Wu et al.

Journal of Biomedical Science (2024) 31:4

https://doi.org/10.1186/s12929-024-00999-7

RESEARCH Open Access

Exploring the relationship

between metabolism and immune
microenvironment in osteosarcoma based
on metabolic pathways
Changwu Wu1,2, Jun Tan1,2*, Hong Shen2,3,4* , Chao Deng5, Christian Kleber6, Georg Osterhoff6 and
Nikolas Schopow6

Abstract
Background Metabolic remodeling and changes in tumor immune microenvironment (TIME) in osteosarcoma are
important factors affecting prognosis and treatment. However, the relationship between metabolism and TIME needs
to be further explored.
Methods RNA-Seq data and clinical information of 84 patients with osteosarcoma from the TARGET database
and an independent cohort from the GEO database were included in this study. The activity of seven metabolic super-
pathways and immune infiltration levels were inferred in osteosarcoma patients. Metabolism-related genes (MRGs)
were identified and different metabolic clusters and MRG-related gene clusters were identified using unsupervised
clustering. Then the TIME differences between the different clusters were compared. In addition, an MRGs-based risk
model was constructed and the role of a key risk gene, ST3GAL4, in osteosarcoma cells was explored using molecular
biological experiments.
Results This study revealed four key metabolic pathways in osteosarcoma, with vitamin and cofactor metabolism
being the most relevant to prognosis and to TIME. Two metabolic pathway-related clusters (C1 and C2) were identi-
fied, with some differences in immune activating cell infiltration between the two clusters, and C2 was more likely
to respond to two chemotherapeutic agents than C1. Three MRG-related gene clusters (GC1-3) were also identified,
with significant differences in prognosis among the three clusters. GC2 and GC3 had higher immune cell infiltra-
tion than GC1. GC3 is most likely to respond to immune checkpoint blockade and to three commonly used clinical
drugs. A metabolism-related risk model was developed and validated. The risk model has strong prognostic predic-
tive power and the low-risk group has a higher level of immune infiltration than the high-risk group. Knockdown
of ST3GAL4 significantly inhibited proliferation, migration, invasion and glycolysis of osteosarcoma cells and inhibited
the M2 polarization of macrophages.
Conclusion The metabolism of vitamins and cofactors is an important prognostic regulator of TIME in osteo-
sarcoma, MRG-related gene clusters can well reflect changes in osteosarcoma TIME and predict chemotherapy

*Correspondence:
Jun Tan
[email protected]
Hong Shen
[email protected]
Full list of author information is available at the end of the article

© The Author(s) 2024. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecom-
mons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 2 of 26

and immunotherapy response. The metabolism-related risk model may serve as a useful prognostic predictor.
ST3GAL4 plays a critical role in the progression, glycolysis, and TIME of osteosarcoma cells.
Keywords Osteosarcoma, Metabolism, Tumor immune microenvironment, Prognosis, Vitamin and cofactor,
Treatment response, ST3GAL4

