Bmp9 regulates Notch signaling and the temporal dynamics of angiogenesis via Lunatic Fringe

Summary

Sprouting angiogenesis and blood vessel stabilization require precise coordination between endothelial cells (ECs) and pericytes. Bone Morphogenic Protein 9 (Bmp9), whose signaling through Activin receptor-like kinase 1 (Alk1) is dysregulated in several diseases, was thought to regulate these processes by independently activating Notch target genes in an additive fashion with canonical Notch signaling. Here, through predictive computational modeling validated in mice, zebrafish, and human cell lines, we uncover that Bmp9 enhances Notch activity synergistically by upregulating Lunatic Fringe (Lfng) in ECs. Specifically, Bmp9-induced Lfng enhances Notch receptor activation, most strongly when Delta-like ligand 4 (Dll4) is also present. This Lfng regulation alters vessel branching by modulating the timing of EC phenotype selection and rearrangement during angiogenesis. Lfng also contributes to pericyte-driven vessel stabilization by mediating Jagged1 upregulation in Bmp9-stimulated ECs. In summary, Bmp9-upregulated Lfng enhances Dll4-Notch1 signaling in ECs and Jag1-Notch3 activation in pericytes, shaping angiogenic sprouting and stabilization outcomes.

Graphical Abstract

eTOC Blurb

Ristori et al. show that Lunatic Fringe (LFng) mediates the crosstalk between Bone Morphogenic Protein 9 (Bmp9) and Notch signaling during angiogenesis. This crosstalk impacts the temporal dynamics of endothelial cell identity during sprouting and endothelial cell-pericyte interactions during vessel stabilization, which may provide a therapeutic target in Bmp9-Alk1 diseases.

Introduction

The formation of new blood vessels from pre-existing ones, termed angiogenesis, is a complex biological process crucial in tissue development and healing. Dysregulated angiogenesis is correlated with several diseases such as cancer and hereditary hemorrhagic telangiectasia (HHT)^1,2. HHT is characterized by incorrect development of blood vessels and formation of arteriovenous malformations (AVMs), caused by mutations of the activin receptor-like kinase 1 (ACVRL1/Alk1). Alk1 binds with high affinity to Bmp9, a circulating factor predominantly produced by the liver. The Bmp9-activated Alk1 then forms a binding complex with endoglin and BMP type II receptor (BMPR-II)^3,4. This induces phosphorylation of Smad1/5/9, followed by recruitment of Smad4, their nuclear translocation, and regulation of Alk1 target genes. The Alk1-Bmp9 pathway has been shown to be involved in the regulation of angiogenesis and vascular density^5,6. Alk1 deletion leads to embryonic death by E11.5 in mouse models, due to vascular development defects⁷. Deletion of the Alk1 gene (Acvrl1) in postnatal mice causes AVM formation, severe internal bleeding, and death⁸, highlighting Alk1 role in homeostasis. However, the exact contribution of the Bmp9/Alk1 pathway to the regulatory mechanisms of angiogenesis is not fully clear.

It is well established that the crosstalk between vascular endothelial growth factor (Vegf) and Notch signaling (Fig. 1A) is a fundamental regulatory mechanism of blood vessel formation^9-12. At the onset of angiogenesis, Vegf released by nearby hypoxic tissue is detected by Vegf-receptors present on endothelial cells (ECs)¹³. This stimulates the EC migratory response ¹³ as well as upregulation of the ligand Delta-like 4 (Dll4)^9,14. The consequential Notch1 activation in neighboring ECs leads transcription factors to modulate their Vegf receptor expression (Fig. 1A, blue pathway), such that these Notch1-activated cells obtain a decreased ability to sense Vegf and activate their migratory behavior^9,12,15-17. This cycle repeats over time, determining the formation of a pattern of highly migratory and Dll4-expressing cells (termed “tip cells”), alternated with less migratory and Notch1-activated cells (termed “stalk cells”)¹³. This ensures that only a few leading tip cells emerge from the existing vessels, with regular spacing, while proliferating stalk cells contribute to elongate the sprout¹³. The determination of these tip/stalk identities is nevertheless dynamic, since ECs overtake each other at the tip of sprouts (“cell shuffling”), continuously competing for the tip position via Notch signaling^18-20.

Figure 1: Bmp9 upregulates LFng expression in ECs.

(A) Schematic of Bmp9 (green) and Vegf-Notch (blue) crosstalk-signaling in ECs, and hypothesized LFng upregulation by Bmp9 (red).

(B-D) In vitro experiments (full color) and simulations (striped color) of RT-qPCR analysis of HEY1 and LFNG in HUVECs, induced by Bmp9 treatment and/or Dll4-coating after 24h stimulation. (B, D) n = 5 biological replicates.

(E) Western blot analysis of NICD and LFng compared to the safe-keeping gene beta-actin in HUVECs with Ctrl (left) or LFNG siRNA treatment (right) upon stimulation with 10ng/ml Bmp9, up to 24h. Relative quantification normalized over unstimulated data (bottom). n = 3 biological replicates.

(F) RT-qPCR analysis of Lfng, Hey1, and Hes1 expression in ECs isolated from wild-type and Alk1 EC retinal vessels (n = 3 biological replicates; 2 mice/replicate, for a total of 6 mice/group corresponding to 12 retinas/group), with representative IsolectinB4 and LFng staining images (n = 3 mice/group). Scale bar, 200 μm.

(G) In vitro experiments (full color) and simulations (striped color) of RT-qPCR analysis of HES1, HEY1 and LFNG in HUVECs, induced by Bmp9 treatment and/or Dll4-coating after 24h stimulation, with (shaded color) or without (bright color) LFNG siRNA transfection. n = 6 biological replicates.

All values are mean +/− SD. *p < 0.05; **p<0.01; ***p<0.001; ****p<0.0001; (B, D, G) ANOVA, (F) Unpaired t-test. (B, D) No significance was found for the comparisons without the symbols *.

Previous studies have shown that the activation of Alk1 by Bmp9 can independently activate Notch target genes to trigger stalk-cell markers and a consequential inhibition of angiogenesis^6,21. However, reported double-inhibition or gain-of-function studies on the Dll4-Notch and Bmp9-Alk1 pathways combined⁶ demonstrated a synergistic increase or decrease in angiogenic branching, respectively, that the prevailing additive, independent-activation of Notch target genes model (Fig. 1A, green pathway) of crosstalk between them cannot explain. Thus, we hypothesized that additional, inter-dependent crosstalk must exist between the two pathways, such that modulation of one can create positive feedback to enhance the effects of the other.

Here, we examined the possibility that the Bmp9 and Notch pathways might be linked by Fringe proteins, previously shown to be important for sprouting angiogenesis. Lunatic, Manic, and Radical Fringe are enzymes that increase Notch1 and Dll4 binding affinity by glycosylating the Notch extracellular domain. Lunatic Fringe (LFng) has a dominant role in the regulation of Notch1 signaling²², and its knockout (KO) is known to cause vascular hypersprouting resulting from Notch inhibition in ECs²³. Despite its key role in angiogenesis, however, the regulators of LFng during angiogenesis are currently largely unclear. Previous studies have shown that LFng expression can be controlled by Notch activity in somites²⁴. In ECs, it has been shown that LFng is modulated by the endothelial transcription factor ERG²⁵. Interestingly, ERG also regulates the effects of Bmp9 on ECs²⁶. Moreover, a recent study showed that LFng expression is downregulated in Alk1-mutated ECs²⁷. Overall, these studies suggest a possible link between Bmp9/Alk1 signaling and LFng expression in ECs.

By integrating predictive simulations with in vitro and in vivo experiments, we identified that BMP9 upregulates LFng in a Smad-dependent manner, which in turn enhances Notch signaling in an interdependent crosstalk manner and affects the temporal dynamics of Notch patterning. We previously showed that modulation of Notch pattern timing can delay or accelerate tip/stalk cell selection and shuffling, resulting in sparser or denser branch spacing during angiogenesis and thereby impacting the resulting vascular network topology^28-31. For example, a slower process of tip/stalk cell selection led to a sparser vascular network topology, e.g. when the anti-angiogenic, tissue derived factor semaphorin3E, which co-regulates Dll4, was lost³¹. We have previously found via theoretical modeling that LFng alone could potentially modulate Notch pattern timing³². Here, via extended modelling integrated with experiments, we demonstrate that Bmp9 regulation of LFng expression can alter Notch pattern timing, branching, and shuffling in silico, in vitro and in vivo. Simulations with an agent-based model, compared with additional in vitro experiments, also highlighted that the LFng overexpression can overrule the effects of Alk1 deficiency on cell clustering in mosaic sprouting competition assays. Finally, we show that Bmp9-mediated upregulation of LFng potentiates Notch crosstalk between pericytes and ECs, which can affect the vital pericyte-mediated stabilization of ECs at the end of the angiogenic process. The discovery of LFng as a strong enhancer of the Bmp9-Notch crosstalk and temporal modulation of tip selection and branch spacing, points to a potential target for medical therapies aimed at regulating and improving vascularization in disease.

Results

Simulations predict Notch ligand-receptor binding rate as a candidate for Bmp9-Notch1 synergy

We hypothesized that a missing, interdependent crosstalk link must exist between the Bmp9/Alk1 and Vegf/Notch pathways. This was motivated by previous RT-qPCR analysis of human umbilical vein endothelial cells (HUVECs) exposed to either Bmp9, Dll4-coated substrates, or both, which highlighted a synergistic increase upon double stimulation that cannot be explained by the prevailing (additive) independent crosstalk model (Fig. 1A, green-blue schematic)⁶. Here, we observed the same synergistic effect (Fig. 1B). HUVECs upregulated HEY1 expression in response to Dll4 (~5-fold increase) and Bmp9 (~18-fold increase) and, most interestingly, the upregulation resulting from the two stimuli combined (~55-fold increase) was far greater than a simple addition of the effects of the two taken alone (Fig. 1B).

To test our hypothesis, we utilized computational modeling, extending our existing ordinary differential equation (ODE) model of Vegf/Notch signaling during angiogenesis³² (see the STAR Methods for the ODE model description). Previous studies validated this model’s predictive capability in vivo³⁰. This model³² considers Notch transactivation, but neglected Notch receptor-ligand cis-inhibition³³. To test whether cis-inhibition may play a role here, we performed experiments with HUVECs cultured sparsely, to minimize possible confounding effects from Notch transactivation. These HUVECs were treated with DLL4 small interfering RNA (siRNA), to elicit Dll4 expression knockdown (KD), and were then stimulated with external Dll4-coating to activate Notch. If cis-inhibition played a relevant role in our cell culture system, the decrease of intracellular Dll4 induced by DLL4 siRNA would decrease intracellular Notch1 cis-inhibition, with a consequentially higher Notch1 availability and therefore Notch1 activation by the external Dll4-coating. This was however not the case: Dll4 downregulation had no significant effects on the expression of Hey1 and Hes1, indicating a lack of involvement of cis-inhibition (Fig. S1A). Therefore, cis-inhibition was not required in the computational model, as per analogous studies where simulation results were validated by angiogenesis experiments^19,20,29,34.

We therefore adapted the original model³² to simulate cell signaling among a line of 10 to 50 cells being exposed to Dll4-coating. We performed a parameter exploration to test how different upstream Bmp9-Notch crosstalk interactions would impact the Dll4-coating effects on Notch activation levels (Fig. S1B). This exploration predicted that the synergistic effects of Dll4 and Bmp9 on Hey1 can be best explained by either an increase in the Dll4-Notch1 binding rate, Notch1 expression, or by a decrease in Notch1 or Notch Intracellular Domain (NICD) degradation rates (Fig. S1B). For example, when assuming that Bmp9 increases not only Hey1 expression, but also the Dll4-Notch1 binding rate (Fig. 1A), the simulations can now accurately replicate the unexplained experimental synergistic result under Dll4-Bmp9 co-stimulation (Fig. 1B-C).

