Circulating fibronectin affects bone matrix, whereas osteoblast fibronectin modulates osteoblast function

Abstract

The bone matrix is composed mostly of collagen, but the initial and continuous presence of fibronectin was found to be crucial for collagen matrix integrity in vitro. It has been assumed that osteoblasts produce the fibronectin required for bone matrix formation. Using transgenic mice, we conditionally deleted fibronectin in the osteoblasts and in the liver using the cre-loxP system. We also used mice with mutated fibronectin and conditionally deleted β₁-integrin in osteoblasts to identify the receptor involved in fibronectin effects on osteoblasts. Conditional deletion of fibronectin in the differentiating osteoblasts [using the 2.3 kb collagen-α1(I) promoter] failed to show a decrease in fibronectin amount in the bone matrix despite evidence of successful deletion. Using these mice we established that osteoblast-derived fibronectin solely affects osteoblast function. This effect was not mediated by integrins that bind to the RGD motif. Conditional deletion of fibronectin in the liver showed a marked decrease in fibronectin content in the matrix associated with decreased mineral-to-matrix ratio and changed biomechanical properties but had no effect on osteoblasts or osteoclasts. In conclusion, osteoblast fibronectin affects osteoblasts function. This does not seem to be mediated by the RGD motif on fibronectin. In contrast, liver-derived fibronectin affects bone matrix properties without affecting osteoblast or osteoclast function. A novel role for liver-derived circulating fibronectin thus was defined and delineated from that of locally produced fibronectin. © 2010 American Society for Bone and Mineral Research

Introduction

The extracellular matrix of bone is the scaffold on which mineral is deposited.1 The osteoblasts lay down the extracellular matrix and then mineralize it, which results in new bone formation.2, 3 The most abundant protein in the extracellular matrix is collagen type I, which comprises up to 90% of the proteinaceous matrix.2 Collagen type I is secreted as triple helices, which then assemble to form fibrils and fibers.2, 4 This is not, as originally thought, a self-assembly process. In vitro studies had shown a key role for fibronectin (FN), a protein found in small amounts in bone matrix, in the assembly of collagen.5 Fibroblasts unable to produce FN and cultured in the absence of FN failed to assemble collagen unless FN was added to the medium.4 Furthermore, it was shown that not only is FN needed during the initial steps of the polymerization of collagen, but the continuous presence of FN is also required for the integrity of the matrix.6 The importance of FN in collagen matrix assembly suggests that its absence results in matrix defects resembling those seen in osteogenesis imperfecta. This disease is caused by either quantitative or qualitative abnormalities in collagen type I resulting in abnormal matrix formation, and its severity ranges from stillbirth owing to multiple fractures to osteoporosis in adulthood.7

Osteoblasts produce FN during proliferation and differentiation at the same time that they produce collagen type I, which implies that the osteoblasts are responsible for the provision of FN during active bone formation.5 The source of continuous FN provision will have to be solved differently. The most likely source of continuous FN provision would be the liver. In fact, proteins produced exclusively by the liver, such as albumin (69 kDa) and α₂-HS glycoprotein (48 kDa),8, 9 have been found incorporated in the bone matrix. This, however, would require the diffusion of plasma FN, a large molecule of 480 kDa, into the bone matrix in sufficient amounts. FN itself is required by the osteoblasts to form mineralized nodules in vitro.10–12 Taken together with the role of FN in collagen assembly, these findings suggest a complex role for this molecule in modulating bone.

FN is a dimer that contains one or more of the following three alternatively spliced domains [extra-domain-A (EDA), extra-domain-B (EDB), and the variable region]. FN classically has been viewed as two separate entities: (1) The soluble form is called plasma FN, is produced by hepatocytes in the liver and circulates in the bloodstream, and contains neither the EDA nor the EDB, but in one of its two chains the variable region is present in its complete length, and (2) the cellular form is produced by a variety of cells, presumably contains one or more of the alternatively spliced domains, and gets incorporated into the matrix.13, 14 Even though the soluble plasma FN always was used in the studies of collagen matrix assembly, it has always been assumed that the cellular forms of FN are the ones that are operative in the matrix in vivo.15 FNs can bind to up to 11 different integrins.16 Osteoblasts have been shown to express 6 of these: α4β1, α5β1, α8β1, αvβ1, αvβ3, and αvβ5.11, 17–20 It is currently unknown, however, which integrin(s) is the primary adhesion molecule for osteoblast binding to fibronectin.

FN knockout mice die in utero at embryonic day 8.5, underlining FN's importance in the formation of a normal scaffold of extracellular matrix in vivo.21 Therefore, the study of the role of FN in vivo has to proceed through its deletion in specific tissues. Introduction of two loxP sites in the FN gene has resulted in successful deletion of FN in a variety of cells22, 23 using promoters attached to the cre recombinase gene.

In this report we used six types of conditional knockout (cKO) mice to determine the role of FN originating from the osteoblasts and to delineate that from the effect of circulating FN.

