The response of anosteocytic bone to controlled loading

Bone is a hierarchical composite biomaterial with impressive stiffness and toughness (Currey, 2002; Launey et al., 2010). The bones of all vertebrates share the same basic building blocks: nanoplatelets of carbonated hydroxyapatite, which are arranged in layers within a matrix consisting of collagen type I fibrils, water and small amounts of non-collagenous proteins (Dunlop and Fratzl, 2010; Reznikov et al., 2014; Weiner and Wagner, 1998). The bones of vertebrates also share the same basic organizational unit, the lamella (Ascenzi and Lomovtsev, 2006; Gebhardt, 1905; Giraud-Guille, 1988). Despite its inert appearance, bone is a metabolically active tissue, with the remarkable ability to adapt its structure to changing loads and to self-repair accumulated fatigue damage, by the processes of modeling and remodeling, respectively (Currey, 2002; Robling et al., 2006).

The concept of bone modeling as a response to changing mechanical loads is credited to Wilhelm Roux and Julius Wolff, and was formulated towards the end of the 19th century (van der Meulen and Huiskes, 2002). In the last two decades major advances have been made in the understanding of this process, including some insights into the cellular mechanisms responsible for its regulation (Robling and Turner, 2009). Modeling consists of bone material deposition by osteoblasts or resorption by osteoclasts, with these processes usually not occurring at the same location (Robling et al., 2006). The result is a change in the size and shape of bones, such that their new shape is more favorably adapted to the altered mechanical requirements imposed upon them.

While it is clear that the trigger for bone modeling is mechanical loading, the precise cellular mechanisms responsible for the initiation and orchestration of modeling, including the activation and deactivation of the cells that carry out these processes (osteoblasts and osteoclasts), are not fully understood. It is, however, a widely accepted paradigm of bone biology that osteocytes, former osteoblasts that become trapped in the matrix they deposit, play a crucial role in the regulation of all these diverse dynamic processes (Bonewald, 2011; Schaffler and Kennedy, 2012).

Osteocytes are the most abundant of bone cells, constituting 95% of all bone cells, and are distributed throughout the bone tissue. They are housed in lacunae and interconnect with neighboring osteocytes and bone surfaces via numerous dendritic cell extensions, which are housed in tiny cylindrical canaliculi (Schaffler and Kennedy, 2012). The means by which osteocytes detect the load applied to bone are currently under intense investigation, and several signals are being considered as triggering them, such as electromagnetic fields, shear stress due to fluid flow caused by loading and bone deformation causing local strain or strain energy density concentration (Cowin, 2007; Cowin and Cardoso, 2015; Jacobs et al., 2010; Stern and Nicolella, 2013). The stimulated osteocytes then regulate the activation or deactivation of osteoblasts and osteoclasts by releasing soluble autocrine and paracrine signals, and by direct cell–cell contact (Chen et al., 2010; Nakashima et al., 2011). Recent studies are starting to shed light on the crucial role played by osteocytes in the molecular pathways responsible for the regulation of the modeling process, for instance via nuclear factor kappa β (RANKL β) and sclerostin (Compton and Lee, 2014; Moustafa et al., 2012; Robling et al., 2008; You et al., 2008).

Considering the central role ascribed to osteocytes in the biology of bones, the observation that the bones of many fish species do not contain osteocytes at all is surprising. This observation was first made more than 150 years ago, by the Swiss-German anatomist Kölliker (1859). Even more surprising was the discovery that the anosteocytic state is derived: the bones of less advanced fish like the basal teleosts contain numerous osteocytes, while it is the advanced teleosts (more than half of all extant fish) that totally lack osteocytes. It was suggested, therefore, that the loss of osteocytes confers some advantage on the skeleton of advanced teleosts (Moss, 1961). However, the nature of this putative advantage remains elusive.

