Daily mitochondrial dynamics in cone photoreceptors

We performed detailed three-dimensional (3D) analyses of cone mitochondrial clusters using 50 nm Z-sections collected by serial block-face scanning electron microscopy (SBFSEM) (Fig. 1B and Movie S1). Image stacks were analyzed using either a rapid dot quantification method or manual segmentation to compare mitochondrial number, size, shape, and location between day (15:00, Zeitgeber Time 6 [ZT6]) and night (01:00, ZT16). Individual red-green, blue, and UV cone image stacks with corresponding 3D renderings of mitochondrial clusters and OSs at ZT6 and ZT16 are presented in Fig. 1C. SI Appendix, Fig. S2 presents 3D renderings with mitochondrial numbers and cluster volumes for all cells in this study.

We found that at night, blue and UV cone mitochondria increase in number by 17 ± 10% and 57 ± 6%, respectively (Fig. 1 D, Top). These changes did not coincide with an increase in cluster volume (Fig. 1 D, Middle); on average, volumes of individual blue and UV cone mitochondria decreased at night by 29 ± 3% and 31 ± 6%, respectively (Fig. 1 D, Bottom). One animal in the study (SI Appendix, Fig. S2, fish 7) had overall larger cones than all others, and its blue cones were identified as outliers using a robust regression followed by outlier identification test (transparent markers at ZT16 in Fig. 1D). For transparency we included these two cones in the analysis, but this did not affect the conclusions. Green cones, which in zebrafish exist in red-green double cones, appeared in two populations at both timepoints. Four animals in our study have very large green cones (>700 mitochondria), while the green cones of four other animals are smaller (300 to 450 mitochondria) (Fig. 1D and SI Appendix, Fig. S2). Despite this bimodal population, mitochondrial volume for green cones is maintained between 0.4 and 0.7 µm³ at ZT6 and ZT16. This suggests that single and double cone subtypes undergo different cycles of mitochondrial dynamics.

Mitochondrial cluster shape was also examined for each cone subtype using transgenic zebrafish expressing YFP targeted to cone mitochondria (gnat2:mito-cpYFP) (28), counterstained with antibodies targeting components of the mitochondrial respiratory chain (SI Appendix, Fig. S1A and Table S1 lists antibodies used). Individual cone mitochondrial cluster lengths and circularity ratios were calculated. For all cone types, cluster length increases at night by ∼50% (SI Appendix, Fig. S1 B, Top), but cluster width, reflected by the circularity ratio, decreases at this time (SI Appendix, Fig. S1 B, Bottom). Both LD and DD groups exhibited cyclical changes in cluster morphology.

Surface area and volume of each manually segmented mitochondrion was used to calculate mitochondrial complexity index (MCI), a size-insensitive measure of morphological complexity (29). Mitochondria within clusters were heterogeneous in volume and MCI (Fig. 2A), and MCI was poorly correlated with volume (Fig. 2B). Four individual mitochondria with the same volume but different MCIs are presented in Fig. 2C. Compared to mouse cones, zebrafish cone mitochondria are simpler on average but occupy a larger volume (SI Appendix, Fig. S3).

Our analyses also revealed morphological distinctions between cone subtypes. Clusters in green cones are longer (SI Appendix, Figs. S1C and S2) and have smaller mitochondria (Fig. 1D). Fig. 2D and Movie S2 depict 3D renderings of manually segmented green, blue, and UV cone clusters. At ZT6, green, blue, and UV mitochondria have significantly different distributions of mitochondrial volume, with mitochondrial volumes largest in UV cones and smallest in green cones, but cone subtypes do not have significantly different MCI distributions (Fig. 2E).

To compare mitochondrial volume and MCI between day and night, we analyzed manually segmented green, blue, and UV cones at ZT6 and ZT16. At ZT16, mitochondria are significantly smaller in blue and UV cones, but not green cones (Fig. 2F). The MCI of all cone subtypes shifts significantly toward having more simple mitochondria at night. These results suggest that green cones can alter their mitochondrial complexity without changing volume.