Materials and Methods

Mice

Transgenic mice

Mice possessing a construct of the 2.3 kb proximal region of the collagen α1(I) promoter driving cre recombinase expression (Col-cre) were used to delete fibronectin (FN) and β₁-integrin in osteoblasts.24 Mice possessing a construct of the inducible Mx promoter or the albumin promoter driving cre recombinase expression were used to delete FN in hepatocytes, and induction of Mx was performed as described previously.23 Transgenic mice carrying loxP-flanked (floxed) FN (FN-fl/fl), enhanced green fluorescent protein (EGFP) transcribed when the cell is exposed to cre, floxed β₁-integrin (β₁-fl/fl), or an RGD to RGE mutation in FN have been reported.22, 23, 25, 26 Total FN was determined by ELISA using a mouse pFN standard (Dunn, Asbach, Germany).27 Four-week old mice were injected with BrdU 100 mg/kg 1 hour before euthanasia and used for the analysis of osteoblast proliferation and apoptosis. Osteocalcin was measured using ELISA (Biomedical Technologies Inc., Stoughton, MA, USA). Bones were lysed in 4 M guanidine-HCl, 10 mM EDTA, and 50 mM Tris, pH 7.4. Protein was evaluated with BCA (Thermo Fisher Scientific, Bonn, Germany).

Wild-type mice

Plasma FN was isolated as described previously.27 For the labeling experiments, adult mice were injected with alizarin complexone on day 0 (Fluka, Munich, Germany). Then 1 mg of protein was labeled with Oyster-500 (Denovo-Biolabels, Muenster, Germany) on three consecutive days (days 1 to 3) and injected on each of the 3 days intraperitoneally. On day 5, mice were euthanized. To examine the effect of unlabeled plasma FN injection, five mice were injected daily with 1 mg FN, and 10 controls received NaCl 0.9% for 10 days. The amount was increased in later experiments to 3 mg FN and the duration to 15 days.

The studies in mice were approved by the Animal Protection Committee of the University of Heidelberg.

Cell culture

Mouse calvarial osteoblasts were isolated and cultured as described previously27 using 10% FN-depleted fetal calf serum. Once confluent, mineralization was induced with 5 mM β-glycerophosphate, 50 µg/mL ascorbic acid, and 10 nM dexamethasone added with each medium change every 2 to 3 days.27 Polymerase chain reaction (PCR) for FN on genomic DNA was performed using the following primers: 5′-TCGCACCCGCTGCGCTGCA-3′ (sense) and 5′-ATTGTCAAAACAGCCAGCTGCGA-3′ (antisense), followed by nested primers 5′-AGGGTGTGAGCCGGACAACT-3′ (sense) and 5′-CAACTGACTTGGTGAGCCTGCA-3′ (antisense). Nodule formation was confirmed using von-Kossa staining.10 Reverse-transcription PCR was performed using the following primer pairs: EDB 5′-TCACTGACCTAAGCTTTGTTGATA-3′ (sense) and 5′-CGTTTGCTGTCAGTGTAGT-3′ (antisense), EDA 5′-ACATTGATCGCCCTAAAGGACT-3′ (sense) and 5′-CAGGGGCTGGCTCTCCATA-3′ (antisense), V120 5′-CCACACCCCAATCTTCATGGA-3′ (sense) and 5′- GATTGAGGCCCGGAACATGA-3′ (antisense), fibronectin 5′-GTCAGTGTCTCCAGTGTCTA-3′ (sense) and 5′-GGCTTGCTGGCCAATCAGT-3′ (antisense), osteocalcin 5′-ACAGACAAGTCCCACACAGCA-3′ (sense) and 5′-TAGCGCCGGAGTCTGTTCACT-3′ (antisense), and GAPDH 5′-CTGGCCAAGGTCATCCATGA-3′ (sense) and 5′-GTCAGATCCACGACGGACA-3′ (antisense). EGFP-expressing cells were sorted using fluorescence-activated cell sorting (FACS) Vantage (Becton Dickinson). FN isoforms were examined by ELISA.27 Lysates were immunoprecipitated with antibody on protein G–sepharose (GE, Healthcare, Munich, Germany).

Immunohistochemistry

Cells or bones were fixed in 4% paraformaldehyde. Bones were decalcified with 5% EDTA, pH 7.0. The following antibodies were used: mouse anti-cre recombinase, sheep anti-BrdU and rabbit anti-active caspase-3 (Abcam, Cambridge, UK), mouse anti-Cy3 (Dianova, Hamburg, Germany), rabbit anti-mouse FN (Millipore, Schwalbach, Germany), and goat anti-rabbit Alexa 555 antibody (Invitrogen, Karlsruhe, Germany). DAPI was used as a counterstain.

Mass spectrometry

Immunoprecipitated plasma FN and osteoblast FN obtained from conditioned medium of confluent cells cultured in the absence of fibronectin in the serum were analyzed at the core facility of the German Cancer Research Center (DKFZ). ESI-MS/MS data were acquired on an LTQ Orbitrap mass spectrometer coupled with a nanoHPLC system. Database search was performed against the NCBInr database using the Mascot search algorithm.

