Fluid flow as a driver of embryonic morphogenesis

Much effort has been made to understand how fluid flow and its accompanying physical forces affect embryonic development. The activation of the right molecular pathway at the right time and place drives correct embryonic development, a process partly dependent on fluid flow. Fluid flow can create a morphogen gradient, driving important morphogenic changes during embryonic development. On the other hand, the physical forces created by fluid flow can be translated into a biological response by mechanosensitive cell types, which activates mechanosensitive pathways (summarised in Box 1 and Tables 1 and 2).

Even one century ago, it was clear that fluid flow affects embryonic development. The first such publication regarded the role of flow in cardiovascular development, where it was shown that removing the heart from a chick embryo impaired normal vascular development (Chapman, 1918). However, until the advent of modern imaging techniques and genetics (Table 3), these findings were largely observational. By now it is clear that fluid flow is not limited to cardiovascular development, but affects every organ system that is filled with fluids at some point in development. Recent work has shed further light on the contributions of fluid flow to embryonic development, at the tissue, cellular and molecular level. In this Primer, we describe the importance of fluid flow during early, cardiovascular and neurological embryonic development, while highlighting the mechanosensitive signalling pathways involved.

Fluid flow is either laminar or turbulent (Fig. 1A). In laminar flow, adjacent layers of fluid (lamina) slide along one another without mixing. Each concentric layer reduces the friction on the wall for the next most inner layer, creating a parabolic flow profile. Laminar flow can be classified as steady, pulsatile or oscillatory (Fig. 1B). In steady flow patterns, there is a constant unidirectional flow present. In pulsatile flow patterns, the velocity of the laminar flow changes over time but remains unidirectional, whereas in oscillatory flow, backwards flow can also be observed. Turbulent flow is characterised as being irregular, with a chaotic mixing among the layers, which blunts the flow profile. Both oscillatory flow and turbulent flow are considered forms of disturbed flow.

Fluid flow creates a frictional force against the wall of the lumen, known as wall shear stress. In this case, ‘stress’ refers to physics terms for a force per unit area, and not a biological ‘stress’, such as that created by reactive oxygen species. Shear stress depends on the shear rate and the viscosity of the fluid. The rate of change of velocity defines the shear rate, with flow closer to the wall being slower than flow further away. Viscosity is the resistance of the flowing fluid; whole blood is more viscous than plasma, and thus shear stresses are greater for whole blood than for plasma.

Perhaps the earliest influences of fluid flow can be observed in the establishment of left-right asymmetry shortly after gastrulation, although how fluid flow controls symmetry breaking is still highly debated. Left-right asymmetry is necessary for the development of mature organ shapes and the proper positioning of the organs (Blum and Ott, 2018). Left-right asymmetry has been extensively studied in chick embryos, where it is characterised by distinct gene expression patterns on each side of the embryo, with Nodal, Lefty and Pitx2 expression on the left side of the lateral plate mesoderm, and Cerberus expression on the right (Fig. 2A) (Levin et al., 1995; Logan et al., 1998). In particular, left-sided expression of Pitx2 is crucial for left-right patterning, as alterations of Pitx2 expression in chick and Xenopus embryos affect the direction of body rotation, and Pitx2^−/− mouse embryos present with laterality defects (Ryan et al., 1998).

Breaking of lateral symmetry is suspected to be initiated by cilia-driven fluid flow in the vertebrate embryo. Motile cilia are located in the left-right organiser (LRO) of the embryo. For mouse embryos, this is located on the ventral surface of the node; for zebrafish, on the inner surface of Kupffer's vesicle (Nonaka et al., 1998, 2002; Okabe et al., 2008; Okada et al., 2005). In mouse and Xenopus embryos, motile cilia are positioned at the posterior side of the LRO cells by the planar cell polarity (PCP) pathway, creating a posterior tilt of the cilium (Nonaka et al., 2005). In mouse embryos, this process is additionally driven by posterior Wnt5a/b expression patterns, as polarisation of the node cells is impaired in Wnt5a^−/− and Wnt5b^−/− embryos (Minegishi et al., 2017). The created posterior tilt is crucial for generating an effective leftward ciliary stroke, resulting in a right-to-left flow of the extracellular fluids (Nonaka et al., 2005). Computational analysis of cilia-driven flow has shown that, without this tilt, vertical cilia would create small recirculating flows around each cilia, rather than a directional flow across the whole node (Cartwright et al., 2004). The overall distribution of cilia is also biased, with a higher density of cilia in the anterior region of Kupffer's vesicle, which creates a faster fluid flow (Ferreira et al., 2017; Wang et al., 2011a). The PCP pathway (driven by Wnt signalling) further specifies the posterior tilt of motile cilia through Vangl2 in zebrafish embryos (Borovina et al., 2010; Song et al., 2010). Indeed, cilia are dispersed randomly in Vangl1/2^−/− zebrafish embryos and nodal flow is turbulent, resulting in disruption of left-right patterning (Borovina et al., 2010; Song et al., 2010).