Introduction as new options with great potential for application [10,

Osteosarcoma is a malignant bone tumor that pre- 11]. Indeed, recent studies have proven that immu-
dominantly affects children and young adults. Despite notherapy has shown advantages over conventional
advances in treatment, the prognosis for patients with interventional strategies in inhibiting osteosarcoma
osteosarcoma remains poor, with a 5-year survival rate metastasis and recurrence, and satisfactory efficacy in
of approximately 60–70% for localized disease and less inhibiting the progression of advanced osteosarcoma
than 30% for metastatic disease [1]. The development and [12–14].
progression of osteosarcoma is a complex process that The interaction between tumor metabolism and immu-
involves multiple molecular and cellular mechanisms. nity has been intensively studied and it is generally rec-
In recent years, there has been increasing interest in the ognized that oncogenic transformation can lead to the
role of metabolic reprogramming and the tumor immune adaptation of a well-characterized metabolic pheno-
microenvironment (TIME) in osteosarcoma, and their type in cancer cells that can profoundly affect the TIME
potential as therapeutic targets. [15]. Specifically, in addition to affecting cancer cells
Tumor cells have to modify their metabolic program directly, metabolic reprogramming of tumors also alters
to support the energy and macronutrient requirements the TIME by affecting the behavior of other cell types,
of rapid proliferation. Metabolic reprogramming is now such as immune cells and stromal cells [16]. For exam-
recognized as a hallmark of cancer and is one of the most ple, the acidic microenvironment created by aerobic gly-
critical biological differences between tumor cells and colysis can suppress the immune system and promote
normally differentiated cells [2]. For example, in many the growth of blood vessels, which can in turn promote
tumor cells, altered carbohydrate metabolism, repre- tumor growth and metastasis [17]. In particular, the lac-
sented by the Warburg effect, provides a proliferative tate produced by aerobic glycolysis can induce the infil-
advantage for tumor cells [3]. Osteosarcoma cells exhibit tration of regulatory T cells and the M2 polarization of
a variety of metabolic alterations, including increased macrophages in tumors, thereby promoting immune
glucose uptake, altered mitochondrial function, and suppression [18]. Tumor metabolic heterogeneity refers
increased reliance on glycolysis for ATP generation [4]. to the significant differences in metabolic characteristics
These metabolic changes are driven by a variety of sign- that exist between different tumors or within the same
aling pathways, including the PI3K/AKT/mTOR path- tumor tissue. It is an important aspect of tumor het-
way and the HIF-1α pathway [5]. They provide potential erogeneity and is mainly driven by different genotypes
therapeutic targets for the treatment of osteosarcoma, or microenvironments [19–21]. The research by Feng
as inhibition of key metabolic pathways could poten- et al. demonstrates that within the same type of tumor,
tially starve cancer cells of the nutrients they need to patients can be divided into subgroups suitable for dif-
proliferate. ferent treatment methods based on different metabolic
TIME plays a critical role in a variety of biologi- characteristics [22]. Therefore, identifying the metabolic
cal processes including proliferation, metastasis, and profile of different osteosarcoma patients will not only
treatment response (including chemotherapy, radia- explore the impact of different metabolic landscapes
tion therapy and immunotherapy) in osteosarcoma on TIME, but also guide treatment decisions. However,
[6–8]. Previous studies have shown that patients with few studies have been conducted to genomically analyze
different TIME status within their osteosarcoma have osteosarcoma from a global perspective of metabolic het-
very different prognoses [9]. Specifically, patients with erogeneity. The few previous studies have been limited
“hot” tumors that have more immune cell infiltration to a specific metabolic pathway [23, 24]. In this study, we
in TIME have a better prognosis, while patients with focused on the seven most prominent metabolic super-
“cold” tumors that have less immune cell infiltration pathways with the aim of comprehensively assessing
have a worse prognosis. Therefore, therapeutic agents metabolic pathways of prognostic importance in osteo-
that modulate TIME, transform “cold” tumors into sarcoma, identifying tumor subtypes with different met-
“hot” tumors, and use existing immunity to destroy abolic profiles and exploring the heterogeneity of TIME
osteosarcoma cells are increasingly being considered profiles and treatment response across tumor subtypes.
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 3 of 26

