Dual function of perivascular fibroblasts in vascular stabilization in zebrafish

Blood vessels are vital to sustain life in all vertebrates. While it is known that mural cells (pericytes and smooth muscle cells) regulate vascular integrity, the contribution of other cell types to vascular stabilization has been largely unexplored. Using zebrafish, we identified sclerotome-derived perivascular fibroblasts as a novel population of blood vessel associated cells. In contrast to pericytes, perivascular fibroblasts emerge early during development, express the extracellular matrix (ECM) genes col1a2 and col5a1, and display distinct morphology and distribution. Time-lapse imaging reveals that perivascular fibroblasts serve as pericyte precursors. Genetic ablation of perivascular fibroblasts markedly reduces collagen deposition around endothelial cells, resulting in dysmorphic blood vessels with variable diameters. Strikingly, col5a1 mutants show spontaneous hemorrhage, and the penetrance of the phenotype is strongly enhanced by the additional loss of col1a2. Together, our work reveals dual roles of perivascular fibroblasts in vascular stabilization where they establish the ECM around nascent vessels and function as pericyte progenitors.


Introduction
The vascular system is crucial to the survival of vertebrates. Blood vessels must rapidly expand and contract in response to systemic cues, but also be sturdy to withstand the stress of blood flow. To maintain their integrity, blood vessels are supported by a highly specialized perivascular architecture comprised of blood vessel associated cells and the surrounding extracellular matrix (ECM) [1,2]. Compromised vascular integrity can result in devastating human diseases, such as aneurysms, vascular malformations, and hemorrhagic strokes [3][4][5][6]. However, how blood vessels are stabilized by different perivascular components is still poorly understood.
The prevailing model is that vascular stability is maintained at three different levels: endothelial cells, mural cells and the surrounding ECM [2]. First, blood vessels are lined by endothelial cells. Adherens and tight junctions between endothelial cells provide the primary barrier to passage of fluids, cells, and macromolecules between blood and tissues [7]. Second, vascular smooth muscle cells (vSMCs) and pericytes, collectively known as mural cells, closely interact with and stabilize the endothelium [8]. vSMCs form a continuous protective sheath around large diameter blood vessels, whereas pericytes are solitary cells partially covering smaller blood vessels such as capillaries. Defects in mural cell specification or recruitment result in widespread vascular leakage and early lethality in mice [9][10][11][12][13]. Lastly, the vascular ECM provides structural support for blood vessels [1,14]. Mutations in ECM proteins, such as collagens and laminins, lead to early death due to blood vessel rupture in mice [15][16][17][18]. In humans, defects in fibril forming collagens (type I, III and V), as well as molecules involved in collagen modification and processing, have been associated with Ehlers-Danlos syndrome (EDS) [19]. EDS patients are characterized by a significant vascular fragility that often leads to spontaneous rupture of blood vessel walls. Thus, work in both animal models and human patients shows the importance of collagen in maintaining vascular stability. However, one unresolved question is how newly formed blood vessels are stabilized in the early time window before the differentiation of mural cells.
vSMCs and pericytes are the most well-studied blood vessel associated cells and are classically defined by the expression of alpha smooth muscle actin (ACTA2) and platelet-derived growth factor receptor b (PDGFRb), respectively [20,21]. Lineage tracing in mice and chickquail chimeras demonstrate a heterogeneous developmental origin of mural cells [20][21][22][23]. For example, vSMCs covering the trunk aorta originate from either the neural crest or the sclerotome of the somite, depending on the axial position. Other than mural cells, several types of perivascular cells have been identified, including adventitial cells [24], fibroblasts [25], immune cells [26], and astrocytes [27]. However, whether these different cell types play roles in vascular stabilization is unknown.
Recent single cell transcriptional profiling studies in mice have revealed the presence of perivascular fibroblast-like cells in the adult mouse brain that are not labeled by classical mural cell markers [28][29][30][31]. In contrast to mural cells, perivascular fibroblast-like cells adhere loosely to blood vessels and show robust expression of many ECM components, such as collagens. Previous studies have described similar cell populations in the mouse central nervous system that contribute to fibrotic scar formation after injury [32][33][34][35]. However, where these perivascular fibroblast-like cells originate from during development and how they regulate blood vessel development and stabilization have not been explored.
Zebrafish is a powerful model to study human cardiovascular diseases [36][37][38]. The organization, development and function of the vasculature, including mural cells, are highly conserved between zebrafish and mammals [36,[39][40][41]. In the zebrafish trunk, intersegmental vessels (ISVs) sprout from the dorsal aorta (DA) at around 24 hpf (hours post fertilization) and fully establish the metameric pattern by 36 hpf [42]. Interestingly, pericytes, the primary mural cells along the ISVs, only emerge at 60 hpf [43]. This raises the question how nascent blood vessels are stabilized prior to the differentiation of pericytes. In our work, we describe a novel population of collagen-expressing perivascular fibroblasts that become closely associated with ISVs soon after vessel formation. Perivascular fibroblasts originate from the sclerotome and are distinct from pericytes in their morphology, distribution, and marker expression. Using a combination of in vivo live imaging, genetic ablation and CRISPR mutant analysis, we demonstrate that perivascular fibroblasts play dual roles in stabilizing nascent ISVs by regulating the vascular ECM and later functioning as pericyte progenitors. Together, our work provides important insights into the development of perivascular support structures and suggests a molecular and cellular basis of Ehlers-Danlos syndrome.

Perivascular fibroblasts express several collagen genes
We previously developed a sclerotome-specific transgenic line nkx3.1:Gal4; UAS:Nitroreductase-mCherry (nkx3.1 NTR-mCherry , similar designations are used for all Gal4/UAS transgenic lines in this paper) [44]. This reporter labels the initial sclerotome domains as well as their descendants due to the perdurance of the mCherry protein. Examination of the nkx3.1 NTR-mCherry line in combination with an endothelial reporter, kdrl:EGFP, revealed that a population of mCherry + cells was closely associated with intersegmental vessels (ISVs) at 48 hpf ( Fig 1A). This result suggests that the sclerotome contributes to a population of perivascular cells well before the appearance of pericytes at 60 hpf [43]. Interestingly, nkx3.1 NTR-mCherry -expressing perivascular cells at 52 hpf were also labeled by a pan-fibroblast reporter, col1a2:GFP [44] ( Fig  1B). On average, 37% of mCherry + perivascular cells were positive for col1a2:GFP (S1A Fig), which likely reflects the mosaic nature of the col1a2:GFP line. Consistent with this result, fluorescent in situ hybridization showed that ISV-associated perivascular cells co-expressed several ECM genes, including col1a2 (collagen 1a2) and col5a1 (collagen 5a1) at 48 hpf (Figs 1C and S1B). These collagen-expressing perivascular cells are reminiscent of perivascular fibroblastlike cells recently identified in the adult mouse brain by single cell RNA sequencing [28][29][30][31]. We therefore refer to this population of collagen-expressing ISV-associated cells as 'perivascular fibroblasts'.

