Cardiac forces regulate zebrafish heart valve delamination by modulating Nfat signaling

In the clinic, most cases of congenital heart valve defects are thought to arise through errors that occur after the endothelial–mesenchymal transition (EndoMT) stage of valve development. Although mechanical forces caused by heartbeat are essential modulators of cardiovascular development, their role in these later developmental events is poorly understood. To address this question, we used the zebrafish superior atrioventricular valve (AV) as a model. We found that cellularized cushions of the superior atrioventricular canal (AVC) morph into valve leaflets via mesenchymal–endothelial transition (MEndoT) and tissue sheet delamination. Defects in delamination result in thickened, hyperplastic valves, and reduced heart function. Mechanical, chemical, and genetic perturbation of cardiac forces showed that mechanical stimuli are important regulators of valve delamination. Mechanistically, we show that forces modulate Nfatc activity to control delamination. Together, our results establish the cellular and molecular signature of cardiac valve delamination in vivo and demonstrate the continuous regulatory role of mechanical forces and blood flow during valve formation.


Introduction
Heart valves are structures critical for ensuring unidirectional blood flow, and heart valve disease is a significant cause of illness and death worldwide. Given the intimate relationship between cardiac forces and cardiovascular development [1][2][3][4][5], a better understanding of how mechanical forces regulate heart valve morphogenesis can yield valuable insights into the origins of heart valve disease.
Heart valve development is initiated by the endothelial-mesenchymal transition (EndoMT) of a subset of endocardial cells, which migrate into the cardiac jelly (CJ) and subsequently a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 except near the base of the valve and near the edges of the lumen (Fig 1D-1D"'). Surprisingly, using one of several transgenic lines that label endocardial cells (Tg(fli1a:gal4ff;UAS:EGFP-CAAX), Tg(fli1a:gal4ff;UAS:Kaede), and Tg(fli1a:LifeAct-EGFP)), we found that valve leaflets are formed as early as 72 hpf (Fig 1E-1E""' and 1G, S3 Movie). To increase spatial resolution, we imaged Tg(fli1a:gal4ff;UAS:EGFP-CAAX) together with BODIPY-TR Ceramide, a red fluorescent dye, which counterstains the blood plasma and CJ. This allowed us to determine valve morphology at cellular resolution and perform cell counting. Approximately 11 cells migrate into the CJ by 65 hpf. At 72 hpf, valve leaflets were 2 cell layers thick except at the base of the leaflet, where 3 to 4 cells remain in the CJ and form the abluminal hinge cells of the leaflet ( Fig  1E-1E""' and 1H). Until 80 hpf, we find that valve leaflets tend to fold in stopped hearts such that the inner layer of valve endocardial cells is in contact with endocardial cells of the AVC wall ( Fig 1F-1F""', S2A-A""' and S2B-S2B"" ' Fig).
It has been proposed that valves transform from primitive structures into valve leaflets via the gradual process of elongation, involving abluminal cell rearrangement/VIC invasion, cell proliferation, and the thinning of luminal valve cells [23]. Given the tissue morphologies we observed, we hypothesized that primitive structures instead transition to leaflet morphology via tissue sheet delamination, where the bilayer of cells that have migrated into the CJ splits into 2 separate layers. The fact that we only see 2 main phenotypes-cellular cushions and valve leaflets-between 72 and 80 hpf suggests that this is a very fast process. We verified that delamination occurs and that it can take place within an hour via time-lapse imaging (3D + cardiac cycle + developmental time) of the beating heart ( Fig 1I-1I", S2C-C' Fig, S4 and S5 Movies).

Abluminal hinge cells give rise to VICs
Abluminal cells of the bilayer have been proposed to become future VICs [23,26]. We thus wondered if the few abluminal cells at the hinge of newly formed superior AV valve leaflets can give rise to the much larger population of endocardial-derived VICs seen at 98 hpf and 144 hpf. To address this, we performed photoconversion analysis [27], where we photoconverted valve progenitors at 48 hpf and imaged the beating heart at 80 hpf when most valves have gained leaflet morphology. As expected, we found that the endocardial cells leading the initial collective migration into the CJ form abluminal cells of the hinge, while cells that enter the CJ behind the leading cells form the AVC wall and the inner layer of the valve leaflet (S3A-S3A" ' Fig). We then repeated our photoconversion experiments but stopped hearts at 98 hpf (S3B Fig and 144 hpf (Fig 2A-2A"") observe the origin of VICs. We find that luminal cells of the valve leaflet remain luminal at later stages, suggesting that the few abluminal hinge cells proliferate at the hinge and their progeny migrates toward the distal part of the valve and give rise to all endocardial-derived VICs (S3C Fig, Fig 2B).

Characterization of valve EndoMT/MEndoT in zebrafish
Valve delamination requires most abluminal cells of the bilayer to become luminal again, suggesting that these cells undergo MEndoT. We thus sought to examine the cellular signature of valve delamination by examining various endothelial and mesenchymal cell markers.
We began by using immunohistochemistry to study the dynamics of VE-cadherin, the major endothelial adhesion molecule. [28] We found that VE-cadherin is expressed between all cell-cell interfaces between 48   To confirm that VE-cadherin is reexpressed at 80 hpf, we examined VE-cadherin expression using the transgenic line Tg(ve-cad:ve-cadTS) (Fig 3). The higher spatial resolution obtained here also reveals that VE-cadherin between luminal cells at the site of delamination is down-regulated between 65 and 80 hpf (Fig 3).
To examine tight junction dynamics, we performed immunostaining of zonula occludens-1 (ZO-1). We found that ZO-1 is expressed at cell-cell interfaces of all cells from 60 hpf through to 80 hpf (Fig 4). To confirm that tight junctions are present at 65 hpf when VE-cadherin is down-regulated in abluminal cells, some 65 hpf embryos were co-stained with the junctional protein Esama. We found that ZO-1 colocalized with Esama in abluminal cells (S5A Fig). Fibronectin deposition is considered a biomarker of EMT [29,30]. Previous studies have shown that fibronectin is expressed at the basal side of endocardial cells at 48 hpf [21] and surrounds early VICs at 120 hpf [23]. Using immunohistochemistry, we show that while fibronectin is consistently and strongly deposited along the basal side of AVC luminal cells from 48 hpf through to 98 hpf, the presence of fibronectin around other valve progenitors varies Photoconverted cells can be seen between the 2 luminal layers of the valve at 144 hpf. (A"') Embryos where the deepest cell inside the CJ and the atrial edge of the AVC are photoconverted. At 144 hpf, photoconverted cells can be seen between the 2 luminal layers of the valve, presumably derived from the cell that was deepest inside the CJ at 48 to 50 hpf. Luminal photoconverted cells can also be seen at the tip of the valve, presumably derived from the atrial side of the AVC at 48   Cyan arrow points to low VE-cadherin signal between the inner layer of the leaflet and the endocardial cells comprising the AVC wall. Orange arrows point to interfaces where VE-cadherin is up-regulated after being down-regulated following cell migration. Scale bar: 10 μm. AVC, atrioventricular canal; hpf, hours postfertilization. To examine whether valve cells lose and regain apical-basal polarity, we sought to examine the localization of podocalyxin, a CD34-related sialomucin protein that is commonly used as an apical cell surface marker in epithelial cells [31,32], including endothelial cells [33][34][35]. We thus generated a transgenic line where EGFP-tagged podocalyxin is expressed under the fli1a promoter, Tg(fli1a:EGFP-Podxl), and crossed it with Tg(fli1a:myr-mCherry), a transgenic line that labels endothelial cell membranes with mCherry, to see how podocalyxin localization changes between 48 and 80 hpf (S7 Fig). By analyzing z-slices corresponding to the middle half of the valve, we found that podocalyxin tends to be apically localized in endocardial cells of the These findings suggest that cells that migrate into the CJ lose apical-basal cell polarity and, except for hinge cells that remain in the CJ, regain it just prior to delamination.
We next analyzed the expression of the classical EMT markers snail1/2 and twist1 that are known to be important for heart valve formation in amniotes [36,37]. We examined the expression of snail1b and twist1b in hearts between 48 hpf and 80 hpf using RNAscope and found that both snail1b and twist1b are enriched at the AVC and outflow tract at all time points analyzed (Fig 5). Looking more closely at the superior AV valve region, snail1b is expressed in both luminal and abluminal endocardial cells to a similar extent (Fig 5A and 5B-5D). Twist1b is also expressed in both luminal and abluminal endocardial cells but shows greater differences in expression level depending on the valve region and is also expressed in the myocardial tongue. At 48 hpf, twist1b is expressed specifically in the region where we expect to find cells that are initiating migration into the CJ (Fig 5A), delimitating the cell pool that will eventually form the abluminal hinge cells of the valve. Between 55 and 65 hpf, twist1b is expressed at higher levels in abluminal cells (Fig 5A, 5E and 5F). At 80 hpf, when delamination is presumed to have occurred, twist1b expression is particularly high near the valve base, where we expect abluminal cells to reside (Fig 5A and 5G). As an alternative method to assess twist1b expression, we used the transgenic line TgBAC(twist1b:GFP) and imaged embryos at 75 hpf. Since GFP is stable, we were able to observe patterns of cumulative twist1b expression. In examining the AV valve forming region, we found that, consistent with our RNAscope analyses, twist1b is expressed in valve cells that have at an earlier point in development migrated into the CJ as well as cells of the myocardial tongue (S8 Fig).
Overall, we find that while cells transiently lose some endothelial characteristics/gain mesenchymal characteristics after migrating into the CJ, delamination comes with most cells regaining their endothelial characteristics. That ZO-1 remains expressed between these cells both during and after their migration into the CJ shows that cells do not become individualized, but rather stay in their highly organized layers as they remodel their adhesive contacts during delamination.