There are two proposed mechanisms for how interstitial fluid flow initiates left-right patterning, either through mechanosensation (Fig. 2B) or by creating a morphogen gradient (Fig. 2C). A fraction of cilia are immotile and it has been suggested that the immotile cilia sense flow (McGrath et al., 2003; Sampaio et al., 2014). Crown cells in the node of mouse embryos and Kupffer's vesicle in zebrafish have immotile cilium equipped with mechanosensitive Ca²⁺ channels, such as transient receptor potential polycystin 2 (Trpp, previously known as polycystin 2 or Pkd2) (McGrath et al., 2003; Schottenfeld et al., 2007; Yoshiba et al., 2012) (Fig. 2B). The activation of Trpp2 induces Ca²⁺ oscillations in the cilium, resulting in leftward Nodal expression and the initiation of vertebrate left-right asymmetry (Yuan et al., 2015). In the zebrafish embryo, the increased cytoplasmic Ca²⁺ levels block Cerberus expression, so that Nodal is expressed on the left side of the zebrafish embryo (Piccolo et al., 1999). Xenopus embryos lacking trpp2 fail to develop left-side oriented Nodal expression (Vick et al., 2018). However, it remains unclear how the primary cilium senses the nodal flow because, in mouse and Xenopus embryos, for example, the direction of nodal flow defines left-right patterning. It has therefore been hypothesised that primary cilia either detect an asymmetric flow velocity (Sampaio et al., 2014) or detect the specific direction of the flow (McGrath et al., 2003), or that motile cilia sense their own movement by amplifying the asymmetry of the LRO (Cartwright et al., 2020; Ferreira et al., 2018, 2019). But it remains unclear whether primary cilia can sense the low flow present in the LRO or whether they can distinguish the direction of flow (Cartwright et al., 2020; Ferreira et al., 2019).

As mentioned, rather than sensing flow, a chemo-sensory mechanism of left-right patterning has also been proposed. Here, cilia-driven fluid flow distributes nodal vesicular particles containing retinoic acid and sonic hedgehog towards the left side of the embryo, creating a morphogen gradient (Ferreira et al., 2017; Okada et al., 2005; Tanaka et al., 2005) (Fig. 2C). However, mathematical models suggest that nodal flow rates are insufficient to transport large vesicles and are simply too strong to create a gradient of small proteins, instead creating a symmetric distribution (Omori et al., 2018).

The presence and necessity of nodal flow remains unclear in some species, such as pig and chick embryos. In the chick embryo, cells of the Hensen's node rearrange themselves asymmetrically, creating a leftward movement of the cells around the node (Gros et al., 2009). The relative displacement establishes an asymmetric gene expression pattern of Shh and Fgf8, and initiates left-right patterning (Gros et al., 2009). Although differences occur, left-right patterning is well conserved between vertebrate species.

Although the initial specification of the cardiovascular system is determined by genetic cues, physical forces created by blood flow are crucial for its proper maturation as soon as the heart begins to beat (Jones et al., 2006). Mechanotransduction, the sensing of physical forces and their translation into a biological response, is an inherent function of endothelial cells (Jones, 2011). To enable proper mechanotransduction, endothelial cells are equipped with various mechanosensors that activate a number of signalling pathways (summarised in Box 1). During cardiovascular development, physical forces sensed by endothelial cells drive proper heart morphogenesis, vascular remodelling, arterial-venous specification, lymphatic valve formation and the development of haematopoietic stem cells (HSCs).

In mice, the primary heart tube starts rhythmically contracting at E8.25, which initiates blood flow through the vasculature of the yolk sac and embryo proper (Nishii and Shibata, 2006). This blood flow is crucial for sculpting the heart and great vessels, the dorsal aortae and cardinal vein. As such, disrupting flow within the developing heart induces both cardiac and vascular malformations (Hove et al., 2003; Hu et al., 2009).