Perivascular fibroblasts originate from the sclerotome
The expression of nkx3.1 NTR-mCherry in perivascular fibroblasts suggests that the sclerotome is the embryonic source of these cells. We have previously shown that the sclerotome in zebrafish has a bipartite organization with a ventral domain and a smaller dorsal domain [44]. The ventral sclerotome domain further gives rise to a population of notochord associated cells. At 24 hpf, the sclerotome consists of three compartments: the dorsal sclerotome domain, the ventral sclerotome domain, and notochord associated cells derived from the ventral domain (Fig 2A).
To determine the contribution of these three compartments to perivascular fibroblasts, we performed confocal time-lapse imaging in nkx3.1 NTR-mCherry ; kdrl:EGFP embryos. The nkx3.1 NTR-mCherry reporter labels sclerotome progenitors and their progeny, while the endothelial specific kdrl:EGFP line allows us to visualize the ISVs. Perivascular fibroblasts can be distinguished from tenocytes based on their perivascular locations as well as the lack of long tree-like processes extending into the intersomitic space, a characteristic unique to tenocytes [44]. The trunk region of embryos (somite [12][13][14][15][16][17][18] was imaged laterally at 6-9 minute intervals starting at 25 hpf, immediately after the emergence of ISV sprouts (Fig 2B and S1 Video). At 30 hpf, most ISVs became visibly lumenized. During ISV sprouting and lumenization, mCherry + cells can be observed associated with the kdrl:EGFP-labeled endothelium (Figs 2B and S2A; S1 Video). By the end of the movie at 49.5 hpf, mCherry + perivascular fibroblasts 'decorated' the entire ISVs along the dorsal-ventral (D-V) axis. Interestingly, in the absence of ISVs, numerous mCherry + sclerotome derived cells can still populate the interstitial space between muscles, the spinal cord and the notochord (S2B Fig), suggesting that the migration and proliferation of sclerotome progenitors and their descendants are independent of the vasculature. By retrospective cell tracing, we showed that perivascular fibroblasts can be traced to all three compartments of the sclerotome. Of 170 perivascular fibroblasts traced, 28 cells originated from the dorsal sclerotome domain (16%), 112 from the ventral sclerotome domain (66%), and 30 from notochord associated cells (18%). Since notochord associated cells are derived from the ventral sclerotome domain, this result suggests that the ventral sclerotome is the main contributor of perivascular fibroblasts (142/170 cells, 84%).
To better compare the contribution of sclerotome progenitors from different compartments, we subdivided perivascular fibroblasts into two domains based on their final D-V positions along the ISVs. The position of the horizontal myoseptum was used as a landmark to define perivascular fibroblasts at dorsal or ventral ISVs (Fig 2A). The 170 perivascular fibroblasts can be traced back to 122 sclerotome progenitors. These sclerotome progenitors typically underwent 1-2 cell divisions over the 24-hour period to give rise to multiple daughter cells, at least one of which became associated with the neighboring ISV. We categorized sclerotome progenitors into three groups based on the final positions of perivascular fibroblasts they generated: 1) only at the dorsal ISV, 2) only at the ventral ISV, or 3) at both dorsal and ventral ISV ( Fig 2C). Sclerotome progenitors in the dorsal domain gave rise to exclusively perivascular fibroblasts along dorsal ISVs (20/20 cells, 100%) ( Fig 2C). Similarly, the majority of sclerotome progenitors around the notochord (i.e., sclerotome derived notochord associated cells) (21/23 cells, 91%) contributed to only perivascular fibroblasts at the dorsal ISV ( Fig 2C). By contrast, 48% of sclerotome progenitors in the ventral domain (38/79 cells) gave rise to only perivascular fibroblasts at the ventral ISV, 35% of ventral sclerotome progenitors (28/79 cells) generated only dorsal perivascular fibroblasts, while the remaining 16% of ventral progenitors (13/79 cells) contributed to perivascular fibroblasts at both dorsal and ventral ISV regions (Fig 2C). Similar to the contribution of tendon fibroblasts (tenocytes) [44], the dorsal sclerotome generates only dorsally positioned perivascular fibroblasts, whereas the ventral sclerotome contributes to perivascular fibroblasts along the entire D-V axis of ISVs (the combined bar in Fig 2C). Moreover, perivascular fibroblasts along a given ISV were always derived from the sclerotome of the same overlying somite (170/170 cells, 100%). Together, our results demonstrate that

