Animals

All experimental procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of California, Irvine (Protocol #AUP-23-017), and were conducted in accordance with national and institutional guidelines for the care and use of laboratory animals. Eight-week-old Swiss Webster male mice were obtained from Charles River (North Carolina, USA). The mice were maintained on a 12-h light/dark cycle and provided with food and water ad libitum.

Brain tissue collection

Mice were deeply anesthetized using isoflurane and subsequently transcardially perfused with a combination of 0.9% saline and 4% paraformaldehyde (PFA) [22]. We used a total of 56 mice, perfusing 4–5 mice at every 2-h interval over a 24-h period, resulting in 12 time points. This sample size was chosen to provide sufficient statistical power to detect significant cilia morphological differences, as validated for accuracy and reproducibility in our earlier publications. Following perfusion, we harvested the brains, preserved them in PFA overnight, and then transferred them to a 30% sucrose solution. The brains were then coronally sectioned at a thickness of 30 μm using a microtome.

Immunohistochemistry and imaging

For immunohistochemistry preparation, 3−4 sections were selected from various levels across the brain. The sections were then blocked using goat serum in PBS containing 0.3% Triton X-100 for 1 h. Following this step, the brain sections were incubated in blocking buffer solution containing the rabbit primary anti-adenylate cyclase 3 (ADCY3) antibody (LSBIO-C204505), at a dilution of 1:500. Post-primary antibody incubation, the sections were rinsed with PBS and incubated with the secondary antibody, Invitrogen AlexaFluor546 goat anti-rabbit, at a dilution of 1:500, along with DAPI at a dilution of 1:10,000 (Thermo Fisher Scientific). The sections were then washed with PBS and mounted onto slides. The cilia were imaged using the Keyence BZ-9000 fluorescence microscope with a 20× objective lens. Brain regions analyzed include: Anterior hypothalamic area (AHC), Arcuate nucleus (ARC), Dorsomedial hypothalamus (DMH), Lateral hypothalamus (LH), Median preoptic area (MPO), Hypothalamic paraventricular nucleus (PVN), Suprachiasmatic nucleus (SCN), Supraoptic nucleus (SON), Ventromedial hypothalamus (VMH), Caudal striatum (CS), Nucleus accumbens core (NAc), Nucleus accumbens shell (NAs), Rostral striatum (RS), Olfactory tubercle (TU), Cornu Ammonis 1 (CA1), Cornu Ammonis 3 (CA3), Dentate Gyrus (DG), Motor cortex (MC), Prefrontal cortex (PFC), Piriform cortex (Pir), Somatosensory cortex (SSC), Medial amygdala (MA). Dual-staining immunohistochemistry was performed to verify the base and tip of primary cilia. The basal body was detected using a rabbit anti-Centrin 1 polyclonal antibody (Proteintech, 12794-1-AP) at a 1:200 dilution, while the cilia were stained with a chicken anti-ADCY3 antibody (EnCor Bio, CPCA-ACIII) at a 1:500 dilution. Secondary antibodies included Alexa Fluor 488 goat anti-rabbit (Invitrogen, A-11008) and Alexa Fluor 564 goat anti-chicken (Invitrogen, A-11040), both at a 1:200 dilution. Images were captured using fluorescence microscopy at 40x magnification.

