Disorganization of chondrocyte columns in the growth plate does not aggravate experimental osteoarthritis in mice

Osteoarthritis (OA) is a multifactorial joint disease mainly affecting articular cartilage (AC) with a relevant biomechanical component. During endochondral ossification growth plate (GP) chondrocytes arrange in columns. GPs do not ossify in skeletally mature rodents. In neonatal mice, an altered joint loading induces GP chondrocyte disorganization. We aimed to study whether experimental OA involves GP disorganization in adult mice and to assess if it may have additional detrimental effects on AC damage. Knee OA was induced by destabilization of the medial meniscus (DMM) in wild-type (WT) adult mice, and in Tamoxifen-inducible Ellis-van-Creveld syndrome protein (Evc) knockouts (EvccKO), used as a model of GP disorganization due to Hedgehog signalling disruption. Chondrocyte column arrangement was assessed in the tibial GP and expressed as Column Index (CI). Both DMM-operated WT mice and non-operated-EvccKO showed a decreased CI, indicating GP chondrocyte column disarrangement, although in the latter, it was not associated to AC damage. The most severe GP chondrocyte disorganization occurred in DMM-EvccKO mice, in comparison to the other groups. However, this altered GP structure in DMM-EvccKO mice did not exacerbate AC damage. Further studies are needed to confirm the lack of interference of GP alterations on the analysis of AC employing OA mice.

www.nature.com/scientificreports/ activates the Glioma-Associated Oncogene (GLI) transcription factors, which induce the expression of Runx2 and promote chondrocyte hypertrophy. Therefore, there is a balance between proliferation and hypertrophy in the GP chondrocyte population dependent on negative feedback between Ihh and PTHrP action 21,22 . The primary cilium is a mechanosensing non-motile structure that projects from the cell and plays a crucial role in the transduction of a variety of extracellular signals. After chondrocyte division in the proliferative zone, the cilium of the resulting cells orients and cells polarize 23 . Then, by successive rotational movements, chondrocytes in the GP align parallel to the longitudinal axis of the bone and stack in columns 23 . Mutations affecting ciliogenesis or the function of proteins anchored to its structure, result in a group of conditions known as ciliopathies, many of which include skeletal phenotypes 24 . Ellis van Creveld (EvC) syndrome is a ciliopathy characterized by disproportionate short stature, short limbs, narrow chest and post-axial polydactyly, among other features 24,25 . Loss-of-function mutations in EVC and EVC2 cause the EvC syndrome. EVC and EVC2 protein products localize at the base of the primary cilium and regulate Hh signal transduction 25 .
It has been previously demonstrated that disruption of Hh signalling by Evc deletion alters bone development in mice. Evc -/embryos exhibit shorter proliferative and hypertrophic chondrocyte regions in the GP, and overall, shorter bones. Also, Evc -/mice at postnatal day 16 (P16) show abnormal tibia shape and disorganized epiphyseal chondrocytes 26 . An altered biomechanical force on chondrocytes during bone development also evokes inaccurate GP function and defective bone growth [26][27][28] . In neonatal mice, a reduced mechanical joint loading due to hemiplegia induces the disorganization of chondrocyte columns in the GP and diminished tibial length 27 . Similarly, joint immobilization in 6-week-old rats that mimicked an unloading state, reduced the length of the tibial GP, resulting in abnormal bone growth 28 . In a different approach, in a mouse model of spondyloepiphyseal dysplasia congenita, which resulted in an increased mechanical loading on the shoulder, a loss of GP chondrocyte column organization in the humerus was described 29 .
As chondrocytes in hyaline cartilage have mechanosensors, variations in the pressure of their environment can affect the interactions between cells and with the extracellular matrix (ECM).
