Ciliary Beating Compartmentalizes Cerebrospinal Fluid Flow in the Brain and Regulates Ventricular Development.

Motile cilia are miniature, propeller-like extensions, emanating from many cell types across the body. Their coordinated beating generates a directional fluid flow, which is essential for various biological processes, from respiration to reproduction. In the nervous system, ependymal cells extend their motile cilia into the brain ventricles and contribute to cerebrospinal fluid (CSF) flow. Although motile cilia are not the only contributors to CSF flow, their functioning is crucial, as patients with motile cilia defects develop clinical features, like hydrocephalus and scoliosis. CSF flow was suggested to primarily deliver nutrients and remove waste, but recent studies emphasized its role in brain development and function. Nevertheless, it remains poorly understood how ciliary beating generates and organizes CSF flow to fulfill these roles. Here, we study motile cilia and CSF flow in the brain ventricles of larval zebrafish. We identified that different populations of motile ciliated cells are spatially organized and generate a directional CSF flow powered by ciliary beating. Our investigations revealed that CSF flow is confined within individual ventricular cavities, with little exchange of fluid between ventricles, despite a pulsatile CSF displacement caused by the heartbeat. Interestingly, our results showed that the ventricular boundaries supporting this compartmentalized CSF flow are abolished during bodily movement, highlighting that multiple physiological processes regulate the hydrodynamics of CSF flow. Finally, we showed that perturbing cilia reduces hydrodynamic coupling between the brain ventricles and disrupts ventricular development. We propose that motile-cilia-generated flow is crucial in regulating the distribution of CSF within and across brain ventricles.

Introduction

Cilia are microscopic hair-like structures that extend from the surface of various cell types across the animal kingdom [1]. Although immotile primary cilia are sensory organelles [2], motile cilia act as nanomachines that move entire cells in their environment or fluids along epithelial surfaces in multicellular organisms [3]. In vertebrates, motile cilia are involved in the establishment of left-right asymmetry [4, 5], the removal of debris from the lungs [6], and the movement of fluid in the kidney ducts [3, 7]. Motile cilia cover a range of structures in the nervous system, including the nasal cavity [8, 9], the central canal of the spinal cord [10, 11, 12, 13], and the brain ventricles [12, 14, 15, 16]. Motile-cilia-generated fluid flow is thought to be important for brain development and physiology, as human ciliopathy patients and animals lacking motile cilia develop neurological features, including aberrant sensory computations [9], defective spine curvature [17], and hydrocephalus [3, 15].

In the brain ventricles, the cerebrospinal fluid (CSF) flow, generated by the ciliary beating of ependymal cells, transport nutrients and signaling molecules and remove waste products [15]. CSF flow is also suggested to support neural stem cell proliferation [18] and directional neuronal migration [19]. Motile-cilia-mediated flow patterns in the brain ventricles are surprisingly complex and proposed to be regulated by the circadian rhythm [14] and brain-derived neuropeptides [20]. Moreover, recent work showed that the composition of CSF varies across the ventricles due to the distinct transcriptomes of the CSF-producing organs known as choroid plexuses [21, 22]. Such regionalization was suggested to promote different differentiation programs along the rostro-caudal axis of the brain [21]. All these findings indicate that the nervous system can specifically regulate both the compartmentalization of CSF within the ventricles and the exchange of signaling molecules, nutrients, and waste across brain regions. However, it remains poorly understood how CSF flow in the brain is generated and how motile ciliary beating and other physiological factors contribute to the regulation of such CSF flow.

In this study, we took advantage of the transparent brain of a small vertebrate, the zebrafish larva [23], to understand how ciliary beating generates CSF flow and contributes to brain development. We observed that motile cilia-bearing cells appear at around 32 hr post-fertilization (hpf) in specific locations of the brain ventricular system. Next, we found that motile ciliary beating generates a directional CSF flow, which is restricted within individual ventricular cavities. Interestingly, we observed that other physiological factors, such as heartbeat and bodily movements, contribute to the exchange of CSF across brain ventricles through the connecting ducts. We also showed that motile ciliary beating and the hydrodynamic forces generated by the heartbeat and bodily movements work in parallel but on different components of the CSF flow, thereby regulating the distribution of CSF across the brain. Finally, we revealed that the genetic ablation of ciliary function and motility lead to aberrant development and organization of the ventricular system.

Results

Motile Ciliated Cells Line the Ventricles of the Developing Zebrafish Brain

From fish to mammals, the brain ventricles of vertebrates are decorated with motile ciliated cells [12, 14, 15, 16, 17, 24, 25, 26, 27]. It is, however, less clear when these motile ciliated cells appear during development and whether they occupy specific spatial locations, which could facilitate their function. In order to identify the developmental time point and the specific location of motile ciliated cells in the zebrafish brain, we performed in situ hybridization for the master regulator of motile ciliogenesis Foxj1 (Figures 1A and 1A’), a commonly used marker of motile ciliated cells [28, 29, 30], including ependymal cells [27, 31, 32]. In zebrafish, the Foxj1 gene is duplicated into two orthologs: foxj1a and foxj1b [33, 34, 35]. As the ventricular system undergoes changes during brain development [24, 36, 37], we analyzed developmental stages, from 30 hpf, when ventricles are wide (Figure 1B), to 4 days post-fertilization (dpf), when ventricles are compact and embedded tightly in the brain (Figure 1B’). We observed that the ventricular system is compartmentalized into three cavities connected by ducts (Figures 1B and 1B’), constituting the telencephalic ventricle (TV), the diencephalic-mesencephalic ventricle (hereafter referred to as the diencephalic ventricle [DV]) [24, 36], and the rhombencephalic ventricle (RV). In situ hybridization indicated that foxj1a and foxj1b are expressed in distinct yet partially overlapping cell populations. We observed that foxj1a labels cells in the ventral part of the diencephalic ventricle along the midline (Figures 1A and 1A’). In contrast, foxj1b labels cells on the dorsal part of the diencephalic ventricle from 2 dpf (Figure 1A) and in the choroid plexus at 4 dpf (Figure 1A’). Both foxj1a and foxj1b are expressed in the subcommissural organ (Figure 1).

Figure 1

Motile Ciliated Cells Decorate the Ventricles of the Developing Zebrafish Brain

(A and A’) foxj1a and foxj1b genes, which are the two master regulators of motile ciliogenesis, are expressed in different populations of cells in the developing brain as shown by in situ hybridization in 2 (A) and 4 dpf (A’) zebrafish larvae. The foxj1a-expressing cells are situated along the ventral midline (white arrow) of the optic tectum (OT), while foxj1b expression is observed rostrocaudally along the dorsal midline (black arrowhead) of the diencephalon at 2 (A) and 4 dpf (A’). Both foxj1a and foxj1b are expressed in the subcommissural organ (sco), although only foxj1b is expressed in the choroid plexus (cp) at 4 dpf (A’; n = 28; n = 13; n = 13; and n = 8).

(B and B’) T2BGSZ10 Tg(foxj1b:gfp) transgenic larvae show GFP-labeled ciliated cells on the dorsal wall of the diencephalic ventricle (white arrowhead). GFP-positive cells are in direct contact with the brain ventricles labeled by ventricular injection of 70-kDa Rhodamine B isothiocyanate (RITC)-dextran (magenta). Confocal microscopy image of the sagittal midline of anesthetized larvae at 2 (B) and 4 dpf (B’) is shown. The zebrafish ventricular system is composed of three cavities: the telencephalic ventricle (TV); the mesencephalic/diencephalic ventricle (DV); and the rhombencephalic ventricle (RV), which are connected by ducts. Although the compartmentalization of the ventricular system persists from 2 to 4 dpf, the size of the cavities decreases over development (n2 dpf = 7 and n4 dpf = 11).

(C and C’) GFP-positive cells extend their cilia into the ventricular cavity in both dorsal and ventral regions of the diencephalon, as shown by immunostaining with a glutamylated tubulin antibody (magenta) and confocal microscopy. Nuclei are labeled with DAPI (blue; n2 dpf = 8 and n4 dpf = 8).

Scale bars are 50 μm. cp, choroid plexus; OT, optic tectum; pg, pineal gland; sco, subcommissural organ. See also Figure S1 and Video S1.