Cilia length and angle measurement

Our first step involved pre-processing the images by converting them to grayscale, to make them conducive for analysis and to mitigate color-related complications. We then applied Gaussian blur to the images to minimize noise or minor details that might compromise the accuracy of our measurements. Next, we used canny edge detection to find all the edges in the images. This allowed us to identify the cilia more easily and accurately. We applied a threshold value to the canny edge detection output to obtain a binary image where the edges were represented by white pixels and the background is represented by black pixels. After obtaining the binary image, we used the findContours() function in OpenCV to detect all the cilia in the image. This function identifies and segments all the distinct objects in the image, in this case, the cilia, and returns a list of their contours. However, some of these contours may be noise or not relevant to our analysis. Therefore, we filtered them based on their size, aspect ratio, and other factors to only capture the ones that are likely to be cilia. For instance, we filter the contours based on their length, as cilia are typically longer than they are wide. We set a minimum and maximum length threshold for the contours to keep only the ones that fall within this range. In our implementation, we use a minimum length of 1 µm and a maximum length of 15 µm. Furthermore, we filter the contours based on their aspect ratio, as cilia are generally slender structures with a length-to-width ratio greater than 3:2. We set a minimum aspect ratio threshold for the contours to remove any that do not meet this criterion. In our implementation, we used a minimum aspect ratio of 1.5. For each contour, we fit a bounding box around it using the cv2.boundingRect() function. This gave us the length and angle of the cilia. The length was calculated as the distance between the top and bottom points of the bounding box, and the angle was calculated as the angle between the major axis of the box and the horizontal axis. However, we also needed to determine the base of the cilia accurately. To achieve this, we used the cv2.minMaxLoc() function to find the brightest pixel in the grayscale image within the bounding box of the cilia. The cilia’s base is typically the most illuminated section, and this method enables us to pinpoint it with precision. To further improve the accuracy of our angle measurements, we combine the location of the base of the cilia with the center point of the bounding box. By comparing the relative positions of these points, we were able to determine which quartile the cilia were facing and adjust the angle measurement accordingly. Finally, we saved all the information about the cilia, including their lengths, angles, and positions, in a CSV file. This allowed us to analyze the data more easily and systematically. The two programs we developed and used are provided as supplementary materials (S1 Code), referred to as Program 1 and Program 2. A detailed, step-by-step description of the analysis pipeline is outlined below, and a visual walkthrough demonstrating each step is available in supplementary materials (S1 Video).

Analysis pipeline

The cilia detection pipeline consists of three main phases: contours extraction, contours filtering, length, and angle measurement. The specific OpenCV functions used for each step are included.

Circular angles’ mean calculation

Due to the circular nature of angles, the traditional arithmetic mean method is not applicable for calculating angle means. Therefore, we computed the circular mean, also known as the angular mean, to calculate meaningful statistics for our angle dataset. The circular mean of our angle dataset was calculated as follows: First, the angle values were converted to radians, and all the data were subsequently transformed into sines and cosines. The means of the sines and cosines were calculated by summing the values and dividing them by the total count of each value. The circular mean was determined using the atan2 function, a variant of the arctangent function that accepts two arguments and yields results within the appropriate quadrant. The obtained results were then converted back to degrees using the equation: Degrees = radians × (180/π). If negative outputs were obtained from the atan2 function and the data ranged from 0 to 360, 360 was added to those values to align them within the correct range.

Quantification of cilia density across mouse brain

To determine the cilia density in each of the 22 brain regions, we calculated the percentage of ADCY3-stained cilia relative to the number of DAPI-stained cells within the same region.

In order to detect cells in DAPI-staining images, the pipeline processes images to detect and analyze contours, focusing on the identification of significant structures within the image. The methodology consists of three main phases: contours extraction, contours clustering, and contours filtering.

Phase 3: Contours filtering.

1. Minimum and maximum area:
   • Purpose: Ensure that only contours of relevant size are considered for analysis.
   • Method: Filter contours based on area (in pixels) using OpenCV’s cv2.contourArea function with minimum area = 20 pixels and maximum area = 1000 pixels. This step excludes contours that are too small or too large to be relevant.
2. Minimum circularity:
   • Purpose: Eliminate contours that do not meet the minimum shape criteria, focusing on more circular objects.

Method: Calculate the circularity of each contour and retain only those with a circularity above the specified threshold using OpenCV’s cv2.arcLength function and a minimum circularity of 0.01. Circularity is calculated as:

Verification of the automated measurements of cilia lengths and angles

To validate the automated measurements of cilia lengths and angles, and to verify the accuracy of our pipeline in identifying and measuring cilia structures, we conducted manual checks on random samples, selecting at least 10 cilia from each of the 22 brain regions. The measurements were performed blindly using both manual and automated methods. We then compared the results by calculating the recall and mean absolute error (MAE) for both length and angle measurements.

Since cilia length and orientation were measured from 2D images, projecting 3D structures like cilia onto a 2D plane can introduce measurement errors. To address this, we estimated the potential errors introduced by this approach using basic geometric principles, including the Pythagorean theorem. Detailed calculations are provided in the supplementary material (Figs A–C in S1 Text).

Comparative and correlational analysis of cilia lengths and angles

We analyzed the data using two different approaches: first, by assessing the lengths and angles of individual cilia to provide a detailed frequency distribution across all examined brain regions; and second, by calculating the mean length and angle of cilia in each section of the left and right hemispheres, treating these averages as single values for further analysis.

