and 120 fps using the MicroZebraLab apparatus created by Viewpoint Life Sciences (ViewPoint Behaviour Technology). Movies were analysed using the Zebrablood program. Movies of the heart were used to measure heart rate over a minimum of 30 s. The average velocity across the vessel diameter, ⁠, was also measured.

Heatmaps of the velocities along the flow path in zebrafish embryos were created by using Microsoft Excel's built-in conditional formatting and image processing via PaintTool SAI (SYSTEMAX Software Development - PaintTool SAI), referencing an overlaid still image from MicroZebraLab detailing the boundaries of the measured region. The flow parameters were measured at 21-27 different locations per embryo, depending on flow path complexity and whether quality data could be obtained in the determined locations. Average velocity heatmaps were generated using the mean velocities from multiple embryos at predetermined positions across the dorsal aorta and caudal venous plexus. Filopodia were counted from images of equivalent areas and data anonymized for analysis.

map2k2a and rras cDNA were isolated from wild-type 3 dpf zebrafish by amplification with primers including attB1 and attB2. Overlap-extension PCR mutagenesis primers was used to mutate map2k2a^S219D and rras^Q61R (Table S1). Products were purified using NucleoSpin Gel and PCR Clean-up Kit (Macherey-Nagel, 740609.50) before BP Gateway into pDONR221 plasmids by incubating with pDONR221 using BP Clonase II mix (Invitrogen, 11790-020). BP products were transformed into OneShot Top10 cells (Invitrogen, C404003). For LR cloning, the p5E kdrl promoter or p5E mrc1a promoter (kind gifts from Dr Brant Weinstein, National Institute of Child Health and Human Development), was cloned with either pME map2k2a^S219D or pME rras^Q61R inserts and p3E polyA into pDESTTol2mKate, containing cryaa:mKate as an injection marker, using LR Clonase II mix (Invitrogen, 11791-020). LR products were transformed into OneShot Top10 cells. pME of activated human KRAS^G12V was purchased from Addgene (156406) and LR Gateway cloning was used to incorporate the p5E kdrl promoter and pME KRAS^G12V into pDESTTol2mKate.

One-cell embryos from incrosses of wild type or rasa1a^−/− Tg(kdrl:EGFP;gata1a:dsRed) were injected with 50 pg of construct and 50 pg of Tol2 transposase mRNA. Embryos were incubated at 28°C until 48 hpf and screened for strong expression of the cryaa:mKate injection marker before confocal imaging as described previously.

Mutants are always compared with either sibling controls (rasa1a or rasa1b heterozygous incrosses), uninjected controls (injections), DMSO-treated controls (inhibitor studies) or non-mutant transgenic lines (rasa1 mutants). Datapoints were not excluded in these studies. GraphPad Prism8 was used for all statistics. Two-tailed unpaired t-tests were used for two group comparisons (pERK comparisons) and one-way ANOVAs for multiple comparisons with P-values from Sidak's multiple comparisons reported unless otherwise indicated. Paired t-tests were used when analysing data for vessel measurements over time, or velocity measurements from the same embryo. Velocity and pulsatility data were graphed as normalized to baseline measurements unless otherwise indicated. The Chi-squared test was used for rasa1^−/− survival. Multiple t-tests with correction for multiple comparisons using the Holm-Sidak method were used for positional velocity and pulsatility data with adjusted P-values being reported. All data are represented as mean. All statistical analysis used P-values of 0.05 as a cut-off for significance (*P<0.05, **P<0.005, ***P<0.0005).

We thank Childs’ lab members, including Jae-Ryeon Ryu, Thomas Whitesell, Charlene Watterston, Nabila Bahrami and Suchit Ahuja, for thoughtful feedback throughout this project and for reviewing the manuscript. We thank the Alberta Children's Hospital Research Institute Imaging Core and the Center for Health Genomics and Informatics Sequencing core for equipment and support.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.201820.reviewer-comments.pdf