α5 and αv integrins cooperate to regulate vascular smooth muscle and neural crest functions in vivo

Vascular smooth muscle cells (vSMCs) are specialised cells found wrapped around arteries, arterioles and large veins. Most vSMCs are derived from the mesoderm (Mikawa and Gourdie, 1996; Wasteson et al., 2008); however, vSMCs surrounding the ascending aorta and aortic arch are derived from the neural crest (Jiang et al., 2000), whereas vSMCs at the aortic root originate from the secondary heart field (Waldo et al., 2005). During development, a major function of vSMCs is to synthesise and assemble large quantities of extracellular matrix (ECM) around newly formed vessels to provide the vasculature with elasticity and structural strength to withstand the pulsatile blood flow from the heart. The ECM deposited around the vasculature has many functions beyond this structural role however (Hynes, 2009). Previous studies have shown that the ECM within the vessel wall regulates the attachment, contractility, motility, proliferation and differentiation of vSMCs (Raines, 2000). Furthermore, it is now clear that the ECM plays an important role in the patterning and development of the vascular system by binding to, and regulating the availability and activity of, numerous growth factors and morphogens.

The interaction of transforming growth factor β (TGF-β) with the ECM in particular appears essential for regulating cardiovascular morphogenesis and function (ten Dijke and Arthur, 2007; Horiguchi et al., 2012). TGF-β is synthesised as an inactive complex bound to a latent associated protein (LAP) and is incorporated into the ECM through covalently binding to latent TGF-β-binding proteins (specifically LTBP-1, -3 and -4; Ltbp1, 3 and 4, respectively – Mouse Genome Informatics) that interact with fibrillin-containing microfibrils within the vessel wall. This interaction not only localises TGF-β to specific sites but also regulates its activation. Mice lacking the long form of Ltbp1 (Ltbp1L) (Todorovic et al., 2007), just like Tgfb KO mice (Molin et al., 2004) and neural crest-specific Tgfbr2 mutants (Choudhary et al., 2009), die around birth from defects in septation of the heart, cardiac outflow tract and abnormal remodelling of the pharyngeal arch arteries (PAAs), due to decreased TGF-β signalling (Todorovic et al., 2007).

Interestingly, a recent study has shown that assembly of the glycoprotein fibronectin, rather than fibrillin 1, is essential for incorporation of Ltbp1 into the ECM by vSMCs (Zilberberg et al., 2012). Fibronectin is one of the first ECM proteins to be expressed around the vasculature and is essential for cardiovascular development (Astrof and Hynes, 2009). Fibronectin-null mice die at embryonic day (E) 9.5 with defects in the formation of the heart, dorsal aortae and yolk sac vasculature (George et al., 1993,, 1997). Assembly of fibronectin occurs predominantly through the binding of the heterodimeric cell surface adhesion receptor integrin α5β1 to the Arg-Gly-Asp (RGD) motif in fibronectin. In its absence, however, αv integrins (αvβ1, αvβ3, αvβ5, αvβ6 and αvβ8), which also recognise the RGD tripeptide, can relocate to focal contacts previously occupied by α5β1 and assemble fibronectin (Yang and Hynes, 1996; Takahashi et al., 2007; van der Flier et al., 2010). The fibronectin fibrils produced by αv integrins, however, appear short and thick, rather than the long, thin dense fibrillar network produced by cells expressing α5β1 (Takahashi et al., 2007; van der Flier et al., 2010).

α5 and αv integrins play key roles in the development of the vasculature (Hynes, 2007). Integrin-α5 KO mice die at E10.5 with severe vascular defects (Yang et al., 1993; Francis et al., 2002; Mittal et al., 2013), whereas genetic ablation of all five αv integrins leads to placental defects and haemorrhaging within the brain (Bader et al., 1998). Endothelium-specific deletion of both α5 and αv subunits, however, fails to replicate the angiogenic defects observed in global KO-mice, but instead leads to defects in the remodelling of the heart and great vessels (van der Flier et al., 2010). Interestingly, despite the requirement for both α5 and αv integrins for assembly of fibronectin in vitro, fibronectin fibrils are clearly present within the basement membrane of endothelium-specific α5/αv mutants (van der Flier et al., 2010), suggesting that fibronectin is assembled by α5 and αv integrins expressed on vSMCs, or by other fibronectin-binding integrins present on the endothelium, such as α4β1 and α9β1.

In vitro, both α5 and αv integrins are essential for regulating the functions of vSMCs (Moiseeva, 2001). Integrin α5β1 is highly expressed on vSMCs and has been shown to promote proliferation, migration and switching of vSMCs from a more ‘contractile’ to a ‘synthetic’ phenotype (Hedin and Thyberg, 1987; Barillari et al., 2001; Davenpeck et al., 2001; Rensen et al., 2007). Similarly, αv integrins (in particular αvβ3) have been implicated in controlling vSMC migration, de-differentiation, contractility, proliferation and apoptosis (Liaw et al., 1995; D'Angelo et al., 1997; Panda et al., 1997; Dahm and Bowers, 1998). However, the role of α5 and αv integrins on vSMCs in vivo remains unclear. Deletion of α5 using Pdgfrb-Cre, which targets both pericytes and vSMCs (Foo et al., 2006), results in embryonic lethality due to lymphatic, rather than vSMC defects (Turner et al., 2014), whereas arteries in the retinas of mice completely deficient in αvβ3 display only minor delays in recruitment and attachment of vSMCs (Scheppke et al., 2012). Therefore, to gain further insight into the functions of both α5 and αv integrins in vivo, we have generated mice that lack Itga5 and Itgav specifically within their vSMCs.

