Suppression of transcytosis regulates zebrafish blood-brain barrier function

1Blood vessels in the vertebrate brain are composed of a single layer of endothelial cells that possess distinct functional properties that allow the passage of necessary nutrients yet prevent unwanted entry of specific toxins and pathogens into the brain. This specialized endothelial layer forms the blood-brain barrier (BBB) and restricts the passage of substances between the blood and the brain parenchyma via two primary mechanisms: 1) specialized tight junction complexes between apposed endothelial cells to prevent intercellular transit and 2) suppressing vesicular trafficking or transcytosis to prevent transcellular transit. BBB selectivity is further refined with the expression of substrate-specific transporters that dynamically regulate the influx of necessary nutrients and efflux of metabolic waste products. While the BBB is comprised of endothelial cells, the surrounding perivascular cells including pericytes and astroglial cells, play a critical role in forming and maintaining barrier properties. Collectively, endothelial cells and the surrounding perivascular cells form the neurovascular unit.

2As the simplest genetic model organism with an endothelial BBB, zebrafish offer a powerful tool to study the cellular and molecular properties of the vertebrate BBB. Zebrafish have served as a great model system to study vascular biology due to their large clutch size, rapid and external development, and transparency for in vivo whole organism live-imaging. Additionally, with the advent of CRISPR-Cas9 technology, zebrafish provide an efficient genetic toolkit for targeted mutagenesis. However, the subcellular and molecular mechanisms governing the formation and maintenance of the zebrafish BBB are only beginning to be characterized. Expanding our understanding of the zebrafish BBB can thus reveal the mechanistic similarities between the zebrafish and mammalian BBB to further evaluate the position of zebrafish as a model organism for studying the BBB.

3Barrier properties of brain endothelial cells are induced by extrinsic signals from other cells in the surrounding microenvironment during development. In rodents, the BBB becomes functionally sealed in a spatiotemporal gradient, with the hindbrain and midbrain barriers becoming functional before the cortical barrier. Within the cortex, barrier function is acquired along a ventral-lateral to dorsal-medial gradient. In the zebrafish, existing studies have disagreed over the timing of zebrafish barrier formation, with some suggesting that BBB maturation occurs concurrently with angiogenesis and others providing a wide range beginning at 3 dpf and extending to 10 dpf. These conflicting reports may be due to regional developmental gradients of barrier acquisition, differences in definitions of BBB development, which range from functional tracer restriction to gene expression of BBB-specific selective transporters, such as glut1b, or variations in the experimental approaches used to assess BBB permeability such as the molecular weight and properties of tracers and circulation time. Therefore, we aim to resolve some of these discrepancies through detailed regional and subcellular characterizations of functional barrier acquisition in zebrafish.

4Recent work in the mammalian blood-retinal barrier has indicated that the suppression of transcytosis governs functional barrier development. Interestingly, endothelial cells at the leaky neonatal angiogenic front possess functional tight junction complexes halting the intercellular passage of the tracer protein Horseradish Peroxidase (HRP) at the so-called ‘kissing points’. In contrast, these endothelial cells exhibit high levels of HRP-filled vesicles compared to functionally sealed proximal vessels. Moreover, these areas of elevated vesicular trafficking continue to correspond with barrier permeability at the angiogenic front until the barrier seals. Work in the mouse BBB has also demonstrated the importance of suppressing transcytosis in determining barrier permeability. Mice lacking the major facilitator super family domain containing 2a (Mfsd2a) lipid transporter exhibit increased levels of caveolae vesicles in CNS endothelial cells, resulting in increased barrier permeability. Whether this suppression of transcytosis also determines BBB function in zebrafish remains unknown.

5Here in zebrafish, we find a spatiotemporal gradient of barrier acquisition, and capture the dynamics of BBB leakage as it matures during development using time lapse live imaging. We further find a conserved role for transcytosis suppression in determining barrier function, both during normal development and in mfsd2aa mutants.