Histomorphometry

Mice were injected with 30 mg/kg calcein (Sigma-Aldrich, Munich, Germany) 10 and 3 days before euthanasia. Tibiae were fixed and embedded in methyl methacrylate. Sections were deplasticized and stained for Masson-Goldner with hematoxylin (Gill II, Carl Roth, Karlsruhe, Germany), acid fuchsin–ponceau xylidine, and phosphomolybdic acid–orange G to stain the cells and osteoid and light green to stain the mineralized matrix. Primary cancellous bone was defined as the 120 µm band below the growth plate. Cancellous bone was defined as the remaining trabecular area that extends down 2 mm. The same sections were used for dynamic and static histomorphometry. American Society for Bone and Mineral Research (ASBMR) nomenclature was used.28 The following measurements are mentioned: osteoid surface (OS), bone surface (BS), osteoblast number (Ob.N), osteoclast number (Oc.N), erosion surface (ES), adjusted apposition rate (Aj.AR = mineralizing surface × mineral apposition rate/osteoid surface, or MS × MAR/OS), and mineralization lag time (Mlt = osteoid thickness/Aj.AR), bone formation rate (BFR = MS × MAR/BS, mm²/mm/year). Proliferating osteoblasts were corrected to bone surface. Number of osteocyte lacunae was corrected to the area. ImageJ was used (Wayne Rasband, NIH).

Bone mineral density measurements

Bone mineral density (BMD) was measured using peripheral quantitative computer tomography (pQCT) with a XCT Research SA⁺ machine (Stratech Medizintechnik, Pforzheim, Germany). Trabecular bone density was examined at a site located at 7.5% of the bone length from the growth plate.

Infrared spectroscopy

Bone sections were placed on barium fluoride windows (Korth Kristalle, Altenholz-Kiels, Germany). Spectra were recorded using a Bruker IRScope II infrared microscope coupled with a Bruker 66vs FTIR spectrophotometer. The resolution used was 4 cm⁻¹, acquiring 1024 scans in the transmission mode. Nine sites were measured per mouse. Data were analyzed using Excel. Normalization and baseline corrections were performed. Ratios were evaluated for mineral to matrix (900 to 1200/1580 to 1720 cm⁻¹) and acid phosphate content (1117/900 to 1200 cm⁻¹).29

Microhardness

Microhardness was determined using a Vickers indenter (Buehler Micromet 5104, Buehler, Dusseldorf, Germany), applying 10 g over 10 seconds30 on polished embedded samples. Ten indentations separated by 2.5 times the size of one impression were measured per sample. Microhardness (expressed in kg/mm²) was calculated with the formula H_v = 1854.4 × P × d⁻², where P is the test load in grams and d is the mean length of the two diagonals of the impression in millimetres.

Transmission electron microscopy

Bones were fixed in 4% paraformaldehyde, decalcified, washed, transferred to 2.5% glutaraldehyde–0.05 M sodium cacodylate, and processed as described previously.31 Micrographs of cortical sections were taken with a Zeiss EM-10 electron microscope at 80 kV. The diameter of round collagen fibrils was measured using ImageJ. A total of 459 control and 754 cKO fibrils were analyzed.

Statistical analyses

Analyses were performed using SPSS (Version 14.0; SPSS, Inc., Chicago, IL, USA). Comparisons were performed using the Student's t test or Wilcoxon paired test as appropriate. For ease of presentation, males and females were combined. BMD and histomorphometric data were paired with littermates. Results are expressed as mean ± SEM.

Results

Deletion of fibronectin (FN) in the osteoblasts does not affect FN content in bone

Since osteoblasts lay down the extracellular matrix containing FN during bone formation, we deleted FN in the osteoblasts using cre driven by the 2.3 kb collagen-α1(I) promoter,24 which is active in differentiating osteoblasts and osteocytes in adult mice.32 Deletion of FN in osteoblasts of the osteoblast cKO mice (Col-cre/+–FN-fl/fl) (FN^−/−-OB-cKO) was confirmed in vitro. Cultured calvarial osteoblasts from FN^−/−-OB-cKO mice stained positively for cre and exhibited deletion of FN at both the DNA and the protein levels (Fig. 1A–C). Despite the apparent successful deletion of FN in differentiating osteoblasts, neither immunostaining for FN in bone sections nor Western blotting of lysates showed a difference between FN^−/−-OB-cKO and controls (CT) (see Fig. 1D and data not shown), suggesting that most of the FN in bone matrix originates from a different source.