To investigate the role of α5 and αv integrins on vSMCs in vivo, we crossed female double-homozygous Itga5/Itgav-floxed mice to male double-heterozygous Itga5/Itgav KO mice carrying the SM22α-Cre recombinase. SM22α-Cre is expressed within differentiated vSMCs throughout the vasculature (supplementary material Fig. S1) (Holtwick et al., 2002; Yang et al., 2010). This generated four informative genotypes: (1) control mice, (2) α5 integrin-deficient mutants (hereafter referred to as Itga5^SM22-Cre), (3) αv integrin-deficient mutants (hereafter Itgav^SM22-Cre) and (4) α5/αv conditional double-KO mutants (hereafter Itga5/av^SM22-Cre).

Examination of offspring revealed that both Itga5^SM22-Cre and Itgav^SM22-Cre mutants survive embryonic development, but display high rates (∼50%) of postnatal lethality (supplementary material Fig. S2). SM22α-Cre-mediated deletion of both α5 and αv integrins, however, resulted in embryonic lethality from E16.5, and no living Itga5/av^SM22-Cre mice were ever obtained after birth (supplementary material Fig. S2). Analysis of late gestation mice revealed development of extensive vasculature and no obvious haemorrhaging or oedema in any of the mutant embryos (supplementary material Fig. S3); however, a small number of Itga5/av^SM22-Cre mice did appear slightly growth retarded and occasionally pale in colour.

To determine the cause of lethality in Itga5/av^SM22-Cre mice, mutant embryos were isolated at E17.5 and imaged using a micro-CT scanner. This revealed that Itga5/av^SM22-Cre embryos develop severe cardiovascular defects (Fig. 1). In contrast to control, Itga5^SM22-Cre and Itgav^SM22-Cre embryos, in which the aorta arises from the left ventricle and arches over the heart and descends posteriorly as the descending aorta (Fig. 1A), in Itga5/av^SM22-Cre mice, the ascending aorta misses its connection to the descending aorta (Fig. 1B,C). As the missing portion of the aortic arch is between the left carotid and subclavian arteries, this is equivalent to the clinically defined type-B interrupted aortic arch. Micro-CT scans also revealed that E17.5 Itga5/av^SM22-Cre embryos displayed a large aneurysm at the brachiocephalic artery and the proximal region of the right carotid artery (Fig. 1B). Interestingly, interrupted aortic arch type-B is also seen in several TGF-β signalling mutants (Molin et al., 2004; Todorovic et al., 2007; Choudhary et al., 2009). Furthermore, neural crest-specific Tgfbr2 mutants (herein referred to as Tgfbr2^Wnt1-Cre) display an interrupted aortic arch and an aneurysm in the brachiocephalic region just as observed in Itga5/av^SM22-Cre embryos (Choudhary et al., 2009).

In addition to abnormal remodelling of the PAAs, Itga5/av^SM22-Cre embryos also displayed defects in septation of the outflow tract (conotruncus) and ventricles of the heart (Fig. 1D-G). In wild-type mice, cardiac septation is usually complete between E13.5 and E14.5 (Savolainen et al., 2009) (Fig. 1D). However, by E17.5, Itga5/av^SM22-Cre mutants had failed to separate the most proximal region of the outflow tract into the aorta and pulmonary artery (Fig. 1E), a defect known as persistent truncus arteriosus (PTA), and were missing the rostral proportion of the ventricular septum (Fig. 1G). Surprisingly, despite these severe defects in development of the heart and aortic arches, no obvious vascular defects were observed in other tissues analysed. Whole-mount immunofluorescence staining of blood vessels within the skin revealed that α5/αv-deficient mesodermal vSMCs appeared indistinguishable from control cells. vSMCs surrounding dermal arteries in Itga5/av^SM22-Cre embryos appeared aligned, tightly attached and expressed high levels of the contractile protein α-smooth muscle actin (αSMA) (Fig. 1H).

During early embryonic development (at ∼E9.5 in mice), cardiac neural crest cells (CNCCs) from the dorsal neural tube migrate along the third, fourth and sixth PAAs and invade the cardiac outflow tract of the heart. Here, they proliferate, condense and form the aorticopulmonary septum, which divides the single-tubed vessel into the ascending aorta and pulmonary trunk (Kirby et al., 1983; Kirby and Waldo, 1995; Jiang et al., 2000). In addition, CNCCs covering the PAAs have a separate role, as they differentiate into vSMCs and help co-ordinate patterning of the aortic arch arteries. The physiological importance of CNCCs in development of the cardiovascular system can be seen in ablation studies carried out in chicken embryos. Loss of CNCCs leads to abnormal remodelling of the aortic arches, defects in the septation of the outflow tract and disrupted formation of the thymus (Kirby and Waldo, 1995).