The pivotal role attributed to osteocytes in the processes of modeling (adaptation of external morphology to changing loads) and remodeling (replacement of small packets of damaged bone material within the bone bulk) implies that anosteocytic bone is incapable of adapting its structure in response to load. Because of the importance of skeletal adaptation for bone function, and considering such objective criteria of success as their long life span and species diversity, it would be surprising if fish with anosteocytic bones were indeed incapable of these processes. Anosteocytic bone was recently shown to undergo remodeling under extreme loading conditions, for example in the heavily loaded bone of the bill of billfish. These bones were shown to have a dense population of secondary osteons, which can only form by remodeling (Atkins et al., 2014). Similarly, several studies have shown evidence of modeling of anosteocytic bone in response to changing loads, such as the structural changes observed in the pharyngeal jaw of the cichlid Astatoreochromis alluaudi when their diet was changed from hard to soft food (Huysseune et al., 1994). This study showed that the lower pharyngeal jaw of fish eating hard food became substantially thicker and stronger than that of the fish on a soft food diet. Another example is a study which showed that when the spine of Dicentrarchus labrax (sea bass) became excessively curved ventrally (lordotic), the vertebrae adapted to the increased load by morphological changes that decreased the strain energy density in their vertebrae (Kranenbarg, 2005). However, in neither of these reports were the bones directly loaded in a controlled manner and the response measured quantitatively.

Here, we present the results of a controlled experiment of loading of anosteocytic bone of tilapia (Oreochromis aureus), and report the resulting structural and compositional changes, as well as the mechanical properties of the loaded bone.

Eight tilapia [Oreochromis aureus (Steindachner 1864)], 2 years of age, were obtained 4 weeks before the start of the experiment and raised in a controlled environment of 25°C and a light/dark regime of 12 h/12 h. The fish were fed appropriate commercial fish feed for the duration of the experiment. The experiment was approved by the ethics committee of the Hebrew University of Jerusalem.

Fish were anesthetized with clove oil (Syzygium aromaticum) at a concentration of 60 mg l⁻¹. Then, two holes were drilled in the mid-operculum along a line of dorso-ventral orientation 12 mm apart using a hand-held drill (Dremel, Racine, WI, USA) and a 0.5 mm drill bit. Two self-threading 6 mm long stainless steel screws of 1 mm diameter were implanted in the predrilled holes and a 9 mm long nickel titanium (NiTi) orthodontic spring (Dynaflex, St Ann, MO, USA) was attached between them. This spring, made of a memory-shape material and designed to provide a consistent and constant 2 N force, created compression in the area between the two screws (Fig. 1B). In each fish the contralateral operculum had two similar holes drilled but was not implanted with screws and spring (Fig. 1A, right side), and served as a control.

All fish were quadruple-labeled with a sequence of fluorochromes (Calcein Green, Calcein Blue and Alizarin Red; Sigma Chemical Company, MO, USA) by intraperitoneal injection at different, predetermined time points: Calcein Green was administered 30 days prior to implantation of the spring (day −30), Calcein Blue was administered on the day of spring implantation (day 0), Alizarin Red was administered on day 21 of the experiment, and Calcein Green was administered (again) on day 42 of the experiment. For each injection, the fish were anesthetized with clove oil at a dose of 40 mg l⁻¹. A 1% aqueous solution of each of the different fluorochromes (2–3 ml) was prepared and administered at the following dosages: Calcein Green, 25 mg kg⁻¹; Calcein Blue, 30 mg kg⁻¹; and Alizarin Red, 40 mg kg⁻¹. Prior to administration, the solutions were sterilized using 0.2 µm Minisart high-flow filters (Sartorius, Goettingen, Germany).

