Anatomical basis of drug delivery to the inner ear

Abstract The isolated anatomical position and blood‐labyrinth barrier hampers systemic drug delivery to the mammalian inner ear. Intratympanic placement of drugs and permeation via the round‐ and oval window are established methods for local pharmaceutical treatment. Mechanisms of drug uptake and pathways for distribution within the inner ear are hard to predict. The complex microanatomy with fluid‐filled spaces separated by tight‐ and leaky barriers compose various compartments that connect via active and passive transport mechanisms. Here we provide a review on the inner ear architecture at light‐ and electron microscopy level, relevant for drug delivery. Focus is laid on the human inner ear architecture. Some new data add information on the human inner ear fluid spaces generated with high resolution microcomputed tomography at 15 &mgr;m resolution. Perilymphatic spaces are connected with the central modiolus by active transport mechanisms of mesothelial cells that provide access to spiral ganglion neurons. Reports on leaky barriers between scala tympani and the so‐called cortilymph compartment likely open the best path for hair cell targeting. The complex barrier system of tight junction proteins such as occludins, claudins and tricellulin isolates the endolymphatic space for most drugs. Comparison of relevant differences of barriers, target cells and cell types involved in drug spread between main animal models and humans shall provide some translational aspects for inner ear drug applications. HighlightsInner ear fluid compartments are separated by barriers with varying leakiness.Oval and round window are permeable for several chemical compounds.Endolymph compartment is tightly sealed and vulnerable to sensory epithelium damage.Hair cells containing Cortilymph is permeable to perilymph via the basilar membrane.Active transport mechanisms by mesothelial cells provide a pathway to neurons.


Why inner ear microanatomy favors local drug delivery
Various compounds were identified to rescue neurons and hair cells (HCs) from ototoxic insults and mechanical trauma in animal experiments (Mukherjea et al., 2011;Nguyen et al., 2017). Since endogenous regeneration of mammalian inner ear HCs is very low and limited to the vestibular system, gene therapies were developed to transdifferentiate supporting cells into HCs or utilize stem cells to replace lost receptors (Atkinson et al., 2014;Mittal et al., 2017). Despite this knowledge derived from in vitro and in vivo animal experiments we still lack an effective pharmacological treatment of the human inner ear to combat hearing loss, tinnitus, Meniere's disease and vestibular deteriorations. Delivery of genes/ vectors or chemical compounds to target cells in the inner ear is still a challenge. Ideal chemical formulation and combination of suitable compounds with an appropriate mode of drug delivery are the current tasks. Knowledge about microanatomy and functional characteristics of cell types in the inner ear is indispensable for a successful pharmacotherapy.
The anatomical location of the human inner ear within the hardest bone isolates the vestibulo-cochlear sensory organ. Additionally low blood flow rates and a limited passage through a blood-labyrinth barrier similar to the blood-brain barrier interfere with systemic application of drugs, so a local administration appears attractive. Local absorption of drugs from the middle ear was first described by Schuknecht using intratympanic streptomycin for treatment of Meniere's disease (Schuknecht, 1956) and proved suitable also for several other compounds (Berjis et al., 2016;El Kechai et al., 2015;Staecker and Rodgers, 2013;Staecker et al., 2017). Minimally invasive approaches to bypass the tympanic membrane with e.g. cannulas to place drugs close to the inner ear proved safe in routine clinics and for human trials with growth factors (Nakagawa et al., 2014). From the middle ear pharmacologically active compounds need to overcome several barriers to reach HCs, spiral ganglion neurons (SGNs) or other target cells. Fluid filled spaces separated by thinnest membranes, tight cellular barriers and very compact bone results in a compartmentalization of the mammalian inner ear that requires several strategies for drugs to penetrate. Barriers between compartments vary in the degree of permeability which complicates assessment of pharmacokinetic spread and elimination. A profound knowledge of the microanatomy, fluid flow and characteristics of cell types in the inner ear is important to predict routes for small molecules and bigger proteins or genes.
Here we provide a review on the inner ear architecture relevant for drug delivery with some additional new data on the human inner ear. Comparison of relevant differences between main animal models and humans shall unravel variations in the mammalian inner ear architecture.

Basic anatomy
A rigid outer wall forms the bony labyrinth that comprises the cochlea, vestibule and semicircular canals. The otic capsule develops early as mesenchymal tissue around the otocyst and is in human adults embedded in the petrous portion of the temporal bone. It is one of the densest bones in the body and retains a considerable fraction of its primordial cartilage as islands of chondral tissue that directly calcifies (Fig. 1a). This results in a meager vascularized bone with low turnover impeding callus formation after fractures. The degree of encapsulation of the otic capsule is very different in mammalian species with a thick encasement in human (Fig. 1a) and cat and only a thin bony jacket in rodents like mice or guinea pigs (Fig. 1b). Especially the bony layer in the apical portion of the guinea pig is very thin and allows penetration of substances into the apex of the cochlea (Mikulec et al., 2009). The perilymphatic fluid space dominates the inner ear, fills the major part of the bony labyrinth in the vestibular system that passes into the scala tympani (ST) of the cochlea. The apical helicotrema interconnects ST with the scala vestibuli (SV) thereby forming the biggest connected fluid space across the inner ear. Within the bony labyrinth a much smaller membranous labyrinth interpenetrates this canal system and joins the apical poles of the inner ear sensory epithelia to a common compartment. Stereocilia and kinocilia from the vestibular HCs are bathed in this endolymph fluid. Vestibular and cochlear endolymph compartments connect via the small reunion duct. Both fluid systems serve as relay media to transmit vibrations, positional changes and motion acceleration but seem also to be important for the transport of oxygen, nutrients and waste to compensate for low blood supply of e.g. the cochlear sensory epithelium. The endolymphatic scala media (SM) of the cochlea adds an electrochemical gradient boosting sensitivity of mechanotransduction to movements in the range of atomic size.

