Experiments were performed on 16 healthy inbred Wistar rats of either sex, for a total of 25 temporal bones (Table 1). Animals were housed with 12 h/12 h light/dark cycle, food and water provided ad libitum; sacrifices were performed during the light phase. This study was carried out in accordance with the recommendations of Act 26/2014, Italian Ministry of Health. The protocol (number 155/2017-PR) was approved by the Italian Ministry of Health and University of Pavia Animal Welfare Office (OPBA). All efforts were made to minimize the number of animals used and animal suffering. Three samples (R3, R8b and R16a) were used for detailed 3D morphological reconstruction of the bony and membrane labyrinth and the relative segmentations were published as datasets (17). Throughout the paper, terminology of temporal bone structures follows that by (18).

Rats were anesthetized as in (13), and transcardially perfused with cold heparinated PBS through a peristaltic pump. After complete blood removal (assessed by mucosal whitening) rats were perfused with 4% PFA until optimal tissue hardening and decapitated. Skin and muscle were removed from the head, which was post-fixed for 24 h and subsequently trimmed removing most parietal and occipital bone. Fixed trimmed heads were cryoprotected in 30% sucrose for one week. After cryoprotection, heads were decalcified with buffered 5% EDTA (with shaking, daily changes) until softening of the squamous temporal bone (3–4 weeks at room temperature or 2–3 weeks at 37°C). For rat R8, microwave-aided decalcification with buffered 5% EDTA was performed (8 h for 3 days at 40°C, with change every hour, using Milestone MLS 1200 Mega high-performance microwave). In some cases, the brain was used for different experiments, and the temporal bone was dissected free, whereas in others the whole posterior part of the head (cranium and brain, sectioned at the midbrain level) was processed, allowing access to the dura and its sinuses in their entirety.

Tissue clearing was performed as in (19); samples were immunolabeled with rabbit anti- Collagen IV (ColIV; Abcam ab6586, 1:200) sheep anti-von Willebrand factor (vWf; Genetex GTX74137, 1:100), mouse anti-smooth muscle actin (SMA; Abcam amab7817, 1:200), and rabbit anti-Iba1 (WAKO 019–19,741, 1:200); secondary antibodies (Invitrogen Donkey anti-sheep, anti-rabbit and anti-mouse conjugated with Alexa 488, 555 or 647) or nanobodies (20) (Synaptic System nanobodies anti-mouse and anti-rabbit conjugated with ATTO Fluor 565 and 647 dyes) were all used at 1:200. Incubations with nanobodies required half the time than with regular antibodies. TO-PRO labeling (2 h, Sigma T3605, 1:500 or 1:1000) was used to stain cell nuclei at the end of secondary antibody incubation. Unfortunately, several microvascular markers did not label our cleared samples (PECAM-1: Santa Cruz mouse anti-mouse and predicted rat, sc-376764; RECA-1: abcam, mouse anti-rat, ab9774 and ab264524, PROX-1: Sigma, mouse anti-rat, p0089; LYVE-1: R&D systems, sheep anti-rat, af7939); therefore, lymphatic vessels and marrow sinusoids were not traced.

Cleared samples were imaged using a custom-made mesoscale selective plane-illumination microscope [mesoSPIM (21)] system at the Brain Research Institute, University of Zurich and at the Wyss Center for Bio and Neuroengineering in Geneva, Switzerland, which allowed imaging of sample in its entirety (travel range 44 × 44 × 100 mm), providing near-isotropic resolution 3D datasets. Higher resolution scans were made on a CLARITY-optimized light-sheet microscope [COLM (22)] at the Wyss Center for Bio and Neuroengineering in Geneva, Switzerland, as in (19).

The microCT sample was counterstained with iodine by Lugol (Merck 1.0926.1000) immersion until solution clearing, for soft tissue staining. MicroCT observations were performed on the local core facility SkyScan 1,276 CMOS (Bruker, Kontich, Belgium) using step and shoot acquisition with a 0.2-degree rotation step. Images were reconstructed using the software Bruker Nrecon 1.7.4 with ring artifact correction and beam hardening, and Gaussian smoothing to limit rogue voxel artifacts (23). The sample was observed with source settings of 85 kV/47 μA and 1 mm Al filter. Images were 3272 × 3092 × 2189 pixels, at a resolution of 6.4 μm.

