The growth and expansion of meningeal lymphatic networks are affected in craniosynostosis

Meningeal lymphatic vessels (mLVs) are implicated in waste clearance and in modulating the immunological state of the central nervous system (CNS) (Alves de Lima et al., 2020). MLVs reside in dura mater – the outermost meningeal layer encasing the CNS – along both the dural venous sinuses and meningeal arteries (Antila et al., 2017). Although the functions of para-arterial mLVs have yet to be discerned, perisinusoidal lymphatics along the venous sinuses uptake cerebrospinal fluid (CSF) (Louveau et al., 2018; Ahn et al., 2019). During steady state, mLVs transport macromolecules (Louveau et al., 2015; Aspelund et al., 2015) and dendritic cells (Louveau et al., 2018) from the cranium to the deep cervical lymph nodes (dcLNs). As such, mLVs regulate CNS homeostasis and normal brain functions, and their dysfunction is associated with neurodegeneration, cognitive impairment and changes to neuroinflammatory processes (Da Mesquita et al., 2018; Louveau et al., 2018; Alves de Lima et al., 2020; Ding et al., 2021).

Given their anatomical location, mLVs evaded our attention for centuries until their ‘rediscovery’ in 2015 (Aspelund et al., 2015; Louveau et al., 2015). MLVs have now been reported in humans, non-human primates and rodents (Louveau et al., 2015; Antila et al., 2017; Absinta et al., 2017). In mice, mLVs grow postnatally by penetrating dura through foramina at the skull base, and begin forming preliminary networks along both the meningeal arteries and dural venous sinuses (Antila et al., 2017). Para-arterial mLVs begin at the pterygopalatine artery and complete their growth along the middle meningeal artery (MMA) by P28. Meanwhile, perisinus lymphatic networks begin developing along the sigmoid (SgS) and petrosquamosal sinuses (PSS), before ascending the transverse sinuses (TVS). Development continues until ∼P30, by which time coverage along the superior sagittal sinus (SSS) is complete (Antila et al., 2017; Izen et al., 2018).

MLVs require Vegfc and Vegfr3 signaling for their growth and maintenance (Antila et al., 2017). Vascular smooth muscle is postulated to be a local source of Vegfc and the growth of mLVs along blood vessels is concurrent with venous smooth muscle development (Antila et al., 2017). Differential gene expression in meningeal versus peripheral lymphatics suggests mLVs are unique according to gene pathways related to extracellular matrix (ECM) interactions, focal adhesion, and responses to endogenous and exogenous stimuli (Da Mesquita et al., 2018). Thus, processes that govern the growth and maturation of mLVs are likely specific to their unique meningeal environment. Although the detailed mechanisms of mLV development remain unknown, these likely include tissue-specific interactions with blood vessels and dura.

We recently reported that humans with craniosynostosis caused by mutations in the transcription factor TWIST1 (Saethre-Chotzen syndrome) have dural venous sinus malformations (Tischfield et al., 2017). Modeling the disorder in mice by inactivating Twist1 in the cranial sutures and dura via Sm22a-Cre (Twist1^Flx/Flx:Sm22a-Cre) revealed that Twist1 regulates the growth and remodeling of the dural venous sinuses in a non-cell autonomous manner (Tischfield et al., 2017). In the present study, we show that the growth and expansion of mLV networks are also perturbed in these models. Our findings now suggest that individuals with craniosynostosis may be at risk for functional impairments to mLV networks.

We generated craniosynostosis models by inactivating Twist1 in cranial mesenchyme (Twist1^Flx/Flx:Sm22a-Cre, i.e. Twist1^CS), as previously reported (Tischfield et al., 2017). Sm22a-Cre is expressed by E10.5 in neural crest and cranial mesoderm, giving rise to dura and arachnoid tissue (El-Bizri et al., 2008; Tischfield et al., 2017). Lineage labeling shows strong Cre activity in sutural mesenchyme, especially the coronal and sagittal sutures (Fig. 1A). These regions of non-ossified tissue give rise to osteoblasts and are commonly affected in Saethre-Chotzen syndrome (Twigg and Wilkie, 2015; Ottelander et al., 2021). Skull malformations were imaged using three-dimensional x-ray microscopy, revealing varying degrees of suture fusion (Fig. 1B). In a small subset of Twist1^CS animals (13%, 4/30), the skull was intact with bilateral fusion of the coronal sutures. In ∼60% of animals (17/30), the skull showed bilateral fusion of the coronal sutures and absence of bone along the dorsal midline, affecting the frontal, parietal and interparietal bones (Fig. 1B). Areas that lacked mineralized bone instead contained a thin, hardened, semi-translucent material that covered the underlying soft tissue (Fig. S1A). Regionalized absence of bone was also often observed where the sagittal and coronal sutures are normally located (Fig. 1B, Fig. S1B). In the remaining animals, bone absence was more severe and stretched further away from the midline (Fig. S1B). Skull length from the nasal to occipital bones was reduced by ∼25% (Fig. S1B). Thus, loss of Twist1 in sutural mesenchyme via Sm22a-Cre models a severe form of craniosynostosis marked by premature fusion and/or partial agenesis of the cranial sutures and regionalized absence of bone.

