LINC00473‐modified bone marrow mesenchymal stem cells incorporated thermosensitive PLGA hydrogel transplantation for steroid‐induced osteonecrosis of femoral head: A detailed mechanistic study and validity evaluation

Abstract The pathogenesis of steroid‐induced osteonecrosis of the femoral head (SONFH) involves a glucocorticoid‐induced imbalance of osteogenic and adipogenic differentiation, and apoptosis of bone marrow mesenchymal stem cells (BMSCs). An increasing number of genes, especially noncoding RNAs, have been implicated in the function of BMSCs. Our previous studies have confirmed the key role of LINC00473 and miR‐23a‐3p on the osteogenic, adipogenic differentiation, and apoptosis of BMSCs. However, the underlying mechanism of this process is still unclear. Based on bioinformatics analysis, here we investigated the effects of LINC00473 on the LRP5/Wnt/β‐catenin signaling pathway in the osteogenesis and adipogenesis of BMSCs, as well as the PEBP1/Akt/Bad/Bcl‐2 signaling pathway in dexamethasone‐ (Dex‐) induced apoptosis of BMSCs. Our data showed that LINC00473 could promote osteogenesis and suppress the adipogenesis of BMSCs through the activation of the miR‐23a‐3p/LRP5/Wnt/β‐catenin signaling pathway axis, while rescuing BMSCs from Dex‐induced apoptosis by activating the miR‐23a‐3p/PEBP1/Akt/Bad/Bcl‐2 signaling pathway axis. Notably, we observed that LINC00473 interacted with miR‐23a‐3p in an Argonaute 2 (AGO2)‐dependent manner based on dual‐luciferase reporter assay, AGO2‐related RNA immunoprecipitation, and RNA antisense purification assay. Furthermore, injectable thermosensitive polylactic‐co‐glycolic acid (PLGA) hydrogel loaded with rat‐derived BMSCs (rBMSCs) modified by LINC00473 were used for the treatment of SONFH in a rat model. Our results demonstrated that PLGA hydrogels provided a suitable environment for harboring rBMSCs. Besides, transplantation of PLGA hydrogels loaded with rBMSCs modified by LINC00473 could significantly promote the bone repair and reconstruction of the necrotic area at the femoral head in our SONFH rat model. Surprisingly, compared with the transplantation of BMSCs alone, the transplanted rBMSCs encapsulated within the PLGA hydrogel could migrate from the medullary cavity to the femoral head. In summary, LINC00473 promoted osteogenesis, inhibited adipogenesis, and antagonized Dex‐induced apoptosis of BMSCs. Therefore, LINC00473 could provide a new strategy for the treatment of SONFH.

