β1 Integrin regulates convergent extension in mouse notogenesis, ensures notochord integrity and the morphogenesis of vertebrae and intervertebral discs

The notochord is a rod-like midline structure of mesodermal origin that forms during gastrulation. Its formation and extension are among the earliest and most obvious events of axis development during embryogenesis (Jurand, 1974; Rossant and Tam, 2002; Stemple, 2005). In vertebrates, the notochord has two primary functions during embryonic development. Firstly, it serves as the axial skeleton of the embryo until vertebrae form. Secondly, it acts as a signaling center that induces the formation of floor plate in the neural tube (Jessell et al., 1989) and the differentiation of the ventral somatic derivatives into the sclerotome (Pourquie et al., 1993). Unlike in some lower vertebrates, such as lampreys, and some primitive fish, such as sturgeons, for which the notochord persists throughout life (Stemple, 2005; Corallo et al., 2015), the notochord in higher vertebrates is a transient structure that remodels during vertebral body formation leaving remnants in the intervertebral discs (IVDs) as nuclei pulposi (NP) (Stemple, 2005; Choi et al., 2008; McCann et al., 2012).

Time-lapse live imaging and cell lineage tracing of the mouse notochord revealed three distinct morphogenetic origins along the anterior-posterior (A-P) axis (Yamanaka et al., 2007). The anterior notochord arises through condensation of dispersed cells anterior to the forming node; the trunk notochord derives from cells of the ventral layer of the bilaminar node that move via convergent extension; and the tail notochord develops from peripheral node cells (Sulik et al., 1994; Kinder et al., 2001; Ezin et al., 2003; Yamanaka et al., 2007; Ukita et al., 2009). The convergent extension of the notochord is stimulated by mechanical forces exerted by the expanding amniotic cavity as the mouse embryo elongates along the A-P axis (Imuta et al., 2014). These forces were shown to trigger membrane accumulation of VANGL2 and, thereby, the planar cell polarity (PCP) pathway. Whether and how these forces, the accompanying convergent extension and subsequent remodeling of the notochord to form NP are regulated by cell-matrix interactions is not known.

Notochordal cells are surrounded by a sheath of extracellular matrix (ECM), which contains glycoproteins such as fibronectin (FN), collagens and proteoglycans (Gotz et al., 1995; Smits and Lefebvre, 2003). In Sox5 and Sox6 double knockout mice, the notochord lacks the ECM sheath, resulting in impaired cell survival, lateral displacement of notochord cells and absence of the NP (Smits and Lefebvre, 2003). A major component of the notochordal sheath is collagen II, secreted by notochordal cells. In mice lacking collagen II, the notochord is enlarged in diameter, but is not remodeled and remains as a rod-like structure until birth, resulting in the failure to develop an IVD and NP (Aszódi et al., 1998). These defects indicate that the notochordal sheath is key for the reorganization of the notochord and IVD development, and support the hypothesis that notochordal cell-ECM interactions are probably mediating mechanical forces required for notochord reorganization (Aszódi et al., 1998; Lawson and Harfe, 2015).

Integrins are α/β heterodimeric cell adhesion molecules that mediate the interactions of cells with ECM proteins including collagens, FN and proteoglycans (Hynes, 2002). They are bidirectional signaling molecules that are first activated from within a cell, then bind ligands and transduce biochemical and physical properties across the plasma membrane into cells (Sun et al., 2016). Genetic studies in Xenopus, zebrafish and mice implicated FN, laminin (LN) and the α5 integrin in notochord development and elongation along the A-P axis (George et al., 1993; Parsons et al., 2002; Pulina et al., 2014; Trapani et al., 2017). The β1 integrin subunit associates with any of 12 α subunits, and thereby forms the largest subfamily of ECM-binding integrins. However, as the constitutive deletion of the Itgb1 gene encoding β1 integrin leads to peri-implantation lethality (Fassler and Meyer, 1995; Stephens et al., 1995), the role of β1 integrin during notochord development has not yet been elucidated.

To investigate the role of β1 integrin in notochord morphogenesis, we developed a transgenic Cre recombinase deletor strain to inactivate Itgb1 in the mouse notochord. Characterization of these mice demonstrated Cre activity either in the notochord alone or in the notochord and the tail bud. The resulting mutant mice (designated Itgb1^ΔND) developed notochordal and vertebral column defects that resembled those of α5 integrin- and FN-deficient mice. We also report that β1 integrin promotes convergent extension movements and the asymmetric accumulation of the core PCP component VANGL2 in the node pit cells. Our findings assign β1 integrin-mediated cell-matrix adhesion as a key function in notochord morphogenesis.

In order to ablate Itgb1 specifically in the notochord, we developed a notochord-specific Cre (Foxa2mNE-Cre) transgenic mouse, in which bicistronic Cre and lacZ expression is driven by the Sox9 promoter and a minimal notochord enhancer element isolated from the Foxa2 gene (Sasaki and Hogan, 1996) (Fig. S1A). The earliest transgene expression visualized by lacZ staining was at around embryonic day (E) 7.5 in the node, extending into the notochord or in both the notochord and tail bud at later stages (Fig. S1A, black arrows). In crosses of the Foxa2mNE-Cre with an EGFP Cre reporter strain (zEG) (Novak et al., 2000), EGFP expression was activated in the notochord as early as E8.0, albeit in a mosaic pattern (Fig. S1B, white arrows), and in crown cells from the peripheral node, but not in pit cells located in the node center (Fig. S1B, highlighted region).