We aimed to study whether the induction of experimental OA could disturb GP organization in adult mice. Moreover, we wanted to assess if GP alteration triggers a further deleterious effect on AC damage. For the latter

Results
Gp structure is altered during oA. AC damage was evaluated in all joints obtained at sacrifice, 8 weeks after DMM surgery. Post-traumatic damage due to surgical DMM accounted for the AC lesions observed in the DMM-WT mice, that were characterized by proteoglycan loss, cartilage fibrillation and vertical clefts. Cartilage erosions and clefts were mainly located in the bearing region of the tibia (Fig. 1a). As expected, DMM-WT mice showed a higher OARSI score than that observed in NO-WT animals (Fig. 1b). We then studied whether experimental knee destabilization was able to modify the columnar structure of chondrocytes in the GP. While NO-WT mice showed GP chondrocytes arranged in the characteristic columns, DMM-WT mice presented an altered columnar structure, and clusters of cells were observed (Fig. 1c). The percentage of cells in columns in DMM-WT animals was significantly lower than that found in the NO-WT group, as well as its CI (Fig. 1d). We observed a negative association between the percentage of cells in columns and the OARSI score (ρ = -0.55, p = 0.01), as well as between the CI and the OARSI score (ρ = -0.52, p = 0.02) (Fig. 1e). In addition, the number of columns, the mean column length, the tissue cellularity, and the number of cells per column also decreased in the DMM-WT group in comparison to NO-WT animals ( Supplementary Fig. S2).

Gp chondrocyte disorganization driven by Evc deletion does not induce Ac damage. In order
to evaluate whether GP alteration per se could induce OA-like cartilage damage, we employed Evc cKO mice. It has been previously observed that newborn Evc -/mice display a clear disarrangement in the GP 26 due to Hh signalling disruption. Therefore, we used adult Evc cKO mice, in which both Evc and Evc2 expression were deleted after skeletal maturity, as a model of GP disorganization in adult mice. In order to assure that potential AC lesions were not caused by bone alterations due to genetic manipulation in Evc cKO mice, we first assessed bone growth and quality in these mice. Evc cKO mice showed no apparent morphological defect and normal skeletal development. No differences were observed in the body weight between WT and Evc cKO mice at 12 weeks of age, just before DMM surgery ( Supplementary Fig. S3). Consequently, we analyzed GP chondrocyte arrangement in NO-Evc cKO animals, and observed that this group showed a clear GP chondrocyte disorganization in the adult stage ( Fig. 2c), with a decreased CI and percentage of cells in columns in comparison to NO-WT, (Fig. 2d), as well as a lower GP cellularity, mean column length and number of cells per column than those measured in NO-WT mice ( Supplementary Fig. S2). Nevertheless, NO-Evc cKO animals did not show AC damage, and presented a similar OARSI score to that observed in NO-WT mice (Fig. 2a,b).

Gp chondrocyte disorganization evoked by Evc deletion does not aggravate Ac damage in a
DMM-induced oA model. In order to establish whether GP alteration could accelerate AC damage evoked by OA, we studied the GP chondrocyte arrangement and the AC histological damage in DMM-Evc cKO mice. The pattern of disorganization in the GP chondrocytes due to OA was recurrent among these animals. Accordingly, DMM-Evc cKO mice showed a significantly lower CI than NO-Evc cKO mice, together with a decreased percentage of chondrocytes organized in columns, although the latter was not statistically significant (Fig. 3c,d).
As expected, DMM surgery induced significant OA-like lesions in the AC of the Evc cKO mice, as it is shown in the OARSI score assessment (Fig. 3a,b). As described for WT animals, Evc cKO mice also showed a negative association between the percentage of cells in columns and the OARSI score (ρ = -0.59, p = 0.02), and between the CI and OARSI score (ρ = -0.58, p = 0.02) (Fig. 3e).
