Architecture of the osteocyte network correlates with bone material quality


  • Michael Kerschnitzki,

    1. Max Planck Institute of Colloids and Interfaces, Department of Biomaterials, Potsdam, Germany
    2. Berlin-Brandenburg School of Regenerative Therapies (BSRT), Berlin, Germany
    Search for more papers by this author
  • Philip Kollmannsberger,

    1. Max Planck Institute of Colloids and Interfaces, Department of Biomaterials, Potsdam, Germany
    2. ETH Zurich, Laboratory of Applied Mechanobiology, Department of Health Sciences and Technology, Zurich, Switzerland
    Search for more papers by this author
  • Manfred Burghammer,

    1. European Synchrotron Radiation Facility, Grenoble, France
    Search for more papers by this author
  • Georg N Duda,

    1. Berlin-Brandenburg School of Regenerative Therapies (BSRT), Berlin, Germany
    2. Charité, Julius Wolff Institute and Center for Musculoskeletal Surgery, Berlin, Germany
    Search for more papers by this author
  • Richard Weinkamer,

    1. Max Planck Institute of Colloids and Interfaces, Department of Biomaterials, Potsdam, Germany
    Search for more papers by this author
  • Wolfgang Wagermaier,

    1. Max Planck Institute of Colloids and Interfaces, Department of Biomaterials, Potsdam, Germany
    Search for more papers by this author
  • Peter Fratzl

    Corresponding author
    1. Berlin-Brandenburg School of Regenerative Therapies (BSRT), Berlin, Germany
    • Max Planck Institute of Colloids and Interfaces, Department of Biomaterials, Potsdam, Germany
    Search for more papers by this author

  • MK and PK contributed equally to this work.

Address correspondence to: Dr. Peter Fratzl, PhD, Max Planck Institute of Colloids & Interfaces, Department of Biomaterials, Am Mühlenberg 1, D-14424 Potsdam, Germany. E-mail:


In biological tissues such as bone, cell function and activity crucially depend on the physical properties of the extracellular matrix which the cells synthesize and condition. During bone formation and remodeling, osteoblasts get embedded into the matrix they deposit and differentiate to osteocytes. These cells form a dense network throughout the entire bone material. Osteocytes are known to orchestrate bone remodeling. However, the precise role of osteocytes during mineral homeostasis and their potential influence on bone material quality remains unclear. To understand the mutual influence of osteocytes and extracellular matrix, it is crucial to reveal their network organization in relation to the properties of their surrounding material. Here we visualize and topologically quantify the osteocyte network in mineralized bone sections with confocal laser scanning microscopy. At the same region of the sample, synchrotron small-angle X-ray scattering is used to determine nanoscopic bone mineral particle size and arrangement relative to the cell network. Major findings are that most of the mineral particles reside within less than a micrometer from the nearest cell network channel and that mineral particle characteristics depend on the distance from the cell network. The architecture of the network reveals optimization with respect to transport costs between cells and to blood vessels. In conclusion, these findings quantitatively show that the osteocyte network provides access to a huge mineral reservoir in bone due to its dense organization. The observed correlation between the architecture of osteocyte networks and bone material properties supports the hypothesis that osteocytes interact with their mineralized vicinity and thus, participate in bone mineral homeostasis.


Biological tissues are characterized by interactions between cells and their environment. This environment is usually extracellular matrix formed and controlled by the cells. A particularly interesting case of the interplay between cells and surrounding extracellular matrix are osteocytes in bone. Osteocytes are the most abundant bone cell.[1] They derive from bone-forming osteoblasts that get embedded in the material they synthesize.[2] During the deposition of unmineralized bone matrix, which subsequently mineralizes, osteocytes form dendritic cell processes, thereby establishing connections with their neighboring cells and the bone surface.[3] After the tissue has mineralized, these processes reside within a dense network of lacunar and canalicular cavities.[4] Osteocytes are multifunctional and are known to orchestrate the complex interplay of bone resorption and formation by osteoclasts and osteoblasts on the bone surface[5-8] by translating mechanical strains into biochemical signals.[9] For this mechanotransductive function, the intricate network between the cells seems of decisive importance. Moreover, osteocytes are considered to play an active role in mineral homeostasis by directly interacting with their mineralized vicinity.[1, 10] Here again, the dense network of lacunar and canalicular spaces is crucial as it provides abundant bone surfaces exposed to the cells. This surface area is estimated to be more than 100-fold larger than the area available to osteoblasts and osteoclasts.[11]