Dura mater underlies calvarial bone and sutural mesenchyme and secretes growth factors that are crucial for skull development and suture maintenance (Ito et al., 2003; Lee et al., 2006). As the dural venous sinuses and mLVs also reside in dura, we examined the development of dura at or adjacent to the dorsal midline in E16.5 embryos, by which time the meninges have differentiated and bone mineralization is commencing. In affected embryos, meningeal tissue closer to the skull base was relatively normal and three layers were recognized. In mildly and more moderately affected embryos, however, the dura and meninges were intact but became noticeably hypoplastic at or near the dorsal midline, as detected by co-staining with Crabp2 and connexin 43, which marks dura and arachnoid tissue and the arachnoid membrane, respectively (Fig. 1C-E). Additionally, condensed osteogenic mesenchyme was hypoplastic (Fig. 1C,D,D′). The apical expansion of bone growth was also delayed and/or absent adjacent to the midline as E-cadherin/N-cadherin positive osteoblasts had not yet developed (Fig. 1F). These findings were more pronounced in severely affected embryos and extended further away from the dorsal midline (Fig. S1C). These results show that calvarial bone absence parallels the loss of dura, which is more significant dorsolaterally and/or at the dorsal midline of the head where the dural venous sinuses and mLV networks reside.

MLVs are typically examined by scraping dura off the skull, followed by immunohistochemistry, flat mounting and imaging (Louveau et al., 2015). This was unfeasible in Twist1^CS animals as the hypoplastic dura did not stay intact during scraping. Instead, we performed a dorsal craniotomy and used chromogenic immunohistochemistry to stain vessels with Lyve-1. In control animals, mLV networks were patterned along the SSS and TVS, and were more complex with numerous branches at the sinus confluence and at particular locations along the TVS (Fig. 2A,A′). These regions have been proposed to be ‘hotspots’ specialized for the uptake of CSF (Louveau et al., 2018). By contrast, mLV networks were poorly developed in Twist1^CS animals (Fig. 2B,B′). Networks along the TVS appeared less complex and/or atrophic, and growth towards the sinus confluence was diminished. Branched vessel networks along the SSS were also hypoplastic and/or missing (Fig. 2B′). Lineage labeling showed that Sm22a-Cre is not expressed in mLVs (Fig. 2C). Furthermore, Twist1 is not expressed in meningeal endothelial cells and Twist1^FLX/FLX:Tie2-Cre embryos have normal vascular development with no signs of edema (Tischfield et al., 2017). The development of lymphatic networks in other tissues, such as ear skin, was normal (Fig. 2D). Thus, the observed phenotypes are non-cell autonomous and result from changes in the local environment that affect lymphatic networks.