To label motile ciliated cells in vivo, we used the T2BGSZ10 transgenic zebrafish line, which has an insertion of GFP in the endogenous foxj1b gene [33]. Confocal images revealed that this zebrafish line labeled the cells along the dorsal wall of the diencephalic ventricle (Figures 1B, 1B’, S1A, and S1B; Video S1), overlapping with our in situ hybridization results. We also showed that GFP-positive cells in the T2BGSZ10 (foxj1b:gfp) line are in contact with the ventricles (Figures 1B and 1B’) and harbor cilia that are enriched in the ciliary marker glutamylated tubulin (Figures 1C and 1C’) [38, 39]. Interestingly, we did not observe cells bearing large brushes of cilia, as seen in other multiciliated epithelia [9, 12, 27], suggesting that these cells are not multiciliated at this developmental stage. For visualizing foxj1a-positive cells, we analyzed two different lines expressing GFP under the 0.6-kb [9] or 5.2-kb foxj1a promoters [17]. These transgenic lines did not label ventrally located cells, which were positive for foxj1a according to our in situ hybridization. Instead, these lines labeled the dorsally located cells in the diencephalic and telencephalic ventricle and the subcommissural organ, overlapping with the expression pattern of the foxj1b:gfp transgenic line (Figures S1C–S1D’’, Video S1, and data not shown). Therefore, we used only the foxj1b:gfp transgenic line, which showed a reliable expression pattern in all our investigations. Interestingly, we observed that glutamylated tubulin does not label all cilia in the brain but specifically the cilia extending to the ventricles from foxj1b:gfp-positive cells (Figures 1C, 1C’, S1E, and S1E’). We also observed glutamylated tubulin staining in the ventral diencephalic ventricle, which corresponds to the location of the foxj1a in situ hybridization pattern (Figures 1A, 1A’, 1C, and 1C’). Altogether, our results suggest that there are at least two populations of ciliated cells in the brain ventricles of the larvae, with different spatial distribution and gene expression patterns for foxj1a or foxj1b.

Video S1. Analysis of T2BGSZ10 (foxj1b:gfp) and 0.6kbfoxj1a:gfp Transgenic Lines, Related to Figure 1

The ventricular system is composed of three interconnected cavities: the telencephalic ventricle, the diencephalic ventricle, and the rhombencephalic ventricle. The GFP positive cells in both the T2BGSZ10 (foxj1b:gfp) and Tg(foxj1a:gfp)nw6Tg transgenic lines are located along the dorsal wall of the diencephalic ventricle, as shown by 3D reconstruction of confocal scans obtained upon ventricular injection of 70 kDa RITC-dextran at 2 dpf.

The Cilia Located around the Ventricles Are Motile and Beat with a Rotational Movement

Our results suggest that cells located along the ventricles may likely bear motile cilia, as they express Foxj1 orthologs, which are prominent motile cilia markers. To investigate the motility of cilia at these locations, we performed light-sheet fluorescence microscopy in a transgenic zebrafish line, which expresses the ciliary protein Arl13b fused to GFP in the cilium of all cells (β-actin:arl13b-gfp) [40] (Videos S2 and S3). To determine the ciliary beating frequency (CBF), we applied a fast Fourier transform to the intensity time course of each pixel across the image sequence and reported the highest frequency peak as the CBF. Regions with robust CBF were located along the diencephalic ventricle (Figures 2A–2C) and on the dorsal wall of the telencephalic ventricle (Figure 2C), in the same locations where foxj1a and foxj1b were expressed. Our CBF analysis revealed distinct patches of frequencies corresponding to individual cilia (Figures 2A and 2B; Video S3). Upon quantification, we showed that CBF ranges from 15 Hz to 55 Hz with an average of 29.6 Hz at 2 dpf and 24.3 Hz at 4 dpf (Figures 2C, 2C’, S2A, and S2B). Interestingly, we observed a significant difference in CBF across brain regions (Figures 2C’ and S2C) and developmental stages (Figures S2B and S2C). Moreover, the beating angle and width of the ciliary beating along the dorsal wall of the diencephalic ventricle were significantly smaller at 4 dpf compared to 2 dpf (Figure S2F). Altogether, our data suggest that the properties of ciliary beating are different across ventricles. These differences may be related to ventricular size, because both the ventricular volume (Figures 1B and 1B’) and the ciliary beating properties change during development.

Figure 2

The Cilia Located along the Telencephalic and Diencephalic Ventricles Are Motile and Beat with a Symmetrical Movement

(A) Detection of ciliary beating along the dorsal and ventral walls of the DV of a 2-dpf β-actin:arl13b-gfp larvae, as shown by frequency analysis of a light-sheet microscopy recording. The ciliary beating frequency (CBF) was detected by fast Fourier transform and reported for each pixel of a 1,024-frame light-sheet recording (frame rate 100 Hz). The frequency of the highest peak in the frequency spectrum is reported as the CBF and color coded from 15 to 50 Hz. Recording is along the midline from a sagittal viewing angle.

(B) Recording (frame rate 400 Hz) along the dorsal wall of the diencephalic ventricle from a horizontal viewing angle reveals heterogeneous frequency patches, which correspond to individual cilia.

(C) Motile cilia were identified in the dorsal TV (green arrow), dorsal DV (blue arrow), and ventral DV (yellow arrow) upon light-sheet microscopy. Pie charts indicate the percentage of all analyzed cilia in different ventricular locations (color coded) at 2 dpf (n = 675 cilia from 9 larvae).

(C’) Quantification of beating frequencies from individual cilia at 2 dpf revealed small but significant differences between the TV and DV when comparing the mean of individual larvae; p value was calculated by the Wilcoxon rank-sum test. Mean ± SD is indicated for all data. The mean for each larva is indicated by a red cross.

(D–D’’) Quantification of the width and angle of ciliary beating, as represented on a manually reconstructed cilium (D). The width is slightly higher in sagittal recording compared to horizontal recording (D’), but not the angle (D’’). p value was calculated by the Wilcoxon rank-sum test. Mean ± SD is indicated.

Scale bars are 50 μm (A) or 10 μm (B). A, anterior; D, dorsal; L, left; P: posterior; R, right; V: ventral. See also Figure S2 and Videos S2 and S3.

Video S2. Cilia Located in the Dorsal and Ventral Walls of the Diencephalic Ventricle Are Motile, Related to Figure 2

Light-sheet recording (frame rate: 100 Hz) of the sagittal midline of a 2 dpf β-actin:arl13b-gfp transgenic larvae. The analysis of this recording is shown in Figure 2A.

Video S3. Motile Cilia Point toward an Anterior Direction and Beat with a Rotational Movement, Related to Figure 2

Light-sheet recording (frame rate: 400 Hz) along the dorsal wall of the diencephalic ventricle. Anterior to the left, posterior to the right. The analysis of this recording is shown in Figure 2B.

Most motile ciliated cells with a single cilium are shown to beat with a rotational movement [41, 42]. In contrast, multiciliated cells commonly display an asymmetric and planar beating pattern [9, 12]. To characterize the beating patterns and the beating direction of the motile cilia, we acquired video recordings of ciliary beating along the dorsal diencephalic ventricle from a horizontal and a sagittal viewing angle. The dorsally located cilia were oriented toward an anterior direction, demonstrating that ciliary beating is directional (Video S3). Next, we hypothesized that, if the ciliary beating is planar, both the angle and width of ciliary beating (Figure 2D) would be different for cilia when imaged from a horizontal and a sagittal viewing angle. However, we did not observe significant differences for the rotation angle of cilia between recordings obtained from a horizontal and a sagittal plane (Figures 2D’, 2D’’, and S2E), suggesting that the beating waveform is not planar.

To quantify the ciliary beating pattern more precisely, we imaged β-actin:arl13b-gfp larvae with sparsely labeled cilia at 2 dpf by light-sheet microscopy. Using a ciliary tracking software [43], we recovered the x, y, and relative z positions as well as the curvature angle for each time point and location along the cilium arc length (Figure 3A). Our results revealed that the changes in curvature angle (Figure 3B) and z position (Figure 3C) travel along the arc length of the cilium from the base to the tip in a very regular manner. These observations were further supported by the presence of only one major beating frequency along the entire arc length of all analyzed cilia (Figure 3D). Moreover, we observed that the initial segment of all cilia was more rigid compared to the distal end (Figure 3E). Finally, we predicted that, if the beating pattern were rotational, a given point on the arc length of the cilium would move on a circle when analyzed in the y and z dimensions. Hence, the relative y position of this given point would be zero when z is largest or smallest and conversely. This relationship would appear as a 90° phase shift between the y and z position oscillations for any point along the arc length of a rotating cilium. In line with our prediction, we observed a phase shift of 85.6° ± 27.6° between the y and z position oscillations (Figures 3F and 3G), providing further evidence that the beating of ventricular cilia is rotational. Altogether, our results demonstrate that motile ciliated cells in the developing brain are directionally organized and that their cilia beat with a rotational movement.

Figure 3

The Cilia Located along the Diencephalic Ventricle Beat with a Rotational Movement

(A) Time projection of a reconstructed cilium in x and y, obtained by SpermQ.

(B) Kymograph representing the curvature angle along the arc length over time for a representative cilium. The curvature angle indicates the angle between the tangential at a given point of the arc length and the tangential at another point located 1.5 μm upstream.

(C) Kymograph representing the width of the cilium along the arc length over time for a representative cilium. The width of the cilium is used to infer the z position according to the point spread function. The scale bar is in arbitrary units (a.u.), wherein the low values correspond to “in focus” and the high values imply “out of focus.”

(D) The same beating frequency is detected along the entire arc length of each cilium (n = 5) upon fast Fourier transform analysis of the curvature angle time course.

(E) The initial segment of the cilium is stiffer compared to the distal end of the cilium, as shown by the slowly increasing curvature angle amplitude along the arc length. Mean ± SD is indicated (n = 5).

(F) Curvature angle and z position over time at an arc length of 2.925 μm on the cilium. A phase difference (Δphase) is observed between curvature angle and z position for a representative cilium.

(G) Absolute phase differences for all analyzed cilia (85° ± 27°; n = 5). Mean ± SD is indicated.