Next, we analyzed the Cre activity by tracking the GFP signals in ∼400 Foxa2mNE-Cre;zEG embryos obtained from different male breeders. Three types of Cre activities were observed at E9.5 (Fig. S1C). Type I (50%) showed strong activity in the entire notochord and some activity in the tail bud from E9.5, which probably reflects the contribution of node cells to the tail bud (Cambray and Wilson, 2002). The remaining 50% of embryos exhibited two types of notochord-specific Cre activities: one is strong (Type II, 37%) and the other is weak (Type III, 13%) (Fig. S1D). Cre expression ceased around E14.5, and GFP+ notochord descendants were found in the NP at adult stages (Au et al., 2020), which is consistent with published reports demonstrating a notochordal origin of NP cells (Choi et al., 2008; McCann et al., 2012). Thus, we utilized this Foxa2mNE-Cre mouse line to study gene function in cells of the developing notochord.

Next, we crossed mice homozygous for the floxed Itgb1 (β1^F/F) allele with heterozygous Foxa2mNE-Cre transgenic mice. The latter were used as controls in our study. Immunostaining revealed significant reduction of β1 integrin expression at E8.0 (∼1 somite) (Fig. 1A) in the developing notochord marked by brachyury (T). By the ∼23 somite stage (E9.5), β1 integrin expression was absent in mutant cells in notochordal segments displayed as Region-1 (Fig. 1B, Region-1), indicating that β1 integrin protein has a maximal lifetime of ∼1.5 days in the notochordal tissue. Consistent with a combination of mosaic Cre activity and the stability of β1 integrin, β1 integrin expression was still detected at variable levels in notochordal regions relative to the surrounding non-notochordal cells (Fig. 1B, Region-2 and Region-3). Detailed analysis of the trunk and tail notochord from three embryos at E9.5 showed variable mosaicism in mutants ranging from ∼16-30% (Fig. S2A,B). Cre activities at later embryonic stages (E11.5 and E12.5) were also tracked. Notably, only notochord and notochord-descendants were affected, not the surrounding SOX9+ cells at the level before the hindlimb bud (Fig. S3A,B). At relatively posterior regions, such as levels of the sacral and tail vertebrae, the expression of β1 integrin in the surrounding somatic regions was also slightly decreased compared with the control (Fig. S3C). Approximately half (54%, 111 out of 206) of Itgb1^ΔND mice could be distinguished from their littermates at birth by short and distorted tails (Fig. 2A), which may be representative of the Type I Cre activity. The remaining 46% of Itgb1^ΔND mutants had normal tails, which is in line with Type II and III Cre activities (Fig. S1D). Itgb1^ΔND mutants were viable, fertile and had normal life spans.

At postnatal day (P) 20, control mice had straight tails. Vertebrae had similar morphology and sizes, and were connected via IVDs (Fig. 2Aa). In sharp contrast, the severely affected Itgb1^ΔND mice resulted from Type-I Cre activity showed misaligned and malformed tail vertebrae (Fig. 2Ab). Some of the mutant tail vertebrae resembled hemi-vertebrae (Fig. 2-Ab′ and Ab″), seen in a rare congenital spinal malformation in human, in which only one side of the vertebral body develops, resulting in deformation of the spine (Varras and Akrivis, 2010). In some regions, the entire skeletal elements were absent (Fig. 2Ab, black arrow).

Histological analysis showed that each IVD in the control mice contained an NP in the core of the IVD (Fig. 2B, black arrows) sandwiched by two cartilaginous end-plates and surrounded by an inner and an outer annulus fibrosus. In the Itgb1^ΔND mutants, the NP were either smaller or absent from IVDs (Fig. 2B, red arrows). Detailed histological analysis at every fifth section at P10 revealed that the NP were missing at random intervals along the spine from thoracic level to posterior tail vertebrae (Table S1; Fig. S4). In addition to the NP absence observed at the postnatal stage, the notochord remnants at E12.5 were also revealed to be abnormally displaced from the center of the intervertebral mesenchyme (IM), compared with the centrally localized notochord remnants in the control embryo (Fig. S5, red arrows). These two mutant phenotypes suggest that a lack of β1 integrin specifically in the notochord results in abnormal NP formation, defective IVDs and curved upper spine (e.g. in the thoracic level), and that a lack of β1 integrin in the notochord and tail bud results in misshapen vertebrae in the tail.

To investigate the cause of NP loss within the IVD, we assessed notochord development at different stages of embryogenesis (E8.0 to E10.5, Fig. 3A). In control embryos, the notochord marked by T was a continuous midline structure from the anterior to almost the tip of the tail. The notochord in Itgb1^ΔND mice was interrupted at different locations along the A-P axis, which was evident as early as the 3-4 somite stage (Fig. 3A; Fig. S7A, red arrows). This discontinuity in Itgb1^ΔND mice was confirmed by whole-mount in situ hybridization (WISH) with two additional notochord markers, Shh and Noto (Fig. S6A). Larger interruptions were found in the posterior regions and at later stages, such as E9.5 (Fig. 3A, red arrows).

In order to distinguish whether the discontinuous T expression is due to cell fate change or defective structural integrity, we sectioned E8.25 embryos after WISH (Fig. S6B) and E9.5 embryos for immunostaining (Fig. S6C). Whereas control E8.25 embryos displayed intense T RNA signals below the folding neural tube, the Itgb1^ΔND embryos lacked T expression in the interrupted region (Fig. S6B). At E9.5, the notochord structure from control embryos showed DAPI- and T-positive nuclei, whereas Itgb1^ΔND embryos lacked DAPI-positive notochordal cells beneath the floor plate in some regions (Fig. S6C). Despite the interruptions in the Itgb1^ΔND notochord structure, the patterning of the floor plate and the neural tube were unaffected, as reflected by the expression patterns of molecular markers such as PAX6, PAX7 and NKX2.2 (Fig. S6D; Yamada et al., 1991; Roelink et al., 1995; Ericson et al., 1997; Briscoe et al., 1999, 2000).