Perivascular fibroblasts are distinct from pericytes
Mural cells, including pericytes and vSMCs, are known blood vessel associated support cells. We next asked whether perivascular fibroblasts represent a different perivascular cell population from mural cells. In the zebrafish trunk, pericytes are associated with ISVs, while vSMCs are localized to larger vessels such as the dorsal aorta [40,41,43]. Platelet derived growth factor (PDGF) signaling is known to play an important role in pericyte recruitment and the platelet derived growth factor receptor-beta (pdgfrb) is a well-established pericyte marker [43,45,46]. We utilized the pdgfrb:GFP [47] and pdgfrb:Gal4FF; UAS:NTR-mCherry (referred to as pdgfrb NTR-mCherry ) [48] lines to label pericytes. Interestingly, while both pdgfrb:GFP and pdgfrb NTR-mCherry lines labeled pericytes (GFP high mCherry + expression) at 4 dpf, pdgfrb:GFP but not pdgfrb NTR-mCherry also marked perivascular fibroblasts (GFP low mCherryexpression) at 2 and 4 dpf (days post fertilization) (S3A and S3B Fig). Fully differentiated pericytes can be reliably identified based on their elevated pdgfrb:GFP expression (2.5 fold) compared to perivascular fibroblasts (S3C Fig). The expression of pdgfrb:GFP in sclerotome descendants is consistent with the previous report that the sclerotome expresses pdgfrb [49], suggesting that pdgfrb:GFP is a more sensitive reporter than the pdgfrb NTR-mCherry line. Indeed, pdgfrb NTR-mCherry labeled about 62% of pdgfrb:GFP-positive pericytes at 4 dpf (S3D Fig). Using the two pdgfrb lines for pericytes as well as the nkx3.1 NTR-mCherry and col1a2:GFP reporters for perivascular fibroblasts, we found that perivascular fibroblasts and pericytes represent distinct cell populations. First, they showed different developmental timing and cell numbers. Consistent with our time-lapse movies, nkx3.1 NTR-mCherry -positive perivascular fibroblasts were present along ISVs at 2 dpf ( Fig 3A). They were equally distributed along arterial and venous ISVs, and remained largely constant in number (ranging from 9.7 to 10.6 per ISV) from 2 to 4 dpf ( Fig 3A). By contrast, pericytes marked by high level of pdgfrb:GFP expression did not appear on ISVs until 3 dpf. They were more concentrated along arterial ISVs than venous ISVs, and increased in cell number to 2.2 per aISV and 1.0 per vISV by 4 dpf (Fig 3A), similar to what has been previously described [43]. Second, perivascular fibroblasts and pericytes showed distinct marker expression (Fig 3B). Perivascular fibroblasts labeled by col1a2: organization of the zebrafish sclerotome and the generation of perivascular fibroblasts. At 24 hpf, the zebrafish sclerotome in each somite is divided into three compartments: the dorsal sclerotome, the ventral sclerotome, and sclerotome derived notochord associated cells. Note that notochord associated cells originate from the ventral sclerotome and are located about half-somite posterior to the corresponding somite. At 48 hpf, perivascular fibroblasts appear along the length of the intersegmental vessels (ISVs, green). The dotted line indicates the position of the horizontal myoseptum, which divides any given ISV into a dorsal half and a ventral half. Perivascular fibroblasts were quantified based on their final locations along each ISV: dorsal ISV (above the horizontal myoseptum) or ventral ISV (below the horizontal myoseptum). Sc: sclerotome. (B) Snapshots from time-lapse imaging of a nkx3.1 NTR-mCherry ; kdrl:EGFP embryo from 25 hpf to 49.5 hpf. Perivascular fibroblasts along ISVs were retrospectively traced to determine their cell of origin. One cell from the ventral sclerotome domain (cyan arrows) and one cell from sclerotome derived notochord associated cells (yellow arrows) were traced over 24.5 hours with their daughter cells indicated by the same colored arrows/arrowheads. Both sclerotome progenitors divided at least once to give rise to one perivascular fibroblast (arrowheads) as well as several interstitial cells (arrows). A schematic representation of color-coded traced cells at the last time point is shown with perivascular fibroblasts indicated by asterisks. The corresponding time-lapse movie is shown in S1 Video. n = 6 embryos. GFP did not express the pdgfrb NTR-mCherry reporter at 4 dpf. Conversely, pdgfrb NTR-mCherry -positive pericytes at 4 dpf showed minimal col1a2:GFP expression. Further, while perivascular fibroblasts appeared globular, pericytes were more elongated with long cellular processes ( Fig  3B). Perivascular fibroblasts were also more laterally located along ISVs compared to pericytes, with pericyte processes often sandwiched between the neighboring perivascular fibroblasts and the endothelium (Fig 3B). To further examine the morphological differences between perivascular fibroblasts and pericytes, we performed high resolution confocal microscopy at the single cell resolution at 3 dpf. pdgfrb NTR-mCherry -positive pericytes showed the typical pericyte morphology: small and flat cell bodies with several elongated cellular processes that extended longitudinally along and tightly wrapped around the ISV, labeled by kdrl:EGFP ( Fig 3C). Since perivascular fibroblasts are more abundant along ISVs compared to pericytes, we took advantage of a mosaic col1a2:Gal4; UAS:Kaede line (referred to as col1a2 Kaede ) [50] to sparsely label single perivascular fibroblasts for easier visualization. Unlike the tight association between pericyte and the endothelium (10/10, Fig 3C), 36% of perivascular fibroblasts (5/14) showed visible gaps between the cell body and the underlying endothelium (Fig 3D), suggesting a looser association. In contrast to pericytes, perivascular fibroblasts had a globular morphology with short processes that wrapped diametrically around the associated ISV in an 'awkward hug' (Fig 3D). Quantification of cellular morphologies revealed key differences between these two cell types (Fig 3E-3I). In contrast to perivascular fibroblasts, pericytes displayed a significantly larger aspect ratio (Fig 3F), suggesting that they were more elongated along the long axis of the ISVs. Moreover, perivascular fibroblasts possessed more but shorter cellular processes compared to pericytes (Fig 3G-3I). Taken together, our results suggest that perivascular fibroblasts and pericytes represent two distinct populations of perivascular cells with unique marker expression, morphology, distribution and developmental timing.

A sub-population of perivascular fibroblasts functions as pericyte progenitors
While occupying the similar perivascular space, perivascular fibroblasts appear along ISVs at least one day before pericytes (Fig 3A). We hypothesized that perivascular fibroblasts act as progenitors for ISV pericytes. To test this possibility, we conducted time-lapse imaging of pdgfrb NTR-mCherry ; col1a2:GFP embryos from 54 to 73 hpf during which pericytes started to appear on ISVs. Interestingly, a subset of col1a2:GFP-positive perivascular fibroblasts gradually lost GFP expression, switched on the pericyte reporter pdgfrb NTR-mCherry , and developed long cellular processes characteristic of pericytes at this stage ( Fig 4A and S2 Video). Indeed, of 15 venous ISVs (vISVs) at 2, 3, and 4 dpf. The number of perivascular fibroblasts and pericytes were scored in nkx3.1 NTR-mCherry ; kdrl:EGFP and pdgfrb:GFP; kdrl:mCherry embryos, respectively. Arterial and venous ISV identity was determined based on the connection to either the dorsal aorta (DA) or posterior caudal vein (PCV), respectively. Each data point represents the average cell number of 8-10 ISVs from an individual embryo. Data are plotted with mean ± SEM indicated. n = 11-15 (nkx3.1 NTR-mCherry ; kdrl:EGFP) and 17 (pdgfrb:GFP; kdrl: mCherry) embryos at each time point. (B) pdgfrb NTR-mCherry ; col1a2:GFP embryos were imaged at 4 dpf to visualize perivascular fibroblasts and pericytes. No overlap in marker expression was observed between pdgfrb NTR-mCherry -positive pericytes (red, arrows) and col1a2:GFPpositive perivascular fibroblasts (green, arrowheads). Unlike perivascular fibroblasts, pericytes also displayed elongated cellular processes (notched arrowheads). n = 14 embryos. (C) pdgfrb NTR-mCherry ; kdrl:EGFP embryos were imaged at 3 dpf to visualize individual pericytes (green, asterisks) with long cellular processes (notched arrowheads) that wrapped around the ISV (red). (D) Mosaic col1a2 Kaede line was imaged to visualize a single perivascular fibroblast (green, asterisks) associated with an ISV (red) in col1a2 Kaede ; kdrl:mCherry embryos at 3 dpf. (E-I) Cell morphology, number of processes, and process length were quantified and graphed for single perivascular fibroblasts and pericytes using ImageJ as shown in (E). Perivascular fibroblasts showed an overall globular morphology as indicated by smaller aspect ratio (F) with more abundant (G) but shorter processes (H, I) compared to pericytes.  newly formed pericytes traced in our time-lapse movies, 60% were derived from col1a2:GFPexpressing perivascular fibroblasts (9/15 cells). This result suggests that a sub-population of perivascular fibroblasts can differentiate into pericytes. Since col1a2 transgenic lines such as col1a2:GFP are mosaic (S1A and S3E Fig) [50], the number above likely under-estimates the proportion of pericytes derived from perivascular fibroblasts. To further test our model, we performed time-lapse experiments in nkx3.1 NTR-mCherry ; pdgfrb:GFP embryos. Similar to movies in pdgfrb NTR-mCherry ; col1a2:GFP embryos, cell tracing in nkx3.1 NTR-mCherry ; pdgfrb:GFP embryos showed that about 8% of mCherry + GFP low perivascular fibroblasts (100/1208 cells) at 54 hpf slowly upregulated GFP expression, and extended pericyte-like cellular processes wrapping around the ISV (S3 Video). Of 125 newly formed GFP high pericytes, 90% of them (113/ 125 cells) were derived from mCherry + cells, including perivascular fibroblasts residing along the ISV (101/113 cells), cells migrating from near the dorsal aorta (5/113 cells), as well as some cells with unclear origin due to incomplete imaging (7/113 cells). This result confirms that a sub-population of perivascular fibroblasts functions as pericyte precursors. Since perivascular fibroblasts originate from the sclerotome, we conclude that most pericytes along ISVs are also derived from the sclerotome.
To further test if perivascular fibroblasts serve as pericyte precursors, we performed genetic ablation of perivascular fibroblasts using the col1a2:Gal4; UAS:NTR-mCherry line (referred to as col1a2 NTR-mCherry ). Nitroreductase (NTR) converts metronidazole (MTZ), an otherwise harmless prodrug into a cytotoxic compound, resulting in the death of NTR-expressing cells [51]. To ablate perivascular fibroblasts prior to pericyte differentiation, col1a2 NTR-mCherry ; pdgfrb:GFP embryos were treated with MTZ from 25 to 54 hpf, after which the drug was washed off and embryos were imaged for pericytes at 4 dpf ( Fig 4B). In contrast to watertreated controls, we observed a complete loss of perivascular fibroblasts in MTZ-treated embryos, with only residual mCherry + debris visible in the trunk ( Fig 4C). Interestingly, we observed a substantial loss of pericytes on both arterial ISVs (aISVs) and venous ISVs (vISVs) in MTZ-treated embryos compared to water-treated controls ( Fig 4D). This result is consistent with our model that perivascular fibroblasts function as pericyte precursors on both arteries and veins as depletion of perivascular fibroblasts leads to universal decrease in pericyte number on all ISVs. Together, our work suggests that the sclerotome generates perivascular fibroblasts, some of which further differentiate into pericytes along the ISVs.