Figure 1. (A) Cre staining of cultured osteoblasts. Positive staining was detected in the osteoblasts located in the nodules in cultures from fibronectin-osteoblast conditional knockout mice (FN^−/−-OB-cKO). DAPI staining showed the presence of nuclei inside and outside the nodules because of the production of cre in late but not in preosteoblasts (Inset). Bars represent 50 µm. (B) PCR of DNA obtained from FN^−/−-OB-cKO and control osteoblasts in culture confirming the deletion of the part of the FN gene between the two loxP sites in FN^−/−-OB-cKO mice (results verified by sequencing) but not in the control mice (CT). (C) Bar graph showing a decrease in FN concentration in conditioned medium (adjusted to osteocalcin) of the FN^−/−-OB-cKO osteoblast cultures compared with control. The remaining FN in the conditioned medium of FN^−/−-OB-cKO cultures was most likely due to the presence of preosteoblasts in which the promoter is not active yet (D). No difference in FN immunostaining in bone sections between control and FN^−/−-OB-cKO mice was seen. FN was stained red. The nuclei of osteocytes were counterstained with DAPI and overlayed. Bars represent 10 µm.

Deletion of FN in bone affects osteoblast number and function

Since in vitro data had suggested that FN plays a pivotal role in matrix mineralization,10–12 we examined whether osteoblast function was affected. Static histomorphometry showed a 27% increase in osteoid (CT 27.6 ± 2.0 versus FN^−/−-OB-cKO 34.9 ± 0.5 OS/BS, p < .05; Fig. 2A, B) and a 41% increase in osteoblast number (CT 12.3 ± 0.8 versus FN^−/−-OB-cKO 17.3 ± 1.4 Ob.N/BS, n = 11, p < .01; see Fig. 2C). Thus more matrix seemed to be laid down by an increased number of osteoblasts. However, new bone formation at the tissue level was not increased (BFR/BS: CT 0.34 ± 0.05 versus FN^−/−-OB-cKO 0.32 ± 0.05 mm²/mm/year, n = 14, p = NS; not displayed). Two further measurements explained this apparent contradiction: (1) The adjusted apposition rate, which estimates the activity of a team of osteoblasts at the cellular level, was decreased by 34% (CT 0.77 ± 0.04 versus FN^−/−-OB-cKO 0.51 ± 0.04 µm/day, p < .01; see Fig. 2D), and (2) the mineralization lag time showed a delay in mineralization by 58% (CT 8.3 ± 0.7 versus. FN^−/−-OB-cKO: 13.2 ± 1.0 days, p < .01; see Fig. 2E). Follow-up experiments failed to show that the increase in osteoblast number was due to increased proliferation or decreased apoptosis (proliferation: CT 6.4 ± 0.7 versus FN^−/−-OB-cKO 6.6 ± 0.7%, p = NS; apoptosis: CT 6.2 ± 0.5 versus FN^−/−-OB-cKO 6.7 ± 0.3%, n = 8, p = NS; not displayed). These results raise the possibility that osteoblast maturation was affected. However, the number of osteocytes, which are terminally differentiated osteoblasts, was not increased (CT 460 ± 25 versus FN^−/−-OB-cKO 460 ± 11 osteocyte/mm², n = 12 and 11, p = NS; not displayed). Similarly, circulating osteocalcin levels as a marker for differentiated osteoblasts did not differ (CT 46.4 ± 3.9 versus FN^−/−-OB-cKO 47.1 ± 6.6 ng/mL, n = 15 and 10, p = NS; not displayed). These data suggest that the increase in the number of differentiating osteoblasts does not result in an increase in the number of differentiated osteoblasts.

Figure 2. Total bone histomorphometry of the proximal tibia in FN-osteoblast cKO mice (FN^−/−-OB-cKO). (A) Masson-Goldner staining of bone sections showing marked increase in osteoid thickness (in pink) in FN^−/−-OB-cKO mice. Bars represent 50 µm. (B) Increased osteoid surface adjusted to bone surface (OS/BS) in FN^−/−-OB-cKO mice compared with controls (CT). (C) Increased osteoblast number adjusted to bone surface (Ob.N/BS) in FN^−/−-OB-cKO mice. (D) Decreased adjusted apposition rate in FN^−/−-OB-cKO mice (Aj.AR represents the formation rate of a team of osteoblasts at the basic multicellular unit). (E) Increased mineralization lag time (Mlt) in FN^−/−-OB-cKO mice (Mlt represents the time needed to mineralize osteoid).

Osteoblasts produce fibronectin isoforms that are distinct from plasma fibronectin

Mice carrying the EGFP reporter and collagen-α1(I)-cre will express EGFP in cells exposed to cre.24, 25 Calvarial osteoblasts from these mice were sorted by FACS. Expression of the EDB, EDA, and complete variable region was found in sorted cells on reverse-transcriptase PCR (Fig. 3A). Immunoprecipitation of FN in osteoblast lysates followed by blotting showed that osteoblast FN did not have the same distinctive double-band pattern seen in plasma FN in the circulation (Fig. 3B). ELISAs of osteoblasts conditioned medium were performed. The medium contained higher concentrations of EDA- and EDB-containing isoforms than the circulation (adjusted to total FN) (Fig. 3C), and mass spectrometry confirmed the presence of the EDA and EDB sequences in osteoblast FN but not in plasma FN. Finally, the FN produced contained at least one O-glycosylation site in the variable region at the site defining the isoform oncofetal FN (oFN) (see Fig. 3C). Thus the osteoblasts produce FNs that differ in some aspects from plasma FN.