Notochord-derived signaling is also known to induce the differentiation of the ventral somatic derivatives into the sclerotome (Pourquie et al., 1993). Thus, we also examined the somite differentiation from E8.5 to E10.5 using WISH for Uncx4.1 (Uncx) and Pax1. Consistent with the different Cre activities observed, somite formation was normal up to E9.5, but Pax1 expression was interrupted in the tail somites of some mutant embryos at E10.5 (Fig. S7A-C), which could be caused by the activity of Type I Cre. This observation correlates with an obvious tail abnormality in the mutants from E12.5 (Fig. S8).

To understand the developmental basis of the NP phenotype, we focused our further analyses on early embryonic stages when expression of Cre was restricted to the notochord. The notochord is initially formed as a plate of cells embedded inside the dorsal gut endoderm. As soon as the cells fold off from the endoderm roof, they form a cylindrical rod-like structure, called the definitive notochord (Jurand, 1974; Sulik et al., 1994; reviewed by Balmer et al., 2016). Later, as the surrounding mesenchymal cells penetrate the space between the floor plate and the notochord, the notochord separates from the neural tube. At the more anterior region of control E9.0 embryos (∼16-17 somites), the notochord adhered to the midline structures on both sides, the overlying neural tube and the underlying foregut (level Ba in Fig. 3B). Slightly more posteriorly, the notochord had detached from the endoderm but still adhered to the ventral side of the neural tube (level Bb in Fig. 3B), leaving the notochord at the center of the dorsal-ventral (D-V) midline axis (Fig. 3B, dotted lines). In Itgb1^ΔND embryos, the notochord had separated from the neural tube and displayed prolonged contact with the endoderm, resulting in an abnormally large, acellular space between the neural tube and the notochord and, hence, displacement of the notochord from the D-V midline axis (Fig. 3B).

In control E9.5 embryos, the rod-like notochord structure was located at the left-right (L-R) midline axis between the neural tube and foregut endoderm, whereas in Itgb1^ΔND embryos the notochord was displaced to the lateral side of the neural tube, away from the L-R midline axis (Fig. 3C). In line with these findings, WISH analysis revealed that T signals were displaced in some regions from the midline axis (yellow arrows in Fig. 3A), which possibly resulted in the displaced notochord remnants observed at E12.5 (Fig. S3C, asterisk; Fig. S5, red arrows). These data suggest that the absence of β1 integrin in Itgb1^ΔND embryos leads to cell loss in the notochord. The consequential interruptions and defective structural integrity and alignment in the notochord affect its positioning.

Integrin-mediated cell-ECM adhesion signaling supports cell survival (Frisch and Ruoslahti, 1997) and cell proliferation (Legate et al., 2006). As the Itgb1^ΔND embryos consistently displayed notochord defects with interruptions (∼70%) (Table S2), we next examined whether the cell loss was due to mesoderm deficiency or impaired cell proliferation or survival. EdU (E8.5, Fig. 4A) and BrdU (E9.5, Fig. 4D) staining showed that at E8.5 ∼30% (Fig. 4B) and at E9.5 ∼22% (Fig. 4G) of the cells within the notochordal plate incorporated the DNA dyes in both control and Itgb1^ΔND embryos. In line with a previous report (Bellomo et al., 1996), numerous cells surrounding the notochord and in the primitive streak were also actively proliferating in control and Itgb1^ΔND embryos (Fig. 4A).

To test whether the notochord interruptions are due to increased apoptosis, we performed whole-mount immunostaining for active caspase 3 (actCas3) at E8.5. In control and Itgb1^ΔND embryos, scattered actCas3 signals were detected in the mesenchyme surrounding the notochord (Fig. 4Ca′,Ca″,Cb′,Cb″, yellow arrows), but not within the notochord or in regions with gaps (Fig. 4Cb′,Cb″, white arrows). Staining in the posterior region of E9.5 Itgb1^ΔND embryos (Fig. 4E), in which the defective notochord phenotype was particularly prominent (Fig. 4F), revealed very few actCas3-positive cells in the notochord region, which was also the case in control embryos (Fig. 4H). However, Itgb1^ΔND embryos had occasional aggregates of T-positive cells at the tip of interrupted segments (Fig. 4I, oval outline).

FN is the major ligand of α5β1 integrin and a major component of the notochordal sheath (Gotz et al., 1995; Smits and Lefebvre, 2003). In control embryos, FN was strongly enriched in the sheath surrounding the notochord; however, it was significantly decreased in the notochord sheath of Itgb1^ΔND embryos with notochord displacement, suggesting that β1 integrin loss is the reason for the weakened/reduced FN sheath formation (Fig. 4J). Moreover, some Itgb1^ΔND embryos displayed an acellular ‘ghost’ notochord surrounded by an LN-containing sheath (Fig. 4K).

In chordate notogenesis, convergent extension is one of the major processes by which the cells converge towards the midline, driving the elongation of the notochord along the A-P axis (Ezin et al., 2003; Sausedo and Schoenwolf, 1994; Yamanaka et al., 2007). During the lengthening of the node along the A-P axis, cells in the axial midline intercalate along the medial-lateral (M-L) axis, resulting in a notochordal plate consisting of two adjacent rows of cells (Yamanaka et al., 2007). To determine whether impaired convergent extension contributes to the notochordal defects of Itgb1^ΔND embryos, we compared cellular intercalation and the length/width ratio (LWR) of the developing notochord of control and Itgb1^ΔND embryos. In controls, T immunostaining showed the notochordal plate in the midline containing two rows of cells up to the node level (Fig. 5A, white lines before the node). By contrast, in stage-matched Itgb1^ΔND embryos, the notochordal plates were wider, contained more rows of cells (Fig. 5A, highlighted by yellow lines), had more cells per somite (Fig. 5B, P<0.05) and a reduced LWR within the somite region (∼25% lower than controls; Fig. 5C, P<0.005). This correlates with the obviously wider notochord at E9.5 (Figs 3C, 5D, P<0.0001).