Scale bar is 2 μm. A, anterior; P, posterior. See also Videos S2 and S3.

The Ventricular Flow Is Compartmentalized and Regulated by Multiple Factors, including Ciliary Beating, Heartbeat, and Bodily Movement

As directional beating of motile cilia is a common mechanism to generate fluid flow [4, 44, 45], we next characterized the properties of CSF flow in the larval brain. To visualize the CSF flow, we injected fluorescent particles into the ventricles of anesthetized larvae [24, 36] and recorded the movement of particles by confocal microscopy. Our recordings show that the ventricular CSF flow, which is established between 30 and 34 hpf (Figures S3A–S3A’’), is complex and includes multiple components (Video S4). To analyze the CSF flow, we used particle-image velocimetry (PIV), a commonly used tool in fluid dynamics [9, 46]. Our analysis revealed two major components of the CSF flow: a highly pulsatile movement located in the ducts and a directional flow along the ventricular walls (Figure 4). The directional flow runs posterior to anterior dorsally and anterior to posterior ventrally in the diencephalon (Figures 4A, 4C, and S3A’), as well as along the dorsal telencephalic wall (Figures S4A and S4B). Analysis of sequential recordings, taken at plane intervals of 5 μm, revealed that the directional flow is stronger at the midline along the ventral and dorsal walls of the diencephalic ventricle (Figures S4A and S4C; Video S6). In contrast, we observed that the CSF flow is present along most of the dorsal wall of the telencephalic ventricle (Figures S4A and S4B; Video S6).

Figure 4

Two Main Components Contribute to the CSF Flow of Larval Zebrafish: a Directional Flow near the Ventricular Walls and a Pulsatile Movement across the Ducts and in the Center of the Ventricle

CSF flow along the sagittal midline of the diencephalic ventricle (DV) is visualized by confocal microscopy upon ventricular injection of fluorescent beads.

(A and C) Particle image velocimetry (PIV) analysis revealed a directional flow along the walls at 2 (A) and 4 dpf (C).

(B and D) The pulsatile CSF movement is visualized by displaying the relative power (in %) at the most abundant periodic frequency identified by fast Fourier transform, equivalent to 2.27 Hz at 2 dpf (B) and 2.17 Hz in 4 dpf (D). This pulsatile CSF movement is prominent mainly near the two ventricular ducts and in the middle of the DV. Note that the pulsatile movement is less dominant at 4 dpf (D) than at 2 dpf (B).

(A’–D’) Upon ablation of heartbeat by microdissection of the heart, the directional, near-wall flow persists (A’ and C’), and the pulsatile CSF flow is absent (B’ and D’; n2 dpf = 22 and n4 dpf = 10).

Scale bars are 50 μm. See also Figures S3, S4, and S5 and Videos S4, S5, and S6.

Video S4. The CSF Flow Patterns of 2- and 4-dpf Larval Zebrafish, Related to Figure 4

The CSF flow of 2 and 4 dpf larval zebrafish is contributed by two main components; a directional, near-wall flow and a pulsatile flow, as shown by confocal imaging along the sagittal midline of the diencephalic ventricle upon ventricular co-injection of 70 kDa RITC-dextran and fluorescent particles (Φ = 160 nm). Acquisition frequency: 11.6 Hz.

Video S5. The CSF Flow Is Contributed by a Directional, Near-Wall Flow and a Pulsatile Flow, Related to Figure 4

Interfering with the heartbeat (upon heart microdissection) or ciliary beating (upon mutation of elipsa, smh and foxj1a) affects the CSF flow dynamics at 2 dpf, as shown by confocal imaging along the sagittal midline of the diencephalic ventricle upon ventricular co-injection of 70 kDa RITC-dextran and fluorescent particles (Φ = 160 nm). Acquisition frequency: 11.6 Hz.

Video S6. 3D Analysis of CSF Flow Patterns in the Telencephalic and Diencephalic Ventricles, Related to Figure 4

CSF flow patterns and pulsations in telencephalic ventricle (TV) and diencephalic ventricle (DV) were obtained by confocal microscopy upon ventricular injection of fluorescent particles (Φ = 160 nm) at 2 dpf. Sequential recordings were obtained every 10 μm from a horizontal viewing angle. The pulsatile CSF movement is visualized by displaying the relative power (in %) at the most abundant periodic frequency identified by Fast Fourier transformation (indicated in Hz). The flow fields were analyzed by particle image velocimetry (PIV). Left: position of the analyzed frame on a sagittal z projection of the ventricular system. Middle: pulsatile analysis. Right: PIV analysis. Scale bars are 50 μm.

To quantify the pulsatile component of the flow, we analyzed the recordings by using a fast Fourier transform and determined the most prominent frequency peak. We identified that the pulsatile flow presented a frequency of 2.10 ± 0.11 Hz in 32–34 hpf and 2.39 ± 0.34 Hz in 2 dpf larvae (Figures S3A’’ and S3D). By reporting the power of the most prominent frequency, we observed that the pulsations were strongest at the ducts and in the center of the diencephalic ventricle (Figures 4B and S3A’). At 4 dpf, when the ventricular space becomes tighter, the pulsations were apparent in 40% of the analyzed larvae and restricted to the ducts (Figure 4D; Video S4). Yet the frequency remained similar (2.14 ± 0.15 Hz) compared to younger larvae (Figures S3A’’ and S3D).

Because the CSF pulsation frequency is similar to the reported heartbeat frequency [47], we measured the CSF flow after stopping the heartbeat upon microdissection (Figures 4A’–4D’; Video S5) or by incubation with 2,3-butanedione monoxime (BDM) [10, 24] (Figures S3B and S3C). Upon ablation of the heartbeat, we detected neither pulsatile CSF flow nor any power at the heartbeat frequency, thereby confirming the contribution of the heartbeat to the pulsatile CSF flow. The absence of heartbeat did not affect the directions or velocities of the CSF flow fields (Figures 4A’, 4C’, S3B, and S3C), indicating that the heartbeat pulsations do not generate any net directional CSF flow. Moreover, we observed that the directional (Figures 4A and 4C) and pulsatile flow components (Figures 4B and 4D) are strongest in different parts of the ventricular system. Similarly, we did not observe any net directional CSF flow across the brain ventricles by tracking individual particles in larvae with and without the heartbeat (Figure S5).

As the directional flow component was most prominent near the location of motile ciliated cells, we next tested whether cilia loss or immotility abolishes the directional CSF flow component by using different cilia mutants. In the elipsa (traf3ip1) mutant [48], which lacked all primary and motile cilia (Figures 5B, 5E, 5G, and 5J), the directional flow was abolished at 2 (Figures 5B’ and 5E’; Video S5) and 4 dpf (Figures 5G’ and 5J’). Surprisingly, in oval (ift88), an ift88 mutant presenting ciliogenesis defects [49, 50], the motile ciliated cells retained a few cilia at 2 dpf, which resulted in the reminiscence of a weak flow at 2 dpf (Figures S6B–S6C’). At 4 dpf, the number of cilia decreased further and CSF flow was absent (Figures S6D–S6F’). These results demonstrate that cilia are the driving force of the directional flow in the ventricles. To verify the causal relationship between ciliary beating and CSF flow, we performed similar experiments in the mutant schmalhans (smh, ccdc103) [51] with a defect in cilia motility, but not in primary cilia. Our glutamylated tubulin staining confirmed the presence of cilia in the brain ventricle (Figures 5C, 5E, 5H, and 5J), and light-sheet recordings further validated the immotility of the cilia in smh mutants (n = 6; data not shown). As we did not detect any directional CSF flow in the smh mutants (Figures 5C’, 5E’, 5H’, and 5J’; Video S5), these results confirm that cilia motility is required for the directional CSF flow. In foxj1a mutant zebrafish, only motile cilia of the ventral diencephalic wall were absent (Figures 5D, 5E, 5I, and 5J). Loss of foxj1a did affect neither the number (Figure 5E) nor the beating properties (Figure 5F) of the motile cilia located along the dorsal wall of the diencephalic ventricle. Yet, this spatially confined ablation of motile ciliary beating reduced the overall velocity of the CSF flow, without affecting its directionality (Figures 5D’, 5E’, 5I’, and 5J’; Video S5). Importantly, the heartbeat-mediated pulsations remained similar in all analyzed mutants (Videos S4 and S5; Figure S3D). Altogether, our results demonstrate that the directional ventricular flow is generated upon beating of motile cilia located along the dorsal and ventral walls of the diencephalic ventricle.

Figure 5

The Directional CSF Flow near the Diencephalic Ventricular Wall Is Generated by Motile Ciliary Beating

In order to establish the causal link between ciliary beating and the directional CSF flow near the ventricular walls, we analyzed different mutants with ciliary defects.

(A–J) For each mutant line, cilia were labeled by glutamylated tubulin staining (magenta, DAPI in blue) at 2 (A: nctrl, 2 dpf = 21; B: n = 8; C: n = 8; D: n = 8) and 4 dpf (F: nctrl, 4 dpf = 20; G: n = 8 ; H: n = 9; I: n = 8) and counted at 2 (E) and 4 (J) dpf.