Consequently, it is conceivable that interactions between osteocytes and the enormous bone surface they are facing could significantly contribute to mineral homeostasis, thus maintaining circulating ion levels in the body during events of severe ion demands.[11] In recent years, evidence has accumulated that osteocytes are involved in controlling body phosphate levels[12] and might be capable of resorbing their pericellular bone mineral.[13, 14] However, the proposed interactions of the cells with their surrounding material should leave traces of their activity in the mineralized matrix. Therefore, a promising approach to better understand the relationship between osteocytes and their surrounding mineralized matrix is to spatially correlate the osteocyte network architecture with bone material properties describing bone quality[15] in the proximity and far away from the cell network.

To date, this approach was hardly followed because it requires a combination of imaging methods with high spatial resolution. A suitable methodology at the crossover between biology and material science has to provide at once, detailed information on the three-dimensional structure of osteocyte lacunar-canalicular systems (OLCS)[16] and on the mineral properties of the adjacent bone material, both with at least a micrometer resolution. To tackle this, we developed a new method allowing a precise quantitative correlation between the cell network and the surrounding bone material. We used confocal laser scanning microscopy (CLSM) combined with rhodamine staining[17] to visualize and quantify the OLCS in undemineralized bone sections (Fig. 1). Complementary synchrotron radiation in a microfocus setting was employed to scan the very same precharacterized bone samples to measure the small-angle X-ray scattering (SAXS) signal with a submicrometer resolution. With the mineral particles having a thickness of only a few nanometers, SAXS is the appropriate method to determine nanoscopic material properties such as the mineral particle arrangement (ρ) and thickness (T) parameters in a position resolved way.[18] These parameters describe bone mineralization and organization, both closely associated to bone material quality.[19]

Figure 1.

Osteocyte network visualization and quantification. (A) Bone samples are stained with rhodamine, which diffuses through all microscopic and nanoscopic extracellular spaces, representing the osteocyte network. The location of the dye is measured with confocal laser scanning microscopy. (B) z-Projection of a confocal image stack. (C) Segmentation of the data into mineralized matrix (black), microscopic bone features such as blood vessels and osteocyte lacunae (shades of blue), and the medial axis skeleton of the canalicular network (shades of red). (D) Subvolume showing confocal raw data (gray), segmented osteocyte lacunae (blue), and skeletonized canalicular structures (shades of red depicting the z-value of each canalicular voxel). (E) Conversion of osteocyte network data to a graph structure consisting of cell lacunae (blue), canalicular nodes (yellow dots), and edges (black lines). Edges represent connections between nodes of the canalicular network. (F) Combination of raw image and processed network data from D and E with segmented cell lacunae (blue) and skeletonized canalicular network (shades of red) in a representative subvolume of 25 × 40 × 4 µm. Scale bars = 20 µm.

With this approach, we succeeded to measure a direct correlation between the osteocyte network architecture and bone material quality. In this work, we show in particular a detailed three-dimensional visualization of the osteocyte network in bone, which allows us to (1) characterize cell-cell connectivity via complex canalicular structures. These data not only reveal present network logistics, but also allows (2) determining average distances between the cell network and the mineralized bone matrix. Finally, these distance data are (3) correlated to the bone material quality elucidated by SAXS measurements to better understand the relationship between osteocyte network architecture and bone mineral embedded in their proximity inside the extracellular matrix.

Materials and Methods

Bone samples and sample preparation

Native ovine fibrolamellar (plexiform) bone samples, not yet showing secondary osteons due to remodeling, were collected from the midshaft region of the femur of a mature 5-year-old sheep. For initial preparation, samples were anhydrously cut with a low-speed diamond saw (IsoMet; Buehler GmbH, Duesseldorf, Germany). The resulting plates with a cross section of 5 × 5 mm and an initial thickness of 200 µm were further polished to a final thickness of 20 µm using an automatic polisher (Logitech PM5; Logitech Ltd., Glasgow, UK). During all steps of sample preparation, ethylene glycol was used as an anhydrous sample preparation medium. After polishing, samples were washed in pure ethanol to remove remaining ethylene glycol. Bone samples were examined to identify regions of interest (ROIs) and subsequently marked by creating artificial notches with laser micro dissection (PALM MicroBeam C; P.A.L.M. Microlaser Technologies GmbH, Bernried, Germany). Tagging the identified ROIs allows precise matching of data.


