Controlled Mechanical Loading Affects the Osteocyte Transcriptome in Porcine Trabecular Bone in situ

Abstract

Introduction:

Osteocytes modulate bone adaptation in response to mechanical stimuli imparted by the deforming bone tissue in which they are encased by communicating with osteoclasts and osteoblasts as well as other osteocytes in the lacuna-canalicular network through secreted cytokines and chemokines. Understanding the transcriptional response of osteocytes to mechanical stimulation in situ could identify new targets to inhibit bone loss or enhance bone formation in the presence of diseases like osteoporosis or metastatic cancer. We compared the mechanically regulated transcriptional response of osteocytes in trabecular bone following one or three days of controlled mechanical loading.

Methods:

Porcine trabecular bone explants were cultured in a bioreactor for 48 h and subsequently loaded twice a day for one day or 3 days. RNA was isolated and sequenced, and the Tuxedo suite was used to identify differentially expressed genes and pathway analysis was conducted using Ingenuity Pathway Analysis (IPA).

Results:

There were about 4000 differentially expressed genes following in situ culture relative to fresh bone. One hundred six genes were differentially expressed between the loaded and non-loaded groups following one day of loading compared to 913 genes after 3 d of loading. Only 45 of these were coincident between the two time points, indicating an evolving transcriptome. Clustering and principal component analysis indicated differences between the loaded and non-loaded groups after 3 d of loading.

Discussion:

With sustained loading, there was a nine-fold increase in the number of differentially expressed genes, suggesting that osteocytes respond to loading through sequential activation of downstream genes in the same pathways. The differentially expressed genes were related to osteoarthritis, osteocyte, and chondrocyte signaling pathways. We noted that NFkB and TNF signaling are affected by early loading and this may drive downstream effects on the mechanobiological response. Moreover, these genes may regulate catabolic effects of mechanical disuse through their actions on pre-osteoclasts in the bone marrow niche.

Graphical abstract:

Introduction

Bone architecture and mass are actively adapted to the local mechanical environment through the coordinated activities of osteoblasts, osteoclasts, and osteocytes [1]. Osteoblasts form new bone matrix and initiate mineralization while osteoclasts demineralize and degrade existing bone. Osteocytes are terminally differentiated osteoblasts that have been entombed in the mineralized matrix. There, they sense deformation of the surrounding bone tissue and microdamage which alter their expression of signaling molecules that regulate osteoblast and osteoclast activity, thereby targeting resorption and formation to maintain bone mechanical properties [2–5]. This is particularly important to understand in trabecular bone, which is highly metabolically active and susceptible to decreased bone density with aging, osteoporosis, and metastatic disease.

Osteocytes transmit signals to osteoblasts and osteoclasts through cytokines that are transcriptionally and post-translationally regulated by mechanical loading. For example, sclerostin, which suppresses osteoblast differentiation by inhibiting the canonical Wnt signaling pathway, is transcriptionally down regulated and post-translationally degraded in response to mechanical loading in osteocytes [4, 6–10]. While these well-known signaling proteins and their molecular pathways have been widely studied, upstream transcriptional events may provide insight into cellular structures that initiate osteocyte signaling in response to mechanical loading and provide new targets to treat bone loss or unbalanced remodeling due to aging or disease.

Transcriptional profiling has identified several novel genes that are regulated by mechanical loading. C-X-C motif ligands were differentially expressed in osteocyte-like MLO-Y4 cells following 2 h of oscillatory fluid flow stimulation in comparison to non-loaded controls, while genes commonly associated with bone adaptation such as SOST and DKK1 were not affected [11]. These results contrast with in situ osteocyte gene transcription in mice [12]. The transcriptional response changes rapidly following loading. In an in vivo murine tibial loading model, the number of differentially regulated genes (DRGs) nearly doubled from 3 h to 24 h following a single loading bout, and half of the genes responding at 3 h were no longer differentially regulated at 24 h [12]. Similarly, there were only 19 DRGs after a single day of in vivo tibial loading in young mice compared to the non-loaded control limb, which increased to 552 and 1460 DRGs by days 3 and 5, respectively [13]. The response was attenuated in older mice, with no DRGs after one day of loading, 314 after three days, and 403 after five days. It is not clear if this represents a delayed response of osteocytes to loading or the time lag in the activation of downstream genes. The transcriptional response of osteocytes also differs in cancellous vs. cortical bone, both basally and following mechanical loading [12]. This may reflect either unique osteocyte phenotypes or the heterogeneity of deformation in trabecular bone, as osteocytes located in trabeculae subjected to low strains may have attenuated or no transcriptional response, thereby obscuring differences when the osteocytes are sequenced in bulk.