Before any analysis, lightsheet image stacks were processed to reduce z-axis anisotropy and lightsheet stripe artifacts. For z-axis anisotropy, image stacks were deconvolved using the MATLAB script from (24), with a NA of 0.14 for MesoSPIM (21) and 0.28 for COLM (22) and the refractive index of DBE (1.562) and scaled to isotropy with FIJI (25). For stripe artifact minimization, we filtered datasets with the SPIM-Matlab-Destripe-Deshake macro from (26) where necessary.

After deconvolution and destriping, segmentation of inner ear labyrinth, blood vessels, bone marrow clusters and bone channels was performed with FIJI (25), ITK-SNAP (27) and 3D-Slicer (28) with the following pipeline:

CLAHE (FIJI macro, optional);

This pipeline allowed segmentation of a labyrinth in a few hours; for other structures, which had a more variable shape and less sharp demarcation, time was longer. Segmentation was performed by three independent researchers and a consensus was reached upon comparisons.

In all animals, we analyzed bone marrow clusters, dural sinuses, inner ear labyrinth, bone channels, and blood vessels. Cleared tissues display autofluorescence, which is often used as a guide for identifying structures in large 3D samples (29). In our samples, autofluorescence allowed the segmentation of marrow clusters and bony and membranous labyrinths. Autofluorescence evoked by 647 nm light also allowed to delineate bone matrix differences between endochondral and endosteal bone (see Results). Given the complexity of inner ear 3D structures, in where necessary we show 3D reconstructions of the autofluorescence signal of samples R1b and R3 to clarify where the structures of interest are located.

Bone marrow was observed as granular content within bone cavities in the auto-fluorescence or TO-PRO signal. To measure volumes, we segmented bone cavities rather than the associated cellular masses, since tissue shrinking due to the iDISCO+ technique [which is estimated at about 10% (16)] could lead to marrow collapse, especially in larger lobules.

Volumes were obtained as scaled voxel counts from ITK-SNAP or 3D-Slicer; for a visualization of their spatial congruence, labyrinths were registered to each other using 3D Slicer landmark-based registration. The pipeline was as follows:

Point clouds were obtained from R16b bone labyrinth segmentation files, and landmarks were added with SlicerMorph pseudoLMgenerator (30);

Spatial congruence of marrow clusters in different temporal bones was observed by superimposing each cluster 3D volume reconstruction to the volume sum of all clusters (with the Union operator by 3DSlicer). Ectotympanic bone marrow volume was only quantified in 4 samples, since in most samples the bulla was opened to ease fixation and immunolabeling and was therefore incomplete. Samples with incompletely visible marrow clusters (due to excessive trimming, lightsheet artifacts or suboptimal clearing in the temporal bone region of interest) were not used for quantitative observations.

Volumes were compared between clusters and gender. Significance of differences was tested with factorial ANOVA using R (33).

Blood vessels (including dural sinuses) were visualized by ColIV labeling [which labels all basal laminae (34)], vWf [which labels vascular endothelium (35) and megakaryocytes (36)], and SMA [which labels smooth muscle (37)]. Autofluorescence at 647 nm delineated bone channels. Bone and vascular channels were traced throughout the temporal bone from mesoSPIM and COLM scans with a resolution from 1 to 4 μm; channel diameters were measured by selecting 10 random optical sections perpendicular to the channels, binarizing them, and measuring diameters of the object with circularity >0.9.

Since most structures in lightsheet stacks were visualized by the combination of bright and dark details (e.g., bright vessel wall and dark vessel lumen) and displayed convoluted pathways, in several images we employed Maxmin stack visualization, which was performed as follows (see also Supplementary Figure S1).