DMM-Evc cKO mice also showed a lower percentage of cells in columns and a lower CI than that observed in DMM-WT mice, exhibiting an absolute loss of chondrocyte organization in the GP (Fig. 3h, i). Additionally, cellularity, total number of columns, and number of cells per column were further decreased in DMM-Evc cKO mice when compared to DMM-WT animals ( Supplementary Fig. S2). However, DMM-WT and DMM-Evc cKO mice exhibited equally severe lesions in the AC, thus showing a similar OARSI score (Fig. 3f, g). Therefore, the higher disorganization of the GP observed in the DMM-Evc cKO group did not induce additional deleterious effects on the AC lesions observed in these animals.  Fig. S3). Furthermore, both cortical and trabecular BMD measured in the femur, tibia and vertebrae, respectively, was similar in the four groups studied ( Supplementary Fig. S3), indicating a healthy bone growth and development in genetically modified animals throughout the study. As expected, we observed an increase in BMD in the knee joint of DMM operated animals, which has been described as subchondral bone sclerosis, both in mice 30 and in OA patients 31 . The increase in knee BMD was similar between DMM-WT and DMM-Evc cKO mice ( Supplementary Fig. S3).
GP COL-X histological distribution is modified during OA. The GP thickness was similar in NO-WT and DMM-WT animals, as well as the proteoglycan content in this localization (Fig. 4a, b). Likewise, there was no difference in the amount of MMP-13 and COL-X in the GP between these groups (Fig. 4c-f). However, OA accounted for a certain modification in COL-X localization in the GP, as can be observed in Fig. 4e. In NO-WT, COL-X was mainly located in the ECM of the upper area, where cells exhibited less COL-X staining when compared to the enlarged hypertrophic chondrocytes in the lower area. In DMM-WT mice, an intense ECM staining was observed, and all the cells in the GP seemed to be equally stained (Fig. 4e). ness was similar between NO-WT and NO-Evc cKO mice, although the GP of these animals seemed to contain a higher presence of proteoglycans than that observed in NO-WT (Fig. 5a, b). Both MMP-13 and COL-X staining and localization were comparable in these groups ( Fig. 5c-f). DMM surgery did not seem to modify these parameters ( Fig. 5g-l).
Gp pathological state may interfere in molecular analyses of Ac damage. Our next objective was to assess whether the presence of GP cartilage could disturb the measurement of AC proteins in mouse models that employ the femoral condyle for these studies. As can be observed in Supplementary Figure S4, a significant proportion of the hyaline cartilage comes from the GP when the condyle is isolated for analysis. Therefore, we analysed the presence of AGG and COL-II, two AC proteins usually analysed in OA studies, in mouse condyles with or without AC. We observed that femoral condyles without AC showed a tendency of lower AGG protein levels compared to femoral condyles with intact AC, although no statistical significance was observed (Fig. 6a, b). According to these results, a high percentage of total AGG in the mice femoral condyle, around 60-70%, comes from the GP. Furthermore, the amount of COL-II protein in the femoral condyles analysed indicates that around 95% of this protein comes from the GP cartilage when the condyle is studied, since identical COL-II presence was observed in tissues with or without AC (Fig. 6c, d).

Discussion
A relevant percentage of preclinical studies employing animal models are not reproducible in humans, which results in a worldwide cost overrun in biomedical research 32,33 . For this reason, the search for specific causes that may bias the face value of the results and their interpretation emerges mandatory. Our results show that, besides the damage in the AC, OA induced by knee destabilization in adult mice altered the characteristic columnar structure of the chondrocytes in the GP. Eight weeks after OA induction, columns look shorter and not homogenously organized within the GP, but instead, in irregular clusters. Endochondral ossification is an active process in mice until 10 week-of-age, when bone development concludes and mice show a mature skeleton 9 , even though epiphyses do not ossify in skeletally mature rodents as it happens in humans 18 . In addition, growing mice are resistant to joint damage 9 . In order to abolish any effect due to bone immaturity, we studied GP alterations www.nature.com/scientificreports/ induced by joint destabilization in 12-week-old mice, once bone growth was completed. Despite being considered a mature GP, our results show that this tissue is profoundly altered by OA in a relatively short period of time. Different causes may be responsible for these modifications, although those related to mechanical alterations seem of great impact. Limb paralysis in neonatal mice that resulted in a reduced mechanical joint loading, also evoked chondrocyte column disorganization at the GP 27 . We contemplate that chondrocyte disorganization and the loss of the column arrangement during DMM-induced OA might be prompted by an alteration of joint biomechanics that results in unbalanced forces. These forces may affect subarticular tissues 34 and even the GP, and modify cell-to-cell and cell-to-ECM interactions 35 , as well as chondrocyte polarization in the GP. Both connexins and integrins have been described as mechanotransducers in the cartilage, and modulate cell responses after mechanical stimulation 36,37 , and could therefore be responsible for the alteration in GP structure after joint destabilization. These proteins are especially relevant in chondrocytes, playing an essential role in cellular polarity and structure of the collagen network in the GP, and modulating cell-to cell and cell-to matrix interaction [37][38][39][40] . Connexins and integrins can be localized in the chondrocyte primary cilia, where they regulate cell interactions with the surrounding matrix [41][42][43] .