Bmp9 upregulates the expression of LFng

We next experimentally tested the potential points of Bmp9-Notch crosstalk identified by the simulations (Fig. S1B). RT-qPCR data ruled out significant effects of Bmp9 stimulation on HUVEC Notch1 expression (Fig. S1C). By subjecting cells to cycloheximide chase, Notch1 and NICD degradation were also ruled out, as no significant difference in protein temporal dynamics were observed for Bmp9-treated HUVECs (Fig. S1D). Therefore, the simulations and experiments indicated possible effects of Bmp9 on Notch ligand-receptor binding rate.

The Notch1 ligand-receptor binding rate is positively correlated with affinity³⁵, in turn strongly influenced by the enzymes Lunatic, Manic, and Radical Fringe^36,37. These enzymes have a key role for angiogenesis²³, and previous modeling from our group indicates LFng modulation could influence the temporal dynamics of tip/stalk selection³². Previous studies have shown that Smad1/5, effectors of Bmp9-Alk1 signaling, are involved in the regulation of LFng³⁸. Therefore, we hypothesized that the synergistic effects of Bmp9 and Notch modulation on angiogenesis are mediated by Fringes.

We therefore checked whether Bmp9 modulates the expression of the Fringe enzymes. Considering that feedback loops could possibly cause temporal oscillations of the enzyme expression^30,31,39, multiple time points were analyzed. RT-qPCR analysis showed that, while Manic and Radical Fringe expressions are unaffected by Dll4 and Bmp9 stimulation (Fig. S2A), Lunatic Fringe (LFNG) is strongly and rapidly upregulated by Bmp9 and unaffected by Dll4 (Fig. 1D). Focusing on Bmp9 stimulation, we observed that a non-significant increase in LFNG expression is already observed after 30 minutes of Bmp9 stimulation, while a significant upregulation is detected after 1 hour (Fig. S2B). The effects of Bmp9 on LFNG peak at 1.5-3 hours, followed by a gradual return to untreated levels after 24 hours (Fig. 1D, S2B). This was verified by Western Blotting (Fig. 1E), which confirmed at the protein level that Bmp9 effects on LFng are time-dependent and relatively rapid: LFng protein was upregulated and increased at 3h after Bmp9 exposure, while the effects decayed later. This rapid change in LFng expression following Bmp9 stimulation is consistent with previously reported rapid transcriptional activation of Bmp/Smad target genes such as ID1–ID3, HES1, and others^40-42. Given the rapid upregulation and previous data indicating Smad1/5 as LFNG regulators³⁸, we hypothesized that Bmp9 regulates LFNG through Smad-mediated transcription. In agreement with this, SMAD4 KD through HUVEC transfection with SMAD4 siRNA showed that Bmp9-induced expression of LFNG at 3 hours is Smad signaling dependent (Fig. S2C). Therefore, our coupling between simulations and experiments uncovered that Bmp9 triggers Smad-mediated upregulation of LFNG expression in a time-dependent manner.

Bmp9 induces endothelial Notch1 cleavage dependent on Dll4 and LFng but not Jag1

Next, we assessed whether Bmp9-mediated LFNG upregulation affects Notch activation. Western Blotting for NICD indicated a strong increase in Notch cleavage at 6h and 10h after Bmp9 exposure. This increase was lost when LFng or Dll4 were reduced by siRNA treatment (Figs. 1E, S3A). Despite Bmp9 upregulating the expression of the other Notch ligand Jagged1⁶ (JAG1), JAG1 siRNA-mediated KD did not alter the Notch1 cleavage levels induced by Bmp9 24-hour exposure (Fig. S3B). This is consistent with data showing that Jag1 has a lower potential to activate Notch1 compared to Dll4, in the presence of LFng⁴³. Moreover, consistent with a negligible role of Jag1 in this context, the synergistic effects of Bmp9 and Dll4 in the expression of HES1/HEY1 were also unaffected by JAG1 siRNA (Fig. S3C). As such, our data indicates that Bmp9 induces Notch1 cleavage by upregulating LFng to potentiate Dll4-mediated Notch1 activation, independent from Jag1.

To validate Alk1 signaling regulation of LFng and Notch activation in vivo, we analyzed the retina of wild-type and Alk1ΔEC mice (Fig. 1F). RT-qPCR analysis of the retinas showed that the conditional heterozygous deletion of endothelial Acvrl1, confirmed by immunostaining (Fig. S3D) in agreement with previous studies^44-46, caused significantly lower Lfng levels in Alk1ΔEC mice compared to control. Accordingly, gene expressions of Hes1/Hey1 also decreased. This was supported by LFng immunostaining images; while the vasculature was still distinguishable (e.g. LFng staining is clearly visible on arteries) in the control group, indicating LFng expression in the vascular endothelium, the vasculature was no longer distinguishable in Alk1ΔEC retinas. Overall, these in vivo data show that Bmp9-mediated activation of Alk1 regulates LFng expression in ECs in vivo.

LFng enhances Bmp9-mediated upregulation of endothelial Notch target genes

We next checked how much Bmp9-mediated LFng regulation contributes to the synergistic increase in target gene expression seen with Dll4 and Bmp9 co-stimulation (Fig. 1B). We simulated cells exposed to Bmp9 and Dll4-coating, with and without LFng KD. As expected from the previous parameter exploration (Fig. S1B), the simulations predicted that LFng KD decreases the synergistic effects of Bmp9 on the expression of Notch target genes Hes1/Hey1 (HE in the model) (Fig. 1G). These computational results therefore suggested that LFng mediates a proportion of the Bmp9 and Notch effects on ECs, especially enhancing them when Dll4 is present.

To verify these computational findings, we measured key Notch target gene expressions in HUVECs exposed to Bmp9 with or without Dll4-coating, previously transfected with control (Ctrl) or LFNG siRNA inhibiting LFNG expression (Fig. S3E-F). Consistent with simulations, Bmp9-mediated upregulation of Notch target genes HES1/HEY1, observed in previous studies^6,21,47, was significantly reduced by LFNG siRNA treatment when HUVECs were exposed to both Bmp9 and Dll4-coating (Fig. 1G), by approximately 22% and 62% for HEY1 and HES1, respectively. LFNG KD significantly decreased HES1 expression also when only Bmp9 was present, with approximately a 64% reduction, while this was not the case for HEY1. Consistently, statistical analysis confirmed that the interaction (synergistic) effect of Dll4 and Bmp9 significantly affected the expressions of both HES1 and HEY1 for Ctrl siRNA samples (Table S1). For LFNG siRNA samples, the interaction effect lost significance for HES1, indicating LFng strongly mediates this interaction. For HEY1, the interaction effect remained significant with LFNG siRNA, but with a lower percentage of total variation compared to Ctrl. This suggests that, while LFng contributes to the Dll4-Bmp9 synergy regulating Hey1, other synergistic mechanisms yet to be discovered, may play a role. This is consistent with previous studies showing that, in synergy with cell serum, Bmp9 induces Hey1 expression, but not Hes1, in a Notch-independent fashion³⁹. Overall, our results indicate that the LFng upregulation by Bmp9 mediates a significant proportion of the effects that Bmp9 has on ECs.

Bmp9 and LFng are temporal regulators of tip/stalk cell selection in silico

Previous experimental evidence has demonstrated that the temporal dynamics of tip/stalk cell selection at the onset of angiogenesis strongly affects the resulting vascular topology^28,31,48, where slower tip selection leads to sparser branch spacing. Therefore, we next performed simulations to investigate whether Bmp9 and LFng are temporal regulators of tip/stalk selection induced by Vegf exposure.

We found that the time to establish a tip/stalk “salt and pepper” pattern was indeed condition-dependent (Figs. 2A-B, S4). Cells exposed to Bmp9 (and Vegf) retained stalk-cell features for longer (Figs. 2A-B). Simulations also predicted that increasing LFng levels generally slows tip/stalk patterning (Figs. S4A, S4E). However, major effects could be seen only with relatively large changes in the Dll4-Notch1 binding rate; halving this parameter, corresponding to the LFNG siRNA simulation in Fig. 2A-B, did not yield major effects compared to control. Hey1/Hes1 levels had a similar monotonic effect; increasing their level slowed down patterning (Fig. S4F). This occurred because Hes1/Hey1 directly downregulate Vegfr2⁴⁹ and, therefore, they decelerate Vegf detection and the consequential filopodia formation driving tip cell selection²⁹. Likewise, increasing Bmp9 levels also exhibited a monotonic, increasing effect on the time to pattern (Fig. S4G), consistent with Bmp9 inducing upregulation of both LFng and Hes1/Hey1 in the model. In our simulations, LFng KD upon Bmp9 exposure could partially rescue patterning speed (Fig. 2B, S4H), as Bmp9 can then only enhance Hey1/Hes1 expression directly. In conclusion, the simulations indicate that Bmp9 and LFng are temporal regulators of tip/stalk selection, slowing it down with Bmp9 addition, consistent with the sparser branching network topology observed in vitro and in vivo when Bmp9 is added^6,21.

Figure 2: LFng mediates Bmp9 effects on angiogenesis.

(A) Representative simulations of tip/stalk pattern formation for a row of 10 ECs exposed to Vegf with and without Bmp9 and LFng siRNA treatment. Amount of filopodia in each cell shown green (high, tip cells) and purple (low, stalk cells).

(B) Pattern formation rate predicted by n = 50 simulations/condition.

(C-D) Quantification and (E) representative images of HUVEC bead assays, with or without Bmp9 stimulation (1 ng/ml) and LFNG siRNA treatment. n = 6 biological replicates. Scale bar, 100 μm.

All values are mean +/− SD. *p < 0.05; **p<0.01; ***p<0.001; ****p<0.0001; ANOVA.

LFng mediates Bmp9 effects on in vitro angiogenic sprouting

To verify that LFng mediates Bmp9 effects on sprouting angiogenesis, we performed bead sprouting assays with HUVECs transfected with LFNG siRNA or Ctrl siRNA, with and without addition of Bmp9 (Fig. 2C-E). In agreement with previous in vivo studies⁶ and the slowed tip/stalk patterning predicted by simulations (Fig. 2B), Bmp9 stimulation significantly decreased the length and number of sprouts compared to control (Fig. 2C-E).

Our computational results predicted that inhibiting LFNG upon Bmp9 treatment should partially rescue tip selection speeds and therefore vascular branching density (Fig. 2B). Consistently, LFNG downregulation did partially rescue branching, significantly increasing sprout length (Fig. 2D-E) and generating a comparable number of sprouts per bead (no significant difference) between the control samples (without BMP9) and the LFNG siRNA samples with BMP9 stimulation (Fig. 2C). Taken together, these results strongly suggest that LFng mediates the expression of Notch target genes induced by Bmp9, which in turn regulates the inhibition of angiogenesis caused by Bmp9. The trends observed in simulations (Fig. 2B) and the results obtained with the bead assay are strikingly consistent (Fig. 2C-D), strongly suggesting that the regulation of branch length and density is driven by changes in the Notch patterning temporal dynamics, allowing more tip cells to be selected faster (or slower) to generate more (or less) well extended sprouts (Fig. 2).