6To determine when and how the zebrafish BBB becomes functional in different brain regions, we performed intracardiac injections of fluorescently conjugated tracers (1 kDa NHS and 10 kDa Dextran) simultaneously at different developmental stages and imaged live fish after 1 hr of tracer circulation (Figure 1A and B). NHS is widely used to assess mouse BBB permeability. Additionally, NHS has been used successfully to assess junctional defects in occludin- and claudin-deficient animals, and was previously shown to be restricted within the adult zebrafish cerebral vasculature. We used a combination of different molecular weight tracers to tease apart potential avenues of leakage, as tight junctional defects result specifically in the leakage of low-molecular-weight tracers 1 kDa and below into the brain parenchyma. To assess BBB permeability, we measured tracer fluorescence intensity in the brain parenchyma and normalized to circulating levels of each tracer, that is to the fluorescence intensity of tracer within brain blood vessels, to account for between fish variation in tracer injections or circulation (details in the Materials and methods section). At 3 dpf, we observed the presence of both NHS and Dextran throughout the brain parenchyma (average of 8.5 ± 0.3 Tracer Intensity in the midbrain and average of 7.4 ± 0.4 NHS and 5.5 ± 0.6 Dextran Intensity in the hindbrain; Figure 1; Figure 1—figure supplement 1). These leakage assays revealed that the injected tracers were able to permeate into the brain parenchyma, suggesting that the BBB was not functionally sealed. In addition to the use of exogenous injected fluorescent tracers, we also assayed BBB permeability with an endogenous transgenic serum DBP-EGFP fusion protein (Tg(l-fabp:DBP-EGFP)) to account for injection artifacts. At 3 dpf, we observed similar leakage patterns with the transgenic serum protein as we did with the injected tracers (average of 8.9 ± 0.8 DBP-EGFP Intensity in the midbrain and average of 6.1 ± 0.4 DBP-EGFP Intensity in the hindbrain; Figure 1C and D; Figure 1—figure supplement 1). At 4 dpf, we observed a significant decrease in tracer intensity in the hindbrain (average of 4.1 ± 0.2 NHS Intensity, 3.9 ± 0.4 Dextran Intensity and 4.0 ± 0.4 DBP-EGFP Intensity; p<0.0001, p=0.0115 and p=0.0034 compared to 3 dpf for NHS, Dextran and DBP-EGFP by two-way ANOVA; Figure 1—figure supplement 1). However, the midbrain barrier remains leaky (average of 7.6 ± 0.5 NHS Intensity, 7.5 ± 0.8 Dextran Intensity and 6.8 ± 0.9 DBP-EGFP Intensity; p=0.66, p=0.89 and p=0.079 compared to 3 dpf for NHS, Dextran and DBP-EGFP by two-way ANOVA; Figure 1C and D). At 5 dpf, the midbrain parenchymal tracer intensities are dramatically reduced (average of 3.3 ± 0.3 NHS Intensity, 3.6 ± 0.2 Dextran Intensity and 3.2 ± 0.3 DBP-EGFP Intensity; p<0.0001, two-way ANOVA; Figure 1B and C) and no significant changes were observed in the hindbrain (Figure 1—figure supplement 1). No significant change was observed in midbrain tracer uptake from 5 to 6 dpf or from 5 to 10 dpf (Figure 1C and D), indicating that the midbrain barrier becomes sealed at 5 dpf. Of note, all three tracers showed nearly indistinguishable patterns of uptake, with similar parenchymal tracer intensity levels (Figure 1; Figure 1—figure supplement 1), suggesting that the leakage is not due to tight junctional defects but may rather be due to an increase in vesicular trafficking.

7The vast majority of studies interrogating BBB permeability have relied on observing leakage in fixed static images from mutant mice, which does not reveal the dynamic nature of the BBB. A key advantage of zebrafish is the ability to examine biological processes in vivo in real-time, providing us with the unique opportunity to observe the process of functional barrier maturation (Figure 1). To examine the developmental dynamics of barrier leakage, we injected larvae at 3 dpf with fluorescently labeled 10 kDa Dextran and performed time lapse live imaging for 1 hr following injection, with time 0 being an average of 8 min post-injection. We observed a gradual and diffuse increase in extravascular Dextran intensity over time within the brain parenchyma (slope of 0.13 intensity/minute; Figure 2—videos 1 and 2; Figure 2A, C and E). In addition to the time-dependent increase in overall Dextran intensity in the brain parenchyma, we also observed parenchymal cells taking up tracer, both directly adjacent to blood vessels and at a small distance away (Figure 2A and C). To validate that the Dextran was in fact being taken up by cells rather than filling the extracellular space surrounding cells, we performed these same tracer injections in a transgenic line with fluorescently labeled cell membranes (Tg(actb2:MA-Citrine)) at 3 dpf and observed several tracer-filled cells, that is tracer confined within cell membranes outside of the vasculature (Figure 2—figure supplement 1). Similarly, when we performed these Dextran injections in a nuclear labeled transgenic line (Tg(actb2:Hsa.H2B-Scarlet)) at 3 dpf, we also observed Dextran co-localization with nuclear signal (Figure 2—figure supplement 1), further indicating that the injected tracers are taken up by scavenger parenchymal cells in addition to permeating the extracellular spaces.

8Interestingly, at 3 dpf, we observed an actively sprouting cell that extended away from the established vessel in a sporadic fashion (Figure 2—Video 2; Figure 2A). After 20 min of rambling migration, this sprout suddenly released a large bolus of Dextran that appeared to be taken up by a single parenchymal cell. Two min later, this same sprout released a second bolus of Dextran on the opposite side (Figure 2—Video 2). These rare large bursts of leakage were not unique to this sprout; they were also sporadically observed from established blood vessels (Figure 2—Video 1; Figure 2A). While this particular fish displayed three of these bursting events, defined as parenchymal Dextran intensity increasing to luminal levels in less than 1 min, only two of the other five measured 3 dpf fish displayed any bursts within the hour post tracer injection, one from a sprouting tip cell and one from a stable vessel, further indicating the infrequency of these large bursts. Cumulatively, this time lapse data revealed two discrete types of leakage occurring during this early developmental stage: steady and diffuse Dextran leakage into the parenchyma that makes up the vast majority of observed leakage and rare large bursts of leakage.