While both in vitro and in vivo studies have detected genes that are differentially expressed in osteocytes following mechanical stimulation, it is difficult to fully isolate the effect of mechanical loading in osteocytes [11, 13, 14]. Osteocyte cell lines cultured in monolayer may respond to loading differently than in situ, as they only have focal adhesions on their basal side while shear stress is applied to a free surface, resulting in deformation of the cell membrane near the nucleus and potentially deflection of the primary cilia [15]. In contrast, the primary mode of osteocyte stimulation in situ is believed to occur along the processes within canaliculi [16–18]. This may explain why cells in monolayer culture require 3–5% strain to affect cytosolic calcium mobilization [19], which is more than 10 times the typical strain seen in cancellous bone during activities of daily living [20]. In vivo studies retain the appropriate cell context, but the loading is superposed on the animal’s daily activities, which may confound the interpretation of mechanical response. The external loading may also affect muscle and soft tissue inflammation, which could result in crosstalk that affects osteocyte gene expression, and the stress associated with handling and anesthesia may also affect circulating factors that interact with osteocytes. Indeed, gene expression was affected in the contralateral limbs of mice subjected to in vivo tibial loading relative to sham controls [12].

The response of osteocytes to mechanical loading depends on the context – even between cortical and trabecular bone – and on the loading level. There are differences in the pathways that regulate bone adaptation between cortical and trabecular bone. For example, signaling of IL-6 family cytokines in osteocytes controls bone formation in opposite ways on trabecular and periosteal surfaces [21]. It is particularly important to study bone in a large animal model in order to ensure that the biomechanics and biology are relevant to humans. In particular, porcine bone exhibits homology with humans in terms of biomechanics in bone tissue.

The response of osteocytes in trabecular bone to physiological loads is essential to understand normal bone health, as bone diseases preferentially affect the more actively remodeling trabecular bone. The goal of this study was to quantify the transcriptional response of osteocytes in trabecular bone solely due to mechanical loading of the bone within a physiological range. Specifically, we 1) quantified transcriptional changes in osteocytes in trabecular bone in bioreactor culture in comparison to the native state; 2) quantified the effects of mechanical loading on the transcriptome following short- and multi-day loading; and 3) identified and compared signaling pathways that were activated following short- and multi-day loading.

Methods:

Cervical spines from 8 female pigs around the age of 9 months were obtained from a local abattoir (Martin’s Custom Butchering, Wakarusa, IN) within one hour of slaughter. Male pigs were not included, because they are typically castrated at a young age, which may have unknown effects on bone physiology. Two cylindrical trabecular bone explants, 8 mm in diameter, were excised from each of the C3, C4, and C5 vertebrae using a diamond coring bit under constant irrigation with cold saline supplemented with 3% antibiotic/antimycotic (10,000 units/mL of penicillin, 10,000 μg/mL of streptomycin, and 25 μg/mL of Amphotericin B, Corning Inc.). The cores were cut parallel to the axis of the vertebral body – which is the primary loading axis of the bone in the body and aligned with the principal trabecular orientation – taking care to avoid the growth plates. Both ends of the explants were cut flat and parallel using a diamond wafering saw to a final length of 10 mm.

Prior to culture, all explants were imaged using micro-computed tomography to capture the geometry for finite element analysis. Images were acquired at 70 kVp and 114 mA current and 200 ms integration time. Images were Gaussian filtered with a variance of 0.8 and support 2.0 and a threshold value of 210/1000, which corresponds to a mineral density of 252 mg HA/cm³. The voxel size was 0.02 mm.

The prepared explants were cultured in a custom compression bioreactor. Samples were bathed in high glucose (4.5 g/L) DMEM (Corning) with 10% FBS (Hyclone), and 1% AB/AM (Corning) circulating at a rate of 45 mL/hr. Samples were equilibrated for 48 h with no load. It is unclear whether osteocytes have a transcriptomic memory of their loading history. However, the period of equilibration was intended to allow the osteocytes to reach a baseline state with no bone strain.

After equilibration, six explants were subjected to 1200 cycles of compressive loading using an electromagnetic load frame (TA Instruments, Electroforce 5500, New Castle). A rest inserted triangle wave from 10 N to 200 N (0.20 to 3.98 MPa) at 4 Hz was applied in two 5-minute bouts with 1 h between bouts (Fig. 1). Based on the average volume fraction of porcine vertebral trabecular bone from previous studies and published regression models [22], we expected deformations of 2100 μɛ. Six control explants were cultured for 3 d without loading.

Fig. 1.

a) Bone explants were prepared from the cervical vertebra of young female pigs. b) Samples were cultured in a bioreactor. c) Two groups of six explants were subjected to mechanical loading while six served as non-loaded cultured controls. Six additional samples were not cultured or loaded (day 0 controls). d) Loaded explants were subjected to 1200 cycles of a rest inserted triangle wave of compressive loading from 10 N to 200 N at 4 Hz [20]. e) Explants were loaded twice daily on day 3 and day 4 with a one-hour rest period in between and one time on day 5. RNA was isolated from the explants 6 h after the final loading bout.

For the multi-day loading study, six explants were cultured for five days subject to two bouts of loading on day 3 and day 4 and one bout of loading on day 5. Six additional control explants were cultured 5 d without loading. The samples cultured for 3 d were taken from four pigs (Supplemental Table 1) and those for the multi-day loading study from the remaining four pigs.