Abnormal cardiac forces can block or delay valve delamination
Considering mechanical forces are key modulators of valve morphogenesis [5,38,39], we next sought to better understand the role of cardiac forces on valve cell behavior underlying tissue delamination. We first assessed whether valve delamination is dependent on normal flow by performing 2 types of flow manipulation: The first involves surgically inserting an approximately 30-micron diameter bead into the ventricle of the heart and thereby occluding blood flow (S9A Fig), and the second involves transferring embryos to media with 15 mM BDM, a myosin II inhibitor, to the water. Because prolonged perturbation of heartbeat and blood flow causes the heart to unloop and the chambers to collapse, we limited the length of intervention to 4 hours, performing the interventions at 72 hpf and analyzing the effect of the intervention at 76 hpf.
Inserting a bead into the ventricle predictably causes a drastic change in the flow profile (S9B Fig, S6 Movie), as well as a reduction in heartrate (S9C Fig). In imaging the beating heart, we found that inserting a bead into the ventricle completely stops the delamination process (S9D and S9E Fig). We confirm our results by immunostaining bead-inserted and sham Given that bead injection and BDM treatment causes severe changes to cardiac forces and complete abolition of valve delamination, we searched for ways to manipulate flow more subtly. We, therefore, examined gata1 mutants where red blood formation is inhibited. A previous study approximating the heart as a linear tube has shown that the absence of red blood cells leads to a decrease in maximal wall shear stress of 2 to 3 times at 48 hpf [13]. Using a recently developed motion estimation algorithm [40], we were able to extract the position and motion of the cardiac wall (and blood cells in wild-type controls) in our movies of the beating heart and model wall shear stress in wild-type and gata1 mutants at 65 hpf. At this stage, when heartbeat is faster and stronger, we found that maximal wall shear stress at the AVC is approximately 3 to 4 times lower in gata1 mutants (S10 Fig).
Gata1 mutant hearts appear normal at 48 hpf and, unlike gata1 morphants, which have increased numbers of AVC endocardial cells [41], we found no differences in the number of AVC endocardial cells in gata1 mutants compared to controls (S11A and S11B Fig). Searching for possible delamination defects, we found that we could screen out the majority of gata1 mutants with early superior AV valve defects by removing embryos with pericardial edema at 65 hpf (S11C-S11H Fig). Imaging the beating heart of gata1 mutants in the Tg(fli1a:gal4ff; UAS:Kaede) background at 80 hpf, we found that only 17% of gata1 mutant superior AV valves manage to delaminate by 80 hpf compared to 92% of gata1 control superior AV valves (S12A and S12B Fig, S7 and S8 Movies). Superior AV valves that have failed to delaminate by 80 hpf were imaged again at 98 hpf (S12A and S12B Fig, S8 and S9 Movies). In doing so, we found that for the majority of mutants, delamination is delayed rather than completely prevented. Taken together with the bead injection and BDM experiments, these results show that perturbation of cardiac forces can block or delay delamination.

Abnormal hemodynamics causes delamination defects and thick valve phenotypes
In imaging the beating heart, we found that even delaminated gata1 mutant superior AV valves appear thick (S12A Fig). We confirm that the majority of gata1 mutant superior AV valves are thick by imaging stopped/fixed hearts at 98 hpf (Fig 6A and 6B). Segmenting gata1 mutant valves showed that they are also enlarged (Fig 6A and 6C), and cell counting showed that they are hyperplastic (Fig 6D). By performing fibronectin immunostaining, we found that extra abluminal cells in gata1 mutants express fibronectin, suggesting that these cells proceed to differentiate into VICs (Fig 6A, S10 and S11 Movies).
We hypothesized that the cellular origins of the thick valve phenotype could be due to abnormal delamination, where cells originally destined to become luminal cells end up remaining in the CJ. To test this, we immunostained 80 hpf gata1 mutants for Cdh5 and ZO-1. We found that while 75% gata1 controls reexpress Cdh5 and maintain expression of ZO-1, 100% of gata1 mutants do not properly reexpress Cdh5 (Fig 6E, 6F-6F' and 6G). In 41% of gata1 mutants, ZO-1 signal appears diffuse or is absent around abluminal cells, which lose regions shown in (A) and indicates where the measurement was performed. p-Values were determined using unpaired t test in (B, C, E, F) and using 1-way ANOVA in (D,G