Figure 3. Fibronectin isoforms produced by calvarial osteoblasts in vitro. (A) Qualitative reverse-transcriptase PCR shows that osteoblasts produce the EDB domain, the EDA domain, and the complete variable region of fibronectin. (B) Immunoprecipitation of fibronectin in osteoblast lysates reveals a band that is different from the characteristic double band resulting from circulating plasma fibronectin. (C) Concentration of EDA, EDB, and oncofetal fibronectin (oFN) adjusted to total fibronectin (ng/µg total FN) in conditioned medium from osteoblasts sorted based on EGFP production when collagen-α1-cre is expressed and in the circulation. These values are an underestimation because the standards used for the isoforms but not for the total fibronectin are human standards.

The effect on osteoblast function is not mediated by binding through the RGD motif

FN usually mediates its effects by binding to integrins. Five of the six integrins expressed on osteoblasts bind to RGD in the central cell-binding region of FN. These are α5β1, α8β1, αvβ1, αvβ3, and αvβ5.11, 17–20 Furthermore, the presence of the EDA domain, which is found in osteoblast fibronectin, was shown to enhance the interaction of RGD on fibronectin with α5β1 integrin.33 We therefore chose mice carrying a mutation in the RGD sequence of FN to RGE, rendering it unable to bind to RGD-binding integrins (FN-RGE). Owing to the central role of FN binding to α5β1 and αvβ1 during embryonic development, homozygote mice (FN-RGE/RGE) die in utero.26 We therefore made col-cre/+-FN-RGE/fl mice, in which activation of the cre recombinase in osteoblasts leads to the deletion of one FN allele (the floxed allele) and the production of an FN that contains the mutated RGE sequence (FN^RGE/–-OB-cKO) and compared these with FN-fl/fl littermates. Mice in which osteoblasts produced only mutated FN (FN^RGE/–-OB-cKO) did not show statistically significant differences compared with controls in the tested parameters (Ob.N/BS, Aj.AR, and Mlt) (Fig. 4A–C). Therefore, binding of FN originating from osteoblasts to integrins via the RGD motif does not seem to be critical for their number or function.

Figure 4. Bone histomorphometry of the proximal tibia in mice lacking FN production in osteoblasts (FN^−/−-OB-cKO), mice producing FN containing a mutation from RGD to RGE (FN^RGE/–-OB-cKO), and mice lacking β₁-integrin expression on osteoblasts (β1^−/−-OB-cKO). The different controls were set to 100% for ease of presentation. (A) Increased osteoblasts number adjusted to bone surface (Ob.N/BS) is significant in FN^−/−-OB-cKO mice and shows a trend for an increase in β1^−/−-OB-cKO mice. (B) Decreased adjusted apposition rate in FN^−/−-OB-cKO and β1^−/−-OB-cKO mice (Aj.AR represents the formation rate of a team of osteoblasts at the basic multicellular unit). (C) Increased mineralization lag time (Mlt) in FN^−/−-OB-cKO and β1^−/−-OB-cKO mice (Mlt represents the time needed to mineralize osteoid).

Two integrins expressed on osteoblasts can bind to sites in FN unrelated to the RGD sequence. αvβ3 can bind at a site in the amino terminus of FN, and the relevance of this binding has been established in vivo.26 α4β1, on the other hand, has been shown to bind to the EDA domain or to the variable region.34, 35 In order to differentiate between these two integrins, we established mice that are cKO for the β₁-integrin subunit in osteoblasts (Col-cre/+-β1-fl/fl) (β1^−/−-OB-cKO) and compared these with littermate controls (β1-fl/fl). We found a trend to more osteoblasts in cKO mice (CT Ob.N/BS = 20.1 ± 2.8 versus β1^−/−-OB-cKO 26.1 ± 2.9/mm, 27% increase, n = 8, p = .09; Fig. 4A), a significant decrease in the adjusted apposition rate (CT 0.70 ± 0.10 versus β1^−/−-OB-cKO 0.47 ± 0.05, 32% decrease, p < .05), and an increase in mineralization lag time (CT 10.9 ± 1.4 versus β1^−/−-OB-cKO 14.9 ± 1.9, 36% increase, p < .05; Fig. 4B, C). The direction of the changes in β1^−/−-OB-cKO thus corresponded with the changes seen in FN^−/−-OB-cKO (Fig. 4A-C). This raises the possibility, but does not prove, that β₁-integrin may mediate FN actions on osteoblasts, in which case α4β1 could be a possible candidate for FN effects on osteoblasts.