The development of heart and the vessels are inherently linked, as exemplified by the development of the aortic arch (Fig. 3). Initially, six pharyngeal arches are present that experience the same flow pattern. During cardiac looping, the outflow tract of the primary heart tube shifts to the right and takes on a C-shape, introducing a more complex geometry through which blood flows (Taber, 2006) (Fig. 3A). As a consequence, the pharyngeal arch that experiences the highest flow grows in diameter, while the others regress. In the chick embryo, physical disruption of blood flow by ligation of the vitelline vein or left atrium disrupts aortic arch selection by altering blood flow patterns in the pharyngeal arches (Hu et al., 2009; Kowalski et al., 2012). In the mouse embryo, PITX2 drives the rotation of the outflow tract during heart looping; loss of PITX2 results in randomised heart looping and randomised laterality or even doubling of the aortic arch (Yashiro et al., 2007) (Fig. 3B).

Shear stress in the heart modulates gene expression of endocardial cells: the endothelial cells lining the inside of the heart. The embryonic heart initially grows by trabeculation: the formation of columns of muscular tissue within the ventricles (Wittig and Münsterberg, 2016). The endocardium communicates with the underlying myocardium (heart muscle tissue), such that if the endocardium is absent, trabeculation does not occur even though blood flow is normal (Peshkovsky et al., 2011). In zebrafish embryos, the primary cilium of endocardial cells facilitates shear stress-induced klf2a expression in the endocardium and subsequent notch1b expression (Samsa et al., 2015). In mouse embryos, mechanosensation by endocardial primary cilia induces myocardial compaction (Slough et al., 2008). The shear stress-driven klf2a expression of the endocardium is necessary for compaction of the trabeculae into denser heart tissue and for myocardial wall integrity (Dietrich et al., 2014; Rasouli et al., 2018). The absence of myocardial compaction and wall integrity in klf2a mutant zebrafish embryos can be rescued by specific overexpression of klf2a in the endocardium, demonstrating the tight communication between both (Dietrich et al., 2014; Rasouli et al., 2018). This endocardial-myocardial signalling cascade is also activated by altered haemodynamic forces created by cardiac injury in zebrafish (Gálvez-Santisteban et al., 2019). Haemodynamic activation of notch1b induces erbb2 expression and activates BMP signalling, which in turn promotes cardiomyocyte reprogramming and heart regeneration (Gálvez-Santisteban et al., 2019).

Blood flow through the ventricle coordinates growth and maturation of the myocardial cells lining the ventricle. Indeed, weak atrium (wea) mutant zebrafish have reduced blood flow without loss of ventricular contraction, resulting in reduced myocardial cell size and myofibril content (Lin et al., 2012). Furthermore, the elongation and orientation of cardiomyocytes in the outer aortic curvature is driven by blood flow and partially facilitates heart looping (Auman et al., 2007). As flow-driven elongation is lost, wea mutants have reduced curvatures and more cuboidal cell morphology in the outer curvature of the aorta (Auman et al., 2007).

During cardiac valve formation, oscillatory flow is necessary for proper valve development. The zebrafish heart has two chambers with three sets of valves: the outflow tract (OFT), the atrioventricular canal (AVC) and the inflow tract (IFT) valve (Fig. 4). As the heart takes shape, endocardial cells converge into the AVC to contribute to the cardiac valves, a migration driven by oscillatory shear stress (OSS) (Boselli et al., 2017). High OSS levels in the AVC of zebrafish embryos open mechanosensitive Trpp2 and Trpv4 (transient receptor potential cation channel subfamily V member 4) channels, which induce klf2a expression (Heckel et al., 2015; Vermot et al., 2009). Furthermore, Piezo1 Ca²⁺ channels in endocardial cells are involved in OFT valve development and also induce klf2a expression (Duchemin et al., 2019). Haemodynamic endocardial klf2a activation drives different processes during cardiac valve formation. In the AVC and OFT, endocardial klf2a expression drives notch1b expression and the subsequent deposition of fibronectin, which is necessary for the rearrangement of cardiac valve leaflet cells (Duchemin et al., 2019; Steed et al., 2016; Vermot et al., 2009) (Fig. 4A,B). Both in the OFT and AVC region, endocardial klf2a activation induces wnt9b expression, which stimulates the cell proliferation and subsequent condensation of the remodelling cardiac cushion in underlying mesenchymal cells (Goddard et al., 2017) (Fig. 4A,B). As the OFT is surrounded by smooth muscle progenitors, endocardial klf2a activation ensures Yap1 nuclearisation, although the necessity of Yap1 remains unknown (Duchemin et al., 2019) (Fig. 4B). Besides klf2a, OSS also induces miR-21 expression in the AVC, which prevents spry2 expression, an inhibitor of the MAPK signalling pathway. With spry2 expression inhibited, MAPK signalling is active, driving the proliferation of valve-forming cells and inhibiting notch1b signalling (Banjo et al., 2013). As such, OSS both upregulates and inhibits notch1b expression depending on the pathway, possibly implying feedback loops that operate on different time scales.

