Abstract

   Rationale: Osteosarcoma (OS) is the most common primary bone tumor with
   a poor prognosis, but the detailed mechanism is still unclear. A
   comprehensive investigation of tumor microenvironment (TME) of OS might
   help find effective anti-tumor strategies. Single-cell transcriptomics
   is a powerful new tool to explore TME. Therefore, this study is
   designed to investigate the TME and gene expression pattern of primary
   and recurrent OS at the single-cell level.

   Methods: The single-cell RNA sequencing and bioinformatic analysis were
   conducted to investigate the cellular constitution of primary,
   recurrent, and lung metastatic OS lesions according to the datasets of
   [33]GSE152048 and [34]GSE162454. TIMER database was used to investigate
   the role of LOX in the prognosis of sarcoma. The functions of related
   cells and markers were further confirmed by in vitro and in vivo
   experiments.

   Results: Cancer associated fibroblasts (CAFs) were found with a higher
   infiltrating level in recurrent OS, and were enriched in the
   epithelial-mesenchymal transition (EMT) pathway. CAFs showed remarkably
   increased expression of LOX, which might lead to EMT and poor prognosis
   of OS. Mechanically, LOX regulated the function of CAFs and macrophage
   polarization to remodel the tumor immune microenvironment. Moreover,
   LOX inhibitor could inhibit migration and promote apoptosis of OS both
   in vitro and in vivo.

   Conclusions: This study revealed the heterogeneity of recurrent OS and
   highlighted an innovative mechanism that CAFs regulate EMT of OS via
   LOX. Targeting LOX of CAFs showed promising efficacy in remodeling TME
   and treating recurrent OS.

   Keywords: Single-cell transcriptomics, Cancer associated fibroblasts,
   Tumor immune microenvironment, Osteosarcoma

Introduction

   Osteosarcoma (OS) is one of the most common bone tumors, and mainly
   affects adolescents [35]^1. In spite of substantial achievements in
   surgeries and chemotherapies, the prognosis of OS patients remains
   unsatisfactory [36]^2^, [37]^3. Recurrent OS is one major clinical
   challenge and recurrent OS patients have a poor prognosis [38]^4.
   Accordingly, it is urgent to explore the mechanisms of OS progression
   and enhance the understanding of recurrent OS thoroughly.

   Tumor microenvironment (TME) mainly consists of various cell types and
   extracellular matrix, that communicate with cancer cells and promote
   cancer progression [39]^5. Cancer-associated fibroblasts (CAFs), as the
   most common cells in TME [40]^6, can lead to cancer progression via
   cancer stem-cell renewal, blunting immunotherapy efficacy, and
   chemoresistance [41]^7^, [42]^8. However, the roles and mechanisms of
   immunosuppressive features of CAFs in OS deserve further investigation.

   Lysyl oxidase (LOX) is a secreted enzyme in the extracellular matrix.
   Collagen and elastin as the substrates of LOX, are the main components
   of bone and lung, indicating that LOX may play a role in OS
   tumorigenesis and metastasis [43]^9. Researchers have found that LOX is
   dysregulated in OS cell lines and confirmed its role in OS
   tumorigenesis and metastasis [44]^10^, [45]^11. However, Xu et al.
   suggested that LOX acts as a tumor suppressor to inhibit the growth of
   OS cells [46]^12. Therefore, the above contradictory outcomes indicate
   that the role of LOX is poorly understood in OS.

   Single-cell RNA sequencing (scRNA-seq) is a powerful new tool to detect
   genetic and functional heterogeneity, reconstruct cellular components
   and their interactions, and detect rare subpopulations of TME [47]^13^,
   [48]^14. ScRNA-seq is applied to identify and analyze the precise
   cellular clusters and compositions of OS [49]^15^, [50]^16. However,
   all these studies had focused on cellular components, but had not
   clarified the roles and mechanisms of TME in OS progression clearly.

   This study utilized scRNA-seq and bioinformatics technology to reveal
   the regulative roles of CAFs in TME of recurrent OS based on previous
   studies of [51]GSE152048 and [52]GSE162454. It might promote the
   understanding of the roles and mechanisms of CAFs in TME of recurrent
   OS at the single-cell level, which is deemed to provide novel
   therapeutic targets for OS.

Materials and methods

Single-cell RNA sequencing and bioinformatic analysis

   The raw scRNA-seq data of [53]GSE152048 and [54]GSE162454 were
   downloaded and used for further bioinformatic analysis. We performed
   the data integration and the dimensionality reduction, cell-clustering
   and annotation, differentially expressed genes (DEGs) identification,
   and GO enrichment analysis, respectively.