Cones of several species (30–32), including zebrafish (33), contain large megamitochondria. To examine the distribution of megamitochondria in zebrafish cone clusters, we 3D rendered the three largest mitochondria in manually segmented cones in their respective clusters. In all cone subtypes, megamitochondria localize within the cluster core, often in direct contact (Fig. 3A). We did not detect striking differences in megamitochondria between day and night, but note that the volume, location, and complexity of megamitochondria vary between cone subtypes (SI Appendix, Fig. S4). Megamitochondria in green cones localize to the top of the cluster and are relatively uniform in size and of average complexity. Single cones contain distinct megamitochondria whose volume comprises 12 to 22% of the cluster. We observed several megamitochondria with projections extending toward the cluster periphery, and UV cone megamitochondria are unusually complex.

To examine the distribution of mitochondrial sizes across the cluster, each mitochondrion from SBFSEM analyses was plotted as a single point in 3D using its center X-Y-Z coordinates (Fig. 3B). For manual segmentation, point size was proportional to mitochondrial volume. For dot quantification, point size was proportional to the number of dots needed to track each mitochondrion; larger points represent larger or more branched mitochondria. Points were separated into peripheral and core populations according to distance from the cluster 3D central axis. Similarly, top and bottom populations were defined by proximity to the OS or cluster base. Mitochondrial point sizes were quantified in relation to cluster position (Fig. 3C), and ratios of core/periphery and top/bottom were used to assay relative size distribution (Fig. 3D).

At ZT6, cones have significantly more large mitochondria at the cluster core, compared to the periphery (Fig. 3 C and D, Left). At ZT16 average mitochondrial point size decreases 30 to 45% in the core and ∼30% in the periphery; size also decreases ∼35 to 50% at the bottom of the cluster (Fig. 3C). Conceptually, this represents a shift toward more small mitochondria at the base of the cluster at night (Fig. 3D). Compared to single cones, green cones underwent similar changes to mitochondrial size distribution, but maintain a significantly larger population of small mitochondria at the cluster base (Fig. 3 D, Right). These data suggest that in daytime all cones maintain a population of small mitochondria at the cluster periphery; at night mitochondrial size becomes more uniform across the cluster.

In other cells the organelle endoplasmic reticulum (ER) initiates mitochondrial fission (34, 35), a process that occurs at the end of mitogenesis (36). Zebrafish cone ER primarily contacts mitochondria at the cluster base and periphery (28), where mitochondrial size is most dynamic. We examined the cone ER network surrounding mitochondria at all timepoints. Using sections from LD or DD transgenic zebrafish expressing GFP targeted to cone ER (gnat2:er-GFP) (37), we performed IHC with antibodies against the mitochondrial respiratory chain (Fig. 3E). To quantify potential ER contacts for individual blue and UV cone clusters, ER–mitochondrial apposition ratios were calculated by dividing the longest length of ER adjacent to the cluster by the cluster length.

In the daytime, ER tightly associates with the entire cluster surface and is densely packed around its base (Fig. 3 E, Top). At night, apposition ratios decrease as the ER becomes more diffuse and extends less toward the OS (Fig. 3 E, Bottom and Fig. 3F). This shows that the ER network changes to accommodate the mitochondrial cluster and is consistent with the shift toward larger mitochondria at the top and periphery of single cone clusters at night (Fig. 3D). At most timepoints ER–mitochondrial apposition ratios are lower for the DD group (Fig. 3F), suggesting that light exposure drives tighter ER–mitochondrial associations.

Increased mitogenesis could result in more cone mitochondria at night. Mitogenesis is controlled at the transcriptional level (38, 39): upstream signals lead to mitochondrial DNA (mtDNA) replication, protein synthesis, and fission creating new mitochondria (Fig. 4A). We measured mRNA transcript levels of mitogenesis genes using qPCR with whole retinas (40) (SI Appendix, Table S3 lists primers; SI Appendix, Table S4 lists fitting parameters for cosinor curves). The early nuclear transcription factors pgc1α and pgc1β rise in the morning, while the mitochondrial transcription factor tfam, DNA polymerase polg1, and deacetylase sirt3 all peak before night onset (Fig. 4B). Together this suggests that canonical mitogenesis increases at night in whole retinas.

Transcripts encoding the mitochondrial fusion protein mfn2 rise in the morning (Fig. 4 C, Top), suggesting fusion could mediate the corresponding decrease in mitochondrial number in cones; a recent study found that cones contribute roughly half of the mitochondrial proteins in adult zebrafish retina (41). No significant changes were detected in the zebrafish cone mitochondrial enlarging factor es1 (42) (Fig. 4 C, Middle). As a control we examined aanat2 (43), which displays robust circadian changes in expression (Fig. 4 C, Bottom). Transcript levels between LD and DD groups were similar, and tfam, polg1, and sirt3 met statistical cutoffs for rhythmicity, indicating that retinal expression of mitogenesis genes is regulated primarily by the circadian clock.