Genetic cues initiate arterial and venous cell fate in endothelial cells that form the primitive vascular plexus. In early zebrafish vasculature, presumptive arterial endothelial cells express efnb2, notch1b, notch3/5 and gridlock (grl), and the expression of venous marker flt4 is already restricted to the posterior cardinal vein (Lawson et al., 2001; Zhong et al., 2000). In the early mammalian vasculature, Cx37 (Gja4), Dll4 and Hey1 are expressed in arteries before the onset of flow, whereas there are no genes specifically restricted to the venous vessels (Chong et al., 2011). However, the arterial-venous identity of endothelial cells is susceptible to modification by fluid flow. In chick embryos, transplanting arterial endothelial cells into a vein, or vice versa, adapts them to the new environment (Moyon et al., 2001). In zebrafish, the number of arterial and venous intersegmental vessels is pre-defined by heterogeneous Notch activation, but the balance between arterial and venous intersegmental vessels is fine-tuned by shear stress (Geudens et al., 2019). After connecting the intersegmental vessels to venous circulation, venous endothelial cells migrate against flow to replace arterial endothelial cells of the intersegmental vessel (Weijts et al., 2018). In presumptive arterial intersegmental vessels with higher flow, flow-induced Notch signalling prevents this venularisation (Weijts et al., 2018). Similarly, in mice, blood flow is necessary for arterial-specific Notch1 expression (Jahnsen et al., 2015).

A primitive vascular plexus forms at E7.0 in mice and consists of capillaries lacking mural cell support or an extensive basement membrane. The vasculature is subsequently remodelled to efficiently supply the growing embryo. Shortly after the first irregular primitive heart tube contractions, erythrocytes enter circulation and the subsequent increase in viscosity initiates vascular remodelling (Al-Roubaie et al., 2011; Lucitti et al., 2007; McGrath et al., 2003). When erythrocytes are blocked from entering circulation (e.g. by trapping them in the blood islands), shear stress levels do not increase and the yolk sac vasculature does not remodel (Lucitti et al., 2007). The same phenotype is observed when haematopoiesis is impaired and red blood cells do not differentiate in the blood islands (e.g. mutations in Rapgef2 or Myc) (He et al., 2008; Satyanarayana et al., 2010). However, injecting a starch solution, which is viscous and therefore restores blood viscosity, rescues vascular remodelling, suggesting that it is shear stress that initiates vascular remodelling and not erythrocytes themselves (Lucitti et al., 2007).

Although shear stress is well acknowledged to drive remodelling, the role of mechanotransduction is less straightforward, as embryos still remodel surprisingly well even in the absence of certain mechanosensors or mechanosensitive pathways. Loss of Klf2, one of the principal transcription factors driving mechanotransduction, is embryonic lethal but only after initial remodelling has occurred (Kuo et al., 1997). Similarly, endothelial-specific Yap ablation does not prevent remodelling between E8.5 and E9.5 but leads to later embryonic lethality (Zhang et al., 2014). Similarly, impairment of some mechanosensors results in remodelling defects but impairment of others does not. Endothelial-specific deletion of Ift88 is non-lethal (Dinsmore and Reiter, 2016), whereas endothelial ablation of Piezo1 does prevent remodelling, although differences in genetic background appear to allow survival of the embryo in the absence of the protein, likely due to differences in the stage at which the protein is first deleted (Li et al., 2014a; Nonomura et al., 2018). Overall, these conflicting results indicate that there is a lot of redundancy in mechanotransduction, such that deleting any individual pathway or mechanosensor cannot prevent remodelling as a whole.

Flow also influences vascular remodelling, which includes sprouting angiogenesis, intussusceptive angiogenesis, and regression and fusion of vessels (Fig. 5). During sprouting angiogenesis, VEGF gradients initiate vascular sprouts from existing vessels (Gerhardt et al., 2003). However, flow dynamics fine-tune the VEGF-driven sprouting process, with sprouts forming exclusively at shear stress minima (Ghaffari et al., 2015; Song and Munn, 2011) (Fig. 5A). How these shear stress minima are sensed is not known, but may involve cilia; primary cilia are equipped with Trpp2 Ca²⁺ channels and loss of Trpp2 in zebrafish reduces the number of tip cells in the developing vasculature (Goetz et al., 2014). Although VEGF initiates sprouting, VEGF distribution is not simply defined by passive diffusion, but is determined by interstitial flow (Ghaffari et al., 2017). Increased interstitial flow in the direction of sprouting enhances sprouting and sprout extension follows the interstitial flow patterns (Akbari et al., 2019; Ghaffari et al., 2017; Kim et al., 2016). Furthermore, sprout elongation follows the interstitial flow patterns and the speed of elongation is proportional to the magnitude of interstitial flow (Ghaffari et al., 2017).