In order to understand the cellular basis of the wider notochordal plate, we performed time-lapse live imaging of the node region using the EGFP reporter mouse line (zEG) (Novak et al., 2000). Embryos with 1-2 somites were cultured and imaged overnight for ∼17 h (Fig. S9A). At the end of overnight culture, the embryos had usually developed into stages with 8-9 somites, depending on the starting age of the embryo (Fig. S9A). To exclude any effects introduced by photo-toxicity during overnight imaging, the embryos were dissected to check their morphology and only healthy ones were selected for further analysis. In an E8.0 embryo with ∼1 somite, the anterior notochordal plate had already formed. Single cell tracking in the node peripheral region with crown cells, in which the most active morphogenesis occurs (Fig. S9C, white arrows), revealed that the convergent movement of node cells to the midline is coupled with directionally persistent movement in an A-P direction along the midline in the control embryos (Fig. 5E, upper panel; Movie 1). In contrast, in mutant embryos, the node cells displayed a tumbling behavior near their original positions during the first 3 h (Fig. 5E, ∼0-200 min, lower panel; Movie 2). Ultimately, movement towards the posterior did occur, albeit with lower directionalities (Fig. 5F, P<0.0005). Quantitative analysis revealed no significant difference in velocities and accumulated migrating distances between control and β1 integrin-deficient node cells (Fig. 5G,H). However, owing to the reduced directionality, mutant node cells traveled less far in Euclidean distances (Fig. 5I, P<0.0004). Thus, the compromised notochord convergent extension observed in the mutants is due to reduced directional persistence of cell movement upon reduced β1 integrin expression (Fig. S10).

To assess the impact of loss of β1 integrin on intercellular adhesion of the mouse notochord, we analyzed N-cadherin expression in the mutants with defective convergent extension. Compared with the control notochord, the N-cadherin signal was enhanced in the mutant notochord from anterior to posterior at E8.5 (Fig. 6A,C, P<0.0001) and at E9.5 (Fig. 6B,C, P<0.0001).

Wnt/PCP signaling is a key conserved mechanism coordinating polarized cell behaviors (Gray et al., 2011; Yang and Mlodzik, 2015). In vertebrates such as zebrafish, it regulates convergent extension along the embryonic A-P axis (Heisenberg et al., 2000; Park and Moon, 2002; Kilian et al., 2003), and in Xenopus, through the assembly of the FN matrix (Dzamba et al., 2009). In mouse, the pit cells in the center of the node establish PCP as early as E7.5 (Hashimoto et al., 2010; Mahaffey et al., 2013) and show clear asymmetric VANGL2 localization at E8.25 (Song et al., 2010; Mahaffey et al., 2013). In line with no Foxa2mNE-Cre activity in pit cells using the current transgenic Cre tool (Fig.S1B), we detected asymmetric VANGL2 localization in both control and Itgb1^ΔND mutants (Fig. S11A,B), suggesting that PCP establishment in the central node is a cell-autonomous behavior, which is not affected by the Itgb1 deletion in the surrounding crown cells.

In order to determine whether β1 integrin is required for PCP establishment, we cultured wild-type embryos from E7.25 to E8.25 (Fig. S12A) in the presence of control or β1 integrin functional blocking antibodies. At the end of culture, the embryo overall morphology and body length between two groups were similar (Fig. S12A,B). As shown in Fig. 7A, embryos cultured overnight with the isotype-control antibody displayed clear asymmetric localization of VANGL2 in the center of the node in which the pit cells are localized. In sharp contrast, the number of cells showing asymmetric VANGL2 distribution was significantly reduced in pit cells of embryos cultured with β1 integrin functional blocking antibodies (Fig. 7A,B, arrows indicate cells with asymmetric VANGL2 and asterisks indicate cells without clear VANGL2 asymmetry). The few cells with normal, asymmetric VANGL2 distribution (Fig. S12C,D) were likely not fully blocked with the β1 integrin antibody. Altogether, these data point to a role of β1 integrin signaling in the establishment of PCP in the pit cells of the node.

The notochord of an E9.5 mouse embryo is a tiny structure with ∼1500 cells. Notochordal cells isolated via cell sorting do not proliferate and cannot be maintained in vitro under normal culture condition (S.S.G. and K.S.E.C., unpublished). To obtain enough cells for in vitro analysis, we immortalized notochordal cells via retroviral transduction with the SV40 large T antigen. Antibody staining with two notochord markers, Foxa2 and T, suggested that notochordal characteristics were maintained in the immortalized cells (Fig. S13A, upper panel). Expression profile analysis of integrins in the notochordal cells show that α5β1 (FN receptor) is the most abundant integrin, followed by the less abundant αvβ3 [vitronectin (VN) receptor] and α6β1 (LN receptor), and low collagen receptor expression (α1β1) (Fig. S13C).

To test whether cell-matrix interactions were changed once β1 integrin expression was lost, we deleted Itgb1 using the CRISPR/Cas9 system. Gene deletion was verified by immunostaining (Fig. S13B), sequencing the frame shift (Fig. S13D) and western blot (Fig. 8E,F). Although cells lacking β1 integrin (β1-KO) are smaller than parental control cells, notochordal identity was maintained, as defined by Foxa2 and T expression (Fig. S13A, lower panel).