On day 80 of the experiment (50 days after spring implantation), all fish were killed. Both opercula of each fish (eight pairs) were manually cleaned of soft tissue and preserved in two ways: (1) three opercula pairs were stored at −20°C, to be used for nanoindentation testing as described below; (2) five opercula pairs were fixed with phosphate-buffered 4% paraformaldehyde (pH 7.4) for 12 h. They were then halved along a rostro-caudal line which bisected the distance between the screws holding the spring, using a water-cooled rotary diamond saw (Isomet^® low speed saw, Buhler, Plymouth, MN, USA). Five halves were stored at −20°C, to be used for structural studies (mostly microscopy), while the other five halves were used for histological staining.

Samples were prepared from five pairs of half-opercula as shown in Fig. 1A using a water-cooled rotary diamond saw (Isomet^® low speed saw, Buhler). The coronal surface of each sample was ground with emery paper (800, 1200, 2400 and 4000 grit), then polished with 3 and 1 µm diamond suspensions to a final thickness of 100 µm in the dorso-ventral direction. The samples were viewed by a reflective light microscope (Olympus^® BX 51 microscope, Olympus, Japan), and high-resolution images were obtained with a dedicated camera (Olympus DP71).

The same samples were then studied by linear polarized light microscopy (Nikon Eclipse E600-POL), and images were captured with a digital camera (Nikon DXM 1200 Eclipse).

The same samples were also viewed with a confocal microscope (Zeiss LSM-510) to study their fluorescence labeling. Excitation and emission wavelengths were calibrated for each fluorochrome: Calcein Green, 495 nm excitation and 515 nm emission; Alizarin Red, 530–560 nm excitation and 580 nm emission; and Calcein Blue, 370 nm excitation and 435 nm emission. The distance between the labels was measured (micrometers) and divided by the time span (days) to quantify mineralized tissue apposition rate between the different injection time points (micrometers per day).

Four of the remaining five pairs of opercular halves were de-mineralized with 4% EDTA for 2 weeks. After demineralization, they were dehydrated in graded ethanol solutions and Histo-clear^© (National Diagnostics, Atlanta, GA, USA), then embedded in paraffin (SPI Supplies, West Chester, PA, USA), and sections 5 µm thick were cut with a rotary microtome (Leica^® model RM2245, Solms, Germany) and mounted on glass slides (Superfrost, Thermo Scientific, Barrington, IL, USA). The mounted sections were de-paraffinized in xylene (Sigma Chemical Company, MO, USA), washed twice with ethanol, rehydrated through a graded series of ethanol solutions, and stained with the following stains: (1) hematoxylin and eosin (H&E), which stains proteins and cytosol pink and cell nuclei dark, (2) tartrate-resistant acid phosphatase (TRAP; Sigma), which is an enzyme marker for osteoclast activity, and (3) alkaline phosphatase (ALP), which is an enzyme marker for osteoblast activity (Life Technologies, Carlsbad, CA, USA).