Circulating FN is incorporated in the bone matrix

Since the amount of FN deposited in bone was not decreased despite the deletion of FN in the matrix-synthesizing osteoblasts, we hypothesized that plasma FN originating from hepatocytes in the liver is incorporated in the bone matrix. Therefore, plasma FN was fluorescently labeled and administered to mice intraperitoneally, and the presence of the labeled FN in the bone matrix was verified (Fig. 5A, left panels). Of note is that its deposition was not limited to the newly mineralized extracellular matrix marked by alizarin complexone (see Fig. 5A, right panels).

Figure 5. (A) Oyster-500-labeled plasma FN (pFN) uptake in the proximal tibia (lower left). Shown are longitudinal sections of tibia. The deposition was not limited to areas of active mineralization marked with alizarin complexone (lower right). The control (upper panels) was injected with Oyster-500 and alizarin complexone. Autofluorescence of bone but no uptake of the green fluorescent dye was seen. Bars represent 125 µm. (B) Concentration of circulating FN in the plasma of liver FN–cKO mice lacking circulating FN (Mx-cre/+-FN-fl/fl: Mx-cKO; albumin-cre/+-FN-fl/fl: Alb-cKO) compared with littermate controls (FN-fl/fl). Please note the significantly lower levels of circulating FN in Mx-cKO compared with Alb-cKO mice, suggesting a better deletion of FN using the Mx promoter. (C) Marked decrease in FN staining in bone sections of liver FN–cKO mice (Mx-cKO). The stronger staining at the edges is an artifact. Bars represent 10 µm. (D) Marked decrease in FN signal on Western blot of bone matrix in the Mx-cKO mice compared with control. Equal protein amounts were loaded.

Deletion of FN in the liver affects FN content in bone

In order to determine whether the deletion of FN production in the hepatocytes will reduce the amount of FN deposited in bone, we made Mx-cre/+-FN-fl/fl mice (Mx-cKO), which delete FN in hepatocytes and a variety of other cells after induction of the Mx promoter, and albumin-cre/+-FN-fl/fl mice (Alb-cKO), which delete FN in hepatocytes only.23 Circulating FN was decreased by more than 95%, as evidenced by ELISA (Mx-cre line: CT 157.1 ± 12.8 versus Mx-cKO 4.6 ± 0.2 µg/mL, n = 7 and 26, 2.9%, p < .001; albumin-cre line: CT 131.1 ± 8.1 versus Alb-cKO 6.2 ± 0.4 µg/mL, n = 7 and 41, 4.7%, p < .001; Fig. 5B). Deletion using the Mx promoter was significantly better than using the albumin promoter (Mx-cKO 4.6 ± 0.1 versus Alb-cKO 6.2 ± 0.4, p < .01; see Fig. 5B) and occurred, on average, 4 to 6 weeks earlier. FN in the bone matrix was decreased in Mx-cKO compared with control mice, as shown by immunostaining and Western blotting (Fig. 5C, D).

Circulating FN affects bone density without affecting bone formation or resorption

Despite the decrease in bone matrix FN, no significant effects on bone cells were seen. None of the following parameters differed significantly between conditional knockout and control mice in both the Mx line and the albumin line: osteoclast numbers (Oc.N/BS), erosion surface (ES/BS), osteoblast numbers (Ob.N/BS), osteoid surface (OS/BS), mineralization lag time (Mlt), adjusted apposition rate (Aj.AR), or bone-formation rates (BFR/BS). This suggests that loss of liver-derived FN and the amount of FN deposited in the matrix have no effect on either osteoblast or osteoclast function. It therefore came as a surprise to find a significant decrease in trabecular BMD using pQCT in the Mx-cKO (CT 205.4 ± 6.2 versus Mx-cKO 177.9 ± 5.4 mg/cm³, n = 34 and 24, respectively, 13% decrease, p < .005) and Alb-cKO mice (CT 209.2 ± 4.8 versus Alb-cKO 193.9 ± 6.0 mg/cm³, n = 27 and 25, respectively, 7% decrease, p = 0.05; Fig. 6A). The effect on BMD seemed dose-dependent, with a greater change in the Mx line in which the circulating FN levels were significantly lower than in the albumin line. This could suggest that an increase in FN will affect BMD too. However, no consistent increase in BMD in wild-type mice injected with up to 3 mg/day for 15 days of plasma FN was found, possibly because of the small relative degree of change in circulating plasma FN and its short duration. While the deletion of FN resulted in more than 95% decrease (20-fold) in circulating FN, injection of plasma FN only caused a relatively short-lived 300% increase (threefold). Another potential explanation might be the presence of a “saturation effect,” after which the amount of FN in the blood does not noticeably affect BMD.

Figure 6. Characteristics of bones from liver-cKO mice. (A) Trabecular BMD as measured by pQCT was diminished in both liver-cKO mice lacking circulating FN (Mx-cKO and Alb-cKO). (B) Mineral-to-matrix ratio measured by infrared microspectroscopy was decreased. (C) Microhardness was increased. (D) The distribution of collagen fibril diameter showed a shift toward thicker fibrils in the liver-cKO mice (white bars) compared with controls (black bars).