New vessels can also be formed by intussusceptive angiogenesis, when a vessel is split longitudinally. Here, endothelial cells form pillars that grow across the vessel lumen (Burri and Djonov, 2002). These pillars are invaded by fibroblasts and pericytes, until eventually the vessel splits (Burri and Djonov, 2002). High shear stress levels facilitate rapid pillar growth, although pillars fuse to each other in low shear stress regions (Djonov et al., 2002; Karthik et al., 2018) (Fig. 5B).

The developing vasculature produces an excess of vessels that are then pruned by selective regression. Vessels experiencing sufficiently high shear stresses are maintained, whereas those unable to meet this threshold regress (Chen et al., 2012; de Jong et al., 2012; Kochhan et al., 2013; Korn and Augustin, 2015). In well-perfused vessels, endothelial cells elongate and align with flow; their Golgi apparatus is located upstream of the nucleus, polarising the cell against the flow (Franco et al., 2015). During vessel regression in low flow vessels, axial polarity is misaligned, after which these endothelial cells actively migrate out of the regressing vessel segment to contribute to higher flow segments (Chen et al., 2012; Franco et al., 2015; Udan et al., 2013) (Fig. 5C).

Vascular fusion, or reverse-intussusception, is a major method of vessel enlargement during development (Chouinard-Pelletier et al., 2013; Udan et al., 2013). During fusion, smaller vessels merge to share one lumen, thereby increasing the diameter of a vessel rapidly. Vascular fusion occurs in regions of high flow but is driven by low shear stress levels (Chouinard-Pelletier et al., 2013). Avascular pillars that disturb the flow stream create a low shear stress downstream of the avascular region. VE-Cadherin phosphorylation occurs in these low shear stress regions and rearranges the junctional complexes of endothelial cells, allowing two nearby vessels to fuse (Caolo et al., 2018) (Fig. 5D).

In the final phase of vascular remodelling, smooth muscle cells and pericytes are recruited to stabilise the vasculature, a process facilitated by a number of mechanosensitive pathways. The G protein-coupled S1P receptor 1 (S1PR1) governs pericyte and smooth muscle cell recruitment during embryonic development and is activated by laminar shear stress (LSS) (Gaengel et al., 2009; Jung et al., 2012). In the absence of S1P1, flow-mediated signalling is reduced in endothelial cells in vitro, and S1p1^−/− embryos present with vascular leakage and dysregulated blood flow (Jung et al., 2012). Furthermore, in zebrafish embryos, shear stress sensed through primary cilia activates the Notch signalling pathway in arterial endothelial cells and promotes Foxc1b expression (Chen et al., 2017). Foxc1b drives vascular smooth muscle cell recruitment by activating mesodermal cells present in close proximity to arterial-fated vessels (Chen et al., 2017). After recruitment of pericytes, interaction between endothelial cells and pericytes is stimulated by flow-mediated miR-27b, which inhibits semaphorin expression and accompanied repulsive signals (Demolli et al., 2017).

Besides shear stress, blood flow also creates a circumferential stretch, induced by the beating heart. Stretch requires a vessel to be elastic. Hydrostatic pressure, the absolute level of pressure, is not extremely biologically active. Rather, the biologically stronger signal is the expansion and contraction that occurs as elastic vessels are exposed to pulse pressure. Much less is known about the role of stretch compared with shear stress, and in vivo, it is extremely difficult to separate the two. Several mechanosensitive Ca²⁺ channels, including Piezo1 and Trpp2, are stretch-activated channels. However, shear stress also creates tension on endothelial cell membranes, similar to the effects of stretch. This makes it difficult to assess whether the major role of these channels in vivo relates to sensing shear stress or stretch (Melchior and Frangos, 2010; Yamamoto and Ando, 2013).