one hand, glucocorticoid can induce BMSCs to differentiate into osteoblasts and chondrocytes 11,12 but this effect is concentration-and time-dependent. Long-term exposure to high doses of glucocorticoid can disturb the balance of osteogenic and adipogenic differentiation of BMSCs, inhibit cell proliferation, and induce the apoptosis of BMSCs, 13,14 which may be relevant to the pathogenesis of SONFH.
Various signaling pathways are involved in the biological function of BMSCs, such as the Wnt/β-catenin, 15 TGF-β/BMPs, 16,17 and MAPK 18 signaling pathways. However, many upstream genes selectively regulate these pathways that have not been explored with regards to SONFH.
Among these upstream genes, noncoding RNAs (ncRNAs) have been proposed to be critical regulators of downstream gene expression, although they do not code proteins. 19 Long noncoding RNAs (lncRNAs) are ncRNAs with more than 200 nucleotides and are capable of affecting the differentiation, proliferation, and apoptosis of BMSCs. 20 It was reported that osteogenic and adipogenic differentiation of BMSCs could be regulated by several lncRNAs, including lncRNA RP11-154D6, 21 lncRNA TCONS_00041960, 22 and LncRNA LOXL1-AS1. 23 Other lncRNAs, such as lncRNA-p21, 24 lncRNA HULC, 25 and lncRNA NORAD 26 27,28 The regulatory effects of miRNAs on BMSCs depend on their target genes. miR-20b was found to inhibit adipogenic differentiation of BMSCs by binding to its target gene PPAR-γ, and promoting the osteogenic differentiation of BMSCs through the transcription of Runx2. 29 In addition, it was reported that miR-29a could inhibit the expression of DKK1 and activate the Wnt/βcatenin signaling pathway, and thereby promote the osteogenesis of BMSCs. 30 On the contrary, several miRNAs, such as miR-138, miR-139, and miR-133, inhibit the osteogenic differentiation of BMSCs. [31][32][33] Notably, lncRNAs and miRNAs can interact to regulate cellular processes. For instance, lncRNAs can act as competing endogenous RNAs (ceRNAs) to bind to specific miRNAs and regulate their functions. [34][35][36] Despite all this, our current knowledge of the roles of lncRNAs in SONFH remains limited.
According to our previous microarray analysis, miR-23a-3p was identified as an up-regulated differentially expressed miRNA in dexamethasone-induced BMSCs, 37 and was confirmed to target lowdensity lipoprotein receptor-related protein 5 (LRP5) to inhibit osteogenic differentiation of BMSCs through the inactivation of the Wnt/β-catenin signaling pathway. 38 In addition, it has been reported that the inhibition of miR-23a-3p could decrease osteonecrosis incidence in a rat model. 39 We also identified differentially expressed lncRNAs in BMSCs from patients with SONFH, and found that LINC00473 was significantly downregulated and could rescue BMSCs from dexamethasone-induced apoptosis through the activation of the PEBP1/Akt/Bad/Bcl-2 signaling pathway. 40,41 Furthermore, bioinformatics analysis based on the Encyclopedia of RNA Interactomes (ENCORI; http://starbase.sysu.edu.cn/index.php) 42 showed that both miR-23a-3p, LRP5 and PEBP1 are involved in the ceRNA interaction network of LINC00473. Therefore, there may be a cascade reaction between LINC00473 and both miR-23a-3p/LRP5/Wnt/ β-catenin and miR-23a-3p/PEBP1/Akt/Bad/Bcl-2 signaling pathway axes.
In view of this, this study was designed to investigate the above predictions. Our data showed that LINC00473 could bind miR-23a-3p in an Argonaute 2 (AGO2)-dependent manner. It also promoted osteogenic differentiation and suppressed the adipogenic differentiation of BMSCs through the activation of LRP5/Wnt/β-catenin signaling pathway.
LINC00473 was also shown to rescue BMSCs from the Dex-induced apoptosis through the activation of PEBP1/Akt/Bad/Bcl-2 signaling pathway.
Moreover, our previous study has confirmed that LINC00473 overexpression has no effect on the proliferation and cell cycle of human-derived bone marrow mesenchymal stem cells (hBMSCs), 39 which was positive for maintaining the stemness of BMSCs and achieving the best therapeutic effect in the cell therapy. Accordingly, the role of LINC00473 in stem cell therapy for SONFH was further explored in the present study. To our knowledge, stem cell transplantation combined with tissue engineering has attracted increasing interest and shown a promising prospect in the treatment of SONFH. 43,44 Injectable thermosensitive PLGA hydrogels were often used as cell-delivery carrier in the tissue engineering, 45,46 due to the excellent biocompatibility, convenience, and minimally invasive procedure. However, the fate of transplanted stem cells in vivo is still controversial. It is uncertain whether the transplanted cells will spread with blood flow after transplantation, and undergo mutation, apoptosis, or even necrosis. Therefore, we further assessed the effects of the transplantation of PLGA hydrogels loaded with BMSCs modified by LINC00473 on SONFH in a rat model as well as the fate of the transplanted cells. We observed that this method could significantly attenuate the progression of SONFH in vivo. Surprisingly, transplanted rat-derived BMSCs (rBMSCs) encapsulated within the PLGA hydrogel could migrate from the medullary cavity to the femoral head. In summary, our data propose novel aspects for the pathogenesis of SONFH, and provide a reference for the exploration of hip-preserving surgery to treat SONFH.

