Abstract

Background

   Osteosarcoma is a highly aggressive cancer, and the efficacy of
   existing therapies has plateaued. Multiomics integration analysis can
   identify novel therapeutic targets for various cancers and therefore
   shows potential toward osteosarcoma treatment. This study aimed to
   leverage multiomics integration to develop a new risk model,
   characterizing the immune features of osteosarcoma to uncover novel
   therapeutic targets.

Methods

   Metabolomics profiling was conducted to identify key metabolites in
   osteosarcoma. Transcriptomic sequencing datasets were analyzed to
   identify prognostic genes related to key metabolic pathways and develop
   a prognostic risk model. Patients were then divided into high-risk and
   low-risk groups with distinct clinical outcomes based on the risk
   model. The single-sample gene set enrichment analysis, Estimation of
   Stromal and Immune cells in Malignant Tumor tissues using Expression
   data (ESTIMATE) algorithm, and xCell algorithms were used to evaluate
   the immune cell infiltration and activity. Single-cell RNA sequencing
   was used to explore cell-to-cell interactions within the tumor
   microenvironment. In vitro coculture functional assays were performed
   to validate the role of macrophage migration inhibitory factor (MIF) in
   macrophage polarization and chemotaxis. In vivo studies were used to
   evaluate the effectiveness of MIF inhibition in combination with immune
   checkpoint blockade in murine models.

Results

   Elevated lactate levels in osteosarcoma patients correlated with poorer
   overall survival. We identified SLC7A7 and CYP27A1 as prognostic
   lactate metabolism genes and developed a risk model to stratify
   patients into high-risk and low-risk groups with distinct outcomes.
   Bioinformatics analyses highlighted the differences in immune
   infiltration patterns and activity between the groups. Notably, the
   infiltration and phenotype of macrophages varied significantly between
   the groups, and MIF was identified as a critical mediator in this
   process. In osteosarcoma cells, lactate regulated MIF expression
   through histone H3K9 lactylation. Combining the MIF inhibitor 4-IPP
   with a programmed cell death 1 (PD-1) monoclonal antibody treatment
   demonstrated a significant antitumor effect.

Conclusion

   MIF acts as a novel therapeutic target by regulating macrophage
   polarization and chemotaxis. Lactate regulated MIF expression through
   histone lactylation. Targeting MIF holds promise for enhancing the
   efficacy of anti-PD-1 treatment.

   Keywords: Immune modulatory, Macrophage, Immunotherapy, Bone Cancer
     __________________________________________________________________

WHAT IS ALREADY KNOWN ON THIS TOPIC

     * Immune checkpoint inhibitors show limited efficacy in osteosarcoma
       treatment, which was due to the immunosuppressive tumor
       microenvironment.

WHAT THIS STUDY ADDS

     * Lactate was correlated with macrophage migration inhibitory factor
       (MIF) expression in osteosarcoma and combined treatment with MIF
       inhibitor and programmed cell death 1 (PD-1) demonstrated a
       significant antitumor effect in osteosarcoma.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

     * This study offered a rationale for combining MIF inhibitors with
       PD-1, which may influence the clinical trials and treatment for
       osteosarcoma.

Introduction

   Osteosarcoma is a highly malignant bone tumor predominantly affecting
   children and adolescents. The application of neoadjuvant chemotherapy
   combined with surgical treatment significantly improves the 5-year
   overall survival rate of patients with osteosarcoma. However, the
   efficacy of these conventional therapies for osteosarcoma has reached a
   plateau, with no significant recent improvements. Approximately 35% of
   patients develop recurrence and metastasis during treatment, leading to
   an unfavorable prognosis with an overall 5-year survival rate of less
   than 25%.[47]^1 2 Consequently, new promising treatments are required
   to enhance the long-term survival of patients with osteosarcoma.

   Cancer treatment has been revolutionized in the last decade with the
   advent of immune checkpoint inhibitors (ICIs) that show remarkable
   clinical responses across various tumor types.[48]^3 4 Regrettably,
   several phase II trials on the programmed cell death 1 (PD-1) inhibitor
   pembrolizumab in metastatic or advanced osteosarcoma demonstrated that
   only approximately 5% of patients achieved a significant response,
   highlighting its limited effectiveness.[49]^5 6 Osteosarcoma has been
   characterized as a “cold” tumor because of insufficient immune
   infiltration and lack of tumor immunogenicity.[50]^7 These factors
   limit the potential for meaningful responses to ICIs alone.
   Furthermore, identifying reliable biomarkers to guide enhanced ICI
   treatment continues to be a challenge.

   Accumulating evidence highlights cancer as a metabolic malady and
   metabolic reprogramming in cancer and immune cells is therefore
   essential for modulating the antitumor immune response.[51]^8 This
   modulation occurs through releasing metabolites that can affect immune
   molecule expression and effector function execution.[52]^9 The
   integration of metabolomics with other multiomics approaches, such as
   transcriptomics, has been widely adopted to unravel the intricate and
   key immune interactions within the tumor microenvironment
   (TME).[53]10,[54]12 This comprehensive approach has been proved to be
   valuable in revealing novel biomarkers that may serve as targets to
   enhance the efficacy of ICI therapies.[55]^13 14

   To characterize osteosarcoma immune features and discover biomarkers
   for improving immunotherapy, we employed a multiomics approach
   combining metabolomics, bulk transcriptomics, and single-cell
   transcriptomics. Metabolomics profiling revealed elevated lactate
   levels in patients with osteosarcoma, which correlated to poorer
   prognosis. Subsequently, we developed a risk model incorporating
   prognostic lactate metabolism-related genes (LMRGs) SLC7A7 and CYP27A1.
   The integration of bulk and single-cell transcriptomics data into our
   risk model identified disparities in immune cell infiltration, and
   notable differences were observed in macrophage polarization and
   chemotaxis, regulated by macrophage migration inhibitory factor (MIF).
   Lactate regulated MIF expression through histone H3K9 lactylation in
   osteosarcoma cells. Consequently, we found that combining the MIF
   inhibitor 4-IPP with a PD-1 monoclonal antibody exerted a notable
   antitumor effect. This study offers new insights into the immune
   heterogeneity of osteosarcoma and may inform more targeted treatment
   strategies.

Methods

Serum non-targeted metabolomics analysis

   Previously, we applied serum non-targeted metabolomics using the
   ultrahigh-performance liquid chromatography coupled with
   high-resolution mass spectrometry platform to assess samples from 65
   patients with osteosarcoma and 30 healthy adults.[56]^15 Metabolomics
   profiling in this study was based on further analysis of those data.
   Partial least squares discriminant analysis and metabolic pathway
   enrichment analysis were performed using MetaboAnalyst (widely used
   web-based platform dedicated for comprehensive metabolomics data
   analysis, interpretation and integration with other omics data.
   Website: [57]https://www.metaboanalyst.ca/.

Bulk transcriptome sequence and public bulk transcriptome dataset download

   We used proprietary transcriptome data comprizing 45 osteosarcoma
   tissue samples and 25 paired peri-osteosarcoma tissue samples, which
   had been collected and processed at the First Affiliated Hospital of
   Sun Yat-sen University. We defined this as the Sun Yat-Sen Cohort (SYS
   cohort). For transcriptome sequencing, TRIzol was used to extract RNA
   from tumor tissues. For the qualified total RNA samples, 1–3 µg of
   total RNA was used as the starting material for each sample to
   construct a transcriptome sequencing library. Sequencing was performed
   using the BGISEQ-500 platform by running a paired-end sequencing
   program (PE), and 150 bp PE sequencing reads were obtained.

   For public bulk transcriptome datasets, the Therapeutically Applicable
   Research to Generate Effective Treatment-Osteosarcoma (TARGET-OS)
   dataset (n=85) was obtained from the UCSC Xena browser
   ([58]http://xena.ucsc.edu/). The Gene Expression Omnibus (GEO) dataset
   [59]GSE21257[60]^16 (n=53) was downloaded from the National Center for
   Biotechnology Information ([61]https://www.ncbi.nlm.nih.gov/). The
   Chromatin Immunoprecipitation sequence (ChIP-seq) and CUT&Tag
   data[62]17,[63]24 of histone lactylation was also downloaded from GEO.

Single-cell transcriptome sequence data download and processing

   A single-cell RNA-sequencing (scRNA-seq) dataset comprizing six
   treatment-naïve osteosarcoma samples was obtained from GEO datasets
   ([64]GSE152048).[65]^25 Quality control (QC) and subsequent analysis
   were conducted using the “Seurat” R package (a widely used and powerful
   R package designed for QC, analysis, exploration and of scRNA-seq data)
   following standard protocols. Cells with <300 genes or >10%
   mitochondrial content were excluded as were genes detected in less than
   three cells. After filtering out housekeeping and mitochondrial genes,
   the study encompassed 25,640 cells.