The depth region containing the structure of interest was chosen in the image stack;

Temporal bone marrow distribution could be visualized in lightsheet scans of cleared samples and microCT scans of iodine-contrasted samples (Figure 1). In cleared samples, autofluorescence allowed the segmentation of bone marrow as well as bony and membranous labyrinths. Bone marrow was well delineated using the autofluorescence signal excited at 488 nm (Figure 1A), but the autofluorescence signal excited at 647 nm also displayed local intensity variations within bone matrix (Figure 1B). Given that endochondral and membranous bone differ in collagen I and osteocalcin expression patterns (38), and that both proteins have been observed to be autofluorescent in cleared bone (14), these intensity differences appear to correlate with endochondral bone distribution. As further support of this, around the cochlea the geometry of low-autofluorescence bone (Figures 1C,D) was similar to that of endochondral bone (39) and at high resolution osteon-like layered patterns could be seen in this region (Figure 1E), consistent with the presence of osseous globuli, which predominate in the rat endochondral bone (40). On the other hand, microCT allowed to visualize the distribution of bone marrow cavities but did not allow to observe local variations of bone matrix structure or cavity contents (Figure 1F).

In rat, differently from human, temporal bone ossification centers do not merge into a single bone in adulthood: in particular, the rat shows a petrosal bone containing the inner ear, an ectotympanic bone forming the middle ear bulla, and a squamosal bone (18). Rat petrosal bone marrow could be located outside the otic capsule (Figures 1A,B) or within its endochondral layer (Figures 1C,D). Since petrosal marrow (Figure 2A) was contained within several separate cavities rather than in a single cavity like in calvarial bones (see Supplementary Figure S2), we called its units “clusters.”

Within petrosal endochondral bone, marrow was found clustered in a region reaching from the cochlear base to the endolymphatic sac (VEM, vestibular endochondral marrow, Figures 2B,E) and at the cochlear apex (CAEM, cochlear apex endochondral marrow, Figures 2C,E). In our data, all samples displayed separate CAEM and VEM, but in 3 cases we observed a “trail” of small marrow lobules with vascular connections to CAEM following the inferior petrosal sinus pathway around the cochlea (white line in Figure 2E). Outside endochondral bone, a large marrow cluster surrounded semicircular canal arms, extending into the crista petrosa (PNEM, petrosal non-endochondral marrow, Figure 2D). Other marrow of potential interest to inner ear responses was found in the occipital bone, across the distal portion of the endolymphatic sac (Figure 2B), and in the ectotympanic bone in association with the middle ear bulla (Figure 2C). Both ectotympanic and occipital marrows were connected to the dural sinuses which surround the petrosal bone (Figure 2E).

Temporal bone marrow distribution from different animals (Figure 3) was consistent, displaying a cluster at the cochlear apex (CAEM, Figure 3A), one around semicircular canal arms, outside the otic capsule (PNEM, Figure 3B), and one around the base of the cochlea and the vestibule, extending around the crus commune (VEM, Figure 3C). Endochondral bone marrow clusters (Figure 3D), were significantly smaller than PNEM [CAEM = 0.06 ± 0.05 mm³ (n = 21), VEM = 0.09 ± 0.13 mm³ (n = 15), PNEM = 3.18 ± 1.26 mm³ (n = 11)] and displayed sex-linked volume differences (Figure 3D).

Temporal bone marrow could be visualized (Figure 4) within the autofluorescence signal evoked at 647 nm (Figure 4A) and 488 nm (Figure 4B) as bone cavities filled with granular substance; TO-PRO labeling (Figure 4C) showed that these cavities contain a very high density of cell nuclei. The 647 nm autofluorescence signal also displayed bone channels departing from marrow clusters (Figure 4A), which included small (mean diameter 20.9 ± 4.1 μm, n = 635) and large (mean diameter 35.0 ± 5.3 μm, n = 47) populations. Some of these channels contained ColIV+ blood vessels (Figure 4B) and SMA+ arterioles (Figure 4D). Bone marrow sinusoids were not visible due to lack of selective labeling for the rat (see Methods); larger marrow diploic veins (diameter > 50 μm) were instead visible as tubular voids in marrow autofluorescence (Figure 4E).