9When we performed these same time lapse experiments at 5 dpf, we observed significantly less overall tracer accumulation in the brain parenchyma over the course of the hour in addition to reduced rates of tracer accumulation (slope of 6.833e-6 intensity/sec; Figure 2—Video 3; p<0.001, Mann-Whitney test; Figure 2B, D and E). Furthermore, 5 dpf fish never exhibited bursts of tracer leakage as observed at 3 dpf. These data suggest that at early stages, most tracer leakage occurs broadly through vessel walls via unrestricted transcytosis with an occasional transient rupture to endothelial integrity that leads to these large bursts of leakage. Once tracer is leaked into the intercellular space of the parenchyma it can be taken up by select parenchymal cells (Figure 2—figure supplement 1). As the BBB matures both sources of leakage sharply decrease.

10Given the observed differences in brain permeability during larval development, we next sought to determine the subcellular mechanisms underlying the development of a functional BBB. Therefore, we assessed BBB properties by performing intracardiac injections of electron-dense NHS-gold nanoparticles (5 nm) followed by transmission electron microscopy (TEM) at different developmental stages when the BBB is leaky (3 dpf) and when it is functionally sealed to circulating fluorescent tracers (5 and 7 dpf; Figure 3). All the blood vessels analyzed had a maximal diameter of 5 µm to enrich for capillaries and small veins. At 3 dpf, we found blood vessels in direct contact with neurons and pericytes, with the pericytes sharing the endothelial basement membrane (Figure 3A), as in the mammalian neurovascular unit. During this leaky developmental stage, the endothelial basement membrane was filled with the electron-dense gold nanoparticles (Figure 3A and B; turquoise arrowheads) with an average gold intensity of 1.29 ± 0.07 (luminal gold intensity normalized to 1.0; Figure 3G), further demonstrating that the BBB is functionally immature at 3 dpf. To decipher how the gold nanoparticles traverse from the lumen to the basement membrane, we first looked at the tight junctions to see if the nanoparticles passed between apposed endothelial cells. We consistently observed that the nanoparticles were halted at the ‘kissing points’ between endothelial cells, indicating that tight junctions were functional prior to formation of a functional BBB (49/49 functional tight junctions; Figure 3B; green arrowhead). The presence of functional tight junctions at this stage is in agreement with the previously reported expression of tight junction proteins ZO-1 and Claudin-5 in cerebral blood vessels at 3 dpf. As transcytosis has been implicated in the maturation of the blood-retinal barrier, we next examined the levels of luminal and abluminal flask-shaped vesicles filled with gold nanoparticles as a means of assessing transcytosis. Quantification of gold-filled luminal and abluminal flask-shaped vesicles revealed an average of 0.21 ± 0.03 and 0.17 ± 0.02 vesicles/µm, respectively (Figure 3H and I). These data reveal that the basement membrane becomes filled with gold nanoparticles via vesicular transport rather than intercellular passage between immature tight junctions.

11When we repeated this assay at 5 dpf, when the barrier becomes less permeable to fluorescently conjugated tracers (Figures 1 and 2), we still observe close contacts between endothelial cells and pericytes (Figure 3C). At 5 dpf, the basement membrane was noticeably lacking gold particles as compared to 3 dpf (Figure 3D), with a reduced average gold intensity of 0.85 ± 0.02 (Figure 3G). Like at 3 dpf, the tight junctions remained functional at 5 dpf, based on their capacity to halt gold nanoparticles at the kissing points between neighboring endothelial cells (76/76 functional tight junctions; Figure 3D). However, the levels of vesicles both luminally and abluminally were notably decreased to 0.08 ± 0.01 and 0.07 ± 0.01 vesicles/µm, respectively (p<0.001 (luminal) and p<0.0001 (abluminal), nested two-way ANOVA; Figure 3H and I). At 7 dpf, the neurovascular cellular interactions remained constant with endothelial cells in close contact with pericytes and neurons, and with an unfilled basement membrane (average gold intensity of 0.91 ± 0.02; Figure 3E and G). Importantly, the tight junctions remained functional (51/51 functional tight junctions), and the low vesicular densities observed at 5 dpf remained comparably low at 7 dpf with 0.07 ± 0.01 vesicles/µm, both luminally and abluminally (Figure 3F, H and I). Taken together with the fluorescent tracer data, our data suggests that the zebrafish BBB becomes functionally impermeable at 5 dpf via the suppression of vesicular trafficking.