Bone samples were stained over night with rhodamine-6G (Sigma Aldrich GmbH, Taufkirchen, Germany) that was dissolved (0.002% wt) in methyl methacrylate (MMA) solution. As described,[20] the fast diffusion of rhodamine through the bone samples leads to a fluorescent labeling of all internal surfaces such as those from blood vessels, osteocyte lacunae, and canaliculi. After washing with pure ethanol three times, confocal z-stacks of the previously identified ROIs were acquired (Fig. 1A) using a Leica DM IRBE (Leica Mikroskopie und Systeme GmbH, Germany) equipped with a 100× oil immersion microscope objective with a numerical aperture of 1.4. The resulting images (Fig. 1B) have a diffraction-limited resolution of about 250 to 300 nm. The average spacing of structures in the lacunar-canalicular network was found to be on the order of 2 µm using focused ion beam–scanning electron microscope (FIB-SEM)[21] and much smaller distances are likely to occur only around junctions; therefore, the analysis of the network structure is not compromised by the resolution limit.

Thresholding and segmentation of image data

The original data were Gaussian filtered (5 × 5 × 5 kernel, SD = 0.65) before automatic thresholding was applied to convert the intensity stack into a binary volume with foreground (lacunar-canalicular network) and background (bone matrix) voxels (Fig. 1). The threshold was determined by fitting a double exponential to the stack histogram, corresponding to image data and noise, respectively, and setting the threshold to the intensity value at which the noise contribution fell below 1%. Because the analysis is robust to the choice of the threshold within a certain range (see Supporting Information Fig. S1), other thresholding algorithms described in the literature will also work, depending on the level of noise in the images. Discretization noise was removed by morphological closing (dilation followed by erosion using a 3 × 3 × 3 kernel). The binary volume was further cleaned by discarding small isolated foreground clusters and by converting isolated background components (voids) to foreground.

In the next step, larger structures (cell lacunae or blood vessels) were automatically separated from the canalicular network. First, the distance transform of the foreground was calculated, assigning to every foreground voxel its Euclidean distance to the nearest background voxel. Subsequently, all connected regions with a distance of at least 1 µm from the background and a size of at least 16 µm3 were defined as cells or blood vessels. Finally, the identified cells and blood vessels were expanded (dilation with a 3 × 3 × 3 kernel) until reaching a distance of 0.6 µm to the background to approximate the original cell shape. The choice of all parameter values was validated on a number of image stacks and then kept constant throughout the analysis of all data.

Medial axis skeletonization and network analysis

The structure of the cell network was derived from the binarized images by computing the medial axis of the foreground. Voxels were iteratively removed from the surface of the volume if removing them did not change the connectivity of their 3 × 3 × 3-neighborhood and if they had more than one neighbor in their 3 × 3 × 3-neighborhood (otherwise they are endpoints). We implemented the homotopic thinning algorithm described by Lee and colleagues,[22] using parallel rechecking in eight subvolumes with nonoverlapping neighborhoods. The result was a one-voxel-thick medial axis skeleton that is homotopic to the original image (Fig. 1C).

In the next step, the voxel skeleton was converted into a network topology described by nodes, links, and cells. Clusters of voxels with more than two neighbors each were defined as a node, and groups of connected voxels with two neighbors each were defined as links or branches (if connected to one node only). All branches shorter than 1 µm and all loops and isolated nodes were removed (Fig. 1D, E, F).

This topology was used to calculate length and degree distributions of the network, and to assign the distance to the nearest cell through the network to every voxel in a link. A link matrix (dimension N × N, where N = number of nodes) was calculated, with entries aij denoting the distance between node i and node j if there was a direct link between the two. From this matrix, the shortest path between two nodes within the network was determined using Dijkstra's algorithm.