RNA was isolated from all samples 6 h after the final load. RNA was also isolated from six explants on the day of slaughter to quantify changes in gene expression due solely to bioreactor culture. These explants came from the same four pigs as the samples that were loaded for 1 d.

Explants were not paired across groups because some contained portions of the growth plate and poor-quality RNA (RIN < 6.4) required some to be excluded (Supplemental Table 1). However, each group had at least one sample from each animal and samples from each of C3, C4, and C5. We do not expect differences in osteocyte behavior between the adjacent vertebrae where the trabecular structure is similar. We did not adjust the statistics for multiple samples from each animal, because the groups were not balanced between animals and treated each explant as an independent replicate.

RNA Isolation and Sequencing

RNA was isolated from osteocytes by first removing the marrow from the explants followed by gradient separation and purification [12, 23, 24]. Briefly, the explants were warmed to 37° C and centrifuged three times at 13000xg for 7 min, with incubation at 37° C in trypsin for 7 min in between each centrifugation. The removal of the marrow was verified by H&E histology on a separate group of 2 samples (Supplemental Fig. S1). After marrow removal, the sample was immediately snap frozen in liquid nitrogen, crushed using a mortar and pestle, placed in 10 mL of Trizol^™ Reagent (Invitrogen), homogenized with a syringe and needle, and frozen at −80° C until RNA was isolated. The samples were later thawed, and 2 mL of chloroform was added. The solution was centrifuged at 12000xg separating it into a clear aqueous phase, a middle interphase, and a pink organic phase. The clear aqueous phase containing the RNA was carefully removed using a pipette and purified using RNEasy kit (Qiagen, Germantown, PA) according to the manufacturer’s instructions. The quality of the purified RNA was measured on an Agilent 2100 Bioanalyzer as well as NanoDrop. RNA samples with an RIN greater than 6.4 [13] were submitted for paired-end sequencing on an Illumina NextSeq 550 with a read depth of 30 million reads per sample. Samples with an RIN less than 6.4 were not sequenced. As such, while each group had the same number of explants from each animal, we were not able to match by both animal and vertebral body.

RNA-Seq Data Analysis

The library was processed using the Tuxedo suite to identify DRGs [25]. Briefly, the sequences were trimmed using Trimmomatic to remove the Illumina adapters and bases from the sample that did not meet the quality standard. The sus scrofa genome (University of California Santa Cruz Genomics Institute) was indexed using Bowtie, and reads were aligned to the genome. Exonexon splice junctions were identified using TopHat. The transcriptome was assembled from the aligned reads using Cuffliinks and merged with the files used for alignment using Cuffmerge. Finally, Cuffdiff, a program in R, was used to identify DRGs. Differentially expressed genes with p-values less than 5×10⁻³ were considered statistically significant [26–29]. Principal component analysis was performed in R. Heatmaps of the z-score, used to normalize the expression for each gene, were also created in R.

Differential expression of four genes was verified by qRT-PCR at each time point. Briefly, custom primers were designed using IDT’s PrimerQuest tool. RNA was reverse transcribed to cDNA using the High-Capacity cDNA Reverse Transcription Kit with RNase Inhibitor (Applied Biosystems, Waltham, MA). Quantitative RT-PCR was carried out using iTaq Universal SYBR Green Supermix (Bio-Rad Laboratories, Inc.) on a CFX Connect Real-Time PCR Detection System (Bio-Rad). Expression was normalized to ACTB (Supplemental Fig. S2).

Ingenuity Pathway Analysis (Qiagen, Hilden, Germany) was used to identify pathways that were affected by culture and mechanical loading. Differentially expressed genes and their corresponding fold change information was loaded into the IPA software. The analysis identified pathways that were differentially regulated, the genes within the pathway that were affected, the p-value, and the activation z-score which indicated whether the pathway was activated or deactivated.

Finite Element Modelling

Finite element models of each loaded explant were created from the micro-CT scans to quantify the strain in the bone tissue. Voxel-based meshes were created directly from the thresholded volume. The bone was modeled as an isotropic linear elastic material with a modulus of 12 GPa and Poisson’s ratio of 0.3. The average peak displacement measured by the loadframe during loading on day 1 was corrected for machine compliance and converted to strain based on the sample height. A compressive displacement boundary condition was applied to the superior surface of the model, while the bottom was constrained vertically with free lateral expansion. The models were solved with our custom software [30] and the deviatoric strain was calculated per element.

Results

Mechanical loading

The applied apparent strain in the loaded samples was 3010 ± 550 μ-strain. The median tissue level strain was 1834 ± 205 μ-strain, and was nonnormally distributed (Fig. 2). The calculated peak reaction force for the samples when subjected to the measured strain was 215 ± 33 N, which was not significantly different from the applied load of 200 N. The structure of the bone was not significantly different between the groups (Table 1).

Fig. 2.

a) Example deviatoric strain distribution. The tissue level strains ranged from 0 – 5000 μ-strain with a median of 1834 ± 205 μ-strain. b) The histogram shows the strain distribution, with red bars for explants loaded for one day and blue bars for explants loaded for three days. Most of the tissue was subjected to strains below 2000 μ-strain, and the distributions were similar for all samples. c) The fraction of tissue by volume with deviatoric strain exceeding a range of magnitudes. (Error bars are 1 std. dev.)