As vascular defects were observed in Itga5/av^SM22-Cre embryos only in areas known to be dependent on CNCCs, we crossed female double-homozygous Itga5/Itgav-floxed mice to male double-heterozygous Itga5/Itgav KO mice carrying the Wnt1-Cre recombinase, to test whether the phenotype of Itga5/av^SM22-Cre embryos was in fact due to defects in neural crest-derived cells. As expected, Wnt1-Cre was expressed throughout the neural crest (supplementary material Fig. S4A) and in vSMCs surrounding the ascending aorta, aortic arch and carotid arteries (Fig. 2A).

Consistent with an essential role for α5 and αv integrins in neural crest function (Mittal et al., 2010), most Itga5^Wnt1-Cre, Itgav^Wnt1-Cre and Itga5/av^Wnt1-Cre mutants died at around birth (supplementary material Fig. S5). However, only Itga5/av^Wnt1-Cre mutants displayed cardiovascular defects (Fig. 2B-G). Just as observed in Itga5/av^SM22-Cre mutants, Itga5/av^Wnt1-Cre embryos developed PTA (Fig. 2B-E) and ventricular septal defects (Fig. 2F,G), confirming that the phenotype of Itga5/av^SM22-Cre is largely due to loss of α5 and αv integrins on CNCCs. Genetic deletion of both α5 and αv integrins did not appear to affect neural crest cell distribution in early Itga5/av^Wnt1-Cre mutants however. Analysis of whole-mount X-Gal-stained Itga5/av^Wnt1-Cre embryos containing the Rosa26 lacZ reporter revealed an overtly normal pattern of neural crest cells in mutant mice (supplementary material Fig. S4A-D). Moreover, CNCCs were clearly present in the outflow tracts of E11.5 Itga5/Itgav^Wnt1-Cre mutants (supplementary material Fig. S4B,D), indicating that loss of both α5 and αv integrins does not compromise CNCC migration. However, in contrast to Itga5/av^SM22-Cre embryos, Itga5/av^Wnt1-Cre embryos rarely developed defects in remodelling of the aortic arch (1/6) and never developed the large aneurysms at the brachiocephalic artery (Fig. 2C), suggesting that the remodelling of the PAAs is dependent on both mesodermal and neural crest-derived vSMCs.

In addition to cardiovascular defects, Itga5/av^Wnt1-Cre mice also displayed defects in the positioning of their thymus. At E17.5, the two lobes of the thymus should be positioned above the heart (Fig. 2H). In Itga5/av^Wnt1-Cre embryos, however, at least one lobe of the thymus was often ectopically located in the cervical region (Fig. 2I). Once again, mirroring the defects in Tgfbr2^Wnt1-Cre mutants (Ito et al., 2003), Itga5/Itgav^Wnt1-Cre mice also developed a cleft palate (Fig. 2J,K). Fusion of palatal shelves is normally complete by E14.5 in control mice. Palatal shelves in Itga5/Itgav^Wnt1-Cre mice, however, remained small and failed to fuse at the midline by E17.5 (Fig. 2K).

A crucial step in the formation of the great arteries is the differentiation of CNCCs into highly contractile vSMCs. Neural crest-specific deletion of Tgfbr2 (Wurdak et al., 2005), Smad2 (Xie et al., 2013), Notch (High et al., 2007) or the myocardin-related transcription factor B (Li et al., 2005) all lead to defects in vSMC differentiation and abnormal development of the outflow tract and aortic arch arteries. Failure to differentiate CNCCs into vSMCs was not the cause of the defects in Itga5/av^SM22-Cre embryos however. Expression of the early vSMC markers αSMA and smooth muscle myosin heavy chain 11 (Myh11) were clearly visible in CNCC-derived cells surrounding the brachiocephalic artery (supplementary material Fig. S6A) and PAAs (data not shown) at E11.5, and continued to be expressed around the great vessels until lethality at E17.5 (supplementary material Fig. S6B-D).

Although the differentiation of vSMCs appeared unaffected by the loss of α5 and αv integrins, immunofluorescence staining of E17.5 embryos with anti-αSMA antibodies did reveal striking abnormalities in the structure of the ascending aorta, carotid and brachial arteries in Itga5/av^SM22-Cre mice. In control embryos, vSMCs appeared long, compact and were organised into 3-4 concentric lamellar units within the vessel wall (Fig. 3A,B). By contrast, vSMCs in the ascending aorta (data not shown) and right carotid/brachiocephalic artery of mutant embryos appeared round, disorganised and formed up to 18 layers of cells (Fig. 3C,D). Furthermore, in the most severely affected mutants, Pecam1-positive endothelial cells were undetectable within the aneurysm (Fig. 3C). Large regions of the great vessels appeared unaffected by the loss of both α5 and αv integrins however. vSMCs around the left carotid artery of Itga5/av^SM22-Cre embryos, for example, appeared well organised despite often displaying an increased number of lamellar units (supplementary material Fig. S6B). In addition, no obvious defects in vSMC proliferation could be detected around the brachiocephalic or carotid arteries (supplementary material Fig. S7).

Previous studies have shown that the assembly of ECM proteins within the vessel wall is essential for maintaining the structural integrity of the great vessels. We therefore examined whether the dilated brachiocephalic/carotid artery in Itga5/av^SM22-Cre embryos was due to abnormal ECM deposition.