Finally, the logical structure was converted back into a cleaned voxel skeleton, which was used for analyzing the distance of the bone matrix from the cell network. For this purpose, the distance transform of the bone matrix to the cell network was calculated, assigning to each matrix voxel its Euclidean distance to the nearest voxel in a cell, and to the nearest voxel in the lacunar-canalicular network including cells.

All image analysis and quantification algorithms were implemented in MATLAB R2009a (The MathWorks, Natick, MA, USA) using the image processing toolbox and the MatlabBGL graph library. The MATLAB code is available from the authors on request.

SAXS measurements

Bone samples were scanned at the Nanofocus beamline ID13 at the European Synchrotron Radiation Facility (ESRF), Grenoble, France, using a monochromatic 14-keV X-ray beam (λ = 0.886 Å) with a beam size of 500 nm. An ESRF FReLoN camera (Kodak) as detector, with 2048 × 2048 pixels and a corresponding pixel size of 51.5 × 50.7 µm2 was used to record the scattering patterns. Bone sections were stabilized with a lead frame and screwed on a sample holder that was magnetically mounted to the sample stage. During scanning, the stage was translated perpendicularly to the beam with a step size of 1 µm. Typical scan areas had a dimension of 200 × 200 µm, including around 40,000 recorded SAXS patterns.

Data correction

The accurate sample to detector distance and beam center position were determined from the diffraction patterns of hydroxyapatite (HAP) standard powder using Fit2d[23] software. Scattering intensity data were corrected in relation to empty beam (EB) intensities following the equation:

display math

with t as the correction factor, defined as,

display math

Here, t1 is the correction of the transmission intensity (measured with a beam diode), which is the ratio of the intensity at the sample (Isample) and the empty beam intensity (IEB) and t2 as the correction of the decreasing X-ray beam intensity during scanning, measured with a beam monitor. For t1 the decrease of the ring current during the transmission scan was neglected because transmission scans were very short.

Data analysis

Scattering data were analyzed using Autofit (custom-made analysis software; C. Li, Max Planck Institute of Colloids and Interfaces, Potsdam, Germany). For calculation of the average mineral particle orientation in the exposed sample volume, the scattering intensity was integrated over q as a function of the azimuthal angle χ. The distribution of the integrated intensity along q over χ describes the average direction (peak position of both intensity maxima) and the fraction of aligned mineral particles in the exposed sample volume, described by the rho parameter, which corresponds to:

display math

The amount of aligned particles is represented by A1 + A2. The baseline area A0 represents the amount of unaligned mineral particles residing within the exposed sample volume (Supporting Information Fig. S2). The alignment of mineral particles corresponds to the orientation of the collagen matrix as the collagen molecules are directing mineral particle growth.[19]

For calculation of the mineral particle thickness parameter (T parameter), the decay of the scattering intensity integrated over all χ at larger q-values was analyzed. The T parameter is defined as:

display math

with φ as the volume fraction of the mineral phase, σ as the surface area of mineral particles per unit volume, P as the Porod constant (intercept of the Porod fit) determined from the Porod plot, and Ī as the integration of the Kratky plot (Supporting Information Fig. S3), defined as:

display math

which further simplifies to:

display math

as for biphasic systems:

display math

Fitting parameters for the Porod plot were set to Porodmin = 1.5 and Porodmax = 2.8 nm−1. The fit parameter Kratkymin, from which the Kratky plot is interpolated for small q-values (due to insufficient q-resolution), was set to 0.3 nm−1.


Extensive osteocyte interconnectivity via canalicular structures

The graph representation of the canalicular network first allows the analysis of potential cell-cell connectivity via the canalicular structures (Fig. 2). Our analysis showed a strong interconnection between cells (Fig. 2A) but surprisingly, this interconnection extends not only to the neighboring cell lacunae, but to all lacunae within the analyzed bone volume (Fig. 2B). In addition, all cell lacunae have direct access to their neighboring blood vessels via the network, which is crucial for nutritional supply (Fig. 2A, B). To quantify average transport lengths along canalicular structures toward the cells, we estimated the network region controlled by each cell by assigning the nearest cell to each network voxel (Fig. 2C). The distribution of the shortest distance of a network voxel to its adjacent cell shows that almost 70% of the canalicular network is within a transport distance of only 10 µm (Fig. 2E).

Figure 2.