Table 1:

Trabecular architecture of the explants (mean ± S.D., N=6 per group). The architecture did not differ between groups (p > 0.05).

	One day		Three day
	Loaded	Non-loaded	Loaded	Non-loaded
BV/TV (−)	0.239 ± 0.058	0.232 ± 0.073	0.220 ± 0.052	0.247 ± 0.056
Tb.Th. (mm)	0.121 ± 0.017	0.124 ± 0.023	0.118 ± 0.016	0.123 ± 0.014
Tb.N. (mm−1)	2.054 ± 0.252	1.945 ± 0.146	1.954 ± 0.280	2.070 ± 0.175
Tb.Sp. (mm)	0.458 ± 0.057	0.485 ± 0.042	0.491 ± 0.068	0.458 ± 0.046
SMI (−)	0.534 ± 0.365	0.577 ± 0.535	0.647 ± 0.308	0.436 ± 0.508

BV/TV=Bone volume per total volume; Tb.Th.=Trabecular thickness; Tb.N.=Trabecular number; Tb.Sp.=Trabecular spacing; SMI=Structural Model Index

RNA-Seq

We detected over 75,000 unique transcripts coding for 18,899 unique proteins along with 9,789 non-protein coding genes in the isolated RNA. Several CA2+ binding proteins belonging to the S100 family were highly expressed. Other highly expressed genes included the C-X-C motif ligands CXCL2, a negative regulator of osteoblast differentiation, and CXCL8 which stimulates bone resorption. ACTB, a housekeeping gene that codes for β-actin, a protein that makes up the cytoskeleton of cells, was also highly expressed. Several matrix metalloproteinase genes were highly expressed including MMP9, MMP13 and MMP14, which are essential for the maintenance of bone homeostasis [31–33].

Effects of Culture

We compared the gene expression in osteocytes from both loaded and non-loaded trabecular explants cultured for three days to uncultured explants. There were 3568 and 4055 DRGs between day 0 and non-loaded control and following 1 d loading, respectively, with 3044 genes in common between the groups (Fig. 3). Day 0 samples clustered distinctly from the cultured samples in principal component analysis (PCA) using unsupervised clustering. Similar differences were seen at day 5, following 3 d of loading (Supplemental Fig. S3).

Fig. 3.

a) Day 0 compared to cultured conditions. RNA isolated from Day 0 samples were compared to the loaded and non-loaded samples harvested on day 3. Red dots signify genes considered statistically significant (p < 0.005) b) 4055 genes were differentially expressed between loaded samples and day 0 controls; 3568 genes were differentially expressed between non-loaded samples and day 0 controls. c) Principal component analysis showed clustering of Day 0 vs. cultured samples. d) There were 3044 genes common between the loaded and non-loaded cases, suggesting their differential expression is due solely to the culture conditions and not loading. e-g) Top pathways affected in fresh bone relative to cultured and loaded bone (e); cultured and non-loaded bone (f) and fresh bone relative to all cultured bone (g). The horizontal axis indicates the p-value, the size of the circle is proportional to the number of DRGs, and the color indicates the IPA z-score, which is positive for activated pathways and negative for deactivated pathways. The CREB Signaling, Osteoarthritis, Pulmonary Fibrosis, Axonal Guidance, Cardiac Hypertrophy, and Hepatic Fibrosis pathways were differentially expressed in all groups. With the exception of Axonal Guidance, pathways had lower z-score magnitudes (less difference from Day 0) and fewer associated DRGs in the loaded group compared to fresh bone than bone cultured without loading compared to fresh bone.

Pathway analysis identified 103 significantly regulated pathways associated with the DRGs. Among the most significantly differentially regulated pathways, the Osteoarthritis Pathway was downregulated in day 0 bone relative to cultured bone while the CREB signaling, and Hepatic Fibrosis pathways were both upregulated in culture regardless of loading (Fig. 3).

Effects of short-term loading

We compared the gene expression of osteocytes in loaded vs. non-loaded explants after 2 d of non-loaded culture followed by two bouts of loading. There were 29 downregulated genes and 77 upregulated genes (Fig. 4). Of these genes, 19 were unique to the loaded vs. non-loaded group while the rest were also differentially expressed between the cultured and day 0 groups. The 30 genes that differed between day 0 and the unloaded samples were rescued by in situ loading to become similar to the uncultured bone. Fourteen additional genes had smaller differences relative to baseline explants, yet still significantly different. The expression of the remaining 62 genes changed more than the difference between day 0 and unloaded, cultured bone. While thousands of genes remained different from day 0 bone, suggesting that mechanical loading was not sufficient to fully rescue the in-vivo osteocyte behavior, but at least some of the differences in gene expression in cultured bone can be attributed to lack of loading.