Target cells -are they equal across mammalian species?
Mammalian animal models may differ in several aspects from primate inner ears. Common ancestors from rodents and primate species separated approximate 75 million years ago (Mouse Genome Sequencing et al., 2002)-enough time to evolve changes in orthologue genes or gene activities that may influence efficacy of drugs. Differences in shape and length of the cochlea and semicircular canals reflect adaptation for frequency spectrum, resolution and sensitivity of motion detection necessary for proper mode of locomotion. Few studies imply functional aspects on a comparative basis of inner ears across mammalian species including man (Makimoto et al., 1980;Nadol, 1988). More knowledge will be necessary to understand why a pharmacological trial fails in human when animal experiments show positive results. Even inbred mouse strains used in otology research show marked differences in hearing performance and susceptibility to damage (Ohlemiller Fig. 1. Gross anatomy of the human and rodent cochlea and the organ of Corti: aeb: Section through a human (1a) and guinea pig (1b) inner ear. The cochlear nerve (CN) runs through the inner acoustic meatus (IAM) into the central modiolus (MO). ST; scala tympani, SV scala vestibuli. c: Human organ of Corti in the apical turn (approx. 200 Hz region). On a thin and wide basilar membrane (BM) plenty of different cell types form a complicated arrangement of sensory cells (IHC; inner hair cell, OHC; outer hair cell) and supporting cells (DC; Deiters cells, IP: inner pillar, HC; Hensen cells, OP; outer pillar). A thick tympanic covering layer (TCL) populates the scala tympani (ST) surface of the BM. Tunnel of Corti (TC) and Nuel's space (NU) represents another fluid space termed cortilymph. SM scala media, TM; tectorial membrane. Zheng et al., 1999). However, certain cell types in the inner ear may have specialized more than others during the last 75 million years of evolution as adaptation to specialization and communication skills. In the following section we compare the most important target cells for a pharmaceutical intervention between human and common animal models in inner ear research.

Hair cells
Protection and regeneration of our inner ear receptor cells, the HCs, are one of the main tasks for drug-and gene therapies, since most HCs in the mammalian inner ear do not recover. Only limited regeneration in vestibular end organs was found (Li et al., 2016;Warchol et al., 1993). Apart from a more irregular order of cochlear HC rows in humans compared to other mammalian species (Glueckert et al., 2005b), there do not seem to be relevant differences to rodents or cats. Specialization into a primary inner hair cell (IHC) and motile outer hair cells (OHCs) is the same and typical tonotopical gradients regarding microanatomy of the sensory epithelium are present in all higher mammals. Stereocilia of the cochlear mechanotransducers and cell body of OHCs are longer in the apex (Fig. 1c) and shorter in the base. In high sensitivity frequency regions, IHCs show more afferent contacts. Gradients in susceptibility for ototoxic substances and degeneration pattern following acoustic trauma are common across different mammalian species (Lee et al., 2013;Ohlemiller and Gagnon, 2004;Sha et al., 2001).
HC receptor cells seem to be evolutionarily quite conserved in several aspects, so that even toxicity studies from fish lateral line neuromast HCs may predict vulnerability in inner ear HCs (Esterberg et al., 2013;Ou et al., 2012). Marked differences imply their regeneration capacity that distinguish the primal features of fish lateral line HCs. Such data underline the significance of model systems chosen to screen for compounds that may protect our valuable sensory cells. This does not replace drug efficacy studies in mammalians including primates.

Spiral ganglion neurons
The majority of bipolar neurons belong to type I SGNs. They transform IHC receptor potentials into action potentials that travel to the ascending neuronal pathway in the central nervous system (CNS). The 3e5% smaller type II neurons provide afferent innervation of OHCs. With ageing, humans loose about 100 SGNs per year (Makary et al., 2011). Decline of innervation may be independent from HC loss (Felder and Schrott-Fischer, 1995;Viana et al., 2015) and affect dynamic range and speech understanding before a profound hearing loss develops. Animal experiments that applied ototoxic drugs to wipe out HCs often resulted in a nearly total loss of neurons following deafness. Toxicity to neurons may be underestimated in humans due to delayed and slow effects. Humans retain many of their neurons decades after complete deafness (Glueckert et al., 2005a). New transgene animal models selectively ablate HCs without ototoxic drugs changing our picture of a retrograde loss of SGNs following HC loss (Kurioka et al., 2016).
Like sympathetic and sensory ganglia in our body, SGNs are completely enveloped by glia cells (Fig. 2). Unlike CNS neurons there is no direct contact of the neuron with extracellular fluid or matrix. Schwann cells wrap around peripheral and central axons and satellite glia cells (SGCs) sheath the soma (Fig. 2aed). SGCs have been found to play a variety of roles, including control over the microenvironment by regulating the diffusion of molecules across the cell membrane (Hanani, 2010a(Hanani, , 2010b. Because of this uninterrupted sheath seen in ganglia, a similar role as the bloodebrain barrier for larger molecules was suggested (Ten Tusscher et al., 1989). "Naked" portions at nodes of Ranvier are at least covered by nodal microvilli emanating from myelinating Schwann cells and its basement membrane (Fig. 2e). A striking difference of SGNs in human and most other mammals is the lack of myelination of the neuron somata (Fig. 2aeb). Perisomatic myelination of the SGCs may have developed for fastest neurotransmission and timing relevant for most precise sound localization. Humans presumable lost this myelination as an adaptation for speech taking into account a possible delay in action potential propagation (Rattay et al., 2013). Common SGCs form clusters of SGNs in humans (Fig. 2aeb) that can be found to a lesser extent in some apical SGNs in mice. Together with the presence of a gap junction system that couple SGCs these data suggest that SGCs may play a role in signal processing of functional SGN units (Glueckert et al., 2005a;Liu et al., 2015b). Perisomatic myelination may act as an additional barrier ( Fig. 2ced) not present in human, since transport across several membrane layers may take longer than crossing only two membranes to reach the neuron. Differences in permeability or pharmacokinetics for drugs are not easy to evaluate. There is a mouse model where most SGCs are not myelinated that could be compared with its C57BL/6 background strain to address this issue (Jyothi et al., 2010). Most type II neurons are completely unmyelinated but yet not described to be a primary target for drug delivery.