Next, we performed cell adhesion assays on different substrates (Fig. 8A) and found that β1-KO cells failed to adhere to collagen I and LN, and to a lesser extent to FN and VN when compared with control cells. The reduced adhesion of β1-KO cells on FN was associated with an inability to spread (Fig. 8B) and to migrate in scratch assays (Fig. 8C). Detailed cell tracking analysis revealed that migrating distance, speed and directionality were impaired in β1-KO cells (Fig. 8D). The pronounced spreading and migration defects of β1-KO cells were accompanied by reduced GTP-Rac1 and GTP-Cdc42 activities (Fig. 8E). Moreover, consistent with the observation in vivo, N-cadherin expression level was also upregulated in the β1-KO cells (Fig. 8F).

Although a key characteristic of the notochord is its midline position, it is unclear to what degree deviation from this feature contributes to its functional attributes and influences the overall patterning and alignment of the vertebral bodies and IVDs of the spine. Further, the role of cell-matrix interactions mediated by β1 integrin, the core subunit of the largest subfamily of ECM-binding integrins, in notochord induction and morphogenesis, and notochord integrity and signaling function, is unknown because notochord-specific manipulations of the gene have been lacking. In this study, we employed a notochord-specific enhancer from the mouse Foxa2 gene to drive Cre recombinase-mediated ablation of β1 integrin expression specifically in notochordal cells at early somite stages. We report that loss of β1 integrin in the Itgb1^ΔND mutants impacts the structural integrity and midline positioning of the notochord, which in turn affects the proper development of vertebrae and IVDs in the spine. Our study shows that these defects arise because convergent extension of the notochord during its morphogenesis and its activity in vertebral body patterning are impaired.

Cell lineage tracing studies have suggested that all cells of the mouse NP, from tissue formation through to skeletal maturity, are of notochord origin (Choi et al., 2008; McCann et al., 2012). A striking abnormality observed in the Itgb1^ΔND mutants is the interruption of the notochord at multiple locations along the A-P axis, displacement from the midline axis and the presence of notochordal sheath segments devoid of cells. These abnormalities correlate with absence of NP in the IVD at postnatal stages and are consistent with this concept of the developmental origin and contribution of notochordal cells to the mature NP.

As Foxa2mNE-Cre expression commences at ∼E7.5-8.0, a stage at which the node structure has already matured (Mahaffey et al., 2013), and the half-life of the β1 integrin protein exceeds 24 h (Brakebusch et al., 2000), Foxa2mNE-Cre effectively ablates β1 integrin function in the notochord from ∼E8.5. Therefore, we can exclude defective development and morphogenesis of the node as cause for the notochord displacement from the midline axis. Our experiments also ruled out cell loss by apoptosis and/or reduced cell proliferation as the cause of the fragmented and dislocated notochord in Itgb1^ΔND embryos.

Depletion of α5 and αV integrins in zebrafish results in a curled notochord that detaches from the paraxial mesoderm during trunk elongation (Dray et al., 2013). Constitutive deletion of the FN (FN1) gene in mice results in impaired mesoderm development and absence of the notochord (George et al., 1993). However, although mesoderm development in Itgb1^ΔND embryos initiated normally, we noted reduced deposition of FN around the mutant notochord. As FN is assembled by α5β1 integrins into a stable and biologically active fibrillar structure, the absence of α5β1 integrins in Itgb1^ΔND notochordal cells could contribute to the reduced FN in the notochordal ECM sheath marked by LN.

Mechanical forces exerted by the expanding amniotic cavity during mouse embryo elongation along the A-P axis have been shown to drive convergent extension of the mouse notochord (Imuta et al., 2014). Our findings suggest that notochord integrity and midline position depend on normal β1 integrin signaling, which mediates mechanical coupling within the notochord tissue itself, rather than on the node geometry (Pulina et al., 2014). It is possible that the loss of β1 integrin in our mutants changed the structural composition of the notochord sheath, thereby affecting cell-matrix interaction and cell-cell adhesion, and altering the inter-tissue mechanical coupling between the notochord and the paraxial mesoderm. The defective integrin-mediated connection between notochord and peri-notochord tissue would render the notochord unable to withstand the mechanical forces generated during elongation, leading to fragmentation and the displaced midline structure.

Although deficiency in ECM composition of the notochordal sheath could contribute to the interruptions and displacement of the notochord in Itgb1^ΔND mutants, it is interesting that FN deficiency in the ascidian Ciona does not interfere with the development of a sheath surrounding the notochord, but profoundly affects intercalation and migration of notochordal cells and therefore convergent extension (Segade et al., 2016). Our experiments provide strong evidence that reduced directional persistence of the movement of notochordal cells lacking β1 integrin account, at least in part, for the defective notochordal convergent extension, midline positioning and acellular interruptions observed in Itgb1^ΔND mice. The defective cell movement could be caused by impaired cell-cell or cell-matrix adhesion and/or altered biomechanical properties of the notochord and its sheath. In Xenopus, inhibition of β1 integrin function alters C-cadherin-mediated cell adhesion, affecting cell intercalation and inhibiting convergent extension and axial elongation of gastrulating embryos (Marsden and DeSimone, 2003). The abnormally high intercellular expression of N-cadherin observed in the Itgb1^ΔND mutants and in the β1-KO notochordal cells suggests imbalance of cell-cell and/or cell-matrix adhesions between notochordal cells and the ECM sheath, contributing to the defective convergent extension.

The PCP pathway has been shown to be involved in the assembly of ECM proteins such as FN in Xenopus gastrulation (Goto et al., 2005). The importance of the PCP pathway for notochord development was illustrated in embryos lacking the core PCP pathway genes Vangl2 (lp/lp mice) and Ptk7, in which the notochordal plate widened (Yen et al., 2009; Mahaffey et al., 2013; Imuta et al., 2014). PCP is established in the pit cells of the node, which give rise to the anterior notochord/notochordal plate (Hashimoto et al., 2010; Song et al., 2010). However, VANGL2 was properly localized in the node of Itgb1^ΔND embryos. This result is probably because the notochord-specific Cre line used in our study is not active in pit cells and does not prevent β1 integrin expression until ∼E8.0-E8.5, which is after the node has matured and PCP is established (Mahaffey et al., 2013). It is also of note that Vangl1 but not Vangl2 is expressed in the developing notochord (Torban et al., 2008) and Vangl1 mutants do not display an embryonic PCP phenotype (Torban et al., 2008; Song et al., 2010).