Surprisingly, despite the requirement for both α5 and αv integrins for fibronectin fibrillogenesis in vitro, fibronectin fibrils were clearly visible in the vessel walls of both early (supplementary material Fig. S8) and late (Fig. 4A) gestation Itga5/av^SM22-Cre embryos. In contrast to control mice, these fibrils displayed a disorganised lamellar organisation at E17.5 (Fig. 4A). The assembly of microfibril ECM proteins was also disrupted in Itga5/av^SM22-Cre embryos. In the dilated carotid arteries of mutant mice, fibrillin 1-containing microfibrils were almost undetectable in the tunica media, and organisation of fibulin 5 into lamellar structures appeared impaired (Fig. 4B,C). As observed in most human aneurysms, Itga5/av^SM22-Cre embryos also displayed elastin fragmentation within their aneurysm (Fig. 4D). Furthermore, collagen IV fibrils, which are also found in the medial layers, were largely absent from the aneurysmal region of mutant mice (Fig. 4E). Thus, the disorganised pattern of vSMCs is accompanied by defects in organisation of several ECM proteins.

To gain further insight into the molecular mechanisms underlying the Itga5/av^SM22-Cre phenotype, vSMCs from the aortae of adult Itga5/av^flox/flox mice were isolated and immortalised with the SV40 large T antigen. vSMC identity was confirmed by immunofluorescence staining for αSMA and smoothelin (Fig. 5A). Integrin α5/αv-floxed cells were then infected with either an empty vector or Cre-expressing adenovirus to generate control and integrin α5/αv-deficient vSMCs (ΔItga5/av), respectively. Efficient excision of the floxed alleles was confirmed by PCR (data not shown) and loss of both α5 and αv proteins verified by immunoblotting (Fig. 5B). Consistent with the binding affinities of α5 and αv, ΔItga5/av cells failed to attach to either fibronectin or vitronectin (Fig. 5C) and showed reduced adhesion to both laminin and collagen I substrates (supplementary material Fig. S9A), despite expressing similar levels of their cognate integrin receptors (supplementary material Fig. S9B). ΔItga5/av cells, however, adhered efficiently to Matrigel (Fig. 5C), but appeared smaller, more rounded and had fewer protrusions than control cells 24 h after plating (Fig. 6A-C).

Visualisation of the contacts between the cells and the ECM revealed that loss of both α5 and αv had dramatic effects on the formation of mature adhesions. In control vSMCs, focal adhesions are visible in cell protrusions and are present as fibrillar adhesions in the cell body (Fig. 6A,D). By contrast, ΔItga5/av cells contained only nascent adhesions and focal complexes throughout the cell (Fig. 6B,E). As a result, ΔItga5/av cells show reduced activation of the focal adhesion kinase (FAK/PTK2), markedly reduced levels of paxillin phosphorylation and reduced phosphorylation of the Crk-associated substrate p130 (CAS) (Fig. 6F; quantifications in supplementary material Fig. S9C). These deficits were rescued by re-expression of either ITGA5 or ITGAV within ΔItga5/av cells (Fig. 6F), suggesting that either integrin can compensate for loss of the other.

Given the similarity of the cardiovascular defects observed in Itga5/av^SM22-Cre mutants and knockouts of Ltbp1L (Todorovic et al., 2007) or TGFβ2 (Molin et al., 2004), and in mice in which the Tgfbr2 gene was conditionally ablated in vSMC precursors (Choudhary et al., 2009), we examined whether TGF-β signalling was disrupted in our ΔItga5/av cells. Immunoblotting for downstream mediators of the canonical TGF-β signalling pathway revealed that phosphorylation of SMAD2, but not of SMADs 1, 5 and 8 (Fig. 7A), was reduced in ΔItga5/av cells, despite a moderate increase in Tgfb1 expression (supplementary material Fig. S9D). The extent of this reduction in SMAD signalling appeared dependent on the Matrigel batch used, with some experiments producing only small changes in the level of pSMAD2 on specific Matrigel preparations (data not shown). To our surprise, immunofluorescence staining revealed that pSMAD2 levels were increased in the aneurysmal region of Itga5/av^SM22-Cre embryos, when compared with the right carotid artery of control E17.5 mice (Fig. 7B). Interestingly, this increase was also apparent, just before the onset of the large dramatic aneurysm, in the brachiocephalic/carotid region in E12.5 embryos (supplementary material Fig. S10). However, no obvious differences in pSMAD2 signalling were observed in any of the other regions of Itga5/av^SM22-Cre embryos examined (Fig. 7B).

TGF-β is secreted as an inactive form and requires release from LAP to exert its biological functions. Previous studies have shown that integrins regulate TGF-β signalling by interacting with the RGD motif contained in the LAPs of TGF-β1 and TGF-β3, thus releasing TGF-β from its latent complex (Munger et al., 1999; Annes et al., 2002; Yang et al., 2007). To determine whether loss of both α5 and αv prevented binding to LAP, we assessed the ability of control, ΔItgav, ΔItga5 and ΔItga5/av vSMCs to adhere to recombinant human TGF-β1 LAP-coated plates. Consistent with previous data (Munger et al., 1998,, 1999; Ludbrook et al., 2003), loss of integrin αv significantly reduced binding to LAP (Fig. 7C). Adhesion to LAP was also reduced in ΔItga5 vSMCs, albeit to a lesser extent than in ΔItgav cells, whereas binding to LAP was almost completely blocked in cells lacking both α5 and αv integrins (Fig. 7C).