Other target cells
Other cell types to target are more in the spotlight for gene delivery strategies. Exogenous expression of the transcription factor Atoh1 in sensory epithelium supporting cells (Fig. 1c) is sufficient to induce the trans-differentiation into a HC phenotype and may recover function in the cochlea and vestibule (Baker et al., 2009;Izumikawa et al., 2005). Overexpression of nerve growth factors may promote neural survival or regrowth of peripheral processes of SGN (Fukui et al., 2012;Kawamoto et al., 2003). Mesothelial cells lining perilymphatic spaces are often the main "target" to accomplish endogenous production of nerve growth factors through viral infection or electroporation (Wang et al., 2012). A gene delivery system was developed to rescue hearing in a mouse model of Connexin 26 deletion (Iizuka et al., 2015); here cochlear supporting cells as well as lateral wall fibrocytes serve as primary targets. There are no reports to our knowledge about interspecies differences in these cell types regarding pharmacological interventions.

Blood supply and blood-labyrinth-barrier
The inner ear blood supply is quite complex and lacks collaterals that make it vulnerable to ischemic effects. A single labyrinthine artery emanating from the anterior inferior cerebellar artery is the only blood vessel into the inner ear. The main cochlear artery spirals up within the modiolus and sends out radial arterioles through the SV wall to form a capillary network at the stria vascularis (Fig. 3a). This three cell layered epithelium presents highest density of capillaries to ensure the electro-chemical gradient between perilymph and endolymph in the cochlea. Venules below the spiral prominence ( Fig. 3a) gather into bigger veins that travel within the spiral ligament of the ST into the modiolus. These vessels coalescence to the inferior cochlear vein and courses through a small bony channel parallel to the cochlear aqueduct.
Maintenance of inner ear fluid homeostasis may have favored the formation of a barrier function between the vascular system and the inner ear fluid. Similar to the blood-brain-barrier, endothelial cells lack fenestrations and seal their lumen with vast tightand adherens junctions (Fig. 3eed). Pericytes as well as perivascular resident macrophage-like melanocytes are in intimate contact with endothelial cells and add multiple basement membrane layers (Fig. 1e) to tightly regulate exchanges from the blood to interstitial fluid. This is especially pronounced within the stria vascularis (Zhang et al., 2012) and reviewed recently (Shi, 2016). Strial basement membranes were found to be negatively charged (Suzuki and Kaga, 1996) that may establish a charge selective barrier similar to the glomerular basement membrane in the kidney that restricts the transmembrane flux of anionic proteins. Especially highly positive charged substances might obstruct the strial ultrafilter of blood-labyrinth-barrier (BLB) capillaries.
More and more data deliver direct evidence on impaired function of the BLB in Meni ere disease. Transport of chelated gadolinium (Gd) is compromised in Meniere's patients likely due to endolymphatic hydrops (Shi et al., 2014). Meni ere disease was recently associated with a deteriorated BLB evaluating MRI postcontrast measurements of signal intensities (Pakdaman et al., 2016). Concurrently specific ultrastructural changes of capillaries constituting the BLB were identified in Meni ere patients' utricles taken out during surgery (Ishiyama et al., 2017). Hence, systemic pharmacotherapy should not aim to disrupt this tightly regulated homeostasis function, also because its regulation is largely unknown and severe inner ear disorders like Meni ere disease are associated with.  , 2006). By its ionic composition perilymph reveals as a typical extracellular fluid, the source is still under debate. It may be produced as a blood ultrafiltrate, originated from the cerebrospinal fluid (CSF) and ced: Mouse spiral ganglion neurons are much smaller than human and fully myelinated, despite completely unmyelinated type 2 (II) neurons innervating outer hair cells. Satellite glia cells (SGC) form a myelin layer that is thinner than the myelin sheet formed by Schwann cells (SC) isolating the axons. There is no unmyelinated AIS present in these neurons (2d). e: Human Ranvier node (RN) lacks myelin and is often surrounded by Schwann cell processes (SCP) forming a micro environment. transported via the cochlear aqueduct (CA) (Fig. 4a) or is a mixture of both. Although interconnected, composition of this fluid varies between ST and SV (Juhn et al., 2001;Wangemann, 2006) due to active homeostatic mechanisms that vary along the tonotopical axis.
The very dense endosteal bony layer of the otic capsule encompasses this fluid space interrupted by the round-and oval window as well as the cochlear-and vestibular aqueduct (Fig. 4a). The CA is a small bony canal that originates at the floor of the ST close to the round window (RW) and establishes a communication between the perilymphatic and the subarachnoid space of the posterior cranial cavity. CA is approximately 1 cm in length (range from 2.4 to 14.6 mm) the narrowest diameter (isthmus) varies between 0.06e0.3 mm (Bachor et al., 1997;Guo et al., 2016). In guinea pig the CA is only 2 mm long (Shinomori et al., 2001) but not narrower than in humans (Ghiz et al., 2001). Flow may not be restricted to one direction. Communication between perilymph and CSF was described (Kaupp and Giebel, 1980) but may have been favored by large volumes introduced into perilymph of such studies (Ghiz et al., 2001). A main function of the CA may also be a pressure release route from the cochlea (Carlborg et al., 1982). Complete bony obstruction of the CA in human was reported (Rask-Andersen et al., 1977).
A parallel accessory canal with a diameter of 0.13e0.27 mm in human (Guo et al., 2016) guides the inferior cochlear vein to drain blood from the cochlea to join the internal jugular vein (Fig. 4a).
Marker studies suggest that inner ear fluids are relatively unstirred and solely dependent on diffusion rates (Salt et al., 2015). Even in prolonged local application of drugs, concentration gradients remain in the fluids (Salt and Ma, 2001). Another source for CSF fluid may occur via the inner ear canal (inner acoustic meatus) and the vestibulocochlear nerve. This nerve is surrounded by a perineurium that progresses into the dura of the brain. At its distal ending the VIII th nerve penetrates the bone in the fundus region of the human inner ear via fenestrations of various sizes that guide central axons (Fig. 5aeb). In other mammals such as guinea pigs this opening is more prominent (Fig. 1b) without bony restrictions towards the modiolus. CSF fluid may pass into the modiolus and volume fluctuations may open the reverse path. Big pressure changes into the inner ear must be avoided otherwise it may be very likely that drugs introduced into the inner ear may show up in the brain and even to the contralateral ear (Stover et al., 2000).
Dimensions of fluid spaces vary considerably across species (Thorne et al., 1999) and also within a single species as so in human (Avci et al., 2014;Erixon et al., 2009). We evaluated the fluid spaces in 24 human inner ears acquired with a microcomputed tomography (microCT) at 15 mm voxel resolution and manual segmentation.
ST length ranges from 28 to 40 mm in humans (Wright et al., 1987) and are around 17 mm in guinea pigs and 4.3 mm in mice (Nadol, 1988). As fluid spaces of the human cochlea are also considerably longer than those of animal models a single site applications of a drug may act only at a limited region. Insertion of catheters to provide a more even spread implies the high risk of damage while elution of drugs from a cochlear implant may add few additional risk of trauma. Enhanced substance spread could also be performed with superparamagnetic nanocarries in an external magnetic field before they release their payload (Barnes et al., 2007;Ramaswamy et al., 2015).
SGNs reside in a bony canal of the modiolus termed Rosenthal's canal (named after the neuroanatomist Friedrich-Christian Rosenthal 1780e1846). Bony columns guide the peripheral axons of the bipolar neurons as fascicles towards the osseous spiral lamina (OSL) where nerve fibers fan out to innervate the sensory epithelium (Figs. 4c, 5a-b). The ST portion of the bony modiolar wall is surprisingly porous with plenty of fenestrations forming a trabecular meshwork with broad communication canals into the central modiolus (Fig. 5a). Likewise, the delicate osseous spiral lamina reveals also plenty of bony fenestrations. SV face presents a different view with a smooth bony surface poor in fenestrations. Prominent holes guide the spiral modiolar arterioles to the apical portion of SV wall and further to stria vascularis. Trabecular bone close to the OSL provides bony opening into the modiolus, but to a lesser extent than in the ST (Fig. 5a).
There are suggestions for direct communication routes from perilymph to the modiolus in humans (Kucuk et al., 1991;Rask-Andersen et al., 2006;Shepherd and Colreavy, 2004). In animals these may also exist as reported also in cat (Shepherd and Colreavy, 2004). Results from the distribution of nanocarriers such as liposomes and polymerosome particles suggest even intramodiolar routes from ST basal turn into the modiolus towards ST& SV in more apical turns (Buckiova et al., 2012). A radial communication between ST and SV via the spiral ligament has also been demonstrated in animals and humans (Salt et al., 1991a(Salt et al., , 1991bZou et al., 2005) suggesting cells facing perilymphatic spaces allow fluid flow into the central modiolus across all turns.