PLOS BIOLOGY
Tissue delamination and mechanical forces during valve morphogenesis  Fig 6F' and 6G). This supports our hypothesis as it suggests that abluminal bilayer cells in gata1 mutants often fail to undergo MEndoT. We then performed photoconversion experiments, which confirmed that some valve cells in gata1 mutants are incorrectly assigned to abluminal cell fate (Fig 7, S13 Fig).
Misassignment of cell fate alone cannot completely account for hyperplasticity in the valve region in gata1 mutants ( Fig 6D). We thus wondered if cells misassigned from luminal to abluminal fate overproliferate. To test this, we first performed cell counting in 98 hpf embryos. We confirmed that while 98 hpf gata1 mutant valves have decreased numbers of cells in the inner layer of the valve and the AVC wall, this decrease cannot account for all of the additional VICs ( Fig 8A). We then performed EdU staining from 74 to 98 hpf to assess cell proliferation ( Fig 8B). We found that gata1 mutants valves have greater numbers of EdU + abluminal cells ( Fig 8C) and a greater fraction of valve cells that are EdU + (Fig 8D). By measuring the fraction of EdU + abluminal and luminal cells, we found that the difference in proliferation is not statistically different between abluminal cells in gata1 mutants and controls ( Fig 8E). However, abluminal cells were found to be more proliferative than luminal cells in both mutants and controls ( Fig 8E). Together, these results suggest that cells misassigned from the inner layer of the valve/AVC wall to the valve interstitial space in gata1 mutants become more proliferative, thereby exacerbating the overabundance of VICs and causing valve hyperplasticity ( Fig 7D).
To test our hypothesis that wall shear stresses regulate valve delamination, we injected a nanoemulsion into the bloodstream of gata1 mutants at 60 hpf (S14A Fig). This increased the blood viscosity of these mutants by approximately 17%, according to microfluidic rheometry measurements. We found that these nanoemulsion injections were able to partially rescue the gata1 phenotype-while we did not see a significant decrease in the total number of thick valves in injected embryos (S14B and S14C Fig), injected embryos had significantly fewer abluminal cells compared uninjected embryos (S14B and S14D Fig). Altogether, these results suggest that abnormal wall shear stresses in gata1 mutants could lead to delamination defects and thick, hyperplastic valves.

Thick valves are associated with decreased valve function
We next sought to see if thick valves could decrease valve function. We find that all gata1 mutants have pericardial edema by 98 hpf (S15A Fig), a phenotype commonly associated with poor heart function. However, gata1 mutants at 80 hpf also have decreased heartrate (S15B Fig), making it difficult to isolate the role of valve morphology alone. This, as well as the difficulties in assessing flow profiles in gata1 mutants, led us to examine embryos with mutations in the transcription factors klf2a and 2b, 2 genes that are well known to have redundant roles in the endocardial cell response to cardiac forces [42]. It has been previously shown that at 96 hpf, about half of klf2a morphants have abnormally thick superior or inferior AV valves, while the other half have defects suggestive of failed EndoMT [21]. It is known that klf2a klf2b double mutants (henceforth referred to as simply klf2 mutants in text) have increasing incidences of pericardial edema as they age from 2 dpf to 5 dpf [43]. Excluding approximately 50% of embryos that have pericardial edema at 65 hpf, we were able to screen out the majority of cadherin and ZO-1 at 80 hpf. (F) In 16/27 embryos, some abluminal cells appear to have reexpressed VE-cadherin normally (red arrows), while others are immunopositive for ZO-1 but fail to reexpress VE-cadherin (cyan arrows). (F') In 11/27 embryos, there is only residual ZO-1 signal in abluminal cells (purple arrows). In regions where ZO-1 signal is absent, gaps are formed between cells (red asterisks). (G) Graph showing quantification of Cdh5 and ZO-1 staining. p-Values are calculated using Fisher exact test for normal expression Cdd5 and ZO-1 expression pattern (Cdh5 +ve, ZO-1 +ve) versus abnormal expression Cdd5 and ZO-1 pattern (Cdh5 -ve, ZO-1 -ve; Cdh5 -ve, ZO-1 -ve). Scale bars: 20 μm. AVC, atrioventricular canal; FN, fibronectin; hpf, hours postfertilization; VIC, valve interstitial cell; ZO-1, zonula occludens-1.
We found that klf2 mutants have lower survival rates compared to controls (S15H Fig). Brightfield imaging of the beating heart at 500 frames per second allowed us to visualize the movement of red blood cells over the cardiac cycle and demonstrated that both klf2 mutants hearts were stopped using BDM and the ventricle and the ventricular side of the AVC were photoconverted. The embryos were then returned to normal media and allowed to grow normally until 65 hpf, when the heart was stopped again using BDM and the valve imaged (first and second columns in (B-C')). They were then returned to normal media and allowed to grow normally until 98 hpf, when the heart was stopped using BDM and they were imaged again (third and fourth columns in (B-C')). with thick superior AV valves and those with normal superior AV valves show increased reversing flow compared to wild-type controls, most likely because mutants with normal superior AV valves may still have abnormal inferior AV valves. However, the amount of reversing flow is greatest in mutants with thick superior AV valves (S15I, S15J-S15J"', and S15K Fig, S12 and S13 Movies), thus suggesting that thick valves are associated with decreased valve function.

Flow-dependent Nfatc signaling is required for proper valve delamination
We then proceeded to determine the mechanotransduction pathways involved in delamination. Since we have already confirmed that both klf2a −/− klf2b −/− and klf2a +/− klf2b −/− mutants have thick valve defects (S15E-S15G Fig), we wondered if down-regulation of these genes could be responsible for the delamination defects seen in the gata1 mutant. However, while we found that klf2 mutants have both delayed delamination and cell fate maps suggesting defects in delamination similar to that found in gata1 mutants (S16A, S16B and S16F Fig), klf2a and klf2b appear up-regulated rather than down-regulated in gata1 mutants (S16C and S16E Fig).
Searching elsewhere, we wondered if abnormal Nfat signaling could underlie the gata1 mutant phenotype. In mice, Nfatc2/3/4 are expressed in the AVC myocardium, where it suppresses the expression of VEGF, thereby allowing the initiation of EndoMT in the AVC endocardium [44]. By contrast, Nfatc1 is expressed in the endocardium [44], and targeted deletion of nfatc1 in the endocardium results in an overabundance of EndoMT-derived cushion mesenchyme, decreased cardiac neural crest-derived mesenchyme, and subsequent defects in valve remodeling [45]. In zebrafish, nfatc1 is expressed in valve precursors in the endocardium [22,23], and endocardial Nfatc has recently been shown to be activated in blood flowdependent manner [46].
To see if flow forces could regulate delamination via Nfat signaling, we first sought to visualize Nfat activity in gata1 mutants and controls. Using the Nfat binding element reporter line Tg(4xnfbr:d2EGFP), which expresses d2EGFP in response to the binding of nuclear-localized Nfat protein [46], we found that Nfat is activated in the region corresponding to the outer layer of the future superior AV valve leaflet at 65 hpf and that the number of Nfat activated cells is decreased in gata1 mutants compared to controls ( Fig 9A). These results suggest that flow forces regulate Nfat activity in luminal endocardial cells at the stage when valves are preparing for delamination.
Next, we examined whether inhibition of Nfat signaling could lead to delays in delamination. To do so, we used FK506, an immunophilin ligand that down-regulates the mRNA and protein levels of the phosphatase calcineurin. Since calcineurin activity is necessary for the nuclear translocation of Nfatc proteins, FK506 inhibits Nfat activation [47][48][49][50]. Using the Tg(4xnfbr:d2EGFP) line, we confirmed that FK506 effectively suppresses Nfat activity in this system ( Fig 9B). Then, to see if reduced Nfatc activity causes delays in delamination, we treated embryos with FK506 from 60 to 80 hpf and imaged the beating heart. We found that, unlike the gata1 mutant, FK506-treated embryos do not show delays in delamination (S17A Fig).
We then sought to determine whether inhibition of Nfat signaling could cause delamination defects. To do so, we treated embryos with FK506 either from 60 to 98 hpf or from 60 to 80 hpf. We found that FK506 treatment resulted in thick, hyperplastic superior AV valves ( Fig 9C, 9F and 9G, S17B and S17C Fig) and increased incidence of pericardial edema (S17D and S17E Fig). Then, to see if these thick valve phenotypes are caused by delamination defects, we performed VE-cadherin and ZO-1 immunostaining. We found that embryos treated with FK506 from 60 to 84 hpf results in an extra layer of VE-cadherin negative, ZO-1 positive cells situated between the 2 luminal layers of the valve leaflet in FK506-treated embryos, suggesting that, like gata1 mutants, some abluminal cells may have failed to undergo MEndoT (S17G Fig). Cell counting in 98 hpf, FK506-treated embryos suggests that thick valves have an overabundance of VICs and a reduced number of cells at the AVC wall (S17F Fig), suggesting that some cells of the abluminal bilayer may be misassigned from luminal to abluminal cell fate and that cells of the AVC wall compensate by helping form the inner layer of the valve leaflet. We confirm this via photoconversion experiments (Fig 10). Together, this suggests that inhibition of Nfat signaling could cause defects in delamination, leading to thick valve phenotypes.
Finally, we questioned if inhibition of Nfat signaling could cause increased cell proliferation. To test this, we performed EdU staining in FK506-treated embryos and their DMSO treated controls. Our results suggest that, like gata1 mutants, the superior AV valves of FK506-treated embryos have increased cell proliferation and that this increase is most likely attributable to the increased proliferation of cells misassigned from luminal to abluminal cell fate ( Fig 9D, 9H and 9I).
Altogether, these results suggest that lower maximal wall shear stresses in the gata1 mutant lead to lower Nfatc activity, which, in turn, leads to defects in delamination, increased cell proliferation, and thick, hyperplastic valves.