| The ceRNA interaction network of LINC00473
To explore the regulatory mechanisms of LINC00473 in BMSCs, we constructed the ceRNA interaction network of LINC00473 by predicting its target mRNAs and miRNAs based on data from ENCORI F I G U R E 1 The ceRNA interation network of LINC00473. (a) The network diagram of LINC00473 as a ceRNA (only 50 target mRNAs for miR-23a-3p were selectively included due to the limitation of space); (b) The predicted binding sites of miR-23a-3p on LINC00473; (c) The predicted binding sites of PEBP1 on miR-23a-3p; (d) The predicted binding sites of LRP5 on miR-23a-3p. All data were obtained from the ENCORI database. ceRNA, competing endogenous RNA; ENCORI, encyclopedia of RNA interactomes; LINC00473, LINCRNA-00473; miR-23a-3p, MicroRNA-23a-3p; LRP5, low-density lipoprotein receptor-related protein 5; PEBP1, phosphatidylethanolamine-binding protein 1 ( Figure 1a). A total of 36 miRNA-targeting sites on LINC00473 were identified and, among them, miR-23a-3p was reported to inhibit osteogenic differentiation of BMSCs in our previous study 38 (the predicted binding sites of miR-23a-3p on LINC00473 are shown in Figure 1b, and the correlation between them in 32 types of cancer are shown in Table 1). Furthermore, we detected a total of 3681 mRNAtargeting sites on miR-23a-3p, including LRP5 and PEBP1. We have shown previously that LRP5 is a target of miR-23a-3p and that it inhibits osteogenic differentiation of BMSCs 38 (the predicted binding sites of LRP5 on miR-23a-3p are shown in Figure 1c). Moreover, we have shown that PEBP1 is involved in the antagonism of LINC00473 on Dex-induced apoptosis of BMSCs 41 (the predicted binding sites of PEBP1 on miR-23a-3p are shown in Figure 1d). In addition, we also detected mRNA-targeting sites on LINC00473, including only ZNF639 and C8orf34, not LRP5 and PEBP1. In view of this, there may be a cascade reaction between LINC00473 and both the miR-23a-3p/LRP5/Wnt/β-catenin and miR-23a-3p/PEBP1/Akt/Bad/Bcl-2 signaling pathways.

| Acquisition and identification of hBMSCs
The hBMSCs were isolated from the bone marrow tissue, and were purified and expanded in vitro, according to a previous description. 40 After three passages, the cells showed a fibroblast-like phenotype, whirlpool-like growth, and spindle-shaped morphology T A B L E 1 Correlation between LINC00473 and miR-23a-3p in 32 types of cancer  Figure 3a,b). After the hBMSCs were infected with up-LINC00473 and sh-LINC00473 lentivirus, respectively, green fluorescent probe was detected in more than 90% of cells ( Figure 3a). In addition, the relative expression of LINC00473 in hBMSCs was significantly up-regulated by up-LINC00473 lentivirus (543%) and was down-regulated by sh-LINC00473 lentivirus Our results showed that the overexpression of LINC00473 could promote osteogenic differentiation and suppress adipogenic differentiation of hBMSCs, while LINC00473 knockdown had exactly opposite effects.
2.4 | LINC00473 promoted osteogenic differentiation and suppressed the adipogenic differentiation of hBMSCs through activating the Wnt/β-catenin signaling pathway Subsequently, a blocking experiment was designed to investigate the role of the Wnt/β-catenin signaling pathway in the regulation of LINC00473 on osteogenic and adipogenic differentiation of hBMSCs.
We first confirmed that the overexpression of LINC00473 exerted the same effects above described (Figure 4a Our results showed that LINC00473 promoted osteogenic differentiation and suppressed the adipogenic differentiation of hBMSCs by activating the Wnt/β-catenin signaling pathway. 2.5 | miR-23a-3p is a key regulator of the LINC00473-activated LRP5/Wnt/β-catenin signaling pathway in hBMSCs hBMSCs cells were co-transfected with miR-23a-3p mimics and recombinant lentiviruses overexpressing LINC00473 to investigate the underlying mechanism of LINC00473 on the regulation of the Wnt/β-catenin signaling pathway. The efficiency of transfection was determined by qRT-PCR. After the hBMSCs were transfected with miR-23a-3p mimics, red fluorescent with Cy3 was detected in more than 90% of cells ( Figure 5a). In addition, the relative expression of miR-23a-3p in hBMSCs was significantly up- Due to that the targeted binding between miR-23a-3p and LRP5 had been confirmed in our previous study, 38 the above data indicated that miR-23a-3p regulated the LINC00473-activated LRP5/Wnt/ β-catenin signaling pathway axis in hBMSCs cells.
2.6 | miR-23a-3p is involved in LINC00473activated PEBP1/Akt/Bad/Bcl-2 signaling pathway axis in Dex-induced apoptosis of hBMSCs Based on our previous study, LINC00473 could rescue hBMSCs from Dex-induced apoptosis through the activation of the PEBP1/ Akt/Bad/Bcl-2 signaling pathway axis. 41 Moreover, the previous predictions indicated that there were target sites between miR-23a- The quantification of integrated density in target bands normalized to GAPDH. Data are presented as mean value ± standard deviation of three independent experiments. (b) ** P < 0.01 compared with the Control group and Vector group, and ## P < 0.01 compared with the Control group and sh-Control group. (d,e,g) ** P < 0.01 compared with the Control group, ## P < 0.01 compared with the OS group and OS + Vector group, and && P < 0.01 compared with the OS group and OS + sh-Control group. (i,j,l) ** P < 0.01 compared with the Control group, ## P < 0.01 compared with the AS group and AS + Vector group, and && P < 0.01 compared with the AS group and AS + sh-Control group. ALP, alkaline phosphatase; AS, adipogenic stimulation; BSPII, bone sialoprotein II; CEBP-α, CCAAT/enhancer-binding protein-α; FABP4, fatty acid binding protein 4; hBMSC, human-derived bone marrow mesenchymal stem cells; LINC00473, LINCRNA-00473; OPN3, osteopontin 3; OS, osteogenic stimulation; PPAR-γ, peroxisome proliferator-activated receptor-γ; qRT-PCR, quantitative real-time polymerase chain reaction; Runx-2, runt-related transcription factor-2; TG, triglyceride 3p and LINC00473, as well as miR-23a-3p and PEBP1. Therefore, miR-23a-3p could act as a bridge to connect LINC00473 and PEBP1. In the present study, we used 10 À6 mol/L Dex to induce the apoptosis of hBMSCs, and employed recombinant lentiviruses overexpressing LINC00473 to transfect hBMSCs. Subsequently, cell proliferation from day 1 to 7 was evaluated by the CCK-8 assay, and cell apoptosis was observed by Hoechst 33342 staining and flow cytometry assay for Annexin V-PE/7-AAD double-staining. As Furthermore, the interactions between miR-23a-3p and PEBP1 were verified by a dual-luciferase reporter assay system. The plasmid-PEBP1 vectors were constructed by inserting the sequences of wildand mutant-type (without miR-23a-3p-binding sites) PEBP1 upstream of the luciferase gene ( Figure 6h). We found that miR-23a-3p mimics significantly decreased the luciferase activity of wild-type PEBP1 but not mutant-type PEBP1 or empty vector in hBMSCs, while miR-NC had no effect (Figure 6i), showing that miR-23a-3p could bind to PEBP1 directly.