Fig. 4.

a) Differential gene expression following 1 bout of loading. Red symbols signify genes considered statistically significant (p < 0.005). b) There were 106 DRGs between the loaded and non-loaded groups. c) Principal component analysis shows separation between the loaded and non-loaded groups. d) One cluster of genes was downregulated by mechanical loading (top portion of heat map), while the majority were upregulated. One of the loaded samples (Loaded 2) exhibited much greater upregulation of these genes and clustered distinctly from all of the remaining samples. e) The top five significantly differentially regulated pathways (Ingenuity Pathway Analysis). The osteoarthritis pathway was deactivated by loading, while other pathways had insufficient information to identify activation or deactivation. The horizontal axis indicates the p-value, the size of the circle is proportional to the number of DRGs, and the color indicates the IPA z-score, which is positive for activated pathways and negative for deactivated pathways.

Principal component analysis and clustering analysis indicated substantial overlap between the loaded and non-loaded groups. Five pathways were significantly affected by loading including the Osteoarthritis and the Hepatic Fibrosis/Hepatic Stellate Cell Activation pathways, as well as pathways related to Rheumatoid Arthritis (Table 2). Several genes in these pathways directly affect bone adaptation. Within the Osteoarthritis pathway, differentially expressed genes included FRZB, IHH, DKK1, CNMD, SFRP and GDF5. DKK1, SFRP, and FRZB are inhibitors of the canonical WNT pathway, which is a major regulator of osteoblast differentiation [33]. In the Hepatic Fibrosis/Hepatic Stellate Cell Activation Pathway, the genes affected included collagen 9, CD14, and IL-1R2.

Table 2:

Pathways that were differentially regulated by mechanical loading after one and three days of loading, along with the key genes in the pathway.

One day		Three days
Upregulated	Downregulated	Upregulated	Downregulated
Osteoarthritis Pathway (deactivated)
CNMD, FRZB, GDF5, IHH, MATN3, MMP3, TIMP3, COL2A1,	DKK1, IL- 1R	ADAMTS5, SMAD1/5/8, VEGF, WNT11, WNT5	ACAN, Adipokine, ALP, CEBPB, CNMD, COL2A1, ELF3, FGFR3, FRZB, Frizzled, IL-1R, Integrin, MATN3, MMP12, NAMPT, NfκB, NKX3–2, RBPJ, SOX9, SMAD2/3, TNFα, Caspase, SOST
Osteoclasts, Osteoblasts, and Chondrocytes in Rheumatoid Arthritis (indeterminate)
MMP3, SFRP	DKK1, IL- 1R	IL-6, p38, MAPK	CCR1, CCR5, CD14, IKK, IL-1R, IL1, IL-10, IL-10RA/B, TNFα, NfκB,
Hepatic Fibrosis/Hepatic Stellate Cell Activation (indeterminate)
Il-1R2 CD14	Collagen	IGF1, PDGF, VEGF, IL-6, VEGFR, PDGFRB, IGFBP5, EDNRA, Myosin	TNFα, Collagen, SMAD2/3, CD14, IL- 10, CCL5, CCL2, ICAM1, NfκB,

Effects of multi-day culture

Over 900 genes were differentially expressed between loaded and control groups after 3 d of loading, of which only 45 were in common with those differentially expressed after 1 d of loading (Fig. 5). The loaded group was clustered distinctly from the non-loaded group. Sixty-eight pathways were differentially regulated (p < 10⁻³), including the Osteoarthritis, Hepatic Fibrosis/Hepatic Stellate Cell Activation, and Rheumatoid Arthritis pathways (Table 2). In contrast to 1 d loading, more genes – including many downstream genes – were differentially expressed in each pathway, while primarily upstream genes were affected following short-term loading. For example, in the Hepatic Fibrosis/Hepatic Stellate Cell activation pathway, CD14 and IL-1R2 were downregulated after 1 d of loading and the downstream genes Nf-kB, ICAM1, and CCL5 were downregulated after 3 d. Conversely, several DRGs were no longer differentially expressed after 3 d of loading, including DKK1 and IHH (Table 3).

Fig. 5.

a) Differentially expressed genes between loaded and non-loaded samples after 3 d of loading. Red symbols signify genes considered statistically significant (p < 0.005) between conditions. b) 913 genes were differentially regulated between loaded and non-loaded groups. Forty-five of these genes were also differentially expressed after 1 d of loading. c) Principal component analysis showed separation between the loaded and non-loaded groups. d) Heatmap of z-score. Unsupervised clustering indicated that the loaded samples are clustered. e) Following 3 d of loading, the hepatic fibrosis signaling, osteoarthritis, and wound healing pathways were deactivated (z-score < 0) while the pulmonary fibrosis pathway was activated relative to non-loaded explants (z-score > 0). The horizontal axis indicates the p-value, the size of the circle is proportional to the number of DRGs in the pathway, and the color indicates the IPA z-score, which is positive for activated pathways and negative for deactivated pathways.

Table 3:

DRGs with either > 2X or <0.5X expression in loaded vs. unloaded samples after 1 or 3 d loading.