   We pinpointed the 2,000 most variable genes for data normalization.
   Principal component analysis was applied to distil the data into 30
   principal components, and the Harmony package was used to mitigate
   batch effects. Cells were visualized in a two-dimensional space via
   t-distributed stochastic neighbor embedding with the uniform Manifold
   Approximation and Projection (uMAP) algorithm. Seurat’s FindAllMarkers
   function identified signature genes for each cluster, aiding in cell
   type characterization. The FindAllMarkers function was employed with a
   logarithmic fold change threshold of 0.25 (logfc.threshold=0.25) and
   set to identify only positive markers (only.pos=TRUE) to detect markers
   for each cluster. CopyKAT (a widely-used and powerful tool identifying
   genome-wide aneuploidy in single cells, which can separate tumor cells
   from normal cells, and tumor subclones rapidly using high-throughput
   scRNA-seq data) was used to assess the benign or malignant nature of
   cells.

Prognostic LMRG identification

   We used the term “lactic” as the search keyword in the Molecular
   Signatures Database (MSigDB;
   [66]https://www.gsea-msigdb.org/gsea/msigdb/index.jsp) and determined
   five priority lactate related gene (LRG) sets, namely, Gene Ontology
   Biological Process (GOBP) lactate metabolic process, human phenotype
   (HP) increased serum lactate, HP lactic acidosis, HP lactic aciduria,
   and HP severe lactic acidosis. All gene sets used in this research were
   listed in [67]online supplemental table 1. After deleting duplicates,
   284 LMRGs were identified. The differential expression of LMRGs was
   assessed between normal and tumor groups in the SYS cohort using the
   “limma” package in R, with significance defined as p<0.05 and
   |log2FC|≥0.50. Significant differentially expressed LMRGs underwent
   subsequent univariate Cox regression to ascertain their association
   with overall survival. Variables with p<0.05 were evaluated by Least
   Absolute Shrinkage and Selection Operator (LASSO) analysis via the
   “glmnet” package (a widely used and convenient package that fits
   generalized linear and similar models), identifying candidate genes
   with the optimal penalty parameter λ determined by the 1—SE criterion.
   Finally, multivariate Cox analysis was used to determine the prognostic
   LMRGs for the construction of the optimal lactate metabolism-related
   risk model based on a minimum Akaike information criterion.

Lactate metabolism risk model construction and related immune landscape
analysis

   Risk scores were calculated by aggregating the expression levels and
   the corresponding coefficient of each prognostic LMRG. Risk scores were
   calculated using the following formula that incorporates the expression
   levels and the corresponding coefficient of two key genes: Risk
   score=SLC7 A7 × (−0.2948) + CYP27 A1 × (−0.3376). The “survminer”
   package (a widely used and powerful package that provides functions for
   facilitating survival analysis and visualization) was used for survival
   analysis and visualization. The risk group was stratifying patients
   into low-risk and high-risk groups according to the risk score. In
   addition, time-dependent receiver operating characteristic curves were
   used to assess the efficacy of the model in predicting 2-year, 3-year,
   and 5-year survival for patients. The [68]GSE21257 cohort served as an
   external validation set for the generalizability.

   The levels of cancer-related hallmarks raised by Hanahan and
   Weinberg,[69]^26 such as “Cell cycle,” “Hypoxia,” and
   “Epithelial–mesenchymal transition,” were quantified using the
   single-sample gene set enrichment analysis (ssGSEA) algorithm (a widely
   used and comprehensive tool to characterize cell state in terms of the
   activity levels of biological processes and pathways rather than
   through the expression levels of individual genes); this analysis was
   based on transcriptome data and gene sets from MSigDB and published
   literature. The ESTIMATE algorithm was employed to assess the
   infiltration ratio of tumor, immune, and stromal cells across all
   samples. In addition, the ssGSEA algorithm was used to evaluate the
   abundance of 29 immune cells and immune-related molecules. Then, the
   components of TME cells were analyzed using the xCell algorithm as
   previously reported. Intercellular communication analysis was conducted
   using the CellChat package (a widely used and convenient R package
   designed for inference, analysis, and visualization of cell-cell
   communication from single-cell data). Cancer Single-cell State Atlas
   (CancerSEA) was performed to assess 14 functional states of single-cell
   data from cancer cells.[70]^27 These states encompass stemness,
   invasion, metastasis, proliferation, epithelial–mesenchymal transition,
   angiogenesis, apoptosis, cell cycle, differentiation, DNA damage, DNA
   repair, hypoxia, inflammation, and quiescence.

Immunohistochemistry

   Immunohistochemistry (IHC) staining was performed on 5 mm sections of
   paraffin-embedded osteosarcoma tissue samples. Briefly, slides were
   deparaffinized, treated for antigen retrieval, and then
   catalase-inactivated with 3% H[2]O[2]. Slides were then incubated
   overnight at 4°C with primary antibodies, including anti-CYP27A1
   (1:1000, Cat. 14 739–1-AP, Proteintech), anti-SLC7A7 (1:1000, Cat.
   ab126785, Abcam), anti-MIF (1:500, Cat. 87501, Cell Signaling
   Technology), anti-CD68 (1:1000, Cat. ab303565, Abcam), anti-CD206
   (1:2000, Cat. 81 525–1-RR, Proteintech), and anti-CD8α (1:400, Cat.
   85 336S, Cell Signaling Technology). Antirabbit/mouse–horseradish
   peroxidase conjugate was then added as the secondary antibody, and
   immunoreactive proteins were detected using 3,3ʹ-diaminobenzidine.
   Finally, sections were counterstained with hematoxylin. The score for
   MIF, SLC7A7, and CYP27A1 staining was based on the integrated staining
   intensity and the proportion of positive cells. IHC-positive staining
   of CD68, CD206, and CD8α was used for the calculation of positive cell
   numbers per high field. All percentages/numbers of positive cells were
   expressed as the average of five randomly selected microscopic fields.

Immunofluorescence

   Cells were seeded in chamber slides, then fixed with 4%
   paraformaldehyde, and permeabilized with 0.1% Triton X-100 for 15 min.
   Cell samples were then blocked with 5% bovine serum albumin (BSA) for
   1 hour at 25°C, followed by incubation with primary antibodies
   overnight at 4°C, including anti-iNOS (1:500, Cat.18 985–1-AP,
   Proteintech) and anti-Arg1 (1:800, Cat.66 129–1-lg, Proteintech).
   Slides were then washed three times with phosphate-buffered saline
   (PBS) and incubated with secondary antibodies for 30 min at 25°C.
   Nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI) for
   visualization. Digital images were taken using a fluorescence
   microscope (Nikon, Japan).

Western blotting analysis

   Total cellular proteins were extracted with radio immunoprecipitation
   (RIPA) lysis buffer (Biosharp) supplemented with protease and
   phosphatase inhibitors. Equal amounts of cellular proteins were
   resolved via sodium dodecyl sulfate–polyacrylamide electrophoresis
   followed by electrotransfer onto the 0.2 µm polyvinylidene fluoride
   membrane for 45 min at 4°C and 100 V. Subsequently, membranes were
   incubated overnight at 4°C with the specific primary antibody
   (anti-MIF, 1:2000, Cat. ab175189, Abcam). After washing the membranes
   three times with TBST, membranes were incubated with the corresponding
   secondary antibodies. The immunoreactive signal was visualized using an
   enhanced chemiluminescence kit.

RNA isolation, reverse transcription, and real-time PCR

   Total RNA from cultured cell lines was isolated using the TRIzol
   reagent (Takara), and complementary DNA was synthesized using primers
   and the Takara PrimeScript reverse transcription reagent kit. Real-time
   PCR was performed using the RT-qPCR SYBR Green Kit (Takara) with a 7500
   Fast Real-Time PCR System (Applied Biosystem). GAPDH expression was
   used as an internal reference. The following primers were used:
   CD86-RE, 5ʹ-CTGCTCATCTATACACGGTTACC-3ʹ and

   5ʹ-GGAAACGTCGTACAGTTCTGTG-3ʹ; CD163-RE, 5ʹ-TTTGTCAACTTGAGTCCCTTCAC-3ʹ
   and 5ʹ-TTTGTCAACTTGAGTCCCTTCAC-3ʹ; ARG1-RE, 5ʹ-TGGACAGACTAGGAATTGGCA-3ʹ
   and 5ʹ-CCAGTCCGTCAACATCAAAACT-3ʹ; IL10-RE,
   5ʹ-GACTTTAAGGGTTACCTGGGTTG-3ʹ and 5ʹ-TCACATGCGCCTTGATGTCTG-3ʹ; iNOS-RE,
   5ʹ-AGGGACAAGCCTACCCCTC-3ʹ and 5ʹ-CTCATCTCCCGTCAGTTGGT-3ʹ; and TNFa-RE,
   5ʹ-CCTCTCTCTAATCAGCCCTCTG-3ʹ and 5ʹ-GAGGACCTGGGAGTAGATGAG-3ʹ.
   MIF-ChIP-RE 5ʹ- CGTCACAAAAGGCGGGACCACA-3ʹ and
   5ʹ-CTTCCTGTCCCCTCCCGGCAAA-3ʹ

ELISA assay

   The concentration of MIF, CXCL9 and CXCL10 in the culture media was
   detected using an ELISA kit (Cat.E-EL-H6170, Cat.E-EL-H6062, Cat.
   E-EL-H0050 Elabscience). Culture media was centrifuged at 1000 rcf for
   10 min, and the ELISA assay was performed as per the provided protocol.
   The average of three independent experiments was used in the analysis.