Loss of perivascular fibroblasts results in dysmorphic ISVs
Our work suggests that perivascular fibroblasts function as progenitors to generate pericytes to support the vasculature. However, only a small fraction of perivascular fibroblasts actually differentiates into pericytes, raising question about the function of the remaining perivascular fibroblasts. Moreover, the emergence of perivascular fibroblasts occurs concurrently with ISV development (Figs 2B and S2A; S1 Video), well before pericyte differentiation at 60 hpf [43].
The time stamps are indicated in the hh:mm format. Schematic drawings of the merged images at each time point are shown at the bottom. The corresponding timelapse movie is shown in S2 Video. n = 7 embryos. (B) Schematic showing experimental timeline of perivascular fibroblast ablation. col1a2 NTR-mCherry ; pdgfrb:GFP embryos were incubated in either water or metronidazole (MTZ) from 25 to 54 hpf following which embryos were washed at 54 hpf and imaged at 4 dpf to visualize pericytes. (C) Representative images of water (left) or MTZ (right) treated col1a2 NTR-mCherry ; pdgfrb:GFP embryos at 4 dpf. Compared to water-treated control embryos, MTZ-treated embryos showed complete ablation of mCherry + cells with only residual mCherry + debris (notched arrowheads). Fewer pdgfrb:GFP high pericytes (arrowheads) can be seen in MTZ-treated embryos compared to water-treated controls. (D) Quantification of pericyte number after perivascular fibroblast ablation along aISVs and vISVs. Pericytes were identified based on high level expression of the pdgfrb:GFP reporter as shown in (C) and S3A and S3C Fig These observations raise the possibility that perivascular fibroblasts play an early role in stabilizing nascent blood vessels prior to pericyte formation. To test this idea, we examined the impact of early ablation of perivascular fibroblasts on ISV development using the nitroreductase-based system. nkx3.1 NTR-mCherry ; kdrl:EGFP embryos were incubated in water or MTZ for a 24 hour period starting at 38 hpf when ISV formation was complete and blood flow had commenced in ISVs (Fig 5A). ISV morphology was examined after the drug treatment at 62 hpf. Water-treated control embryos showed stereotypical ISVs with many mCherry + perivascular fibroblasts (Fig 5B). MTZ treatment resulted in a complete loss of perivascular fibroblasts, with only some mCherry + cell debris remaining (Fig 5B). Compared to control ISVs, most ISVs in ablated embryos showed visible distortions in their diameters, shrinking and dilating dramatically at different points along their length (Fig 5B). To quantify this variability, the diameter of each ISV was measured at four equidistant points along its length and standard deviation from the mean diameter was graphed as a readout for ISV diameter variability ( Fig 5C). Indeed, ISVs in the absence of perivascular fibroblasts were significantly more variable than those of water-treated controls ( Fig 5C). Examination of larger blood vessels in the trunk revealed that MTZ treated embryos also showed significantly reduced mean diameter for both the dorsal aorta (DA) and posterior cardinal vein (PCV), with substantial increase in PCV diameter variability (S4 Fig). These results suggest that sclerotome derived cells play a crucial role in maintaining the trunk vasculature during the early stage. Interestingly, similar ablation experiments using col1a2 NTR-mCherry ; kdrl:EGFP during later stages between 4 and 5 dpf did not significantly alter the ISV morphology (S5A- S5C Fig). Together, our results suggest that perivascular fibroblasts play an early role in the stabilization of nascent blood vessels but are dispensable for maintaining ISV morphology at later stages.