Bmp9 and LFng are temporal regulators of tip/stalk cell shuffling

ECs continually rearrange during sprouting, with tip cells typically overtaken every 4-6 hours^18,19. This process is driven by Notch-regulated differential adhesions²⁰. Previously, we found that slower Notch patterning also reduced cell overtaking rates in silico and in vitro³¹. Thus, we next investigated whether Bmp9 could impact the dynamics of tip/stalk cell shuffling via LFng.

We adapted our previously validated spatial agent-based model (the “memAgent-spring” model incorporating Cellular Potts Model, MSM-CPM) of Vegf/Notch-regulated cell shuffling^20,31,50. The model enables tracking of the position of cells as they rearrange within a sprout exposed to a Vegf gradient along the longitudinal direction (Fig 3A). Similar to previous computational results with Notch inhibition²⁰, our simulations predicted that LFNG KD strongly increased the frequency of tip/stalk shuffling (Fig. 3B-C). LFNG KD weakened the feedback loops of the signaling network because of low Notch activation and, consequently, tip cells could not efficiently inhibit their neighbors via Notch, allowing the neighbors to actively compete and often overtake them.

Figure 3: Lfng is a temporal regulator of tip selection and tip/stalk shuffling.

(A) Representative simulation of the MSM-CPM. Over time (horizontal axis) cells on a sprout (different colors) form filopodia towards the Vegf source (top) and compete for the tip position of the sprout by exchanging positions (“shuffling”, red arrows).

(B) Kymograph of the center of mass of individual cells (different colors) shuffling in the simulated sprout, with (bottom graph) or without (top graph) Lfng KD.

(C) MSM-CPM shuffling rate with and without LFng KD, normalized over the control rate (wild-type). n = 10 simulations/condition.

(D) In vivo quantification of EC nuclei shuffling, (E) number of ECs selected to branch, and (F) representative time-lapse images of EC nuclei dynamics in intersegmental vessel (ISV) sprouts in either control (MOC) or lfng KD (lfng) Tg(kdrl:nlsEGFP)^zf109 embryos from 19 hours post-fertilization (hpf). In (F), nuclei are pseudocolored according to the order of their emergence from the dorsal aorta (blue: first; red: second; orange: third), with dividing cells retaining the same color and primary number as their parental cell (e.g. 1.1 and 1.2, top row). Bracket indicates a dividing cell. Shuffling events are indicated via dashed lines. Scale bar, 20 μm. n = 11 for MOC, n = 23 for lfng.

All values are mean +/− SD. *p < 0.05; **p<0.01; ***p<0.001; ****p<0.0001; Mann-Whitney test.

To verify these results, we performed in vivo experiments tracking EC nuclei dynamics in zebrafish embryos, during intersegmental vessel (ISV) sprouting from the dorsal aorta (DA) (Fig. 3D-F, supplementary videos 1-2). Lfng KD triggered significantly more cell shuffling compared to control, validating simulations (Fig. 3C-D). Moreover, a higher number of ECs emerged from the DA of lfng KD embryos, with some lfng KD ISVs exhibiting 3 emerging ECs per ISV, compared to a maximum of 2 emerging ECs in control embryos (Figs. 3E). These observations are consistent with simulations, showing that major LFng inhibition can limit Notch activity leading to faster selection of more tip cells (Fig. S4A, S4E). Representative in vivo results are shown in Fig. 3F. In the control, a single tip cell (1, blue) emerges from the DA and, after division, is followed by the resulting cell (1.2, blue) and a stalk cell (2, red) without shuffling (Fig. 3F, top row). In lfng KD, three cells were selected to emerge, and the third (3, orange) sequentially overcame the first two to reach the sprout tip (Fig. 3F, bottom row). Overall, these in vivo results demonstrate that LFng regulates the temporal dynamics of EC tip selection and shuffling.

LFng loss can explain dysregulated cell rearrangement resulting from impaired Bmp9-Alk1 signaling

Next, we investigated whether the Bmp9-mediated upregulation of LFng could have physiological relevance in the vascular dysfunctions associated with impaired Alk1 signaling, such as AVMs present in HHT⁵¹. AVMs tend to be comprised of cells low in Notch signaling⁵², therefore pointing to a possible relevant role of LFng. Previous mosaic experiments with Alk1-inhibited cells mixed with wild-type cells uncovered that Alk1 inhibition induces the inhibited cells to cluster at sprout front in vitro⁶ and form AVMs in vivo⁵³.

Therefore, we next simulated cell shuffling dynamics in mosaic experiments, where cells were randomly assigned either a wild-type (red) or LFng KO (green) cell type, comparing to a control sprout with only wild-type cells (randomly labelled green and red) (Fig. 4A). As expected, the control case presented equal shuffling rates and dynamics of green and red wt:wt cells (Fig. 4A, panels i-ii), whereas LFng KO cells had a far higher forward shuffling rate, causing them to cluster at the sprout front, outcompeting wild-type cells, which were pushed to the sprout rear (Fig. 4A, panels iii-iv). This stems from the fact that Notch receptors present in LFng KO cells have lower likelihood to bind to Dll4 in neighboring cells; therefore, they have a generally lower Notch activation compared to wild-type cells, making them more motile and less adhesive. A gradual increase in LFng KD resulted in a consistent, gradual cell clustering (Fig. 4B).

Figure 4: LFng KO and KD cells spatially cluster in silico and in vivo in mosaic vessels.

(A) Dynamics of cell shuffling in a MSM-modelled vessel sprout: (i, ii) all wild-type ECs (labelled green and red); (iii, iv) mosaic mix of wild-type (red) and LFng KO (green) ECs. (i, iii) Kymograph showing cell centre of mass position in the sprout over time. (ii, iv) Averaged cell centre of mass, 10 runs, mean and SD.

(B) Mean +/− SD position (vertical axis) of the center of mass of MSM-simulated cells in a sprout with a mixed cell population (wild-type = red, LFng KD = green), varying the fraction of LFng that was KD in each cell per simulation.

(C) Mosaic sprouting experiment of HUVEC-GFP cells transfected with Ctrl or LFNG siRNA, mixed with HUVEC-mCherry. Images taken 4 days after microcarrier beads were embedded in fibrin gels. n = 3 biological replicates.

(D) Mosaic bead sprouting assay of GFP or GFP/LFNG HUVECs transfected with either ACVRL1 or Ctrl siRNA, together with mCherry HUVECs, with relative quantification of the percentage of cells present at the tip of the sprouts. n = 3 biological replicates. Scale bars, 100 μm.

To validate these computational findings, we performed in vitro mosaic sprouting competition assays with GFP-labelled HUVECs transfected with LFNG or Ctrl siRNA, mixed with mCherry-labelled HUVECs. Consistent with simulations, LFNG KD indeed conferred an advantage in terms of cell localization at the sprout tip, which contained more LFNG KD HUVECs compared to control HUVECs (Fig. 4C).

Therefore, while we have previously ascribed the clustering of Alk1 KD cells to the sprout front in mosaic experiments to the direct targeting of Hey1 by Alk1⁶, our present simulations (Fig. 4A-B) and experiments (Fig. 4C) indicate the loss of LFng in Alk1 KD cells (Fig. 1) as another contributing mechanism. To validate this theory, we tested whether LFng overexpression (OE) can compensate for loss of LFng in Alk1 KD cells (Fig. 1) and therefore rescue the effects of Alk1 KD on endothelial sprouting. Western blot demonstrated that lentiviral delivery successfully led to LFng OE in HUVECs (Fig. S5A). Alk1 KD was obtained via ACVRL1 siRNA transfection (Fig. S5B). In vitro mosaic sprouting competition assays were performed with LFng OE and/or ACVRL1 KD HUVECs mixed with wild-type cells. Quantification of cells at the sprout tip showed that LFng OE dominates over the ACVRL1 KD effects: as expected from our theory and previous experiments⁶, ACVRL1 KD HUVECs were in a higher percentage present at the sprout tip (Fig. 4D); consistent with our model, LFng OE counteracted this effect and prevented Alk1-deficient cells from reaching the tip position (Fig. 4D). Our model argues that LFng OE prevents these cells from being selected at the tip position by increasing Notch1 activation. Indeed, we observed that LFng OE in HUVECs resulted in a significant increase in Notch1 cleavage compared with cells transduced with the empty vector (Fig. S5A).

Bmp9-mediated expression of LFng and Jag1 in ECs ensure contemporary Notch activation in ECs and pericytes

Bmp9 is known to strongly upregulate the expression of Jag1³⁸, another Notch ligand highly involved in cardiovascular (and angiogenesis) regulation^23,54-57. Previous research has shown LFng-mediated Notch1 glycosylation inhibits Jag1-Notch1 activation^36,37,58 and thereby ensures a pro-angiogenic role of Jag1 expression^23,59. The concomitant increase of Jag1 and LFng, induced by Bmp9, together with the anti-angiogenic role of Bmp9, thus appears counterintuitive. Jag1 plays a crucial role in EC-pericyte crosstalk signaling^60,61, fundamental for endothelial stabilization^62,63. Thus, we hypothesized that Jag1 and LFng act in synergy to potentiate the crosstalk between ECs and pericytes, with LFng acting as a “railroad switch” diverting Jag1 from activating Notch1 on ECs, to activate Notch3 on pericytes, allowing the Bmp9-upregulated Jag1 levels to improve, rather than hinder, vessel stability.

Motivated by previous experiments^{6,23,36,37,61}, we extended our ODE model of EC signaling to include the Jag1-mediated EC crosstalk with pericytes (Fig. 5A). In the model, Bmp9 upregulates EC expression of Jag1⁶, in turn activating Notch1 in ECs and Notch3 in neighboring pericytes^23,61. Notch3 activation induced pericytes to express Dll4⁶¹, which activates Notch1 in ECs, establishing a full feedback loop between pericytes and ECs. Based on previous experiments^36,37,64, and distinct from previous models^65,66, LFng was assumed to increase the binding of both Dll4 and Jag1 to Notch1, while decreasing the chance of Jag1-Notch1 activation post-binding (Fig. 5A). As shown in previous studies⁶⁷ and confirmed via quantitative RT-qPCR analysis (Fig. S5C), pericytes express little ACVRL1; therefore, Bmp9 was assumed to have no direct effect on pericyte Hes1/Hey1.

Figure 5: Bmp9-mediated LFng upregulation potentiates EC-pericyte interactions.

(A) Schematic of EC-pericyte crosstalk signaling in simulations. Full names correspond to unbound ligands and receptors (e.g. Notch1). Capital letters with numbers correspond to bound receptors (e.g. N1.D4 is the bound Notch1-Dll4 complex).

(B-G) Experiments (full color) and simulations (striped color) of RT-qPCR analysis of HES1 and HEY1 in HUVECs and pericytes, alone or in co-culture, induced by Bmp9 (10 ng/ml) after 24h stimulation, with (shaded color) or without (bright color) JAG1 or LFNG siRNA treatment only on HUVECs. n = 5 biological replicates for all groups except F (n = 6). In the computational model, Hes1 and Hey1 expressions were merged in a single variable (HE).

(H) RT-qPcR analysis of LFNG and JAG1 in HUVECs induced by Bmp9 (10 ng/ml) after 24h stimulation, with or without LFNG siRNA treatment on HUVECs. n = 7 biological replicates.

(I) RT-qPCR analysis of JAG1 in HUVECs stimulated by Bmp9 (10 ng/ml) and/or Dll4 (10 ng/ml) at different time points, up to 24h. n = 7 biological replicates.