To analyze differences in cilia lengths and angles across various brain regions at different times of the day, we performed one-way ANOVA. Multiple comparisons were adjusted by controlling the False Discovery Rate (FDR) using the two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli. Additionally, we computed the Pearson correlation coefficient (r) for each pair of region-specific cilia lengths and angles across 24-h time points, generating a correlation matrix to examine the relationships among all brain regions. Data were analyzed using GraphPad Prism 9.4, with statistical significance set at p < 0.05, and expressed as mean ± standard error of the mean (SEM).

Network analysis

Using the Allen Mouse Brain Connectivity Atlas data [51], we built a structural network of the regions used in our study. We normalized projections from source regions to other regions of interest by dividing the projection density by the injection volume. The normalized projection density was averaged across all experiments for each pair of source and target regions to generate a connectivity matrix. These values were log normalized to produce a more effective distribution for use as edge weights in a network. Community detection was carried out using the Louvain algorithm implemented in scikit-network with a resolution parameter of 1 [52]. We identified three main clusters of brain regions: purple (NAc, NAs, and others), green (PFC, MC, and others), and red (CA1, CA3, and DG). We examined the correlation between cilia length and angle with various network features. We examined all pairs of regions within and across different clusters, identifying varying fractions of region pairs correlated by either cilia length or angle. We also analyzed whether these fractions would likely occur if regions were correlated randomly, shuffling all correlations across region pairs for this purpose. We constructed cilia correlation networks using a log-scaled R value to achieve a desirable distribution for edge weights. Network analysis was performed in Python using NetworkX [53], with network visualizations created using Plotly [54].

Circadian rhythmicity analysis

BIO_CYCLE, a deep-learning model designed for assessing periodicity of signals, was utilized to determine significant circadian patterns in cilia length and angle, as well as their oscillation’s amplitude and phase [55,56]. BIO_CYCLE is trained using synthetic and real biological time series datasets with labeled periodic and aperiodic signals. This tool comprises a deep neural network (DNN) for classification, which distinguishes periodic from non-periodic signals, and another DNN for regression estimating the signal’s period, phase, and amplitude. The oscillation status of a parameter is defined based on a p-value (with a cutoff of 0.05) computed by BIO_CYCLE as follows: N aperiodic signals are first generated from the synthetic time series datasets, and the N output values V(i) (i = 1, …, N) of the classification DNN on these aperiodic signals are calculated. The N output values of the classification DNN on these aperiodic signals are then calculated, forming the basis for the null hypothesis distribution. The output value V of the new signal s is then compared to V(i) (i = 1, …, N), producing the estimate for the probability of obtaining an output of size V or greater (p-value), assuming that the signal is part of the null distribution (aperiodic signals). A smaller p-value suggests a higher likelihood of the signal being periodic. The related q-values are calculated via the Benjamini and Hochberg procedure. BIO_CYCLE is freely accessible from the CircadiOmics web portal: http://circadiomics.igb.uci.edu.

Methodological verification

To validate the accuracy of our automated system for measuring cilia length and angle, we conducted a comparative analysis with manual measurements. In each of the 22 brain regions analyzed, 10 cilia were measured manually. The average of these manual measurements was then compared to the automated measurements. This comparative analysis showed a recall of 93.46%, indicating high reliability in the automated detection of cilia. The MAE for cilia length was 1.33 µm, and the MAE for angle measurements was 7.759 degrees (S1 Data). We also estimated the instrumental error introduced by measuring 3D cilia structures in 2D Detailed calculations are provided in the Supplementary results (Figs A–C in S1 Text).

Region-dependent variations in cilia density, length, and angle in the mouse brain

Our analysis demonstrated marked variability in cilia density across the 22 brain regions examined. The lowest density was observed in the LH, where only 35% of cells were ciliated, while the highest density was found in the CA1 region of the hippocampus, with 90% of cells containing cilia (Fig 1E and S1 Data).

We assessed the average cilia length within specific brain regions using two distinct approaches. First, in the direct regional average method, we calculated the average length of each cilium individually within each region, revealing a variation in cilia lengths from 4.79 µm in the NAs to 6.33 µm in the TU (panel a of Fig E in S1 Text). Across all brain regions, including over 10 million cilia, this method resulted in an overall mean cilia length of 5.15 µm.