Perivascular fibroblasts deposit collagens around ISVs
Our work suggests that perivascular fibroblasts are crucial for the stabilization of nascent ISVs. Since perivascular fibroblasts show high-level expression of collagen genes, including col1a2 and col5a1, we predicted that perivascular fibroblasts are the main source of the vascular ECM around newly formed ISVs. To test this model, we generated a UAS:Col1a2-GFP construct to express GFP-fused Col1a2 in live embryos. A similar Col1a2-GFP reporter has been previously used to visualize collagen deposition during skin development and repair in zebrafish [52]. nkx3.1:Gal4; UAS:NTR-mCherry embryos injected with the UAS:Col1a2-GFP plasmid showed many mCherry + GFP + perivascular fibroblasts surrounding ISVs at 56 hpf ( Fig 5D). Indeed, numerous thin GFP + 'strings', likely corresponding to collagen fibers, wrapped around the ISV with the intensity increasing from 56 to 72 hpf, suggesting continuous Col1a2-GFP protein deposition (Fig 5D). To determine whether perivascular fibroblasts directly contribute to collagens around ISVs, we injected nkx3.1 NTR-mCherry embryos with a low dose of the UAS: Col1a2-GFP plasmid to achieve mosaic labeling of perivascular fibroblasts. Imaging of individually labeled mCherry + GFP + perivascular fibroblasts showed local Col1a2-GFP deposition around the ISV within 1-2 cell diameters from the Col1a2-GFP-expressing perivascular fibroblast ( Fig 5E). Our results suggest that perivascular fibroblasts contribute to collagens in the ECM surrounding nascent ISVs that likely contribute to vessel stabilization.
To visualize the dynamics of collagen deposition, we combined cell ablation experiments with a stable UAS:Col1a2-GFP transgenic line. nkx3.1:Gal4; UAS:NTR-mCherry; UAS: Col1a2-GFP (nkx3.1 NTR-mCherry;Col1a2-GFP ) embryos were incubated in water or MTZ from 38 to 62 hpf (Fig 5F). Embryos were imaged individually before and after the water/MTZ treatment to compare collagen deposition around the same ISVs (Fig 5G). Water-treated control embryos showed an average of 27% increase in Col1a2-GFP protein deposition around ISVs ( Fig 5G and 5H). By contrast, complete ablation of perivascular fibroblasts with MTZ treatment resulted in a 21% decrease in Col1a2-GFP signal (Fig 5G and 5H). This result suggests that perivascular fibroblasts play an active role in contributing to collagens in the vascular ECM during early stages of ISV development. To determine whether perivascular fibroblasts continue to deposit collagens at later stages, we performed similar cell ablation experiments in col1a2 NTR-mCherry;Col1a2-GFP embryos from 4 to 5 dpf (S5D Fig). Water-treated controls showed 8% increase in Col1a2-GFP level around ISVs, whereas collagen deposition in MTZ-treated embryos remained largely unchanged (S5E and S5F Fig). In summary, these results show that the vascular collagen network undergoes active remodeling during initial development, but becomes more stable at later stages, highlighting the early requirement of perivascular fibroblasts.

collagen mutants show severe ISV hemorrhage
Our work suggests that perivascular fibroblasts secrete collagens and stabilize nascent ISVs. We next asked whether Col1a2 and Col5a1, which are expressed in perivascular fibroblasts (Figs 1C and S1B), are required to maintain the integrity of nascent blood vessels. Using CRISPR/Cas9 editing [53], we generated mutants for col1a2 (col1a2 ca108 , 1-bp deletion) and col5a1 (col5a1 ca109 , 4-bp deletion) (S6A and S6B Fig). Both alleles resulted in frame shifts in the corresponding coding sequences leading to premature stops (S6C and S6D Fig). For simplicity, we designate wild-type, heterozygous or homozygous fish as +/+, +/-, and -/-, respectively. For example, heterozygous carriers of col1a2 ca108 are referred to as col1a2 +/-. Whole mount in situ hybridization showed that mutant mRNAs for both genes underwent nonsensemediated decay (Fig 6A), suggesting that both col1a2 ca108 and col5a1 ca109 are null alleles. Crosses of heterozygous col1a2 +/parents gave rise to progeny segregated in roughly Mendelian ratios at all stages examined (S7A Fig). Although homozygous col1a2 -/mutants were viable as adults, they were substantially smaller than their wild-type siblings and often showed severe spine curvature (S7B Fig). This phenotype is reminiscent of those described in other type I collagen mutants [54,55]. By contrast, from crosses of heterozygous col5a1 +/parents, homozygous col5a1 -/mutants were under-represented at 11 dpf, and completely absent by 14 dpf (S7C Fig). These results suggest that Col5a1 is a more critical component of the ECM compared to Col1a2, and it is required for the survival of the fish to adulthood. Strikingly, a small percentage of embryos from crosses of heterozygous col5a1 +/parents (2/88: 2%) showed spontaneous hemorrhage in the trunk region at 2 dpf and were genotyped as homozygous col5a1 -/- (Fig 6B), suggesting compromised vascular integrity in the absence of Col5a1.
To further characterize the hemorrhage phenotype in collagen mutants, we developed a phenotype screening procedure (Fig 6C). In this assay, we analyzed both single mutants as well as double mutants to test genetic interactions between col1a2 and col5a1. In addition, we tested whether the increase in physical stress exacerbates the hemorrhage phenotype by raising embryos in a high-viscosity medium (0.6% methylcellulose, MC) [56]. Briefly, heterozygous adults harboring either one or both of the collagen mutations (col1a2 +/-, col5a1 +/-, or col1a2 +/-; col5a1 +/-) were intercrossed to generate wild type, heterozygous, and homozygous sibling embryos for different combinations (Fig 6C). The resulting embryos were then incubated in either water or the viscous methylcellulose solution from 48 to 80 hpf, during which embryos were imaged and screened for hemorrhage in the trunk. Embryos were then genotyped to correlate the genotype with the phenotype. Similar to their wild-type col1a2 +/+ siblings (water: 0/ 14; MC: 0/27) and heterozygous col1a2 +/siblings (water: 0/39; MC: 0/38), homozygous col1a2 -/mutants did not show any trunk hemorrhage in either water (0/22) or the methylcellulose solution (0/21). This result suggests that loss of col1a2 alone does not compromise vascular stability even with increased physical stress. By contrast, while neither wild-type col5a1 +/ + embryos (0/11) nor heterozygous col5a1 +/siblings (0/39) showed any hemorrhage under normal conditions, 12% of col5a1 -/embryos (3/25) displayed obvious hemorrhages in the trunk at 2 dpf, which was worsened substantially by physical stress (17/26, 65%). Analysis of different allele combinations of double mutants revealed several key characteristics of the hemorrhage phenotype (Fig 6D). First, the loss of both col5a1 wild-type alleles was required to cause the trunk hemorrhage phenotype. Second, even though loss of col1a2 alone in col1a2 +/or col1a2 -/embryos was not sufficient to cause hemorrhage, loss of one or two col1a2 wildtype alleles progressively increased the penetrance of the hemorrhage phenotype in col5a1 -/mutants, from 23% in col1a2 +/-; col5a1 -/embryos (5/22) to 58% in col1a2 -/-; col5a1 -/embryos (7/12) (Fig 6D). Lastly, increased physical stress by incubating embryos in the viscous methylcellulose solution substantially exacerbated both the penetrance and the severity of the hemorrhage phenotype (Figs 6D, 6E and S7D). It is interesting to note that stress-induced blood vessel rupture is typical of the vascular phenotypes associated with classical Ehlers-Danlos syndrome in humans. Together, our mutant analysis suggests that Col5a1 and Col1a2 function redundantly to maintain blood vessel integrity with Col5a1 being a more critical component.