Mesothelial cell layer-more than a pavement!
Perilymphatic spaces are delineated by mesothelial cells (MCs) (Fig. 5cei) that derive from mesodermal tissue (Mutsaers, 2004). They commonly form a very thin monolayer and rest on a basement membrane supported by a complicated collagen meshwork (Rask-Andersen et al., 2006) (Fig. 5d and e). Although perilymphatic fluid is the primary media for local pharmacotherapies and MCs are the first cellular barrier, only little is known about this cell type in the inner ear. MCs line most internal organs and cavities such as the peritoneal, pleural and pericardial. It is the first line of defense against invading microorganisms by initiating inflammatory and immune responses and provide a slippery non adhesive surface (Mutsaers et al., 2016). They form tight junctions to avoid paracellular flow (Fig. 5d) or appear more loosely arranged between Fig. 4. Fluid spaces in the inner ear: aeb: 3D reconstructions of segmented datasets based on micro computed tomography from human (4a) and guinea pig (4b) inner ears. Scala vestibuli (SV) and perilymph compartment in the vestibular systems forms the biggest fluid space (green) that is interconnected to the scala tympani (ST) in the cochlear apex. In human we segmented also the cochlear aqueduct (CA) and the parallel running accessory canal (AC). Endolymphatic duct and sac (ES) is accompanied by a blood vessel in a parallel bony canal (colored orange). Endolymphatic compartments (colored red) are much smaller. The inner acoustic meatus (IAM) houses several big nerves and proceeds into the facial nerve canal (FNC). Oval window (OW) and round window (RW) represent most interesting gates for local pharmacotherapies. SM; scala media, CN; cochlear nerve, white color; vestibular nerves. c: Volume rendering of another human inner ear depicting nerve fiber pathways. Neurons reside in Rosenthal's canal (RC) and radiate through the osseous spiral lamina (OSL). utr.; utricle and innervating nerve fibers.
bony columns (Fig. 5e) and underneath the basilar membrane (BM) where they may form thick layers called tympanic covering layer (TCL) (Fig. 5i). The TCL layer is more prominent apically where they form a thick meshwork of cells (Fig. 1c) while in higher frequency regions this layer becomes thinner with flat MCs (Cabezudo, 1978;Liu et al., 2015a) (Fig. 7c). Acellular debris can frequently be found between TCL cells (Fig. 5i). MCs secrete phosphatidylcholine, the major constituent of lamella bodies and pulmonary surfactant (Mutsaers, 2002). For the inner ear they may provide a frictionless, lubricant surface for smooth fluid movements. They are able to transport fluid and cells thereby "cleaning" their luminal compartment through pinocytotic vesicles, intracellular cavities or stomata. Stomata are cavities at the junctions of MCs, 3e12 mm in diameter that allow for rapid removal of fluid and cells (Ohtani et al., 2001) and provide a direct access to underlying space in the modiolus (Fig. 5c). In pathological or severe damage conditions MCs develop adhesion, may convert into fibrocytes and even trigger fibrosis (Jia et al., 2016). MCs have the ability to convert their phenotype comparable to changes seen in the epithelialmesenchymal transition (EMT), hence they express vimentin and desmin together with cytokeratins characteristic for epithelial cells. They are important for wound healing as MC proliferation was reported as reaction of ruptures of the round window membrane (RWM) (Sone, 1998). Fibrosis following cochlear implant insertion trauma is likely also be related to this EMT and may influence transcellular transport mechanisms of MCs.
Especially the region between the bony columns in the basal turn seems to provide a direct access into the modiolus (Fig. 5a, b,e and f). Various melanocytes populate the perilymphatic surface area. These dendritic cells with its typical brown endogenous pigments play a key role in innate immune responses. Little is known about the physiology of these melanocytes. In the stria vascularis the pigmented cells share characteristics of both macrophage and melanocyte phenotypes and control the integrity of the intrastrial fluideblood barrier by affecting the expression of tight-and adherens-junction proteins (Tapia et al., 2014). Their location directly underneath the MC layer (Fig. 5feh) suggests similar functions or a combination of immune response and tight junctional permeability control.
TCL cells show an incomplete junctional seal underneath the basilar membrane. Thicker layers in the apical turns (Liu et al., 2015a) present like a labyrinth in the labyrinth (Figs. 1c and 5i). A more or less dense meshwork of spindle shaped cells appears to be highly interwoven into each other. Tight junctions alternate with free passage canals to the BM. Different textures of the TCL in different turns suggested the function as a damping layer (Angelborg and Engstrom, 1974) to support mechanical movements of the BM. Their high phagocytic activity is more relevant for drug studies. We often found dense agglomerations of nanocarriers tagged with fluorochromes in the TCL (Glueckert et al., 2015). In human material acellular substances often fill the intercellular space (Fig. 5i). Like with other MCs a regulatory function on the border between compartments may be an explanation. These cells form a micro environment underneath the BM and may regulate transport between fluid spaces of the sensory epithelium and the ST. Other functions may involve repair or maintenance of the BM.
Taken together the ability of fast transport mechanisms of the MCs cell layer and a highly fenestrated bone especially at the ST suggests that the passage of perilymph may occur through the modiolus into the inner acoustic meatus. Mammalian mesothelium is considered essentially similar regardless of species or anatomical site (Mutsaers, 2004), so tissue from big body cavities could serve as models to analyze drug uptake/permeation mechanisms of MCs. Their enormous flexibility in phenotypical appearance from epithelial to mesenchymal characteristics and phagocytic activity may attribute MCs as main candidates for drug clearance and metabolic degradation of chemical compounds. Lack of blood vessels in the scalae and the BLB argues for a vital role of this cell type in clearance of drugs from perilymph. Elimination of drugs that are locally applied to perilymph is one of the major factors influencing both the local concentration achieved and spread towards the apex. A better understanding of MC physiology may be valuable to prolong drug activity and enhance spread within the inner ear.