Once observed the alteration of chondrocyte columns in the GP induced by OA, we wanted to study whether this pathologic disarray could produce AC damage or increase the AC lesions induced by knee destabilization in mice. Evc -/embryos and newborn mice had loss of chondrocyte column structure in the GP 26 , together with severe skeletal abnormalities that replicate the human disorder described as EvC syndrome. However, most of them die a few days after birth or cannot feed normally 26 . Here, we have generated a new experimental model for conditional Evc deletion in adult mice in order to approach different physio-pathological aspects in OA 26,44 . Evc cKO mice showed no discernible defects neither at 10 weeks of age, not after TAM-induced complete Evc deletion. At 10 weeks of age, Evc cKO mice showed normal skeletal development, and similar body weight to that observed for their WT littermates, in spite of the loss of Evc expression and carrying only one Evc2 allele (unpublished data), in line with the skeletal properties described for Evc2 +/heterozygous mice 26,44 . TAM administration in adult mice predictably induced a complete disruption of Hh signaling, according to previous analyses in Evc -/embryonic GP 26 . Adult NO-Evc cKO mice did not show any evident skeletal alteration, their tibia and femur length were normal, and BMD measurements at different bone locations were similar to those observed for WT mice. However, they showed a disorganized GP architecture, with a lesser proportion of chondrocytes in columns to that observed in WT animals. This disorganized GP architecture did not predispose to cartilage damage, since NO-Evc cKO mice did not exhibit any OA-like lesion in the AC. Furthermore, in DMM-Evc cKO mice, we observed that knee destabilization induced a higher disorganization in the GP to that observed in DMM-WT or in NO-Evc cKO mice, although it did not aggravate AC lesions, which were similar to those observed in DMM-WT animals. Therefore, a profound disorganization in the GP did not have any additional detrimental effect on the AC. However, it cannot be discarded that a longer monitoring of these animals, resulting in more severe and chronic joint instability, could presumably display detectable OA-like cartilage lesions, or even the appearance of spontaneous OA.
Since both AC and GP cartilage are damaged during experimental OA, molecular studies using the whole mouse femoral condyle could be misinterpreted due to the presence of a damaged GP. The analyses in these experimental models are not usually performed by the precise isolation of the AC, due to size limitations. Therefore, the whole femoral condyle is commonly employed in gene and protein analyses, which may partially contain GP cartilage. To elucidate this point, we studied the presence and distribution of classic markers of cartilage damage in the GP. While proteoglycan content and MMP-13 presence did not change between healthy and OA WT mice, we observed a different distribution pattern of COL-X in the GP during OA. This result guided our next experiment considering the possibility that molecular analyses employed as usual markers of hyaline cartilage damage may be biased and overestimated by the interference of tissue coming from the GP. We analysed whether the relative presence of AGG and COL-II proteins was different in femoral condyles that contain or not the AC. In both cases, the origin of the protein was mainly the GP, since the lack of the AC did not significantly change the relative presence of these proteins in the femoral condyle analysis. Around 71% AGG and 97% COL-II protein may be originated in the GP. Although the contribution of GP cartilage could be variable between the different samples, our results showed that molecular analysis of the cartilage in mice samples containing the GP should be interpreted with caution.
The limitations of this study include the impossibility of assessing the GP bias on AC during OA, due to the unfeasible duplication of the experimental model to obtain the femoral condyles for this purpose.