| LINC00473-targeted miR-23a-3p as a ceRNA
According to our previous results of fluorescent in situ hybridization (FISH), 40 LINC00473 was located in both the cytoplasm and nucleus of hBMSCs, suggesting that it induce a posttranscriptional process. Considering that numerous lncRNAs act as ceRNAs to regulate the distribution of miRNA, we explored whether there was a similar interaction between LINC00473 and miR-23a-3p.
First, we investigated the mutual regulation between LINC00473 and miR-23a-3p. qRT-PCR analysis showed that overexpression of LINC00473 decreased the enrichment of miR-23a-3p, and knockdown of LINC00473 significantly increased the expression of miR-23a-3p in hBMSCs. Furthermore, miR-23a-3p mimics down-regulated the expression of LINC00473, and miR-23a-3p inhibitors up-regulated the expression of LINC00473 in hBMSCs. These data revealed that there was a negative interaction between LINC00473 and miR-23a-3p in hBMSCs (Figure 7a,b).
We employed a dual-luciferase reporter assay system to verify the interactions between LINC00473 and miR-23a-3p. The plasmid-LINC00473 vectors were constructed by inserting the sequences of wild-and mutant-type (without miR-23a-3p-binding sites) LINC00473 downstream of the luciferase gene ( Figure 7c). We found that miR-23a-3p mimics significantly decreased the luciferase activity of wildtype LINC00473 but not mutant-type LINC00473 or empty vector in hBMSCs, while miR-NC had no effect ( Figure 7d). These data demonstrate that miR-23a-3p could bind to LINC00473.
It was shown that the protein AGO2 could bind to miRNAs and mRNAs and regulate the abundance of miRNAs in post-transcription, thereby affecting the biological function of miRNAs. [47][48][49] Therefore, to investigate whether AGO2 was involved in the modulation of miR-34a-5p by LINC00473, we performed AGO2-related RNA immunoprecipitation (AGO2-RIP) to pull down endogenous miRNAs and lncRNAs bound to AGO2 and detected the expression of LINC00473 and miR-23a-3p in the pull-down complex by qRT-PCR. Our results showed that endogenous LINC00473 and miR-23a-3p were specifically enriched by the AGO2 protein in hBMSCs, but not by a control IgG protein ( Figure 7e). Moreover, we conducted AGO2-RIP in hBMSCs transfected with miR-23a-3p mimics. Surprisingly, the enrichment of endogenous LINC00473 pull-down by AGO2 was increased in hBMSCs transfected with miR-23a-3p mimics compared with that of miR-NC ( Figure 7F).
We also performed RNA antisense purification (RAP) to further confirm the specific binding between LINC00473, miR-23a-3p, assessed. As expected, our data indicated that both AGO2 protein and miR-23a-3p were specifically enriched by LINC00473 probes, but not by NC-probes (Figure 7g,h). Taken together, these findings revealed the interaction between LINC00473 and miR-23a-3p in an AGO2-dependent manner.