12Given the important role of suppressing transcytosis in determining the developmental maturation of the mouse BBB, and a similar correlation of the timing between the suppression of transcytosis and functional barrier formation in zebrafish, we wondered whether the key mammalian barrier regulator Mfsd2a, which suppresses caveolae mediated transcytosis, plays a conserved role in zebrafish. Zebrafish contain two paralogues of Mfsd2a, mfsd2aa and mfsd2ab, both of which are expressed in the developing zebrafish CNS. In order to resolve the vascular expression of both paralogues, we performed fluorescent in situ hybridization (FISH) in kdrl:mCherry transgenic fish sections at 3 and 5 dpf and in adult brain sections. To control for off-target signal, we also performed FISH with a mouse Gfap probe, which should not bind to the zebrafish transcript (Figure 4). Vascular expression levels for mfsd2aa or mfsd2ab were then background corrected using the measurements from the Gfap probe. As a positive control, we also performed FISH for the BBB-specific vascular marker slc2a1a (also known as glut1b), and observed consistent slc2a1a signal in cerebral vessels at all time points examined (Figure 4—figure supplement 1). FISH revealed undetectable levels of mfsd2aa expression in cerebral vessels at 3 dpf, higher levels at 5 dpf, when BBB endothelial cells suppress transcytosis (Figure 3), and intermediate levels in adult blood vessels (Figure 4). Mfsd2ab, on the other hand, is highly expressed in cerebral vessels at 3 and 5 dpf and lowly expressed in adult vessels (Figure 4). Mfsd2aa is 61% identical to human MFSD2A and 62% identical to mouse Mfsd2a (Figure 4—figure supplement 2A). Mfsd2ab is 64% identical to human MFSD2A and mouse Mfsd2a (Figure 4—figure supplement 2A). The two paralogues are only 68% identical to each other, but they both contain the lipid binding domain that is critical for governing barrier properties (Figure 4—figure supplement 2A; ). Given the lack of a clear paralogue that most closely resembles Mfsd2a, we generated CRISPR mutants for both paralogues independently. Mfsd2aahm37/hm37 mutants have a 7 bp deletion in exon 2 (Figure 4—figure supplement 2B) that is predicted to lead to a premature stop codon at amino acid 82 (Figure 4—figure supplement 2A; black box) and exhibit reduced levels of mfsd2aa mRNA (Figure 4—figure supplement 2C). Homozygous mfsd2aa mutants are viable and fertile. While mfsd2aa mutants appear in Mendelian ratios at 5 dpf (105/408 fish), a reduced number of mfsd2aa mutants survive to adulthood (44/248 fish). Mfsd2abhm38/hm38 mutants have a 19 bp deletion in exon 5 (Figure 4—figure supplement 2B) that is predicted to lead to a premature stop codon at amino acid 175 (Figure 4—figure supplement 2A; black box) and exhibit reduced levels of mfsd2ab mRNA (Figure 4—figure supplement 2C). Homozygous mfsd2ab mutants are also viable and fertile and appear in Mendelian ratios both as larvae (51/215 fish) and in adulthood (33/124 fish). Neither mutant displayed obvious angiogenic defects, similar to the normal vasculature observed in mouse Mfsd2a mutants. In addition to BBB defects, Mfsd2a knockout mice also exhibit microcephaly. However, neither mutant displayed decreased larval brain size.

13To investigate whether either paralogue was necessary for barrier formation, we performed NHS tracer injection assays as we did to identify the developmental timeline of barrier formation and assayed for BBB leakage at 5 dpf. In addition to the use of the exogenous injected fluorescent tracer (1 kDa NHS), we also assayed the leakage of the endogenous transgenic serum DBP-EGFP fusion protein (Tg(l-fabp:DBP-EGFP)). Mfsd2aa mutants displayed at least a two-fold increase in midbrain and hindbrain parenchymal tracer leakage, both for the injected 1 kDa NHS and the endogenous serum transgene 80 kDa DBP-EGFP, at 5 dpf compared to wild-type sibling controls (Figure 5). These data suggest that mfsd2aa plays a similar role to mouse Mfsd2a in determining barrier properties. Conversely, mfsd2ab mutants displayed similarly low levels of parenchymal tracer intensity to wild-type controls at 5 dpf (Figure 5—figure supplement 1), indicating that mfsd2ab is dispensable for functional barrier formation. Similarly, when we injected different guides to target either mfsd2aa or mfsd2ab and assessed BBB permeability in F0 crispants, we observed increased BBB permeability in mfsd2aa crispants but not mfsd2ab crispants (Figure 5—figure supplement 2), confirming the mutant phenotypes. To see if the loss of both paralogues resulted in increased barrier permeability, we investigated mfsd2aahm37/hm37; mfsd2abhm38/hm38 double mutants for BBB permeability using the endogenous serum DBP-EGFP fusion protein (Figure 5—figure supplement 3). While mfsd2aa single mutants displayed the previously observed increase in barrier permeability and mfsd2ab single mutants did not, mfsd2aahm37/hm37; mfsd2abhm38/hm38 double mutants displayed similar levels of increased barrier permeability to the mfsd2aa single mutants (Figure 5—figure supplement 3). These data correspond with previously reported mfsd2aa morpholino knockdown leading to increased BBB permeability. However, neither our mfsd2aa or mfsd2ab mutants display brain hemorrhage or increased embryonic lethality, as reported in the morphants. Taken together, these data suggest that the two paralogues, while structurally similar, play different roles, with zebrafish mfsd2aa and mammalian Mfsd2a sharing a conserved role in the BBB.