Macrophage generation and differentiation

   THP-1 cells were pretreated with 100 ng/mL Phorbol 12-myristate
   13-acetate (PMA) for 24 hours (MedChemExpress, USA) to generate M0
   macrophages. To mimic the generation of tumor-associated macrophages
   (TAMs), M0 macrophages were cocultured with osteosarcoma cells in a
   24-well Transwell coculture system (0.4 µm pore size, Corning, USA),
   where 2.5×10^5 THP-1 cells were seeded in the bottom well while
   2.5×10^5 osteosarcoma cells were seeded in the upper chamber. After
   48 hours, cocultured macrophages were collected to obtain TAMs for
   further analysis.

Phalloidin-staining assay

   After discarding media, cells were fixed with 4% paraformaldehyde for
   15 min and then washed with PBS three times. Cells were then incubated
   with 0.1% Triton X-100 for 15 min to increase permeability. Finally,
   cells were stained with 200 nM FITC-Phalloidin, sealed with DAPI
   Fluoromount-G, and observed under a laser scanning microscope (Nikon,
   Tokyo, Japan).

Migration assay

   THP-1 cells (1×10^5) were suspended in a serum-free medium and seeded
   into the upper chambers of 24-well inserts (8.0 µm), and osteosarcoma
   cells were added to the lower chambers. Cells were co-cultured for
   48 hours, and migrated macrophages were fixed with 10% formaldehyde,
   stained with crystal violet, and counted under a microscope.

Isolation and activation of peripheral blood mononuclear cells

   Peripheral blood mononuclear cells (PBMCs) of three healthy donors were
   isolated via Ficoll–Hypaque density gradient centrifugation and then
   cultured in Roswell Park Memorial Institute (RPMI)-1640 medium
   supplemented with 10% fetal bovine serum (FBS). In addition, T cells in
   PBMCs were activated by adding 25 µL/mL of human CD3/CD28 T-cell
   activator (Cat.10991, STEMCELL Technologies) in combination with
   1.5 ng/mL IL-2 (Cat.791902, BioLegend) for 72 hours at 37°C.

T-cell activation assay

   Macrophages were differentiated from THP-1 cells and precultured with
   osteosarcoma cells before coculturing with T cells. Following 48 hours
   of coculture, osteosarcoma cells were removed, and preactivated T cells
   were seeded in six-well plates at 1×10^6 cells per well. After an
   additional 48 hours of coculture, suspended T cells were harvested and
   resuspended in a staining buffer (Dulbecco's Phosphate-Buffered Saline,
   DPBS containing 3% FBS) along with the indicated antibodies: APC
   antihuman CD8 (Cat.344721, BioLegend), PE antihuman/mouse granzyme B
   (Cat.372207, BioLegend), and Brilliant Violet 605 antihuman
   interferon-gamma (IFN-γ) (Cat.502535, BioLegend), followed by flow
   cytometry analysis.

T-cell proliferation assay

   Preactivated T cells were labeled with fluorescent violet dye and then
   cultured with macrophages for 1 week in RPMI-1640 medium with 10% FBS,
   100 mM L-glutamine, and 100 mM sodium pyruvate. T cells in suspension
   were harvested and resuspended in the staining buffer (DPBS containing
   3% FBS). The violet or mean fluorescence intensity was detected as an
   indicator of T-cell proliferation via flow cytometry.

ChIP-qPCR assay

   The ChIP assays were performed using the Pierce Magnetic ChIP kit
   (Thermo Fisher Scientific, #26157) with Anti-L-Lactyl-Histone H3 (Lys9)
   (PTM-1419RM) antibody. In brief, cells were incubated with 1%
   formaldehyde in PBS for 10 min, then washed two times with PBS. Cells
   were lysed for 10 min at 4°C in lysis buffer with protease inhibitors,
   followed by centrifugation at 12,000 rpm for 10 min at 4°C to remove
   debris. Samples were incubated overnight at 4°C with the antibody.
   Immunoprecipitated complexes were enriched using protein A/G Sepharose.
   DNA from immunoprecipitates was amplified by PCR and analyzed
   quantitatively with qPCR.

Animal models

   All animal experiments were conducted in accordance with established
   guidelines for the Use and Care of Laboratory Animals and approved by
   the After receiving approval from the Institutional Animal Care and Use
   Committee of Guangzhou LingFu TopBiotech (LFTOP-IACUC-2024–0069). All
   mice were randomly divided into per groups. The murine osteosarcoma
   K7M2 cell line was purchased from Zhongqiaoxinzhou Biotechnology
   (Shanghai, China). A tumor-bearing mouse model was established by
   inoculating 2×10^6 K7M2 cells with Matrigel matrix (Sigma-Aldrich
   E1270) into the right flank of 4-week-old female BALB/c mice.
   Macrophages were depleted via tail-vein injection, once every 3 days
   with 200 µL PBS (control) or clodronate liposome (5 mg/mL,
   Cat.CP-005–005, Liposoma, Holland). Mice were intraperitoneally
   administered a vehicle control (5% dimethyl sulfoxide/40% polyethylene
   glycol 300/5% Tween-80/45% saline) or 4-IPP at doses of 10 mg/kg once
   every 3 days. Mice were monitored every 3 days for 3 weeks. Five mice
   from each group were euthanized at random when tumors attained an
   approximately spherical shape. Subcutaneous and xenograft tumors were
   harvested, photographed, and weighed. Tumor size was calculated using
   the formula width[71]^2×length × π/6 as described previously.[72]^28

Statistical analysis

   Statistical analyses and graph visualization were conducted using R
   V.4.2.1 ([73]http://www.r-project.org) or GraphPad Prism (V.9.0). All
   experiments were performed in triplicate, and representative data from
   one experiment are presented. Statistical analyses were conducted using
   Student’s t-test and one-way analysis of variance with Dunnett’s
   multiple comparisons test or Tukey’s multiple comparisons test. Overall
   survival was analyzed using Kaplan-Meier (KM) analysis and compared
   using the log-rank test. Differences with p<0.05 were considered
   significant.

Results

Elevated lactate predicts poor prognosis in patients with osteosarcoma

   To investigate prognostic metabolic signature of osteosarcoma, we
   analyzed the metabolomics data of serum samples from osteosarcoma
   patients and healthy individuals in our previous study.[74]^15
   Integrating clinical information from the patients, we identified
   significant differences in serum metabolite profiles between patients
   with osteosarcoma and healthy adults ([75]figure 1A), with lactate
   showing the highest Variable Importance in Projection (VIP) score among
   120 analyzed metabolites ([76]figure 1B). The pathway enrichment
   analysis revealed lactate metabolism pathways—particularly the Warburg
   effect, pyruvate metabolism, and gluconeogenesis—were significantly
   enriched in patients with osteosarcoma ([77]figure 1C). Serum lactate
   levels significantly elevated in patients with osteosarcoma and
   positively correlated with tumor size (R^2=0.3379, p<0.0001)
   ([78]figure 1D and E). Patients with elevated lactate levels had
   significantly reduced 5-year overall survival compared with those with
   lower levels ([79]figure 1F). Multivariate analysis confirmed that the
   serum lactate level was an independent prognostic indicator for adverse
   outcomes in patients with osteosarcoma ([80]figure 1G). Further
   quantification of lactate in fresh tumors and adjacent tissues from a
   cohort of 30 patients with osteosarcoma revealed a significant increase
   in lactate within tumor tissues (p<0.001) ([81]figure 1H). Similarly,
   lactate levels positively correlated with tumor size (R^2=0.4074,
   p=0.0001; [82]figure 1I). Multivariate Cox regression analysis
   identified hypoxia as a significant risk factor for overall survival
   among various hallmarks of cancer defined by both Hanahan and Weinberg
   in the TARGET-OS dataset ([83]figure 1J). Consistently, GSEA analysis
   revealed that glycolysis—a lactate metabolism-related pathway—was
   enriched, whereas oxidative phosphorylation was downregulated in
   osteosarcoma tissues compared with adjacent tissues based on the SYS
   cohort ([84]figure 1K). These results suggested that elevated lactate
   levels serve as a prognostic indicator and reflect enhanced lactate
   metabolism in patients with osteosarcoma.