Cell-cell interconnectivity and canalicular network analysis. (A) Yellow lines depict direct connections between individual segmented cells (blue) via the canalicular network structures. (B) Exemplarily for one cell, the shortest direct connection through the canalicular system to each neighboring cell is depicted in yellow. (C) Color coding of canalicular network voxels depending on their nearest cell structure as a measure of the network area that is controlled by each cell. (D) Classification of canalicular junctions (nodes) according to the number of attached connections (degree). Red points depict interwoven nodes featuring a degree > 5. (E) Histogram (black) and cumulative plot (red) of the travel distance from canalicular voxels to the nearest cell through the network, showing that almost 70% of the canalicular network resides within a 10-µm travel distance. (F) The degree distribution of canalicular junctions is exponential (black line: fit ≅ exp(–βd), β = 0.87) revealing the single-scale character of the canalicular network. There is a high abundance of junctions with dendritic character (65% of nodes show d = 3), but junctions with up to 10 connecting canaliculi are also present. The mean degree of nodes is 3.25. (G) The length distribution of individual canalicular structures is exponential (black line: fit ≅ exp(–αl), α = 0.7 µm−1) with individual lengths up to 15 µm. The mean canalicular length between two nodes is 2.15 µm. Scale bars = 20 µm.

Second, we quantified topological parameters of the network, which are of importance for its function as a communication network and yield information on the physical constraints present during its formation.[25] The edge lengths l in the network follow an exponential distribution p(l) ≅ exp(–αl) with a slope α = 0.7 µm−1; the mean canalicular length (travel distance) between the nodes is about 2.15 µm (Fig. 2G). The degree d of nodes (ie, the number of edges, respectively canaliculi, that connect to it) is also exponentially distributed, p(d)exp(–βd) with a slope β = 0.87, the mean degree is 3.25 (Fig. 2F). The large amount of branching nodes (with degree 3) shows the mainly dendritic nature of the canalicular network. However, 35% of nodes show a rather interwoven type, because four or more than 10 individual canaliculi connect to these junctions (Fig. 2D).

Majority of mineral resides within a micrometer from the cell network

The processed osteocyte network data also contain quantitative information about the amount of mineral residing in the vicinity of the osteocyte network (Fig. 3). For this quantification, the Euclidian distance of every matrix voxel to the closest cell (Fig. 3A) was calculated. Distance distributions show that 80% of all matrix voxels reside below a distance of 10 µm to the adjacent cell (Fig. 3C). However, the distance distribution changes dramatically by considering not only the cell lacunae, but the entire network including the canaliculi (Fig. 3B). We found that within a distance as small as 1 µm, already 60% of the mineralized matrix are located, and within 1.4 µm, 80% of the mineralized matrix are located (Fig. 3D).

Figure 3.

Distance distribution between bone mineral and the osteocyte network. z-Projection of the Euclidean distance of each mineralized matrix voxel to the closest cell voxel (A) and to the closest network voxel (B) (either cell or canalicular structure). Distance distributions show that 80% of the matrix voxels reside within a distance of 10 µm to the closest cell (C) and within only a distance of 1.4 µm to the closest canalicular network structure (D) (the gray dashed line shows the distance from the closest cell for comparison). Due to the sample thickness of 20 µm, distance values above 10 µm (depicted with the dashed line in C) are not reliable and could be artifacts from the sample edges. Due to short mineral-cell network distances, these artifacts are not present in D. Scale bars = 20 µm.

Bone material quality depends on network architecture

Position resolved nanoscopic mineral properties such as the mineral particle thickness parameter (T) and orientation (ρ) describing bone material quality (see Materials and Methods) were measured in a microfocus setting by scanning SAXS in precharacterized bone sections featuring distinct osteocyte network organizations (Fig. 4). In areas of dense, highly aligned osteocyte network (Fig. 4A), the mineral particle thickness (Fig. 4B) and particle orientation (Fig. 4C) are increased showing values of T > 2.5 nm and ρ > 0.5, respectively. Within bone regions that exhibit loose, unorganized network structures, mineral particles are thinner (T < 2.5 nm) and poorly oriented (ρ < 0.5). The direction of mineral particle alignment is always parallel to the osteocyte long axis, perpendicular to the direction of the canalicular alignment. Moreover, in the direct vicinity of osteocyte lacunae within regions of a highly organized network, lower mineral particle thicknesses T and orientations ρ are apparent (Fig. 4D).