14Since mfsd2aa mutants contain a leaky BBB (Figure 5), we wanted to further investigate the dynamic nature of the leakage in mfsd2aa mutants using time lapse live imaging to determine how tracers enter the mutant brain and whether or not this would mirror the leakage observed in larval fish (Figure 6). In 9 mfsd2aa heterozygote control fish (Figure 6—Videos 1 and 2), we observed similarly low rates of Dextran accumulation to that observed in wild-type 5 dpf fish (Figure 2). In 9 mfsd2aa mutant fish (Figure 6—Videos 3 and 4), we also never observed any bursting directly from blood vessels as sporadically observed in 3 dpf fish (Figure 2). However, we uncovered a significantly increased rate of diffuse Dextran accumulation in the brain parenchyma in mfsd2aa mutants (slope of 0.098 intensity/min) compared to heterozygote controls (slope of 0.019 intensity/min; Figure 6). Interestingly, the increased Dextran permeability observed in mfsd2aa mutants closely resembles the rate and overall levels of Dextran permeability of the immature 3 dpf larvae (Figure 2).

15Given the leakage phenotype in mfsd2aa mutants at 5 dpf, we next wanted to examine whether the leakage phenotype persisted into adulthood. To address this, we assessed the localization of the serum DBP-EGFP fusion protein in adult brains sections of mfsd2aa mutant and wild-type controls. As expected, the wild-type siblings retained the DBP-EGFP within their blood vessels throughout the brain (Figure 7). However, mfsd2aa mutants exhibited DBP-EGFP extravasation into the brain parenchyma (Figure 7), suggesting that the leakage phenotype was not limited to larval fish. Finally, to determine whether this increased permeability was due to increased transcytosis as in Mfsd2a knockout mice, we measured vesicular density in capillaries with luminal diameters less than 5 µm from adult mutant and wild-type siblings using TEM. The mfsd2aa mutant blood vessels appeared morphologically normal by TEM, composed of a thin single layer of endothelial cells in close contact with pericytes, as observed in their wild-type siblings (Figure 8A and B). A closer examination of the endothelial cells revealed electron-dense tight junction complexes between all apposed endothelial cells in both wild-type and mfsd2aa mutant fish (Figure 8C and D). However, while wild-type fish display similarly low levels of luminal (0.1 vesicles/µm) and abluminal (0.11 vesicles/µm) vesicular densities to those observed at 7 dpf (Figure 8C–F), mfsd2aa mutant fish display a significant increase in luminal and abluminal vesicular densities (0.29 and 0.27 vesicles/µm, respectively) compared to wild-type siblings (Figure 8C–F). Interestingly this increase in vesicular abundance is even higher than that observed during early barrier development at 3 dpf (Figure 3). These data also suggest that the increased tracer leakage observed in mfsd2aa mutants results from an increase in vesicular trafficking across the BBB, further supporting a conserved role for mfsd2aa in determining barrier properties. In contrast to mfsd2aa mutants, mfsd2ab mutants displayed similar levels of abluminal and luminal vesicular pit density to wild-type siblings (Figure 8—figure supplement 1), further demonstrating that the paralogue mfsd2ab does not play a conserved role in determining barrier properties in zebrafish.

16One of the major advantages to studying the BBB in zebrafish is the ability to perform live imaging of tracer permeability dynamics. We provide some of the first data on the dynamics of immature barrier leakage during early development and of barrier leakage seen in a genetic mutant fish at the time when the BBB should be functionally mature. At 3 dpf, we observed two types of leakage, the gradual overall increase in the parenchyma and rare transient bursts, both from a migrating sprout and established blood vessels. We attribute some of the gradual diffuse increase in parenchymal tracer uptake to the high levels of vesicular trafficking observed in our TEM analyses in which the endothelial basement membrane became filled with gold nanoparticles. Time lapse analysis of mfsd2aa mutants revealed similar diffuse tracer leakage levels and rates to that observed in the leaky 3 dpf larvae. Since mfsd2aa mutant fish have normal angiogenesis but elevated endothelial transcytosis levels, the Dextran leakage at 5 dpf, when their siblings have a functional BBB, provides a direct measure of the in vivo rates and mode of BBB leakage due to unsuppressed transcytosis. The rarer bursts of leakage, on the other hand, could be due to fleeting tight junctional ruptures between neighboring endothelial cells rather than increased vesicular trafficking. As these events were extremely scarce during our time lapse imaging, it would be nearly impossible to capture these potential junctional breaches by TEM to confirm this hypothesis. However, with continuing advances in the resolution of fluorescence microscopy, we could use zebrafish to resolve whether these two types of leakage, gradual and bursting, proceed through the same or different subcellular routes into the brain parenchyma, either paracellularly or transcellularly.