Endolymphatic compartment-splendid isolation
Endolymph composition is unique in mammals with concentrations high in potassium and low in calcium, similar to intracellular fluid. The endolymphatic compartment is much smaller than the surrounding perilymphatic space and comprises only 17.3% of total labyrinth fluid in human, the SM is only 3.3% of the lymphatic volume in human according to our own data (Table 1). Via the vestibular aqueduct this compartment is connected to the endolymphatic sac (Fig. 4a). Functions imply regulation of the volume and pressure of endolymph, immune response of the inner ear, and the elimination of endolymphatic waste products by phagocytosis (Couloigner et al., 2004;Rask-Andersen and Stahle, 1980;Rask-Andersen et al., 1991;Salt, 2001).
Most compounds soluble in perilymphatic fluid should behave the same in the endolymph. Endolymph shows even lower protein concentrations (Wangemann et al., 1995) that may enhance effectiveness of a drug. Exchange of substances between perilymph and endolymph is strictly limited to transcellular routes (Juhn et al., 2001). However, the transport of horseradish peroxidase from periplymph to endolymph was reported previously (Saijo and Kimura, 1984). Uninterrupted strands of tight junctions seal uncontrolled paracellular transport. Loss of the EP driving force results in deafness. Various claudin proteins form a complicated expression pattern with variable barrier functions across different areas. Many cell borders show co-expression of different claudin proteins whereas in the basal cells and adjacent fibrocytes of the stria vascularis only claudin 11 is present to insulate the cochlear battery ( Fig. 6c) (Kitajiri et al., 2004b;Liu et al., 2017). Deletion of claudin 11 leads to a loss of the EP and deafness but not to a decreased K þ concentration in the endolymph (Gow et al., 2004;Kitajiri et al., 2004a). The EP seems to be generated in the basal cell area whereas maintenance of high K þ levels are attributed to marginal cells of the stria vascularis (Wangemann et al., 1995).
Although very thin, the 2-cell layered Reissner's membrane (RM) provides a potent barrier for ions and chemical compounds (Fig. 6aeb). Selective transport for sodium was found that contributes to maintaining the low Na þ concentration in endolymph fluid (Yamazaki et al., 2011a) similar to semicircular canal duct epithelial cells (Yamazaki et al., 2011b). Trans-epithelial exchange of Na þ , K þ ,Cl À and Ca þþ via the RM was proposed (Lang et al., 2007) and water transport mediated by aquaporin channels calculated (Eckhard et al., 2014). However for bigger molecules RM and epithelium of the vestibular portion of the endolymphatic compartment seems to be rather impermeable. A water shunt region in the cochlear apex was proposed that may explain the experimentally determined phenomenon of endolymphatic longitudinal flow towards the cochlear apex (Eckhard et al., 2014;Hirt et al., 2010).