Figure 1. Elevated lactate levels in patients with osteosarcoma correlate
with poor prognosis. (A) Scores plot from orthogonal partial least squares
discriminant analysis showing differences in serum metabolite patterns
between patients with osteosarcoma and healthy adults. (B) VIP scores
indicate the serum metabolite differences between osteosarcoma patients and
healthy adults. (C) Enrichment analysis of serum metabolites between patients
with osteosarcoma and healthy adults. (D) Comparison of serum lactate levels
between patients with osteosarcoma and healthy adults. (E) Correlation
between serum lactate levels and tumor size. (F) Overall survival analysis
was generated for patients stratified according to serum lactate levels,
grouping based on the median of lactate (G) Multivariate analysis of
prognostic factors for overall survival in patients with osteosarcoma. (H)
Comparison of lactate levels in osteosarcoma tumors and adjacent tissues
using a colorimetric assay (n=30). (I) Correlation analysis between the
lactate concentration in osteosarcoma tissues and tumor size. (J)
Multivariate Cox regression analysis among various hallmarks of cancer
defined by both Hanahan and Weinberg in the TARGET-OS dataset. (K) GSEA
analysis of glycolysis and oxidative phosphorylation pathways between the
tumor and normal tissues based on the SYS cohort. Data are presented as the
mean±SD, ***p<0.001 and ****p<0.0001, by Student’s t-test (D and H). CMPF,
2-(2-carboxyethyl)-4-methyl-5-propylfuran-3-carboxylic acid; HA, healthy
adult;EMT, epithelial–mesenchymal transition; GSEA, gene set enrichment
analysis; NES, normalized enrichment score; OPLS-DA, orthogonal partial least
squares discriminant analysis; OS, osteosarcoma; SYS cohort, Sun Yat-Sen
Cohort; TARGET-OS, Therapeutically Applicable Research to Generate Effective
Treatment-Osteosarcoma; VIP, variable importance in projection.

   [85]Figure 1
   [86]Open in a new tab

The development and validation of a risk score model based on LMRGs

   To further investigate the role of lactate metabolism in osteosarcoma
   prognosis, we constructed a prognostic model prioritizing LMRG. First,
   five LMRG sets, comprizing 284 genes, were collected from the MSigDB
   database. Principal component analysis using these genes successfully
   distinguished osteosarcoma tissues from normal controls in the SYS
   cohort ([87]online supplemental figure S1A). Among these genes, 68
   LMRGs exhibited significant differential expression ([88]figure 2A and
   [89]online supplemental figure S1B). Univariate Cox regression analysis
   identified eight genes significantly associated with overall survival
   in the TARGET-OS dataset ([90]figure 2B, [91]online supplemental figure
   S1C and D). LASSO regression combined with multivariate Cox analysis
   further narrowed down the candidate genes, identifying CYP27A1 and
   SLC7A7 as the most suitable genes for constructing a lactate
   metabolism-related prognostic risk score model ([92]figure 2C). KM
   analysis further confirmed that patients with low expression levels of
   SLC7A7 and CYP27A1 were associated with poor overall survival
   ([93]figure 2C and [94]online supplemental figure S1E). IHC staining of
   62 osteosarcoma specimens from our department consistently demonstrated
   that lower expression of these genes correlated with poorer 5-year
   overall survival and lung metastasis-free survival ([95]figure 2D–K).

Figure 2. Identification of prognostic lactate metabolism-related genes and
construction of a risk model. (A) Volcano plots showing differentially
expressed lactate metabolism-related genes between osteosarcoma and normal
tissues in the SYS cohort, with important prognostic genes labeled. (B)
Univariate Cox regression analysis in the TARGET-OS dataset identified eight
prognostic genes associated with patient survival. (C) Multivariate Cox
regression analysis and Kaplan-Meier survival analysis identified SLC7A7 and
CYP27A1 retained to construct the risk model for TARGET-OS dataset. (D)
Representative immunohistochemistry staining images of high and low SLC7A7
expression in tumor specimens scale bars: 100 µm (left) and 50 µm(right). (E
and F) Analysis of overall survival (E) and lung metastasis-free survival
(F), grouping based on median SLC7A7 expression levels. (G) Univariate
analysis in the TARGET-OS dataset. (H) Representative immunohistochemistry
staining images of high and low CYP27A1 expression in tumor specimens scale
bars: 100 µm (left) and 50 µm(right). (I and J) Analysis of overall survival
(I) and lung metastasis-free survival (J) based on CYP27A1 expression levels.
(K) Multivariable analysis in the TARGET-OS dataset. (L) KM and receiver
operating characteristic analysis for predicting 2-year, 3-year, and 5-year
overall survival based on the stratified risk model in the TARGET-OS dataset.
(M) KM and receiver operating characteristic analysis for predicting 2-year,
3-year, and 5-year overall survival based on the stratified risk model in the
[96]GSE21257 cohort. Data are presented as the mean±SD, by Log-rank test (C,
E, F, I, J, L and M). FC, fold change; FPR, false positive rate; KM,
Kaplan-Meier; SYS cohort, Sun Yat-Sen Cohort; TARGET-OS, Therapeutically
Applicable Research to Generate Effective Treatment-Osteosarcoma; TPR, true
positive rate.

   [97]Figure 2
   [98]Open in a new tab

   We then integrated clinical parameters, including age, gender, tumor
   site, and metastasis status, alongside the expression levels of SLC7A7
   and CYP27A1, into multivariable Cox regression analyses. The analysis
   revealed that the risk score exhibited the highest hazard ratio
   (HR=6.37, p=0.002), demonstrating its prognostic value for patient
   survival outcomes ([99]figure 2K). Using the optimal cut-off value,
   patients were classified into low-risk and high-risk groups. In the
   TARGET-OS dataset, low-risk patients exhibited significantly better
   survival, with area under the curve (AUC) values of 0.68, 0.73, and
   0.76 for 2-year, 3-year, and 5-year survival predictions, respectively
   ([100]figure 2L). Similarly, in the [101]GSE21257 cohort, application
   of the same formula successfully stratified patients into high-risk and
   low-risk groups, with survival analysis indicating significantly better
   survival in the low-risk group ([102]figure 2M). The AUC values in this
   cohort were 0.69, 0.68, and 0.69 for 2-year, 3-year, and 5-year
   survival, respectively, further validating the model’s prognostic
   accuracy ([103]figure 2M). Furthermore, the risk model exhibited
   prognostic power for overall survival across several cancer types among
   the 33 pancancer types, including Lung adenocarcinoma (LUAD, p=0.026;
   [104]Online supplemental figure S2A), Sarcoma (SARC, p=0.096;
   [105]Online supplemental figure S2B), Adrenocortical carcinoma (ACC,
   p=0.022; [106]Online supplemental figure S2C), Mesothelioma (MESO,
   p=0.016; [107]Online supplemental figure S2D), Lymphoid Neoplasm
   Diffuse Large B-cell Lymphoma (DLBC, p=0.012; [108]Online supplemental
   figure S2E), and Skin Cutaneous Melanoma (SKCM, p=0.00025; [109]Online
   supplemental figure S2F). These findings suggest that our LMRG-based
   model has prognostic potential for osteosarcoma and possibly other
   cancer types.

Distinct immune microenvironment features in LMRG-derived osteosarcoma risk
groups

   To dissect the distinct immune microenvironmental traits associated
   with different risk groups in osteosarcoma, we performed gene set
   variation analysis (GSVA) for pathway enrichment. Risk scores were
   negatively correlated with pathways involved in lymphocyte-mediated
   immune regulation, immune response initiation, B cell immune
   modulation, lymphocyte and macrophage chemotaxis, effector CD8+ T cell
   activation, T-cell cytotoxicity regulation, T-cell differentiation,
   macrophage inflammatory cytokine production, and T-cell costimulation
   in the SYS cohort ([110]online supplemental figure S3A). Similarly, in
   the TARGET-OS and [111]GSE21257 cohorts, these immune-related pathways
   showed inverse correlation with risk scores and were more enriched in
   the low-risk group than in the high-risk group ([112]online
   supplemental figure S3B and C).

   In addition, ESTIMATE scores revealed that the high-risk group had
   significantly lower estimates for immune, stromal, and overall ESTIMATE
   scores (p<0.001) while the tumor purity was significantly higher
   (p<0.001), in contrast to the low-risk group ([113]figure 3A). Similar
   results were observed in the [114]GSE21257 cohort ([115]online
   supplemental figure S4A). Using the ssGSEA algorithm, we scored 29
   immune-related gene sets across samples, which reflects the diverse
   immune cell populations and their functional activities. The analysis
   showed that as risk scores decreased, the abundance of various immune
   cells—including plasmacytoid dendritic cells (DCs), immature DCs, B
   cells, and CD8^+ T cells—increased significantly, along with enhanced
   immune response activity ([116]figure 3B). This pattern was
   consistently observed in both the SYS and TARGET-OS datasets
   ([117]figure 3C; [118]Online supplemental figure S4B). In addition,
   xCell deconvolution analysis revealed a notable enrichment of
   macrophages and monocytes, which were the most differentially abundant
   immune cells in the low-risk group ([119]figure 3D). Macrophage
   infiltration was higher in the low-risk group within the [120]GSE21257
   cohort ([121]online supplemental figure S4C). KM survival analysis
   demonstrated that the higher macrophage infiltration in osteosarcoma
   tissues was associated with improved survival outcomes in both the
   TARGET-OS and [122]GSE21257 cohorts ([123]figure 3E and F).