17While the BBB constitutes the largest interface for blood-brain exchange in adults, and represents the most well-studied cellular barrier, it is not the only brain barrier. The blood-cerebrospinal fluid (CSF) barrier is created by the polarized epithelial cells of the choroid plexus (CP) that projects directly into the brain ventricle and creates a barrier between the fenestrated capillaries of the CP and the CSF. During development, these CP epithelial cells become selectively less permissive by expressing tight junction proteins, including Claudin-5 and Occludin, and nutrient transporters, including Mrp1, Mdr1 and Oat3. In zebrafish, this blood-CSF barrier has been shown to become completely impermeable by 4 dpf to various molecular weight tracers including 3 kDa fluorescein, 10 kDa and 40 kDa Dextrans. However, while CP epithelial cells are initially permeable to 3 kDa fluorescein and 10 kDa Dextran at 2 dpf, they never allow the passage of the larger 40 kDa Dextran. Since we saw comparable BBB permeability for all tracers used during the developmental time course, including the 80 kDa DBP-EGFP serum transgene, this strongly suggests that the tracers entered the brain predominantly through leaky BBB vessels rather than through an immature blood-CSF barrier. The avascular arachnoid epithelium directly beneath the dura that completely ensheaths the CNS acts as a third barrier between the extracellular fluids of the CNS and the rest of the body. Previous HRP tracer injections have shown that the zebrafish arachnoid barrier matures after 9 dpf. Since we did not observe any differences in tracer impermeability between 5 and 10 dpf, when the arachnoid barrier is still leaky, this suggests that the immature arachnoid barrier does not significantly contribute to the observed tracer extravasation into the brain earlier in development. Finally, the similarities between the observed amounts and dynamics of leakage in 3 dpf immature brains and mfsd2aa mutant brains at 5 dpf, which contain BBB-specific endothelial transcytosis defects, further suggests that the observed tracer leakage at 3 dpf is predominantly arising through the BBB. Taken together, while alternative routes for tracer entry into the brain exist, including an immature blood-CSF or arachnoid barrier, our methods allowed us to specifically evaluate BBB functionality, and other routes are not significantly contributing to the tracer leakage we observed in the larval brain.

18Among the various clathrin-independent transcytotic pathways, caveolae are particularly abundant in vascular endothelial cells. Caveolae are caveolin-coated 50–100 nm flask-shaped invaginations of the plasma membrane. Furthermore, the suppression of caveolae-mediated transcytosis regulates the development and function of both the mouse blood-retinal barrier and the BBB. Taken together, this suggests that the increased transcytosis during early larval stages is most likely caveolae-mediated. Interestingly, the linear profile of Dextran uptake in the midbrain at 3 dpf closely resembles the caveolae-mediated uptake of a fluorescently conjugated aminopeptidase P (APP) antibody in the mouse lung. While similarly linear, the scale is on the order of minutes in the larval zebrafish brain versus seconds in the mouse lung. This discrepancy in timing is most likely due to the large difference in the vesicular densities between the immature BBB endothelium (average of 2 vesicles/µm2) and the continuous endothelium of peripheral tissues, which ranges from 30 to 98 vesicles/µm2 on the luminal membrane of diaphragm and myocardial endothelium. Since loss of caveolin 1 (Cav1) is sufficient to precociously seal the mouse blood-retinal barrier at the angiogenic front, it would be interesting in future work to examine whether zebrafish cav1 mutants, which are viable as homozygotes, also exhibit an earlier onset of BBB maturation.

19BBB permeability is tightly regulated by endothelial cell interactions with pericytes and astroglial cells. For the first time, we visualized their locations under TEM in zebrafish throughout barrier development. Pericytes have been shown to be essential for establishing the mammalian BBB. Similarly, pericyte-deficient notch3fh332 fish display increased BBB permeability in a tight junction independent manner like pericyte-deficient mice. Our TEM data show that zebrafish pericytes are in close contact with brain endothelial cells (Figures 2 and 5), even at the earliest stage examined (3 dpf; Figure 2), and are embedded within the endothelial basement membrane as in mammals. Our subcellular localization data is in line with a growing body of evidence for the conserved role and molecular profile of pericytes in the zebrafish BBB.

20Taken together this study provides a thorough characterization of the development of the zebrafish BBB, highlighting regional differences in timing of functional maturation and capturing the dynamics of the immature BBB. Furthermore, our developmental TEM series demonstrates the role of vesicular trafficking in regulating zebrafish BBB function during development. We have also shown that this down-regulation of vesicular trafficking is necessary for functional BBB formation, as mfsd2aa mutants display increased barrier permeability due to unsuppressed transcytosis. Finally, our time lapse imaging of mfsd2aa mutants provides the first direct measure of tracer permeability in vivo that results from increased transcytosis. We hope that this work will serve as a launching pad for future studies using zebrafish to understand the molecular regulators of BBB development and homeostasis in vertebrates.

21Zebrafish were maintained at 28.5°C following standard protocols. All zebrafish work was approved by the Harvard Medical Area Standing Committee on Animals under protocol number 04487. Adult fish were maintained on a standard light-dark cycle from 9 am to 11 pm. Adult fish, age 3 months to 2 years, were crossed to produce embryos and larvae. For imaging mutant larvae, 0.003% phenylthiourea (PTU) was used beginning at 1 dpf to inhibit melanin production. These studies used the AB and Casper (mitfaw2/w2; mpv17a9/a9) wild-type strains and the transgenic reporter strains Tg(l-fabp:DBP-EGFP)Iri500, Tg(kdrl:HRAS-mCherry)s896, abbreviated as Tg(kdrl:mCherry) in the text, Tg(actb2:MA-Citrine)hm26 and Tg(actb2:Hsa.H2B-Scarlet)hm38.