PLOS BIOLOGY
Tissue delamination and mechanical forces during valve morphogenesis

Twist1b is down-regulated in both FK506-treated embryos and gata1 mutants
Since Nfat is activated in luminal cells, we wondered how a decrease in Nfat activity could lead to defects in MEndoT in abluminal cells. Since we have shown that snail1b and twist1b are expressed in abluminal cells in wild-type embryos (Fig 5), we wondered if their expression is changed in gata1 mutants and FK506-treated embryos. To test this, we first assessed snail1b and twist1b expression in gata1 mutants using RNAscope and found that twist1b, but not snail1b, is up-regulated in gata1 mutants (Fig 11A-11C). Next, we sought to confirm that twist1b is up-regulated in gata1 mutants by analyzing the TgBAC(twist1b:GFP) line in the gata1 mutant background. We found that the number of cells expressing high levels of GFP (indicative of high cumulative expression of twist1b throughout the life of the embryo) is increased in mutants compared to controls (Fig 11D and 11E). Cell counting in the same embryos showed that total superior AV valve cell number is unchanged, indicating that the increase in the number of cells expressing high levels of GFP is due to changes in twist1b expression and not in increased proliferation of twist1b positive cells ( Fig 11F). Finally, we performed twist1b RNAscope and analyzed TgBAC(twist1b:GFP) embryos treated with FK506. In doing so, we found that twist1b is up-regulated in FK506-treated embryos (Fig 12). We thus propose a model (Fig 13) whereby blood flow-derived wall shear stress is necessary to activate Nfatc in luminal cells prior to delamination and that the transcription of Nfat target genes leads to the inhibition of twist1b in abluminal cells, thereby preventing these abluminal cells from undergoing further EndoMT. Perturbation of this pathway leads to defects in delamination, resulting in thick, hyperplastic valves and decreased heart function.

Discussion
Understanding the mechanobiology of developing valves is challenging due to the dynamic changes in force over 2 timescales: (1) Developing valves experience some of the largest and most rapid changes in mechanical force in the embryo within each heartbeat; and (2) cardiac forces due to heartbeat also change throughout development in concert with heart morphogenesis and changes in heart material composition [21,[51][52][53]. Elucidating the overarching mechanisms by which cardiac forces govern valve morphogenesis thus requires careful examination of their role at different developmental stages, and currently, the role of cardiac forces during post-EndoMT stages of valve morphogenesis is poorly understood. Here, we show for the first time that zebrafish superior AV valve morphogenesis involves a delamination step after endocardial cells migrate into the CJ (Fig 1, S2 Fig). We additionally show that valve delamination in gata1 mutants is delayed (S12 Fig) and that delamination, when it does occur, often occurs incorrectly, resulting in misassignment of cells from luminal to abluminal cell fate (Figs 6 and 7, S13 Fig). Misassigned cells proliferate, resulting in thick, hyperplastic valves with an overabundance of VICs (Fig 8). Finally, we show that decreased Nfatc activation in the endocardium is in part responsible for delamination defects in the gata1 mutant (Figs 9 and 10, S17 Fig) and that the well-established EndoMT transcription factor twist1b is up-regulated in both gata1 mutants and in embryos where Nfatc1 activity is inhibited (Figs 11 and 12), pointing toward a possible mechanotransduction pathway (Fig 13).

Zebrafish AV valves form via tissue sheet delamination
The debate as to how zebrafish AV valves form has in part been fueled by experimental data and also by the fact that valves in different parts of the circulatory system form via different mechanisms. Since MEndoT has not been demonstrated in amniotic valves, the idea that all

PLOS BIOLOGY
Tissue delamination and mechanical forces during valve morphogenesis cells that migrate into the CJ undergo EndoMT and become future VICs is intuitive [26]. Perhaps due to this, various studies that have analyzed superior AV valve morphology between 72 and 80 hpf in fixed or stopped hearts have presumed that most valve cells at these stages are abluminal [21,22,25,54,55], contributing to a model whereby AV valves attain their leaflet morphology via the gradual process of elongation [20]. Such a model is attractive as it draws parallels to the formation of the septal leaflet of the mitral heart valve in amniotes [56]. Inconsistent with this idea, an earlier study has shown that thin superior AV valve leaflets can be seen when one images the beating heart at 72 hpf [24]. Most recently, a study analyzing stopped and fixed hearts has suggested that elongation is preceded by "VIC invasion," whereby abluminal cells first undergo cellular rearrangement [23]. The authors note that only some abluminal cells become future VICs [23], leaving open the possibility that other mechanisms may be at play.
Here, we combine analyses of fixed/stopped hearts, beating hearts, and XYTZR time-lapse imaging to uncover how cellularized endocardial cushions transition into valve leaflets. We show that zebrafish superior AV valves first attain leaflet morphology via delamination, a rapid process that relies neither on the lengthening of the endocardial cushion nor cellular rearrangements (Fig 1, S2 and S3 Figs). Rather, delamination relies on the remodeling of cell-cell adhesions of cells that have already migrated to their correct location (Figs 1 and 3, S2 and S3 Figs). Delamination can occur as early as 72 hpf (thus explaining the observation of free-moving valve leaflets in the beating heart at this stage [24]) and is followed by valve elongation and the proliferation of abluminal cells at the valve base, the progeny of which migrate into the valve interstitial space to become VICs (Fig 2). This way of valve formation resembles the formation of mural heart valves and the septal leaflet of the tricuspid valve in amniotes [56]. However, among valve systems studied so far, the zebrafish superior AV valve appears unique in that delamination occurs via an apoptosis-independent process between 2 layers of endocardial-derived abluminal cells. Inherent to this process is the requirement that some cells transition toward a more mesenchymal state before transitioning back toward an endothelial cell state. Given that cells do not lose tight junctions throughout early migration and delamination stages (Fig 4), we suspect that EndoMT must be tightly controlled to ensure cells do not transition too far toward the mesenchymal state and retain their relative positions. While MEndoT has not been reported in any other valve system thus far, there is some evidence that MEndoT may be involved in cardiovascular processes such as angiogenesis [57] and artery reassembly [58,59]. Given that MEndoT is understudied relative to EndoMT, further investigation of zebrafish superior AV valve delamination may be a useful model to study how cells transition between EndoMT and MEndoT programs.

