There is no difference in incidence or location of AVMs in venous-specific and pan-endothelial Eng-iKO retinas suggesting that loss of arterial ENG is not involved in driving the formation of AVMs. Instead, AVMs result from ENG defects in venous and/or capillary ECs. Furthermore, AVMs can form in the retinal primary plexus after the main neo-angiogenesis events are complete and arteries have become muscularised. This finding suggests that AVMs are not restricted to areas of neo-angiogenesis, but can also occur in regions of vessel remodelling and maturation. We have recently reported a similar scenario where aberrant remodelling of mature peripheral vessels of the adult mouse leads to major AVMs following pan-endothelial ENG depletion [
29]. Further work is required to determine whether loss of ENG in mature vessels leads to AVMs via reduced EC migration against flow, a process previously shown to underpin AVM formation in actively angiogenic vessels [
13], or whether an abnormal increase in EC proliferation is the primary cause of AVMs in pre-established blood vessels.
Our work also helps to address a long-standing controversy regarding whether AVMs form because of loss of arterial identity. This idea has come from the fact that Notch signalling maintains arterial identity and loss of function Notch mutations lead to AVMs [
16,
17]. Furthermore, Notch and ALK1 signalling activities synergise, in that SMAD1/5/8 activity promotes transcription of Notch gene targets, such as those required for arterial identity. It has therefore been proposed that HHT may result from a loss of arterial identity due to reduced Notch activity. However, although crosstalk between Notch and Smad1/5/8 signalling is critical for the stalk cell phenotype [
14] it does not appear that arteries play a direct role in AVM formation in HHT. We show here that even when arteries are unaffected, AVMs develop in the same way and at a similar frequency to pan-endothelial depletion of ENG. This finding is in agreement with a venous and angiogenesis role for ENG, rather than an arterial role, and is supported by the fact that ENG expression is stronger in veins and proliferating capillaries compared with arteries [
13,
22]. We find the increased venous expression of ENG is maintained at P11, although ENG expression in the established capillaries of P11 retinas is now reduced below that of arteries (Supplementary Fig. 2b). Furthermore, we recently showed that the AVM phenotype in Eng-iKO
e mouse retinas persists despite overexpression of human (h) ENG in arteries, and it was not until hENG was also overexpressed in veins and capillaries that the AVM phenotype was rescued [
13]. Our findings also align with recent work in zebrafish showing that Alk1 is not required for arterial identity, and AVM formation in
alk1 mutants does not result from loss of Notch activity [
30]. As the molecular role of ENG is to promote ALK1 signalling by facilitating ligand binding it is likely that ALK1, like ENG, will also not have an arterial role in protecting against AVMs in mammals, although supporting evidence currently awaits further studies. It is interesting to consider in this context our original observation that ENG expression is decreased in
Alk1-iKOe retinas [
12]. This raises the question of whether the reverse is true such that ALK1 expression is reduced in
Eng-iKOe retinas. In fact, we observed that ALK1 expression appears to increase in the high flow AVMs of Eng-iKO
e retinas using our previous knockout protocol [
22] (Supplementary Fig.
3). These findings are entirely consistent with previous studies showing that ENG is downstream of BMP9 signalling and is therefore be reduced in the absence of ALK1 [
12,
31], whilst in contrast, ALK1 expression has not been reported to be regulated by BMP9, but is upregulated by sheer stress [
32,
33], which is anticipated to be increased due to the high blood flow through AVMs. Although endothelial loss of ENG or ALK1 lead to AVMs in preclinical models, the phenotype of ALK1-iKO
e mutants is more severe than that of Eng-iKO
e mutants [
7]. This is likely because loss of ALK1, the signalling receptor, leads to elimination of this signalling pathway, whilst loss of ENG, the co-receptor, reduces the availability of ligand to the signalling receptor [
3], thereby dampening rather than completely obliterating signal. In HHT patients, however, the differences between HHT1 (
ENG) and HHT2 (
ALK1) clinical phenotypes are more complex and await a better understanding of the relative contribution of second hits and additional factors such as inflammation to fully unravel the aetiology of this disease.
Finally, we show for the first time that AVMs can form in the murine retinal blood vessels during the second week of life, after the primary plexus is established, which considerably extends the previous time frame available for these studies, opening up further options for therapeutic rescue experiments.