Between E6.5 and E9.0 of mouse development, the first blood cells (both erythroid and myeloid cells) form in the extra-embryonic yolk sac (Kasaai et al., 2017; Samokhvalov et al., 2007; Tanaka et al., 2012; Yoder et al., 1997). Later, HSCs emerge from the aorta-gonad-mesonephros (AGM) region, before colonising liver and finally bone marrow (Medvinsky and Dzierzak, 1996; Müller et al., 1994; Yokomizo and Dzierzak, 2010). Time-lapse imaging has demonstrated that HSCs can derive directly from specialised aortic endothelium called the ‘haemogenic endothelium’ (Lam et al., 2010; reviewed by Adamo and García-Cardeña, 2012). Given their vascular localisation, haemogenic endothelium and differentiating and emerging HSCs are exposed to the same forces of flow as vascular endothelial cells.

Shear stress-driven nitric oxide (NO) production is crucial for the formation and release of HSCs into the circulation of the embryo. In the absence of blood flow, there is a severe reduction of circulating HSCs in the zebrafish embryo (North et al., 2009). Shear stress activates klf2a expression in haemogenic endothelium, where it promotes the expression of nos1/2b, thus increasing NO production (Adamo et al., 2009; Wang et al., 2011b). If NO production is blocked, fewer haemogenic cells are present in the AGM region at E10.5, even if blood flow remains (Adamo et al., 2009). Similarly, in the zebrafish embryo, either the loss of klf2a expression or loss of heartbeat results in reduced expression of Runx1, an important regulator of haematopoiesis, a phenotype rescued by external NO delivery (North et al., 2009; Wang et al., 2011b). Shear stress also induces the expression of RUNX1 and Yap1 nuclearisation in induced pluripotent stem cell-derived haemogenic endothelial cells in vitro (Lundin et al., 2020). Moreover, in zebrafish embryos, haemodynamic-driven Yap1 nuclearisation supports HSC specification and maturation, independent of the usual hippo signalling activation (Lundin et al., 2020; Nakajima et al., 2017).

Lymphatic endothelial cells (LECs) that line lymphatic vessels are highly sensitive to the low levels of shear stress created by lymph flow (Baeyens et al., 2015). VEGFR3, a component of the junctional mechanosensory complex is abundantly expressed on LECs and specifically tunes the high sensitivity of LECs to the low shear stresses of lymph flow (Baeyens et al., 2015). The mechanosensitive protein FOXC2 and its upstream regulators β-catenin and GATA2 are necessary for the stabilisation of the developing lymphatic vasculature (Cha and Srinivasan, 2016; Sabine et al., 2015; Sweet et al., 2015). Furthermore, increased LSS induces sprouting in LECs, likely reflecting the need for increased draining in high lymph flow regions (Choi et al., 2017).

Lymph flow also drives lymphatic valve formation as lymphatic valves are absent in mouse embryos that lack lymph flow due to loss of C-type lectin-like receptor 2 (Clec2), where lymph flow is opposed by the backflow of blood (Sweet et al., 2015). Lymphatic valves form in areas of maximal OSS, where Prox1 and subsequently Foxc2 expression are highest. Here, OSS induces Wnt/β-catenin signalling, which in turn controls Foxc2 expression (Sabine et al., 2012). As Foxc2 has a slightly different expression pattern than Prox1, the up- and downstream valve leaflet surfaces become distinct (Cha et al., 2016; Sabine et al., 2012) (Fig. 6B). Moreover, Prox1 and Foxc2 together modulate Cx37 expression, which in turn regulates Nfatc1 signalling, a crucial pathway for the maturation and stabilisation of lymphatic valves (Sabine et al., 2012) (Fig. 6C). As the valves mature, Cx37 expression is driven particularly by the recirculating flow downstream of elongating leaflets of the lymphatic valves (Sabine et al., 2012), which further stimulates the differentiation of lymphatic valves. Endothelial-specific deletion of Piezo1 impairs lymphatic valve protrusion (Nonomura et al., 2018). However, loss of Piezo1 does not disrupt expression of Prox1, Foxc2 or Nfatc1 at valve forming regions, suggesting that Piezo1 functions in a separate molecular pathway, likely related to cytoskeletal re-organisation (Nonomura et al., 2018).

Lymph flow is necessary for the expansion and maturation of the lymph node. Primordial lymph nodes form by E16.5 (Blum and Pabst, 2006; van de Pavert and Mebius, 2010). Interstitial lymph flows through the developing lymph node and amplifies Cxcl13 expression, promotes the retention of inducer cells and is necessary for the formation of the subcapsular sinus, as well as for recruitment of macrophages to the sinus (Bovay et al., 2018). In the absence of FOXC2, lymphatic valves fail to form, so that unidirectional lymphatic flow does not occur (Petrova et al., 2004). Instead, lymphatic flow remains bidirectional, preventing lymph node maturation as inducer cells are not retained within the forming lymph node (Bovay et al., 2018).