Figure 3. Characterization of the immune microenvironment disparities
associated with the LMRG-based risk model. (A) Comparison of ESTIMATE scores
between high-risk and low-risk groups. (B and C) ssGSEA analysis of immune
cell infiltration and immune function activity in the SYS and TARGET-OS
datasets. (D) Comparison of the immune cell relative abundance between
high-risk and low-risk groups in the TARGET-OS dataset using xCell
algorithms. (E and F) Macrophage infiltration on patient survival in the
TARGET-OS dataset, grouping based on median macrophage infiltration score.
(G) Representative immunohistochemistry staining images of high and low CD68+
macrophage infiltration in tumor specimens, 100 µm (left) and 50 µm(right).
(H and I) Analysis of overall survival (H) and lung metastasis-free survival
(I) based on CD68+ macrophage infiltration levels, grouping based on median
IHC score. (J) Comparison of proinflammatory TAM and anti-inflammatory TAM
macrophage phenotypes between high-risk and low-risk groups in the TARGET-OS
dataset. (K and L) Correlation analysis of risk scores with total macrophage
abundance and proinflammatory TAM macrophage abundance in the SYS cohort and
the TARGET-OS dataset. (M and N) Correlation of risk scores with expression
of macrophage function, T-cell function, and immune-related genes in the SYS
and TARGET-OS datasets. Data are presented as the mean±SD, by Student’s
t-test (A and J), by Log-rank test (E and F). DCs, dendritic cells; aDCs,
activated dendritic cells; APC, Antigen-presenting cell; CCR, C-C motif
chemokine receptor; ESTIMATE, Estimation of STromal and Immune cells in
MAlignant Tumor tissues using Expression data; FC, fold change; IFN,
interferon; IHC, immunohistochemistry; LMRG, lactate metabolism-related gene;
MHC, major histocompatibility complex; MSC, mesenchymal stem cell; NK,
natural killer; pDCs, plasmacytoid dendritic cells; ssGSEA, single-sample
gene set enrichment analysis; SYS cohort, Sun Yat-Sen Cohort; TAMs,
tumor-associated macrophages; TARGET-OS, Therapeutically Applicable Research
to Generate Effective Treatment-Osteosarcoma; Th1, T helper cell 1; Th2, T
helper cell 2; TIL, tumor-infiltrating lymphocyte; Treg, regulatory T cell;
Tcm, central emory memory T cell; Tem, effective memory T Cell.

   [124]Figure 3
   [125]Open in a new tab

   To further examine the effect of macrophage infiltration on prognosis,
   we performed IHC on 62 osteosarcoma specimens from our institution
   ([126]figure 3G). These findings confirmed that patients with high
   levels of macrophage infiltration had significantly better 5-year
   overall survival and lung metastasis-free survival (p<0.001)
   ([127]figure 3H and I). Macrophage polarization analysis revealed a
   predominance of the anti-inflammatory TAMs phenotype in the high-risk
   group, whereas the proinflammatory TAMs phenotype dominated in the
   low-risk group[128]^29 30 ([129]figure 3J). Spearman correlation
   analysis in the SYS cohort demonstrated an inverse correlation between
   risk scores and macrophage infiltration, particularly with the
   proinflammatory TAMs phenotype (R=−0.68 for total macrophages, p<0.001;
   R=−0.47 for proinflammatory TAMs macrophages, p=0.0011) ([130]figure
   3K), a trend that was also observed in the TARGET-OS and [131]GSE21257
   datasets ([132]figure 3L; [133]Online supplemental figure S4D). The
   analysis further indicated that as risk scores decreased, the
   expression of genes associated with macrophage activation and antigen
   presentation increased, enhancing the capacity of macrophages to
   stimulate cytotoxic T lymphocytes and thereby strengthen antitumor
   immunity ([134]figure 3M and N, [135]online supplemental figure S4E).
   This was supported by the increased expression of T-cell activation and
   trafficking genes in the low-risk group, consistently observed across
   both cohorts ([136]figure 3M and N, [137]online supplemental figure
   S4E). Our findings revealed distinct macrophage profiles between risk
   groups. Specifically, the high-risk group exhibited reduced macrophage
   infiltration with predominant anti-inflammatory TAM polarization, while
   the low-risk group showed increased infiltration of proinflammatory
   TAMs.

Insights from single-cell transcriptomics on immune microenvironment
diversity between LMRG-based risk groups

   To further investigate the immune microenvironment differences between
   risk groups, we performed a single-cell transcriptomic analysis on data
   from six chemotherapy-naïve primary patients with osteosarcoma from the
   [138]GSE162454 dataset. After QC, a total of 25,640 cells were analyzed
   using Harmony for batch correction, CopyKAT for malignant cell
   prediction and UMAP for dimensionality reduction and visualization
   ([139]figures5A[140]D). We identified 10 distinct cell subtypes based
   on their signature gene expression: macrophages (C1QC, APOE, APOC1,
   C1QA, and C1QB), monocytes (FCN1, S100A9, S100A8, and CXCL8), DCs
   (CST3, HLA-DRA, and HLA-DQA2), natural killer (NK)/T cells (CD2, CD3D,
   CD3E, CD3G, and NKG7), osteoblastic cells (ALPL, RUNX2, and IBSP),
   osteoclasts (ACP5, CTSK, and MMP9), cancer-associated fibroblasts
   (COL1A1, ACTA2, and DCN), plasmocytes (IGHG1 and MZB1), endothelial
   cells (eg,FL7 and PLVAP), and B cells (MS4A1 and CD79A) ([141]figure
   4A).

Figure 4. Single-cell transcriptomic analysis of immune microenvironment
disparities between LMRG-based risk score groups. (A) UMAP plot illustrating
the identified 10 cell types in the [142]GSE162454 dataset. (B) Columns and
pie charts show the distribution of cell types in high-risk and low-risk
groups related to (A). (C) Gene set variation enrichment analysis pathway of
differential enrichment between macrophages in high-risk and low-risk score
groups. (D–O) Comparison of the functional status score of hypoxia (D),
proliferation (E), metastasis (F), invasion (G), stemness (H), angiogenesis
(I), EMT (J), inflammation (K), apoptosis (L), differentiation (M), DNA
damage (N), DNA repair (O) in tumor cells between high-risk and low-risk
groups based on the CancerSEA database. Data are presented as the mean±SD, by
Student’s t-test (D–O). CAFs, cancer-associated fibroblasts; CancerSEA,
Cancer Single-cell State Atlas; DC, dendritic cell; EMT,
epithelial–mesenchymal transition; GO, Gene Ontology; LMRG, lactate
metabolism-related gene; NK, natural killer; NES, normalized enrichment
score; UMAP, uniform manifold approximation and projection.

   [143]Figure 4
   [144]Open in a new tab

Figure 5. Effects of osteosarcoma cells on macrophage phenotype polarization
and migration capacity. (A and B) Interaction strength among various cell
types within the osteosarcoma microenvironment. (C) Classification of
osteosarcoma cell lines into high-risk and low-risk cells using the NTP
method. (D) Schematic of a co-culture system between osteosarcoma cells and
macrophages. (E) Phalloidin-FITC staining of macrophage morphology after
coculture with 143B or U2OS cells. +143B: macrophages cocultured with 143B
cells; +U2OS: macrophages cocultured with U2OS cells. Scale bar: 50 µm. (F)
RT-qPCR analysis of mRNA levels for iNOS, CD86, TNFa, ARG1, IL10, and CD163
in macrophages (n=3). (G) Immunofluorescence detection of ARG-1 and iNOS
protein expression in macrophages, scale bar: 100 µm. (H) Relative
fluorescence quantification of ARG-1 protein expression (n=3). (I) Relative
fluorescence quantification of iNOS protein expression. (J) Migration assay
comparing the chemotactic effect of osteosarcoma cells on macrophages, scale
bar: 200 µm. (K) Relative quantification of migrated macrophages. Control:
medium alone in the lower chamber; +143B: macrophages cocultured with 143B
cells; +U2OS: macrophages cocultured with U2OS cells (n=3). Data are
presented as the mean±SD. *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001, by
one-way ANOVA with Tukey’s multiple comparisons test (F, H, I and K). ANOVA,
analysis of variance; CAFs, cancer-associated fibroblasts; DAPI,
4′,6-diamidino-2-phenylindole; IL, interleukin; DC, dendritic cell; mRNA,
messenger RNA; NK, natural killer; ns, not significant; NPT, nearest template
prediction; RT-qPCR, reverse-transcription quantitative PCR; TAMs,
tumor-associated macrophages.