(A’–J’) The directional CSF flow in the diencephalic ventricle (DV) of each mutant line was measured by confocal imaging of fluorescent beads, analyzed by PIV, compared to controls at 2 (A’: nctrl, 2 dpf = 12, B’: n = 6, C’: n = 8, D’: n, 2 dpf = 8) and 4 dpf (F’: nctrl, 4 dpf = 14, G’: n = 7, H’: n = 11, I’: n, 4 dpf = 7) and quantified at 2 (E’’) and 4 dpf (J’).

(B and G) Elipsa mutants lack all primary and motile cilia, as demonstrated by the absence of glutamylated tubulin staining in a dorsal and ventral section of the diencephalon at 2 dpf (B, quantified in E) and 4 dpf (G, quantified in J). As a result, the directional CSF flow was absent in the diencephalic ventricle of elipsa at 2 (B’, quantified in E’) and 4 dpf (G’, quantified in J’).

(C and H) The cilia mutant smh shows no significant difference in the number of cilia at 2 dpf (C, quantified in E) and 4 dpf (H, quantified in J), but the directional CSF flow was abolished (C’, H’, E’’, and J’).

(D and I) In the cilia mutant foxj1a, the number of cilia along the ventral midline of the DV was reduced (indicated by ∗) at 2 dpf (D) and 4 dpf (I), but not along the dorsal DV (E and J). The difference in cilia number along the ventral DV is more apparent at 4 dpf (I) than at 2 dpf (D). The beating frequency of the cilia in the dorsal DV was not significantly different from controls at 2 dpf when comparing mean of individual larvae (indicated by a red cross); p value was calculated by the Wilcoxon rank-sum test (nctrl = 7, n = 14; E’). The directionality of the CSF flow in foxj1a was similar to controls, yet the velocity was reduced (D’, E’’, I’, and J’).

(E and J) Quantification of the cilia number along the dorsal DV at 2 (E) and 4 dpf (J).

(E’’ and J’) Quantification of the flow velocity along the dorsal DV at 2 (E’’) and 4 dpf (J’), as indicated in the inset for a control example. All controls from the various mutant clutches are pooled. Horizontal lines indicate the mean of the respective sample groups; vertical lines represent the mean ± SD. All p values were calculated with a Kruskal-Wallis (E and J) or one-way ANOVA (E’’ and J’) test followed by pairwise comparisons.

Scale bars are 50 μm (flow fields) or 25 μm (stainings). White areas in flow fields were not included in the PIV analysis, due to the presence of brain tissue and absence of particles. See also Figure S6 and Video S5.

Despite the directional CSF flow generated by the motile cilia and the pulsatile displacement by the heartbeat, we did not observe any net CSF flow across the brain ventricles. Such absence of CSF exchange across the individual ventricles was rather surprising and suggests a compartmentalization of the ventricular system. During our experiments, we noticed a strong influence of bodily movements on displacement of CSF across ventricular boundaries. When the animal contracted upon bodily movement, we observed a fast surge of particles from the rhombencephalic to the telencephalic ventricle (Video S7; Figures 6A and 6A’). This was followed by a slower return of particles in the opposite direction (Figure 6B’), as shown by an increase of a caudally oriented flow in the all ventricles (Figures 6C–6F’’’’), before the baseline flow was reinstated (Figure 6B’’). We observed a similar but weaker contribution of bodily movement at 4 dpf (Figure S7). Altogether, our results demonstrate that the ventricular flow is compartmentalized when the animal is immotile. Yet the distribution of CSF is strongly influenced by bodily movement that eliminates the stringent boundaries of the ventricular cavities.

Figure 6

Upon Bodily Movement, the Stringent Compartmentalization of the Larval Ventricular System Is Temporarily Eliminated

(A–A’’) Upon bodily movement, the amount of fluorescent particles transiently increases in the telencephalic ventricle (TV), as shown by confocal microscopy of a fluorescent particle-injected larval brain before (−0.1 s), shortly after (0.6 s), or 10 s after bodily movement at 2 dpf (A). The onset of movement corresponds to time 0. The change of fluorescence (dF/F) in the TV (encircled in green in A) upon a single body contraction reveals a fast increase of fluorescence upon bodily motion followed by a slow return to baseline levels (A’). The average (n = 7 movements) is indicated in black.

(B–B’’) PIV analysis of fluorescent particle recordings indicates that the CSF flow fields are strongly influenced by movement at 2 dpf. Prior to bodily movement (t = −20–0 s), the directional CSF flow along the ventricular walls is evident, referred to as the baseline flow (B). After the bodily movement (t = 0–4.0 s), a strong surge of CSF backward from the TV to the rhombencephalic ventricle (RV) is prominent (B’). About 4.0 s (t = 4.0–24 s) later, the baseline flow as seen in (B) is re-established (B’’). Representative example of n = 10 is shown.

(C–F””) In order to quantify the impact of bodily movement on the CSF flow kinetics, we measured the flow velocity and the flow direction in all consecutive PIV-analyzed frames in the TV (C–C’’’’), the ventral DV (D–D’’’’), the dorsal DV (E–E’’’’), and the RV (F–F’’’’). The velocity of CSF flow increases transiently in the TV, DV, and RV following bodily movement, indicated by a dashed line at time 0, and returns to baseline after 5 s. Black is average of all the movement-induced flow (C’–F’).

(C’’–F’’’’) Bodily movements affect the directionality of CSF flow across the ventricles, as indicated by polar histograms of multiple frames preceding movement (−20–0 s; C’’–F’’), shortly after movement (0–4 s; C’’’–F’’’) and following movement (4–24 s; C’’’’–F’’’’). In the TV, the polar histograms show little flow directionality before movement (C’’), a strong directionality of 30° pointing caudally toward the DV shortly after movement (C’’’), and a re-establishment of baseline flow directionality 4 s after the bodily movements (C’’’’). In the ventral DV and RV, the CSF flow is highly pulsatile before and after bodily movement, as indicated by the axial distribution of flow direction pointing toward 180° and 0° (D’’ and F’’). Just after the movement, the flow is primarily oriented caudally toward 0° (D’’’ and F’’’). In the dorsal DV, the flow direction is oriented rostrally with an angle of 180° (C’’ and C’’’’), but after bodily movements, the flow is directed caudally (C’’’).

Scale bars are 50 μm. Each individual movement is represented (n = 7 movements). Mean is shown in gray. See also Figure S7 and Video S7.

Video S7. Upon Bodily Movement, the Stringent Compartmentalization of the 2-dpf Larval Ventricular System Is Temporarily Disrupted, Related to Figure 6

The baseline cilia-mediated, unidirectional flow along the diencephalic ventricular walls of zebrafish larvae compartmentalizes the CSF to the separate ventricles, with little interventricular exchange of fluid. Upon bodily movement (at time 0), there is a major displacement of fluid toward the telencephalic ventricle, followed by a slower surge backward to the rhombencephalic ventricle, temporarily eliminating the stringent compartmentalization of the brain ventricular system. Prior to and after movement, the common unidirectional flow along the ventricular walls is evident. Acquisition frequency: 11.6 Hz.

Motile-Cilia-Mediated Flow Is Required for Ventricular Development

Previous work has reported that patients and animal models displaying cilia dysfunction ultimately develop an enlarged ventricular system, which is a condition known as hydrocephalus [15, 17, 52, 53, 54]. To investigate whether the lack of motile-cilia-mediated flow is sufficient to induce ventricular defects in zebrafish larvae, we measured the ventricular sizes in animals devoid of either both primary and motile cilia or motile cilia only. Analysis of 2 dpf larvae revealed no significant differences between all mutants and controls (Figures S6H–S6K). However, at 4 dpf, our results indicated that only larvae lacking both primary and motile cilia (oval, elipsa) were more susceptible to develop enlarged ventricles (Figures 7A–7D and S6G). Interestingly, animals with defects in motile cilia only (smh, foxj1a) did not develop enlarged ventricles (Figures 7A–7D), suggesting that the lack of primary cilia affects brain development independent of motile-cilia-mediated flow. Instead, we observed that zebrafish larvae with impaired motile-cilia-mediated flow demonstrated a significantly higher probability of ventricular duct occlusion. This was measured as a reduced diffusion of dextran and fluorescent particles to the telencephalon in motile cilia mutants (Figures 7E and 7F). Even though foxj1a mutants retained some weak motile-cilia-mediated flow, this flow was not sufficient to keep the connecting ducts open to allow diffusion of molecules (Figures 7E and 7F). The diffusion to the telencephalon in elipsa mutants was also significantly reduced compared to the controls, albeit the effect was less strong than observed in smh and foxj1a mutants (Figures 7E and 7F). This may result from the enlargement of the ventricular system in elipsa (Figures 7A–7D), which might facilitate the passage of dye and particles.