Discussion
In this study, we characterize the origin and function of perivascular fibroblasts, a novel population of blood vessel associated cells in zebrafish. Our work provides three main conclusions. First, perivascular fibroblasts originate from the sclerotome and are distinct from pericytes. Second, a subset of perivascular fibroblasts functions as pericyte progenitors. Third, cell ablation and mutant analysis reveal that perivascular fibroblasts stabilize nascent blood vessels through deposition of collagens, including Col1a2 and Col5a1. Together, we propose a dual role of perivascular fibroblasts in vascular stabilization: they first establish the ECM around nascent blood vessels for initial stabilization and then function as pericyte progenitors to generate pericytes that further contribute to vessel integrity (Fig 7). Our work provides new insights into the molecular and cellular mechanisms underlying vascular stabilization and vascular diseases, such as Ehlers-Danlos syndrome. Model of perivascular fibroblasts in vascular stabilization in zebrafish. Perivascular fibroblasts, characterized by the expression of fibrillar collagens col1a2 and col5a1 as well as low levels of pdgfrb, become associated with intersegmental vessels (ISVs) in the zebrafish trunk by 1.5 dpf. Perivascular fibroblasts deposit a network of collagen fibers to establish the vascular ECM and stabilize nascent ISVs at 2 dpf and continue to deposit collagen until at least 3 dpf. In the same time window, a subset of perivascular fibroblasts functions as pericyte progenitors. They gradually upregulate the expression of the classic pericyte marker pdgfrb and develop elongated cellular processes. By 3 dpf, these 'pericyte progenitors' have completely differentiated into mature pericytes, showing robust pdgfrb expression distinct from collagen-expressing perivascular fibroblasts. Together, perivascular fibroblasts perform dual functions in vascular stabilization by depositing collagens to support nascent blood vessels and acting as pericyte progenitors. https://doi.org/10.1371/journal.pgen.1008800.g007

Perivascular fibroblasts are a unique population of perivascular cells
Pericytes are well studied blood vessel support cells associated with intersegmental vessels in the zebrafish trunk. Four lines of evidence suggest that collagen-expressing perivascular fibroblasts represent a unique population of perivascular cells, distinct from mature and fully differentiated pericytes. First, these two cell populations show distinct marker expression, with fully differentiated pericytes labeled by high level of pdgfrb expression, whereas perivascular fibroblasts marked by the expression of collagen genes, such as col1a2 and col5a1. Second, perivascular fibroblasts become associated with ISVs as early as 31 hpf, whereas pericytes do not differentiate until at least 60 hpf [43]. Third, perivascular fibroblasts are much more abundant than pericytes, and are associated with both arterial and venous ISVs in similar numbers unlike pericytes. Finally, perivascular fibroblasts are positioned abluminal to pericytes and appear more globular with numerous shorter processes that extend around the endothelium. Based on the differences in marker expression, developmental timing, cell number, and cell morphology, we propose that perivascular fibroblasts represent a novel perivascular cell population, distinct from pericytes. Intriguingly, several single cell RNA sequencing studies have revealed a similar perivascular 'fibroblast-like' cell population in the adult murine brain [28][29][30][31]. Murine fibroblast-like cells show robust expression of ECM genes, including collagen genes col5a1 and col1a2, but much less mural cell markers such as pdgfrb [31,57]. Furthermore, similar to zebrafish, murine perivascular fibroblast-like cells are loosely adhered to the blood vessels, abluminal to mural cells and the basement membrane [31]. Combined with our work, these studies suggest that perivascular fibroblasts represent a unique and evolutionarily conserved population of blood vessel associated cells across vertebrates.

The sclerotome is the embryonic origin of perivascular fibroblasts and pericytes
The sclerotome is the embryonic compartment that contributes to the axial skeleton of vertebrates. Our previous and current studies demonstrate that the sclerotome also gives rise to several populations of tissue support cells, including tenocytes (tendon fibroblasts) [44] and perivascular fibroblasts. The zebrafish sclerotome has a unique bipartite organization with a larger ventral domain and a smaller dorsal domain. Using time-lapse imaging and cell tracing, we have built a spatial map of origin for perivascular fibroblasts. Similar to tenocytes, different sclerotome domains contribute to perivascular fibroblasts in a stereotypic spatial pattern, with the ventral sclerotome domain being the major contributor (84%). The ventral sclerotome domain generates perivascular fibroblasts along the entire length of the ISV, whereas the dorsal sclerotome domain contributes only locally to perivascular fibroblasts associated with the dorsal half of the ISV.
Since perivascular fibroblasts also function as pericyte precursors (see below), our results suggest that most pericytes associated with ISVs also originate from the sclerotome. Consistent with our results, vSMCs around the dorsal aorta have been traced back to cells derived from the sclerotome in zebrafish [49]. In addition, mural cells in the zebrafish trunk vessels have been shown to be dependent on the mesoderm but not the neural crest lineage [58]. Thus, blood vessel support cells in the zebrafish trunk, including perivascular fibroblasts and mural cells, likely originate from the sclerotome. Combined with previous lineage analysis in mouse and chick [59,60], these data suggest a model where mural cells in the brain are neural crest derived, whereas mural cells in the trunk are sclerotome derived.

Perivascular fibroblasts function as pericyte progenitors
Although perivascular fibroblast-like cells have been identified in the mouse brain by single cell RNA sequencing, their biological functions are still unclear. In our study, in vivo time-lapse imaging reveals that the majority of newly born pericytes on ISVs at 3 dpf can be traced back to perivascular fibroblasts at 2 dpf. In a complementary approach, we show that genetic ablation of perivascular fibroblasts results in a general reduction in the number of pericytes on both arterial and venous ISVs. Together, these results indicate that some perivascular fibroblasts serve as pericyte precursors, contributing to most, if not all, pericytes associated with ISVs in the trunk. Incomplete labeling or ablation likely reflects the mosaic nature of the col1a2 transgenic lines [50].
It is interesting to note that there are about 10 perivascular fibroblasts associated with each ISV, however, only about 8% of these cells differentiate into pericytes. It thus raises the question of how these perivascular fibroblasts are 'selected' as pericyte progenitors. There are two possibilities. In the first scenario, all perivascular fibroblasts have the equal potential to differentiate into pericytes, but extrinsic cues determine which cell to switch on the pericyte fate. Notch signaling has been implicated in the regulation of pericyte formation in both mouse and zebrafish [58,[61][62][63][64]. In particular, recent work in zebrafish has shown that the activation of Notch signaling in naïve mesenchymal cells (likely corresponding to perivascular fibroblasts in our study) is crucial to induce pericyte differentiation around ISVs [58]. Therefore, active Notch signaling might bias some perivascular fibroblasts towards the pericyte lineage. Alternatively, perivascular fibroblasts might represent a heterogenous population of cells, containing a small fraction of pericyte progenitors. Supporting this idea, recent scRNA sequencing studies have shown that perivascular fibroblast-like cells in the mouse brain can be further divided into several subtypes based on unique gene expression signatures [30,31].