Reticular lamina & cortilymph: shear stress proof tight junctions and the third lymph
Tricellulins are important especially in the reticular lamina (Fig. 7a) that seals the SM against the organ of Corti. Vibrations and shear stress sets special requirements for these mosaic pattern   6. Tight junctional seals in the human cochlea: a: The endolymphatic compartment is sealed by Reissner's membrane (TEM image), a delicate two cell sheet layer with cochlear duct epithelium facing the scala media (SM) and a mesothelial (MC) layer facing the scala vestibuli (SV). b: Immunostaining for the tight junction protein occludin depicts the uninterrupted seal by both cell types in Reissners membrane. scala media; SM, scala vestibuli; SV. c: Claudin 11 and Na þ -K þ -ATPase immunostaining identifies this important tight junction protein in strial basal cells and adjacent layers of type I fibrocytes of the spiral ligament (SL). Marginal strial cells are highly positive for Na þ -K þ -ATPase to maintain of high K þ levels in the scala media (SM). BV; blood vessel. Table 1 Temporal Bone (TB) dimensions of fluid spaces from 24 human individuals evaluated by manual segmentation of 15 mm voxel resolution microCT data sets. R and L represent left and right inner ears. vestib. EL, vestib. PL refers to endolymphatic and perilymphatic spaces in the vestibular system, Fluid spaces in the scala media endolymph (S. media EL), scala tympani perilymph (S. tymp. PL), scala vestibuli perilymph (S. vest. PL) and total perilymphatic (PL total) and endolymphatic fluid (EL total). Fluid total; total inner ear fluids.  shaped cell borders. Oscillation induced forces set highest demands on junctions between Deiters cells and motile OHCs (Fig. 7b). This cell-cell contact is unique in a way that the OHCs-Deiters cells connection form a novel hybrid tight junction with adherens junction organization (Nunes et al., 2006). Loss of tricellulin leads to HC degeneration and non-syndromic deafness (DFNB49) in human but does not impair EP or endolymphatic ion composition (Kamitani et al., 2015;Riazuddin et al., 2006). Nuel's space surrounds the lateral surfaces of OHCs and Deiters' phalangeal processes. Together with the fluid filling the tunnel of Corti this compartment is termed cortilymph, the third fluid space (Fig. 7a, c). Composition is similar to perilymph regarding K þ concentration at rest but changes with sound evoked stimulation as extracellular K þ accumulates by HC activation (Johnstone et al., 1989). K þ is then recycled back into the stria vascularis by a connexin gap junction network.
A leaky reticular lamina would intermingle endolymph with fluid spaces of the sensory organ and in turn lead to a loss of HCs. Hence, any damage of the sensory epithelium surface, especially in combination with acoustic stimulation must be avoided for any safe drug application.

Basilar membrane-leaky to cortilymph?
A communication between ST and cortilymph was suggested to occur at the base of the Hensen-Deiters' junction (Beagley, 1965). TEM imaging and tracer studies suggest this slit as a patent pathway between ST and the organ of Corti even able to generate hydrops of the sensory epithelium (Nomura et al., 2016). We recently analyzed the human BM and confirm thinnest portions situated underneath the OHCs/Deiters cells (Liu et al., 2015a). Further a discontinuity in the basement membrane between the Hensen cells was found in the apical region in cat (Cabezudo, 1978). These gaps provide structural evidence of the free communication between the intercellular spaces in the organ of Corti and the ST through the BM, as previously reported by several authors (Altmann and Waltner, 1950;Duvall, 1972;Ilberg and Vosteen, 1969;Masuda et al., 1971;Tonndorf et al., 1962). At this level in the pars pectinata the upper layer of BM filaments was very thin and a discontinuity between the bundles was observed. This pathway appears extremely attractive to target sensory HCs via the perilymphatic compartment. Whether this Hensen-Deiters slit is the only path from ST through the BM to cortilymph is not clear. The slit is located next to prominent septa of the BM that extend (5e6 mm in height and diameter) into the organ of Corti (Santi and Johnson, 2013). These acellular protrusions likely serve as an anchoring support for Deiters cells to mechanically support or uncouple Hensens' cells base. In chemically fixed specimens it seems to be a site where adherent junctions are less developed and so para-cellular gaps appear (Figs. 1c and 7c, d and e). The protrusions are most prominent in the apex where additional smaller protrusion support Deiters cells (Fig. 7cee). To what degree such intercellular gaps occurs as a fixation artefact not present at physiological conditions or act as patent pathways into cortilymph needs to be further elucidated. We always found several basement membrane layers at this slit region in human that may act as a filter for certain high molecular weight compounds (Fig. 7e).
A natural port through the BM is the habenula perforata where nerve fibers pass into the sensory epithelium. These oval to round holes (1e3 mm in diameter; Iurato, 1967), underneath the IHC are occupied by unmyelinated nerve fibers and few specialized glia cells that are surrounded by a basement membrane (Fig. 7feg). Adjacent supporting cells add another basement membrane but leave a small cleft open as seen in TEM images (Fig. 7g) that may even be smaller under in-vivo conditions. Drugs that are able to enter to intercellular clefts may be able to enter cortilymph, although gap junctions present in the organ of Corti narrow these gaps. Connexin proteins form hemi-channels as well as cell-cell communication gap junctions vital for cochlear function. Gap junctions are present in organ of Corti supporting cells, glia cells, fibrocytes, and other cell types. Fast K þ recycling into the stria vascularis is accomplished via these intercellular communication routes. Few studies focus on permeation of cytoplasmic molecules through connexin channels, so it is unknown how gap junctions contribute to drug distribution in the inner ear. Limiting factors are molecule size and charge. More care should be taken on drugs not to interfere with gap junction action. Several small compounds have already been found to inhibit connexin function. One example represents the amino acid taurine that is able to directly and reversibly inhibit homomeric and heteromeric channels that contain Cx26 (Locke et al., 2011).
"Leakiness" of the BM may largely depend on molecular formulation of drugs. Kinetics of this passage may also vary between the locations of application. A thicker and narrower BM in the basal turn could influence penetration speed compared to a wider and thinner BM in the apical region. The thicker TCL layer in the apical region full of MCs may compensate for that higher speed of penetration or act as a cleaning filter for compounds close to the basilar membrane.