Figure 7

The Brain Ventricular System Is Differentially Affected upon Loss of Primary versus Motile Cilia

(A–D) Increased incidence of hydrocephalus is observed in the cilia mutant elipsa, but not in smh and foxj1a. Various ventricular hallmarks were measured in larvae injected with 70-kDa RITC-dextran at 4 dpf. These included the height of the telencephalic (TV; A) and the diencephalic (DV; B) ventricles and the width of the TV-DV duct (C) and the DV-RV duct (D). The controls for all mutants are pooled into one group and compared to mutants lacking all cilia, elipsa (green), or motile cilia mutants, smh and foxj1a (blue). Outliers (encircled in red) were defined as measurements outside the 1.5 interquartile ranges of the pooled control group (light blue box). The p value above the respective mutant line reports its significant difference compared to the pooled control group. Elipsa mutant larvae demonstrated increased numbers of outliers and differed significantly from controls for the measurements of the TV (A), DV (B), and DV-RV duct (D), but not for the TV-DV duct (C). smh and foxj1a mutant larvae show no significant difference compared to controls (A–D; nctrl = 29, n = 16, n = 18, and n = 7).

(E and F) Motile cilia mutants (smh and foxj1) were more susceptible to obstructed ventricular ducts than the primary cilia mutant elipsa at 4 dpf. This was measured as the relative diffusion of large 160-nm-diameter particles (E) or of 70-kDa RITC-dextran (F) from the diencephalon to the telencephalon upon injection into the rhombencephalic ventricle. All controls were pooled for comparison to the cilia mutants. Horizontal lines indicate the mean of the respective sample groups. Vertical lines represent the mean ± SD. The blue box identifies the lower and upper threshold of the pooled control group used to determine outliers (encircled in red). All p values were calculated with a Kruskal-Wallis test followed by pairwise comparison (nctrl = 28, n = 17, n = 14, and n = 10).

Scale bars are 50 μm. See also Figure S6.

Altogether, we demonstrated that the mutations affecting both primary and motile cilia or motile cilia only alter the development and compartmentalization of the brain ventricles in different ways. Our results indicate that primary cilia mutants develop ventricular enlargement and hydrocephalus, and larvae with aberrant motile cilia do not. Moreover, we showed that motile cilia are crucial for generating a directional flow, which prevents the obstruction of ventricular ducts and thereby maintain the communication of CSF across ventricles.

Discussion

Our results revealed the presence of motile ciliated cells in the developing brain ventricles of zebrafish larvae. Based on their function, their direct contact with the ventricles, and their molecular profile, these cells appear homologous to ependymal cells in mammals. However, there are a few notable differences. First, in mammals, ependymal cells differentiate primarily postnatally [12, 55], although, in our study, we identified motile ciliated cells already at the embryonic stage. However, a recent report indicated the presence of motile cilia at embryonic stage in distinct locations across the mouse brain [54]. In accordance with these findings, Foxj1 is expressed in a few cells in the developing mouse brain [31]. Second, most ependymal cells in mammals are multiciliated, bearing 30–300 cilia per cell [12, 16], which beat in an asymmetric stroke, similar to the zebrafish nose pit [9]. In contrast, motile ciliated cells in the larval zebrafish brain ventricles appear to be monociliated and beat in a rotational movement, which is typical for monociliated cells, including sperm [41] and cells in the left-right organizer [42]. Few studies reported multiciliated ependymal cells in the adult zebrafish, in the ventricle wall located between the corpus cerebelli and the medulla oblongata [17] and in the diencephalic and telencephalic ventricle [25, 26], where foxj1a is expressed [56]. It will be exciting to investigate whether these multiciliated cells and the monociliated cells we identified at larval stage arise from the same progenitor pool and co-exist in the adult zebrafish brain.

We have identified two different subpopulations of motile ciliated cells based on the expression pattern of the two Foxj1 orthologs and the analysis of ciliary mutants. Our results obtained with a 0.6kbfoxj1a:gfp and 5.2kbfoxj1a:gfp [17] transgenic lines demonstrated that these promoter regions are not sufficient to drive the specific expression pattern of foxj1a in the larval brain and suggest that other regulatory elements may be involved in its transcriptional regulation [57]. Nevertheless, our data revealed that both foxj1a- and foxj1b-positive cells are needed to generate a robust CSF flow in the ventricles. In mammals, previous studies have also shown that ependymal cells can be of different types, the multiciliated type I and the bi-ciliated type II [58]. Interestingly, these cells exist in different regions across the mammalian ventricular system [59], similar to the two populations we have identified in zebrafish larvae. It will be exciting to ascertain whether foxj1a- and foxj1b-expressing cells populate distinct brain regions in adult zebrafish and whether they have different functions, similar to the type I and type II mammalian ependymal cells.

Because of the small size and transparency of the larval zebrafish brain, we were for the first time able to measure and quantify all components of the ventricular flow in a living and intact vertebrate brain. Our results revealed that the CSF flow is generated and modulated by multiple parameters, including ciliary beating, heartbeat, and body contractions. Our findings are in line with previous modeling work of human brain ventricles, which suggested that motile ciliary beating contributes specifically to the near-wall flow, while the heartbeat, blood vessel pulsations, and CSF secretion generate a pulsatile bulk flow [60].

CSF is primarily produced and secreted by the choroid plexus in all vertebrates [22, 37]. In mammals, all four ventricles have one choroid plexus [22]. In contrast, zebrafish exhibit only two choroid plexuses located along the dorsal walls of the telencephalic and rhombencephalic ventricles [37, 61]. It is surprising that we do not observe a directional flow from the choroid plexus-bearing ventricles toward the diencephalic ventricle. It is possible that the fluid flow generated by the choroid plexuses is very small at the early developmental stage. Indeed, previous studies suggested that the choroid plexuses become functional at 4 dpf [62], which corresponds to the second stage investigated in this study. Interestingly, our results indicate that, at least in the telencephalon, choroid plexus cells, which are foxj1b positive, extend motile, glutamylated tubulin-positive cilia, in agreement with a previous report [61]. Interestingly, the foxj1b-positive motile ciliated cells located in the dorsal diencephalic ventricle may also express claudin5a, a marker of the choroid plexus [61]. Yet, these cells are not in close contact with blood vessels, which is a commonly recognized feature of the choroid plexuses [22, 37], suggesting that these cells may have an ependymal-like function rather than being a part of the choroid plexus. The mammalian choroid plexuses also express Foxj1 [31] and present some motile cilia during late embryonic stage and early postnatal period in mice [63]. Although the contribution of motile cilia to the function of the choroid plexus remains poorly understood, previous work revealed that primary cilia are required for the development of the choroid plexus by supporting ion transport and CSF generation [53, 64]. It will be interesting to further study the role of motile cilia in the functioning of the choroid plexus and to investigate the significance of choroid-plexus-mediated flow in the embryonic brain.

Our results show that motile cilia compartmentalize the CSF flow by establishing ventricular boundaries, with very little CSF exchange across the ventricles. Only bodily contractions during animal movement overcome the compartmentalization and allow exchange of molecules past these ventricular boundaries. These results are very exciting, as they suggest an important role of motile-cilia-mediated flow in regionalizing molecules. Recent studies in mammals identified that the proteome of the CSF varies across brain ventricles as a result of transcriptional differences between the telencephalic and the rhombencephalic choroid plexuses [21, 22] and support a targeted neuronal differentiation program [21]. The motile-cilia-mediated CSF flow, which we characterized in this study, would certainly facilitate such compartmentalized differentiation programs by limiting the CSF exchange across ventricles. Concurrently, motile-cilia-mediated flow would ensure a proper mixing of CSF within individual ventricles.

Even though we did not observe any motile-cilia-mediated CSF flow in the ventricular ducts, our results demonstrate that ciliary beating is crucial in maintaining these ducts open. It is possible that a lack of flow causes aggregation of molecules and other components in the CSF, which eventually could clog the ducts. Yet it is surprising that zebrafish larvae with obstructed ducts do not develop a hydrocephalus. It may require time to build up water pressure upon obstruction of the ducts. Alternatively, the CSF may be absorbed into the brain parenchyma over the ventricular walls [65, 66]. In this case, the CSF would still be able to diffuse through the neuronal tissue and thereby prevent the accumulation of ventricular water pressure causing hydrocephalus. Our results indicate that only larvae with defective ciliogenesis, which lack both primary and motile cilia, have a higher probability of developing an enlarged ventricular system. The difference between motile cilia and primary cilia mutants suggests that hydrocephalus results from a combination of several factors beyond motile-cilia-mediated flow, including excess CSF production, reduced CSF absorption, or impaired brain development. Indeed, even though ependymal dysfunction is proposed to be the underlying cause of hydrocephalus in several mouse mutants [12, 52, 54, 67], neurogenesis [68] or choroid plexus defects [53] can also result in hydrocephalus in mutant mice. Interestingly, ciliary loss appears to more prominently affect the secondary wave of neurogenesis and gliogenesis when major brain areas are being established [69]. It is also important to note that, although most mouse mutants develop hydrocephalus, the prevalence of hydrocephalus in human primary ciliary dyskinesia patients is rare [52]. This suggests that the size of the ventricular system or the genetic disparity between species may account for the sensitivity of each species to hydrocephalus and advocates for further work on the function and regulation of CSF flow in various animal models.