During early development, the neural tube closes and bends to form three fluid-filled cavities: the X, Y and telencephalic ventricles (Lowery and Sive, 2009). In amphibians and mammals, the telencephalic ventricle further matures into two lateral ventricles to form four ventricles in the adult brain (Lowery and Sive, 2009). During early embryonic development, these cavities are already filled with cerebrospinal fluid (CSF), secreted by the choroid plexuses, after which it enters the spinal cord. Here, we present the key concepts in CSF flow, focussing on development (extensively reviewed by Fame and Lehtinen, 2020; Ringers et al., 2020).

CSF flow in the brain ventricles consists of a macrofluidic component, which makes up the bulk of flow in the middle of the ventricles, and a microfluidic component close to the ventricle wall (Ringers et al., 2020; Siyahhan et al., 2014). The major bulk of flow is driven by a number of cues, such as the pressure gradient created by CSF secretion, exchange of CSF with interstitial fluid, cardiac and respiratory cycles, and even body movements, which therefore make CSF flow difficult to study (Dreha-Kulaczewski et al., 2015; Feinberg and Mark, 1987; Olstad et al., 2019; Ringers et al., 2020; Spassky et al., 2005; Xu et al., 2016). The bulk CSF flow is primarily thought to nourish the brain and maintain brain homeostasis (Ringers et al., 2020).

Microfluidic flow close to the ventricle wall has been studied more extensively during development and is at least partly supported by motile cilia present on ependymal cells lining the ventricle, as demonstrated in zebrafish (Grimes et al., 2016; Olstad et al., 2019), mice (Abdelhamed et al., 2018; Banizs et al., 2005; Ibañez-Tallon et al., 2004), pigs (Faubel et al., 2016) and humans (Worthington and Cathcart, 1966). In both zebrafish and mouse, insufficient CSF flow due to impaired ciliary motility, as well as deficient mechanosensation, induces ventricular defects and eventually hydrocephalus: an enlargement of the brain ventricles (Banizs et al., 2005; Grimes et al., 2016; Ibañez-Tallon et al., 2004; Kramer-Zucker et al., 2005). For example, the dynein-complex driving cilia motion contains Dnah5; Dnah5^−/− mice lack CSF flow and develop hydrocephalus during early postnatal development (Ibañez-Tallon, 2002; Ibañez-Tallon et al., 2004). Similarly, mutations disrupting intraflagellar transport, such as Ift88^−/− and Ift57^−/− in mouse embryos, affect proper formation of both motile and mechanosensory cilia, causing hydrocephalus (Banizs et al., 2005; Kramer-Zucker et al., 2005). However, individuals with primary ciliary dyskinesia barely present with hydrocephalus (Lee, 2013), which suggests apparent differences between species. These differences could possibly be explained by the difference in ventricular size, which could change the contribution of bulk CSF flow compared with near-wall CSF flow (Ringers et al., 2020).

Radial glia cells are neuronal stem cells in the subventricular zone, which is located beneath the ependyma of the lateral ventricle, that give rise to neurons, glia cells and ependymal cells. Ependymal cells arise from radial glia (Spassky et al., 2005). Ependymal cells are multi-ciliated cells that drive CSF flow and, as with the LRO, the polarisation of these cells is important for their function. Translational polarity is defined by the position of the basal bodies on the apical surface of the cell, while rotational polarity is defined by the angle of individual basal bodies (Mirzadeh et al., 2010). Before ependymal cells arise, radial glia cells have already initiated the well-defined ependymal PCP in mice through mechanosensation (Mirzadeh et al., 2010; Ohata et al., 2016) (Fig. 7). The mechanosensors Trpp1 and Trpp2 on the primary cilia sense CSF flow in the embryonic ventricles, driving translational polarity in mice (Guirao et al., 2010; Mirzadeh et al., 2010; Ohata et al., 2015). This translational polarity is preserved when they undergo differentiation into ependymal cells postnatally, which is accompanied by extensive ciliogenesis. Ependymal cells have multiple motile cilia, which gradually align with CSF flow (Guirao et al., 2010). This rotational polarity is regulated through Vangl2 and a positive-feedback loop of CSF flow, as evidenced from cultured ependymal cells, although it remains unclear how these motile cilia sense CSF flow (Guirao et al., 2010; Mitchell et al., 2007; Ohata et al., 2016). Furthermore, Trpp1 and Trpp2 govern a balance between proliferation and differentiation of radial glia cells in mice (Winokurow and Schumacher, 2019). Loss of Trpp1 or Trpp2 stimulates increased proliferation of radial glia cells through Notch signalling and Stat3, while impairing their differentiation to neurons (Winokurow and Schumacher, 2019).