Pre-processing of scRNA-seq data

   For each sample, the raw output data were processed in R software using
   the Seurat package ([55]http://satijalab.org/seurat/) [56]^17. We
   filtered out cells that expressed fewer than 300 genes or more than 10%
   mitochondrial genes of total expressed genes [57]^18. The filtered
   cells were then utilized for the following bioinformatic analysis.

Data integration and dimensionality reduction

   The gene expression was calculated for each sample as the gene's
   fraction and multiplied by 10,000. The results were then converted to
   natural logarithms and normalized by adding 1 to prevent taking the log
   of 0. Before we carried out the principal component analysis (PCA)
   based on these highly variable genes (HVGs), we identified, centered,
   and scaled the top 3000 highly variable genes (HVGs) from the
   normalized expression matrix. On the basis of the top 50 PCA components
   found, the batch effects were eliminated by the R Harmony package
   [58]^19.

Patient samples

   This study collected three paired primary and recurrent OS samples in
   our hospital. Each participant provided written informed consent for
   the data to be utilized in our study.

Ethics committee approval and patient consent

   This study was approved by the institutional review board and the
   medical ethics committee of Wuhan Union Hospital, Huazhong University
   of Science and Technology. Each participant provided written informed
   consent for the data to be utilized in our study. The animal study
   protocol was approved by the Committee on the Ethics of Animal
   Experiments of Huazhong University of Science and Technology.

Western blot

   The proteins were transferred to 0.22 μm polyvinylidene fluoride (PVDF)
   membranes (Merck Millipore, USA) and incubated with anti-N-cadherin,
   anti-E-cadherin, anti-LOX, and anti-GAPDH primary antibodies at 4 °C
   overnight. Secondary antibodies (Proteintech, Rosemont, IL, USA) were
   incubated with the following PVDF membranes before imaging.

Cell migration assay

   Cells were resuspended in 100 μl serum-free medium and were seeded on
   the upper chamber of the Transwell chamber plates (BD Biosciences, St
   Louis, MO, USA). The bottom well was used to store 20% FBS medium.
   After incubation at 37 °C for 24h, cells that migrated through the
   pores of the Transwell were fixed and stained.

Wound healing assay

   The treated tumor cells were seeded into 6-well plates to a final
   density of 1×10^5 cells/well and incubated overnight to permit cell
   adhesion and formation of a confluent tumor cell monolayer. Wounds were
   generated by scratching lines with a 200 μl plastic tip in the cell
   monolayer. The wound closures were observed and recorded under a light
   microscope at 0h and 24h after the scratch was generated.

Immunofluorescence (IF) staining

   After high-temperature antigen retrieval, sections were incubated with
   blocking solution for 10 min at 25 °C, and the primary antibodies were
   applied overnight at 4 °C. The tissues were incubated with Polymer
   HRP-anti-mouse/Rabbit IgG secondary antibody. The sections were
   heat-treated after the application of each fluorophore and primary
   antibody, and then incubated with the secondary antibody and another
   fluorophore working solution. Tissue sections were counterstained with
   DAPI for 5 min.

Immunohistochemistry (IHC) staining

   All tissue sections were deparaffinized, rehydrated, and washed.
   Endogenous peroxidase was blocked using 3% hydrogen peroxide for 10min.
   After water-bath heating for antigen retrieval, slides were incubated
   with primary antibodies followed by HRP-linked secondary antibodies and
   diaminobenzidine staining. Counterstaining was done with hematoxylin.
   Slides were dehydrated with sequential ethanol washes.

Subcutaneous xenograft tumor model

   Nude mice (BALB/c, female, 4 to 5-week-old) were injected
   subcutaneously with 5×10^6 MNNG/HOS cells. Tumor volumes were
   calculated with the length (a) and the width (b): volume (mm^3) =
   ab^2/2. The animals were sacrificed, and hematoxylin-eosin (HE)
   staining, IHC staining (Ki67, LOX, N-cadherin, and E-cadherin), IF
   staining (α-SMA, FAP, F4/80, CD163, iNOS, CD3, CD4, CD8), and TUNEL
   staining were performed to stain tumor tissues.