| LINC00473 promoted osteogenic differentiation and cell migration, and suppressed the adipogenic differentiation and Dex-induced apoptosis of rBMSCs
In consideration of our previous study, 40 indicating that LINC00473 was one of the differentially expressed genes between BMSCs from patients with SONFH and control patients with femoral neck fracture, we intended to perform LINC00473-modified rBMSCs transplantation into rats with SONFH to further reveal the biological function of LINC00473 in vivo. Notably, in vivo experiment for LINC00473 in mice has been reported, 50 although LINC00473 was a primatespecific lncRNA and not normally expressed in rodents. 51 In addition, there was no lack of using rodents to reveal the biological function in vivo for primate-specific lncRNAs. 52 Therefore Interestingly, these results were in line with our findings in hBMSCs. 39 Furthermore, we performed a scratch wound healing assay to observe the effect of LINC00473 on rBMSCs migration. As expected, the rBMSCs overexpressing LINC00473 migrated and recovered into the denuded area in a shorter time compared with control cells (Figure 8j,  We further evaluated the toxicity of PLGA hydrogel on rBMSCs. The effects of a PLGA hydrogel maceration extract on the proliferation and apoptosis of rBMSCs were evaluated by crystal violet and flow cytometry assay, respectively. Continuous treatment with the PLGA hydrogels maceration extract for 7 days did not affect the cell proliferation nor did it induce apoptosis of rBMSCs cells (Figure 9e-h).
F I G U R E 6 Legend on next page.  Increased apoptosis and decreased proliferation of BMSCs induced by long-term exposure to glucocorticoids are considered another potential pathogenesis for SONFH. 13,14 In our previous studies, LINC00473 was shown to rescue hBMSCs from Dex-induced apoptosis through activating the PEBP1/Akt/Bad/Bcl-2 signaling pathway. 41 In the present study, we observed that the overexpression of LINC00473 protected hBMSCs exposed to Dex through the increase of the expression of p-Akt, p-Bad, PEBP1 and Bcl-2, which was in line with our previous reports. 41 Remarkably, these effects of LINC00473 were suppressed by miR-23a-3p mimics. Moreover, the targeted binding between miR-23a-3p and PEBP1 was verified in the present study, suggesting that miR-23a-3p is involved in LINC00473induced activation of the PEBP1/Akt/Bad/Bcl-2 signaling pathway in hBMSCs exposed to Dex.
The above results demonstrated that miR-23a-3p mediated the  miR-23a-3p in hBMSCs. We proved that the specific sequence of miR-23a-3p could bind to the LINC00473 sequence. Interestingly, the Ago2 protein mediates the interaction between lncRNAs and miRNAs. [47][48][49] In this study, RIP and RAP assays revealed that the interaction between LINC00473 and miR-23a-3p occurred in an AGO2-dependent manner.
LINC00473 is reported to be a primate-specific lncRNA and not normally expressed in rodents. 51 In spite of this, the biological function of LINC00473 in mice should still be explored based on the homologous function induced by the overexpression of LINC00473. 50 Therefore, we first investigated the functions of LINC00473 on the differentiation, migration, Dex-induced apoptosis of rBMSCs, and found similar effects to those observed in hBMSCs.
Stem cell transplantation combined with tissue engineering has shown a promising prospect in the treatment of SONFH. 44,53 Injectable thermosensitive PLGA hydrogels are often used as cell-delivery carriers in tissue engineering. 45,46 In this study, PLGA hydrogels showed reliable  60 In our study, we investigated the distribution of rBMSCs in the femur of rats with SONFH after intramedullary transplantation, and found that the transplanted rBMSCs encapsulated within the PLGA hydrogel could migrate from the medullary cavity to the necrotic area of the femoral head, an effect not found in the transplantation of rBMSCs alone. To further reveal whether PLGA hydrogel affects the migration of rBMSCs, we performed a scratch wound healing assay and found that PLGA hydrogels maceration extract had no positive effects on the migration of rBMSCs. Therefore, based on the three-dimensional network structure and the reliable cell biocompatibility, we speculated that PLGA hydrogel could provide a suitable environment for harboring rBMSCs to avoid apoptosis, necrosis, and absorption, which may be beneficial to the homing of rBMSCs.