	One Day	Both	Three Days
Upregulated	ABCC8, CHRDL2, CLEC3A, COL27A1, CYP24A1, CYTL1, EXTL1, FOXA3, GDF5, GJB3, HS3ST3A1, IDO2, IHH, LGI2, MMP3, NRN1, SNORA72, SNORC	WNT11	ADAMTS1, ADAMTS12, ADAMTS2, ADAMTS5, ADAMTS9, CDH5, COL15A1, COL18A1, COL5A3, COL8A1, FGD5, FRMD4A, IGF1, IGFBP2, IGFBP4, IGFBP5, IL33, IL6, LYVE1, MYO1B, NOS3, NOTCH3, NOTCH4, NOX4, NR2F1, PDGFB, PDGFRB, PTGFR, PTPN13, PTPRB, RPS6KA2, SEMA6B, SMAD1, SOX13, SOX17, SOX18, TGFB1I1, TGFBI, TRIM16, TRPM4, VEGFC, WNT5A
Down at 1 d Up at 3 d
CYP1A1
Up at 1 d Down at 3 d
B3GNT7, CNMD, COL11A2, COL9A1, CSPG4, ECRG4, FRZB, GFPT2, HAPLN1, KCNMA1, MATN3, MELTF, NIPAL1, PAX1, PTPRU, SHISA2, SLC7A2, SPINT1, TRIM63, TRPV4, XIRP1, ZNF385D
Downregulated	BANK1, BPI, DKK1, FDSCP	CCL8, CCR2, CD14, CD163, CHI3L2, CTSL, HNMT, IL1R2, MGP, MS4A7, SEMA4A, SERPINB2, XDH	ACAN, ALPL, CCL2, CCL22, CCL3L1, CCL5, CCR1, CCR2, CDH1, CNNM4, COL11A1, COL27A1, COL2A1, COL9A3, CX3CR1, CXCL2, FGFR3, FGR, HAPLN1, ICAM1, IL10, IL10RA, IL10RB, IL1A, IL1RAP, IL22RA2, IL7R, MAT2B, MATN1, MYO7A, NFIL3, NFKB2, NFKBIA, NFKBIE, PTAFR, PTPN22, PTPRC, PTPRCAP, RPS6KA1, RUNX3, SMAD3, SOST, SOX9, TGFA, TNF, TNFAIP3, TNFAIP6, TNFAIP8, TNFSF15, TNFSF8, TRIM63, WEE1

Discussion

We sought to identify osteocyte genes that are differentially regulated by physiological levels of mechanical loading in trabecular bone. We used a model of complete unloading followed by a single day or multiple days of mechanical loading in short bouts. Our in situ culture model caused differential regulation of thousands of genes involved in several key osteocyte signaling pathways compared to bone from a live animal. While many of the same genes were differentially regulated in both non-loaded and loaded bone compared to fresh bone, over 1500 were affected in only one condition, suggesting that baseline mechanical loading played a role in the differential gene expression. The DRGs were associated with Hepatic Fibrosis/Hepatic Stellate Cell Activation, Osteoarthritis, CREB Signaling, and the STAT3 pathways. All of these were down regulated in fresh bone relative to cultured bone, except for the osteoarthritis pathway. The osteoarthritis pathway includes the Wnt signaling pathway, which is associated with mechanotransduction, and activation of this pathway in the fresh bone relative to non-loaded cultured bone suggests that mechanical loading plays a baseline role in maintaining normal bone physiology. Even 3 d of loading did not fully rescue normal gene expression levels (Supplemental Fig. S3).

Our study is unique in that we investigated the transcriptomic response of trabecular bone of a large animal. We were able to use physiological loading levels for trabecular bone by removing the effects of background activities and muscle loading for multiple days prior to loading. However, there are also important limitations to consider. First, we used porcine cervical vertebrae. The pig is not a common model for osteoporosis, but the genome has been mapped and pigs are used to study a range of other diseases. The isolation of the bone and in situ culture clearly affect the normal gene expression, and several important interactions with hormones are lost in this model. Although the bone marrow was intact and signaling molecules from the immune and hematopoietic cells in the marrow may still affect the osteocytes, factors such as calcium regulating hormones, systemic growth regulators, and local growth factors affect bone growth and remodeling, and the lack of these circulating factors may have significantly affected osteocyte gene expression [34, 35]. While the intent of the model is to isolate the mechanical response from other factors, osteocyte gene response may be affected by such circulating factors. These effects could be quantified by incorporating selected cytokines or hormones of interest into the media to quantify their roles. Finally, trypsinization and centrifugation of the bones may have affected the gene expression [36, 37]. This is a necessary step to ensure that our cell population is not contaminated by marrow cells, and is common to study osteocyte gene expression [12, 38].

The response of osteocytes to a single day of mechanical loading was limited. However, expression of several key genes that code for inhibitors of bone formation was downregulated. Genes involved in the Wnt/β-catenin signaling pathway, including FRZB and DKK1, exhibited inconsistent regulation after a single loading bout. FRZB, which binds competitively to Wnt proteins to downregulate Wnt signaling was upregulated. In contrast, DKK1, another antagonist of the Wnt/β-catenin signaling pathway that acts by binding to the LRP6 co-receptor preventing it from activating the WNT signaling pathway, was downregulated following loading.