TIMER database analysis

   TIMER database investigates the immune infiltrates in various cancer
   types ([59]https://cistrome.shinyapps.io/timer/). This study assessed
   the cumulative survival curves according to LOX expression in sarcoma
   tissues and the correlation of LOX with the abundance of immune
   infiltrates. Moreover, correlations between the expression of LOX and
   gene markers of epithelial-mesenchymal transition (EMT) (E-cadherin,
   N-cadherin, SNAI1, and S100A4) were explored via correlation modules.

Statistical analysis

   Statistical analyses were conducted by GraphPad 6.0 software.
   Two-tailed student's t-test was utilized to show a significant
   difference between two groups, and a one-way analysis of variance
   (ANOVA) was utilized for three or more groups. A p-value of < 0.05 was
   considered to indicate a statistically significant result (*p < 0.05
   and **p < 0.01).

Results

Cellular constitution of OS tumor lesions

   This study was designed and investigated as shown in Figure [60]1. We
   conducted scRNA-seq analyses to investigate the cellular constitution
   of primary, recurrent, and lung metastatic OS lesions according to the
   datasets of [61]GSE152048 (BC) and [62]GSE162454 (OS). [63]GSE152048
   (BC) includes 100,987 individual cells from 7 primary, 2 recurrent, and
   2 lung metastatic OS lesions. [64]GSE162454 (OS) includes 29,278 cells
   from six primary OS patients prior to neoadjuvant chemotherapy.
   Furthermore, we recruited three paired primary and recurrent OS samples
   to verify the results of scRNA-seq.

Figure 1.

   [65]Figure 1
   [66]Open in a new tab

   Graphical view of the study roadmap.

   After quality control, the cells of [67]GSE152048 (BC) and
   [68]GSE162454 (OS) were clustered into 14 major clusters via the
   T-distributed stochastic neighbor embedding (t-SNE) method (Figure
   [69]2A). Cluster-specific genes were utilized to annotate cell types
   with classic biomarkers. 37 signature gene expressions across the
   cellular clusters were shown by dot plots. The size of dots indicates
   the proportion of cells expressing the particular marker, and the
   spectrum of color represents the mean expressions of the markers
   (Figure [70]2B). Div-cells referred to the diverse cells with MKI67+
   expression, which have enhanced proliferation ability. The t-SNE plots
   of the 14 cell types in [71]GSE152048 (BC) and [72]GSE162454 (OS) were
   shown (Figure [73]2C and [74]2D). The relative proportion of each cell
   cluster in [75]GSE152048 (BC) and [76]GSE162454 (OS) was indicated
   (Figure [77]2E).

Figure 2.

   [78]Figure 2
   [79]Open in a new tab

   Single-cell transcriptomic analysis of OS lesions. (A) The t-SNE plot
   of 14 identified cell types in OS lesions. (B) Dot plots of the 37
   signature gene expressions in 14 cellular clusters (log1p transformed).
   (C, D) The t-SNE plots of the 14 cell types in [80]GSE152048 (BC) and
   [81]GSE162454 (OS). (E) Relative proportion of each cell cluster in
   [82]GSE152048 (BC) and [83]GSE162454 (OS) was indicated.

CAFs had a higher infiltrating level in recurrent OS

   The t-SNE plot showed the 14 identified cell types in primary,
   recurrent, and lung metastatic OS lesions (Figure [84]3A and [85]3B).
   We further investigated the relative proportion of each cell cluster
   across primary, recurrent, and lung metastatic OS lesions. Remarkably,
   compared with primary samples, recurrent samples showed a significantly
   higher infiltrating level of Div-CAFs (96.42% vs 3.58%). No remarkable
   difference in CAFs was shown between primary and recurrent samples.
   Most of the identified cell types exist across patients. But the
   proportion of cell types varied among primary, recurrent, and lung
   metastatic samples because of the remarkable heterogeneity of TME
   (Figure [86]3C).

Figure 3.

   [87]Figure 3
   [88]Open in a new tab

   CAFs had a higher infiltrating level in recurrent OS. (A, B) The t-SNE
   plot of the 14 identified cell types in primary, recurrent, and lung
   metastatic OS lesions. (C) Relative proportion of each cell cluster
   across OS lesions was indicated. (D) The heatmap of GSEA of the 50
   hallmark gene sets in the 14 cell clusters. (E) The ridgeplot of GSEA
   of EMT hallmark gene sets in the 14 cell clusters.