| Construction of the ceRNA interaction network of LINC00473
The data of the ceRNA interaction network of LINC00473 were obtained by the section of "ceRNA-Network," and the target genes of miR-23a-3p were predicted through the section of "miRNA-Target" in the ENCORI website (http://starbase.sysu.edu.cn/index.php) as well as the predicted binding sites of miR-23a-3p on LINC00473. Briefly, "miRNA-Target" was selected from the navigation bar, and "miRNA-lncRNA" was then selected. Subsequently, "human" was selected in the section of "Genome," and a search for "hsa-miR-23a-3p" was made in the section of "microRNA" to complete the identification of lncRNAs matching with miR-23a-3p. Then, "LINC00473" was typed in the section of "Search" to obtained the binding sites between the miR-23a-3p and the LINC00473, and "Pan-Cancer" was selected to perform the Pan-Cancer analysis across 32 types of cancers. Finally, the network diagram was created using Cytoscape V3.7.2 software (NRNB, USA).

| hBMSCs isolation and culture
The hBMSCs were isolated from the bone marrow tissue as previously reported. 40 The third passage cells were utilized in subsequent experiments. USA) as previously reported. 40 The efficiency of transfection was determined by qRT-PCR after 72 h. Three samples were analyzed in each group and three replicates were made per experiment.

| Osteogenic and adipogenic differentiation of hBMSCs
Third passage hBMSCs were cultured in osteogenic and adipogenic differentiation medium (Fuyuanbio, Shanghai, China) as previously reported. 61

| ALP staining assay
The potential hBMSCs osteogenic differentiation was assessed using an ALP Staining kit (Solarbio, Beijing, China) as previously reported. 61 Three samples were analyzed in each group and three replicates were made per experiment.

| ALP activity assay
The ALP activity of hBMSCs was performed using the ALP Activity Assay kit (Nanjing Jiancheng Bioengineering Institute, Nanjing, China) as previously reported. 61 Three samples were analyzed in each group and three replicates were made per experiment.

| Oil Red O staining assay
The potential hBMSCs adipogenic differentiation was assessed using Oil Red O staining as previously reported. 61 Three samples were analyzed in each group and three replicates were made per experiment.

| TG content assay
Cellular TG content was detected using a Triglyceride Determination kit (Solarbio, Beijing, China) as previously reported. 61 Three samples were analyzed in each group and three replicates were made per experiment.

| Cell proliferation assay for hBMSCs
The proliferation of hBMSCs in the presence of Dex (10 À6 mol/L) was evaluated by a Cell Counting Kit-8 (CCK-8 assay kit) (Solarbio, Beijing, China) as previously reported. 40 Three samples were analyzed in each group and three replicates were made per experiment.

| Assessment of the morphology of apoptotic cells in hBMSCs exposed to Dex
The morphology of apoptotic cells in the presence of Dex (10 À6 mol/ L) was assessed using the chromatin dye Hoechst 33342 kit as previously reported. 40 Three samples were analyzed in each group and three replicates were made per experiment.

| Flow cytometry measurement of apoptosis
Apoptotic cells were analyzed by flow cytometry using an Annexin V-PE/7-AAD kit as previously reported. 40 Three samples were analyzed in each group and three replicates were made per experiment.

| Total RNA isolation and qRT-PCR analysis
Total RNA isolation, reverse transcription, and qRT-PCR were conducted as previously reported. 40 The relative expression of genes was evaluated by the 2 À44Ct method. LINC00473 mRNA was normalized to GAPDH, while miR-23a-3p was normalized to U6. The sequences of primers are listed in Table 2. Three samples were analyzed in each group and three replicates were made per experiment.