Second, in the section-based average method, we calculated the average of lengths of all cilia within each section to obtain a single value per section, then used these sections’ averages to calculate the overall mean cilia length for each region. The brain-wide cilia length average was 5.4 ± 0.01 µm, with regional averages ranging from 4.87 µm in the NAs to 6.3 µm in the TU (Fig 2A and S2 Data). Additionally, a frequency distribution analysis showed that approximately 40% of cilia across all examined regions had lengths within the 4.8–5.4 µm range (Fig 2B and 2C). However, the frequency distribution analysis revealed a varied distribution of cilia lengths across the 22 brain regions. Some regions, like the ARC, MPO, SCN, and NAs have most cilia lengths concentrated around a narrow range, whereas other regions, like the CS and RS, exhibit a broader distribution, with cilia lengths spread across a wider range (panel b of Fig E in S1 Text).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Cilia length analysis in different brain regions.

(A) Histogram showing the means±SE of cilia length in the 22 brain regions based on the average length measured in each brain section. The x-axis represents cilia length in micrometers and the y-axis represents the percentage frequency of occurrence. (B and C) Histograms showing the frequency distribution of cilia length in (B) the entire brain and (C) the 22 individual brain regions, using the section-based average method. For each brain section, the lengths of all cilia were averaged to obtain a single value, which was then used to calculate the overall mean cilia length for each region. The mean cilia length was derived from the average lengths of both left and right hemisphere sections (3–4 sections per region). The x-axis represents cilia length in micrometers, and the y-axis shows the percentage frequency of occurrence; underlying data for Fig 2 are available in S2 Data.

https://doi.org/10.1371/journal.pbio.3003197.g002

Our analysis of individual cilia angles of over 10 million cilia showed a distinct pattern across all regions (Fig 3 and Fig F in S1 Text). Notably, the highest distribution of cilia orientation across the entire brain was toward 180° and 270°, indicating a primary medial and/or ventral orientation of cilia (Fig 3 and Fig F in S1 Text). However, cilia orientation did not exhibit uniform patterns across different brain regions (Fig 3). The RS, CS, MA, and CA3 showed a predominant cilia orientation toward 180°, and a secondary preference for 270°, whereas the NAc, TU, MPO, SCN, DG, and CA1 exhibited a primary orientation toward 270°, followed closely by 180° (Fig 3). A more scattered distribution of cilia orientation was observed in the hypothalamic nuclei, various cortices, and the NAs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Cilia circular angle analysis, based on individual cilia measurements.

Rose plot displaying the frequency distribution of cilia circular angles in individual brain regions from all individual measured cilia at 30° bins, Underlying data are available on the Zenodo database (https://doi.org/10.5281/zenodo.15151271).

https://doi.org/10.1371/journal.pbio.3003197.g003

When using the section-based averaging, where the circular mean of all cilia angles within each section was first determined and treated as a single value, cilia angle means and distributions revealed distinct patterns. Cilia angles predominantly favored the third quadrant (180°–270°), particularly around 250°, when measured relative to the horizontal line in coronal sections (Fig 4A, 4B and panel a of Fig G in S1 Text). The Pir exhibited the highest circular mean cilia angle (267°), whereas the MC displayed the lowest (219°) (Fig 4B and panel b of Fig G in S1 Text).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Cilia circular angle analysis based on section-averaged circular means.

(A) Radial plot showing the frequency distribution of cilia circular angles across the entire brain and individual brain regions. The analysis is based on section-averaged circular means, where the circular mean of all cilia angles within each section was first calculated and treated as a single value. Underlying data are available in S2 Data. (B) Radial plots displaying the mean cilia circular angles for the 22 brain regions, grouped into major categories: hypothalamus, hippocampus, cortices, and striatum. The average angle for each region was determined by calculating the circular angle means for each brain section. Underlying data are available in S2 Data.

https://doi.org/10.1371/journal.pbio.3003197.g004

We found no significant correlation between cilia density and cilia length (Pearson’s coefficient, r = −0.08, P = 0.718), whereas there was a trend toward a positive correlation between cilia density and orientation (r = 0.38, P = 0.084).

Temporal fluctuations and rhythmicity in cilia length and orientation across brain regions

We next analyzed cilia lengths and orientations at 2-h intervals over a 24-h period (S2 Data), and found complex, time-dependent fluctuations in both cilia length and orientation across various brain regions (Fig 5A and 5B). Notably, all regions except the AHC and CS exhibited significant variations in cilia length, with distinct fluctuation patterns observed during the light and dark phases (Fig 5A–5C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Fluctuation of cilia length and cilia circular angle across 24-hours’ time.