Figure 4.

Osteocyte network and nanoscopic mineral particle properties. (A) Visualization of the osteocyte network showing highly organized (dense) osteocyte network structures at the top and the bottom and poorly organized (loose) network structures in the center. Synchrotron small angle X-ray scattering (SAXS) measurements of the mineral particle thickness (T-parameter) (B) and the mineral particle orientation (Rho parameter) (C) along the visualized network structures. Mineral particles in dense network regions are thicker and more oriented compared to the loose regions in the center. The direction of alignment of mineral particles is always perpendicular to that of the canalicular structures. (D) Colocalization of T-parameter maps with the osteocyte network reveals thinner mineral particles around individual osteocyte lacunae. Scale bar = 20 µm.

The correlation between the nanoscopic material properties and the position of material in relation to the network is shown in Fig. 5 using the example of the mineral particle thickness parameter T. Large values of T are only obtained close to the osteocyte network (Fig. 5A, B). The distribution displays a sharp peak of the particle thickness at T = 2.75 nm with 1 SD of 0.1 nm (Fig. 5C). This bone material is located in regions with a dense, organized osteocyte network. For low values of the T parameter, the correlation between T and the distance to the network is much weaker. This occurs in regions with a loose, unorganized osteocyte network. The separation between the two different ways of organization of the osteocyte network is reflected in the bimodal distribution of the mineral particle thickness T. In other words, dense osteocyte networks feature highly oriented bone mineral with homogeneously thick mineral particles, whereas in areas of loose osteocyte networks, bone material is less organized and thus, bone material quality is rather low.[19, 24]

Figure 5.

Correlation of microscopic network organization with nanoscopic mineral particle properties. (A) z-Projection of the Euclidian distance of mineralized matrix voxels to the osteocyte network structures. (B) Map of the according mineral particle thicknesses together with the osteocyte network organization. (C) Correlation of mineral particle thickness (T-parameter) with the distance of these mineral particles to the cell network. The T-parameter values shows a sharp mineral particle thickness distribution with an average of 2.75 ± 0.1 nm in dense network regions and broader distributions with an average T-parameter of 2.3 ± 0.3 nm in loose network regions. Color coding of histograms in C is consistent with the color code of the distance and T-parameter map in A and B, respectively. Scale bar = 20 µm.


Our combined approach of visualizing and quantifying CLSM data together with position resolved SAXS mapping reveal the complex three-dimensional structure of osteocyte networks in mineralized bone tissue in correlation with the distribution of nanoscopic mineral properties. This allows the detection and further analysis of nanoscopic network structures and even the application of theoretical network analysis. This reveals not only extensive osteocyte connectivity, but also typical canalicular network properties such as node point characteristics and mean travel lengths through the network.

Our data show that all osteocytes in the visualized volume—and not only neighboring cells—are interconnected via the extracellular spaces of the canalicular network. This mainly dendritic canalicular network exhibits a strongly interwoven characteristic with a high abundance of connecting canaliculi in the vicinity of osteocytes and within osteocytic layers. The exponential decay of the canalicular node degree distribution reveals the single-scale character of the canalicular network. Single-scale networks are characterized by the fact that physical constraints such as transport costs influence the addition of new links.[25] In bone, this may indicate an optimization of the channel network with respect to transport costs of nutrients and signals. Thus, these dense and multiconnected canalicular structures create an efficient communication network and is also robust against local interruptions as a result of microcracking, for example, which can be a consequence of severe mechanical loading.[26] However, it needs to be kept in mind that our network data describe the complex system of extracellular voids (that is the “highway-system”) but not the cells themselves (which would be responsible for the actual “traffic”). Thus, these findings lead to intriguing considerations concerning the flexibility of cellular connections residing within the canalicular structures, as live-imaging experiments suggest that osteocytes permanently expand and contract their dendrites inside these voids in response to stimuli.[27] This could be of high importance because stimulated formation of temporarily connected dendrites could strongly modify cell network characteristics such as the clustering of or the path length between cells. In network theory these two network parameters are used to classify networks to regular, random, or small-world networks.[28]