22Larvae were immobilized with tricaine and placed in an agarose injection mold with their hearts facing upwards. 2.3 nl of Alexa Fluor 405 NHS Ester (Thermo Fisher: A30000) or Alexa Fluor 647 10 kDa Dextran (Thermo Fisher: D22914) fluorescently conjugated tracers (10 mg/ml) were injected into the cardiac sac using Nanoject II (Drummond Scientific, Broomall, PA). Larvae were then mounted with 1.5% low gelling agarose (Sigma: A9414) in embryo water on 0.17 mm coverslips and imaged live within 2 hr post-injection. For developmental electron microscopy experiments, 2.3 nl of 5 nm NHS-activated gold nanoparticles (Cytodiagnostics: CGN5K-5–1, ~1.114 particles/ml in PBS) were injected into the cardiac sac just as for the fluorescently conjugated tracers. After 5 min of circulation, the fish were fixed for electron microscopy. The brains of transgenic DBP-EGFP adults were fixed in 4% paraformaldehyde in PBS overnight at 4°C. Following fixation the brains were washed three times and sagittally sectioned with a cryostat (30 µm) to reveal DBP-EGFP localization within the brain.

23Fish were anesthetized with tricaine and initially fixed by immersion in 4% paraformaldehyde (VWR:15713 s)/0.1M sodium-cacodylate (VWR:11653). Following this initial fixation, the larval fish and adults with exposed brains were further fixed for 7–14 days in 2% glutaraldehyde (Electron Microscopy Sciences: 16320)/4% paraformaldehyde/0.1M sodium-cacodylate at room temperature. Following fixation, larvae or dissected brains were washed overnight in 0.1M sodium-cacodylate. Entire larval heads or coronal vibratome free-floating sections of adult brains (50 µm) were post-fixed in 1% osmium tetroxide and 1.5% potassium ferrocyanide, dehydrated, and embedded in epoxy resin. Ultrathin sections of 80 nm were then cut from the block surface and collected on copper grids. The adult sections were counter-stained with Reynold’s lead citrate prior to imaging.

24Mfsd2aa mutant fish were generated by injection of Cas9 RNA and a single guide RNA (5’-GGTGTGTTTTGCGATCGGAG-3’) targeting exon 3 into single-cell fertilized AB wild-type embryos. Mfsd2ab mutant fish were generated by injection of Cas9 protein and a single guide RNA (5’-TGAGAGCAGAGTAGGGCACG-3’) targeting exon 5 into single-cell fertilized double transgenic Tg(l-fabp:DBP-EGFP; kdrl:mCherry) embryos. F0 injected fish were raised, outcrossed to wild-type fish and screened for potential mutant founders by PCR and sequencing. Founders were outcrossed to wild-type fish at least three times before leakage analyses. Heterozygous fish were intercrossed for all leakage assays, with the exception of the mfsd2aa time lapse microscopy, where a heterozygous fish was crossed to a homozygous mutant. All larvae were imaged prior to genotyping to identify wild-type and mutant fish. The stable mfsd2aa mutant line was genotyped using 5’-AAATCACCTCTTCCAGTGAGGA-3’ and 5’-ATAGTAACAAAACGATGCTGAGCC-3’ primers followed by a Pvu1 restriction digest, which does not digest the mutant DNA. Mfsd2ab mutants were genotyped using 5’-GTCTACTCCATTTGCTGTACTTTGC-3’ and 5’-AGAGCAGAGTAGGGCACGTGGAAGC-3’ primers for the wild-type allele and 5’-GTCTACTCCATTTGCTGTACTTTGC-3’ and 5’-GTGCTGATGAAGCACTGATGAAGCA-3’ for the mutant allele.

25Mfsd2aa crispant fish were generated by injection of Cas9 protein and two new guide RNAs (5’- GCTCACTAAAAACCCAACGG-3’ and 5’-TTGTAGAAAGAAGCCCAGGG-3’) targeting exons 2 and 3 into single-cell fertilized double transgenic Tg(l-fabp:DBP-EGFP; kdrl:mCherry) embryos. Similarly, mfsd2ab crispants were generated by injection of Cas9 protein and two new guide RNAs (5’-AAGCGAGATGGCAAAAGGAG-3’ and 5’- ACCCACAAACAAGATGATGG-3’) targeting exons 1 and 3 into single-cell fertilized double transgenic Tg(l-fabp:DBP-EGFP; kdrl:mCherry) embryos. F0 crispants were analyzed for leakage at 5 dpf.