(J) Simulation of Hes1/Hey1 expression in pericytes co-cultured with control of LFng-KD HUVECs exposed to Bmp9, with the computational model extended to consider the Jag1 expression in HUVECs dependent on Notch1 activation. All values are mean +/− SD. *p < 0.05; **p<0.01; ***p<0.001; ****p<0.0001; ANOVA.

To investigate the possibly synergistic roles of LFng and Jag1, we performed coupled in silico and in vitro studies featuring mono- and co-culture of HUVECs and pericytes, stimulated with Bmp9, with and without downregulation of LFng or Jag1 in ECs. Together, simulations and experiments showed that Bmp9 has little to no effects on Notch activation in pericytes unless co-cultured with ECs, which resulted in a significant increase of Hey1/Hes1 (Fig. 5B). This was confirmed by Western Blotting, showing increased Notch3 cleavage in pericytes co-cultured with ECs under BMP9 exposure (Fig. S5D). In simulations, co-culture also increased the previously observed Hes1/Hey1 upregulation in ECs induced by Bmp9 (Fig. 5C). This was not observed in the experiments, although a non-significant increasing trend was present (Fig. 5C).

In the model, Bmp9 effects on pericytes are entirely mediated by EC Jag1. Consistently, we observed that culture of pericytes on immobilized recombinant Jagged1 (rhJag1) elicited Notch3 cleavage in pericytes (Fig. S5E). Accordingly, JAG1 KD in ECs was predicted to inhibit Bmp9 effects on pericytes in co-culture (Fig. 5D), with decreasing effects also for ECs (Fig. 5E). RT-qPCR experiments confirmed this, with partial KD of JAG1 causing a decrease of pericyte HES1/HEY1 (Fig. 5D) and EC HES1 (Fig. 5E).

We next performed LFng KD experiments and simulations. The model predicted that LFng KD causes a slight increase in pericyte Hes1/Hey1 expression (Fig. 5F) and a decrease in EC Hes1/Hey1 expression, due to decreased Dll4-Notch1 affinity (Fig. 5G). Experiments were in partial agreement, showing LFng KD in ECs decreases Bmp9-mediated EC HES1/HEY1 expression (Fig. 5G), consistent with earlier results (Fig. 1). However, contrary to simulations, LFng KD decreased pericyte HES1 expression, with a non-significant downregulating trend observed for HEY1 (Fig. 5F).

This disagreement suggested the existence of other underlying mechanisms. Given the important role of EC Jag1 for pericytes, we checked experimentally whether LFng KD affects Jag1 expression in ECs. Interestingly, we observed that LFng mediates the Bmp9-induced upregulation of JAG1 (Fig. 5H); in fact, LFNG KD strongly decreased JAG1 expression resulting from Bmp9. This explained the apparent disagreement between experiments and simulations present in Fig. 5E. Specifically, in simulations, Jag1-Notch1 binding rate in ECs decreased with LFng KD; therefore, more Jag1 was available for pericyte Notch3 activation, driving elevated Hes1/Hey1 (Fig. 5F). However, in the experiments, more Jag1 was not available for Notch3 binding because LFng KD decreased Jag1 expression in ECs (Fig. 5H). The role of LFng in regulating Jag1 expression in response to Bmp9 suggests that a synergy between Bmp9 and Notch signaling might exist also in this context. Indeed, a time course RT-qPCR analysis of JAG1 expression in HUVECs confirmed that Dll4-coating induces a moderate increase of JAG1 in vitro, but when combined with Bmp9 stimulation a stronger increase is observed (Fig. 5I). Confirming the central role of LFng, LFNG siRNA treatment significantly decreased Jag1 at the protein level for HUVECs stimulated for 24h with Dll4, Bmp9 or the two combined (Fig. S5F). Refining the model to include this finding resolved the simulation-experiment discrepancy observed in Fig. 5F, now successfully matching all trends of pericyte-EC co-culture experiments (Fig. 5J, S5G-I). Taken together, we showed Bmp9 upregulation of LFng ensures high Dll4-Notch activation in ECs, while increasing EC Jag1 expression to activate Notch3 on pericytes.

LFng and Jag1 induced by Bmp9 act in synergy to stabilize vasculature

Pericytes are fundamental to stabilize ECs and associated vasculature^62,63,68. We computationally investigated whether LFng and Jag1 synergize to stabilize ECs via pericyte interaction. We simulated ECs exposed to VEGF for 24h, to establish a stable tip/stalk pattern, followed by 24h of exposure to VEGF, Bmp9, and pericyte signaling (Fig. 6A). These simulations approximated the final stages of angiogenesis, where ECs recruit pericytes and are exposed to higher Bmp9 levels in the blood⁵. Vessels were assumed to stabilize when all ECs stably exhibited a stalk phenotype characterized by low filopodia. Without Bmp9, despite pericyte signaling, ECs retained the previously established tip/stalk pattern, without vessel stabilization (Fig. 6B-D). Bmp9 addition induced all ECs to exhibit low filopodia, caused by upregulated LFng and Jag1 expressions leading to increased Hes1/Hey1 and therefore inhibited Vegfr2 expression. Consistent with this, LFng and Jag1 expressions resulting from Bmp9 positively correlated with the rate of stabilization (Fig. 6B-D). Both proteins were essential for this process; decreasing either LFng or Jag1 below a threshold led to loss of stabilization (Fig. 6B, bright green area). Therefore, our simulations strongly suggest that LFng and Jag1 act in synergy to potentiate Notch crosstalk between ECs and pericytes and drive vascular stabilization.

Figure 6: LFng and Jag1 as synergistic temporal regulators of endothelial stabilization.

(A) Schematic of the simulations. Exposure to Bmp9 and pericyte contact occur 24h after Vegf exposure.

(B) Average, normalized rate of EC stabilization for co-varied values of Bmp9 effects on LFng and Jag1, normalized against the original parameter values. n = 50 simulations/condition.

(C-D) Representative cases. LFNG and JAG1 siRNA were mimicked by reducing the associated parameter values 50%. In (D), the amount of filopodia in each cell is colored in green (high) and purple (low).

Discussion

Bmp9-Alk1 and Notch signaling are interconnected in multiple cellular systems⁶⁹. By integrating experiments with simulations, we here provide a key missing link that Bmp9 regulates LFng, which generates a positive-feedback, synergistic, crosstalk between Bmp9 and Notch signaling that better explains observations than the prevailing independent, additive model of bmp9-Alk-Notch crosstalk⁶. Furthermore we find that the upregulation of LFng contributes to: 1) regulation of the temporal dynamics of angiogenesis exerted by Bmp9 over the short timeframe of hours, impacting branch length/density, cellularity and spacing; 2) abnormal EC rearrangement at the tip of sprouts in Alk1 loss-of-function conditions; 3) the upregulation of Jag1 and diversion of Jag1 to activate Notch3 on pericytes rather than Notch1 on ECs, contributing to vessel stabilization and providing an explanation for decreased pericyte coverage and EC Notch activation observed in Alk1 mutant AVMs.

LFng has been shown to play crucial roles for several developmental and cellular processes^24,70-79 due to its dominant effect on Notch regulation²². Temporally cyclic regulation of LFng expression has been reported to regulate the oscillatory nature of Notch signaling during mesodermal segmentation^24,70-72,74. In the context of angiogenesis, we have previously shown that the Notch temporal dynamics influences EC fate selection and shuffling, and the resulting vascular topology²⁸. Several temporal modulators of Notch signaling have been identified, including Dlk1 and Mash1^80,81. During angiogenesis, LFng inhibition leads to increased tip-cell selection and vascular density²³. Combined with our previous simulations³², our experimental and computational results strongly indicate that Bmp-induced LFng influences angiogenesis by acting as a temporal modulator of EC fate selection and rearrangement (Figs. 2-3). Our results moreover suggest that LFng plays a key role for the effects that Bmp9 has on angiogenesis, as demonstrated by the partial rescue of Bmp9-stimulated angiogenic sprouts by LFng KD (Fig. 2) and the overruling of Alk1 KO effects by LFng OE in mosaic experiments (Fig. 4D). Therefore, while the cyclic activity of Bmp9 over the course of days has been shown to be central for the regulation of angiogenesis⁸², our study also points at Bmp9 as a regulator of the temporal dynamics of angiogenesis in the shorter timescales of hours.

It is well recognized that the regulation of angiogenesis and vasculature homeostasis exerted by Bmp9-Alk1 signaling is central in health and disease. Dysregulation of Alk1 can lead to AVMs, characteristics of HHT^6,52. Via mosaic experiments, we have previously shown that Alk1 KO cells cluster at the tip of sprouts⁶, and that AVMs in mosaic in vivo experiments are mainly formed by the Alk1 KD cells⁵³. Our simulations indicate that the latter phenomenon can be explained by the downregulated expression of LFng in Alk1 KD cells (Fig. 4), which makes them less prone to be inhibited by Dll4 present in neighboring wild-type cells. Notch inhibition can cause AVMs^83,84. AVM regions as a result of Alk1 KD are typically characterized by lower Notch activity compared to physiological values⁵², which could also be caused by a downregulation in LFng.

LFng downregulation in ACVRL1 mutated diseases might not only impair Notch among ECs, but also between ECs and pericytes, as demonstrated by our co-culture results (Fig. 5). Interestingly, AVM patients exhibit lower pericyte coverage around the vessels⁸⁵. Moreover, poor Notch signaling in pericytes has been shown to lead to AVM formation⁸⁶. Our co-culture data strongly suggests that the Bmp9-mediated upregulation of Jag1 is fundamental for Notch activation in pericytes, and that LFng contributes to this Jag1 upregulation by increasing Notch activity in ECs (Fig. 5). These data are also in accordance with a previous study showing that endothelial-expressed Jag1 is required for Notch3 signaling in perivascular cells⁶⁰. Overall, our data thus point at Bmp9 as a potentiator of the EC-pericyte Notch signaling communication, via upregulation of both Jag1 and LFng, which act in synergy to induce elevated Notch activation in both pericytes and ECs, thereby facilitating cellular homeostasis and vessel stabilization (Fig. 6). Consistent with this concept, AVMs are characterized by vascular instabilities and lack of homeostasis⁸⁵. Therefore, our data indicate that Bmp9-mediated expressions of LFng and Jag1 in ECs are likely to be key for vasculature homeostasis and, when disrupted, might be linked to AVMs and HHT.

Smad1/5 have been previously identified as regulators of LFng in ECs³⁸. Mechanistically, therefore, it is likely that Bmp9 activates Alk1, whose signal is transmitted via Smad1/5, with a consequential upregulation of LFng (Fig. 1A). Consistent with that, we have shown that LFng upregulation by Bmp9 is mediated by Smad4. Moreover, in other cell lines, Bmp signaling has been shown to induce expression of target genes within similar short-term timescales through transcription^41,87. Similar effects observed for ERG, previously shown to upregulate LFng in HUVECs²⁵, could also be explained with analogous mechanisms, given the promotion of Smad1 transcriptional activity consequential to ERG²⁶. In contrast, we showed that Dll4-mediated Notch activation did not induce relevant LFng expression changes in HUVECs (Fig. 1), differently than in somites^24,88, thereby highlighting the key role of Bmp9-Alk1 signaling for LFng regulation in ECs.

Taken together, these data uncover a role for BMP9-mediated LFng expression in maximizing Notch1 activation by Dll4 in ECs, while concomitantly maintaining Notch3 signaling in pericytes through Jag1, overall demonstrating the functional relevance of the Bmp9-mediated LFng upregulation at different timescales of the angiogenic process in health and disease.