Round window and oval window-gateways to the inner ear
The two natural fenestrations of the periplymphatic space open towards the middle ear and are occupied by a thin membrane and the smallest bone of our body. These openings act in concert as vibrations entering the inner ear through the stapes footplate at the oval window cause vibrations with opposite phase to the membrane of the RW (RWM). This allows the fluid in the inner ear to propagate and sets the BM in motion (Fig. 8a).
The RWM is a continuation of the mucous membrane lining the middle ear followed by a connective tissue layer containing a collagen-and elastic fiber matrix populated with fibroblasts, bloodand lymph vessels and comprises even some nerve fibers (Rask-Andersen et al., 1999. MCs cover the perilymphatic aspect in the ST (Fig. 8bec). While the connective tissue layer is thinnest in rodents resulting in a RWM thickness of 10e30 mm in guinea pigs (Tanaka and Motomura, 1981), 12 mm in rats (Nordang et al., 2003) and below 10 mm in mice (Fig. 8b) it is dominating feline (20e40 mm in cat) and primate RWM, thickening it to about 70 mm in human (Figs. 8c) and 40e60 mm in rhesus monkey (Goycoolea and Lundman, 1997;Sahni et al., 1987). Thickening of the fibrous layer with age suggests decreasing compliance (Goycoolea and Lundman, 1997). Size and shape of the human RW was evaluated recently (Atturo et al., 2014). Ultrastructure of the inner and outer cell layers suggests a participation in absorption and secretion with increase of surface by various processes (MCs) and microvilli (middle ear mucosa) (Goycoolea and Lundman, 1997). The MC layer shows typical morphology and lack of continuity of cellular junctions. Experimental evidence has shown the passage of ferritin and microspheres from the ST into the RWM (Goycoolea et al., 1988a, 1988b. It was early recognized that even the thick human RWM behaves like a semipermeable membrane. Permeability from the middle ear into the ST of a variety of compounds was found, such as antibiotics, anesthetics, toxins, albumin, 1 mm latex spheres but selectivity regarding size (3 mm latex spheres did not enter!), concentration, liposolubility, electrical charge and membrane thickness was found. Results are reviewed here (Goycoolea, 2001;Goycoolea and Lundman, 1997). In early stages of middle ear inflammation permeability increases but established pathological changes make this membrane even less permeable to protect inner ear function (Cureoglu et al., 2005;Goycoolea et al., 1980;Ikeda and Morizono, 1988;Kawauchi et al., 1989;Lim et al., 1990;Schachern et al., 1987). Induction of an inflammatory reaction to enhance drug permeation through the RWM would only be beneficial for a short period of time, but may be an option for single shot applications. Uptake of substances and particles from the middle ear cavity is seemingly decided by the mucosal epithelial cells and mainly a selective and active process. Cell lines or primary culture from the human or murine middle and inner ear epithelial cells may provide good models to predict uptake and transport in vitro (Lim and Moon, 2011;Mulay et al., 2016).
The RW niche, a funnel-shaped depression in the otic capsule (Fig. 8a) and favors retention of substances at the RWM especially when solid gel-like matrices are placed. Clearance through the middle ear Eustachian tube by the mucosa and ciliated epithelia can be delayed thereby. Interestingly, the rhesus monkey was shown to exhibit a secretory immunodefense-like structure residing at the RWM rim at the niche (Engmer et al., 2008). A previous study demonstrated in 33% of human temporal bones an obstructed RWM caused by either a pseudomembrane or a fibrous or fat plug (Alzamil and Linthicum, 2000). In these cases the oval window (OW) could compensate as an entry port.
The OW is nearly fully occupied by the stapes footplate, only a narrow annular ligament enables the smallest bone of our body to transmit vibrations into the inner ear (Fig. 8dee). Permeability of substances via the OW has long been underestimated. Trans-tympanic gentamicin treatment for Meniere's disease gave hints for the clinical significance of this pathway to control vertigo since vestibular sensory structures are relatively selectively ablated. Transport of Gd tracer through the OW in vivo was demonstrated in rats with MRI imaging (Zou et al., 2012). This pathway appears to be even more effective than the RW pathway for cochlear uptake in humans. Animal studies calculated Gd permeation of the combined solute entering through both the RW and stapes routes for the stapes region 90% (King et al., 2011), for the marker ion trimethylphenylammonium placed to the RW niche 65% for OW and 35% for RW (Salt et al., 2012). Selective OW transport seems ideal to target the five vestibular end organs (Figs. 4a and 8d) or may be more effective for certain substances than the RWM application (King et al., 2013). Still we lack knowledge about the exact path of compounds through this opening. Potential sites include the fibrous annular ligament of the OW or the thin bone of the stapes footplate (Tanaka and Motomura, 1981) (Fig. 8e). Middle ear face of the footplate is populated by mucosa cells resting on a connective tissue layer containing various blood vessels (Fig. 8e). The perilymphatic aspect is covered by MCs. MicroCT imaging can provide valuable data for stapes measurements (Elkhouri et al., 2006;Hagr et al., 2004;Sim et al., 2013). Two human stapes from the same individual presented here (Fig. 8fei) depicts how thin the footplate bone appears around its center. Undermined by various canals and lacunae (Fig. 8geh) mineral portion of hydroxylapatite (HA) content measures 926.6 mg HA/ccm and 965.6 mg HA/ccm that is in the range of bigger lamellar bones. Footplate thickness ranges from 30 to 500 mm in the presented human (Fig. 8i). Considerable variability in thickness is present even within the same subject (Fig. 8i). Stapes footplate application of gentamicin proofed effective to deliver a sufficient dose for the vestibular system (King et al., 2017). Since penetration of substances into perilymph through the bone of the otic capsule in guinea pigs was presented (Mikulec et al., 2009), it is likely that the thin bone of the footplate is permeable for several drugs.

Conclusion
Local drug delivery via the RWM is a safe and effective way to target cells in the cochlea. SGNs are accessible via ST as well as HCs very likely via the cortilymph compartment that seems to be leaky towards the ST. The OW offers another path that may be more permeable for certain drug formulations and provides good access for the vestibular system. Spread of substances towards the apical turn in the cochlea turned out to be the larger problem in bigger inner ears like the human. Diffusion rate in perilymph fluid is low but several bypasses exist that enhance transport in the cochlea via the modiolus. The modiolus is a loose arrangement of neurons, glia cells and fibrocytes in a highly trabecular bone allowing rather unopposed spread of substances. These shortcuts rely on active mechanisms of MCs that may even be enhanced if we gain more knowledge about that cell type. Superparamagnetic nanocarriers in an external magnetic field may be useful to accelerate drug flow towards the apex. Current models for drug spread in the inner ear utilize mainly smaller animals like rats or guinea pigs. Bigger animals such as pig or sheep with inner ears closer to human dimensions (Schnabl et al., 2012) could provide useful information about drug efficacy along the tonotopical axis. This should help to address clinical questions for dosage and requirements for drug stability in perilymph.
Targeting the endolymphatic compartment implies risks to damage a barrier that results in degeneration of HCs. Only drugs or nanocarriers that are able to leave these barriers intact will succeed for a safe application. Drug formulation will always be the key for a successful pharmacotherapy. Combinations of drugs and genes with various nanocarriers may be extremely useful especially for the inner ear (Pritz et al., 2013a). They do not only cross solubility barriers but may in future enable cell specific targeting. Considering the active and passive mechanisms of cellular transport, fluid pathways and dimensions in the inner ear, a site selective pharmacotherapy seems feasible in the near future.