A selective form of autophagy called mitophagy clears damaged or unnecessary mitochondria (44). While several pathways can trigger mitophagy (45, 46), all coalesce on recruitment of LC3 to maturing mito-autophagosomes. We performed IHC at all timepoints using LD or DD transgenic zebrafish expressing GFP fused to LC3 in cones (gnat2:GFP-LC3) (37) and antibodies against the mitochondrial respiratory chain (Fig. 4D). LC3-positive autophagosomes overlapping with blue and UV cone mitochondrial clusters were quantified. The number of mitochondrial-associated autophagosomes increases twofold at light onset (Fig. 4E). While changes to autophagosome numbers met statistical cutoffs for circadian rhythmicity, more mitochondrial-associated autophagosomes were present in the DD group in the evening, possibly indicative of enhanced mitochondrial turnover in prolonged darkness.

When viewed using EM, autophagosomes appear as lucent, multivesicular structures (Fig. 4 F, Left, yellow arrowheads), while mitochondria are electron dense and contain cristae folds. We validated our IHC findings by quantifying the number of electron-lucent structures in all cone mitochondrial clusters imaged using SBFSEM. More lucent structures are present in clusters at ZT6 (Fig. 4 F, Right), when more mitochondrial-associated autophagosomes were detected using IHC. Several clusters contain >30 lucent structures in the daytime, suggestive of a mitophagic event at this timepoint.

Degraded mitochondrial material canonically enters the endolysosomal pathway (45), but in neurons it can translocate and leave the cell (47, 48). Mitochondria mislocalize toward the nucleus in degenerating human cones (12) and in a zebrafish cone model of mitochondrial calcium overload (41). Four hours before light onset (05:00, ZT20), we observed mislocalized cone mitochondria using IHC with gnat2:mito-cpYFP transgenic zebrafish and three different mitochondrial respiratory chain markers (Fig. 4G, yellow arrowheads). The structures are feasibly the size of a single mitochondrion (0.5- to 1-µm length), contain both cone mito-cpYFP and respiratory proteins, and lie between the cluster and the nucleus. They were present only at overnight timepoints ZT16 and ZT20, and in multiple animals over two generations. While we could not unequivocally identify these structures in SBFSEM, mitochondrial trafficking may occur between the cluster and cell body prior to light onset.

In nearly all cones, we found dark deposits inside and between mitochondria throughout the day in SBFSEM images (Fig. 5 A, Top). Similar deposits are present in published EM images of cone mitochondrial clusters in other zebrafish (figure 2 of ref. 16), walleye (figures 5 and 7 of ref. 49), frogs (figure 2 of ref. 50), pigeons (figure 16 of ref. 51), shrews (figures 2, 7, and 10 of ref. 52), and ground squirrels (figure 5 of ref. 17), as well as mice and albino zebrafish (SI Appendix, Figs. S3 and S5, yellow arrowheads). When manually segmented and 3D rendered, we found these electron-dense lamellar whorls can span several micrometers, contacting the matrices of multiple mitochondria and/or the OS (Fig. 5 A, Bottom). Most deposits lie inside of single mitochondria, but they appear in the cytosol and occasionally cross the plasma membrane (Fig. 5 B, Top). This apparent extrusion most often occurs as single events, but some cells displayed multiple concurrent exit sites (Fig. 5 B, Bottom).

Other neurons can eject mitochondrial material via nanotunnel-like extensions (47). Three-dimensional analysis revealed that in cones, extruded material either remains near the cell surface or forms stalks that can reach 5 µm, terminating in diffuse lamellar extracellular sacs (Fig. 5 C, Top and Movie S3). The stalks appear as 40- to 90-nm electron-dense rings surrounding a hollow core. In the extracellular space stalks and sacs can connect, creating networks that link discrete populations of mitochondria within a cluster (Fig. 5 C, Bottom and Movie S4). These deposits and their extrusion in darkness were observed in nearly all cones but seldom in rods; in one instance extrusion was observed in two neighboring rods (Fig. 5D).