The three opercular samples harvested for mechanical testing were defrosted, then halved equidistant from the screws holding the spring, using a water-cooled rotary diamond saw (Isomet^® low speed saw, Buhler). Single samples, approximately 4 mm×2.5 mm×1 mm (native thickness of the operculum), were cut from each opercular section, such that they contained the directly loaded area between the screws and an area adjacent to the loaded area (Fig. 1), or the equivalent region in the sample from the contralateral, unloaded operculum (Fig. 1). The sections were then dehydrated in an ethanol series (30%, 50%, 70%, 80%, 90% and finally 100% ethanol, 24 h for each stage) to prevent any influence of changing environmental conditions (i.e. humidity), shrinking and cracking. The sections were then embedded in polymethylmethacrylate (PMMA). Ethanol was replaced with MMA monomer and then polymerized at 40°C. The embedded sections were then cut using a water-cooled rotary diamond saw (Isomet^® low speed saw, Buhler) to expose the opercular thickness, which was ground with emery paper (800, 1200, 2400 and 4000 grit), and polished with 3 and 1 µm diamond suspensions. The resulting sample dimensions were 4 mm×2 mm×1 mm. In each sample, three regions of interest were selected: the lateral surface (to a depth of 60 µm and thus considered newly deposited, based on results obtained from fluorochrome staining), the medial surface (to a depth of 60 µm and thus also considered newly deposited) and an area at mid-thickness (pre-existing bone). The samples were tested using a Scanning Nanoindenter (Ubi 1, Hysitron Inc., Minneapolis, MN, USA) with a Berkovic indenter tip. An optical microscope, aligned with the nanoindenter tip, was used to locate regions of interest on the evenly polished bone surface. Because of the irregular surface in the area directly beneath the load, we elected to test the area immediately adjacent to it. The following load function was used: maximum load of 2.5 mN, loading at 0.5 mN s⁻¹, holding at maximum force for 60 s, and unloading to 0.5 mN at a rate of 0.2 mN s⁻¹, followed by a second holding time of 20 s, and finally unloading to 0 mN at a rate of 0.1 mN s⁻¹. Measured properties included peak load, peak displacement and stiffness. Young's modulus of the material was calculated using the Oliver–Pharr method, based on the slope of the unloading curve in the region between 20% and 95% of the maximum load (Oliver and Pharr, 1992). The indentation modulus and hardness were then calculated from the unloading contact stiffness and the indenter contact area (Lewis and Nyman, 2008). Following indentation, the samples were coated with carbon, then examined with a scanning electron microscope (Jeol^® model JCM 6000 benchtop) to verify the position of the indentations, and the florescent labels were used to confirm whether the indentations were in new or pre-existing bone.

Two opercular halves were briefly immersed in liquid nitrogen, then fractured orthogonally to the paramedian surface and fixed with 2% paraformaldehyde for 1 h, after which residual paraformaldehyde was washed off with DDW. The specimens were then dehydrated with ethanol, coated with carbon and examined with a scanning electron microscope (Jeol^® model JCM 6000 benchtop).

One fresh tilapia operculum (obtained separately from a local fish store) was scanned by a micro-computed tomography scanner (Brucker, model 1174 scanner, Belgium). The X-ray source was set at 50 kVp and 800 µA. A total of 450 projections were acquired over an angular range of 180 deg. The sample was scanned with an isotropic voxel size of 17 µm and integration time of 4200 ms, using a 0.25 mm aluminium filter. Scans were reconstructed using commercial software (NRecon^® Skyscan software, version 1.6.1.2). The resulting stack of images was uploaded into AMIRA (FEI, Berlin, Germany) and a three-dimensional tetrahedral meshed model was created using the ‘tetragrid’ command. The resulting geometric model was exported into a finite element analysis package (Patran, MSC Software Corporation, Newport Beach, CA, USA). Bone material properties were assigned based on the microCT scan and the attenuation values of each voxel. These values were converted to mineral density (MD) values using scanned phantoms of known MD, and the Young's modulus of each voxel calculated based on the relationship described by Keller (1994). For simplicity, the resulting moduli were divided into 20 bins and each voxel assigned one of 20 values based on its association with one of these bins.

The model was loaded by the forces of two muscles (the levator opercula and the dilator opercula) and by the compressive 2 N force simulating the experimental spring, and was constrained at 29 nodes in the region of the articulation with the neurocranium (see below). Careful dissection was carried out on a fresh tilapia head to determine the orientation of pull, angle of pennation, physiological cross-sectional area (PCSA) and location of the insertion of the two muscles. Muscle forces were calculated based on their PCSA, and determined to be 0.24 N (levator opercula) and 2.26 N (dilator opercula). Note that in this model, drag forces acting on the lateral surface of the operculum were ignored.

The von Mises stress distributions within the loaded operculum were determined for two scenarios: (1) a physiologically loaded operculum and (2) a similarly loaded operculum with additional spring loading, simulating the experiment.

Data of mineralized tissue apposition rate and Young's moduli were compared between the control and experimental samples, and analyzed statistically using two-tailed t-test, and variation about the mean was expressed as ±1 s.d. Statistical significance was set at a level of P<0.05.