Petrosal bone marrow displayed “spheroid” lobules (Figures 4G,H), which could be interconnected to each other or isolated, and “reticular” lobules with smaller size and irregular shape, resembling a thickening of bone vessels at their crossing (Figures 4F,I). The latter marrow lobules could only be observed in endochondral bone.

In all clusters, marrow lobules displayed very high cell density and contained cells identified as megakaryocytes (36) due to their large size (see Figures 4G,H) and labeling with vWf (Figures 4I,J), and a heterogeneously-distributed population of rounded Iba1+ cells (Figures 4K,L). The presence of Iba1+ myeloid cells and especially megakaryocytes allowed us to define these spaces within bone cavities as bone marrow.

CAEM (Figure 5) was an isolated marrow cluster with few small spheroid lobules confined to the apicomedial side of the cochlea (Figure 5A; see also Figures 1, 2). CAEM volume was significantly larger in females than males (see Figure 3D). In the 647 nm autofluorescence signal, thin bone channels were visible connecting CAEM to the cochlear spiral ligament of the second turn; when followed in their entirety (Figure 5B), these channels showed looping pathways starting from the marrow, reaching the spiral ligament and coasting it, and then returning to the marrow. Similar loops were seen with ColIV labeling (Figure 5C), suggesting at least some of the loops are vascular. Vascular systems of the cochlea and surrounding bone always appeared separated (Figure 5D). However, bone channel systems were not (Figure 6). The inferior petrosal sinus was connected to CAEM through straight bone channels (Figure 6A) and blood vessels (Figure 6B) but was also connected to the spiral ligament border (Figure 6A). Vascular channels extending to the dural sinus could reach into the sinus lumen (therefore representing venules, Figure 6C) or connect to small vessels within the connective tissue surrounding the sinus (Figure 6D), which could represent arterioles or lymphatic collectors. In addition to the dural sinus, CAEM was unexpectedly also connected through several channels to the tensor tympani fossa of the middle ear (Figure 6E), in all samples where both structures were clearly visible (n = 16). Channels were measured in high resolution image stacks from 647 nm autofluorescence or TO-PRO, giving an average number of 6 ± 2 connections (n = 9) with an average diameter of 19.0 ± 19.9 μm, not significantly different (p = 0.1) from the diameter of small channels connecting bone marrow and dural sinuses. In one sample, it was possible to follow ColIV+ blood vessels in these channels. These connections could allow inflammatory mediators and cells to be exchanged between the cochlea and the tensor tympani muscle.

At the cochlear base (Figure 7) bone channels were more complex, showing a very interconnected network (Figure 7A). Whereas CAEM was associated to the cochlear apex only, a separate reticular or mixed reticular-spheroid bone marrow cluster reached the hook spiral ligament (Figure 7A) and scala tympani periosteum (Figure 7B). Vascular labeling with ColIV (Figure 7C) revealed an asymmetry in bone microvascular density at the cochlear base, with higher density around the crus commune up to the round window, and lower density toward the cochlear modiolus and otolithic maculae. Bone marrow was only seen in the regions of high vascular density, which in the rat correspond to endochondral bone (41).

All macrophage populations previously described in the literature (42) could be identified in the cochlea by Iba1 labeling (Figure 8A). In addition, Iba1+ cells were seen within bone marrow and bone channels connecting CAEM to the spiral ligament (Figures 8B,C) and in endochondral bone at the cochlear base (Figure 8D; see also Figure 1E). At higher magnification, Iba1+ cells in bone marrow and channels displayed different shapes (Figure 8E); marrow Iba1+ cells were rounded, whereas those associated with channels were elongated. Although both Iba1 and autofluorescence signals were excited at 647 nm, Iba1 labeling was brighter than autofluorescence, and displayed distribution patterns which were never observed with 647 nm autofluorescence only (compare Figure 8E with Figures 4A,G,H).