   [145]Figure 5
   [146]Open in a new tab

   Patients were then categorized into high-risk score (n=3) and low-risk
   score (n=3) groups based on the risk scores. The high-risk score group
   exhibited a higher proportion of tumor cells and tumor-associated
   fibroblasts, whereas the low-risk score group had higher proportions of
   macrophages (predominantly), monocytes, B cells, DCs, and NK/T cells
   ([147]figure 4B). GSVA revealed that macrophages in the low-risk score
   group were enriched in pathways related to antigen processing, T-cell
   activation, and lymphocyte proliferation, suggesting a proinflammatory
   TAM phenotype. In contrast, macrophages in the high-risk group were
   enriched in pathways associated with an anti-inflammatory TAM
   phenotype, including leukocyte-mediated immunity suppression and
   vasculature development ([148]figure 4C). These findings were
   consistent with bulk transcriptomic analyses, which also revealed the
   elevated infiltration of immune cells, particularly proinflammatory
   TAMs, in the low-risk group. Conversely, the high-risk score group
   exhibited decreased immune cell infiltration, with a predominance of
   anti-inflammatory TAMs. Further analysis using the CancerSEA database
   revealed that tumor cells in the high-risk group exhibited greater
   activity in pathways related to hypoxia, invasion, stemness,
   differentiation, proliferation, metastasis, angiogenesis, apoptosis,
   DNA damage response, quiescence, and inflammation, compared with those
   in the low-risk group ([149]figure 4D–O). Both bulk and single-cell
   sequencing analyses consistently demonstrated that differences in
   LMRG-based risk scores are primarily reflected in macrophage
   infiltration and polarization, which may underlie the observed
   disparities in immune microenvironment and tumor aggressiveness between
   risk groups.

Heterogeneity of osteosarcoma cell–macrophage crosstalk among LMRG-based risk
groups

   To elucidate the interaction between different cell types, we analyzed
   the communication between macrophages and other cell types using the
   CellChat algorithm. The results revealed significant differences in
   signaling interactions between tumor cells and macrophages ([150]figure
   5A and B and [151]online supplemental figure S5E). We next investigated
   whether tumor cells influence macrophage infiltration and polarization.
   Using the NTP method, we categorized osteosarcoma cell lines into
   groups resembling high-risk and low-risk score patients based on
   transcriptomic similarities.[152]^25 Cell lines such as HUO9, NOS1,
   143B, G292, PSS008, and HSOS1 exhibited gene expression patterns
   similar to the high-risk group. While, cell lines including C396,
   HS860T, PSS131R, MG63, OS252, HOS, CAL72, SAOS2, and U2OS exhibited
   expression patterns typical of the low-risk group ([153]figure 5C).

   After 48 hours of coculture, macrophages exhibited distinct
   morphologies in response to the osteosarcoma cell lines they were
   exposed to. Macrophages cocultured with 143B cells developed elongated
   projections, a morphology associated with anti-inflammatory TAMs.
   Conversely, macrophages cocultured with U2OS cells showed a round
   morphology, characteristic of proinflammatory TAMs[154]^29 30
   ([155]figure 5D and E). RT-qPCR analysis confirmed a gene expression
   profile consistent with anti-inflammatory TAM polarization in
   macrophages cocultured with 143B cells, characterized by increased
   levels of ARG1, IL10, and CD163 mRNA ([156]figure 5F). Conversely,
   macrophages cocultured with U2OS cells showed a proinflammatory TAM
   phenotype, evidenced by elevated expression of iNOS, CD86, and TNFa
   ([157]figure 5F). Protein expression supported these findings, with
   elevated expression of ARG-1 in the 143B coculture system and iNOS in
   the U2OS coculture system ([158]figure 5G–I). Notably, neither 143B nor
   U2OS cell lines expressed ARG-1 or iNOS ([159]online supplemental
   figure 6A). Transwell assays further revealed the contrasting
   chemotactic effects of osteosarcoma cells on macrophages, as 143B cells
   exerted a stronger inhibitory effect on macrophage migration compared
   with that of U2OS cells ([160]figure 5J and K). These cell
   communication analysis and coculture experiments consistently
   demonstrated the significant effect of diverse osteosarcoma cells on
   macrophage polarization and chemotaxis.

Osteosarcoma-derived MIF is a key differential mediator in tumor–macrophage
crosstalk between LMRG-based risk groups

   In order to identify the essential mediators of tumor–macrophage
   crosstalk. We analyzed the cytokine-mediated ligand–receptor
   interactions between tumor cells and macrophages in high-risk and
   low-risk groups. This study revealed significant disparities in MIF
   signaling, particularly in the MIF-(CD74+CXCR4) and MIF-(CD74+CD44)
   axes ([161]figure 6A). KM analysis showed that higher MIF expression
   was associated with poorer overall survival in patients with
   osteosarcoma ([162]figure 6B). Immunoblotting and ELISA assays further
   confirmed that high-risk osteosarcoma cell lines exhibited higher
   expression and secretion of MIF than those in low-risk cell lines
   ([163]figure 6C and D). To investigate the clinical relevance of MIF
   expression in osteosarcoma, we examined the expression of MIF in
   osteosarcoma tissues via IHC staining of paraffin sections. Patients
   with high MIF expression had significantly lower 5-year overall
   survival and lung metastasis-free survival (p<0.001) ([164]figure 6E).
   Spearman’s correlation analysis revealed a significant positive
   correlation between MIF levels and anti-inflammatory TAMs infiltration
   (R=0.31, p=0.0037), with no correlation observed for pro-inflammatory
   TAMs (R=0.16, p=0.14) ([165]figure 6F). We then knocked out MIF (SgMIF)
   in 143B cells and co-cultured them with macrophages. Macrophages showed
   distinct morphological changes depending on MIF expression ([166]figure
   6G–I). The analysis of gene expression indicated increased levels of
   proinflammatory markers (iNOS, CD86, and TNFa) in macrophages
   co-cultured with MIF-knockout cells, whereas those treated with rhMIF
   expressed higher levels of anti-inflammatory markers (ARG1, CD163, and
   IL10) ([167]figure 6J). These results were further supported at the
   protein level, showing increased iNOS expression in MIF-knockout
   co-cultures and elevated ARG-1 levels following rhMIF treatment
   ([168]figure 6K–M). MIF is a multifunctional cytokine that regulates
   macrophage migration, participating in multiple immune responses
   through ligand–receptor interactions.[169]^31 We performed a migration
   assay using a co-culture model. Results demonstrated that MIF knockout
   alleviated tumor cell-mediated suppression of macrophage chemotaxis,
   while rhMIF treatment restored the inhibitory effect ([170]figure 6N
   and O). Overall, these results suggested that osteosarcoma cells
   promote anti-inflammatory macrophage polarization and inhibit
   chemotaxis through MIF secretion.

Figure 6. Osteosarcoma-derived MIF contributes to anti-inflammatory TAM
macrophage polarization and chemotaxis. (A) Analysis of ligand–receptor
interactions between tumor cells and macrophages in high-risk and low-risk
score groups. (B) Survival analysis of MIF expression in the TARGET-OS
dataset, grouping based on median MIF expression. (C) MIF protein levels in
osteosarcoma cell lines. (D) Quantification of MIF in osteosarcoma cell
culture supernatants (n=3). (E) Representative immunohistochemistry staining
images of high and low MIF expression in tumor specimens and survival
analysis 100 µm (left) and 50 µm(right), grouping based on median IHC score.
(F) Correlation analysis between MIF expression and macrophage phenotype in
TARGET-OS patients. (G) Knockout of MIF in 143B cell lines. (H) MIF levels in
143B cells measured via ELISA (n=3). (I) Phalloidin-FITC staining of
macrophage morphology post coculture with MIF-knockout 143B cells, with rhMIF
supplementation; scale bar: 20 µm. (J) RT-qPCR analysis of iNOS, CD86, TNFa,
ARG1, IL10, and CD163 in macrophages after coculture (n=3). (K)
Immunofluorescence detection of ARG-1 and iNOS protein expression in
macrophages, scale bar: 100 µm. (L) Relative fluorescence quantification of
ARG-1 protein expression (n=3). (M) Relative fluorescence quantification of
iNOS protein expression (n=3). (N and O) Migration assay assessing chemotaxis
of macrophages (N) and quantification (O) (n=3), scale bar: 200 µm. Data are
presented as the mean±SD. *p<0.05, **p<0.01, ***p<0.001 and ****p<0.0001, by
Student’s t-test (H), by one-way ANOVA with Tukey’s multiple comparisons test
(D, J, L, M and O), by Spearman correlation analysis (F), by Log-rank test (B
and E). ANOVA, analysis of variance; DAPI, 4′,6-diamidino-2-phenylindole;
IHC, immunohistochemistry; IL, interleukin; MIF, macrophage migration
inhibitory factor; mRNA, messenger RNA; RT-qPCR, reverse-transcription
quantitative PCR; TAM, tumor-associated macrophage; TARGET-OS,
Therapeutically Applicable Research to Generate Effective
Treatment-Osteosarcoma.