To circumvent the lack of Cre activity in the pit cells, wild-type embryos were treated with anti-β1 integrin blocking antibodies, which are in IgM isotype. Fc domains present in both IgG and IgM are known to mediate antibody transmission from mother to young in mammals via the Fc receptor expressed in the surface of the embryo yolk sac (Roberts et al., 1990; Kim et al., 2009; Merad and Wild, 1992). Although we cannot discount possible differences between the transmission of the functional blocking antibodies and the corresponding IgM isotype control, the results of the antibody blocking experiments strongly suggest that β1 integrin is important for the early establishment of PCP in the node, which could impact on the developing notochord.

It is possible that the convergent extension defect and the altered migration of notochordal plate cells observed in 5- to 7-somite-stage mutant embryos may not be caused by a VANGL-mediated PCP defect. Such a scenario would be consistent with other examples, in which the relationship between PCP and convergent extension is not definitive (Tada and Heisenberg, 2012; Sutherland et al., 2020). Later in notochord development, β1 integrin may work with different PCP components to regulate convergent extension (Imuta et al., 2014; Sutherland et al., 2020). In Itgb1^ΔND mutants, convergent extension is impaired because this role is abrogated and/or additional external mechanical forces contribute. Further studies are needed to distinguish between these possibilities and determine whether β1 integrin itself and/or the downstream signaling effectors indeed interact with components of the PCP pathway to regulate mouse notochord convergent extension, for example, by using a mouse line that expresses Cre in the pit cells and works before node formation.

In addition to notochord-specific Cre expression, we found that some of the transgenic mice expressed Cre in the tail bud from E9.5. This expression pattern correlates with the presence of defective tail somites after E10.5 in some mutants. Hemi-vertebrae in the tail and abnormal vertebrae and scoliosis in the upper spine were observed in the Itgb1^ΔND mutants. The hemi-vertebrae in the tail are possibly due to the combination of Cre activity in the notochord and the tail bud at E9.5, which later affects posterior (tail) somitogenesis. In the mouse, the notochord is the source of SHH signals essential for induction of the floor plate, sclerotome differentiation and subsequent formation of the vertebral column (Echelard et al., 1993; Chamberlain et al., 2008). Removal of Shh expression in the notochord resulted in absence of SHH in the floor plate, and loss of the IVDs and vertebral structures, indicating an essential role for a notochordal source of SHH for patterning the spinal column (Choi et al., 2012). It has also been shown that, in the absence of the notochord, the floor plate is sufficient for maintaining differentiation of the somite into the sclerotome and vertebra (Ando et al., 2011). It is interesting that in the Itgb1^ΔND mutants, despite the interruptions in the notochord, induction of the floor plate and neural tube patterning are not perturbed adjacent to regions from which the notochord is absent. This suggests that an initially intact notochord in the Itgb1^ΔND mutants provides sufficient SHH for floor plate induction. This initial source, together with endogenous SHH production within the floor plate, is sufficient to pattern the neural tube.

Although it is not possible to ascribe the tail vertebrae phenotype at birth solely to Cre activity in the notochord, the influence of the notochord on development of the vertebrae cannot be excluded. Lineage tracing for the cells in the node of the late-streak-stage embryo (∼E7.5), have been shown to contribute to the anterior notochord. Node cells from more advanced embryos (E7.5-E7.75), including posterior pit and crown cells, also contribute to the notochord in the upper trunk (Tam et al., 1997; Kinder et al., 2001; Robb and Tam, 2004; Yamanaka et al., 2007). Whether descendants of tail bud cells can contribute to thoracic vertebrae is an open question, especially because fate mapping studies using E9.5-E13.5 tail bud cells show contribution to more posterior somites (Tam and Tan, 1992). Grafting studies have shown that tail bud cells do not incorporate well to the axis, and tend to contribute to short stretches of somites (Cambray and Wilson, 2002). Therefore, as abnormal vertebrae were also observed in thoracic regions postnatally (Fig. 2B), and displaced notochord segments at the level of future trunk vertebrae were found at E12.5 (Fig. S5), a possible influence of the notochord cannot be excluded.

For the tail, Cre activity in the tail bud may impact the differentiation of posterior somites. This impact, together with interrupted SHH production from the fragmented notochord, is not compensated by the floor plate resulting in abnormal development of the tail IVDs and associated vertebrae, culminating in the morphological abnormalities seen, such as malformed vertebrae and hemi-vertebrae. Combined with the thoracic vertebral defect, our data suggest that the notochord midline position is crucial for the site of IVD formation and the alignment of the vertebral column. Such a scenario provides insight into the anomalies present in the upper spine of Itgb1^ΔND mutants at birth.

The phenotypic consequences of the disrupted notochord and tail defects found in the Itgb1^ΔND mutants suggest a novel mechanism for the pathogenesis of the human congenital disorders, hemi-vertebra and scoliosis. The current study implicates β1 integrin-mediated cell-ECM interactions serving as the safe guardian for the notochord along its journey to NP by maintaining its proper convergent extension, structural integrity and fine-tuned midline position that is essential for the proper formation of the NP in the IVDs. Whether developmental defects in the notochord underlie human congenital spine disorders warrants further investigation. In the future, it would be important to clarify a direct correlation between the embryonic midline position of the notochord with effects on sclerotome differentiation and vertebral body patterning at different levels of the spine, and the relationship to scoliosis.