Petrosal non-endochondral bone marrow (PNEM) had a more complex geometry than endochondral bone marrow, following the semicircular canal arms and extending into the crista petrosa (Figure 9; see also Figures 1, 2). Bone marrow in the crista petrosa could be continuous with that surrounding semicircular canals (n = 6 samples) or form a separate island (n = 5 samples; see also Figure 3B). Inhibition of bone remodeling was clearly visible when observing perpendicular sections around simple canal arms (Figures 9A,B) where the latter were surrounded by a circle of dense bone with sparse thin vascular channels connecting them to the surrounding marrow. Bone marrow vessels could penetrate the bony labyrinth through these channels, and reach the canal periosteum, but the vascular network of membranous canals appeared separate (Figure 9C). PNEM was extensively connected to the superior petrosal dural sinus through channels similar to those found for CAEM in the inferior petrosal sinus (Figure 9D). In addition, most commonly around the crus commune confluence, bone marrow opened directly into the dural sinus, with a large junction where marrow cells appeared to be exposed to the sinus lumen (Figure 9E).

The distribution of Iba1+ cells in PNEM was very different from that of endochondral bone: although round Iba1+ cells were present within bone marrow, little or no elongated Iba1+ cells were seen following bone channels (Figure 9F). Semicircular canal periosteum, perilymph, and periosteal bone channels displayed Iba1+ cells, with the ampullae and lateral canal showing higher densities than anterior and posterior canal arms (Figures 9G,H). At the crus commune, endochondral bone was present, and therefore PNEM was replaced by VEM; in most samples, the two clusters were separated, but in two samples they were continuous, gradually changing properties.

Vestibular endochondral bone marrow (Figure 10) displayed a small volume, and its lobules were variable in position, size and shape (see Figures 2, 3): in 5 samples it was reticular (as in Figures 4F,I), in 6 samples it was spheroidal, and in 3 mixed (i.e., contained both small reticular and larger spheroidal lobules); in 2 samples, it was continuous with PNEM so the nature of its lobules was less clear. Reticular marrow was difficult to delineate unless cell distribution was observed with 647 nm autofluorescence, TO-PRO, or vWf; therefore, it is very likely that in microCT observations the contribution of reticular marrow is underestimated. High resolution vWf and TO-PRO labeling (Figure 10A) allowed us to reconstruct the endolymphatic duct and sac together with the vestibular aqueduct vein and sigmoid sinus (Figure 10B), obtaining a similar connectivity as described for brain-associated bone marrow (43), in spite of the small size of marrow lobules. Bone marrow channels around the endolymphatic duct and sac were extremely ramified (Figure 10B), mostly vascular (Figure 10C), and macrophage-rich (Figure 10D), forming an interconnected network reaching the endolymphatic sac, bone surface, crus commune periosteum and sigmoid sinus (Figure 10E). Although not part of the petrosal bone, occipital bone marrow was connected to the sigmoid sinus through diploic veins and thin channels (Figure 10F).

Although the rat middle ear is surrounded by an ectotympanic bone, which is separated from the petrosal bone (18), it appears important to compare middle ear- and inner ear-associated bone marrow features and distribution, since the middle ear represents the principal site of entry of pathogens and drugs for the inner ear (44, 45). Ectotympanic bone marrow (Figure 11) was found in two large islands (total volume 1.54 ± 0.79 mm³, n = 4) surrounding the tubal and tympanic opening (see Figure 2C). The bulla was vascularized by parallel vessels reaching the mucosa (Figure 11A), which passed through bone marrow clusters (Figure 11B). Both large and small bone channels were visible connecting ectotympanic bone marrow to the middle ear mucosa with microCT (Figure 11C). In lightsheet scans, large channels were visible in 488 autofluorescence, but unfortunately, for the samples where the whole bulla was available, 647 signal was unavailable. On the other hand, the ectotympanic marrow closer to the petrosal bone was available in most samples, and thin channels were visible toward the suture between ectotympanic and petrosal bone (Figure 11E), in a position right opposite to CAEM.