Mechanotransduction during AV valve delamination
Previous studies in zebrafish have shown that cardiac forces regulate EndoMT during the early cell migration stage of valve formation [11,46,[60][61][62], but whether they play a role in later stages of valve development is unclear. Here, we focus on valves growing under abnormal flow conditions that have managed to bypass early developmental defects. We find that endocardial Nfat activity is decreased in gata1 mutants (Fig 9A and 9E). This leads to the up-regulation of EndoMT-promoting gene twist1b in abluminal cells (Fig 12), the incorrect assignment of cells from luminal to abluminal cell fate during valve delamination (Fig 10), and thick valves with an overabundance of VICs (Fig 9, S17 Fig). That endocardial Nfatc can suppress EndoMT seems to be conserved in mice [63,64]. Surprisingly, zebrafish mutants in which the sequence that encodes the DNA binding domain of nfatc1 is deleted have been shown to have decreased twist1b expression in the AVC region and an abnormally low number of VICs at 4 dpf [23]. Possible reasons for the difference between the valve phenotypes of nfatc1 mutants and FK506-treated embryos include (1) expression of other Nfatc proteins in the zebrafish valve endocardium whose activity is suppressed by FK506; (2) differences in the period in which Nfatc1 activity is down-regulated; (3) noncanonical actions of Nfatc1 in FK506-treated embryos; and (4) genetic compensation in the nfatc1 mutant.
We find that although gata1 mutant superior AV valves do not down-regulate the flow-sensitive transcription factor klf2, klf2 mutants have thick superior AV valves and delamination delays (S16 Fig). Why klf2 mutants that have bypassed early EndoMT defects could then develop thick valve phenotypes remains unclear, although recent work on how Klf2 functions mechanistically in developing valves allows us to postulate some hypotheses. Klf2a has been shown to act in parallel with Notch to inhibit Flt4, a protein that is thought to be expressed in abluminal cells of zebrafish superior AV valves and strongly expressed in cells undergoing EndoMT in mouse AV valves [55]. Meanwhile, mechanosensitive Notch-Delta-like-4 and Erk5-Klf2-Wnt9a signaling pathways act together to regulate early zebrafish superior AV valve formation [62]. Specifically, Notch-mediated lateral inhibition between endocardial cells is thought to single out Delta-like-4-positive endocardial cells, which acquire competence to respond to Wnt9a. Concurrently, Wnt9a is produced downstream of Klf2, leading to the ingression of Delta-like-4-positive endocardial cells [62]. Finally, Klf2-Wnt/β-catenin signaling is thought to limit mesenchymal cell proliferation in mouse valves during remodeling stages [17,22]. Taken together, one possible explanation for the thick valve phenotype in klf2 mutants is that luminal cells at the delamination site require Klf2-Wnt signaling or Klf2-mediated Flt4 inhibition to break their VE-cadherin adhesions and release the valve leaflet from the AVC wall. Another possibility is that Klf2-Wnt/β-catenin signaling promotes MEndoT in abluminal cells during delamination stages. Most likely, Nfat and Klf2 mechanosensitive signaling pathways, each tuned to specific mechanical stimuli and mechanosensors [46], act synergistically to regulate both early valve EndoMT and valve delamination. However, how these pathways act together to accommodate the changing requirements in valve cell-cell adhesion remains to be determined.
Determining the precise aspects of cardiac force sensed by cells is an ongoing challenge in the field of mechanosensing. Here, we find that maximal wall shear stress is significantly decreased at the AVC at 65 hpf, while heartrate appears unaffected in gata1 mutants (S10 and S11 Figs). Together with our finding that injection of a viscous medium into gata1 mutants after cell migration can partially rescue the thick valve phenotype strongly suggests a role for wall shear stress in regulating valve delamination (S14 Fig). However, we cannot rule out other aspects of cardiac force (e.g., pressure and stretch) playing additional roles, nor do we know if cells are sensing oscillatory aspects of wall shear stress or simply its maximum amplitude. Given that the inferior AV valve leaflet begins to form later than the superior AV valve leaflet but appears to develop at roughly the same speed, there may be insights to be gained by performing more in-depth comparisons of force properties and Nfat activity levels in the superior versus the inferior regions of the AVC.

A cellular model for studying the role of mechanical forces during valve delamination
In summary, we have identified a previously unappreciated step in zebrafish valve formation nestled between early cell migration stages [21,22] and valve maturation stages [23,65] that is critical to understanding how cellularized endocardial cushions morph into primitive, freemoving valve leaflets. We found that this step is sensitive to cardiac forces via a Nfatc-dependent mechanism and that disruption of this step can lead to thick valve phenotypes. Given that there are several congenital cardiac anomalies in humans associated with improper AV valve delamination, including Ebstein's malformation, parachute mitral valve, and parachute-like asymmetric mitral valve [66][67][68], studying how cardiac forces affect delamination would be conducive toward finding better prevention methods and treatments for associated heart diseases.
The Tg(fli1a:EGFP-Podxl) ncv530Tg (referred to as Tg(fli1a:EGFP-Podxl) in the text), was generated using the Tol2 transposon system. The plasmid was generated by fusing EGFP after the signal peptide of zebrafish podocalyxin and then cloned into a Tol2 vector with the fli1a promoter.
The klf2b ig5 line was generated using a TALEN pair (left and right arms: 5 0 GGACATGGCT TTACCT-3 0 and 5 0 AACGTTTGCAAACCAG-3 0 ) were designed to target exon 1 of the klf2b gene and injected into single-cell wild-type (AB) first cell. We identified the alleles generated and confirmed that potential targeting events could be transmitted through the germline by outcrossing the F0 fish with AB animals and sequencing genomic DNA from pools of 6 F1 embryos. After screening the first generation, we focused on a 28-bp deletion mutation (5 0 -GGACATGGCTTTACCTTGCCTTTTGCCT-3 0 ) leading to a premature stop codon in klf2b transcript. Studies were performed from F4 fish and later generations. A PCR-based genotyping strategy was established using the following primers to identify the wild-type and mutant alleles, the length of the deletion allowing their visualization directly on a 3%-agarose DNA gel: forward 5 0 -GGAAAGCGCGTATATTTGGA-3 0 , reverse 5 0 -CAAGTAGGAAATGCAAG TGT-3 0 and sequencing forward primer 5 0 -AGAGCGCACTGTGCCTTATA-3 0 .
For all experiments, embryos with improper heart looping at 48 hpf, curved tails, or leftright defects were excluded from analysis. Embryos with cells from the hatching gland obscuring the light path to the heart were also excluded from analysis. For analysis of wild-type embryos, embryos with pericardial edema were excluded from analysis. For analysis of mutants and FK506-treated embryos, embryos with pericardial edema between 65 and 72 hpf were excluded from analysis unless otherwise stated in the figure legend.