We next examined mLV development in juvenile animals at postnatal (P) day 16, when mLVs are actively growing and remodeling along the TVS. We adapted a higher resolution method for imaging mLV networks by removing the dorsal half of the skull, above the temporal bone, with the meninges still attached. The skullcaps were then decalcified, stained and flat mounted to permit confocal imaging, as previously described (Antila et al., 2017). The TVS was visualized using α-smooth muscle actin (aSMA) to mark the thin layer of smooth muscle that envelopes the venous endothelium, and mLVs were stained with Lyve-1. In controls, mLV networks exhibited numerous sprouts and branches along the TVS (Fig. 3A). By contrast, networks in Twist1^CS animals typically consisted of long, unbranched vessels that were mostly devoid of sprouts, and coverage along the TVS was reduced (Fig. 3B). In more severely affected adults (P60), mLV networks were markedly hypoplastic along the TVS and SSS, and largely missing proximal to the confluence (Fig. 3C,D). Interestingly, hyperplastic lymphatic vessels were observed at the sinus confluence beneath the pineal gland. All affected animals had segmental or complete unilateral loss of the TVS and the intact vessel was hypoplastic in juvenile animals, but not in adults (Fig. 3B-D, Fig. S2A,B). Venous smooth muscle coverage appeared relatively normal in juveniles and on most adult vessels (Fig. 3A-D, Fig. S2A-D). However, coverage was patchy and showed less striation in severely affected animals, and aSMA-positive cells were present in the surrounding tissue, suggesting they failed to migrate (Fig. 3D). aSMA staining also revealed myofibroblasts in tissue surrounding the confluence, suggesting the dura was fibrotic, and Lyve-1-positive macrophages were abundant (Fig. 3D, inset). These results indicate that the growth and sprouting of mLVs are affected in Twist1^CS animals, accounting for loss of mLV networks in adults.

Basal networks along the SgS and PSS were often damaged in dorsal skull flat mounts. We therefore analyzed basal mLV networks using a variation of the aforementioned technique by bisecting the skull along the dorsal midline (Fig. S3). These ‘basal preparations’ preserve mLV networks along the TVS, SgS and PSS, and also allow imaging of mLVs along arteries near the skull base. For these analyses, we crossed Twist1^CS animals with a Prox1^tdTomato BAC-transgenic reporter line that strongly labels mLVs (Fig. 4A) (Hong et al., 2016).

In mildly affected animals, growth along the TVS was more comparable with that in control littermates, as branched networks were observed at ‘hotspot’ regions; however, vessel growth was still absent in areas where the TVS was missing (Fig. 4B,B′). Interestingly, basal networks along the PSS were exuberant in these mildly affected animals, containing long vessels with numerous branches (Fig. 4B″ n=2). In moderately affected animals, hypoplastic networks with rudimentary ‘hotspots’ and less branching were present along the intact TVS, but were largely absent and/or atrophic on the side where the TVS was missing (Fig. 4C,C′, n=3). Basal lymphatic networks along the PSS were less affected than dorsal networks. Long vessels were observed but branching and complexity were reduced compared with more mildly affected animals and controls. In severely affected animals (n=2), mLV growth and complexity along the TVS was significantly affected, and few if any sprouts and branches were observed. mLVs were also present along ectopic veins that branched from the remaining segments of the TVS (Fig. 4D). Basal networks along the PSS were variably affected. Similar to mild/moderately affected animals, long unbranched vessels were sometimes observed, and disorganized hyperplastic networks could accompany these on the contralateral side (Fig. 4D). In general, animals that were missing dorsal networks had exuberant basal networks, suggesting they were compensating for vessel loss along the TVS. Finally, vessels along the PPA and MMA were mildly affected; coverage along distal, dorsal regions of the MMA towards the skull apex appeared reduced in regions where dura was hypoplastic (Fig. S4), and arterial smooth muscle coverage was normal.

MLVs drain macromolecules and waste from CSF, and facilitate the trafficking of immune cells to the dCLNs (Louveau et al., 2018). mLVs also drain to the superficial cervical lymph nodes (scLNs), albeit to a lesser extent (Aspelund et al., 2015). To determine whether drainage to the dCLNs was affected, we infused a 45 kDa Ovalbumin-647 tracer into the CSF by accessing the subarachnoid space through the cisterna magna. We selected mild/moderately affected animals in which the meninges and subarachnoid space were preserved. The amount of tracer in the dCLNs was significantly reduced in Twist1^CS animals 30 min post-infusion, whereas the amount of tracer in the sCLNs was comparable (Fig. 4E). These results suggest the functions of mLV networks are compromised in craniosynostosis.

Our results show that changes to the meningeal environment in craniosynostosis can disrupt the growth and remodeling of mLVs. The model chosen for our current analysis is more affected than what is typically observed in humans, and shows similarities with Sweeney-Cox syndrome, a severe craniofacial disorder caused by dominant-negative mutations in TWIST1 (Kim et al., 2017; Takenouchi et al., 2018). However, homozygous Twist1^FLX/FLX:Sm22a-Cre animals model venous malformations that are absent in heterozygous Twist1^FLX/WT:Sm22a-Cre animals, the latter of which more closely mimic human skull phenotypes (Tischfield et al., 2017). Furthermore, mildly affected animals with bilateral coronal suture fusion approximate the human condition. Thus, mLV anomalies are likely present in humans to varying extents, especially those with syndromic craniosynostosis and venous sinus malformations. Considering venous sinus malformations are present in Apert syndrome caused by activating mutations in FGFR2 (Johnson and Wilkie, 2011), mLV anomalies may be widespread in craniosynostosis.