Finally, our results demonstrate that bodily movements induce a strong fluid displacement in the brain ventricular system that overrules the compartmentalization. Interestingly, neural progenitors extend their primary cilium into the ventricular cavity [70, 71]. Because primary cilia can sense mechanical stress [2], they may detect mechanical forces induced by CSF flow and regulate the proliferation of progenitors. This could also be cilia independent, as mechanical forces promote neuronal proliferation in the subventricular zone of the adult mouse brain through the somatic epithelial sodium channel ENaC [18]. It would therefore be interesting to elucidate the effect of bodily-movement-induced CSF flow on neurogenesis.

Altogether, our results demonstrated that the CSF flow in the zebrafish brain is regulated by multiple factors, including motile ciliary beating, heartbeat, and bodily movements. We propose that these factors work in parallel on different components of the CSF flow and thereby establish a balance between compartmentalization and dispersion of CSF across the brain ventricles. Further studies are needed for investigating the role of spatially organized CSF flow in the development of the brain and its ventricular system.

STAR★Methods

Key Resources Table

Contact for Reagent and Resource Sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Nathalie Jurisch-Yaksi (nathalie.jurisch-yaksi@ntnu.no).

Experimental Model and Subject Details

The animal facilities and maintenance of the zebrafish, Danio rerio, were approved by the Norwegian Food Safety Authority. Fishes were kept in 3.5 L tanks in a Techniplast Zebtech Multilinking system at constant conditions: 28°C, pH 7 and 700 μSiemens, at a 14:10 hr light/dark cycle to simulate optimal natural breeding conditions. Fishes received a normal diet of dry food (ZEBRAFEED; SPAROS I&D Nutrition in Aquaculture) two times/day and Artemia nauplii once a day (Grade0, platinum Label, Argent Laboratories, Redmond, USA). Larvae were maintained in egg water (1.2 g marine salt and 0.1% methylene blue in 20 L RO water) from fertilization to 3 dpf and subsequently in AFW (1.2 g marine salt in 20L RO water).

For experiments, the following fish lines were used: smh [51] (received from Drummond lab, MGH), elipsa [48] (received from J Malicki, University of Sheffield), oval [49] (received from I Drummond, MGH, Harvard), β-actin:arl13b-gfp [40] (received from C Wyart, ICM, Paris), T2BGSZ10 Tg(Foxj1b:GFP) [33] (received from S Roy, A∗STAR, Singapore) and Tg(5.2kbFoxj1a:GFP) [17] (received from D Grimes, Princeton, USA). Tg(0.6kbFoxj1a:GFP)nw6Tg transgenic animals were generated in our laboratory upon co-injection of tol2 transposase mRNA and 0.6kbFoxj1a:GFP plasmid obtained from X Lin [73]. The Tg(0.6kbFoxj1a:GFP)nw6Tg expression pattern presented here was identified in four independent founders. Experiments were performed on embryos of AB and nacre (mitfa homozygous mutant [74]) background.

Smh, oval, elipsa and foxj1a heterozygous adults were genotyped by KASP assays (LGC genomics) or by PCR (foxj1a). Incrosses of genotyped heterozygous parents were used in the experiments. Homozygous mutant animals (25% of the clutch) were identified based on their bent body axis. Straight tail larvae coming from the same clutch were used as controls (corresponding to heterozygous and wild-type larvae).

All procedures were performed on zebrafish larvae up to 4 dpf in accordance with the directive 2010/63/EU of the European Parliament and the Council of the European Union and the Norwegian Food Safety Authorities. The generation of zebrafish mutant lines described hereafter were approved by the Norwegian Food Safety Authorities.

Method Details

CripsR/cas9-mediated mutagenesis

Foxj1a mutants were generated upon CrispR/cas9 mediated mutagenesis according to the following protocol [72]. Briefly, the gRNA sequence was identified using ZiFiT Targeter website (http://zifit.partners.org/ZiFiT/). Annealed oligo was cloned into the pDR274 plasmid (Addgene Plasmid #42250), which was then used as template to in vitro transcribe sgRNA using MAXIscript T7 kit (Invitrogen). Cas9 mRNA was in vitro transcribed from pMLM3613 plasmid (Addgene Plasmid #42251) using mMessage mMachine T7 kit (Invitrogen) and polyA tailed with a polyA tailing reaction kit (Invitrogen). 2 nL of a mixture of 12.5 ng/μL of sgRNA and 300 ng/μL of Cas9 mRNA were injected in one-cell stage embryos. CrispR/cas9-induced mutations and germline transmission of the mutations were initially detected by T7EI assay, and later by melting curve analysis of qPCR amplification using SYBRgreen. Three separate alleles were recovered from one founder. All mutations resulted in a frameshift from amino acid 66 and an early stop codon, which was before the forkhead domain. All homozygous mutants for the allele exhibited a bent body axis. In this study, the foxj1a allele, which carried a 40 bp insertion, was analyzed.

Immunostaining and imaging

Euthanized larvae were fixed in a solution containing 4% paraformaldehyde solution (PFA), 1% DMSO and 0.3% Triton X-100 in PBS (0.3% PBSTx) for at least 2 hr at room temperature. Stainings were performed on cut heads to improve the penetration of the antibodies. Heads were washed with 0.3% PBSTx (3x5 min) and permeabilized with acetone (100% acetone, 10 min incubation at −20°C). Subsequently, samples were washed with 0.3% PBSTx (3x10 min) and blocked in 0.1% BSA/0.3% PBSTx for 2 hr. Larvae were incubated with glutamylated tubulin (GT335, 1:400, Adipogen) overnight at 4°C. On the second day samples were washed (0.3% PBSTx, 3x1 hr) and subsequently incubated with the secondary antibody (Alexa-labeled GAM555 plus, Thermo Scientific, 1:1,000) overnight at 4°C. The third day samples were incubated with 0.1% DAPI in 0.3% PBSTx, Life Technology, 2 hr), washed (0.3% PBSTx, 3x1 hr) and transferred to a series of increasing glycerol concentrations (25%, 50% and 75%). Stained larvae were stored in 75% glycerol at 4°C and imaged using a Zeiss Examiner Z1 confocal microscope with a 20x plan NA 0.8 objective. An antibody against GFP (Millipore, ab16901, and Alexa488-anti-chicken, Thermo Scientific) was used to enhance the GFP signal if needed. Acquired images were processed with Fiji/ImageJ [75] or Zen (Zeiss).

Numbers of cilia were manually counted using the “Cell Counter” plugin for Fiji/ImageJ (Kurt De Vos, Univ Sheffield, Academic Neurology, https://imagej.nih.gov/ij/plugins/cell-counter.html).

Larvae anesthetized in 0.01% MS-222 (Sigma) were imaged upon immobilization in a Fluorodish (World Precision Instruments) in 2% low melting point agarose (Invitrogen) prepared in AFW, using a Zeiss Examiner Z1 confocal microscope with a 20x water immersion NA 1.0 objective.

Ciliary beating measurements by light-sheet microscopy

Acquisition of light-sheet recordings

Ciliary beating was measured in 2-4 days old β-actin:arl13b-gfp larvae. Only 4 dpf larvae were paralyzed by intramuscular injection of α-bungarotoxin (Invitrogen BI601, 1 mg/mL) [77]. Larvae were embedded in 2% low-melting point agarose prepared in AFW in a custom-built mounting chamber. The chamber was filled with AFW to completely submerge the larvae. The larvae were positioned so that they were directly facing the sheet of laser light from the light-sheet microscope and the agarose in front of each larva’s head was removed. Larvae were positioned either horizontally or on their side. Microscopy recordings were obtained by a custom-made light-sheet microscope as previously described [9]. The light-sheet microscope was based on the design described by [76] with a 20x water immersion objective (Olympus, NA 1.00, plan) and a laser of 488 nm wavelength (Cobolt). Images were acquired by the Zebrascope software in LabView [76] and analyzed in Fiji/ImageJ and MATLAB. All light-sheet microscopy recordings were aligned using a custom-written algorithm adapted from [78], which corrects occasional x-y drift and is described in [9].

Analysis of light sheet recording by Fast Fourier Transform

Beating cilia caused a periodic change of pixel intensities over time in the acquired images. We analyzed the frequency of oscillations for each pixel of the recording in the power spectral density estimate obtained by the Fast Fourier Transform algorithm of MATLAB. For each pixel of the recording, the primary frequency was determined as the frequency of the highest peak in the power spectrum between 15 Hz and half the frequency of acquisition. Thereby, a frequency map was generated. Subsequently, a noise estimation and correction method were applied to the frequency map. A custom standard deviation (SD) thresholding-algorithm was used to segment the map into noise and signal regions. In brief, a 3x3 kernel was moved across the entire frequency map. Any pixel belonging to a 3x3 kernel whose SD was below 3 (experimentally determined as an appropriate threshold, data not shown) was considered as signal, whereas pixels belonging only to kernels with an SD above 3 were considered as noise. The average and SD of all power values for frequencies between 15 Hz and half the frequency of acquisition in all power spectra corresponding to noise regions were determined. Next, from all power spectra in signal regions the determined average + SD was subtracted, while restricting the minimum value to 0. Finally, the processed power spectrum was again analyzed for peaks and the position of the highest peak was taken as the final primary frequency. If no peaks were left in the power spectrum, the respective pixel position was excluded from the signal region. Altogether, with this SD-thresholding algorithm, we were able to restrict the analysis only to the regions where cilia were beating. Furthermore, the correction method ensured that only power spectra with a sufficient quality to determine the primary frequency were included into analysis results. To perform statistics on the data, all primary frequency patches, which corresponds to ciliary beating frequencies in the raw recordings, were manually detected from the Fourier-transformation-based analysis of light-sheet microscopy recordings. For Figure 2B, values that were above the maximum beating frequency detected across all recordings were considered as noise and removed.