In conclusion, surgical knee destabilization in mice is responsible for an altered GP structure, accounting for chondrocyte disorganization and loss of column arrangement, although this alteration in the GP does not predispose nor aggravate the development of more severe damage in AC of the affected joint. AC damage and GP disarrangement could be similarly triggered by joint instability, but both seem independent phenomena in terms of their progression and underlying mechanisms. Since adult mice exhibit such a great affectation of the GP structure, molecular measurements might be overestimated by protein content dragged from the GP when employing the whole condyle in order to study AC during OA. Additional research would be necessary to determine this interference and its significance in mouse OA model, in order to prevent bias in experimental models that may subsequently cause a lack of reproducibility in the human disease.

Methods
Animal model of oA. OA was induced in 12-week old female WT mice by DMM as previously described 8 (DMM-WT, n = 12 joints), while non-operated WT mice were simultaneously studied (NO-WT, n = 16 joints) ( Supplementary Fig. S1).
Evc cKO littermates that were genotyped with Evc and Evc2 WT alleles were used as WT animals in the OA model.
Eight weeks following DMM surgery, all mice were anesthetized and bone mineral density (BMD) was assessed by dual-energy X-ray absorptiometry using PIXImus (GE Lunar Corp., USA). Afterwards, mice were euthanized and knee joints fixed in 4% formaldehyde (Sigma-Aldrich, USA), decalcified and embedded in paraffin for histopathological analyses (Supplementary Fig. S1)

Analysis of the interference of Gp cartilage proteins in the study of Ac proteins. Fifteen
3-month-old female C57BL/6 WT mice were euthanized and the femurs were dissected. The AC from left femoral condyles was scraped with a surgical scalpel until completely removed, while right femora were kept intact. Femoral condyle samples from 10 mice were immediately frozen in liquid N 2 for protein analysis, while the remaining 5 samples were fixed in 4% formaldehyde (Sigma-Aldrich), decalcified and embedded in paraffin for histological analyses.
Gp chondrocyte column organization assessment. The analysis of the GP chondrocyte column organization was performed in the knee sections obtained eight weeks after DMM surgery from operated and non-operated animals. 4 µm sagittal paraffin-embedded knee sections were stained with Haematoxylin-Eosin. Joint sections were photographed with a Leica DM3000 LED digital micro-imaging instrument (Leica, Microsystems, Inc. Buffalo Grove, IL, USA) at the central region of the tibia at × 40 magnification, and chondrocyte histomorphometry was evaluated as described elsewhere 27,46 . Briefly, a chondrocyte column was defined as a chondron containing a minimum of three chondrocytes, located no more than 20 pixels apart one from each other, and with angles between them ranging from -155° to -179° or 155° to 180°. These parameters were computed as previously described by Killion et al 27 . Once the columns were established in each sample, the number of cells was quantified using the Image J Software (NIH, USA) cell counter plug-in. Then, the percentage of cells in columns was obtained by calculating the ratio 'cells in columns'/ 'total cells in the GP' . Column Index (CI) was calculated multiplying the percentage of cells in columns by their respective length 27 . Histopathological assessment of the cartilage. AC was evaluated in 4 µm sagittal knee joint sections that were stained with Safranin-O Fast Green. The histological damage was assessed by two blinded observers following the OARSI recommendations for mice samples 47 , which were adapted for sagittal slices. The semiquantitative 0 to 6 points scoring system was applied to one area in the femur and one area in the tibia, and the sum of these two scores was represented as the total score.