MLV growth is dependent upon Vegfc, which is expressed by venous and arterial smooth muscle (Antila et al., 2017). MLVs were rarely observed in regions where the TVS was missing, whereas vessel growth was observed along ectopic veins that developed in the absence of proper venous sinuses. This suggests smooth muscle-derived growth factor signaling from these vessels was sufficient to induce lymphatic growth. Smooth muscle coverage appeared relatively normal along the venous sinuses in most animals and also along the middle meningeal arteries, which are preserved in Twist1^CS animals. We did not observe overt changes to mLVs that grew alongside arterial vessels, with the exception of the more dorsal regions of the MMA, where the dura was hypoplastic. Thus, although some of the observed changes to lymphatic networks may be attributed to venous malformations, loss of smooth muscle and attenuated Vegfc signaling, hypoplastic dura and loss of the extracellular matrix likely has a greater effect in Twist^CS models. Notably, lymphatic development in embryos relies on flow and extracellular activation of integrin β1, which can bind and activate Vegfr3 independently of Vegfc (Planas-Paz et al., 2012). Although not addressed in the present study, elevated intracranial pressure and changes to flow, in combination with hypoplastic dura, may affect mLV networks in Twist1^CS models. In addition, the migratory abilities of lymphatic endothelial cells are influenced by mechanical forces regulated by tissue stiffness (Frye et al., 2018), which may be altered in more severely affected animals with hypoplastic and fibrotic dura. Thus, mLV phenotypes in Twist1^CS animals may manifest from combinatorial changes to the surrounding environment that impinge upon Vegfr3 activation, and the growth and sprouting of lymphatic networks.

Lymphatic drainage to the dcLNs was diminished in Twist1^CS animals. This can be expected to affect immune cell trafficking, as well as CNS waste clearance, especially because ablating mLVs also affects the functions of the perivascular waste clearance system of the brain (i.e. the glymphatic system) (Da Mesquita et al., 2018; Louveau et al., 2017). Impaired CNS waste clearance is associated with amyloid-β plaque build-up and cognitive impairment (Iliff et al., 2012; Da Mesquita et al., 2018), whereas altered meningeal immune cell trafficking can also affect brain function and behavior, in addition to neuroinflammation (Louveau et al., 2018; Alves de Lima et al., 2020). In aging animals, mLV networks naturally deteriorate; dorsal vessels regress, whereas basal vessels become hyperplastic (Antila et al., 2017; Ahn et al., 2019). Interestingly, hyperplastic basal networks were seen in a subset of animals. It is unknown, however, whether craniosynostosis may be associated with a higher prevalence of cognitive decline and dementia. Likewise, to our knowledge, changes to neuroinflammatory processes have not been reported. Given that craniosynostosis is associated with venous hypertension and impaired CSF drainage, our findings now imply that individuals with craniosynostosis, especially syndromic forms, may be at heightened risk for cognitive decline due to impaired waste clearance and/or altered immune surveillance.

The following transgenic mice were used: Twist1^FLX (RRID:MMRRC_016842-UNC), Prox1^tdTomato (RRID:MMRRC_036531-UCD) and Rosa26:Ai14^tdTomato (RRID:IMSR_JAX:007914). For all experiments, male and female mice were included. Animals were maintained on a mixed genetic background (C57Bl/6;FVB;CD1). Embryos obtained from timed matings were considered 0.5 days old upon observance of a plug. The ages of animals in this study include embryonic day (E) 16.5, and postnatal days (P) 16 and 60. Experiments were approved and carried out under IACUC protocol PROTO201702623 (M.A.T.).