The complex organization of the osteocyte network not only seems to be optimized for efficient communication between the cells, but it has the potential to provide access to most of the mineralized bone matrix due to its dense architecture. In highly organized (dense) regions of the osteocyte network, these structures are arranged to feature short distances to the mineral. Indeed, calculations of average cell network–mineral distances show that the majority of bone mineral resides within a distance of only 1 µm to the nearest cell network structure. This leads to a reduction by almost one order of magnitude of the distance that mineral has to diffuse through the bone material before reaching the canalicular network, in which even large molecules can diffuse relatively fast,[29] instead of reaching a cell lacuna. This becomes even more significant when keeping in mind that diffusion times usually scale with the square of the covered distance. Scanning SAXS investigations of nanoscopic bone mineral properties on the same sample sections show a strong correlation between mineral characteristics and the osteocyte network structures. We find that highly organized bone mineral featuring relatively thick and highly oriented particles—associated to high bone material quality—predominantly resides at short distances to the next cell or canalicular structure. In contrast, mineral particles at larger distances from the network are less organized and show broader distributions in thickness and orientation. These findings support the hypothesis that osteocytes directly control bone mineral in their vicinity,[1, 10] but also that a dense and well-organized osteocyte network may be crucial to facilitate this control by either active (cellular) or passive (dissolution/apposition) mechanisms, thereby leading to higher bone material quality.

How cells are influenced by their material environment, such as stiffness[30] or substrate geometry,[31] and reciprocally, how these cells manipulate their environment, is not only of paramount importance in the context of the biological function of cells. An understanding of this bidirectional feedback mechanism between cells and the surrounding material can fundamentally change the future design of in vitro experiments using cell cultures. We believe that the utilization of our approach to combine experimental techniques from biology and materials science will provide important new insights for this interplay. It allows, in the very same biological specimen, to reveal both the organization of the cells together with the impact of cellular action on the material properties of the surrounding extracellular matrix.

The samples used in this study were from primary fibrolamellar (plexiform) bone. However, our findings can be generalized to other types of lamellar bone because the degree of organization of the osteocyte network is similar in different types of lamellar bone of different species, but differs significantly from that found in woven bone.[17] In previous work, we found that osteocyte networks qualitatively mirror bone formation, which in the case of lamellar bone is a layer-by-layer process.[17, 20] Our method for network quantification presented here could therefore be used in the future to study the role of cellular network organization for the formation of lamellar or woven bone, or to relate pathological changes on bone material properties to the characteristics of the osteocyte network structure.

In conclusion, osteocytes form dense and highly organized networks throughout the mineralized matrix, with the majority of bone mineral residing within a distance of 1 µm from it. The degree of organization of the network indicates a high level of communication between cells, which is rather reminiscent of a neuronal network,[32] and was hidden by the all too visible mineralized matrix. Many comparable studies have been carried out on complex brain networks. These described a modification of network characteristics, such as clustering and mean path length, due to cognitive and psychiatric disturbances present in patients with Alzheimer's disease or schizophrenia.[33] Concerning the osteocyte network architecture in bone, in principle it is also conceivable to relate a modification of these network parameters to bone diseases such as osteoporosis. In this regard, disturbances in the osteocyte network organization have a strong impact on fluid flow, and diffusion of ions, hormones, and signaling molecules, all factors promoting cellular communication and mechanosensation via the osteocyte lacunar canalicular network,[34-36] thus ultimately affecting bone material quality.[15]


All authors state that they have no conflicts of interest.


We thank AM Martins and B Schonert from the Max Planck Institute of Colloids and Interfaces for technical assistance and H Schell from the Julius Wolff Institute for providing ovine bone samples collected in the SFB 760. MK is funded by Bundesministerium für Bildung und Forschung (BMBF, grant-number: 01 EC 1006C, project 9133).

Author's roles: Study design: RW, WW, and PF. Study conduct: MK, PK, MB, GD, and WW. Data collection: MK, PK, MB, and WW. Data analysis: MK, PK, and WW. Data interpretation: MK, PK, MB, GD, RW, WW, and PF. Writing the manuscript: MK, PK, GD, RW, WW, and PF. WW takes responsibility for the integrity of the data analysis.