To quantify cone mitochondrial deposits and extrusion events in day and night, we used a ranking system to blindly score clusters imaged with SBFSEM. The presence of mitochondrial deposits is similar between day and night (Fig. 5 E, Left). However, extrusion events occur almost exclusively at night, and most cells had multiple events (Fig. 5 E, Right). While the composition of the deposits is not known, they provide a physical link between cone mitochondria and the interphotoreceptor matrix (IPM).

Mitochondrial metabolism in whole mouse retinas is more active in darkness (3). However, ∼96% of photoreceptors in mouse retinas are rods (53), while zebrafish retinas have ∼50% cones (54, 55). To measure mitochondrial activity in cones, we performed enzyme histochemistry with fresh-frozen retina sections from LD or DD albino zebrafish collected throughout the day. Some albino mouse (56) and zebrafish (57) strains undergo light-induced retinal degeneration, but under normal facility lighting our albino zebrafish display healthy central retinal morphology and mitochondrial clusters (SI Appendix, Fig. S5). Activities of succinate dehydrogenase (SDH, complex II) and cytochrome c oxidase (COX, complex IV) were assayed separately (58); their roles in mitochondrial metabolism are highlighted in Fig. 6A. Histochemistry was performed for all timepoints in parallel, and stain intensities of single cones were measured from light microscopy images (Fig. 6B).

While COX activity remains robust and stable throughout the day, SDH activity in cones increases 25 to 30% in darkness (Fig. 6C). SDH activity in the DD group is elevated at all times of day, suggesting possible repression of SDH activity by light in cones. Additionally SDH activity appears primarily in red-green double cones, compared to strong COX activity observed in all cone subtypes, rods, and the inner retina (Fig. 6B, arrowheads).

As validation, we used metabolite labeling coupled with gas chromatography mass spectroscopy (GC/MS) to assay activity of SDH in light- or dark-adapted zebrafish retinas around ZT6. Whole retinas were incubated with U-¹³C labeled succinate (dark green boxes, Fig. 6A), which SDH converts to U-¹³C fumarate; fumarate is readily interconverted to malate via the enzyme fumarase (59). Labeled fumarate and malate accumulate faster in dark-adapted retinas (Fig. 6D). A 2.75-fold higher initial rate of formation for both fumarate and malate (Fig. 6E) is consistent with higher SDH activity in darkness observed using histochemistry. To determine if succinate uptake was altered in darkness, we assayed labeled metabolites in media after 60 min, but did not find significantly lower amounts of labeled succinate (Fig. 6F). Fig. 6G depicts steady-state metabolite labeling in retinas and media after 60 min in light or darkness; succinate levels are similar in light and dark, but metabolites downstream of SDH accumulate in the media.

In zebrafish cones, hundreds of heterogeneous mitochondria pack between the nucleus and OS (16). In this report we quantified mitochondrial number, size, complexity, and 3D distribution in zebrafish cone subtypes. Compared with mouse cones (60) (SI Appendix, Fig. S3), zebrafish cone mitochondria are more numerous, smaller, and densely packed, with elaborate cristae patterning (61). The dense packing and disparate mitochondrial volumes between cone subtypes may reflect nutrient and oxygen access in the avascular zebrafish retina (20); double cones closer to the RPE and blood supply could fuel more mitochondria. These spatial constraints, the high SDH activity we observed in double cones, and previous studies showing unique metabolic activity in short-wavelength cones (62, 63) suggest that diverse metabolic strategies support cone subtypes.

Half the animals in our SBFSEM study had larger green cone mitochondrial clusters than the others; the same retinal area was imaged for all animals and single cones did not display this feature. Retinomotor movements occur in cone subtypes at slightly different times (26), so our timepoints may have captured intermediate states for green cones. Alternately, zebrafish green cones may exist as multiple subtypes. Blue and UV cones in zebrafish each express one form of the light-sensitive protein opsin; green cones express four opsins across the adult retina (64, 65), but single-cell analysis is needed to determine if green cone subtypes exist.

Mitochondrial size is linked to energetic output; in other cells smaller mitochondria respire less (4, 66). However, in zebrafish cones small, morphologically simple mitochondria may contribute more toward energy production. Zebrafish single cone mitochondria are smaller and simpler at night, when energy demands and mitochondrial respiration are likely highest. Small mitochondria in cones populate the cluster periphery, closest to oxygen and fuels. Cristae in these small mitochondria generally display the organized, linear structure linked to higher energetic output (61).