The Foxa2mNE-Cre transgenic mice were generated by pronuclear injection into F1 C57/Bl6 X CBA zygotes and maintained in the same F1 background (Au et al., 2020). The floxed β1 integrin line (β1^F/F) and the zEG reporter strain have been described previously (Novak et al., 2000; Potocnik et al., 2000). All animals were maintained in the Laboratory of Animal Unit at the University of Hong Kong. Animal care and experiments performed were in accordance with the protocols approved by the Committee on the Use of Live Animals in Teaching and Research of the University of Hong Kong.

P20 control and Itgb1^ΔND mutant mice were eviscerated, fixed in 100% ethanol and transferred to acetone solution. Specimens were rinsed with water and stained in Alcian Blue/Alizarin Red solution (McLeod, 1980) for 10 days before clearing and storing in 100% glycerol for photography.

The frequency of mosaicism was defined by the ratio between the cell numbers with mosaic β1 integrin expression and the total cell numbers that were DAPI+T+. Signal intensities were measured using ImageJ. Cells with an intensity <20 arbitrary units (a.u.) were grouped into ‘complete knockout’, >20 a.u. were grouped into ‘mosaic expression’ (Fig. S2A).

WISH was carried out with modifications on the standard protocol (Wilkinson, 1992) using digoxigenin (DIG)-labeled riboprobes specific for transcripts of T (Wilkinson et al., 1990), Noto (Abdelkhalek et al., 2004), Shh (Echelard et al., 1993), Uncx4.1 (Mansouri et al., 1997) and Pax1 (Deutsch et al., 1988). Anti–DIG-AP antibody (#11093274910, Roche, 1:2000) and BM Purple (#11442074001, Roche, ready-to-use solution) were used for signal detection.

Cell proliferation activity was analyzed using BrdU or EdU labeling assays. Mice were injected intraperitoneally with 200 μg of BrdU or EdU per gram of body weight 2 h before sacrifice. Following fixation, BrdU incorporation in paraffin sections from E9.5 embryos was detected using a BrdU Staining Kit (93-3943, Thermo Fisher Scientific). EdU incorporation on E8.5 embryos was detected using Click-iT™ EdU Alexa Fluor™ 488 Imaging Kit (C10337, Thermo Fisher Scientific).

Tissue sections, cells or embryos underwent immunostaining with the following antibodies: integrin β1 (Millipore, MAB1997, 1:400), brachyury (Santa Cruz Biotechnology, SC-17745, 1:200), Foxa2 (Santa Cruz Biotechnology, SC-101060, 1:200), laminin (Sigma-Aldrich, L9393, 1:800), fibronectin (Millipore, Ab2033, 1:400), vincullin (Sigma-Aldrich, V9131, 1:500), Phalloidin Alexa 546 (Thermo Fisher Scientific, A22283, 1:500), N-cadherin (BD Transduction Laboratories, 610920, 1:800), SOX9 (Millipore, Ab5535, 1:1000), GFP (Abcam, ab6556, 1:1000), active caspase 3 (Cell Signaling Technologies, #9661, 1:100). VANGL2 antibody (rabbit, 1:800) was a kind gift from Dr Matthew W. Kelley (National Institutes of Health, Bethesda, MD, USA).

For whole-mount embryo staining, the embryos were dissected out followed by the removal of the head and the heart for the flat mount at the end. Embryos were permeabilized with 0.5% Triton X-100 (v/v) in PBS for 30 min before antibody application. Both primary and secondary antibodies were incubated overnight at 4°C.

Fluorescent images were obtained using a Carl Zeiss LSM700 or LSM780 microscope maintained by the University of Hong Kong Li Ka Shing Faculty of Medicine Faculty Core Facility.

Imaging was set up following protocols described previously (Yamanaka et al., 2007; Imuta et al., 2014). Embryos at stages between ∼E7.75-E8.0 (∼1 s) were dissected in F12 with 10% fetal bovine serum (FBS) and 15 mM HEPES and were held within the home-made sticky edges on the CoverWell perfusion chamber gaskets (Fig. S9A). Embryos were then cultured in 50% rat serum (v/v), 50% dissection medium (v/v) without Phenol Red supplemented with 100 µM of 2-mercaptoethanol and 1 mM of sodium pyruvate in 5% CO₂ in air at 37°C. A thin layer of miner oil was applied to the medium surface to avoid evaporation.

Embryos were allowed to recover for ∼1 h in the rotating culture system before imaging. Time-lapse live images were acquired using an inverted microscope (LSM510, Carl Zeiss) equipped with an incubator unit, a 488-nm diode laser and a 20× lens (NA=0.75). The cells within both the notochord and the node were imaged for 16.8 h. Images were acquired every 10 min, with ∼16-23 z-positions at each time point, and the section interval was 10 µm. Drift in embryo positioning during imaging was corrected manually based on the non-moving or slow-moving cells and the midline axis at every consecutive time point. Cell moving behaviors were tracked and analyzed via ImageJ. The parameters analyzed are illustrated in Fig. S9B.

Wild-type embryos at stages ∼E7.25 were dissected in the medium mentioned above. Then 20 µg/ml functional blocking antibodies (BD Pharmingen, 555002, 1:50) or the corresponding IgM isotype control (BD Pharmingen, 553957, 1:50) were applied to the embryos and cultured with a rotating system with 5% CO₂ in air at 37°C with the medium described above. Embryos were harvested 24 h later, stained and flat-mounted for confocal imaging.

Notochordal cells were GFP-labeled by intercrossing the Foxa2mNE-Cre line with an EGFP Cre reporter strain (zEG) (Novak et al., 2000). Notochord-containing tissues were collected from ∼20 embryos at E9.5 and trypsin digested for 15 min at room temperature. Notochordal cells were then enriched by GFP cell sorting and immortalized by retrovirally transducing the SV40 large T antigen.