Ethics
Human bodies were donated to the Division of Clinical and Functional Anatomy of the Innsbruck Medical University by people who had given their informed consent prior to death for the use of their bodies for scientific and educational purposes (McHanwell et al., 2008;Riederer et al., 2012). All specimens were anonymized. Post mortem delays until fixation ranged from 6 to 12 h. Surgical human materials from Sweden was approved by the local ethics committee (no. 99398, 22/9 1999(no. 99398, 22/9 , cont, 2003(no. 99398, 22/9 , Dnr. 2013, and subjects gave informed consent. The study adhered to the rules of the Declaration of Helsinki. There was no evidence for any malformation in any human temporal bones. Guinea pig and cat images were taken from archival material used in previous studies (Rattay et al., 2013). Guinea pig were from a study performed in pigmented 250e350 g guinea pigs of both gender from Elm Hill Breeding Labs, Chelmsford, MA in a previous project (Glueckert et al., 2008). Animals were housed in facilities accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International, with free access to food and water throughout the duration of the experiment. Veterinary care and animal husbandry was provided by the Unit for Laboratory Animal Medicine at the University of Michigan, and all protocols were approved by the University Committee for the Use and Care of Animals at the University of Michigan. Experiments were performed in accordance with the guidelines of the National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH Publications No. 80e23, revised 1978). A concerted effort was made to minimize both the number of animals used in this previous study, and the suffering of subjects involved in the study. Re-evaluation from archival celloidine and plastic embedded sections emanate from previous research projects and were published by Spoendlin and others (Spoendlin and Schrott, 1989;Spoendlin and Schrott, 1990).

MicroCT imaging of temporal bone specimens
24 temporal bones from body donors were excised and fixed in Karnovsky's formaldehyde-glutaraldehyde solution for several weeks. To ensure rapid fixative penetration, oval and round windows were penetrated with a needle and the fixative gently perfused with a Pasteur pipette. Specimens were post-fixed in 2% osmium tetroxide (OsO4) for 2 days. After thorough washes in PBS the excess bone was removed in most of these specimens with a drill to meet maximum specimen size for the microCT scanner. Specimens were decalcified in EDTA pH 7.2e7.4 for 6e8 weeks at 37 C in a Milestone ® HISTOS 5 microwave tissue processor and thoroughly washed in PBS for 5 days. Subsequently, they were transferred to 50% and 70% ethanol 3 Â 2 hours each, rotated on an overhead shaker (Heidolph ® Reax) for 2 days and mounted in plastic sample holders again in 70% ethanol. This procedure ensured that air bubbles present in PBS get removed. Scans from the decalcified specimens were acquired using an XRadia MicroXCT-400 at 45 kVp and 109 mA with an isotropic voxel size of 15 mm.
All scans from human inner ears were exported in DICOM format. Scans were imported to Amira ® 6.2 and nerves and structures of the membranous labyrinth were manually segmented switching between the three orthogonal planes using the Segmentation editor. Segmented structures such as the membranous labyrinth, perilymphatic compartments of the whole inner ear, vestibular end organs, vestibulocochlear nerve were visualized using volume and surface renderings, endolymphatic duct as well as the cochlear aqueduct were traced.

MicroCT imaging of stapes specimens
Stapes specimens were vertically mounted in small plastic containers in 1.5% low melt agarose. Quantitative micro-computed tomography was performed using a Scanco mCT35 System (SCANCO Medical AG, Brüttisellen, Switzerland) in order to assess mineralization of the stapes. Image acquisition was performed at 70 kV source voltage and 114 mA intensity with an angular increment of 0.18 between projections and an integration time of 1.6s per projection. Voxel resolution of reconstructed slices was 10 mm (isotropic). Mineral concentrations were measured using the hydroxylapatite-phantom based densitometry calibration of the scanner.
For measurement of the mineral concentrations, image volumes were imported into the 3d software package Amira 6.4 (FEI Visualization Sciences Group, M erignac C edex, France). First, average mineral density was calculated for the whole stapes specimen. Subsequently, the stapes image volume was cropped to the basal plate using the VolumeEdit tool, and average mineral density for the thin parts of the basal plate was calculated. For the cropped basal plate volume, a polygon surface model was created and local surface thickness was measured. Surface thickness values were plotted on the polygon model using a false color lookup table in order to show the thinnest areas of the basal plate.

Transmission & scanning electron microscopy
For detailed protocols the reader is referred to Glueckert et al., 2005a,b (Glueckert et al., 2005a, 2005b. Transmission Electron Microscopy was done at the Institute of Zoology and Center of Molecular Bioscience Innsbruck, University of Innsbruck, Technikerstr. 25, A-6020 Innsbruck, Austria with a Zeiss Libra 120.

Immunostaining
All procedures, antibodies, immunostainings and imaging strategies are described in Liu et al. (2017).

Conflicts of interest
All authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Contributions
R.G. wrote the manuscript, made the figures and participated in scanning-and transmission electron microscopy work; L.J Ch. did segmentation work and fluid space measurements; H.R-A. participated in writing the manuscript, did all immunostaining work and participated in scanning-as well transmission electron microscopy work; W.L. did immunostaining work; S.H. did all microCT imaging and stapes measurements; A.S-F. participated in writing the manuscript and participated in scanning-and transmission electron microscopy work; all authors provided input and proofreading of the manuscript.

Funding
This research was supported by the Austrian Science Fund FWF Austria, project I 3154-B27 (Gapless Man: Machine Interface) and the funding program of the Autonome Provinz Bozen Südtirol, Italy (CUP: B26J16000420003). We thank the Amt der Tiroler Landesregierung, County of Tyrol, Austria for funding us through the K-Regio project VAMEL (Vestibular Anatomy Modelling and Electrode Design).