Integrin binding to LAP alone is insufficient for TGF-β activation. The complex of TGF-β and LAP (small latent complex) also requires anchoring to the ECM through its association with Ltbp1 and Ltbp3 (Horiguchi et al., 2012). For Ltbp1, this incorporation is dependent on the assembly of fibronectin, and is independent of fibrillin expression, whereas incorporation of Ltbp3 is dependent on assembly of fibrillin 1 microfibrils (Zilberberg et al., 2012). To examine whether ΔItga5/av vSMCs could incorporate Ltbp1 into the matrix, we first analysed the ability of ΔItga5/av vSMCs to assemble endogenous fibronectin into fibrils. Immunofluorescence staining revealed that control vSMCs form extensive fibronectin fibrils 24 h after plating (Fig. 7D). By contrast, ΔItga5/av vSMCs plated at confluency formed smaller aggregates of fibronectin (Fig. 7D). As fibronectin mRNA expression levels were reduced (30-40%) in cells lacking both α5 and αv (supplementary material Fig. S11A), we analysed the ability of ΔItga5/av vSMCs to incorporate exogenous biotin-labelled fibronectin into the DOC-insoluble matrix (supplementary material Fig. S11B). This revealed that, even after 5 days in culture, ΔItga5/av vSMCs are unable to assemble exogenous fibronectin into fibrils (supplementary material Fig. S11B). Re-expression of either α5 or αv integrin, however, rescued the ability of ΔItga5/av SMCs to assemble and incorporate fibronectin into the DOC-insoluble matrix (Fig. 7E). Similarly, ΔItga5/av vSMCs also had defects in their ability to assemble the RGD-containing microfibrillar proteins fibrillin 1 and fibulin 5 in vitro (Fig. 7E). Surprisingly, in contrast to our in vivo results, lack of fibrillin 1 and fibulin 5 did not affect the deposition of elastin (Fig. 7E). Loss of α5 and αv integrins did affect incorporation of the RGD-containing Ltbp1 into the DOC-insoluble matrix however (Fig. 7F). After 3 days in culture, Ltbp1 was incorporated into the ECM by control, and to a lesser extent ΔItga5 and ΔItgav vSMCs, but was completely absent from the ECM assembled by ΔItga5/av cells (Fig. 7F). Interestingly, incorporation of Ltbp3, which does not contain an RGD motif, appeared unaffected by loss of α5 and αv and the lack of fibronectin and fibrillin 1 fibrils (Fig. 7F). Thus, absence of these two integrin subunits on vSMCs compromises assembly of ECM proteins and components of TGF-β signalling complexes.

In this study, we have shown that α5 and αv integrins on neural crest-derived vSMCs cooperate to control remodelling of the PAAs, and are essential for the septation of the heart and outflow tract. We have also shown that vSMC expression of both α5 and αv subunits are essential for assembly of ECM within the vessel wall, and that loss of both integrins leads to the formation of large aneurysms within the brachiocephalic/carotid arteries. Expression of α5 and αv integrins, however, appears to be dispensable for the initial assembly of an extensive vascular network and the function of vSMCs surrounding vessels in the skin.

Consistent with our previous study (Turner et al., 2014), genetic ablation of α5β1 from vSMCs failed to replicate any of the vascular defects observed in the global Itga5 KO mice (Yang et al., 1993; Francis et al., 2002). Itga5^SM22-Cre mice survived to birth with no obvious vSMC or cardiovascular defects. Similarly, despite numerous studies suggesting that αv integrins play a key role in controlling vSMC function in vitro (Liaw et al., 1995; D'Angelo et al., 1997; Panda et al., 1997; Dahm and Bowers, 1998), Itgav^SM22-Cre mutants developed normally and contained blood vessels indistinguishable from those in control mice. However, both Itga5^SM22-Cre and Itgav^SM22-Cre mice displayed postnatal lethality, suggesting that loss of either integrin increases susceptibility to cardiovascular defects after birth. Although contradictory to numerous blocking studies in vitro, the lack of major vascular defects in Itga5^SM22-Cre and Itgav^SM22-Cre embryos fits with a growing body of data showing that α5 and αv integrins have overlapping functions and that either integrin can compensate for loss of the other (Yang and Hynes, 1996; Takahashi et al., 2007; van der Flier et al., 2010). Indeed, ablation of Itga5 or Itgav alone caused only a minor reduction in focal adhesion signalling and matrix assembly in our cultured vSMCs. It is also possible that some integrin functions, such as assembly of ECM matrix or activation of TGF-β, are compensated for by integrins expressed on adjacent cells within the vasculature. This might help explain why tissue-specific integrin mutants often display phenotypes less severe than those predicted from in vitro studies.