Perivascular fibroblasts function to stabilize nascent blood vessels
Although mural cells are known to stabilize mature blood vessels, it is not clear how newly formed blood vessels are supported prior to the differentiation of mural cells. Our work provides strong evidence that perivascular fibroblasts function to stabilize nascent blood vessels by depositing collagens in the vascular ECM. Time-lapse imaging shows that the emergence of collagen-expressing perivascular fibroblasts along ISVs occurs concurrently with ISV development, at least one day before the differentiation of the first pericytes. The live Col1a2-GFP reporter reveals that perivascular fibroblasts contribute to a network of collagen fibers around nascent ISVs as early as 38 hpf, which continues to expand until at least 3 dpf. Consistent with this observation, genetic ablation of perivascular fibroblasts from 38 to 62 hpf prior to pericyte differentiation largely prevents the expansion of the collagen network around ISVs, leading to aberrant ISV morphology. These findings suggest that perivascular fibroblasts contribute to the collagen network to stabilize newly formed blood vessels before pericyte differentiation. Consistent with this model, loss of col5a1 results in dysmorphic ISVs with spontaneous hemorrhage in the trunk, which is exacerbated by the additional loss of col1a2 or increasing physical stress. Crucially, we find that these ISV phenotypes can be observed as early as 2 dpf, well before the emergence of first pericytes on ISVs. Consistent with our work, recent study has described a similar population of col22a1 + perivascular fibroblast-like cells associated with the cranial vasculature in zebrafish [65]. Null mutations in col22a1, a minor fibril-associated collagen, result in compromised cranial vessel integrity in both embryos and adults following cardiovascular stress [65], highlighting the importance of collagen and perivascular fibroblasts in vascular stabilization.
Intriguingly, early ablation of perivascular fibroblasts rarely results in the severe hemorrhage phenotype observed in col5a1 -/mutants (under either normal or stressed conditions). Using the live Col1a2-GFP reporter, we find that low level of Col1a2-GFP deposition is already visible around ISVs at 38 hpf prior to MTZ treatment, some of which persists even after ablation at 62 hpf (Fig 5G). It is therefore possible that these residual collagen fibers provide sufficient support to newly formed ISVs and prevent severe hemorrhage seen in collagen mutants. It is also interesting to note that Col1a2-GFP level decreases substantially upon early but not late ablation of perivascular fibroblasts, indicating a higher turnover rate of collagen fibers during early stages of vascular development. This is supported by our observation that late ablation of perivascular fibroblasts from 4 to 5 dpf minimally impacts ISV morphology. Although our work points to perivascular fibroblasts as the main contributor of the vascular ECM, it remains possible that other cell type(s) in the vicinity, such as tenocytes, might contribute to collagen deposition around the ISVs. Development of specific genetic driver lines will allow us to directly test this possibility in the future.

Zebrafish model of Ehlers-Danlos syndrome
Collagen is the most abundant protein in the human body and is a major component of the ECM [66]. Fibrillar collagens, including collagen I and collagen V, constitute the most common subgroup within the collagen family which often co-localize in tissues such as the dermis, tendons and ligaments [66]. Mutations in fibrillar collagens or collagen modification enzymes have been implicated in a number of connective tissue diseases, including Ehlers-Danlos syndrome (EDS). The classical subtype of EDS, caused by mutations in collagen V, and less frequently collagen I, is characterized by skin hyperextensibility, joint hypermobility, and blood vessel fragility [19,67]. Classical EDS patients experience easy bruising and excessive hemorrhaging under normal exertion. By contrast, vascular phenotypes are more prominent in the vascular subtype of EDS, resulting from mutations in collagen III, with most patients experiencing severe spontaneous rupture of large arteries [19]. Interestingly, the zebrafish genome does not contain a collagen III gene, which is likely lost during evolution as in many other teleost fish species [68]. Characterization of zebrafish col1a2 and col5a1 mutants reveals several key features of vascular EDS. Embryos lacking functional col5a1 genes develop spontaneous hemorrhages in the trunk under normal physiological conditions, reminiscent of easy bruising in patients with vascular EDS. One hallmark of vascular EDS is that physical stress often leads to more severe symptoms [69]. Similarly, both the penetrance and severity of the hemorrhage phenotype in col5a1 -/mutants are strongly enhanced by the increase of physical stress. Thus, our results suggest that the lack of collagen III in zebrafish is likely compensated by other fibrillar collagens (types I and V). Our col1a2 and col5a1 mutants represent the first zebrafish model of vascular EDS that we know of.
Analysis of collagen mutants also provides new insights on vascular EDS. First, Col5a1 is a more critical component of the vascular ECM compared to Col1a2. Lack of col5a1 function leads to ISV hemorrhages, whereas loss of col1a2 alone does not result in any obvious vascular defects even with increased physical stress. Intriguingly, collagen I is the most abundant component in collagen fibrils (> 90%), while collagen V constitutes only a minor fraction (< 5%) [70]. Why do col5a1 -/mutants show more severe phenotypes than col1a2 -/mutants? Previous work has shown that Col5a1 plays an important role in proper nucleation and diameter regulation of collagen fibrils [71]. The loss of Col5a1 might result in defects in the assembly of collagen I into fibrils, leading to a more severe phenotype. Second, our mutant analysis reveals genetic interactions between col1a2 and col5a1. Loss of wild-type col1a2 alleles in the col5a1 -/background substantially enhances both the penetrance and the severity of the hemorrhage phenotype in a dosage dependent manner. This result suggests that Col5a1 and Col1a2 function redundantly to maintain vascular integrity. Lastly, as discussed above, our work provides strong evidence that defects in perivascular fibroblasts are the underlying cellular basis for the vascular phenotypes of EDS patients. Perivascular fibroblasts thus represent the cellular targets for potential therapeutic interventions.
In summary, our work identifies perivascular fibroblasts as a novel perivascular cell population with dual function in vascular stabilization: they regulate the ECM of nascent blood vessels and also function as pericyte progenitors. Interestingly, many perivascular fibroblasts remain associated with mature blood vessels even after pericyte differentiation. This raises several questions about the function of these cells at later stages. Do perivascular fibroblasts continue to regulate the ECM around the mature vasculature? Do they function as resident stem cells to replace aging/damaged pericytes? Previous work has suggested that tissue injury in the mouse brain and spinal cord often triggers a fibrotic response by blood vessel associated cells [25,[32][33][34]. Attenuation of the fibrotic response has been shown to limit scar formation and improve axon regeneration post spinal cord injury in mice, suggesting that perivascular stromal cells might be an important therapeutic target [72]. It is plausible that zebrafish perivascular fibroblasts also play an active role in tissue injury repair, an exciting question for future investigations.