Confocal and 2-photon imaging
Confocal and 2-photon imaging was performed using a Leica SP8 upright confocal microscope equipped with lasers operating at 488 nm, 561 nm, and 633 nm, a tunable multiphoton laser, and a Leica HCX IRAPO L, 25 ×, N.A. 0.95 objective. EGFP and Venus signals were imaged using the multiphoton laser tuned to 927-nm rather than the 488-nm laser to improve image quality in all experiments except for bead insertion experiments, where the signal from the bead is much higher than the signal from the EGFP or Venus when irradiated 927-nm light. In all experiments, detectors were set to photon counting mode.

In vivo imaging of the beating heart
Fluorescent imaging of the beating heart was performed using a Leica DMi8 combined with a CSU-X1 (Yokogawa, Tokyo, Japan) spinning at 10,000 rpm, 2 simultaneous cameras (TuCam Flash4.0, Hamamatsu, Shizuoka, Japan), and a water immersion objective (Leica 40X, N.A. 1.1). Embryos were mounted in 0.7% low melting-point agarose (Sigma-Aldrich) in a glassbottom petri dish (MatTek, Ashland, USA, P35G-0-14-C) and imaged at 28.5˚C at 100 frames per second. Tg(fli1a:gal4ff;UAS:EGFP-CAAX), Tg(fli1a:gal4ff;UAS:Kaede), and Tg(fli1a:LifeAct-EGFP) embryos were used to visualize if the AV valve has separated from the inner AVC wall. For experiments designed to visualize cell morphology at single-cell resolution, Tg(fli1a:gal4ff; UAS:EGFP-CAAX) embryos were incubated with 4 μM BODIPY TR Ceramide (Thermo Fisher Scientific, USA) for 20 minutes to stain the blood plasma and the CJ. To stop the heart without unmounting embryos, 2 mL of 100 mM BDM (Sigma-Aldrich) and 0.4% tricaine (Sigma-Aldrich) solution was added to the mounting dish. Once the hearts are seen to stop beating, 2 mL of normal embryo media is added to the dish, and the embryos are imaged within 10 minutes. Where necessary, realignment of the beating heart was performed postimaging using BeatSync2.0 [79].
Brightfield imaging of blood flow in 98 hpf klf2 mutants was performed using a Leica DM IRBE microscope mounted with a Fastcam SA3 camera (Photron, Tokyo, Japan) and a water immersion objective (Leica 20X, N.A. 0.70). Embryos were mounted in 0.7% low meltingpoint agarose (Sigma-Aldrich) in a glass-bottom petri dish (Matek) and imaged at 500 frames per second.

Bead injections
To occlude blood flow at the ventricle and thereby increase reversing flow at the AVC, beads (PureCube Glutathione MagBeads, Cube Biotech, Monheim am Rhein, Germany, 32225) of roughly 30 μm were inserted into the ventricle of the heart at 72 hpf using ultrafine tungsten manipulator probes (Ted Pella, Redding, USA, 13570) and tweezers. Briefly, embryos anesthetized with 0.02% tricaine (Sigma-Aldrich) were mounted in a mold [27] with 0.7% low melting point agarose (Sigma-Aldrich). Beads were then deposited on top of the low melting point agarose. Tungsten probes were used to create a hole in the center of the yolk and a path under the skin for the bead to follow from the yolk to the cardinal vein. A bead is pushed through the agarose and into the yolk using tweezers and then pushed anteriorly into the previously created path using a tungsten probe. The bead is then guided through the path using tweezers by gently pushing the bead anteriorly through the skin of the embryo. Once the bead is situated inside the cardinal vein, suction caused by the beating heart will cause the bead to move into the heart. Only hearts where the bead moves into the ventricle or the outflow tract were used for this study. Injected hearts where the bead remained in the atrium or moved beyond the outflow tract and exited the heart were not included in this study. Hearts with chambers that were collapsed 4 hours after bead injection were excluded from analysis. After insertion of the bead, embryos were allowed to develop normally in a 28.5˚C incubator. For sham embryos, the same surgical procedure is performed except that the cardinal vein was damaged using probes to mimic the damage caused by the bead, and the bead is not guided along the path to the cardinal vein, remaining in the yolk of the embryo instead.

Drug treatments
To slow heart rate, embryos were transferred to 15 mM BDM. For both bead and drug experiments, embryos were stage matched and randomly assigned to control or treatment groups. For both bead and drug experiments, 3 hours 40 minutes after treatment, embryos were placed in 4 μM BODIPY-TR Ceramide (Invitrogen, ThermoFisher Scientific, Waltham, USA, D7540). Just before imaging, embryos were washed and anesthetized in 0.02% tricaine solution and then imaged using a spinning disk microscope to see if the AV valve has delaminated. Heartrate before imaging was measured at room temperature using a stereomicroscope and a timer. Heartrate at the time of imaging was measured based on movies obtained, while the embryos rested inside a 28.5˚C chamber.
To inhibit Nfatc1 activity, FK506 (Sigma-Aldrich) was used at a concentration of 2 μM.

Immunohistochemistry
Embryos were fixed at the desired stage in 4% paraformaldehyde 3 to 4 hours at room temperature or overnight at 4˚C. After washing embryos in 1x PBST (PBS −0.1% Tween-20), the pericardial cavity of embryos was pierced using a 0.25-mm diameter stainless steel needle (Ted Pella), and the tails of the embryo were severed just posterior to the yolk extension. Embryos 48 to 50 hpf were permeabilized in 1x PBST containing 0.5% Triton-X 100 for 30 minutes at room temperature, embryos 55 to 60 hpf were permeabilized overnight at 4˚C in 1X PBST containing 0.5% Triton-X 100 at 4˚C, and embryos older than 60 hpf were permeabilized overnight at 4˚C in 1x PBST containing 1% Triton-X 100. Embryos were then blocked overnight at 4˚C in 1x PBST supplemented with 5% BSA (anti-fibronectin) or 1% BSA and 10% NGS (anti-VECadherin, anti-ZO-1, and anti-Esama). Antibodies were used as follows: rabbit anti-fibronectin (F3648, Sigma-Aldrich) 1:100, rabbit anti-VECadherin [80] 1:500 (kind gift from Affolter lab), mouse anti-ZO-1 (Invitrogen) 1:100, rabbit anti-Esama 1:100 (kind gift from Affolter lab), goat anti-rabbit, and goat anti-mouse Alexa-647 secondary antibodies (Life Technologies, Carlsbad, USA, A-21235) 1:500. For embryos older than 80 hpf embryos, the skin of the embryo covering the heart was surgically removed using tungsten manipulator probes prior to imaging to reduce light scattering. and the background intensity is given by

Background ¼ Integrated intensity of atrial ROI Area of atrial ROI
Note that because there is a small measurement error associated with this method, when the number of dots in a valve region is very close to zero, the measured number of dots can be negative. The outline of the valve and valve cells is mainly determined using the background fluorescein stain, which labels cell membranes faintly. Where necessary, the determination of the outline of cells was also aided by the phalloidin channel, which stains red blood cells and the myocardium, as well as the brightfield channel, which can be useful for identifying the location of atrial endocardial cells.