   Moreover, this study investigated the DEGs in cells between recurrent
   and primary OS lesions. According to the pathway enrichment analysis
   via HALLMARK gene sets, the results showed the remarkably activated EMT
   pathway in recurrent OS tissues, especially in the cell cluster of CAFs
   (Figure [89]3D). The outcomes indicated the heterogeneity of expression
   patterns and activated pathways in cells between recurrent and primary
   OS. The ridgeplot of GSEA showed a significant correlation between EMT
   hallmark gene sets and the cell clusters of CAFs (Figure [90]3E).
   Accordingly, CAFs showed a higher infiltrating level and were enriched
   in the EMT pathway of recurrent OS.

CAFs were correlated with EMT in recurrent OS patient tissues

   To further confirm the results of single-cell RNA sequencing and
   bioinformatic analysis, we investigated the role of CAFs in primary and
   recurrent OS patient tissues. Western blot showed that CAFs had higher
   N-cadherin expression and lower E-cadherin expression when compared
   with OS cells (Figure [91]4A). Furthermore, in the tissues of OS
   patients, we further verified that the stromal activation markers of
   CAFs such as recombinant stromal cell derived factor 1 (a-SMA) and
   fibroblast activation protein (FAP) were significantly higher in
   recurrent OS tissues when compared with primary OS tissues via IHC
   staining (Figure [92]4B). By TIMER database, FAP has a significantly
   positive correlation with EMT markers of N-cadherin (r = 0.137, p =
   2.74e-02) and negative association with E-cadherin (r = -0.365, p =
   1.25e-09) (Figure [93]4C). By IHC staining, recurrent OS tissues were
   observed with higher N-cadherin expression and lower E-cadherin
   expression when compared with primary OS tissues (Figure [94]4D). IF
   staining indicated that recurrent OS tissues had higher a-SMA
   expression and lower E-cadherin expression when compared with primary
   OS tissues (Figure [95]4E). Therefore, CAFs were significantly
   correlated with EMT markers in recurrent OS patient tissues.

Figure 4.

   [96]Figure 4
   [97]Open in a new tab

   CAFs were correlated with EMT in recurrent OS patient tissues. (A)
   Western blot showed that CAFs had higher N-cadherin expression and
   lower E-cadherin expression when compared with OS cells. (B) The
   expressions of a-SMA and FAP were observed in primary and recurrent OS
   tissues via IHC staining (scare bar: 100 µm; **p < 0.01). (C) By TIMER
   database, FAP of CAFs has a significantly positive correlation with EMT
   markers of N-cadherin (r = 0.137, p = 2.74e-02) and negative
   association with E-cadherin (r = -0.365, p = 1.25e-09). (D) The
   expressions of E-cadherin and N-cadherin were observed in primary and
   recurrent OS tissues via IHC staining (scare bar: 100 µm). (E)
   Recurrent OS tissues were observed with higher a-SMA expression and
   lower E-cadherin expression when compared with primary OS tissues by IF
   staining (scare bar: 100 µm; **p < 0.01).

Expression pattern and clinical relevance of LOX expressed by CAFs

   We further examined the DEGs in Div-CAFs and CAFs between recurrent and
   primary OS. The results showed that the expression level of LOX was
   remarkably higher in CAFs and Div-CAFs in recurrent OS (Figure [98]5A
   and [99]5B). IHC staining showing the expression of LOX was
   significantly higher in paired recurrent OS samples (Figure [100]5C).
   qPCR result also confirmed it (Figure [101]5D). Cumulative survival
   curves showed that a higher LOX expression was correlated with a
   significantly poorer prognosis of sarcoma via TIMER database (Figure
   [102]5E). LOX has significant positive correlations with infiltrating
   levels of CD8+ T cells (r = 0.16, p = 1.34e-02) and macrophages (r =
   0.209, p = 1.24e-03) in sarcoma via the TIMER database (Figure
   [103]5F). LOX has a significantly negative association with E-cadherin
   (r = -0.288, p = 4.81e-06) and positive correlations with N-cadherin (r
   = 0.148, p = 2.07e-02), SNAI1 (r = 0.262, p = 3.33e-05), and S100A4 (r
   = 0.251, p = 7.61e-05) (Figure [104]5G). Therefore, LOX was associated
   with immune infiltration levels and EMT states of sarcoma, which may be
   a new target for OS treatment.

Figure 5.