Like other species (51, 67) zebrafish cones form megamitochondria, maintained by the enlargement factor ES1 (31, 42). We did not detect changes to es1 mRNA transcripts in whole retinas, but observed juxtaposed megamitochondria in all cones throughout the day. The biological role of megamitochondria is unknown, but their densely packed cristae suggest roles beyond canonical respiration (68). Membrane lipids concentrated in megamitochondrial cristae can form a conduit for oxygen (69, 70). Many cone megamitochondria exhibit a central body with projections extending toward the cluster periphery; these projections toward the ER could be a site for mitogenesis (67). Further, cone megamitochondria (30, 52) and mitochondrial-derived ellipsosomes (71, 72) in other species guide light toward the OS. Zebrafish visual sensitivity is regulated by light and the circadian clock (73), so daily mitochondrial rearrangements may contribute to vision.

Maintenance of healthy mitochondria and mtDNA is crucial, particularly in the retina. mtDNA mutations (13) and disrupted autophagy (74) are associated with retinal degeneration, and aging human cones accumulate mtDNA mutations and mitochondrial abnormalities (19, 75). In some forms of age-related macular degeneration, cone mitochondrial clusters remodel and mitochondria translocate toward the nucleus (12).

Genes required for mitogenesis (38) undergo circadian changes in retinal expression in a manner that supports mitogenesis at night onset, although it is unknown which retinal cell(s) this occurs in. At ER–mitochondrial contact sites the ER stimulates mitophagy (76) by recruiting autophagic machinery (77) and initiating mitochondrial fission (34). In single cones, ER concentrates around the cluster during the day, when clusters contain more autophagosomes and fewer mitochondria. Together this suggests that cones undergo daily mitochondrial turnover involving ER.

Fewer mitochondria during the day could result in the reduced cone SDH activity we detected using histochemistry. Zebrafish cones have higher SDH activity than other retinal cells, consistent with studies of human (58, 75) and salamander retinas (78). Cone SDH activity increases at night, when regulation from the zebrafish circadian clock (23, 79) could promote mitochondrial succinate uptake (80). However, we did not find evidence of increased succinate uptake in darkness, and frozen sections used for histochemistry don’t require transporters for substrate uptake. Thus, our observations of cone SDH activity likely resulted from changes to expression or posttranslational modifications, rather than succinate availability.

Light (81, 82) and time of day (83, 84) can affect SDH expression in other cells. Additionally, SDH can be regulated by competitive inhibition (85, 86), phosphorylation (87, 88), acetylation (89), and succinylation (90). mRNA transcripts for the mitochondrial deacetylase SIRT3 increase in retinas prior to night onset, but regulation of SDH by SIRT3 in retinas has not been explored. In photoreceptors, both SIRT3 and the mitochondrial desuccinylase SIRT5 are necessary for normal function (91). Further, cristae structure can influence respiration (8, 92), in part by driving SDH supercomplexation (93). SDH catalyzes a reversible reaction in mouse retinas (94); we used succinate to quantify forward capacity of SDH.

In photoreceptors, whorled deposits contact the matrices of multiple mitochondria. The deposits are distinct from melanosomes; they are present in albino zebrafish photoreceptors. We did not investigate their composition, but they resemble osmiophilic structures observed in and around mitochondria of other cells, proposed to have roles in lipid metabolism (figure 8 of ref. 95 and figure 4 of ref. 96), mitochondrial degradation (97, 98), and peroxisome synthesis (99). Further, ER stacks can encircle mitochondria, forming a lamellar structure to initiate mitophagy (77). Collectively this suggests that photoreceptor mitochondrial deposits may mediate lipid homeostasis and/or mitochondrial clearance.

Mitochondrial deposits are extruded from cones at night. The events we report are morphologically distinct from photoreceptor endocytosis (100, 101), exocytosis (102, 103), and from published examples of mitochondrial extrusion in other cells (104–108). Unlike these events, extrusion of cone mitochondrial deposits shows no evidence of fusion with the plasma membrane, and the released material remains associated with the cell surface and underlying mitochondria.