Circulating FN affects bone matrix mineralization and mineral composition

In an attempt to identify the mechanism by which FN content in the bone matrix influenced BMD in the absence of a change in bone cells, we hypothesized that matrix mineralization was affected. Using infrared microspectroscopy, the ratio of mineral to matrix in an area of 20 µm diameter can be determined accurately. This ratio corrects for any irregularities caused by differences in the thickness of the histologic sections.29 Mx-cKO mice had a 12% decrease in mineral-to-matrix ratio (CT 9.91 ± 0.32 versus Mx-cKO 8.76 ± 0.20, n = 6, p < .005; Fig. 6B), whereas OB-cKO mice showed no difference from control (CT 8.57 ± 1.27 versus FN^−/−-OB-cKO 8.39 ± 0.34 days, p = NS; not displayed). Furthermore, acid phosphate content (reflecting the composition of the hydroxyapatite mineral crystals in bone) was increased by 4% in Mx-cKO mice compared with controls (CT 0.0131 ± 0.0001 versus Mx-cKO 0.0136 ± 0.0001, p < .01; not displayed). Injection of plasma FN into wild-type mice resulted in a trend for higher mineral-to-matrix ratio (9.06 ± 0.34 versus 9.86 ± 0.45 in injected mice, n = 3, p = .17), and a 2% significant decrease in acid phosphate content (CT 0.0138 ± 0.0001 versus injected mice 0.0134 ± 0.0001, p < .05; not displayed). This suggests that plasma FN injection leads to changes in mineral composition opposite to its deletion.

Circulating FN affects bone biomechanical properties and collagen fibril characteristics

We then asked whether the mechanical properties of the matrix were affected by these changes. Microhardness testing examines the size of the developing indent when applying a constant force on bone. A significant, albeit small, increase in microhardness was found in Mx-cKO mice (CT 69.2 ± 1.1 versus Mx-cKO 71.8 ± 1.0 kg/mm², n = 12 per group and 10 measurements per mouse, 4%, p < .05; Fig. 6C) but not in OB-cKO mice (CT 66.5 ± 1.7 versus OB-cKO 68.2 ± 2.1 kg/mm², n = 13 per group, p = NS; not displayed). The increase in microhardness in the Mx-cKO mice is similar to findings in a mouse model of osteogenesis imperfecta (oim/oim), in which altered collagen structure leads to increased fractures.36, 37 Increased hardness is either due to an increase in mineral or a change in matrix properties.36 Since the Mx-cKO mice have a decrease in the mineral-to-matrix ratio, the increase in microhardness therefore is likely due to a change in matrix properties. Examination of collagen fibril diameter by transmission electron microscopy revealed that the median diameter was higher in the liver-cKO mice compared with control mice (CT 67.3 nm versus liver-cKO 84.7 nm) but not different between OB-cKO mice and their control (CT 64.4 versus OB-cKO 65.8 nm; not displayed). This raised the possibility of a shift in fibril diameter in liver-cKO mice that was evident when data were plotted (Fig. 6D). In summary, the absence of infiltrating FN affects collagen fibril assembly in a manner similar to osteogenesis imperfecta.38 These two similarities between the absence of circulating FN and osteogenesis imperfecta support a role for circulating FN in collagen assembly in vivo.

Discussion

The principal findings of this study are, first, osteoblast fibronectin controls both the number and the function of osteoblasts. This effect is not mediated by the presence of a functional RGD motif on fibronectin. Second, osteoblast fibronectin has no effect on fibronectin accumulation in the matrix and does not modulate bone matrix properties. Third, circulating fibronectin originating from the liver infiltrates bone matrix and provides the bulk of matrix fibronectin. There it modifies the mineralization and the biomechanical properties of bone matrix and the structure of collagen fibrils without any effect on bone cell numbers or function. The differentiation between the functions of both sources of fibronectin thus is presented for the first time. In particular, these data support the notion that affecting matrix properties can proceed without any effects on bone cell function.

The deletion of fibronectin (FN) in the differentiating and differentiated osteoblasts (including osteocytes) failed to show a significant decrease in the amount of FN in the bone matrix. This suggests that another source is responsible for maintaining FN in the bone matrix. A candidate source would be the liver.27 The injection of labeled plasma FN revealed that this large molecule (480 kDa) is able to infiltrate the whole of the bone matrix without being limited to the newly formed osteoid. In addition, deletion of circulating FN resulted in a marked decrease in the amount of FN in the bone matrix. An interesting subject for future research is whether the decrease in circulating FN levels in advanced liver disease contributes to the bone abnormalities and increased fractures seen in hepatic osteodystrophy.39