   [105]Figure 5
   [106]Open in a new tab

   Expression pattern and clinical relevance of LOX expressed by CAFs. (A)
   The volcano plot indicates DEGs of CAFs between recurrent and primary
   OS. (B) The volcano plot indicates DEGs of Div-CAFs between recurrent
   and primary OS. (C) IHC staining showing the expression of LOX in
   paired recurrent and primary OS samples (scale bars: 50 µm). (D) qPCR
   showing the expression of LOX in paired recurrent and primary OS
   samples (**p < 0.01). (E) Cumulative survival curves according to LOX
   expression in sarcoma tissues via TIMER database. LOX high represented
   a higher LOX expression and showed a significantly poorer prognosis.
   (F) LOX has significant positive correlations with infiltrating levels
   of CD8+ T cells (r = 0.16, p = 1.34e-02) and macrophages (r = 0.209, p
   = 1.24e-03) in sarcoma via the TIMER database. (G) LOX has a
   significantly negative association with E-cadherin (r = -0.288, p =
   4.81e-06) and positive correlations with N-cadherin (r = 0.148, p =
   2.07e-02), SNAI1 (r = 0.262, p = 3.33e-05), and S100A4 (r = 0.251, p =
   7.61e-05).

LOX promoted recurrent OS in vitro

   β-aminopropionitrile (BAPN) is a specific irreversible LOX inhibitor,
   that targets the active sites of LOX or LOXL isozymes [107]^20. Western
   blot showed that LOX inhibition by BAPN could significantly
   downregulate the LOX expression in CAFs. However, no changes in LOX
   expression of OS cells or macrophages were observed (Figure [108]6A).
   Therefore, LOX inhibition might have no direct effects on OS cells or
   macrophages. Furthermore, LOX inhibition contributed to the
   expressional changes in EMT markers such as E-cadherin (Figure [109]6A)
   as well as stromal activation markers of CAFs such as a-SMA and FAP in
   vitro (Figure [110]6B). Therefore, LOX might play a vital role in
   regulating the function of CAFs.

Figure 6.

   [111]Figure 6
   [112]Open in a new tab

   LOX promoted recurrent OS in vitro. (A) Western blot showed that LOX
   inhibition by BAPN could significantly downregulate the LOX expression
   and contribute to the expressional changes in EMT markers such as
   E-cadherin in CAFs. No changes in LOX expression of OS cells or
   macrophages were observed. (B) CAFs in BAPN group had lower expression
   levels of a-SMA and FAP when compared with control group (scale bars:
   50 µm; **p < 0.01). (C) To explore the roles of LOX in the TME, we
   utilized the co-culture system. (D) In the lower chamber, western blot
   showed that BAPN group had lower N-cadherin expression and higher
   E-cadherin expression when compared with control group. The abilities
   of (E, F) migration (scale bars: 50 µm; **p < 0.01) and (E, G) wound
   healing (scale bars: 100 µm) of OS cells were significantly inhibited
   in BAPN group. (H) In BAPN group, the relative proportion of M2
   macrophages (CD163) was significantly decreased (scale bars: 50 µm; **p
   < 0.01).

   To further explore the regulative roles of LOX in the TME of OS, we
   utilized the co-culture system (Figure [113]6C). In the upper chamber,
   CAFs were treated with BAPN. In the lower chamber, we found that OS
   cells of BAPN group had lower N-cadherin expression and higher
   E-cadherin expression when compared with control group (Figure
   [114]6D). Moreover, the abilities of migration and wound healing of OS
   cells were significantly inhibited in BAPN group (Figure [115]6E-G). In
   BAPN group, the relative proportion of M2 macrophages (CD163) was
   significantly decreased (Figure [116]6H). Accordingly, LOX might
   promote recurrent OS via regulating the function of CAFs and macrophage
   polarization.

The anti-OS effects of LOX inhibition in vivo

   The effects of LOX inhibition on subcutaneous xenograft tumor model
   were explored. Mice were treated with PBS (control group), and BAPN
   (BAPN group) for 15 days. Compared with control group, BAPN
   significantly decreased the tumor volume (Figure [117]7A). IHC staining
   showed that the expression of LOX was significantly decreased in BAPN
   group. Furthermore, the lower N-cadherin expression and higher
   E-cadherin were observed in BAPN group (Figure [118]7B). The above
   results suggested that BAPN inhibits the LOX expression and has anti-OS
   effects in vivo.

Figure 7.