The operculum is the bony cover of the gills (see Fig. 1). It is hinged at its rostro-dorsal aspect and moves laterally by contraction of several muscles (see details in Materials and methods). The operculum is flat and thin, 2–3 cm long (in the rostro-caudal direction), ∼4 cm wide (in the dorso-ventral direction) and 0.5–1 mm thick (native opercular thickness). The experiment consisted of studying the bone's response to the load applied by a spring stretched between two screws that were anchored in the opercular bone. The samples for all microscopic studies (light microscopy, SEM and histology) were prepared by removing a section from the operculum by two parallel cranio-caudal cuts, 5 mm apart, made on both sides of the line bisecting the distance between the screws (see Fig. 1A). This section was then rotated 90 deg so the viewed surface was of the thickness of the operculum (coronal surface, see Fig. 1). We studied the part of the bone directly between the screws (area directly under load) and areas immediately cranial and caudal to the area between the screws (area adjacent to the load) – see Fig. 1.

Fig. 2A,B shows light microscopic images of the coronal cross-section (the thickness) of the control and loaded opercula, respectively, of tilapia in this study. The cross-section of the unloaded (control) sample and of the area of the loaded sample adjacent to the load are lamellated (Fig. 2C,E). When viewed in polarized light, the same samples have alternating dark and bright layers, suggesting a typical so-called plywood arrangement, as collagen fibrils oriented orthogonal to the plane appear dark, while layers in which the fibers are oriented more than 45 deg to the plane appear bright (Bromage et al., 2003). This appearance agrees with the results of previous studies of the lamellated arrangement of this bone (Atkins et al., 2015). In contrast, sections of the bone from the area directly under load exhibit an obvious medially directed dip in the lateral aspect of the bone, which indicates net loss of material. This effect seems to result from the combined processes of resorption and deposition of bone material (Fig. 2D). When the same, directly loaded region is viewed by polarized light, the lower (medial) half consists of dark and bright well-organized layers; however, the top half seems much less ordered and surrounds several small (100 µm diameter) circular structures, that are different from the surrounding lamellated architecture (Fig. 2G). These structural elements are circular with several light and dark concentric circular layers. This organization is reminiscent of the appearance of mammalian secondary osteons.

The operculum grows by sequential deposition of layers of bone material on its medial and lateral surfaces. The temporal sequence of the different fluorochrome injections allowed the determination of mineralized tissue apposition rate (MTAR) in the different regions of bone by viewing coronal sections (see Fig. 3, top, for the sequence of injections) (Bloebaum et al., 2007). The fluorescence labeling is visible and quantifiable, particularly on the medial surface (Fig. 3G,I,K). In contrast, on the lateral surface of the bone, in the area directly under load, the differently colored fluorochromes cannot be distinguished from each other (Fig. 3H) because growth lines appear only in well-organized, lamellated areas. MTAR calculation in this area was therefore not possible. In all other areas, the MTAR was determined on the lateral surface and the results show that in the loaded bone, both the area adjacent to the load and the medial surface of the area directly under the load have higher MTAR values than the control bone. The detailed results of the quantification of MTAR for the lateral surface during the entire experiment are shown in Fig. 4.

The lamellated arrangement of the bone in the control operculum is also apparent in SEM images, as shown in Fig. 5A. The overall orientation of the fibers is horizontal. In contrast, in the loaded area (Fig. 5B,D) a much more complex architecture can be seen, with a cylindrical motif of osteon-like structures with a clear central void scattered within a disorganized matrix. In the area adjacent to the load (Fig. 5C), the bone structure is lamellated, as in the control sample.