In contrast to the results for single Itga5^SM22-Cre and Itgav^SM22-Cre mutants, deletion of both α5 and αv integrins by SM22α-Cre caused significant cardiovascular defects. Itga5/av^SM22-Cre embryos developed interrupted aortic arch type-B and a large aneurysm that encompassed the brachiocephalic artery and the proximal region of the right carotid artery. The same phenotype has been observed in mice that lack vSMC expression of all the β1 integrins (Turlo et al., 2012). This suggests that the phenotype of Itga5/av^SM22-Cre mice, and therefore the vSMC-specific Itgb1 mutant, is caused by the loss of α5β1 and specifically αvβ1, rather than any of the other αv (αvβ3, αvβ5, αvβ6 and αvβ8) or β1 (α1β1, α2β1, α3β1, α4β1, α6β1, α7β1, α9β1, α10β1, α11β1) integrin heterodimers. Like Turlo et al. (2012), we also found that migration and initiation of vSMC fate were unaffected by loss of α5 and αv integrins. vSMCs in Itga5/av mutants expressed high levels of the vSMC markers αSMA, SM22α and Myh11, and migrated efficiently to the PAAs and blood vessels within the skin. Deletion of both α5 and αv integrins also appeared to have no obvious effect on the functions of mesodermally derived vSMCs during embryonic development. In contrast to mural cell-specific Itgb1 mutant mice (Abraham et al., 2008), in which vSMCs appeared round, poorly spread and only loosely attached to the subendothelial basement membrane, vSMCs in the cutaneous vasculature of Itga5/av^SM22-Cre embryos appeared indistinguishable from those in control mice. One possible explanation for these conflicting results is that vascular defects in mural cell-specific β1 mutants are due to loss of Itgb1 on pericytes, rather than vSMCs. Alternatively, β1 heterodimers, other than α5β1 and αvβ1, might play important roles in mesodermally derived vSMCs; defects in pericyte and vSMC distribution have been reported in α4 and α7-KO mice (Flintoff-Dye et al., 2005; Garmy-Susini et al., 2005; Grazioli et al., 2006). Arguing against this latter hypothesis, however, is the fact that vascular defects also appear to be restricted to areas populated by neural-crest-derived vSMCs in mice in which Itgb1 has been deleted in all vSMCs, using the same SM22α-Cre line as in this study (Turlo et al., 2012). Nevertheless, large aneurysms are found both in β1 mutants (Abraham et al., 2008; Turlo et al., 2012) and in our Itga5/av^SM22-Cre embryos. As in the vSMC-specific β1 mutants (Turlo et al., 2012), aneurysms were present only in the brachiocephalic and carotid arteries. This could be due to these regions exhibiting the greatest shear stress (Meng et al., 2007; Huo et al., 2008; Wang et al., 2009), especially when flow is redirected to the brachiocephalic and right carotid artery when the aortic arch is interrupted. These regions of the vasculature also correlate with areas of maximum integrin β1 expression (Turlo et al., 2012). It is therefore possible that the aorta, brachiocephalic and carotid artery are more sensitive to the loss of Itga5 and Itgav, and might suggest that both α5β1 and αvβ1 integrins play an important role in providing structural strength to the blood vessels and in resisting the high pulsatile flow from the heart. Alternatively, these regions might be more susceptible, due to being populated by different subsets of CNCCs, compared with other parts of the great vessels, and heightened expression of β1 integrins might simply correlate with a specific population of vSMCs. Interestingly, decreased expression of ITGA5 has previously been linked to formation of human aortic aneurysms (Cheuk and Cheng, 2004), and, similar to some human aneurysms (Pera et al., 2010) and experimentally induced aneurysms in mice (Murphy and Hynes, 2014), Pecam1-positive endothelial cells were reduced in the dilated brachiocephalic artery of Itga5/av^SM22-Cre mice at late stages. It is unclear, however, whether this loss is specifically linked to deletion of both α5 and αv within vSMCs, or merely part of the general pathology occurring within a large aneurysm. It is conceivable that defects in vSMC function (assembly of the basement membrane ECM, signalling, contraction) are directly, or indeed indirectly, causing this dedifferentiation. After all, mechanosensing by Pecam1 is directly linked to integrin engagement to the ECM (Collins et al., 2012), Pecam1 is a ligand for αvβ3 (Piali et al., 1995) and is regulated by TGF-β (Neubauer et al., 2008).

As SM22α-Cre is also expressed in the heart (Turlo et al., 2012), we cannot definitively rule out the possibility that the Itga5/av^SM22-Cre phenotype is secondary to heart defects. Development of the aortic arch has been shown to be linked to the haemodynamics of blood flowing from the heart (Yashiro et al., 2007). However, as septation and PTA defects were also found in Itga5/av^Wnt1-Cre embryos, which lacked aneurysms and aortic arch defects, and because Itga5/av^SM22-Cre mice developed dilated brachiocephalic/carotid arteries before cardiac septation at E12.5, we think that this is unlikely. We believe instead, as both neural crest and mesenchyme-derived cells contribute to the aorta and PAAs (Bergwerff et al., 1998), that the vascular defects are less severe in Itga5/av^Wnt1-Cre embryos, as α5/αv-containing mesenchyme-derived cells can compensate, at least in part, for the loss of both integrins in the neural-crest-derived vSMCs.