   [119]Figure 7
   [120]Open in a new tab

   The anti-OS effects of LOX inhibitor in vivo. (A) Mice were divided
   into control and BAPN groups. The volumes of subcutaneously
   transplanted tumors were detected (**p < 0.01). (B) IHC staining of
   LOX, N-cadherin, and E-cadherin in tumors (scare bar: 50 µm; **p <
   0.01). (C) Relative proportion of CAFs in tissues were shown by IF
   staining (scale bars: 50 µm; **p < 0.01). (D) Relative proportion of
   macrophages in tissues were shown by IF staining (CD163: M2 type of
   macrophages; iNOS: M1 type of macrophages; scale bars: 50 µm; **p <
   0.01, *p < 0.05). (E) Relative proportion of T cells in tissues were
   shown by IF staining (scale bars: 50 µm; n.s.: no significance).

   In the subcutaneous xenograft tumor model, we found that the stromal
   activation markers of CAFs such as a-SMA and FAP in tissues were
   significantly lower in BAPN group via IF staining (Figure [121]7C). In
   BAPN group, the relative proportions of M2 macrophages (CD163) were
   significantly decreased (Figure [122]7D) and the infiltrating levels of
   CD8+ T cells were decreased with no significant change (Figure
   [123]7E). Therefore, LOX inhibition might regulate macrophage
   polarization and remodel the TME of OS in vivo.

Discussion

   As for recurrent tumors, patients often fail to get surgical resection
   and pathological specimens. Therefore, recurrent tumors are mainly
   treated according to the pathological features of primary tumors.
   However, the great differences in TME between primary and recurrent
   tumors have been uncovered [124]^21. Taking the high heterogeneity and
   poor prognosis of recurrent OS into account, it is of great importance
   to reveal the pathological features of recurrent OS as precisely as
   possible.

   With the development of scRNA-seq technology, it holds great promise
   for identifying and analyzing the precise cellular clusters and
   compositions in TME of OS. This study indicated a high CAFs
   infiltration level in recurrent OS. CAFs as the main component of TME,
   contribute to cancer progression [125]^8. CAFs might secrete IL-6 and
   remodel the immunosuppressive TME in esophageal cancer [126]^22.
   Previous researches have indicated that CAFs in TME might reduce the
   immunotherapy response in many cancers [127]^6. Studies revealed that
   CAFs of OS formed three distinct subclusters with novel gene markers
   and functions [128]^15^, [129]^16. In this study, recurrent OS was
   characterized by abundant CAFs infiltration and CAFs were enriched in
   the EMT pathway. Therefore, it provided evidence of OS recurrence
   induced by CAFs.

   To further investigate the underlying mechanisms of CAFs in EMT and
   recurrence of OS, the gene expression pattern of CAFs was detected.
   This study showed that LOX had the highest expression level in CAFs
   among cell types in recurrent OS. Emerging studies have shown that LOX
   upregulation promotes the progression and metastasis of different
   tumors [130]^23. LOX functions in VEGF induction, HIF-1α activation,
   and promoting angiogenesis and EMT process in various tumors, which
   underlines the role of LOX-1 as a potential therapeutic target for
   current antitumor strategies [131]^24. Our study confirmed that LOX
   might contribute to poor prognosis in sarcoma patients. LOX was
   significantly associated with CD8+ T cells and macrophages
   infiltration, and EMT states of sarcoma. In this study, LOX inhibition
   by BAPN could regulate CAFs function and induce macrophage
   polarization, thereby further inhibiting OS (Figure [132]8).

Figure 8.

   [133]Figure 8
   [134]Open in a new tab

   Overview of targeting LOX of CAFs to remodel TME and treat recurrent
   OS.

   In spite of the novel findings in our study, we have to clarify some
   limitations and further studies are warranted to improve them. The
   scRNA-seq data of OS tissues from two public datasets were limited and
   unpaired. We only confirmed the outcomes in three paired primary and
   recurrent OS samples, the outcomes of this study should be further
   validated with larger paired samples. We utilized the LOX inhibitor to
   investigate the role of LOX in OS both in vivo and in vitro. More
   underlying mechanisms and clinical trials deserve further investigation
   to bring novel insights into the treatment of recurrent OS.

Conclusions

   In summary, this study investigated the heterogeneity of recurrent OS
   and highlighted an innovative mechanism that CAFs regulate EMT of OS
   via LOX. Targeting LOX of CAFs showed promising efficacy in remodeling
   TME and treating recurrent OS. Our study indicates a new mechanism of
   CAFs in regulating tumor immune microenvironment and sheds light on a
   unique approach to improving the prognosis of recurrent OS.

Supplementary Material

   Supplementary material.
   [135]Click here for additional data file.^ (773.6KB, rar)

Acknowledgments