Fig. 6 shows the appearance of samples with several histological stains in the loaded and unloaded (control) bones. H&E staining shows, as expected, a total lack of cells in the anosteocytic bone of the tilapia, but nucleated cells are apparent on the bone external surfaces (Fig. 6A–C). The organization in the control bone and in the loaded bone in the area adjacent to the load (Fig. 6A,C) is clearly lamellated, with no cavities. In contrast, in the area directly under the load (Fig. 6B) the bone has several cavities containing cells. TRAP staining, which identifies osteoclasts, demonstrates an increased number of osteoclasts on the surface and within the bone in the area directly under the load (Fig. 6E), while no osteoclasts are detected in the control sample and in the area adjacent to the load (Fig. 6D,F). In the samples stained with ALP (which identifies osteoblasts), some osteoblast activity can be seen on the surface of both the control area and the directly loaded area (Fig. 6G,H), but the area adjacent to the load appears to have higher osteoblast activity (Fig. 6I).

Fig. 7 and Table 1 show the results of nanoindentation testing of opercula in the area adjacent to the load in different regions (newly formed lateral surface, pre-existing center and newly formed medial surface). Average values of indentation moduli were 10.09±1.17 and 12.48±1.04 GPa for the control and loaded opercula, respectively. The regional distinctions were confirmed with fluorescence labeling. Clearly, the loaded bone is stiffer than the control bone in the different regions. The embedding of samples in PMMA was shown to cause an artificial increase in the recorded nanoindentation stiffness results (Bushby et al., 2004). However, from a comparative point of view, the results are valid.

Fig. 8 shows a comparison of results obtained by finite element analysis from an operculum loaded only by muscle forces and an operculum loaded with muscle forces and a spring–screw system simulating the experiment described here. The magnitudes of stresses in the region around the screws of the spring-loaded operculum are clearly higher than those calculated for the same region in an operculum loaded only by muscle forces. This difference is particularly striking in the region directly between the screws, but also extends into the adjacent areas.

The ability of mammalian bone to respond to changing loading circumstances by adapting its structure is a basic paradigm of bone biology (Robling and Turner, 2009). Numerous papers have described this process in detail, as well as its very elegant and efficient regulation (Haapasalo et al., 2000; Lanyon, 1993; Mosley et al., 1997; Rubin and Lanyon, 1985; Schriefer et al., 2005). The widely accepted view is that the primary mechanosensors are the tissue-embedded osteocytes that detect changes in local strains. Once such changes are detected, the osteocytes are also believed to initiate a response by directing and regulating the activation and/or inhibition of osteoclasts and osteoblasts (Bonewald, 2011; Lanyon, 1993; Matsuo, 2008). It could therefore be reasonably expected that bone lacking osteocytes will not be able to respond to load; however, the results presented here clearly establish that this bone can respond to load. This finding leads us to conclude that there must be alternative (non-osteocytic) means to detect load or its effects and to initiate a response.

Here, we show that the bone of the anosteocytic tilapia is able to adapt to applied loads, despite the complete absence of osteocytes. The response takes the form of faster bone deposition in some areas and a combination of resorption and deposition in others, and increased bone material stiffness, compared with unloaded bone. Furthermore, the response to load is not limited to the bone region directly under load, but also occurs in the areas adjacent to the load, and even in the contralateral, unloaded, operculum.

We suggest that osteoblasts are the most obvious candidates to serve as the primary biological mechanosensors in anosteocytic bone. Mature osteoblasts have three potential fates: they can become embedded in the matrix and become osteocytes, they can die by apoptosis, or they can become quiescent and reside on bone surfaces (lining cells) (Kim et al., 2012). Although osteocytes are considered the primary mechanosensors of bone, the osteoblastic lining cells have also been suggested to participate in this function (Mullender and Huiskes, 1997; Robling and Turner, 2009; Witten and Huysseune, 2009). For example, Lanyon's group has shown, using 4-point bending of osteoblast cell cultures, that strain stimulates proliferation of osteoblasts (Damien et al., 1998, 2000). Considering that osteocytes are differentiated osteoblasts that become trapped in the matrix they deposit, this hypothesis is reasonable. It is not clear how surface-dwelling osteoblasts can detect variations in strain deep within the bulk of anosteocytic bone; however, the hypomineralized collagen bundles shown to exist in anosteocytic fish bone may be involved (Atkins et al., 2015). These bundles are located in tubules that traverse the matrix and extend to the opercular surfaces. Cellular extensions of lining cells may extend into these tubules and sense strain concentrations. Because the collagen bundles in the tubules are hypomineralized compared with the surrounding matrix, strains and stresses are amplified in them and could improve the ability of the lining cells to detect increased strains (Reznikov et al., 2014). Furthermore, as the collagen bundles in these tubules are hypomineralized, the resulting stiffness mismatch between them and the surrounding matrix probably contributes to crack deflection and increased toughness of anosteocytic bone (Launey et al., 2010).