Two guide RNA sequences (gRNAs) targeting mouse Itgb1 were synthesized and cloned into pSpCas9(BB)-2A-Puro (PX459) V2.0 (Ran et al., 2013). gRNA-1: CAAACTGATCAATCCAATCC; gRNA-2: TATACATTCTCCGCAAGATT. Puromycin at 3 µg/ml was used to screen the positive clones. Knockout clones from both gRNAs displayed similar morphologies and defects in cell spreading. Data presented here show the knockout clone from gRNA-1.

For integrin expression profiling, after trypsinization and washing, notochordal cells were stained with PE-conjugated or non-conjugated integrin antibodies diluted (1:200) in FACS buffer [1% bovine serum albumin (BSA) in PBS] for 30 min in dark on ice, washed twice with ice- cold FACS buffer. The secondary antibody was incubated for another 30 min if necessary. Flow cytometry was analyzed using a flow cytometer (FACSCanto II; BD Biosciences).

FACS antibodies used: anti-β1 integrin (HMβ1-1, BioLegend, 102207), anti-β2 integrin (C71/16, BD Biosciences, 553293), anti-β3 integrin (2C9.G3, eBioscience, 12-0611), anti-β4 integrin (346-11A, Serotec, MCA2369PE), anti-β7 integrin (M293, BD Biosciences, 557498), anti-α1 integrin (Ha31/8, BD Biosciences, 555001), anti-α4 integrin (R1-2, BD Biosciences, 553157), anti-α5 integrin (5H10-27, BD Biosciences, 557447), anti-αV integrin (RMV-7; BD Biosciences, 551187) and anti-α6 integrin (GoH3, BD Biosciences, 555736). Isotype control antibodies: rat IgG2b (eB149/10H5, eBioscience, 12-4031), rat IgG1 (R3-34, BD Biosciences, 554685), rat IgG2a, K (R35-95, BD Biosciences, 555844), rat IgG2a (YTH71.3, Serotec, MCA1212PE) and hamster IgG (eBio299Arm, eBioscience, 12-4888-83).

All substrates were coated at a concentration of 5 µg/ml overnight at 4°C. The coating substrates used in this assay were fibronectin (Sigma-Aldrich, F1141), vitronectin (Stem Cell Technologies, 07180), laminin (Life Technologies, 23017-15) and collagen (Advanced Biomatrix, 5005-100 ml). A 96-well plate was blocked with 1% BSA in PBS for ∼2 h before cell seeding. Cells were seeded at a density of 50,000 cell/well in DMEM with 1% FBS and placed in the incubator for 60 min for adhesion. Unbound cells were washed away, and the bound cells were fixed and stained at 4°C overnight with 0.1% Cristal-Violet (w/v) in 20% methanol (v/v). Cells were then permeabilized with 0.1% Triton X-100 (v/v) for 5 h at room temperature and the optical density was recorded under 595 nm.

Control or β1-KO cells were allowed to spread on the FN-coated surface (5 µg/ml) for 90 min before the analysis of the focal adhesion formation and the cytoskeleton organization. Cells were allowed to grow into ∼90% confluency in the six-well plastic well before scratching. For each well, two scratches were made using 200 µl pipette tips. Detached cells were washed away with warm PBS, and the wound closure was started in DMEM with 1% FBS. Then the cells were time-lapse live imaged under the wide-field lens using a Perkin-Elmer ultraView ERS spinning disk system attached to a ZEISS Axio Observer microscope with an EM CCD camera. Imaging was acquired with a Plan Neofluar 10×0.3 NA objective in 12 different microscopic fields at a 10 min interval overnight. Cell moving behaviors were tracked and analyzed using ImageJ.

To analyze the activities of Rac1 and Cdc42, control and β1-KO notochord cells were grown to ∼80% confluency and serum starved in DMEM with 0.5% BSA for 24 h. Cells were then trypsinized and suspended for 3 h in serum-free DMEM containing 0.1% BSA and 0.25 mg/ml soybean trypsin inhibitor (Life Technologies, 17075-029). Cells were then plated to dishes that had been coated with 20 µg/ml FN and blocked with 1% BSA. Cells were allowed to adhere to FN for 10 min, then were washed with 10 ml ice-cold PBS and extracted in lysis buffer [25 mM Tris HCl, 150 mM NaCl, 5 mM MgCl₂, 1% NP40 (v/v), 5% glycerol (v/v) (pH 7.2)] with proteinase inhibitors (Roche, 11873580001) and phosphatase inhibitors (sodium orthovanadate at 1 mM, Sigma-Aldrich, S6508). Rac1 and Cdc42 activities were tested via the detection kits from Cell Signaling Technologies (#8815 and #8819) and analyzed via western blot.

Antibodies used for western blot were: integrin β1 (Millipore, MAB1997, 1:2000), N-cadherin (BD Transduction Laboratories, 610920, 1:2000), actin (Sigma-Aldrich, A2228, 1:5000) and GAPDH (Calbiochem, CB1001, 1:10,000).

Statistical analysis was performed using GraphPad Prism 6. For every reported dataset at least three independent experiments were performed, except for Fig. S11. Statistical significance was assumed at P<0.05. *P<0.05; **P<0.005; ***P<0.0005 and ****P<0.0001. Statistical tests used for the experiments are described in the figure legends. Unless otherwise noted, all values are reported as mean±s.d.

We are grateful to Matthew W. Kelly for the gift of VANGL2 antibodies, Xiaochen Lin for the N-cadherin expression analysis in vitro, Ralph Thomas Böttcher for flow cytometry analysis, Zhiqi Sun for advice and for critical reading of the manuscript, Bo Gao and Patrick Tam for helpful and critical discussion.

The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.192724.reviewer-comments.pdf