26qPCR was performed on cDNA obtained from wild-type, mfsd2aahm37, mfsd2abhm38, mfsd2aa and mfsd2ab F0 crispant 5 dpf whole larvae mRNA. An Eppendorf Realplex2 Real-Time PCR system was used for qPCR experiments and gene expression levels were normalized relative to that of a reference gene, 18s using primers 5’-TCGCTAGTTGGCATCGTTTATG-3’ and 5’-CGGAGGTTCGAAGACGATCA-3’. Levels of mfsd2aa were measured using 5’-CTCTTCACTTCGCTAGCCTTCATG-3’ and 5’-CGATGTAAACAGCAGTCTTTTTCCC-3’ primers and mfsd2ab levels were measured with 5’-TCTCGACTCTTAGTCTTGATTTCGC3’ and 5’-GAGTCCGTTTCTGAATCCATCTCG-3’. All reactions were performed in technical duplicates, and the results represent four independent biological samples including the SEM.

27HCR RNA in situ hybridization (Molecular Instruments) experiments on 14 µm cryosections of fixed 3 and 5 dpf larvae and adult brains were performed as previously described. Briefly, sections were air dried and re-fixed in 4% paraformaldehyde in PBS for 10 min at room temperature. Following fixation, sections were further permeabilized using 1 µg/ml Proteinase K (ThermoFisher) for 5 min, followed by PBS washes. Slides were then put through an ethanol dehydration series of 50%, 70% and two rounds of 100% ethanol and air dried for 5 min. Dried slides were put into pre-hyb solution for at least 10 min at 37°C. Subsequently, the probes for mfsd2aa, mfsd2ab, slc2a1a and mouse Gfap were added to the slides at a final concentration of 4 nM to hybridize overnight at 37°C. The next day, slides were washed with a series of wash buffer to 5x SSCT (5x SSC with 0.1% Tween 20). Excess liquid was then removed and samples were immersed in amplification buffer at room temperature for 30 min prior to hairpin amplification, which occurred overnight at room temperature. The next day, slides were washed in 5x SSCT and mounted with Fluoromount-G (Electron Microscopy Sciences).

28All imaging of tracer permeability was performed on a Leica SP8 laser scanning confocal microscope using the same acquisition settings using a 25x water immersion objective. Time lapse imaging was performed on the SP8 using a resonance scanner. FISH sections were also imaged with the Leica SP8 using a 63x oil immersion objective. A 1200EX electron microscope (JOEL) equipped with a 2 k CCD digital camera (AMT) was used for all TEM studies. Images were visualized and quantified using ImageJ (NIH) and Adobe Photoshop. Time lapse videos were visualized as 3D reconstructions and cropped to highlight particular blood vessels of interest using Imaris software (Bitplane).

29All quantifications were performed on blinded image sets. For static images, parenchymal fluorescent tracer intensity was measured using ImageJ in the entire regional parenchyma outside of the vasculature in 30 µm thick maximum intensity projections of the larval brains. These projections began on average 15 µm below the mesencephalic vein to reduce the effects of potential leakage diffusion from the surface vessels and had the vasculature masked and removed for intensity quantification. These parenchymal tracer intensity values were then normalized to the tracer intensity within the vasculature to account for differential amounts of circulating tracer. For time lapse imaging, Dextran intensity was measured in 6 parenchymal regions of average intensity projections of the time lapse videos and averaged as a single value per fish. These values were normalized to the average fluorescence intensity in the lumen at each time point. Z-stacks were collected with 1 µm z-steps using a 25x water immersion objective on a Leica SP8 laser scanning confocal microscope.

30For all TEM quantifications, vesicular density values were calculated from the number of non-clathrin coated flask-shaped vesicles per µm of endothelial luminal or abluminal membrane for each image collected. Larval basement membrane gold intensity was measured in 3 regions per endothelial cell and normalized to luminal gold intensity values within individual images using ImageJ. All images for analysis were collected at 12,000 x magnification on the JOEL 1200EX electron microscope. 10–15 vessels were quantified for each fish, with each color representing a different fish.

31For all FISH experiments, we quantified the number of gene expression puncta within the vasculature of maximum intensity projections using ImageJ. These puncta were normalized to total blood vessel area within the image. Non-specific background levels of mouse Gfap probe binding were similarly assessed at all three time points. These background levels were then subtracted from mfsd2aa and mfsd2ab expression levels and negative values were set to 0.

32All statistical analyses were performed using Prism 8 (GraphPad Software). Two group comparisons were analyzed using an unpaired two-tailed t test. Nested t tests were employed for all electron microscopy comparisons to account for the actual N versus vessels analyzed. Multiple group comparisons were analyzed with two-way ANOVA, followed by a post hoc Tukey’s multiple comparison test. The time lapse leakage dynamics were analyzed with a Mann-Whitney U test to discriminate whether the two patterns of leakage accumulation were different. Sample size for all experiments was determined empirically using standards generally employed by the field, and no data was excluded when performing statistical analysis. Standard error of the mean was calculated for all experiments and displayed as errors bars in graphs. Statistical details for specific experiments, including exact n values and what n represents, precision measures, statistical tests used, and definitions of significance can be found in the Figure Legends.