In the current study, the nature of the tissue response to loading was very different in the bone region directly loaded, the area located between the two anchoring screws, and the areas adjacent to it. The bone that was directly loaded underwent an intense process of combined resorption and apposition, with net bone loss. Loss of periosteal bone as a result of static loading has also been reported in mammals (Dodds et al., 1993; Meade et al., 1984; Robling et al., 2001a). In one experiment, the ulna of growing rats was loaded (non-invasively). It was shown that static loading did not affect endocortical bone formation, while periosteal bone formation was suppressed. In contrast, dynamic loading increased bone formation on both surfaces (Robling et al., 2001b). Meade and colleagues (1984) tested the effect of static loading in adult dogs, using steel springs implanted in their femoral diaphysis (similar to the methodology used in this study), and found that a continuous static load (superimposed on normal loading) led to new bone apposition periosteally, but without internal remodeling or endosteal changes. Furthermore, the new material had lower ash weight and was less stiff and structurally less organized than the original bone material. McDonald and colleagues (1994) used a similar loading regime applied to the tibiae of New Zealand white rabbits, and found that the rate of bone deposition was proportional to the magnitude of the load. They also showed that, similar to what we found here, the response consisted of combined deposition and resorption. They, however, noted a net gain in bone. These results suggest that in mammals the response to loading involves stimulation of both osteoblasts and osteoclasts, ranging between net loss and net gain, and it is likely that the specific response depends on the parameters of loading such as magnitude, duration and location. The differences in the results of our study and those of Meade et al. (1984) and McDonald et al. (1994) could in part be due to the fact that spring compression loading of a mammalian long bone (which is normally loaded mostly in compression) differs from the less physiological compression loading of a small segment of the operculum of fish.

Although efforts were made to place the spring and screws such that no contact occurred between the spring and the lateral opercular surface, contact could have developed in the course of the experiment. As a result, at least in some of the cases, the opercular response in the area between the screws could be the result of a rather complex combination of spring-induced compression, contact pressure on the surface of the bone and muscle forces. Previous studies in rats (Akhter et al., 1992; Turner et al., 1991) have shown that external 4-point bending of the tibia produces a significant periosteal reaction in the form of woven bone deposition on both the periosteal and endosteal surfaces. This loading was compared with simple periosteal pressure without a bending moment, which produced far less periosteal new bone formation and none on the endosteum (Akhter et al., 1998). However, our goal in the present study was to evaluate whether anosteocytic bone can respond to any external load; the exact nature and distribution of the loading forces were of lesser significance.

The precise cellular–molecular mechanism that regulates the response in fishes cannot be elucidated based on the results of this study. Histochemical staining results, however, using TRAP and ALP, show that the numbers of osteoblasts were higher in the areas adjacent to the load compared with the control bone, and osteoclast numbers were higher in the directly loaded area than in the adjacent areas or in the control bone. It is tempting to assume that the same proteins responsible for the regulation in mammalian bone modeling also function here, so that when anosteocytic bone is placed under loads much higher than those it is normally subjected to, osteoblastic lining cells downregulate the synthesis of proteins such as sclerostin and RANKL, and orchestrate the tissue response to load. A large study testing this hypothesis is currently underway.