Proteoglycan content in the GP cartilage was also evaluated in Safranin-O Fast Green stained joint samples. Sections were photographed with a Leica DM3000 LED digital micro-imaging instrument (Leica, Microsystems, Inc. Buffalo Grove, IL, USA) at the central region of the tibia at × 40 magnification. Specific Safranin-O staining in the femur and tibia GP, as well as GP thickness, were measured employing Image J software, and corrected by sample area. immunohistochemical analysis. 4 µm knee joint sections were deparaffinized and rehydrated. Antigen retrieval was performed by incubation with 20 µg/mL proteinase K (Promega, USA). Tissue sections were processed as previously described [48][49][50] , and then incubated with the corresponding primary antibody: anti-mouse type X collagen (COL-X) (1/2000; ab58632, Abcam, UK) or anti-metalloproteinase (MMP)-13 (1/200; ab39012; Abcam). A secondary biotinylated anti-rabbit IgG was used for detection of positive signal through a horseradish peroxidase linked to an avidin/biotin complex (ABC) (Vector Laboratories, USA) using 3,3 diaminobenzidine tetra-hydrochloride as chromogen (Dako, Denmark). Joint sections were counterstained with Haematoxylin, dehydrated and mounted in DPX (Merck Millipore, USA). Sections were photographed with a Leica DM3000 LED digital micro-imaging instrument (Leica, Microsystems, Inc. Buffalo Grove, IL, USA) at the central region of the tibia at × 40 magnification, and specific staining in the GP was quantified using Image J software and corrected by sample area.
Dual-energy X-ray absorptiometry. BMD (g/cm 2 ) was assessed eight weeks after surgery in anesthetized mice, by dual-energy X-ray absorptiometry (DXA) using PIXImus (GE Lunar Corp., USA), as previously described 51 . BMD was assessed in different regions of interest (ROIs). For subchondral BMD, we employed a 11 × 11 pixel ROI focused on the knee joint; cortical BMD was measured in the femur and in the tibia employ- Western blot. Briefly, 20 μg of total protein from tissues were extracted as described elsewhere 50 , separated by electrophoresis in SDS-polyacrylamide gels and transferred to a nitrocellulose membrane in a semidry Trans-Blot device (Bio-Rad, Madrid, Spain). The following primary antibodies were applied: anti-mouse aggrecan (AGG) antibody (1/100; ab3778; Abcam) and anti-mouse type II collagen (COL-II) antibody (1/1,000; ab34712; Abcam) in 3% BSA, overnight at 4 °C. The binding signal was detected by chemiluminescence using Horseradish Peroxidase-linked secondary antibodies. EZBlue gel staining reagent (Sigma-Aldrich) was used for tissue protein loading control. Densitometric measurements were normalized relative to total protein presence in femoral condyles with AC using Quantity One software (Bio-Rad) and expressed as arbitrary densitometric units (A.U.) 49,50,52 . Statistical analysis. Each limb was considered as an independent sample. Based on previously published data 27,48 , the sample size was determined on the expected difference to be detected regarding the percentage of cells in columns. By accepting a significance level (alpha) of 5%, and a statistical power of 80%, a pairwise t-test between conditions requires 6 limbs per group to demonstrate an approximate 35% change (70 vs 25) assuming a standard deviation of 5, calculated using the G-power 3.1 software (G-power, Germany) 53 . Owing to the generation limitations for Evc cKO mice, the required number of samples obtained by statistical prediction was assigned to the NO-Evc cKO group (n = 6 limbs). The n was tripled for the DMM-Evc cKO group (n = 18 limbs), to ensure the required number in case of unexpected deaths. During surgery two mice died due to a defective assimilation of anaesthesia, and another one shortly after surgery, remaining n = 12 limbs. The n assigned to NO-WT and DMM-WT groups was 16 limbs, due to lack of survival limitations. However, two mice died a few days after surgery, remaining 12 limbs in the DMM-WT group.
No adjustment for multiple comparisons between groups was made a priori, but to preserve type one error, individual tests were only undertaken following a significant omnibus ANOVA test. Non-parametric tests were used due to a lack of normality based on a Kolmogorov-Smirnov test. Kruskal-Wallis test was used for comparisons between multiple groups prior to Mann-Whitney U-test for comparisons between two groups, adopting a Bonferroni correction. A P-value of less than 0.05 was considered statistically significant. Associations were expressed using Spearman's rank correlation coefficient (rho, ρ). Statistical analyses of data were performed using GraphPad Prism 6.01 for Windows (GraphPad Software, USA). Data were expressed as median with interquartile range (IQR).

Data availability
All data generated or analysed during this study are included in this published article (and its Supplementary Information files).