Methods and materials Pathway enrichment analysis

Data acquisition, clinical samples and cell lines To quantify the activity of the seven metabolic path-
Standardized RNA-Seq data and clinical information ways in a single tumor sample, gene set variation analy-
for 88 independent osteosarcoma samples from the sis (GSVA) was performed using the R package “GSVA”
TARGET database were obtained from Xena Func- to calculate the enrichment score of each pathway in
tional Genomics Explorer (http://​xena.​ucsc.​edu/), of the individual sample. Subsequently, Kaplan- Meier
which 84 samples with complete survival information curve and log-rank test were used to explore the rela-
were included in this study. In addition, standardized tionship between the seven metabolic pathways and
microarray expression data and clinical information for overall survival (OS) of patients with osteosarcoma.
34 osteosarcoma samples from the GSE16091 cohort The “surv_cutpoint” function of the “survminer” R
were obtained from the GEO database (https://​w ww.​ package was used to determine the optimal cut-off
ncbi.​nlm.​nih.​gov/​geo/) [25]. The gene expression data point of each metabolic pathway based on the maxi-
distribution from the TARGET and GEO databases was mally selected log-rank statistics. In addition, a set of
analyzed using the PCA algorithm, and it was found core biological pathway gene sets closely related to
that there were no apparent batch effects in the data tumors was collected from the study of Mariathasan
(Additional file 1: Figure S1). In addition, gene sets for et al. and the activity of each pathway was calculated by
seven metabolic super-pathways annotated according GSVA [27]. This includes three epithelial mesenchymal
to the Reactome database were collected from a previ- transition (EMT) signatures originating from different
ous study (Additional file 2: Table S1) [26]. Together, publications and composed of different genes. The cor-
these gene sets represent the major metabolic pro- relation of key metabolic pathways with core biological
cesses, including amino acid metabolism, carbohydrate pathways was then calculated using Spearman’s corre-
metabolism, integration of energy metabolism, lipid lation analysis. Gene set enrichment analysis (GSEA)
metabolism, nucleotide metabolism, tricarboxylic acid between the two groups of samples was performed in
(TCA) cycle, and vitamin & cofactor metabolism. Sangerbox (http://​vip.​sange​rbox.​com/) using the GSEA
Tumor samples of 14 primary osteosarcoma patients software (Version 4.1.0) based on the HALLMARK and
who underwent surgical resection between 2018 and KEGG gene sets [28].
2019 and 5 normal tissue samples were obtained from
Xiangya Hospital, Central South University, Hunan,
China. All samples were evaluated by pathologists and Identification of hub genes in the vitamin & cofactor
preserved in paraffin. Only relapse-free survival (RFS) metabolic pathway
data is currently available, as most patients are still alive. The protein–protein interaction (PPI) network was con-
The collection of human tissues was approved by the structed using the STRING database (https://​string-​db.​
Medical Ethics Committee of Xiangya Hospital of Cen- org/) and visualized using the Cytoscape software (Ver-
tral South University (Approval number: 202303046). sion 3.8.2). The hub genes of the vitamin & cofactor
U2OS, MG-63 and THP-1 cell lines were obtained from metabolic pathway were then identified based on the
the Xiangya cell repository and U2OS and MG-63 were PPI network using the “cytohubba” plugin in Cytoscape
cultured in Dulbecco’s modified Eagle’s medium (DMEM, [29]. Gene modules in the vitamin & cofactor metabolic
Biological Industries, Israel) containing 10% fetal bovine pathway were analyzed using the “MCODE” plugin. After
serum (Gibco, USA) at 5% CO2 and 37 °C. THP-1 cell matching the hub genes with RNA-Seq data, univariate
line was cultured in RPMI 1640 medium (Gibco, Thermo Cox regression analysis was performed to determine the
Fisher Scientific, USA). THP-1 cells were differentiated effect of the hub genes on OS in patients with osteosar-
into M0 macrophages by incubation with 100 ng/mL coma. In addition, similar methods were used to analyze
phorbol 12-myristate 13-acetate for 24 h. Typical images the hub genes of other metabolic super-pathways.
of THP-1 cells, M0 macrophages, and M2 macrophages
were shown in Additional file 1: Figure S2. The ST3GAL4
overexpression plasmid was synthesized by Sino Biologi- Identification of metabolic pathway‑related clusters
cal (Beijing, China) and the overexpression efficiency was After identifying key metabolic pathways in osteosar-
shown in Additional file 1: Figure S3. The small interfer- coma, PAM-based unsupervised consensus clustering
ing RNA si-ST3GAL4 and the empty vector si-NC were was used to identify potential metabolic subtypes as we
synthesized by GenePharma (Shanghai, China). A total described previously [30–32]. In brief, 1000 bootstraps
of three siRNAs were validated, among which siRNA-2 were performed and K value was set to 2–10, and the
showed the highest efficiency and was used for subse- optimal number of clusters was defined by the consen-
quent experiments (Additional file 1: Figure S4). sus cumulative distribution function and the consensus
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 4 of 26