The modeling process seen in the area directly under load can occur by resorption of layers by surface osteoclasts combined with laying down of entire lamellated layers by osteoblasts. Another possible mechanism could be a localized sequence of resorption followed by deposition into the resorbed cavity (similar to remodeling of mammalian bone), which will result in osteon-like structures. Close examination of the light microscopy (reflective and polarized) images of the directly loaded area reveals that both mechanisms may have occurred. As can be seen in Fig. 2, most of the newly formed area consists of a wavy but layered architecture, suggesting surface resorption and deposition. However, osteon-like cylindrical structures can also be seen (see inset in Fig. 2D,G), which appear round and contain concentric lamellar layers that surround a central void. We recently showed that acellular bone can, at least in the bill of billfish, contain a dense population of secondary osteons, suggesting that they undergo intense remodeling (Atkins et al., 2014). It may therefore be that if the load applied here had been present for longer periods, the density of the osteon-like structures would have increased.

Load applied to any solid body results in a characteristic three-dimensional distribution of stresses within the entire body. The results of the finite element analysis of the spring-loaded operculum show clearly that increased stresses were induced by the spring in the region between the two anchoring screws. The adjacent areas, however, also experience increased stresses compared with those experienced by the control operculum, with a magnitude that decreases with the distance from the directly loaded region (Fig. 8). It is therefore not unexpected that effects of the spring loading will also be present in the areas adjacent to the loaded area. However, Fig. 3 shows quite clearly that this response was of a different nature. In these areas, the response was limited to accelerated bone deposition, without concurrent resorption, so the resulting structural pattern resembles the layered architecture typical of normal anosteocytic bone (Cohen et al., 2012).

Evaluation of the effect of loading on mammalian bones is often performed by applying load to one limb and considering the difference from the unloaded limb, which serves as an unaffected control (Forwood et al., 1996; Hsieh et al., 2001; Robling et al., 2001a). Surprisingly, we found here that when a spring load was super-imposed on the physiological muscle loads of one operculum, an effect was also noted in the contralateral operculum, which was physiologically loaded and drilled, but not spring loaded. The nature of this effect was seen as accelerated bone apposition, while the original layered structure was maintained. This result suggests that in anosteocytic fish bone the response of local loading is expressed systemically, perhaps by release of active signals in a gradient, with the highest concentration in the immediately loaded area, becoming lower as the distance from this region increases, and also affecting distant parts of the skeleton by the carrying of the signaling molecules into the systemic circulation. We note, however, that other possible explanations may be that the control operculum responded to the drilling of the holes, or the presence of the spring affected the function of the loaded operculum, and thus the contralateral operculum altered its function and, as a result, its structure.

In mammals it has often been shown that when new bone material is deposited in response to loading, this new material is less organized and, even more importantly, less mineralized and therefore less stiff (i.e. it has a lower Young's modulus) (Carter, 1984; Glimcher, 1998; Meunier and Boivin, 1997). Surprisingly, we found that the stiffness of the bone material in the loaded operculum, both in the pre-existing central region and in the newly deposited layers on the medial and lateral surfaces, was significantly stiffer than the control operculum. We are not sure of the mechanism leading to this increase in stiffness; however, the most likely explanation would be that the signals released due to loading increased not only the rate of bone apposition but also the rate of mineralization, thus increasing the material's stiffness. A similar result was reported by Wallace and co-workers (2009), who showed that short-term exercise in mice improved mechanical properties by improving the quality of the bone matrix without affecting bone geometry.

In summary, we have shown that anosteocytic bone is capable of responding to load despite the absence of osteocytes. This response occurs in a mixed mode of surface modeling and remodeling.

We are grateful to Tomer Stern who wrote the code that automatically converts the 3D model created by AMIRA to a finite element, PATRAN-readable format, and allows assigning Young's moduli to groups of elements with user-defined mineral density, based on attenuation values obtained from microCT scans.