Besides direct mechanosensory effects, CSF flow has long been thought to transport molecules and fine-tune the distribution of molecules in the lateral ventricles and the spinal cord. For example, bulk CSF flow guides the migration of newly formed neuroblasts in the adult mouse brain by creating a gradient of chemo-repulsive molecules, such as Slit proteins (Sawamoto et al., 2006). Similarly, Wnt5a is secreted by the choroid plexuses and is guided by CSF flow towards neural progenitors, supporting cerebellar development (Kaiser et al., 2019). Moreover, when ciliary motility is impaired either in brain or spinal cord, none of the beads injected in the ventricles is transported into the central canal of the spinal cord, suggesting that ciliary motility controls long-range CSF transport (Thouvenin et al., 2020).

In the spinal cord, motile cilia are present on floor plate cells, ependymal cells that line the central canal, and CSF-contacting neurons. Motile cilia in the spinal cord have been discovered in a number of species, including zebrafish, chick, mice, rats and humans, but have been most extensively studied in zebrafish.

In 24-30 h post fertilisation (hpf) zebrafish embryos, CSF flows bidirectionally through the spinal cord (Cantaut-Belarif et al., 2018; Sternberg et al., 2018; Thouvenin et al., 2020). Motile cilia primarily line the ventral wall of the central canal of zebrafish embryos, allowing for an asymmetric distribution and, consequently, a bidirectional CSF flow (Thouvenin et al., 2020). This bidirectional CSF flow is enhanced by contractions of the body wall muscles (Olstad et al., 2019; Thouvenin et al., 2020). Additionally, proper ciliary movement drives the formation of the lumen of the central canal, as zebrafish mutants with impaired ciliary motility present with a reduced diameter of the central canal (Thouvenin et al., 2020).

In zebrafish, impaired ciliary motility is often accompanied by improper straightening of the spine (Grimes et al., 2016; Sternberg et al., 2018; Zhang et al., 2018). Straightening of the spine is dependent on the formation of Reissner's fibre, a thread composed of subcommissural organ (SCO)-spondin proteins that are secreted into the central canal of the spinal cord by the SCO (Cantaut-Belarif et al., 2018). Both CSF flow and ciliary motility are necessary for the formation and secretion of Reissner's fibre (Cantaut-Belarif et al., 2018). CSF flow also transports adrenaline from the brain towards the spinal cord of zebrafish embryos, where it stimulates the release of urotensin neuropeptides Urp1 and Urp2 from CSF-contacting neurons (Zhang et al., 2018). The secreted neuropeptides activate slow-twitch muscle fibres and the subsequent contractions are hypothesised to straighten the spinal cord in zebrafish embryos (Sternberg et al., 2018; Zhang et al., 2018). Last, CSF-contacting neurons are equipped with the mechanosensory Trpp2l1 Ca²⁺ channel, which enables them to react directly to CSF flow (Sternberg et al., 2018). Either loss of Trpp2l1, impaired ciliary motility or obstructed urotensin neuropeptide secretion results in severe scoliosis of the zebrafish adult (Sternberg et al., 2018; Zhang et al., 2018). CSF flow and ciliary motility remain important during adulthood to ensure spine straightening, as scoliosis is observed even when motile cilia function is impaired at post-embryonic stages (Buchan et al., 2014; Grimes et al., 2016; Konjikusic et al., 2018). In humans, CSF flow is impaired in individuals with Chiari malformations that suffer from scoliosis and individuals with primary ciliary dyskinesia, who also present with scoliosis (Engesaeth et al., 1993; McGirt et al., 2008; Panigrahi et al., 2004).

Mechanosensitive signalling in response to flow elicits important structural adaptations in the developing embryo. However, mechanosensing is not limited to fluid flow exerted directly onto luminal cells. Other physical forces modulate early embryonic shape changes, later tissue folding events and even musculoskeletal development (Anlaş and Nelson, 2018; Arvind and Huang, 2017; Heer and Martin, 2017). Moreover, the importance of fluid flow is not limited to embryonic development, and is crucial to the progression of many diseases such as atherosclerosis (Souilhol et al., 2019). There is much to learn both from the similarities and the differences between organ systems and developmental stages regarding the role of flow in morphogenesis.