Additional quantifications

Additional quantifications of a subset of cilia were performed using Fiji/ImageJ and included the width and rotational angle (for a precise depiction see Figure 2D). All quantifications were performed in ImageJ/Fiji [75] using the Measure function. The cilia were chosen based on the following criteria: the entire cilium was visible throughout the recording; overlap with other cilia was restricted so that the entire cilium could be easily discerned; the cilium was motile.

Quantification using SpermQ

Processing and analysis was performed in ImageJ/Fiji, using the ImageJ-plugin SpermQ [43], and with the java tool SpermQEvaluator [43]. Time-lapse images were cropped to images of individual cilia. The background was removed using the SubtractBackground function in ImageJ (radius: 5 px). Next, the images were subjected to automated analysis by SpermQ. We used the following settings: thresholding method: Li, Gaussian blur σ: 1.0, repeat Gaussian blur after binarization: no, blur only inside manual selection: no, upscale trace: 3-fold, head center-of-mass added: no; unify start points: yes, filter out points: no, maximum vector length: 20 points, normal vector radius: 2 μm, exclude head points: no, smooth normal: yes, fit-width smoothing: median, accepted xy distance for fit-width-smoothing: 1.5 μm, # (+/−)-consecutive points for xy- and fit-width-smoothing: 3, distance of point to first point to form the reference vector: 3 μm, curvature reference-point-distance: 3 μm, grouped time-steps for Fourier transformation: 1024, initial arc length neglected for Fourier transform: 0 μm, head rotation matrix radius: 10 points. The results were further analyzed using the java tool SpermQEvaluator. The phase of the oscillations of the relative y and z positions were calculated by Fourier Transform in MATLAB.

Brain ventricle injections and imaging

Prior to injections, all larvae were anesthetized in 0.01% MS-222 in AFW for 10-15 min. Injections were done on larvae embedded in 2% low-melting point agarose in AFW and 0.01% MS-222 in AFW a FluoroDish. In order to obtain anatomical z stacks, larvae were mounted horizontally. In contrast, for imaging the ventricular flow, the fish were mounted on their right-hand side, resulting in a sagittal view of the ventricles when imaging.

The injection mixtures contained either 70 kDa rhodamine B isothiocyanate-dextran (RITC-dextran; Sigma-Aldrich, R9379) dissolved in artificial cerebrospinal fluid (aCSF) at a final concentration of 10 mg/mL, or 0.1% w/v fluorescent beads (SPHERO Fluorescent Yellow Particles 1% w/v, Φ = 0.16 μm) diluted in 7.5 mg/mL RITC-dextran in aCSF. aCSF composition was as follows: 124 mM NaCl, 22 mM D-(+)-Glucose, 2.0 mM KCl, 1.6 mM MgSO4 · 7 H2O, 1.3 mM KH2PO4, 24 mM NaHCO3, 2.0 mM CaCl2 · 2 H2O.

The needles used for the injections were pulled with a Sutter Instrument Co. Model P-2000, from thin-walled glass capillaries (1.00 mm; VWR), using the following settings: heat = 785, filament = 4, velocity = 40, delay = 220, pull = 70. The needle tip was cut open with forceps. A pressure injector (Eppendorf Femtojet 4i) was used to inject 1 nL of solution in the rostral rhombencephalic ventricle, as previously described [79]. The pressure and time used for the injection were calibrated for each needle using a 0.01 mm calibration slide for microscopy. Usually, the pressure ranged between 100-150 hPa and the time span of the pressure pulse lasted for 0.30 – 0.70 s.

After injection, the larvae were immediately transferred to the confocal microscope (Zeiss Examiner Z1), and imaged with a 20x water-immersion objective (Zeiss, NA 1.0, Plan-Apochromat) at room temperature. In order to avoid bodily movement during the recordings, larvae were maintained in AFW containing 0.01% MS-222. To measure the impact of bodily movement on CSF flow, the larvae were not anaesthetized prior to ventricular injection; the head was mounted in low-melting point agarose and their tail was freed to allow movement. The health of all larvae was monitored throughout the experiments by visualizing the heartbeat and blood flow with the transmission mode of the confocal microscope. At the end of the experiments, animals were euthanized.

For the bead-injected larvae, both time lapses and z stacks were acquired. The time lapses were acquired either for the telencephalic and diencephalic ventricles at once, or for only the diencephalon, at a frequency of 8.09-12.4 Hz and 10.1-13.5 Hz, respectively. Images were 256 × 200 pixels (diencephalic and telencephalic ventricle) or 256 × 170 pixels (diencephalic ventricle only). Additional zooming was used; 1.1x for the combined telencephalic and diencephalic time lapses, and 1.8x for the diencephalic ventricle. For the recordings of the ducts, images of 256 x 170 pixels were acquired with a zoom of 5x and a frequency of aquisition of 20.22 Hz. In order to stop the heartbeat, larvae were incised with a scalpel or incubated with 2,3-butanedione monoxime (BDM, Sigma) for 30 min. To acquire z stacks of the ventricular system, the laser power and gain were adjusted for depth, and the fish were imaged from about 5 min after injection and onward. A z stack of all analyzed larvae was acquired at the end of the experiment. The dimensions of the ventricular system were measured manually on z stacks obtained from larvae co-injected with RITC-dextran and fluorescent beads using the line tool in Fiji/ImageJ on the stacks acquired at the end of the experiment.

The diffusion of dextran and particles to the forebrain were quantified on the first recording, which included the telencephalic and diencephalic ventricles, following ventricular injection. First, we applied a threshold to the average projection of the recording based on the 70 kDa RITC-dextran signal. The image was then normalized to the maximum pixel intensity. Two regions of interest corresponding to the telencephalic and diencephalic ventricles were manually selected. The % diffusion corresponds to the average normalized fluorescence in the telencephalon divided by the average normalized fluorescence in the diencephalon. All analyses were performed on MATLAB

Analysis of CSF flow

The movement of particles in the ventricles was analyzed using two different methods; particle image velocimetry (PIV) to identify directional flow fields and a custom-made MATLAB script to measure the pulsatile component of the flow. Additionally, for single particle tracking, the manual tracking plugin of Fiji/ImageJ was used and data were plotted using MATLAB.

Directional flow investigated with PIVlab

PIV was used to identify the directional flow of the beads, by help of PIVlab v. 1.41 [46], a generalized user interface-based tool in MATLAB. All recordings (except the ones used for Figure 7) included 1,200 image frames, which were saved as an image sequence in BMP format using Fiji/ImageJ and subsequently imported into the PIVlab user interface. The non-interesting regions (e.g., brain tissue and skin, indicated in white in the respective figures), were masked, leaving only the ventricles to be analyzed. The sizes of the interrogation areas (denoted “passes”) were adjusted as following: Pass 1 = 80, Pass 2 = 40 and Pass 3 = 20. The vector validation was done manually by selecting the densest cloud of points, each point corresponding to a vector, plotted in a scatterplot with the u velocity on the x axis and the v velocity along the y axis. After validation and calibration, the mean vectors of all frames were calculated and plotted as the velocity magnitude (μm/s) by means of a jet heatmap. The average velocity in the dorsal diencephalic ventricle was obtained upon manual selection of a region of interest in PIVlab.

For measuring the changes in flow velocity and direction upon bodily measurement, PIV was performed on 800 images sequences (300 preceding and 500 following bodily movement at 2 dpf) or 400 images sequences at 4 dpf with the following three passes (Pass 1 = 40, Pass 2 = 20, Pass 3 = 10). In order to obtain the mean flow fields before and after movements, the PIV measurements were averaged over consecutives frames corresponding to the three time windows indicated in the respective figures. For 2 dpf larvae, we used 1) 4 – 20 s preceding movement, 2) 0 – 4 s following movement and 3) the 2 – 24 s following movement. The flow velocities and flow directions during the entire time course were extracted for a given area upon manual selection of regions of interest in PIVlab. The flow directions for the time windows preceding or following the movement were plotted using the polarhistogram function of MATLAB with bin sizes of 10° and normalized by a probability distribution. The mean flow direction across individual larvae at given time periods was plotted using the polarhistogram function of MATLAB with bin sizes of 20° and normalized by a probability distribution. For the analysis at 4 dpf, the bin size corresponded to 30°.

In order to measure the effect of movement on the amount of particles in the telencephalic ventricle, the fluorescence intensity was extracted over the course of the recording for a manually-selected region of interest comprising the telencephalic ventricle. To calculate the change of fluorescence induced by bodily movement, dF/F was calculated as follows: (fluorescence (time = t) – mean(fluorescence (time = −20 s – 0 s))/ mean(fluorescence (time = −20 s – 0 s). Bodily movements induced either a single push of particles in the telencephalic ventricle or multiple back-and-forth movements of particles. Only movements with a single contraction were included in the quantification of flow directionality and velocities.