The extruded material can connect to the extracellular space via thin stalks that terminate in diffuse lamellar sacs. Neurons have been reported to release and tether mitochondrial material (47, 48) via thin stalks ∼200 nm wide and terminal multivesiculate sacs of ∼3-µm diameter. The extracellular structures attached to photoreceptors are smaller (∼60-nm stalks and ∼1-µm sacs), and sacs do not contain vesicles. The stalks are smaller than tunneling nanotubes (109), which can transport mitochondria (110–112). Photoreceptors can exchange proteins via an unknown mechanism (113), but we did not observe connections between cones. The extracellular structures could link cones to the IPM, a scaffold of carbohydrates and bound proteins (114, 115) where cones occupy a distinct metabolic microenvironment (116, 117). At night the zebrafish outer retina becomes disorganized by retinomotor movements, so the extracellular structures could physically anchor cones to the IPM.

In summary, we report a quantitative analysis of daily mitochondrial dynamics in cone photoreceptors. At night, when energy demands are likely highest, a mitogenesis event in retinas precedes increases in cone metabolic activity and the number of small mitochondria in single cones. During the day, mitochondrial number decreases, perhaps mediated by fusion and/or mitophagy. We also report a dense lamellar material that is shared between mitochondria and extrudes from the cell at night, sometimes forming extracellular networks. These daily changes to cone mitochondria can support energy production at night, regulated by light and the circadian clock. This study did not include phototransduction mutants or lengthy dark adaptation; differences we report between LD and DD groups could be attributed to effects of light masking and/or master circadian rhythms (118, 119). Elucidating the makeup of mitochondrial deposits, regulation of SDH, role of ER–mitochondrial contacts, and rates and locations of mitochondrial turnover in healthy cones will contribute to an understanding of how mitochondrial abnormalities in aging and disease affect vision.

Research was authorized by the University of Washington Institutional Animal Care and Use Committee. Wild-type, transgenic, and albino adult zebrafish were maintained on a 14-h/10-h light/dark cycle. Twenty-four hours prior to experiments, fish were fasted in either continuous darkness (DD) or standard room light (LD). See SI Appendix for details about fish lines and sample collection.

Wild-type or albino zebrafish eyes were fixed and prepared in resin blocks as described previously (18). Serial sections were cut at 50-nm thickness and imaged in the outer retina, 100 to 200 µm from the optic nerve. Z-stack images were processed and measured using the TrakEM2 plugin for ImageJ (research resource identifier [RRID]:SCR_008954). See SI Appendix for details about sample preparation and image analysis.

Albino or transgenic zebrafish eyes were fixed, cryosectioned, stained, and confocal imaged as described previously (41). Z-stacks from each timepoint were blindly analyzed using ImageJ (RRID:SCR_002285). SI Appendix, Table S1 lists antibodies and stains; see SI Appendix for details about sample preparation and image analysis.

RNA was extracted from wild-type zebrafish eyes, reverse transcribed into cDNA, and analyzed using qPCR as described in SI Appendix. Primer sequences are reported in SI Appendix, Table S3. Cosinor curves fitted to each dataset were used to determine circadian rhythmicity according to cutoff values (more than twofold change and P < 0.05 from ANOVA analysis) (120); data meeting this cutoff display dashed lines. SI Appendix, Table S4 lists fitting parameters and statistics.

Histochemical enzyme activity was assayed similarly to previous studies in human (58) and mouse retina (121). Briefly, albino zebrafish eyes were dissected, frozen in optimal cutting temperature compound and cryosectioned for parallel staining. SDH and COX activity stain solutions were applied for 10 min at 37 °C, then washed, mounted, and imaged with a light microscope. Stain intensity of single cones was blindly quantified using ImageJ. See SI Appendix for details about sample preparation and image analysis.

Light- or dark-adapted wild-type zebrafish eyes were dissected and incubated in 1 mM D-[U-¹³C]-succinate with 1 mM unlabeled glucose for the specified timepoints, then washed in phosphate buffered saline and flash frozen in liquid nitrogen. Metabolites were extracted from retinas, derivatized, and analyzed using GC/MS. See SI Appendix for details about sample collection and metabolite analysis.

Data were processed using Microsoft Excel (RRID:SCR_016137); statistical tests were performed using GraphPad Prism (RRID:SCR_002798). SI Appendix, Table S2 lists experimental numbers (Ns) and statistical analysis methods for each figure. The 3D renderings and animations from SBFSEM were created using Blender (RRID:SCR_008606) (122).