   [171]Figure 6
   [172]Open in a new tab

Lactate regulates MIF expression through histone lactylation

   To further investigate the role of lactate in the immune
   microenvironment, we first explored the effect of lactate on MIF
   expression in osteosarcoma cells with 2-Deoxy-D-glucose (2-DG,
   glycolytic inhibitor) treatment. Treatment with 2-DG significantly
   suppressed MIF expression in osteosarcoma cells and reduced the level
   of protein lactylation ([173]figure 7A and B). Histone lactylation is a
   vital mechanism by which lactate regulates gene expression.[174]^32 We
   further investigated whether lactate modulates MIF expression in
   osteosarcoma through histone lactylation. After exploring the ChIP and
   CUT&Tag data related to histone lactylation from the GEO database, we
   found that lactylation at histone H3K9 (H3K9la) was enriched at the
   region of MIF ([175]online supplemental figure S6B), indicating that
   histone lactylation could regulate MIF expression. Then we assess the
   level of H3K9la in osteosarcoma cells and the results showed 2-DG
   treatment significantly suppressed the H3K9la level ([176]figure 7C).
   ChIP-qPCR using anti-H3K9la antibody was used to explore the role of
   this histone-lactylation site in MIF expression and results exhibited
   H3K9la regulated MIF expression in 143B and G292 cell lines
   ([177]figure 7D). Moreover, we found the lactic acid level was
   positively correlated with MIF expression in patients with osteosarcoma
   ([178]figure 7E). Additionally, 2-DG treatment reduced the secretion of
   MIF by tumor cells ([179]figure 7F). In the tumor immune
   microenvironment, macrophage-derived CXCL9 and CXCL10 play a crucial
   role in recruiting CD8^+ T cells, which are essential for antitumor
   immune responses and the efficacy of anti-PD-1 therapy.[180]33,[181]37
   Then, using the co-culture model, we further examined the effect of
   osteosarcoma-derived MIF on the expression of CXCL9 and CXCL10 in
   macrophages. The results showed that MIF knockdown significantly
   upregulated the expression of CXCL9 and CXCL10, while supplementation
   with rhMIF suppressed their expression ([182]online supplemental figure
   S6C). We then used the co-culture model to investigate whether
   lactate-regulated osteosarcoma cells influence CD8^+ T cell function
   via macrophages. The results showed that osteosarcoma cells treated
   with 2-DG promoted CXCL9 and CXCL10 expression in macrophages
   ([183]figure 7G) and enhanced the expression of granzyme B and IFN-γ in
   CD8^+ T cells ([184]figure 7H). These results demonstrated that lactate
   promotes the MIF expression in osteosarcoma cells through histone
   lactylation modification, which suppresses the secretion of CXCL9 and
   CXCL10 in macrophages, resulting in impairment of CD8^+ T cell
   function.

Figure 7. Lactate regulated MIF expression through histone H3K9 lactylation
(A) RT-qPCR analysis of MIF mRNA expression levels in 143B and G292 cells
treated with 2-DG (n=3). (B) MIF and protein lactylation levels in 143B and
G292 cells treated with 2-DG measured by immunoblotting. (C) H3K9 lactylation
levels in 143B and G292 cells treated with 2-DG measured by immunoblotting.
(D) RT-qPCR analysis of MIF signal from ChIP (using anti-H3K9la antibody) in
143B cell (n=3). (E) Correlation of lactic acid level with MIF expression in
tumor samples from SYS cohort (n=30). (F) MIF secretion in 143B and G292
cells treated with 2-DG (n=3). (G) CXCL9 (left) and CXCL10 (right) level in
macrophage cocultured with 2-DG treated 143B cells measured by ELISA (n=3).
(H) Flow cytometry analysis of the effect of 143B-co-cultured macrophages on
T-cell granzyme B and interferon-γ expression levels (n=3). Data are
presented as the mean±SD, *p<0.01, ***p<0.001 and ****p<0.0001, by Student’s
t-test (A, F and G), by one-way ANOVA with Tukey’s multiple comparisons test
(D and H), by Pearson correlation (E). ANOVA, analysis of variance; CHIP,
chromatin immunoprecipitation; 2-DG, 2-Deoxy-D-glucose; IFN-γ,
interferon-gamma; MIF, macrophage migration inhibitory factor; mRNA,
messenger RNA; RT-qPCR, reverse-transcription quantitative PCR; SYS cohort,
Sun Yat-Sen Cohort; SSC, side scatter.

   [185]Figure 7
   [186]Open in a new tab

MIF-mediated macrophage polarization regulates T-cell activation and
infiltration in osteosarcoma

   Previous studies showed that macrophage polarization affects T-cell
   proliferation and activation. To further explore the role of
   MIF-mediated macrophage polarization in regulating T cell responses, we
   used a co-culture model. Macrophages were first co-cultured for 48
   hours with SgNC 143B, SgMIF 143B, or rhMIF-supplemented SgMIF 143B
   osteosarcoma cells. Tumor cells were then removed, and PBMC-derived T
   cells were added for subsequent co-culture ([187]online supplemental
   figure S7A). The CellTrace assay demonstrated that T cell proliferation
   was partially restored when cocultured with macrophages previously
   exposed to SgMIF 143B cells, compared with those exposed to SgNC 143B
   cells. However, the addition of rhMIF significantly inhibited this
   proliferation ([188]online supplemental figure S7B). Flow cytometry
   further revealed increased expression of granzyme B and IFN-γ in T
   cells cocultured with SgMIF 143B–induced macrophages, which was
   significantly reduced on rhMIF treatment ([189]online supplemental
   figure S7C and D). In addition, immunostaining of osteosarcoma tissues
   revealed that patients with high MIF expression had significantly fewer
   tumor-infiltrating CD68^+ macrophages and CD8a^+ T cells, while markers
   associated with immunosuppressive macrophages, including CD206 and
   CD163, were notably increased ([190]online supplemental figure S7E-I).
   These results showed MIF-mediated macrophage polarization suppressed
   the function and proliferation of T cells.

Antitumor effects of MIF inhibition combined with anti-PD-1 therapy in
osteosarcoma

   To further explore the therapeutic potential of MIF in osteosarcoma
   treatment, we then used K7M2 syngeneic mice to investigate the role of
   MIF in the osteosarcoma immune microenvironment ([191]figure 8A). Mif
   knockout significantly reduced tumor weight and volume ([192]figure
   8A–C). Immunofluorescence staining revealed the increased infiltration
   of CD8a^+ cells ([193]figure 8D) and F4/80^+ cells ([194]figure 8E),
   along with a decrease in CD206^+ cells ([195]figure 8F) in the TME
   following Mif deletion. Moreover, the expression of CXCL9 and CXCL10
   was also increased in the Mif-knockout group ([196]figure 8G and H). We
   then administered the MIF inhibitor 4-iodo-6-phenylpyrimidine (4-IPP)
   (10 mg/kg) intraperitoneally to syngeneic mice every 3 days,[197]^38
   which also led to a significant reduction in tumor weight and volume
   ([198]figure 8I–K). Flow cytometry confirmed the increased macrophage
   infiltration following MIF inhibition ([199]figure 8L and [200]online
   supplemental figure S7J and K). In addition, depleting macrophages with
   clodronate liposomes alongside 4-IPP treatment partially attenuated the
   antitumor effects of 4-IPP, highlighting the critical role of
   macrophages in mediating the efficacy of 4-IPP ([201]figure 8L). Given
   the established association between F4/80^+ and CD8^+ cell infiltration
   and improved outcomes in immunotherapy, we further examined the synergy
   between 4-IPP and anti-PD-1 therapy ([202]figure 8M–P). The combination
   treatment significantly enhanced therapeutic efficacy, demonstrated by
   the increased infiltration of CD8^+ and F4/80^+ cells ([203]figure 8Q
   and R), along with a decrease in that of CD206^+ cells ([204]online
   supplemental figure S8)([205]figure 8S). The expression of CXCL9 and
   CXCL10 was also increased ([206]figure 8T and U) in the TME. Overall,
   these results indicated that the combination treatment of 4-IPP and
   anti-PD-1 therapy significantly boosted the immune response in
   osteosarcoma.

Figure 8. Antitumor effects of MIF inhibition and anti-PD-1 therapy in
osteosarcoma. (A–C) General view, tumor volume, and tumor weight in K7M2
syngeneic tumors with Mif knockout (n=5). (D–H) IF staining of CD8a+, F4/80+,
CD206+, CXCL9 and CXCL10 following Mif knockout (n=5). (I–K) General view,
tumor volume, and tumor weight in K7M2 syngeneic tumors with 4-IPP or
clodronate liposomes treatment (n=5). (L) Flow cytometry analysis of the
macrophage infiltration ratio (n=5). (M) Schematic showing steps for
anti-PD-1 treatment combining 4-IPP treatment. (N–P) General view, tumor
volume, and tumor weight in K7M2 syngeneic tumors with 4-IPP and anti-PD-1
treatment (n=5). (Q–U) IF staining of CD8a+, CD68+, CD206+, and CXCL9 and
CXCL10 cells following 4-IPP and anti-PD-1 treatment. Data are presented as
the mean±SD, ns means no significant, **p<0.01, ***p<0.001 and ****p<0.0001,
by one-way ANOVA with Tukey’s multiple comparisons test (C, D, E, F, G, H, K,
L, P, Q, R, S, T and U), by Pearson correlation (E), two-way ANOVA with
Tukey’s multiple comparisons test (B, J and O). ANOVA, analysis of variance;
DAPI, 4′,6-diamidino-2-phenylindole; FOV, field of view; IF,
immunofluorescence; 4-IPP, 4-iodo-6-phenylpyrimidine; MIF, macrophage
migration inhibitory factor; PD-1, programmed cell death 1.