Photoconversion
Photoconversion was performed as previously described [27]. Briefly, embryos were mounted with the aid of a 3D printed mold in 0.7% low melting-point agarose supplemented with 50 mM BDM to inhibit heart contraction for the duration of the procedure. Photoconversion was performed using the FRAP module on an SP8 confocal microscope and a Leica HCX IRAPO L, × 25, NA0.95 water immersion objective. At 48 or 50 hpf, the region of interest exposed to 405 nm light to convert Kaede protein from its green form to its red form. A z-stack of the photoconverted heart was then acquired in the standard confocal mode to record the starting point of each experiment. Embryos were then removed from the agarose using a glass pipette, placed in fish water for 5 to 10 minutes until heart contraction resumed, and then put at 28.5˚C to develop individually under standard conditions until the time point of interest. Note that for analysis, we assume that photoconverted Kaede in the ventricle is degraded or diluted due to cell proliferation between 98 and 144 hpf.

Electron microscopy
Tg(ve-cad:ve-cadTS) embryos were first imaged using a confocal microscope to ensure that VE-cadherin is down-regulated. The embryos were then fixed by immersion in 2.5% glutaraldehyde and 2.5% formaldehyde solution in 0.1M cacodylate buffer (PH = 7.2)) overnight at 4 C. Embryos were rinsed 2 times in cacodylate buffer and followed by a 1 hour postfixation in 1% osmium tetroxide [OsO 4 ] reduced by 1% potassium ferricyanide [K 3 Fe(CN) 6 ] in dark on ice. Embryos were washed 1 time in cacodylate buffer and after extensive rinses in distilled water. The embryos were incubated in 1% uranyl acetate, for 2 hour on ice, and rinsed in water. Dehydration was performed in graded series of ethanol solutions (50%, 70%, 90%, and 100%; quickly rinsed and incubated for 20 minutes each), to be then infiltrated with epoxy 812. Semithin sections were cut at 2 μm by ultramicrotome (Leica Ultracut UCT) and stained with 1% Toluidine blue in 1% sodium borate, examined by Leica optical microscope (LEICA DMLB, Leica Microsystems, Germany). Ultrathin sections were cut at 70 nm and contrasted with uranyl acetate and lead citrate and examined at 70 kv with a Morgagni 268D electron microscope (FEI Electron Optics, Eindhoven, the Netherlands). Images were captured digitally by Mega View III camera (Soft Imaging System, Münster, Germany).

Analysis of VE-cadherin (Cdh5) expression
In Fig 3B'-3B", we used ImageJ to measure the average signal intensity of each cell-cell interface in the z-slice. For Fig 3B', at 55 and 65 hpf, the average of the average signal intensity for cell-cell interfaces of luminal cells (depicted in purple Fig 3B) is divided by the average of the average signal intensity for cell-cell interfaces of abluminal cells (depicted in orange in Fig  3B). Similarly, at 80 hpf, the average of the average signal intensity of cell-cell interfaces of the outer layer of the valve (depicted in purple in Fig 3B) is divided by the average of the average signal intensity of cell-cell interfaces of the inner layer of the valve and the AVC (depicted in orange in Fig 3B). For Fig 3B", the same method was used except that the average maximum signal intensity for cell-cell interfaces was used instead of the mean of the average signal intensity. Statistical significance relative to a theoretical value of 1 was calculated using the Wilcoxon rank-sum test.

Analysis of podocalyxin localisation
For all podocalyxin quantifications, we first use Imaris to create a new coordinate system [22] and a new image stack (the x-axis points in the direction of the rise of the lumen, the y-axis points in the direction of the length of the lumen, and the z-axis in the direction of the span of the lumen). Z-slices are maximally 2 μm apart. For each z-slice, we use Fiji to measure the average green and red signal along the apical and basal sides of the cells.

Phenotypic analyses
Cell counting and valve volume measurements were performed on Imaris software (Bitplane, Zürich, Switzerland). At 48 hpf, the AVC was defined using morphological landmarks where the edges of the AVC correspond to points of inflection of the endocardial wall. We classify valves as thick when they are 3 or more cell layers thick at the distal end of the valve or more than 3 cell layers thick at the proximal end of the valve. At 98 hpf, the inner layer of the valve may touch the AVC wall when the heart is stopped or fixed if the heart valve is very thick. Thus, for cell counting in these valves, we use the brightness of fli1a driven EGFP (tends to be down-regulated in VICs) and cell shape (organized for luminal cells) in addition to cell position to assign cells to "AVC wall," "VIC," or "Inner Layer" region categories. When imaging the beating heart, valves are counted as "delaminated" when a gap (sinus) is observed between the inner layer of the valve leaflet and the AVC wall at the central region of the valve.

EdU incorporation assay
Gata1 mutants and controls were transferred to 0.5 mmol/L EdU in egg water/0.5% dimethyl sulfoxide (DMSO) from 74 to 98 hpf. For FK506 experiments, EdU was added to media already containing DMSO or FK506 to reach a final concentration of 0.5 mmol/L EdU. The larvae were then fixed with 4% PFA for 3.5 hours at room temperature and washed with 1x PBST. The pericardial cavity was pierced using an ultrafine needle, the tails cut, and the entire larvae permeabilized overnight at 4˚C in 1x PBST containing 1% Triton-X 100. Detection of EdU was performed using a Click-iT EdU Alexa Fluor 647 Kit (Thermo Fisher Scientific), where larvae were incubated in the Click-iT reaction cocktail for 1 hour. Embryos that show no staining in the heart and surrounding tissues were excluded from analysis.

Shear stress modeling
In order to model shear stress at the AVC wall over the cardiac cycle, we treat blood in wildtype as a 2-phase solution, containing red blood cells and plasma, and blood in gata1 mutants as a 1-phase solution, containing only plasma.
To estimate the viscosity of blood plasma, we use the empirical equation from Corcione's work [81], which reviews various experimental works with nanoparticle volume fractions in the range from 0.0001 to 0.071, temperatures in the range between 293 K and 333 K, and particle sizes in the range from 25 to 200 nm. The empirical equation has a 1.84% standard deviation of error. ; where μ f and μ eff are the dynamic viscosity of the base fluid and the effective dynamic viscosity of the nanofluid (N m −2 s), M is the molecular weight of the base fluid, (kg mol −1 ), N is the Avogadro number (6.022e 23 mol −1 ), φ is the volume fraction of the nanoparticles, ρ f0 is the mass density of the base fluid calculated at temperature T 0 = 293 K (kg m −3 ), d p is the diameter of the nanoparticle (m), and d f is the equivalent diameter of a base fluid molecule (m). Zebrafish blood plasma has previously been reported to have a relatively constant dynamic viscosity of 1.5 mPa.s over a range of temperatures [82]. Zebrafish were kept at 28.5˚C (301.65 K), and the density of the plasma was approximated to be 997 kg m −3 . We approximate the molecular weight of the plasma to be 100 kDa by averaging the protein profiles based on weight percentage [83] (using the table provided in the Supporting information of Li and colleagues' work [83]).
To estimate the shear rate and shear stress experienced by cells of the AVC wall, we first image Tg(flk:EGFP; gata1:DsRed) embryos to measure red blood cell velocity and position in 65 hpf wild-type embryos. Images of the AVC wall and red blood cells were captured on the spinning disk at 1,000 frames per second for 2 cardiac cycles. Additional images of the entire heart were captured at 100 frames per second to ensure correct identification of the AVC wall contour. Pairwise image registration was used to track blood cells motion. Free-form image registration was performed with a grid of 5.6 microns, with an intensity mean square metric using Simple Elastix and an initialization with half the value of the previous time step's deformation map to smooth the velocity changes. With image pixels above an intensity threshold identified as blood cells, shear rate is computed assuming a linear flow profile between the blood cell and the AVC surface for each cell that passes through near to the AVC wall. Shear stress values was then calculated (shear stress = shear rate × viscosity).
For the gata1 mutants, XYTZ images of the heart were captured at a frame rate of 100 frames per second for 80 frames, and the beating heart was realigned using BeatSync2.0 [79]. We track the ventricle volumes across time with a motion estimation algorithm [40] and calculate flow rates via the rate of volume changes of the ventricle. Subsequently, AVC shear rate is estimated with the assumption of a parabolic flow profile across a maximum inscribed circle in the cross section of the AVC, thus providing conservatively maximum wall shear rate values in the gata1 mutants. Maximal shear stress values were then calculated (shear stress = shear rate × viscosity).