The following antibodies were used: Lyve-1 (1:300, Abcam, ab14917, RRID:AB_301509), RFP (1:1500, Rockland, 600-401-379, RRID:AB_2209751), αSMA (1:300, Sigma, C6198, RRID:AB_476856), E-cadherin (1:500, Millipore, Mabt26, RRID:AB_10807576), N-cadherin (1:250, Invitrogen, 33-3900, RRID:AB_23113779), mouse Crabp2 (1:100, Millipore, MAB5488, RRID:AB_2085470), rabbit Crabp2 (1:100, Proteintech, 10225-1-AP, RRID:AB_ 2085455), mouse connexin 43 (1:200, Santa Cruz, sc-271837, RRID:AB_10707826) and rabbit connexin 43 (1:2000, Abcam, ab11370, RRID:AB_297976). Nuclei were visualized using Hoechst staining (1:2000, ThermoFisher, H3570) or standard DAPI fluorescence.

Heads from E16.5 embryos were decapitated and fixed overnight in 4% PFA at 4°C. The heads were then washed and dehydrated in steps of 50% and 80% ethanol prior to paraffin wax embedding. Embedded 10 µm sections were subjected to antigen retrieval in citrate buffer (pH 6.0) prior to staining. Antibodies were diluted in PBST with 0.3% triton and 5% normal goat serum and applied overnight at 4°C. Images were obtained using a LSM700 confocal microscope with a 20×0.8 NA objective and a 40×1.4 NA oil objective. For chromogenic 3,3′-diaminobenzidine (DAB) staining, the SignalStain DAB substrate kit was used (Cell Signaling, 8059) according to manufacturer's instructions. Briefly, following application of the primary antibody (Lyve-1), 800 µl of ant-rabbit horseradish peroxidase was added to the excised dorsal skull tissue for 30 min in a 24-well plate at room temperature. DAB (30 µl) was added to 1 ml of the DAB substrate solution. The solution was added to the tissue and allowed to incubate for 2-4 min until the signal was detectable but not saturated. The solution was then discarded and the tissue was washed in distilled water.

The animals were perfused with 4% paraformaldehyde, heads were decapitated, and all skin and musculature were removed from the dorsal half of the skull. The lower jaw was removed by inserting scissors into the oral cavity to cut the mandible and attached muscles. Angled micro scissors were used to cut laterally from the cisterna magna along the temporal bone to each orbit, and the nasal bones were severed to isolate the calvarium. The tissue was post-fixed in 2% PFA overnight at 4°C, washed in PBS and then treated with Dent's Fix (80% methanol, 20% DMSO) overnight at 4°C. Skulls were next decalcified in 14% EDTA for 7 days at 4°C and then cleaned of fat, muscle and connective tissue. Afterwards, the skulls were incubated with primary antibodies for 4 days, followed by secondary antibodies for 2 days at 4°C in 0.5% Triton with 10% normal goat serum. Incisions were made at the corners of the frontal and occipital bones, and the tissue was flat mounted onto slides in mounting medium (SouthernBiotech Fluoromount-G) and cover slipped. Basal flat mounts, which are described in detail in Fig. S3, were otherwise similar to the dorsal flat-mount preparation. The skull was bisected along the dorsal midline into halves by making an incision along the occipital bone midline, inserting the tips of scissors into the nasal bones and opening the scissors to split the skull.

Animals were sacrificed via transcardial perfusion with 4% paraformaldehyde (PFA). The heads were then decapitated and post-fixed overnight in 4% PFA. Hair and skin were removed prior to imaging. Images were obtained using a Bruker Skyscan 1272 X-ray microscope. The following scan conditions were used: image pixel size=13.5 μm, camera=1632 columns×1092 rows, rotation step=0.4 degrees, frame averaging=3, filter=1 mm Al. The resulting images were reconstructed and converted to dicom format with Skyscan Ctan software. Dicom files were opened in Vivoquant for segmentation of teeth and bone from less dense soft tissues.