Three days of loading increased the number of DRGs nine-fold relative to a single loading bout. The larger number of DRGS were associated with a concomitant increase in the number of affected pathways. However, the most significantly affected pathways were the same as those affected by short-term loading, with additional DRGs downstream of those identified for the same pathways following one bout of loading that were consistent with predictions for the pathways. For example, increased expression of CD14 and IL-1R2 after one loading bout was expected to affect the expression of downstream genes NfkB, CCL5, CCL2, and ICAM1, which were all differentially regulated after 3 d of loading. This suggests that sustained loading allows for the progression of the pathway. However, the timing and the durability of the downstream pathways is not clear, as we only looked at the snapshot of RNA differentially expressed at 6 h following the final loading event. It would be interesting to study the evolution of the transcriptome by looking at changes in RNA expression in a time series after a single load [39].

The major pathways affected by loading in our study were the Osteoarthritis, Hepatic Fibrosis/Hepatic Stellate Cell Activation pathways, and pathways associated with rheumatoid arthritis. While most pathways had insufficient information to determine if they were upregulated or downregulated, the Osteoarthritis pathway was consistently downregulated by mechanical loading. While none of the pathways directly relate to bone adaptation, they all involve genes that are involved in bone mechanoadaptation, including the growth factors VEGF, TGFα, IGF1, and PDGFβ. VEGF, which is upregulated after repeated loading, causes osteoblasts to migrate to ossification sites [40]. IGF1 is upregulated by loading, and is essential for the development of bone [41]. Furthermore, a threshold concentration of IGF1 is required to maintain bone homeostasis [42]. TGFα, which regulates differentiation of chondrocytes to an osteogenic phenotype during the process of endochondral ossification was downregulated by loading [43], and PDGFβ increases osteogenic differentiation and promotes mineral deposition [44, 45]. IL-6 promotes osteoclastogenesis. TNF plays a key role in bone adaptation by inhibiting bone formation by suppressing both osteoblast activity and stimulating osteoclast proliferation, acting in synergy with RANKL to promote osteoclast differentiation [46]. As such, the transcriptomic changes due to mechanical loading of cancellous bone has significant overlap with existing reported pathways in the literature.

While the majority of the DRGs following a single day of loading were no longer differentially expressed after sustained loading, there were several genes in common between the two time points, including CNMD, SERPINB2, FRZB, and IL-1R2. This suggests that these genes are directly affected by mechanical loading and drive the multi-day response, while other genes are upregulated after sustained expression of these early genes. Interestingly, DKK1, which inhibits the canonical Wnt pathway by binding to LRP6, was downregulated after 1 d of loading but was not differentially expressed following 3 d of loading.

SOST, which inhibits canonical Wnt signaling by binding both LRP5 and LRP6, was not differentially expressed after 1 d of loading but was downregulated following 3 d of loading. An interesting hypothesis is that the initial autocrine effects on the Wnt pathway in osteocytes decrease SOST expression. Sclerostin protein is also downregulated post-translationally in mechanically stimulated osteocytes, prior to differential gene expression [10]. In fact, that study showed that sclerostin protein is degraded within minutes, suggesting that transcription of sclerostin is not the mechanism by which the protein expression changes in the short term. Other transcriptional studies similarly found that SOST was not differentially expressed following loading [11, 13, 31, 47].

A recently published study that addressed the temporal response of human osteocytes in response to mechanical loading in situ provides additional insight into our results [47]. Similar to our study, they isolated live bone, applied controlled loads, and quantified differential gene expression. They found a small number of DRGs immediately following loading which increased after 6 h and began to decrease at 24 h. Of the differentially expressed genes, only FAF1– a negative regulator of WNT/β-catenin signaling through the promotion of β-catenin degradation [48] – was common between our study and theirs. They did not see the differential expression of known bone mechanotransduction genes, such as SOST or WNT ligands following mechanical loading. There could be a few reasons for this. First, their study used human cortical bone subjected to cyclic three-point bending with maximum strains of 2000 or 8000 μ-strain. In contrast, the apparent strain in our samples was 3057 μ-strain and the average tissue strain was 1907 μ-strain. As their results show, the transcriptome differs between physiological and superphysiological loading. After only one bout of loading, the number of DRGs was relatively constant for 24 h, with only 7 genes differentially expressed between the three timepoints. We found that loading for three consecutive days the number of DRGs increased nearly nine-fold. As such, the osteocyte transcriptome changes rapidly following loading, but many more genes – particularly genes associated with mechanobiological signaling – respond after multiple bouts of loading.