   [207]Figure 8
   [208]Open in a new tab

Discussion

   Metabolic reprogramming has emerged as a hallmark of cancer and is
   crucial for the survival and functional execution of both cancer and
   immune cells.[209]^9 39 40 Accumulating evidence highlights the role of
   metabolic reprogramming as a key determinant of cancer development and
   antitumor immune response.[210]^41 42 Lactate, a glycolysis byproduct,
   functions as a crucial carbon source and signaling molecule. Research
   has indicated a significant correlation between elevated lactate levels
   and increased angiogenesis, tumor growth, nodal and distant metastases,
   and poorer survival across various cancer types.[211]43,[212]45
   Consistent with these findings, our study revealed that high lactate
   levels in patients with osteosarcoma were associated with larger tumors
   and poorer survival, emphasizing the role of lactate metabolism in
   osteosarcoma progression. LMRGs are essential for maintaining lactate
   homeostasis in the TME by regulating synthesis, catabolism, and
   transport within this.[213]^46 Here, we developed a risk model based on
   the prognostic LMRGs SLC7A7 and CYP27A1, serving as an independent
   survival predictor for patients with osteosarcoma. These findings
   suggested that exploring disparities based on the risk model may reveal
   new insights into the lactate metabolism network and highlight
   previously overlooked aspects of osteosarcoma progression.

   In the TME, cancer cells, T cells, NK cells, DCs, and macrophages all
   respond to extracellular lactate levels.[214]47,[215]50 This triggers
   intracellular signals that regulate cellular activities and
   significantly affect their infiltration. Lactate has been shown to
   induce ARG-1 expression in macrophages, promoting an anti-inflammatory
   TAM phenotype and suppressing NK cell function.[216]^48 Moreover,
   regulatory T cells rely on lactate uptake to maintain their
   proliferation and immunosuppressive functions. In addition, lactate
   inhibits T-cell receptor-triggered production of IFN-γ, TNF-α, and
   IL-2, thereby impairing the function of cytotoxic T lymphocytes. These
   findings suggest that the disparities associated with the LMRG-based
   risk score may reflect differences in immune cell infiltration and
   activity. In this study, we applied the risk model to differentiate
   patients with osteosarcoma into high-risk and low-risk groups with
   distinct clinical outcomes. The analysis of independent transcriptomic
   datasets revealed that the high-risk group had lower immune cell
   infiltration and immune activation than the low-risk group. The
   combination of bulk and scRNA transcriptomic analyses confirmed notable
   differences in macrophage abundance and polarization: the low-risk
   group had more macrophages with a proinflammatory TAM phenotype,
   whereas the high-risk group had fewer macrophages with an
   anti-inflammatory phenotype. These findings implied that disparities in
   macrophage infiltration and polarization probably contribute to the
   overall survival differences between the high-risk and low-risk groups.

   TAMs are crucial components in the TME.[217]^51 Reprogramming TAMs from
   the immunosuppressive anti-inflammatory TAM phenotype to the
   proinflammatory TAM phenotype is key to improving antitumor
   immunity.[218]^52 53 Macrophages modify their activation in response to
   microenvironmental signals such as cytokines and pathogen-associated
   molecular patterns. Tumor-secreted cytokines such as IL-4, IL-10,
   IL-13, and M-CSF activate transcription factors including STAT3/6,
   KLF2/4, and IRF3/5, which drive arginine-dependent metabolic pathways
   and macrophage polarization.[219]^54 It is worth noting that while our
   study highlights the immunosuppressive role of MIF via modulation of
   macrophage polarization, TAM in the TME exhibit substantial phenotypic
   and functional heterogeneity beyond the classical anti/proinflammatory
   classification.[220]^29 30 Future studies using single-cell
   transcriptomics or spatial profiling may help further resolve the
   complexity of TAM subsets and their dynamic roles in osteosarcoma
   progression and response to immunotherapy. Our study found notable
   differences in macrophage abundance and phenotype between the high-risk
   and low-risk groups, mainly because of the secretion of MIF by tumor
   cells. Osteosarcoma cells with high MIF levels promoted
   anti-inflammatory TAM polarization in macrophages and reduced their
   chemotactic activity, highlighting MIF as a key mediator in the
   immunomodulatory environment of osteosarcoma.

   MIF is a multifunctional cytokine that regulates cell migration,
   activation, and differentiation in both immune and non-immune
   cells.[221]^31 Elevated MIF levels are commonly found in various
   cancers, including osteosarcoma, and are associated with increased
   tumor burden, advanced grades, and poor prognosis.[222]^55 56 In brain
   tumors, MIF engages CD74 receptors on microglial cells, reducing
   interferon-γ production and shifting them from an anti-inflammatory TAM
   to an anti-inflammatory TAM phenotype. In addition, MIF supports
   macrophage survival and maintains homeostasis by upregulating the
   expression of inflammation resolution genes.[223]^57 Furthermore,
   monocyte-derived MIF drives alternative activation in mouse melanoma
   TAMs, and its inhibition reduces TAM-mediated and myeloid-derived
   suppressor cell-mediated immune suppression.[224]^58 Similarly, in the
   4T1 murine model of breast cancer, tumor-derived MIF promotes
   myeloid-derived suppressor cell accumulation and immunosuppressive
   activity. In osteosarcoma, MIF promoted tumor cell proliferation and
   metastasis via the RAS/MAPK pathway.[225]^59 Our study extended this
   knowledge, demonstrating lactate promoted MIF expression through
   histone H3K9 lactylation. We also found that MIF expression not only
   induced anti-inflammatory TAM polarization in macrophages but also
   reduced their chemotactic activity in vitro and in vivo. Moreover,
   MIF-regulated TAM polarization influenced the infiltration of CD8^+ T
   cells through CXCL9 and CXCL10. These findings highlighted MIF as a
   critical factor in osteosarcoma, suggesting that targeting MIF could
   inhibit tumor growth and metastasis while fostering a supportive
   antitumor immune environment, representing a promising therapeutic
   strategy.

   ICIs have transformed the treatment landscape for chemoresistant
   malignancies, such as melanoma.[226]^60 A notable success of ICIs is
   achieving long-term remission even after treatment discontinuation,
   raising significant hope for potential cures in some patients. Given
   this, the use of immunotherapy to improve survival outcomes in
   osteosarcoma has become an attractive strategy. However, clinical
   trials of ICIs targeting the PD-1 pathway in advanced osteosarcoma have
   yet to yield the expected results. Our study supported these findings,
   showing that monotherapy with a PD-1 monoclonal antibody did not
   significantly reduce tumor growth. Previous studies have shown that the
   clinical benefits and prognostic outcomes of immunotherapy are largely
   influenced by the TME status of the patient.[227]^61 Tumors with low
   levels of tumor-infiltrating lymphocytes are referred to as “cold
   tumors” and respond poorly to ICIs.[228]^62 Our findings highlighted
   that knocking out Mif from murine tumor cells increased effector CD8^+
   T-cell infiltration in the TME. This suggested that targeting MIF in
   osteosarcoma could enhance the effectiveness of immune checkpoint
   blockade. Supporting this, our in vivo studies combining 4-IPP, a MIF
   inhibitor, with a PD-1 monoclonal antibody produced a significant
   reduction in tumor growth, surpassing the effects of either agent
   alone. In addition, this combination approach increased effector CD8^+
   T-cell infiltration in the TME. These results indicated that 4-IPP
   could enhance the antitumor efficacy of PD-1 blockade by improving
   immune cell penetration into the tumor, potentially offering a
   significant therapeutic advantage for patients with osteosarcoma with
   elevated MIF expression.

Conclusions

   Overall, we developed an LMRG-based risk score that distinguishes
   patients with osteosarcoma into two groups with notable differences in
   survival outcomes. Our findings illuminated the disparities in TME
   within the groups, with a focus on immune cell infiltration and
   functional activity. We identified MIF-induced macrophage polarization
   and chemotaxis as crucial factors contributing to the differences
   between the groups and confirmed MIF as a potential target for
   enhancing ICI treatment.

Supplementary material

   online supplemental file 1
   [229]jitc-13-8-s001.jpg^ (1.9MB, jpg)
   DOI: 10.1136/jitc-2024-011091
   online supplemental file 2
   [230]jitc-13-8-s002.docx^ (15.6MB, docx)
   DOI: 10.1136/jitc-2024-011091
   online supplemental file 3
   [231]jitc-13-8-s003.docx^ (35KB, docx)
   DOI: 10.1136/jitc-2024-011091

Acknowledgements