Images were acquired on a LSM700 confocal microscope using a 10×0.3 NA objective. For dorsal and basal lymphatic preparations in Figs 3 and 4, maximum intensity projections (MIPs) were obtained using 7 µm z-stacks. For the quantification of sprouts along the TVS, these were manually counted from four control and experimental animals. Values were obtained from vessels present along the left and right transverse sinuses and averaged. The total vessel length and number of junctions for both the dorsal and basal networks in Fig. 4 were analyzed from MIPs using the AngioTool plug-in for ImageJ. For dorsal networks, a region of interest was centered over the transverse sinus, extending to the sinus confluence. For basal networks, a region of interest was centered over vessels along both the sigmoid and petrosquamosal sinuses (refer to Fig. 4A). The following settings were used in Angiotool: dorsal networks, value diameter and intensity=12; and for basal networks, value diameter and intensity=9, fill small holes=80. Values were obtained from at least one half of the skull for vessels comprising dorsal and basal networks from six control and seven experimental animals. Smooth muscle coverage was assessed along the intact transverse sinus in experimental animals and along the dominant vessel in controls. Vessel segments (2 mm) were traced and thresholded to determine the total percentage area covered along the transverse sinuses. The average width of the transverse sinuses was calculated using α-smooth muscle actin staining and/or dark-field illumination to mark the walls of the vessels, and a Leica M165FC stereomicroscope equipped with a 1× objective and DFC7000T camera. Averages were obtained by measuring the widths of the intact vessel in experimental animals and the dominant vessel in controls at 500 µm intervals across a total length of 3 mm. Changes to the thickness of condensed osteogenic mesoderm in E16.5 embryos was measured at the dorsal midline on consecutive 20 µm sections. Dura measurements were performed by setting a defined region of interest (ROI) centered over the dorsal midline on consecutive 20 µm sections. The area corresponding to Crabp2⁺/connexin 43⁻ staining was traced (as shown in Fig. 1) and calculated as the percentage area covered within the defined ROI. The values represent the average percentage area covered from two consecutive sections from five animals.

Animals were anesthetized using ketamine/xylazine (100 mg/kg). An incision was made on the midline of the head at the occipital crest. Once the skin was excised, curved forceps were used to break through the superficial connective tissue to reveal the underlying muscle. The muscle was carefully separated along the midline to expose the opening to the cisterna magna, and care was taken to not induce tears or bleeding. A 45 kDa ovalbumin-647 tracer (Molecular Probes, Invitrogen) was mixed in artificial cerebrospinal fluid (20 µg/ml solution), and 5 µl of solution was loaded into a 10 µl Hamilton syringe attached to polyethylene tubing and injected into the cerebrospinal fluid via the cisterna magna. The solution was injected using a 28-gauge needle and a Nanomite infusion system (Harvard Apparatus) with an injection rate of 2.5 µl/min. Animals were kept on supplemental oxygen throughout the duration of the experiment to stabilize breathing and minimize hypercapnia, and the needle was glued in place to prevent depressurization. After 30 min had passed, the needle was removed and the animals were immediately euthanized by transcardial perfusion.

Animals were perfused with 4% PFA and the superficial and deep cervical lymph nodes were dissected under fluorescence. The tissue was allowed to post-fix for 12 h in 2% PFA, prior to sinking in 30% sucrose and embedding in Neg-50 medium. Sections (20 µm) were cut from three control and three experimental animals. The sections were imaged using a Leica M165FC stereomicroscope equipped with a 1× objective and DFC7000T camera. Ten 20 µm sections per animal from the superficial and deep cervical lymph nodes were thresholded and analyzed by tracing out the tissue sections and calculating the percentage area covered. The values were averaged for each animal to obtain a single value representing the average percentage area covered for the superficial and deep cervical lymph nodes. Representative images were selected and imaged using a LSM800 confocal microscope with a 20×0.80 NA objective.

Statistics were performed using GraphPad Prism 9.0. For all analyses, unpaired t-tests with Welch's correction were performed. Sample sizes consist of at least three to five biological replicates (animals) for each analysis, and experimental animals were obtained from more than one litter and/or set of parents. No exclusion criteria was applied. To the extent possible, investigators were blinded to genotype. Error bars indicate s.e.m. Box and whisker plots show, at their centers, the median (horizontal line) and the 25th and 75th percentiles (boxed region). The whiskers denote the minimum and maximum data points for each set, and all values that fall below or above the 25th and 75th percentile, respectively.

The authors thank Young Kwon-Hong (University of Southern California) for providing Prox1^tdTomato mice; the Rutgers Molecular Imaging Center (D. Adler and P. Buckendahl) for assistance with skull 3D x-ray microscopy using the Skyscan 1272 micro-CT scanner (funding by NSF Major Research Instrumentation Award 1828332); Michael Falen and Kush Desai for assistance with mouse dissections and immunohistochemistry; and Marianne Polunas and the Rutgers Research Pathology Services Core for assistance with embryo embedding and Hematoxylin/Eosin slide preparation.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200065