Limitations of the Study

Our study reveals that Bmp9-Alk1 signaling upregulates LFng to potentiate Dll4-Notch1 in ECs, with consequences on angiogenesis and EC-pericyte interaction; however, some limitations should be considered. Firstly, much of the molecular work relies on cultured HUVECs, in vitro bead sprouting assays and zebrafish morpholino KD, systems that (while powerful) may not fully recapitulate the cellular complexity, biomechanical context, or compensatory responses of mammalian tissues in vivo. Moreover, RNAi, morpholino and overexpression manipulations can have partial efficacy or off-target effects and may not always mimic physiologic or chronic perturbations. Thus, building on this works foundation, subsequent in vivo, mechanistic and translational studies are needed to establish causality across contexts and to evaluate therapeutic potential. Simulation predictions were validated experimentally, however some parameter assumptions (e.g., ligand–receptor binding and activation rates) may not capture the full biochemical or spatial heterogeneity of tip/stalk dynamics across different vascular beds. These caveats indicate that, while the data are consistent with LFng as a mediator of Bmp9-Notch crosstalk, further work and tissue-specific model calibration will aid understanding of its role across tissue contexts.

STAR METHODS

Experimental model and study participant details

Mouse Model

Tamoxifen-inducible Cdh5-CreErt2 and acvrl1^loxP mice were kindly provided by Ralf Adams and S. Paul Oh respectively, and were back-crossed to C57/Bl6 background (Charles River) for 10 generations. To generate Alk1ΔEC mice, Cdh5-CreErt2 and acvrl1^loxP mice were crossed and injected with 50 mg/kg tamoxifen dissolved in corn oil at P4 and euthanized at P5. Retinas were harvested and stained with IsolectinB4 and selected antibodies as previously described⁶. Throughout the studies, Cdh5-CreErt2-Alk1+/+ (thereafter referred as C5Cre) mice injected with tamoxifen as described above were used as controls. The Maisonneuve-Rosemont Hospital ethics committee, overseen by the Canadian Council for Animal Protection, approved all experimental procedures (protocol number: 2014–18). All the animal experiments were conducted according to the Standard Operation Procedures (SOP) of the Maisonneuve-Rosemont Hospital Animal Ethics Committee.

Zebrafish Model

Embryos and adults were maintained under standard laboratory conditions as described previously²⁹ and experiments were approved by the University of Manchester Ethical Review Board and performed according to UK Home Office regulations. The Tg(kdrl:nlsEGFP)^s896 strain was established previously⁸⁹. Efficient KD of lfng in zebrafish was achieved using a previously validated morpholino oligonucleotide (MO)⁹⁰. Embryos were injected at the one-cell stage with either 8 ng control MO (MOC) or 8 ng lfng MO (lfng). MO sequences were: 5’- CCTCTTACCTCAGTTACAATTTATA -3’ (MOC), 5’- ACCGTGTATACCTGTCGCATGTTTC - 3’ (lfng)⁹⁰. All MOs were purchased from Gene Tools. Movies of zebrafish ECs were acquired as previously described⁹¹. Briefly, live embryos were mounted in 1% low-melting agarose (containing 0.1% tricaine) in glass-bottom dishes and were continually perfused with embryo water supplemented with 0.0045% 1-phenyl-2-thiourea and 0.1% tricaine at 28°C using an in-line solution heater and heated stage controlled by a dual-channel heater controller (Warner Instruments). Embryos were imaged using 40× -dipping objectives on a Zeiss LSM 700 confocal microscope. Cell nuclei were tracked over time. After the emergence of the first tip cell from the DA, over the course of 13 hours, the total number of shuffling events and emerging ECs per ISV were quantified, the latter excluding any additional cell that arose from proliferation.

Cell culture

Human umbilical vein ECs (HUVECs; pooled from male and female donors, C-12203, PromoCell) were cultured at 37 °C and 5% CO₂ in Endothelial Cell Growth Medium-2 (ECGM-2, PromoCell). Prior to stimulation with BMP9 (10 ng/mL), HUVECs were serum-starved overnight in EBM-2 (Lonza) supplemented with 0.1% FBS. The same starvation protocol was applied before stimulation with recombinant soluble DLL4 (sDLL4), which had been precoated onto 6-well plates at 10 μg/mL as previously described⁶. Primary human pericytes (female donor; C-12980, PromoCell) were maintained at 37 °C and 5% CO₂ in Pericyte Growth Medium 2 (C-28041, PromoCell). HEK293T cells (female embryo–derived; ATCC CRL-3216) were cultured at 37 °C and 5% CO₂ in DMEM supplemented with 10% FBS and penicillin–streptomycin. Primary mouse retinal ECs were isolated from mixed-sex litters of neonatal mice and were used for mRNA isolation. For KD experiments in vitro, HUVECs were transfected with siRNAs (QIAGEN) using RNAiMax reagent (ThermoFisher) according to the manufacturer’s instructions, 24 h prior to stimulation with BMP9. Negative controls were performed using AllStars Negative Control siRNA (QIAGEN).

Microbe strains

Replication-incompetent, VSV-G–pseudotyped lentiviral particles were produced in HEK293T cells by transient transfection of transfer, pRSV-Rev (packaging), and pCMV-VSV-G (envelope) plasmids using Fugene, according to the instructions of the manufacturer. Cells were seeded at 70–80% confluence in antibiotic-free DMEM + 10% FBS and culture medium was replaced 12–16 h post-transfection. Viral supernatants were harvested at 48 h and 72 h, clarified by 0.45 μm filtration, and either used fresh or stored at −80 °C. HUVECs (≤ passage 2) were plated at 60–80% confluence in EGM-2 and transduced with viral supernatant. Medium was replaced after 12–16 h, and reporter expression or antibiotic selection (puromycin 1.0 μg/mL, determined by kill curve) was initiated 48–72 h post-infection.

Methods details

Cycloheximide chase assay

Confluent layers of HUVECs, at passage number 5, grown in 35mm culture dishes, were incubated overnight in starvation media (endothelial basal medium, Promocell Cat, with 0.1% FBS). The next day, the media was replaced with fresh starvation media supplemented with 10 ng/ml Bmp9 (0 hours), or only starvation media as control. After 3 hours, 50μg/ml cycloheximide (Sigma-Aldrich) in DMSO was directly added to the culture media. The plates were gently swirled to mix and incubated at 37°C and 5% CO2 until lysis. The cells were lysed at 0, 1, 3, 6, 10, and 24 hours following BMP9 treatment using 3X Laemmli sample buffer (0.1875 M Tris, 0.21 M SDS, 30% glycerol, bromophenol blue) with 3% 2-mercaptoethanol and boiled at 95°C for 10 minutes. The samples were used in Western blotting using primary antibodies, followed by HRP-conjugated secondary antibodies against rabbit or mouse (Vector). The bands were visualized using SuperSignal West Pico PLUS chemiluminescence kit (ThermoFisher Scientific) and iBright FL1000 imaging system (ThermoFisher Scientific). Image Studio Lite version 5.2 (LI-COR) was used for densitometry analysis of the Western blot images. The quantified signal was normalized to β-actin (Sigma-Aldrich Cat) and presented as fold change.

Western blot

Cells were lysed in ice-cold RIPA buffer supplemented with protease and phosphatase inhibitors, clarified (12,000–16,000 × g, 10 min, 4 °C), and protein concentration was determined by BCA assay. Equal amounts of protein (40 μg) were mixed with 4× Laemmli sample buffer + 100 mM DTT, heated (95 °C, 5 min), resolved by SDS–PAGE, and transferred to nitrocellulose membranes using wet transfer. Membranes were blocked (3% BSA in TBST, 1 h, RT), incubated with primary antibodies in blocking buffer (overnight, 4 °C), washed in TBST (3×5 min), and incubated with HRP-conjugated secondary antibodies (1 h, RT). After TBST washes (3×5 min), signals were developed by ECL and imaged on a Azure 600 system. Band intensities were quantified in ImageJ, normalized to housekeeping proteins (β-actin) and expressed relative to control.

Quantitative RT-PCR

Total RNA was isolated (RNeasy, QIAGEN) and quantified by spectrophotometry (A260/280 ≈ 2.0). cDNA was synthesized from 0.5 μg RNA using a cDNA synthesis kit (Bio-Rad). qPCR was performed in 20 μL reactions using SYBR Green (Bio-Rad) master mix with 500 nM primers on a real-time thermocycler (QS7 Flex). Each sample was run in technical triplicate. Gene expression was normalized to ACTB as reference gene and analyzed by the 2^-ΔΔCq method.

Pericyte co-cultures

250,000 HUVECs previously transfected with Ctrl siRNA, or siRNA targeting JAG1 or LFNG, were grown in ECGM-2 medium supplemented with VEGF for 24 hours before the addition of 100,000 human pericytes, with or without Bmp9. After 24 hours, ECs were separated from pericytes using CD31-labeled Dynabeads, and mRNA was harvested using RNeasy MiniKit (QIAGEN), while proteins were extracted by resuspending cells in RIPA buffer. Relative purity of HUVECs and pericytes was confirmed by RT-qPCR analysis of endothelial (PECAM1, CDH5) and pericyte (PDGFRB) markers. As controls, 100,000 pericytes were cultured with or without Bmp9 in the absence of HUVECs.

Fibrin gel bead assay

The bead assay was performed following previous protocols^92,93. Briefly, Cytodex microcarrier beads were swollen in PBS (50 mL/g) and autoclaved at 120°C, to obtain a solution of 60,000 beads/ml. Beads were then coated with approximately 400 HUVECs per bead and transferred to a T25 flask in 5mL of EGM-2. After a day, a (2.5 mg/mL) fibrinogen solution in EBM-2 was prepared, adding aprotinin (Sigma-Aldrich) for a final concentration of 50 μg/ml. The beads were then seeded, for a final concentration of 500 beads/ml. Thereafter, 0.625 Units/mL of thrombin (Sigma-Aldrich) were added to the mix. Human Dermal Fibroblasts (BJ; ATCC) were finally added on top of the fibrin gels, at a concentration of 20000 cells per well. The medium, supplemented with or without 10 ng/ml VEGF and/or 1 ng/ml BMP9 was replaced every 2 days, and sprouts were imaged by fluorescence after 7 days. Single sprouts were defined as continuous cellular protrusions departing from the beads that were longer than the bead diameter. The number of sprouts per bead and their lengths were measured. The latter were averaged for each sample, to obtain single values for independent biological replicates.

Sprouting competition assays were performed analogously as previously described⁶. For the LFNG KD versus wild-type experiments, HUVECs were transfected with either LFNG or Ctrl siRNA (QIAGEN) respectively labelled with GFP and mCherry (obtained via lentiviral infection). For the rescue experiments, HUVECs were transduced with lentiviral particles encoding human LFng and GFP, while control HUVECs were infected with the empty vector containing GFP (pReceiver-Lv201; GeneCopoeia). After puromycin selection (1 ug/ml), HUVECs were transfected with the appropriate siRNA (Ctrl or ACVRL1, QIAGEN). In this case, untreated cells were labelled with mCherry (via lentiviral infection). In both experiments, 24 hours after transfection, GFP- and mCherry-labelled HUVECs were mixed at a 1:1 ratio, then coated on Cytodex3 microcarrier beads and embedded in fibrin gels. After 4 days, pictures of the sprouts were taken to quantify the proportion of GFP-positive cells at the tip position of sprouts.