| Western blotting analysis
Total protein extraction and western blotting analysis were performed as previously reported. 41 The primary antibodies for OPN3, BSPII, Runx-2, PPAR-γ, CEBP-α, FABP4, β-catenin, LRP5, TCF, and AGO2 were purchased from Cell Signaling Technology  After being eluted and purified, the RNA samples were used to detect the expression of miR-23a-3p and U6 by qRT-PCR.

| rBMSCs isolation and culture
The rBMSCs were isolated from bone marrow in the bilateral femurs and tibias of Sprague-Dawley rats as previously reported. 62 The third passage cells were utilized in subsequent experiments.

| Scratch wound assay
Cells at 5 Â 10 4 cells/well were seeded into 6-well plates after rBMSCs were treated according to the groups, and then starved overnight. Cells were scraped with a tip after monolayer fusion and then incubated under conditions of 5% CO 2 at 37 C. After cells were cultured for 0, 6, and 12 h, the wound widths were observed under an inverted phase contrast light microscope (Olympus CKX41, Japan), and the migration rate was calculated by the ImageJ software (version 1.52u). Three samples were analyzed in each group and three replicates were made per sample.

| Morphological observation of rBMSCs in hydrogels
The PLGA hydrogel was dissolved in PBS at 1:5, and was filtered with

| Preparation of PLGA hydrogel maceration extract
The PLGA hydrogel was dissolved in PBS at 1:5, and was soaked in medium containing 10% FBS under 5% CO 2 and 37 C. After being filtered with a 0.22 μm membrane, the PLGA hydrogel maceration extract was incubated with cells to assess cell proliferation and apoptosis.

| Crystal violet staining assay
The third passage cells were seeded in 96-well plates at (6) Control group: Rats were treated with normal saline (n = 10). All surgeries were performed under anesthesia to minimize the suffering, and the appropriate volume of medullary cavity graft was 0.2 ml. After treatment for 12 weeks, the rats were euthanized by carbon dioxide asphyxiation, and the femoral heads were collected and examined by MRI, micro-CT, HE staining, and IHC.
In addition, 20 rats were randomly divided into two groups (BMSCs + PLGA + SONFH group and BMSCs + SONFH group) and treated as described above. After treatment for 1, 3, 7, 15, 30 days, the rats were euthanized by carbon dioxide asphyxiation, and the femoral heads in each group were collected and examined by immunofluorescence.

| MRI analysis
MRI with a phased-array body coil (Skyra 3.0T MRI System, Siemens, Germany) was performed to determine abnormal signals of the femoral head in each group. The turbo spin-echo T 2 -weighted fat-saturated images in the transverse plane were obtained with the following parameters: repetition time (3000 ms), echo time (38 ms), slice thickness (1.4 mm), interslice distance (0.14 mm), field of view (120 mm), and matrix (384 Â 320 pixels).

| Micro-CT analysis
The femoral heads fixed with 4% paraformaldehyde were analyzed by

| Histological analysis
The femoral heads in each group were fixed in 4% paraformaldehyde and decalcified by 10% ethylenediaminetetraacetic acid. After decalcification for 4 weeks, the specimens were dehydrated and embedded in paraffin. Subsequently, 5 μm paraffin sections in the coronal plane were made by a microtome (Leica, Biocut, German), and slices stained with hematoxylin and eosin to investigate the subchondral structure.
In addition, several sections were deparaffinized and used for the immunohistochemistry analysis of the expression of OPN3, Runx2, PPAR-γ, and CEBP-α. Furthermore, sections obtained from the rats treated for 1, 3, 7, 15, 30 days in BMSCs + PLGA + SONFH group and BMSCs + SONFH group, were used in the immunofluorescence analysis for the expression of luciferase. Photomicrographs were obtained by a panorama scanner (3DHISTECH P250 FLASH, Hungary).

| Statistical analysis
The statistical analysis in this study was conducted by the SPSS 19.0 software (IBM, USA). The one-way analysis of variances was preformed to compare the data of more than two groups. Values are expressed as mean ± standard deviation (SD). The homogeneity test for variance was performed. Tamhane's T2 test was adopted when heteroscedasticity was found. P-values <0.05 were considered significant. The statistics charts were made by GraphPad Prism 8 software (GraphPad, CA, USA).