Gata1 mutant rescue experiment
Anesthetized embryos were mounted in a glass bottom petri dish (Matek) with 0.7% low melting point agarose. 4.6 nl of water (controls) or nanoemulsion containing 3.3% of DiI dye salt with tetraphenylborate counterion, DiI-TPB [84], 95 nm red fluorescent lipid droplets in PBS (rescue) was injected into the bloodstream of 60 hpf gata1 mutants via the common cardinal vein near the inflow of the heart. The dye-loaded nanoemulsion was made using a spontaneous emulsification procedure described in [84]. A Nanoject II (Drummond Scientific, Broomall, USA) microinjector, which requires backfilling with mineral oil, was used for injections. Because injected drops of nanoemulsion tend to stay stuck to the glass capilliary, the needle is kept within the common cardinal vein for about a minute after injection until the red nanoemulsion is seen to fully disperse in the circulation before pulling out the needle. After injection, embryos were carefully unmounted and returned to egg water with 0.003% PTU and the 28.5 C incubator. Embryos were left to grow until 98 hpf when their hearts were stopped using BDM, and their superior AV valves were imaged using a confocal microscope (see "Confocal and 2-photon imaging" in the Materials and methods section).
To estimate the increase in blood viscosity caused by the injection of the nanoemulsion into the bloodstream of gata1 mutants, a sample of the nanoemulsion was diluted 1:13 in PBS to match its expected concentration in the blood plasma and the viscosity of the diluted nanoemulsion was measured using a microfluidic rheometer (total blood volume was assumed to be 60 nl based on [85]). We found that the addition of the nanoemulsion to PBS increased its viscosity from 0.92 ± 0.02 cP to 1.10 ± 0.06 cP (n = 20, p < 0.005). The microfluidic rheometer comprised of a polydimethylsiloxane (PDMS, Dow Corning, USA) microchannel bonded to a microscope glass slide (VWR, USA) by oxygen plasma treatment. The channel had a square cross section with a 45-μm width. Flow was driven by a constant input pressure of 200 mbar with a pressure controller (MFCS-EZ, Fluigent, France). The samples were suspended with 2 um polystyrene tracer beads (0.1% v/v concentration, Thermo Fisher Scientific) for flow rate measurement using particle tracking. The microfluidic rheometer was calibrated using viscosity reference standards (Paragon Scientific, UK).

Quantification of GFP high and GFP low cell number in the Tg(twist1b:GFP line)
Images of the whole heart and pericardial cavity were captured using the with laser intensity set so that the entire dynamic range of GFP signal is captured. GFP high cells in the superior AV valve were classified as cells that can be seen to be GFP positive to the naked eye in autocontrasted images. GFP low cells were classified as cells that can only be seen to be GFP positive when the brightness/contrast was further adjusted.

Graphs, statistics, and image orientation
We did not compute or predict the number of samples necessary for statistical differences because the standard deviation of our study's population was not known before starting our analysis. The number of embryos used for each experiment (n) is provided in the figures and/ or figure legends. Notched Tukey boxplots were plotted using the IoSR Matlab Toolbox package (https://github.com/IoSR-Surrey/MatlabToolbox). The notch is centered on the median and extends to 1.58 � IQR/sqrt(N), where N is the sample size. Bar graphs and dot plots in the manuscript were created using GraphPad Prism. The bar in dot plots marks the median. Error bars in bar graphs show standard deviation. No data points were excluded as outliers for analysis of statistical significance. Where exact p-values are not stated, stars represent ranges of p-values: n.s. p > 0.05; � p � 0.05; �� p � 0.01; ��� p � 0.001; ���� p � 0.0001. Images are orientated such that ventricle is on/toward the left, the atrium on/toward the right.   50 hpf, embryonic hearts were stopped using BDM and atrial side of the AVC was photoconverted. The embryos were then returned to normal media and allowed to grow normally until 65 hpf, when the heart was stopped again using BDM and the valve imaged (first and second columns in (B-C')). They were then returned to normal media and allowed to grow normally until 98 hpf, when the heart was stopped using BDM and they were imaged again (third and fourth columns in (B-C')). RF across the AVC was binned into 3 categories: 0%, greater than 0% but less than 20%, and greater than 20%. Graph showing the percentage of klf2 control embryos, klf2 mutant embryos where superior AV valves have normal thickness, and klf2 mutant embryos where superior AV valves are abnormally thick with RF at 98 hpf. p-Values are based on Fisher exact test for embryos with greater than 20% of their cardiac cycle showing RF. The data underlying all the graphs can be found in S1 Data. AV, atrioventricular valve; CJ, cardiac jelly; hpf, hours postfertilization; RF, reversing flow. (TIF) S16 Fig. Klf2 mutants show defects in delamination, but klf2 is up- (F) Graph showing cell numbers by valve region at 98 hpf for embryos treated with DMSO or FK506 from 60 hpf. Statistical significance was determined by multiple t tests. (G) Representative images of embryos that have been treated with DMSO or FK506 from 60 hpf, fixed at 84 hpf, then immunostained for Cdh5 and ZO-1. White asterisks mark additional abluminal cells in FK506-treated embryos that express ZO-1 but not Cdh5. Scale bars: 10 μm. The data underlying all the graphs can be found in S1 Data. AVC, atrioventricular valve; hpf, hours postfertilization; ZO-1, zonula occludens-1. (TIF) S1 Movie. Related to S1B and S1B' Fig. Time-lapse imaging was performed using the Tg (fli1a:gal4ff;UAS:EGFP-CAAX) line starting at 53 hpf. The movie shows how a cell (segmented in blue) migrates into the CJ over 3 hours. The 3D heart was reconstructed for the entire cardiac cycle at each developmental time point, and the point of the cardiac cycle just before atrial contraction is shown. CJ, cardiac jelly; hpf, hours postfertilization. (MOV) S2 Movie. Related to S1C and S1C' Fig