Isolation of Retinal ECs

Following tamoxifen injections at P4, retinas from C5Cre-Rosa^mTmG and Alk1ΔEC-Rosa^mTmG were harvested at P5 and dissociated for 20 min using collagenase type II as previously described⁹⁴. GFP-positive cells were harvested by FACS. For each RT-qPCR measurement, ECs obtained from 2 mice (4 retinas) were pooled before mRNA extraction. Overall, 10,000 retinal ECs were obtained from 4 pooled retinas. Sorted ECs were resuspended in RNA lysis buffer and RNA was isolated (RNeasy MicroKit; QIAGEN) and processed for quantitative RT-qPCR. This process was repeated to obtain independent isolations from 4 (2 wildtype and 2 treated mice) mice each time.

Computational methods

ODE model of EC signaling

Model assumptions

The effects of BMP9 on the Notch-mediated interaction between ECs was simulated by extending a previous ODE model³². The original model was developed based on mass-action kinetics equations. Briefly, the original model assumes that Notch can bind and be activated by Dll4 present in adjacent cells. Notch activation leads to upregulation of Hey1, which in turn downregulates the expression of VEGFR2. VEGFR2 can bind and be activated by VEGF present in the environment, which leads to upregulation of Dll4 and filopodia formation. Finally, filopodia formation is assumed to increase the VEGF that is detected by the cell and thus available for VEGFR2 binding and activation.

Here, this original model was extended by accounting for the lateral movement and diffusion of Notch ligands and receptors along the cell membrane, to enable the simulation of a row of ECs, as well as the effects of Dll4-coating and Bmp9 on Notch. The cell membrane was considered as divided between two edges, left and right. To separately consider proteins present in cell edges in contact with different neighbors in the EC row, the ensemble of Notch receptors and ligands was divided into two groups: proteins on the left cell edge, indicated with the subscript , and proteins on the right cell edge, with subscript. Similar to previous studies^95,96, the movement of unbound Notch proteins between the two edges was included by assuming random movement, and thus by adding an equation term similar to diffusion. Bmp9 was assumed to enhance Hes1/Hey1 and LFng expressions. This was modelled by multiplying the Hes1/Hey1 basal production and Dll4-Notch1 binding rates, given the influence of LFng on Notch receptor-ligand affinity³⁶ and thus their binding rate³⁵.

Model equations

Below the model equations are briefly summarized. The subscript $i\in \mathbb{N}$ indicates a specific cell on the cell row, with $i-1$ and 1 + 1 labelling its left and right neighbors, respectively. Vegfr2 receptors present on the $i^{th}$ EC are exposed to a level of Vegf $V_i$ dependent on the cell filopodia $f_i$:

$$V_i=(1+k_3{f_i}^2)V_0$$ (1)

where $V_0$ is the reference Vegf and $k_3$ scales the filopodia influence. Vegf binds to Vegfr2 ($R_i$) forming a Vegfr2-Vegf complex ($R.V_i$), which increases filopodia formation, while Vegfr2 expression decreases in response to Hes1/Hey1 ($H_i$) and degradation as follows:

$$\frac{d{f}_i}{dt}={\beta }_F+k_{f}V.R_i-k_{-f}{f}_i$$ (2)

$$\frac{d{R}_i}{dt}={\beta }_R-k_1{V}_i{R}_i+k_{-1}V.R_i-k_{inh}{R}_i{H}_i^2-{\varphi }_R{R}_i$$ (3)

$$\frac{dV.R_i}{dt}=k_1{V}_i{R}_i-k_{-1}V.R_i-{\varphi }_{R}V.R_i$$ (4)

where ${\beta }_F$ and ${\beta }_R$ are the basal filopodia formation and production rate of filopodia and Vegfr2, respectively; $k_f$ scales the filopodia formation in response to Vegfr2 activation; $k_{-f}$ represents filopodia turnover; $k_1$ and $k_{-1}$ are respectively protein association and dissociation rates; ${\varphi }_R$ labels the protein degradation rate; and $k_{inh}$ scales Vegfr2 inhibition by Hes1/Hey1.

Vegfr2 activation via Vegf, quantified by $V.R_i$, leads to Dll4 ($D_i$) upregulated expression. This ligand binds to Notch1 ($N_i$) present in neighboring cells, forming the complex $D.N_i$ that induces Notch activation at a rate $k_{cat}$. Notch1 is activated also by external Dll4 ($D_{ext}$), forming associated complexes ($D_{ext}.N_i$). Dll4, Notch1, and the receptor-ligand complexes in the left edge of each cell $i$ then vary according to:

$$\frac{d{D}_{L_i}}{dt}=\frac{1}{2}\left({\beta }_D+\theta \frac{V.R_i^2}{1+V.R_i^2}\right)-2k_{2_{bmp}}{D}_{L_i}{N}_{R_{i-1}}+k_{-2}D.N_{R_{i-1}}-{\varphi }_D{D}_{L_i}+W\left(\frac{D_L+D_R}{2}-D_L\right)$$ (5)

$$\frac{d{N}_{L_i}}{dt}=\frac{{\beta }_N}{2}-2k_{2_{bmp}}{N}_{L_i}\left(D_{R_{i-1}}+D_{ext}\right)+k_{-2}\left(D.N_{L_i}+D_{ext}\cdot N_{L_i}\right)-{\varphi }_N{N}_{L_i}+W\left(\frac{N_L+N_R}{2}-N_L\right)$$ (6)

$$\frac{dD.N_{L_i}}{dt}=2k_{2\_bmp}{N}_{L_i}{D}_{R_{i-1}}-k_{-2}D.N_{L_i}-{\varphi }_{ND}D.N_{L_i}-k_{cat}D.N_{L_i}$$ (7)

$$\frac{d{D}_{ext}.N_{L_i}}{dt}=2k_{2\_bmp}{N}_{L_i}{D}_{ext}-k_{-2}{D}_{ext}.N_{L_i}-{\varphi }_{ND}{D}_{ext}.N_{L_i}-k_{cat}{D}_{ext}.N_{L_i}$$ (8)

$\theta$ scales the Dll4 downregulation in response to Vegfr2 activation; $k_{2\_bmp}$ and $k_{-2}$ are respectively the dissociation and association rates of Dll4 and Notch1; and $W$ scales the movement rate of unbound Notch proteins across the two cell edges. Note that, compared to the original model, the Notch protein production and binding rate were scaled by a factor 2, to accommodate for the two cell edges⁹⁷. The time variations of the proteins on the right, indicated with subscript $R$, are modelled analogously. The Bmp9-mediated upregulation of the Notch receptor-ligand binding rate was modeled assuming a linear relationship:

$$k_{2\_bmp}=h_{2\_bmp}{k}_2$$ (9)

where $k_2$ is the homeostatic binding rate and $h_{2\_bmp}$ linearly scales Bmp9 effects. Finally, Notch activation leads to an increase in Notch intracellular domain ($I_i$) and consequentially Hes1/Hey1 ($H_i$) expression:

$$\frac{d{I}_i}{dt}=k_{cat}\left(D.N_{L_i}+D.N_{R_i}+D_{ext}.N_{L_i}+D_{ext}.N_{R_i}\right)-{\varphi }_I{I}_i$$ (10)

$$\frac{d{H}_i}{dt}={\beta }_{H\_bmp}+\theta \frac{I_i^2}{1+I_i^2}-{\varphi }_H{H}_i$$ (11)

where ${\beta }_{H\_bmp}$ is the Hes1/Hey1 expression as affected by Bmp9:

$${\beta }_{H\_bmp}=h_{H\_bmp}{\beta }_H$$ (12)

where ${\beta }_H$ indicates the Hes1/Hey1 homeostatic basal production rate, linearly scaled by $h_{H\_bmp}$ representing Bmp9 effects. The values $h_{H\_bmp}$ = 1 and $h_{2\_bmp}$ = 1 were chosen to simulate physiological conditions. Values larger than 1 were chosen to simulate Bmp9 injection or Alk1 OE, and lower for Alk1 KD or LFng KD.

Parameter values

While most of the parameter values were left unchanged compared to the original model³², some of the parameter values were updated to reflect more accurate experimental results. The degradation rate of Hes1/Hey1 was assumed to be equal to 8*l0⁻⁴ s⁻¹ to capture the (approximately) 15-minute half-life of these proteins^98-100. The degradation rate of NICD was assumed to be half the one of Hes1/Hey1, consistent with recent reports¹⁰¹ and studies indicating that Notch signaling in ECs is also relatively short-lived due to several signaling modulators such as SIRT1¹⁰². Although the degradation rate of Notch and VEGF proteins is relatively low (in the order of 10⁻⁵-10⁻⁴ s⁻¹)^103,104, their endocytosis and recycling rate is very high (in the order of 10⁻² s⁻¹)¹⁰⁵. Assuming this faster phenomenon to be determinant of the temporal dynamics involved, this value was chosen for the degradation plus endocytosis rate of Notch and Vegf proteins (10⁻² s⁻¹). Cells were assumed to strive to maintain a constant value of receptors on the cell membrane, such that the production and recycling rate of these proteins was assumed equal to this degradation plus endocytosis rate. Moreover, the diffusion rate was assumed to be an order of magnitude lower than the endocytosis rate, reflecting the low diffusion of Notch proteins before being endocytosed¹⁰⁶. The bond lifetime of the Notch1-Dll4 complex was assumed to be extremely short³⁵; thus, a very high dissociation rate and cleavage rate of the Notch1-Dll4 complex were considered (2.5 s⁻¹). The remaining parameter values were left unchanged compared to the original model, except for the basal production of Dll4 and Hes1/Hey1, which were calibrated to reach tip/stalk pattern selection in approximately 8h, as per experimental observations²⁹. The parameter describing the effects of external Dll4 ($D_{ext}$) was calibrated against previous experiments⁶. For simplicity, the unbound Notch protein movement rate ($W$) was kept high enough to ensure fast protein homogenization. The model parameter values are listed in Table S2, where “cu” stands for concentration units.

Simulations

The model was coded in Matlab2023a and solved using ODE15s. The ODEs are deterministic, but the final solutions depend on the initial conditions because of the multiple attractors present in the solution space and the limited simulation time. To account for this, similar to previous studies with analogous Notch models^{65,97,107,108}, the simulations were performed for 50 random initial conditions. Simulations were performed for 10 to 50 adjacent cells with periodic boundary conditions, as in a recent study¹⁰⁸, such that the first cell was assumed to be interacting with the last cell. For more details on the specific ODEs, we refer the reader to the Supplementary Information.

In vitro RT-qPCR experiments with EC monolayers were simulated by assuming that cells were not exposed to VEGF ($V_0$ = 0). In agreement with the actual experiments, the ODEs were solved with a final time equal to 48h, and external Dll4 and/or BMP9 were only added 24h hours after the start of the experiment. The parameter representing external Dll4 ($D_{ext}$) was calibrated such that the Hey1 expression predicted for Dll4-stimulated cells is 5 times higher than the Hey1 expression predicted for starved cells, in agreement with previous in vitro results⁶. The parameter sweep adopted to explore the possible effects of BMP9 on Notch was performed by multiplying and dividing by two the model parameter values associated with Notch signaling (e.g. Notch-Dll4 binding rate, Notch production, etc.). For the specific parameter values that were varied during the parameter sweep, we refer the reader to Eqs. 5-10 and Fig. S1B. After the identification of the effects of Bmp9 on the Notch receptor-ligand binding rate as a possible candidate for explaining previous RT-qPCR experiments⁶, $h_{H\_bmp}$ and $h_{2\_bmp}$ were calibrated to mimic such experiments. LFng KD was simulated by halving the associated value (see $k_2$ in Table S2).