Our results complement findings from tibial loading in young and old mice [13]. There were only 19 differentially expressed genes after 1 d of loading in young mice, which increased to 500 after 3 d, and more than 1400 after 5 d, with many DRGs unique to each time point. For example, 222 genes that were differentially expressed after 3 d were no longer differentially expressed after 5 d, which is consistent with the differences we saw following 1 and 3 d of loading (Supplemental Fig. S4). There were many common DRGs between the two studies. For example, after three days of loading, several Wnt ligands, Serpin, collagen, and fibroblast growth factor were differentially expressed in both their study and ours. The Wnt signaling pathway and the axonal guidance signaling pathway were affected in both our bioreactor model and in mice. However, there were several pathways unique to each study. The hepatic fibrosis pathway was upregulated in our study but not significantly affected in mice, and the angiogenesis pathway was upregulated in the mouse model but was not significantly affected in in situ culture. The latter may reflect that endothelial cells were present in the long bones of the mice, while our trabecular bone would have contained few blood vessels after the marrow was removed.

In another study the number of DRGs increased with sustained loading for 1, 5 and 7 d, with different genes at early and late time points in agreement with our findings [31]. More genes were differentially expressed in the mid-diaphysis, where the strain was approximately 2000 microstrain, similar to the mean value in our trabecular bone models. The WNT pathway was significantly differentially regulated in both their and our data sets, although different WNT ligands were affected. Several collagen related genes were also differentially expressed in both studies.

In contrast to our results, only 55 genes were differentially expressed after 2 h of fluid shear stress stimulation of osteocyte-like murine cells [11]. Moreover, there are several notable differences in the experimental methods that might explain this discrepancy. First, fluid flow across the apical surface of the cells cultured in 2-D causes large deformations of the cell membrane, which may transmit forces directly through cytoskeletal remodeling [49]. Our organ culture model may involve different signaling pathways as the fluid shear occurs primarily on the cell processes and there may be direct deformation of the cell body within the matrix, and this difference in the mechanism of deformation might affect the transcriptional response of osteocytes [2]. Indeed, we saw twice as many DRGs in our experiment. Second, they used a mouse osteocyte cell line, while we studied primary osteocytes in situ in trabecular bone. Although the cell line is well characterized and commonly used to study bone mechanotransduction, the transcriptional response may differ from primary osteocytes or bones of large animals [50]. However, human cells may differ from both species, and experiments on human bone should be pursued [51].

Many of the DRGs following mechanical loading are associated with mechanotransduction. However, pathway analysis associated these gene combinations with pathways that are not associated with the musculoskeletal system, such as the hepatic fibrosis pathway. As such, we used the Ingenuity Pathway Analysis database to propose a bone mechanotransduction pathway (Fig. 6). Two genes appear to be central to the response of osteocytes: TNF and NFkB. It has been theorized that TNF forms a positive feedback loop with NFkB, and inhibition of this feedback loop by mechanical loading may play an important role in initiating adaptive bone remodeling [52]. TNF, which suppresses osteoblast activity and promotes osteoclast proliferation [46], is down regulated by mechanical loading. NFkB is activated by many DRGs that are downregulated by mechanical loading, including TNF, CD14, and IL1R. In turn, NFkB is a transcription factor for other DRGs such as ICAM, IL-6, and binds to elements on the SOST promotor region [52–54]. As such, downregulation of NFkB results in the downregulation of SOST and inflammatory molecules [55]. At the same time, IL-6 – which is upregulated by loading – binds to IL6R and the gp130 coreceptor on osteoclasts, which enhances osteoblast activity while also increasing bone formation in trabecular bone via the coupling factor cardiotrophin 1 [56]. IL-6/gp130 signaling in osteoblasts and osteocytes plays a key role in normal bone formation [57].

Fig. 6.

Hypothesized adaptive bone remodeling pathway. Transcripts indicated on the Unloaded region of the diagram were more highly expressed in the unloaded samples and downregulated in loaded explants. Transcripts indicated in the Loaded region were upregulated in the loaded explants. Transcripts with a thick border were differentially expressed after both 1 d and 3 d of loading, while others were differentially expressed only following 3 d of loading. Loading initially activated WNT signaling through downregulation of DKK1 and upregulation of WNT, followed by sustained signaling by later downregulation of SOST and FRZB. Downregulation of TNF and NFkB signaling is associated with SOST and IL6 gene transcription.

It is noteworthy that a number of genes, including TNF and NFkB, that are more highly expressed in unloaded compared to loaded bone also affect osteoclast maturation and activity. As such, unloading not only suppresses bone formation, but also potentially enhances catabolic activity. Taken together with differences in the transcriptome between fresh bone and cultured bone, which are partially rescued by mechanical loading, osteocytes play a role in bone catabolism and adaptation by enhancing osteoclastogenesis and bone resorption in conjunction with blocking osteoblastogenesis when mechanical signaling is decreased.

Highlights.

• RNA from osteocytes of porcine trabecular bone explants that had been subjected to compression in a bioreactor for 1 or 3 d was sequenced

• 106 genes were differentially expressed in loaded vs. control explants after 1 d of loading while 913 genes were differentially expressed after 3 d

• Gene expression differentiated loaded from unloaded explants after 3 d of loading

• Loading affected pathways associated with multiple inflammatory molecules and caspase

Data availability:

The RNA-Seq data is available on the NCBI Sequence Read Archive (SRA) submission 13889520