(A and B) Diurnal fluctuations of cilia length (A) and angle (B): The plots show the means ± SE of cilia length and angle over the course of 24-h day in different brain regions. The brain regions are grouped into major categories including, the Hypothalamus, Hippocampus, Cortices, and Striatum. The average length and angle are determined based on the measurements’ means in each brain section; Zeitgeber time (ZT). One-way ANOVA test was used to compare the means of lengths and angles at different time points. P values were calculated using the False Discovery Rate (FDR) correction, employing the Two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli for multiple comparisons. (C) P values of one-way ANOVA test, used to assess the variability in cilia length and angle at different time points (in a,b), P < 0.05: significant changes across the different time points over 24-h period. Underlying data for Fig 5 are available in S2 Data.

https://doi.org/10.1371/journal.pbio.3003197.g005

Notably, the MA, LH, RS, CS, TU, AHC, CA3, and NAc reached their maximum cilia lengths at specific zeitgeber times (ZTs) during the light phase, whereas the ARC, DG, NAs, and PFC showed peak lengths at different ZTs during the dark phase (Fig 5A and 5C). Interestingly, both the DMH and VMH exhibited a bimodal pattern, with cilia length peaks occurring at ZT 6 and ZT 18 during the light and dark phases, respectively (Fig 5A and 5C).

We observed significant time-dependent fluctuations in cilia angles within seven brain regions: DMH, PVN, RS, MC, PFC, SSC, and TU (Fig 5B and 5C). Notably, the DMH and PVN, two hypothalamic nuclei, displayed contrasting cilia angle patterns. The DMH reached its highest cilia angle between ZT 16–20, while the PVN showed its lowest angle during this same period. Conversely, at ZT 6, the DMH exhibited its lowest angle, and the PVN reached its highest (Fig 5B and 5C). The PFC’s cilia angle changes were opposite to those in the SSC, with the PFC peaking at ZT 22, while the SSC reached its lowest angle at that time. The MC showed the most extensive range of cilia angle shifts, varying from 174° at ZT 20°–245° at ZT 16 (Fig 5B and 5C).

Our analysis revealed significant periodic fluctuations in cilia length within five specific regions: ARC, DMH, VMH, NAc, and SSC, with peak times varying among them: ARC at ZT 15, DMH at ZT 3.8, VMH at ZT 3.4, NAc at ZT 11, and SSC at ZT 12.16 (Fig 6). However, we found no evidence of rhythmic patterns in cilia angles in any of the regions studied (S3 Data).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Circadian rhythms of cilia length.

BioCycle analysis was used to examine the presence of rhythmic fluctuations in cilia length and revealed significant circadian patterns of cilia length variations across a 24-h period in five brain regions tested: ARC, DMH, VMH, NAc, and SSC, with P < 0.05. Underlying data are available in S3 Data.

https://doi.org/10.1371/journal.pbio.3003197.g006

Correlation of cilia length and circular angle among various brain regions

We next assessed whether fluctuations in cilia length and angle over the 24-h period correlated across different brain regions. Our analysis revealed significant correlations in cilia length changes, with the DMH and VMH showing a particularly strong correlation (r = 9.3). Positive correlations were also observed between the PFC and several regions, including the DMH, VMH, LH, AHC, and MC, while CA1 cilia lengths negatively correlated with the DMH and VMH but positively with the RS, NAc, and DG (Fig 7A). Similarly, cilia angle fluctuations exhibited substantial correlations across various brain regions. The strongest correlation was observed between the DMH and VMH (r = 0.79), with other notable correlations in hypothalamic regions such as AHC-LH, ARC-PVN, and PVN-SCN. Negative correlations were found between the PVN and VMH, and between cilia angles in the CA1 and TU regions (Fig 7B). We next tested the hypothesis that more highly correlated brain regions, either in cilia length or angle, may be preferentially interconnected. Using the Louvain community detection algorithm on a network representation of the anatomical connections among these regions, we found three primary communities. The first and smallest community (“red”) contained the three hippocampal regions DG, CA1, and CA3. The second community (“green”) contained cortical regions, the striatum, and somewhat surprisingly, the SCN. The third community (“purple”) contained all of the other regions, which overall had less strongly defined interconnectivity relationships (Fig 7D). Based on these data, we then asked if regions that had strong length correlations (p < 0.05) were more likely to have stronger interconnectivity than regions with weak length correlations (p > 0.05). We found no significant difference in normalized projection strength between pairs of regions correlated with cilia length versus those not correlated (Fig 7E), indicating that projection strength between regions does not influence cilia length. We next examined all pairs of regions within and across different clusters and identified a bias toward regions having correlated cilia lengths if they were present within particular communities; this relationship was most prominent between regions in the red and green communities (Fig 7F). The fraction of correlated pairs between red-green communities (0.111) was in the 9.1st percentile relative to the random distribution, indicating that cilia lengths were more likely than by chance to be correlated between pairs located in red and green communities (Fig 7F).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Correlation of cilia length and circular angle among various brain regions and relationship between cilia length/angle and network connectivity.