Our experiments suggest a need for both the osteoblast FN and the circulating FN to achieve healthy bones. In vitro experiments offer a possible explanation for the dual source of provision of FN to bone and the ability of circulating FN to infiltrate both newly formed and established matrix. Cells deficient in FN production were unable to form a collagen network unless soluble FN was added.4 Thus the initial presence of FN is a requisite for the initial assembly of collagen. The resulting matrix is in continuous turnover. When FN was removed from the medium after collagen assembly had taken place, FN dissociated from the matrix and was lost.6 Thus the continuous presence of FN is also important for matrix integrity. The osteoblasts are present for a short period of time during bone formation; therefore, they can provide the initial FN required for the first steps of collagen assembly during matrix deposition. They cannot, however, provide the continuous source. The continuous source could be the osteocytes, the bone lining cells, or the circulation. Any of these sources would require the ability of FN to infiltrate into the mineralized bone matrix. Since circulating fibronectin is available throughout life, evolution seems to have elegantly solved the problem by taking advantage of this abundant source of circulating liver-derived FN in the bloodstream.

In our mouse model, the absence of osteoblast FN results in an intriguing effect on the osteoblast itself without affecting the matrix. While in vitro data had suggested that FN is required for normal mineralization by the osteoblasts, our in vivo experiments reveal new findings.10–12 Dysfunctional osteoblasts are increased in number with an apparent correction of the functional defect at the level of whole bone. This reestablished balance keeps bone-formation rates within the normal range. Thus osteoblast FN seems to be part of a feedback loop that controls both the number and function of the osteoblasts. Our data suggest that fibronectin deletion did not take place in preosteoblasts. Thus the effects described in the OB-cKO mice appear to be due to loss of fibronectin in differentiating osteoblasts.

The failure of the liver-derived circulating FN to completely prevent the defect caused by the deletion of FN in the osteoblasts suggests a difference in the function of these isoforms, even though it does not exclude a role for a difference in the amount of fibronectin acting on the osteoblasts. In fact, FN containing the EDA and EDB domains is produced by osteoblasts in relative concentrations that are around nine- and eightfold higher than in the circulation (as shown in Fig. 3C). Interestingly, the presence of the EDA domain supports cell spreading and promotes cell cycle progression more potently than FN lacking the EDA domain.33, 40EDA can bind to α4β1, and its presence can increase the binding of RGD to α5β1.33, 34 No binding partners for EDB FN have been characterized yet. In one instance, an isoform of fibronectin had effects on osteoblasts that differed from plasma fibronectin. We have shown that the presence of oncofetal fibronectin, defined by the presence of an O-glycosylation site in the variable region, was associated with a decrease in the number of osteoblasts and bone formation in vivo and a decrease in osteoblast mineralization in vitro.27 These effects were not seen when using plasma fibronectin. Various modifications in osteoblast FN may affect osteoblast function, and these will need to be examined further. The differences between osteoblast and plasma fibronectin may offer an explanation for the diverse effects of deleting osteoblast or plasma fibronectin. Our data, however, do not prove a causal relationship between the differences in the isoforms and the in vivo findings.

In an attempt to characterize the receptor that mediates the effects of FN on osteoblast function, we were able to exclude a role for RGD-binding receptors. Because two integrins found on osteoblasts (αvβ3 and α4β1) bind to other sites of FN, we then deleted β₁-integrin in osteoblasts. This resulted in effects that resemble the effect of loss of osteoblast FN. While this result may imply that a β₁-integrin is a possible mediator of FN effects on osteoblasts, future experiments will be required to definitely establish that the bone effects seen in the β1^−/−-OB-cKO mice are primarily due to a lack of osteoblast binding to fibronectin. Even though the presence of a low-affinity binding site for α5β1 integrin in the amino terminus of FN has been suggested based on binding assays not involving cells, experiments on cells fail to support a role for this binding.26, 33

At first glance, it seems counterintuitive that even though osteoblasts produce fibronectin, another source for fibronectin would be required to maintain normal bone matrix properties and mineralization. This, however, can be explained by the need for the continuous provision of fibronectin to maintain matrix integrity. Our experiments demonstrate that fibronectin originating from osteoblasts does not significantly contribute to the pool of fibronectin deposited in the bone matrix. Instead, the osteoblast-derived fibronectin isoform seems to be part of a feedback loop that modulates osteoblast differentiation locally without causing appreciable changes in bone matrix properties. On the other hand, circulating plasma fibronectin infiltrates the bone matrix, where it represents the predominant source of fibronectin affecting bone mineralization without measurable effects on the osteoblasts. These data refute the common perception that most of fibronectin in the bone is laid down by the osteoblasts during bone formation and shows a significant role for the liver in contributing to bone health by providing to bone an “extracellular matrix protein” via the circulation.

Disclosures

All the authors state that they have no conflicts of interest.

Acknowledgements

We thank Reinhard Faessler for his invaluable input, Stefan Meuer for his continued support, and Bernd Arnold for the EGFP reporter mice. We acknowledge the competent technical support of Michael Suepfle and Birgit Hub, as well as the use of the Nikon Imaging Center at the University of Heidelbeg. This work was supported by the Max-Planck Society (AB, NK, and IAN) and the University of Heidelberg (AB, NK, and IAN).