In our attempts to further understand the cardiovascular defects in our mutants, we have also investigated the ways in which α5 and αv integrins cooperate to regulate neural crest function. Almost all Itga5^Wnt1-Cre embryos died shortly after birth, whereas Itgav^Wnt1-Cre and, to a greater extent, Itga5/av^Wnt1-Cre mutants, displayed perinatal lethality. The loss of Itga5/av^Wnt1-Cre mice is considerably later than the lethality seen in neural crest-specific Itgb1 mutants (Turlo et al., 2012), confirming the importance of other β1-containing receptors in neural crest cells. Both cardiac and cranial neural crest defects were observed in Itga5/av^Wnt1-Cre mutants. In addition to developing a PTA and ventricular septation defects (VSDs), Itga5/av^Wnt1-Cre mice also displayed mislocated thymi and cleft palates. Previous studies have shown that migration of neural crest cells is dependent on α5 and αv expression, and defects in neural crest migration can lead to heart, thymus and craniofacial defects (Delannet et al., 1994; Alfandari et al., 2003; Keyte and Hutson, 2012). Migration defects do not appear to be the cause, at least of the cardiovascular defects, in Itga5/Itgav^Wnt1-Cre mutants. Although we cannot rule out relatively subtle defects in the migration of specific subsets of neural crest cells, Wnt1-positive CNCCs were clearly present in the PAA and outflow tracts in Itga5/Itgav^Wnt1-Cre embryos as early as E11.5. These results are consistent with a number of studies showing that septation and PAA defects can occur independently from neural crest migration and differentiation defects (Molin et al., 2004; Turlo et al., 2012).

So how do α5 and αv integrins regulate cardiovascular development? Our analysis of ΔItga5/av vSMCs in vitro suggests a number of possibilities. First, our data show that loss of the α5 and αv integrins prevents the formation of mature focal adhesions and, as a consequence, leads to reduced levels of FAK and paxillin phosphorylation. Focal adhesions are essential for maintaining structural integrity of vessels by inducing cytoskeletal rearrangements and alignment of vSMCs in response to mechanical strain. Previous studies have shown that genetic deletion of FAK (Ptk2) in vSMC precursors leads to defects in the patterning of the aortic arch arteries and septation of the heart and outflow tract (Hakim et al., 2007; Vallejo-Illarramendi et al., 2009; Cheng et al., 2011). These defects, however, are caused by abnormal recruitment (Hakim et al., 2007; Cheng et al., 2011) and/or impaired differentiation of vSMCs (Vallejo-Illarramendi et al., 2009), neither of which was seen in our mutants. Less information exists about the role of paxillin in vSMCs. In vitro, paxillin has been implicated in regulating adhesion, proliferation, apoptosis and activation of L-type calcium channels in vSMCs (Wu et al., 2001; Veith et al., 2012). To date, no one has investigated the role of paxillin specifically in vSMCs in vivo. Intriguingly, however, paxillin KO mice die at E9.5 with cardiac and somitic defects resembling those in fibronectin KO mice (Hagel et al., 2002).

The cardiovascular defects in Itga5/av^SM22-Cre mice might also be due to the inability of ΔItga5/av vSMCs to interact correctly with the ECM. ΔItga5/av vSMCs failed to attach to either fibronectin or vitronectin and displayed reduced adhesion to laminin and collagen I. As a result, ΔItga5/av vSMCs often appeared poorly spread in vitro and mirrored the round, disorganised morphology of vSMCs within the ascending aorta, brachiocephalic and right carotid artery of Itga5/av^SM22-Cre embryos. Our data also indicate that α5 and αv integrins are required for the assembly and organisation of the ECM within the vessel wall. In vitro, consistent with α5 and αv being the predominant RGD-binding receptors expressed on vSMCs, ΔItga5/av vSMCs were unable to incorporate fibronectin, fibrillin 1 or fibulin 5 into the DOC-insoluble ECM. Furthermore, organisation of the ECM into concentric lamellar layers around the ascending aorta, brachiocephalic and carotid artery appeared severely compromised in Itga5/av^SM22-Cre mice. Correct assembly of the ECM is essential for maintaining vascular integrity; vSMCs alone are insufficient to resist the mechanical strain generated by pulsatile blood flow (Wagenseil and Mecham, 2009). Defects in assembly of fibrillin, fibulin, collagen and elastin have all been implicated in the formation of aneurysms in vivo (El-Hamamsy and Yacoub, 2009). It is therefore probable that the abnormal assembly of the ECM weakens the vessel wall and directly causes the aneurysms in Itga5/av^SM22-Cre embryos.

A final intriguing possibility is that the defects in Itga5/av^SM22-Cre mice are in fact related to abnormal TGF-β signalling. Previous studies have shown that TGF-β signalling is essential for the patterning of the aortic arch, septation of the heart and outflow tract, and for fusion of the palatal shelves and development of the thymus (Ito et al., 2003; Molin et al., 2004; Wang et al., 2006; Todorovic et al., 2007; Choudhary et al., 2009), defects all present in our Itga5/Itgav mutants. Furthermore, ΔItga5/av vSMCs were unable to bind to LAP or to deposit Ltbp1 into the ECM, and had decreased levels of pSMAD2. Paradoxically, in vivo, Itga5/av^SM22-Cre mice displayed increased levels of pSMAD2 within their aneurysms. This ‘TGF-β paradox’ has also been seen in aortic aneurysms present in patients with Loeys–Dietz and Marfan's syndromes, which, in theory, should also exhibit reduced pSMAD2 levels within their vasculature (Lin and Yang, 2010). The exact mechanism for this paradox remains unclear. However, it might be due to excessive (compensatory) upregulation of TGF-β, increased metalloproteinase expression, dysregulation of TGF-β signalling-feedback loops, or even through non-TGF-β activators of SMADs (Lin and Yang, 2010). There is also the possibility that the defects in Itga5/av^SM22-Cre mice are actually due to reduced non-SMAD TGF-β signalling (Moustakas and Heldin, 2005). As there is extensive cross-talk between TGF-β signalling pathways, integrins, focal adhesions and the ECM, it is extremely difficult to unravel the exact chain of causality leading to the defects seen in Itga5/av^SM22-Cre mice or those in TGF-β signalling.