Tip/stalk selection at the onset of angiogenesis was simulated by adopting the same ODE model. Cells were exposed to no VEGF for the first 24h, thereby mimicking homeostatic conditions, while cells were exposed to relatively high VEGF levels ($V_0$ = 0.1 cu) for the subsequent 24h. For Fig. 2A-B, motivated by our experiments and simulations, Bmp9 injection was modeled by increasing the basal expression of Hes1/Hey1 and the Dll4-Notch1 binding rate by four times ($h_{H\_bmp}$ = 10, and $h_{2\_bmp}$ = 9), while LFng KD was again modeled by dividing the control value by 2 ($h_{2\_bmp}$ = 1/2).

Analysis of in silico tip/stalk pattern formation rate

For the in silico simulations, the time necessary to form the tip/stalk pattern that is characteristic of angiogenesis was evaluated by assigning fates to each cells based on their specific filopodia values, to be consistent with herein experiments and previous computational analysis¹⁰⁹. ECs were assigned a tip cell phenotype if their filopodia exceeded a certain heuristically-determined threshold ($f_{min}$ = 25 c. u.), and a stalk cell phenotype otherwise. A tip/stalk pattern was assumed to be established, similar to previous studies^29,109, when the percentage of tip cells was between 40% and 50%, and no adjacent tip cells were present. To obtain a proxy of the tip/stalk pattern formation rate, the time to pattern was normalized over 1 day, taken as a reference value.

MSM-CPM

Shuffling of cells at the tip of sprouts was simulated with the MSM^109,110, incorporating CPM based modelling of cell rearrangement via differential adhesion²⁰. This model and its parameter values have been previously validated against several in vivo experiments in mouse and zebrafish^20,31,34. The model accounts for the movement of the cell membrane as influenced by VEGF, Notch signaling, filopodia formation and neighboring cells. Individual computational agents (memAgents) represent the EC membrane, which is held together by the actin cortex represented by springs following the Hooke’s law. Cell membrane movement, Dll4 content, and filopodia formation are proportional to the VEGF detected by the cell, which in turn decreases in response to Dll4-mediated Notch activation. In the present study, LFng KD was accounted for by scaling the model parameter representing Notch activation. In particular, to simulate reduced Bmp9 signaling in Alk1 mutant cells, we reduced Notch activation in mutant cells to 25% of its WT value.

In the simulations, the vessel was considered to be oriented in the direction of a VEGF gradient that was constant over time, so that excited tip-like cells moved along the vessel up the gradient. A vessel was composed of 10 cells competing for the tip position, initialized next to each other (one per vessel cross section). We allowed the vessel delta-notch pattern to stabilize for around 8 hours (1000 timesteps of 30s) before allowing cells to rearrange via differential adhesion. We then continued the simulation for either 4000 timesteps (around 33 hours) to evaluate tip cell competition and 14000 timesteps (around 109 hours) to evaluate cell rearrangement. Model parameter values were kept as in previous studies ^20,31, unless otherwise specified. As the model is stochastic, we quantified the average position of the center of mass along the VEGF gradient between mutant and wild-type cells for 10 repeats. At the end of each simulation, each cell position with respect to the vessel longitudinal direction (y-axis) was calculated as the average among the y-coordinate of that specific cell memAgents.

We evaluated LFng KD tip cell competition by creating mosaic vessels with 5 wild-type and 5 mutant cells in random initial positions and quantifying the average position of WT and mutant cells along the gradient. We evaluated the effect of LFng KD on cell rearrangement by creating all-mutant vessels and counting the number of changes of cell order along the gradient. We first filtered the kymograph time series with a uniform filter of length 1 hour to remove short-lived position changes and then counted the number of cell rearrangements over 13000 timesteps (~109 hours), comparing to an all-wild-type vessel.

ODE model of EC-pericyte crosstalk

Model assumptions

The ODE model of EC signaling was extended by adding the Notch signaling crosstalk of ECs with pericytes, as motivated by previous experiments^{6,23,36,37,61}. Pericytes were assumed to express Notch3, which could be activated by Jag1 in neighboring ECs^23,61. The interaction between Jag1 in ECs and Notch3 in pericytes was assumed to be equivalent to the interaction between Dll4 and Notch1, with thus the same model parameters. Notch3 activation following Jag1-Notch3 complex formation caused Dll4 expression in pericytes, in the same fashion as the Vegf-induced Dll4 expression in ECs. The Dll4 ligands that were expressed in pericytes could then bind and activate Notch1 present in neighboring ECs⁶¹, in the same way as the Dll4 expressed in ECs. Bmp9 was assumed to initiate Jag1 expression in ECs⁶, in the same order of magnitude as the VEGF-mediated expression of Dll4 in tip cells. In addition to the Notch3 in pericytes, Jag1 could also bind and activate Notch1 in ECs²³. Notch1-Jag1 binding was assumed to occur at a comparable rate compared to Notch1-Dll4 binding, with this rate linearly proportional with LFng³⁷. Activation of Notch1 by Jag1 was assumed to occur at a much lower rate compared to Dll4-Notch1 activation, with the rate further decreased (10 times) by LFng, in a step-wise fashion³⁷. Finally, the Notch1-Jag1 complex dissociation rate was tuned to observe a considerable inhibiting effect of Jag1 for Notch signaling, in the case of EC monoculture with contemporaneous coating of both Jag1 and Dll4²³.

Model equations

Again, the ensemble of Notch receptors and ligands was divided into two groups, one for each cell edge. We describe here the equations for the left edge with subscript $L$. Analogous equations were used for the right edge. The equations for Vegf signaling and filopodia formation, as well as the equations of Dll4 and Hes1/Hey1 in ECs, and Bmp9 effects on LFng are not reported, since Eqs. 1-5, 8, and 10-12 were left unchanged. The remaining equations are described highlighting the changes compared to the EC signaling model alone.

Bmp9 induced Jag1 ($J_i$) expression in ECs, at a rate linearly dependent on Bmp9:

$${\beta }_{J\_bmp}=h_{J\_bmp}$$ (13)

Jag1 ($J_i$) could bind to: Notch1 ($N_i$) in neighboring ECs, forming protein complexes $J_i.N_i$ at a rate $k_{2\_bmp}$ influenced by Bmp9 via LFng; Notch3 ($N_{3i}$) in neighboring pericytes, forming protein complexes $J_i.N_{3i}$ at a rate independent from Bmp9 ($k_2$). These protein complexes can dissociate at a rate $k_{-2J}$ and $k_{-2}$, respectively, such that the time variation of Jag1 present on the left edge of the cell $i$ can be described as:

$$\begin{gathered} \frac{d{J}_{L_i}}{dt}=\frac{1}{2}\left({\beta }_{J_{bmp}}+{\theta }_J\frac{I_i^2}{1+I_i^2}\right)-k_{2_{bmp}}{J}_{L_i}{N}_{R_{i-1}}+k_{-2J}J.N_{R_{i-1}}-k_2{J}_{L_i}{N}_{3i}+k_{-2}J.N_{3L_i}-{\varphi }_D{J}_{L_i}+ \\ W\left(\frac{J_L+J_R}{2}-J_L\right) \end{gathered}$$ (14)

where ${\beta }_{J_{bmp}}$ describes the basal production of Jag1 in response to Bmp9 signaling while the term ${\theta }_J\frac{I_i^2}{1+I_i^2}$ describes Jag1 upregulation in response to Notch activation. The value ${\theta }_J=0$ sec⁻¹ was used to obtain the simulation results in Fig. 5B-G.

In the model, the Notch3 content in pericytes varies over time according to:

$$\frac{d{N}_{3i}}{dt}=\frac{{\beta }_D}{2}-2k_2{N}_{3_i}\left(J_{R_i}+J_{L_i}\right)+k_{-2}\left(J.N_{3R_i}+J.N_{3L_i}\right)-{\varphi }_N{N}_{3_i}$$ (15)

Upon protein complex formation, Notch3 is activated and leads to an increase in Notch3 intracellular domain ($I_{3_i}$). This upregulates pericyte expression of Dll4 ($P_i$), which can bind in turn to Notch1 in neighboring ECs forming Dll4-Notch1 protein complexes at the interface between ECs and pericytes ($P.N_{L_i}$):

$$\frac{d{I}_{3i}}{dt}=k_{cat}\left(J.N_{3L_i}+J.N_{3R_i}\right)-{\varphi }_I{I}_{3i}$$ (16)

$$\frac{d{P}_i}{dt}=\frac{1}{2}\left({\beta }_D+\theta \frac{I_{3i}^2}{1+I_{3i}^2}\right)-2k_{2_{bmp}}{P}_i{N}_{R_{i-1}}+k_{-2}(P.N_{R_{i-1}}+P.N_{L_{i-1}})-{\varphi }_D{P}_i$$ (17)

Notch1 in ECs can be activated by Dll4 ($D_i$) and Jag1 in ECs, as well as Dll4 ($P_i$) from pericytes, with which it forms protein complexes $P_i.N_i$. The interaction between Notch1 and Dll4 is always the same, irrespective of the cell source, such that the variation of Notch1 and related complexes can be tracked via the following ODEs:

$$\begin{gathered} \frac{d{N}_{L_i}}{dt}=\frac{{\beta }_N}{2}-2k_{2_{bmp}}{N}_{L_i}(D_{R_{i-1}}+P_i)+k_{-2}(D.N_{L_i}+P.N_{L_i})-k_{2_{bmp}}{N}_{L_i}{J}_{R_{i-1}}+k_{-2J}J.N_{L_i}- \\ {\varphi }_N{N}_{L_i}+W\left(\frac{N_L+N_R}{2}-N_L\right) \end{gathered}$$ (18)

$$\frac{dD.N_{L_i}}{dt}=2k_{2\_bmp}{N}_{L_i}{D}_{R_{i-1}}-k_{-2}D.N_{L_i}-{\varphi }_{D}D.N_{L_i}-k_{cat}D.N_{L_i}$$ (19)

$$\frac{dJ.N_{L_i}}{dt}=k_{2\_bmp}{N}_{L_i}{J}_{R_{i-1}}-k_{-2J}J.N_{L_i}-{\varphi }_{D}J.N_{L_{i_i}}-k_{cat\_LF}J.N_{L_i}$$ (20)

$$\frac{dP.N_{L_i}}{dt}=2k_{2\_bmp}{N}_{L_i}{P}_i-k_{-2}P.N_{L_i}-{\varphi }_{ND}P.N_{L_i}-k_{cat}P.N_{L_i}$$ (21)

The parameter $k_{cat\_LF}$ scales the activation of Notch1 by Jag1 upon binding, and it is assumed to be dependent on the amount of LFng in a step-wise fashion. Given that LFng was not modeled explicitly but only via its effects on Notch1 receptor-ligand affinity as quantified by the parameter $h_{2\_bmp}$, the same parameter was used for the effects of LFng on Jag1-mediated Notch1 activation:

$$k_{cat\_LF}=\{\begin{array}{cc}{k}_{cat}∕10,\hfill & h_{2_{bmp}}<h_{min}\hfill \\ k_{cat}∕100,\hfill & h_{2_{bmp}}\ge h_{min}\hfill \end{array}\phantom{\}}$$ (22)

Finally, Notch1 activation led to an increase in Notch1 intracellular domain ($I_i$) as follows:

$$\frac{d{I}_i}{dt}=k_{cat}(D.N_{L_i}+D.N_{R_i}+P_i.N_{L_i}+P_i.N_{R_i})+k_{cat\_LF}(J.N_{L_i}+J.N_{R_i})-{\varphi }_I{I}_i$$ (23)

The values of parameters already present in the EC signaling model were left unchanged. The values of the remaining new parameters are reported in Table S3.