(A) Heatmap showing the correlation of cilia length (based on the mean of the section means) among different brain regions. The correlation coefficient ranges from 1 (indicating the most positive correlation, red) to −1 (indicating the most negative correlation, blue). Positive correlation implies that the cilia lengths in the two regions are on the longer and shorter side simultaneously. Negative correlation indicates that the two regions have opposite cilia length sizes; underlying data are available in S2 Data. (B) Heatmap displaying the correlation of cilia angle (based on the mean of the section circular means) among different brain regions. The correlation coefficient ranges from 1 (indicating the most positive correlation, purple) to −1 (indicating the most negative correlation, green); underlying data are available in S2 Data. (C) Anatomical network of the brain regions in our study. The network is embedded in 2D space using a spring embedding, in which highly connected nodes are pulled together as by a spring. Node size represents betweenness, a measurement of centrality based on the number of shortest paths between nodes that a particular node shows up on. Nodes are colored according to their community assignment, as determined by the Leiden algorithm. (D) Correlogram indicating the strength of input-output connections between all pairs of regions in the dataset. The color bar indicates the log-normalized projection density between any two regions. Regions are grouped according to their community membership. (E) Bar graph comparing projection strength between regions significantly correlated with cilia length vs. those not correlated. Projection strength equals the log normalization of normalized (by injection volume) projection density. (F) Fraction of region pairs significantly correlated by cilia length among all regions pairs between two communities. For example, a fraction of 0.111 means that 11.1% of all region pairs between the red and green communities are correlated by cilia length. (G) Bar graph comparing projection strength between regions significantly correlated with cilia angle versus those not correlated. (H) Fraction of region pairs significantly correlated by cilia angle among all regions pairs between two communities. (I) Cilia length correlation network. The spring embedding puts regions more highly correlated closer together. Betweenness and community membership are shown as node size and color, respectively. (J) Cilia length correlogram shows the groupings of different regions by community, as well as their correlation patterns. (K) Cilia angle correlation network. Betweenness and community membership are shown as node size and color, respectively. (L) Cilia angle correlogram shows the groupings of different regions by community, as well as their correlation patterns. Underlying data of Fig 7C-L are available in S2 Data, and from the publicly available Allen Mouse Brain Connectivity Atlas Data.

https://doi.org/10.1371/journal.pbio.3003197.g007

We next tested for the same relationship between cilia angle and connectivity. As with cilia length, we did not identify a significant difference in projection strength between pairs of regions correlated with cilia angle versus those not correlated (Fig 7G). We then assessed the relationship between community identity and cilia angle correlation and found that the fraction of correlated pairs within the purple community (0.107) was 7.4th percentile relative to the random distribution (Fig 7H). These data together indicate that while the connectivity strength between directly connected pairs did not appear to relate to cilia length or angle, the presence within particular communities did appear to relate to both cilia length and angle, indicating that the network architecture of the brain exhibits some relationship with cilia length and angle correlations across regions.

Given the lack of clear relationship between direct anatomical connectivity and cilia length or angle correlations, we next applied the network approach on the cilia length and angle correlation data directly. In order to do this, we built correlation networks using a log-scaled R value that yielded a nicer distribution for edge weights, and again used the Louvain community detection algorithm to assign communities (Fig 7I–7L). These results yielded four communities that did not distinctly track by anatomical connectivity, as evidenced by a different regional distribution across communities. Interestingly, we observed a number of regions that cluster together in both cilia length and angle correlation networks. Notably, NAc, CA1, CA3, TU cluster together for both, as do AHC, PFC, MC, and SSC (Fig 7I–7L). The NAc and TU are both striatal regions while CA1 and CA3 are both hippocampal, and PFC, MC, and SSC are all cortical regions. However, the exact relationship between these sets of regions that co-cluster is not immediately apparent, again suggesting that factors other than direct connectivity likely influence the non-random correlations in cilia length and angle across regions.