Deletion of both α5 and αv integrins from either vSMCs or endothelial cells results in defects in the remodelling of the great vessels and the heart, but, in both cases, fails to disrupt initial blood vessel development (van der Flier et al., 2010; Turner et al., 2014), or replicate the phenotypes of the global KO mice. These observations suggest that integrins on both mural cells (pericytes, vSMCs) and endothelial cells cooperate to regulate the assembly of ECM proteins and signalling complexes such as those for TGF-β within the ECM. It is, of course, possible that some other cell types expressing these integrins play some essential role, although this seems to us less likely.

Congenital heart defects (CHD) are a leading cause of miscarriage and the most common type of birth defect (Bruneau, 2008). Furthermore, as surgical intervention has advanced, and more children with CHD survive into adulthood, there is an even greater need to understand the molecular pathways that regulate cardiovascular development. Our study has shown that α5 and αv integrins are essential for development of the heart and great vessels. However, many questions about their precise roles in vascular development and homeostasis remain unresolved. Future experiments will need to examine the role of α5 and αv integrins in the adult vasculature, and assess whether loss of either integrin subunit increases susceptibility to aortic aneurysms and other vascular diseases.

All mouse strains were on 129S4:C57BL/6 mixed background. Itga5 floxed (van der Flier et al., 2010), Itgav floxed (Lacy-Hulbert et al., 2007), mTmG (Muzumdar et al., 2007), Rosa lacZ (Soriano, 1999), SM22α-Cre (Holtwick et al., 2002) and Wnt1-Cre (Danielian et al., 1998) mouse lines have all been described previously. Genotyping was performed on DNA isolated from tail snips in house or by Transnetyx. All mice were housed and handled in accordance with approved Massachusetts Institute of Technology Division of Comparative Medicine protocols (IACUC approval 0412-033-15).

4% PFA-fixed embryos were sent to Numira Biosciences for micro-CT imaging. Specimens were stained with a proprietary contrast agent. A high-resolution volumetric micro-CT scanner (µCT40; Scanco Medical) was used to scan the tissue (6 µm isometric voxel resolution, 200 ms exposure, 2000 views and 5 frames per view). The micro-CT generated DICOM files, which were analysed using OsiriX and Volocity software.

vSMCs were isolated from the aortic arch and carotid arteries of Itga5^flox/flox; Itgav^flox/flox mice as per Ray et al. (2001) (see supplementary materials and methods). Upon confirmation of vSMC identity by αSMA- and smoothelin-positive immunofluorescence staining, cells were immortalised using SV40 large T antigen (Zhao et al., 2003) and infected with either empty vector or Cre-expressing adenovirus to generate control or α5/αv-deficient vSMCs (ΔItga5/av), respectively. Following excision, both control and ΔItga5/av cells were grown on plates coated with 20 µg/ml Matrigel (BD Bioscience). To generate vSMCs lacking either α5 or αv integrins, ΔItga5/av cells were infected with retroviral constructs containing ITGAV or ITGA5 (van der Flier et al., 2010) to generate ΔItga5 or ΔItgav cells, respectively.

96-well plates were coated overnight at 4°C with serial dilutions of fibronectin, collagen I, laminin, vitronectin, Matrigel or recombinant human LAP (TGF-β1) (R&D Systems). After blocking with 5% BSA, 20,000 cells/well were plated and allowed to adhere for 24 h. Unattached cells were removed by aspiration, and adherent cells were fixed with 4% PFA and stained with 0.1% Crystal Violet. To quantify adhesion, cells were permeabilised in 50 µl of PBS/0.2% Triton X-100, and OD₅₉₀ was measured in a plate reader.

Cells were seeded at low density and allowed to adhere to Matrigel-coated glass coverslips for 24 h. The cells were then fixed and stained with phalloidin and DAPI. Following imaging of six fields from each coverslip, images were automatically analysed using Volocity software to detect the outline of the cells. Cells touching other cells or the border of the image were rejected from the analysis, and the shape factor was calculated using the formula (4×π×area/perimeter²). A circular cell shape will give a factor of 1, whereas a complex cell shape will give values closer to 0.

vSMCs were seeded at confluence (200,000 cells/ml) on Matrigel-coated plates in fibronectin-depleted medium in the presence or absence of exogenous biotinylated human fibronectin (10 µg/ml). At 1, 3 and 5 days post plating, medium was collected, and cells were lysed in 0.5 ml DOC buffer [2 mM EDTA, 1% sodium deoxycholate, 20 mM Tris pH 8.5, complete mini-protease inhibitors (Roche)] for fibronectin experiments or in 0.25% DOC for investigation of Ltbp1 incorporation into the insoluble matrix fraction. After passing eight times through a 22 G needle, DOC-insoluble material was spun down for 20 min at 20,000 g at 4°C and solubilized in 120 μl 2× reducing SDS-PAGE loading buffer. Reduced (100 mM DTT) samples were loaded onto 4-12% gels.

See supplementary materials and methods for details of histology, immunofluorescence staining, immunoblotting and RT-qPCR.

We thank members of the Hynes laboratory, especially Patrick Murphy, for discussions and advice.