Generation of CRISPR mutants
The col1a2 ca108 and col5a1 ca109 mutant lines were generated using the CRISPR/Cas9 system as previously described [53]. Briefly, target sites were identified using the web program CHOP-CHOP (http://chopchop.cbu.uib.no) [76]. sgRNA target sequences are 5'-GGGGGTTCCATTTGATCCAG-3' (col1a2) and 5'-GGCTCCAGCAGATCATCCAG-3' (col5a1). To assemble DNA templates for sgRNA transcription, gene-specific oligonucleotides containing the T7 promoter sequence (5'-TAATACGACTCACTATA-3'), the 20 base target site, and a complementary sequence were annealed to a constant oligonucleotide encoding the reverse-complement of the tracrRNA tail. sgRNAs were generated by in vitro transcription using the Megascript kit (Ambion). Cas9 mRNA was transcribed from linearized pCS2-Cas9 plasmid using the mMachine SP6 kit (Ambion). To generate mutants, one-cell stage wild-type embryos were injected with a mix containing the appropriate sgRNA at 20 ng/μl and Cas9 mRNA at 200 ng/μl. Injected fish were raised to adulthood and crossed to generate F1 embryos. T7 Endonuclease I assay (NEB) was then used to identify the presence of indel mutations in the targeted region of F1 fish. A col1a2 allele containing 1 bp deletion (col1a2 ca108 ) and a col5a1 allele with 4 bp deletion (col5a1 ca109 ) were identified (S6 Fig). For genotyping, small regions around the mutation sites were specifically amplified by PCR, which was followed by allele-specific restriction enzyme analysis. For col1a2 ca108 , the primers used were 5'-TTTTAAAGACTCACATTTGCCTT-3' (forward) and 5'-CTCCGGGCTAGCTTTA-TATTTCGATT-3' (reverse). For col5a1 ca109 , the primers used were 5'-ACTCTTGTTTGCTGTGCAGGT-3' (forward) and 5'-CTCACCGGTATTGGCCGTGTT-3' (reverse). Following PCR, the products were digested with a restriction enzyme which cuts only the wild type or the mutant allele. The enzymes used were BclI and BstCI (NEB) for col1a2 ca108 and col5a1 ca109 , respectively. The resulting band sizes were analyzed using gel electrophoresis. For col1a2 ca108 , the BclI digest resulted in a single band of 473 bp in wild-type embryos, and two bands of 400 bp and 73 bp in homozygous col1a2 -/embryos. For col5a1 ca109 , the BstCI digest resulted in multiple bands at 89, 44, 30, 18 and 9 bp in wild-type embryos, and bands of 107, 44, 30 and 9 bp in homozygous col5a1 -/embryos.

Plasmid injection
To visualize collagen distribution, we generated a UAS:Col1a2-GFP construct based on a previously published pME-Col1a2-GFP plasmid [52]. To generate the stable transgenic line, UAS: Col1a2-GFP (40 ng/μl) was co-injected with tol2 transposase mRNA (40 ng/μl) into col1a2: Gal4 embryos at the one-cell stage with 1 nl per embryo. Injected embryos were raised to adulthood and founders were identified based on GFP expression in F1 embryos. The UAS: Col1a2-GFP line was subsequently propagated in the absence of the Gal4 driver to prevent silencing of the transgene. For more mosaic labeling, UAS:Col1a2-GFP was injected at 10 ng/μl with tol2 transposase mRNA. Injected embryos were screened for GFP expression at appropriate stages for imaging.

Morpholino injection
To inhibit ISV formation, morpholino oligonucleotides (Gene Tools, LLC) targeting the etsrp gene (etsrp MO : 5'-CACTGAGTCCTTATTTCACTATATC-3') [77] were injected at 0.8 mM into one-cell stage embryos with 1.5 nl per embryo. Injected embryos were imaged at appropriate stages for the distribution of sclerotome derived cells.

In situ hybridization and immunohistochemistry
Whole-mount in situ hybridization and antibody staining were performed according to standard protocols. col1a2 and col5a1 antisense probes were used in this study. Double fluorescent in situ hybridization was performed using digoxigenin (DIG) and dinitrophenyl (DNP) labeled probes. For antibody labeling, the rabbit polyclonal antibody to GFP (1:500, MBL) was used. For fluorescent detection of antibody labeling, appropriate Alexa Fluor-conjugated secondary antibodies were used (1:500, Thermo Fisher).

Time-lapse imaging
Imaging was performed using the Olympus FV1200 confocal microscope as previously described [44]. Embryos older than 24 hpf were incubated in fish water with 1-phenyl 2-thiourea (PTU) to prevent pigmentation. Fish were anesthetized in 0.4% tricaine and mounted in 0.6% low melting point agarose. Z-stack images of the region of interest were then collected using the 20x objective at appropriate time intervals (5-15 mins) for up to 24 hours. Movies and images were processed using the Olympus Fluoview software and the Fiji software [78]. Cell tracing was performed manually using Fiji.

Cell ablation
To ablate perivascular fibroblasts, nkx3.1 NTR-mCherry or col1a2 NTR-mCherry embryos at desired developmental stages were treated with water (control group) or 5 mM metronidazole (MTZ, experimental group) for 24-48 hours. In some experiments, mCherrysiblings treated with 5 mM MTZ in the same time windows were used as additional controls. Embryos were subsequently washed in fish water and imaged to confirm successful ablation of mCherry + cells.

Mechanical stress assay
To test the effect of physical stress on the vascular phenotype of collagen mutants, we adopted a mechanical overloading assay by raising embryos in a high-viscosity medium (0.6% methylcellulose) [56]. Heterozygotes for single mutants (col1a2 +/or col5a1 +/-) or double mutants (col1a2 +/-; col5a1 +/-) were intercrossed to obtain sibling embryos with different allele combinations. Resulting embryos were incubated in water (control group) or 0.6% methylcellulose (MC, experimental group) at 48 hpf. Fish were screened for hemorrhage in the trunk region approximately every 3 hours from 48 to 80 hpf and subsequently genotyped for col1a2 and col5a1.

Quantification of single cell analysis
All measurements of single cell morphology were taken using the 'line selection' tool in Fiji. Cell length measurements were taken along the long axis of the cell, parallel to the neighboring ISV as shown in Fig 3E. Cell height measurements were taken at the midpoint of the cell body, orthogonal to the length measurement. Aspect ratio was calculated as cell length divided by cell height. Process lengths were measured by tracing along single processes using the 'segmented line' option under the line selection tool.

Quantification of blood vessel diameter variability
To quantify blood vessel diameters, we used the endothelial specific kdrl:EGFP reporter to fluorescently label intersegmental vessels. Embryos were imaged laterally in the trunk region (somite 12 to 18) and 8-10 ISVs were quantified per embryo. Using the 'line selection' tool in Fiji, four diameter measurements were taken for each ISV along its length and measurements were averaged to obtain mean vessel diameter. Standard deviation from the mean vessel diameter was plotted as a readout of ISV diameter variability. For dorsal aorta (DA) and posterior cardinal vein (PCV) measurements, diameter measurements were taken at 6-10 equidistant points along their length in the mid-trunk region with one measurement per somite. Mean vessel diameter and standard deviation from the mean were similarly graphed as ISV measurements.

Quantification of fluorescence intensity
To compare pdgfrb:GFP expression between pericytes and perivascular fibroblasts, GFP intensity measurements were taken using the 'oval selection' tool in Fiji. Circular regions of interest (ROIs) of 5.56 μm 2 area were drawn within individual cells and 'mean gray value' was measured. Five pericyte and perivascular fibroblast pairs were measured per embryo. For Col1a2-GFP measurements, a rectangular area of 500 by 160 pixels (393 μm x 126 μm) was drawn using the 'rectangular selection' tool in Fiji. The mean fluorescence intensity of the same region was measured before and after the water/MTZ treatment (GFP before and GFP after , respectively). The percentage change in Col1a2-GFP deposition was calculated using the following formula: (GFP after -GFP before ) / GFP before x 100%.