Simulations

Each EC was assumed to be in contact with one pericyte, and interacting via Notch signaling. In vitro RT-qPCR experiments with EC monolayers and co-culture with pericytes were simulated by assuming that cells were exposed to insignificant Vegf ($V_0$ = 0 cu). The ODEs were solved with a final time equal to 48h, and external Bmp9 was added only 24h hours after the start of the experiment. The parameter values $h_{H\_bmp}$ and $h_{2\_bmp}$ quantifying Bmp9 effects on Hes1/Hey1 and LFng were left unchanged compared to the EC monoculture experiments. Jag1 expression without Bmp9 was assumed to be two orders of magnitude lower compared to the Bmp9-induced value. Jag1 siRNA was simulated by imposing an unchanged expression despite Bmp9 injection. LFng siRNA was simulated analogously.

EC stabilization occurring at the end of angiogenesis was simulated by adopting the same ODE model. For the first 24h, only ECs were simulated, exposed to VEGF ($V_0$ = 0.1 cu) to obtain an initial tip/stalk pattern. For the following 24h, as in the simulation of tip/stalk pattern formation, Bmp9 injection was modeled by increasing the basal expression of Hes1/Hey1 and LFng ($h_{H\_bmp}$ = 20, and $h_{2\_bmp}$ = 3). In addition, Bmp9 induced Jag1 expression in the same fashion as in the RT-qPCR experiments, and pericytes were added to simulate the Jag1-mediated EC-pericyte signaling crosstalk. In the simulations of EC stabilization with co-varied LFng and Jag1, to better track the different roles of the two proteins, the values of the parameters for Jag1 and LFng were modelled independently despite LFng involvement in Jag1 regulation.

Analysis of EC stabilization rate

The time necessary to stabilize ECs was evaluated in the same way as for the tip/stalk pattern formation. ECs were assigned a tip cell phenotype if $f_{min}$ > 25 c. u. and a stalk cell phenotype otherwise. The vessel was assumed to be stabilized when all ECs exhibited a stalk cell phenotype. For an approximation of the EC stabilization rate, analogously to the tip/stalk patterning rate, the time necessary to obtain all stalk cells with different conditions was normalized over 1 day, taken as a reference value.

Quantification and statistical analysis

All statistical analyses were performed using GraphPad Prism (version 10). Data are presented as mean ± standard deviation (SD). The exact value of n, along with the definition of n (e.g., number of biological replicates, number of independent cell culture experiments, or number of animals), and all statistical tests applied to each dataset are reported in the corresponding figure legends. Normality of residuals and homogeneity of variance were assessed using the Shapiro–Wilk test prior to applying parametric analyses. Comparisons between more than two groups were performed using one-way ANOVA, and two-way ANOVA was used to assess the interaction between Dll4 coating and BMP9 treatment on Notch target gene expression. No statistical methods were used to predetermine sample size. Randomization and blinding were not applicable to in vitro cell-based experiments. All experimental replicates were included unless predefined technical failure criteria were met (e.g., RNA integrity). Statistical significance was defined as P < 0.05.

Supplementary Material

1

Video S1: Representative time-lapse confocal movie of EC nuclei dynamics in ISV sprouts in control (MOC) Tg(kdrl:nlsEGFP)^zf109 embryos, related to Figure 3. The video’s total elapsed time was 1 day, 10 hours and 50 minutes.

2

Video S2: Representative time-lapse confocal movie of EC nuclei dynamics in ISV sprouts in lfng-KD Tg(kdrl:nlsEGFP)^zf109 embryos, related to Figure 3. The video’s total elapsed time was 1 day, 8 hours and 37 minutes.

Key Resources Table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse anti-human β-actin	Santa Cruz	Cat# sc-4778
Mouse anti β-actin	Sigma-Aldrich	Cat# A1978
Rabbit anti-human Alk1	Abcam	Cat# ab68703
Goat anti-mouse Alk1	R&D Systems	Cat# AF770
Goat anti-human Jagged1	R&D Systems	Cat# AF1277
Rat anti-human DLL4	R&D Systems	Cat# MAB1506
Rabbit anti-human Notch3 (D11B8)	Cell Signaling	Cat# 5276
Rabbit anti-human SMAD4 (D3M6U)	Cell Signaling	Cat# 38454
Rabbit anti-human, mouse Lunatic Fringe (D6V2V)	Cell Signaling	Cat# 66472
Rabbit anti-human Cleaved Notch1 (Val1744) (D3B8)	Cell Signaling	Cat# 4147
Rabbit anti-human Notch1 (D1E11)	Cell Signaling	Cat# 3608
Donkey anti-Rabbit IgG (H+L) Alexa Fluor™ 594	Thermo Fisher	Cat# A21207
Donkey anti-Rabbit IgG (H+L) Alexa Fluor™ 647	Thermo Fisher	Cat# A31573
Horse Anti-Mouse IgG Antibody (H+L), Peroxidase	Vector	Cat# PI-2000
Goat Anti-Rabbit IgG Antibody (H+L), Peroxidase	Vector	Cat# PI-1000
Horse Anti-Goat IgG Antibody (H+L), Peroxidase	Vector	Cat# PI-9500
Isolectin GS-IB4 From Griffonia simplicifolia, Alexa Fluor™ 488 Conjugate	Thermo Fisher	Cat# I21411
Cell lines
Human Umbilical Vein Endothelial Cells (HUVECs)	Promocell	Cat# C-12203
Human Pericytes	Promocell	Cat# C-12980
HEK293T	ATCC	Cat# CRL-3216
Chemicals, peptides, and recombinant proteins
Recombinant Human BMP-9 Protein, CF	R&D Systems	Cat# 3209-BP/CF
Recombinant Human DLL4 His-tag Protein	R&D Systems	Cat# 1506-D4
Recombinant Human Jagged 1 His-tag Protein, CF	R&D Systems	Cat# 10957-JG
Recombinant Human VEGF 165 Protein	R&D Systems	Cat# BT-VEGF
Cytodex® 3 microcarrier beads	Sigma-Aldrich	Cat# C3275
Fibrinogen from bovine plasma	Sigma-Aldrich	Cat# F8630
Corn oil	Sigma-Aldrich	Cat# C8267
Aprotinin	Sigma-Aldrich	Cat# A3428
Thrombin from bovine plasma	Sigma-Aldrich	Cat# T4648
Tamoxifen	Sigma-Aldrich	Cat# T2859
Bromophenol Blue	Sigma-Aldrich	Cat# B0126
2-Mercaptoethanol	Sigma-Aldrich	Cat# 63689
Dimethyl sulfoxide	Sigma-Aldrich	Cat# D2650
Collagenase Type II	Sigma-Aldrich	Cat# C2-22-1G
Puromycin dihydrochloride	Sigma-Aldrich	Cat# P9620
Lipofectamine RNAiMAX	Thermo Fisher	Cat# 13778030
CD31 Dynabeads	Thermo Fisher	Cat# 11155D
SuperSignal West Pico PLUS Chemiluminescent Substrate	Thermo Fisher	Cat# 34580
RNeasy Mini Kit	Qiagen	Cat# 74104
RNeasy Micro Kit	Qiagen	Cat# 74004
Plasmid Midi Kit	Qiagen	Cat# 12143
Fetal Bovine Serum (FBS)	Gibco	Cat# 10270-106
DMEM, High Glucose	Gibco	Cat# 11965092
Trypsin-EDTA (0.05%), phenol red	Gibco	Cat# 25300062
Penicillin-Streptomycin	Gibco	Cat# 22571038
PBS (10X), pH 7.4	Gibco	Cat# 70011051
iScript Reverse Transcription Supermix	Bio Rad	Cat# 1708840
iQ SYBR Green Supermix	Bio Rad	Cat# 1708880
BJ Fibroblasts	ATCC	Cat# CRL-2522
HEK-293 Cells	ATCC	Cat# CRL-1573
HUVECs	Promocell	Cat# C-12203
EGM-2 Endothelial Cell Growth Medium-2 BulletKit	Lonza	Cat# CC-3162
Endothelial Cell Basal Medium 2	Promocell	Cat# C-22211
Pericyte Growth Medium 2	Promocell	Cat# C-28041
Human Pericytes	Promocell	Cat# C-12980
Fluoroshield™ with DAPI	Sigma-Aldrich	Cat# F6057
Paraformaldehyde 16%	Electron Microscopy Sciences	Cat# 15710
Trypan Blue Solution, 0.4%	Gibco	Cat# 15250061
Fugene HD	Promega	Cat# E2311
Cycloheximide, 95%	Sigma-Aldrich	Cat# 01810
qRT-PCR Primers and siRNA
Hs_HEY1_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_HES1_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_LFNG_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_RFNG_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_MFNG_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_ACTB_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_ACVRL1_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_PDGFRB_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_CDH5_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_PECAM1_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Mm_Hes1_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Mm_Hey1_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Mm_Lfng_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Mm_Actb_1_SG QuantiTect Primer Assay	Qiagen	Cat# 249900
Hs_LFNG_10 FlexiTube siRNA	Qiagen	Cat# 1027417
Hs_ACVRL1_5 FlexiTube siRNA	Qiagen	Cat# 1027417
Hs_JAG1_5 FlexiTube siRNA	Qiagen	Cat# 1027417
Hs_DLL4_5 FlexiTube siRNA	Qiagen	Cat# 1027417
AllStars Negative Control siRNA	Qiagen	Cat# 1027280
Control morpholino oligonucleotide: 5’- CCTCTTACCTCAGTTACAATTTATA -3’	Gene Tools	Cat# 2934999000
lfng morpholino oligonucleotide: 5’- ACCGTGTATACCTGTCGCATGTTTC -3’	Gene Tools	ID# ZDB-MRPHLNO-100510-10
Plasmids
pReceiver-Lv205-hLFNG	Genecopoeia	Cat# EX-H1563-LV205-GS
pReceiver-LV205	Genecopoeia	Cat# EX-Neg-LV205
pLV-mCherry	Addgene	Cat# 36084
pRSV-Rev	Addgene	Cat# 12253
pCMV-VSV-G	Addgene	Cat# 8454
Experimental models: Organisms/strains
Mouse: C57BL/6	Charles River Laboratories	Stock #027
Mouse: Cdh5-CreERT2 Mouse	Ralf Adams	N/A
Mouse: Alk1LoxP	Paul S Oh	N/A
Mouse: Gt(ROSA)26Sortm4(ACTB-tdTomato,-EGFP)Luo/J	Jackson Laboratory	Strain# 007576
Zebrafish: Tg(kdrl:nlsEGFP)zf109	Markus Affolter89	ID : ZDB-ALT-081105-1
Software and algorithms
BioRender	N/A	N/A
GraphPad Prism 10.1.2 ImageStudio Lite	LI-COR	Version 5.2
Matlab	MathWorks	R2023a
Computational code	N/A	DOI: 10.4121/37b88a57-230e-4b04-ba9f-b1fb229181c7

Highlights.

• Bmp9 upregulates LFng expression in endothelial cells, enhancing Notch activation.

• LFng regulates the temporal dynamics of tip/stalk selection and cell rearrangement.

• LFng plays a role in the effects of Alk1 deficiency on vascular formation.

• Bmp9-upregulated LFng mediates endothelial cell-pericyte crosstalk.