Pulsatile flow calculated with a custom-made MATLAB script

To analyze the pulsatile component of the flow, a custom-made MATLAB script was used. This script is based on the oscillatory change of light intensity in any given pixel over time, due to the pulsatile movement of beads. To this end, a Fast Fourier Transform algorithm (fft function in MATLAB) was used to identify the various frequency components of each pixel. The frequency of the highest power (between 1 Hz and half of the frequency of acquisition), henceforth referred to as the peak-frequency (PeakFreq), was extracted for each pixel. Then, the most common PeakFreq was manually selected and attributed to the given sample. The relative power at this given frequency, which was reported for each pixel, corresponded to: (Power(PeakFreq ± 0.1 Hz)-mean(Power(all frequency)))/ mean(Power(all frequency)). For samples where the heartbeat was stopped, the PeakFreq values corresponded to the value obtained in the associated recording when the heart was beating. For data visualization, a Gaussian blur was applied to the results with a kernel of 6x6 pixel and sigma equal to 1.

In Situ Hybridization

In situ hybridization probes were synthetized from plasmids. 1 μg of digested and QiaQuick PCR purified (QIAGEN) plasmids [30, 33] was incubated with 2 μL DIG labeling mix (10 mM ATP, 10 mM GTP, 10 mM CTP, 6.5 mM UTP and 3.5 mM DIG-UTP (Roche) in nuclease-free water), 1 μL Riboblock RNase inhibitor and 2 μL RNA polymerase (20U, SP6 or T7) for 2.5 hr at 37°C in a total volume of 20 μL. Following synthesis, the RNA probe was precipitated upon addition of 1 μL EDTA 0.5 M (Ambion), 1.25 μL 8M LiCl (Sigma) and 75 μL 96% ethanol and incubation at −80°C for at least 30 min. Following centrifugation at 4°C for 30 min, the RNA pellet was washed in 70% ethanol, and resuspended in H2O. The RNA probe integrity was verified by gel electrophoresis and its concentration measured with NanoDrop (Thermo Scientific). Unless otherwise specified, all reagents were purchased at Thermo Scientific.

Larvae were euthanized at the desired age by submersion in MS-222 diluted in AFW at 4°C and transferred to 4% PFA for 2 hr at room temperature for fixation. Following fixation, larvae were dehydrated in methanol (VWR) and kept at −20°C. Larvae were gradually rehydrated through 75% – 50% – 25% methanol steps for 5 min each. Following 3x5 min washes in 0.1% Tween-20 in PBS (PBST), the larvae were digested with proteinase K (10 μg/mL) for 30 min at room temperature, fixed in PFA for 20 min and washed in PBST. Larvae were prehybridized at 65°C for 2-5 hr in hybridization buffer (50% formamide (Ambion), 5X SSC (GIBCO), 50 μg/mL heparin, 500 μg/mL tRNA, 0.1% Tween-20), followed by an overnight hybridization with 1.5 μg/mL probe dissolved in hybridization buffer. The larvae were subsequently washed twice for 20 min in 50% formamide/2X SSCT, one time for 15 min in 2X SSCT, and twice for 20 min in 0.2X SSCT at 65°C. Following one PBST wash, larvae were blocked in blocking solution (5% sheep serum, 2 mg/mL BSA) for 1 hr at room temperature and incubated overnight at 4°C in AP-anti-DIG antibody (Roche) diluted 1:5,000 in blocking buffer. The next day, the samples were washed 6x15 min in PBST, followed by 3x5min in TMNT (100 mM Tris, pH 9, 50 mM MgCl2, 100mM NaCl, 0.1% Tween-20). The in situ hybridization was then stained for 4-12 hr in BCIP/NBT solution (1:400 dilution of BCIP (Roche) and NBT (Roche) in TMNT). The staining was terminated by washes with 1 mM EDTA in PBS and fixation with 4% PFA for 1 hr. Samples were then washed and transferred to 75% glycerol. All reagents were purchased from Sigma unless specified otherwise.

Quantification and Statistical Analysis

Statistical analysis was performed in MATLAB. Wilcoxon rank-sum test was used for non-paired analysis and Wilcoxon signed-rank test for paired analysis. p < 0.05 was considered as statistically significant. Kruskal-Wallis test or one-way ANOVA (kruskalwallis or anova1 function in MATLAB) were used to perform multiple comparisons. The Jarque-Bera test (jbtest function in MATLAB) was used to test the normality of the distribution before performing ANOVA. When significance was obtained upon Kruskal-Wallis or ANOVA test, we conducted post hoc tests to identify which data sample came from a different distribution using the multcompare function in MATLAB. The p value output of the multiple comparison was then reported in its corresponding dataset. The number of observations and statistical tests are indicated in the figure legends.

Data and Software Availability

All analyses were performed with Fiji/ImageJ [75], MATLAB or SpermQ [43] as indicated in the results sections. All custom MATLAB scripts are available upon request.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal glutamylated tubulin (GT335)	Adipogen	Cat#AG-20B-0020-C100; RRID: AB_2490210
Chicken GFP	Millipore	Cat#Ab16901; RRID: AB_11212200
Alexa 488 anti-chicken	Thermofisher	Cat#A11039; RRID: AB_2534096
Alexa-GAM555 plus anti-mouse	Thermofisher	Cat#A32727; RRID: AB_2633276
AP-anti-DIG	Roche	Cat#11093274910; RRID: AB_514497
Plasmid DNA
Cas9 mRNA pMLM3613 plasmid	[72]	RRID:Addgene_42251
gRNA plasmid pDR274	[72]	RRID:Addgene_42250
Foxj1a gRNA plasmid	This study	N/A
0.6kbFoxj1a:GFP plasmid	Lin lab	[73]
Chemicals, Peptides, and Recombinant Proteins
x2 power up SYBR master mix	Thermofisher	Cat#15366158
Alpha-bungarotoxin	Invitrogen	Cat#BI601
70 kDa rhodamine B isothiocyanate-dextran	Sigma-Aldrich	Cat#R9379
SPHERO Fluorescent Yellow Particles 1% w/v, Φ = 0.16 μm	Spherotech	Cat#FP-0252-2
DIG DNA labeling mix	Roche	Cat#11175033910
DIG-11-UTP	Roche	Cat#11209256910
NTPs	Thermofisher	Cat#11863933
BCIP/NBT	Roche	Cat#11681451001
SP6 and T7 RNA polymerase	Thermofisher	Cat#10610301 and 10344450
Formamide	Ambion, Invitrogen	Cat#AM9342
2,3-butanedione monoxime	Sigma-Aldrich	Cat#B0753
Restriction enzymes	Thermo	N/A
Critical Commercial Assays
MAXIscript T7 kit	Invitrogen	Cat#AM1312
KASP assay	LGC genomics	N/A
mMessage mMachine T7 kit	Invitrogen	Cat#AM1344
polyA tailing reaction kit	Invitrogen	Cat#AM1350
Zebrafish lines
Smh ccdc103tn222a	[51]	RRID:ZIRC_ZL906
Elipsa traf3ip1tp49d	[48]	RRID:ZIRC_ZL1731
Oval ift88tz288	[49]	RRID:ZIRC_ZL1219
Foxj1anw3	This study	N/A
Nacre mitfab692	[74]	RRID:ZFIN_ZDB-ALT-010919-2
β-actin:arl13b-gfp, Tg(actb2:Mmu.Arl13b-GFP)	[40]	ZFIN ID: ZDB-TGCONSTRCT-100721-1
Foxj1b:GFP T2BGSZ10	[33]	T2BGSZ10
Tg(Foxj1a:GFP)nw6Tg	This study	N/A
Tg(5.2kbFoxj1a:GFP)	[17]	ZFIN ID: ZDB-TGCONSTRCT-160719-1
Oligonucleotides
All oligonucleotides	IDT	N/A
Foxj1a genotyping forward CTTGGAGGAGGAAGTGGTGACT	IDT	N/A
Foxj1a genotyping reverse GACCAACTATATTGTTTTCCACTGC	IDT	N/A
Foxj1a gRNA sequence GGACGTGTGCTGTCCTGTGC	IDT	N/A
Software and Algorithms
ZiFiT Targeter	Young lab	http://zifit.partners.org/ZiFiT/
ImageJ/Fiji	[75]	[75]
Cell counter plugin for Fiji/ImageJ	Kurt De Vos, University of Sheffield	https://imagej.nih.gov/ij/plugins/cell-counter.html
spermQ	Wachten lab, University of Bonn	[43]
Zebrascope software in Lab View	Ahrens lab, Janelia Farm	[76]
PIVlab v.1.41	Thielicke et al.	[46]
Fast Fourier analysis	MATLAB, this paper	N/A
Other
Pressure injector	Eppendorf	Femtojet 4i
Confocal microscope	Zeiss	Examiner Z1
20x water immersion Plan-Apochromat NA 1	Zeiss	421452-9880-000
20x Plan-Apochromat NA 0.8	Zeiss	420650-9901-000
Stereomicroscope (20x Plan-Apochromat, NA 0.8)	Zeiss	Axio Imager M1
Sutter Laser puller	Sutter	Model P-2000