heatmap. This method was also used for the identifica- Analysis of drug sensitivity and response
tion of clusters based on seven metabolic pathways. to immunotherapy
As previously described [37], normalized gene expression
data of 809 tumor cell lines and response data for each
Immune infiltration analysis cell line to three guideline-based used chemotherapeutic
The ESTIMATE algorithm is a method for inferring agents (cisplatin, cyclophosphamide and gemcitabine)
the overall level of immune infiltration in tumor tissues for osteosarcoma and one targeted agent (sorafenib)
based on gene expression data and has been widely used with clinical application value were downloaded from
in a large number of previous studies [33, 34]. This study the Drug Sensitivity in Cancer (GDSC) database, and the
used this algorithm to infer the ImmuneScore, StromalS- drug response data were converted to the I­ C50. Then the
core and ESTIMATEScore (inversely correlated with ­IC50 of every drug in individual osteosarcoma patient was
tumor purity) of patients with osteosarcoma. In addition, estimated based on oncoPredict algorithm using the gene
the single sample GSEA (ssGSEA) method was used to expression profile of these cell lines and drug response
infer the levels of 28 immune cell infiltration in osteosar- data as the training set. Maeser et al. provided a detailed
coma based on a previous report [35]. explanation of the usage details of the oncoPredict algo-
rithm [38]. Jiang et al. developed TIDE using RNA-Seq
data from tumors treated with anti-PD1 and anti-CTLA4
Identification of metabolism‑related genes (MRGs) therapies, and identified it as an effective predictor of the
in osteosarcoma responsiveness to these therapies [39]. In this study, it
Using the R package “limma” to compare differentially was used to infer the response of osteosarcoma patients
expressed genes between different metabolic pathway- to immune checkpoint blockade (ICB). The TIDE score
related clusters, the threshold was set to p < 0.05 and is based on two mechanisms of tumor immune escape,
­log2|fold change|> 0.25. These genes were considered as including the dysfunction of tumor-infiltrating cyto-
MRGs. Subsequently, Gene Ontology (GO) enrichment toxic T lymphocytes (CTL) and the exclusion of CTLs
analysis and Kyoto Encyclopedia of Genes and Genomes by immunosuppressive factors, and three cell types that
(KEGG) pathway analysis were performed on MRGs. The limit T cell infiltration in tumors, including the tumor-
prognostic role of MRGs in patients with osteosarcoma associated fibroblasts (CAF), myeloid-derived suppressor
was analyzed using univariate Cox regression. After cells (MDSC), and M2 tumor-associated macrophages
obtaining prognosis-related MRGs, potential MRG-asso- (TAM_M2).
ciated gene clusters were identified using unsupervised
clustering analysis.
Single‑cell RNA‑sequencing (scRNA‑seq) analysis
The scRNA-seq dataset GSE152048 containing 11
Construction of the risk model osteosarcoma samples was downloaded from the GEO
In this study, prognosis-related MRGs were downscaled database [40]. The dataset was processed and analyzed
and hub prognostic genes were obtained using the Least according to the standard procedure of the R package
Absolute Shrinkage and Selection Operator (LASSO) “Seurat” (v.4.3.0), and a total of 26,175 genes and 123,322
regression analysis. Subsequently, the hub prognostic cells were included in the study. After performing data
genes were placed into the stepwise multivariate regres- downscaling and clustering, the clusters were annotated
sion to construct risk models, and the risk model with the using the previously reported cellular markers [40]. In
largest C-index was considered the best risk model. The addition, the expression of ST3 beta-galactoside alpha-
risk model was calculated using the following formula: 2,3-sialyltransferase 4 (ST3GAL4) in the scRNA-seq
dataset GSE162454 was also analyzed using the Tumor
risk score = kj × Expi
Immune Single-cell Hub (TISCH) (http://​tisch.​comp-​
where ­kj is the coefficient of each gene in the risk model, genom​ics.​org/) [41].
and ­Expi is the gene expression. The prediction accuracy
of the risk score was quantified by drawing ROC curves
using the “timeROC” R package. This R package is widely Quantitative reverse transcription‐PCR (RT‐qPCR)
used to estimate time-dependent ROC curves and the Adding 1 ml TriPure, chloroform, isopropanol and 75%
area under the time-dependent ROC curve (AUC) in anhydrous ethanol to the 6-well plate to extract cell RNA.
the presence of censoring data [36]. The package uses After quantification, RNA reverse transcription and RT-
the inverse probability of censoring weighting method to qPCR were performed as described previously [42]. The
estimate and handle censoring data. PCR primers used are listed in Additional file 3: Table S2.
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 5 of 26

Lipofectamine® 3000
Cell proliferation assay Then, cells were fixed and permeabilized with the FIX
After cell transfection using the ­ & PERM Kit (MultiSciences Biotech, Hangzhou, China)
(Invitrogen, Carlsbad, CA, USA) according to the man- and stained with CD206 (321104; Biolegend). A FACS
ufacturer’s instructions, the transfected cells were cul- flow cytometer (BD FACS LSRFortessa, USA) was used
tured for 24 h and inoculated in 96-well plates with 2000 for the flow cytometry analysis.
cells per well. After cell walling, cells were incubated for
different time periods (0, 1, 2 and 3 days) and 10 µl of Immunohistochemistry (IHC)
CCK-8 reagent was added to each well. After incubation IHC was carried out as described previously [43]. The
at 37 °C for 3 h, absorbance was measured at 450 nm to rabbit polyclonal antibody to ST3GAL4 was purchased
determine cell viability. from Invitrogen (PA5-62056, 1:200 dilution). Two
blinded pathologists scored the intensity and percent-
Cell invasion and migration assays age of positive cells for ST3GAL4 staining. The intensity
Cell invasion ability was measured using the Transwell was scored as follows: 0 (negative), 1 (weakly positive), 2
assay and cell migration ability was measured using the (moderately positive), and 3 (strongly positive). The per-
scratch assay. The Transwell and scratch assays were car- centage of ST3GAL4-positive cells was scored as follows:
ried out as described previously [9]. The scratch assay 0 (0%), 1 (1–25%), 2 (26–50%) and 3 (> 50%). The IHC
was incubated for 36 h. For Cell invasion assay, the cells score was defined as the sum of the intensity score and
were resuspended in serum-free medium and placed the percentage score of positive cells.
in the upper chamber of the Transwell system. Culture
medium with 10% serum was added to the lower cham- Statistical analysis
ber and was used as a chemoattractant. After 24 h of Differences between two groups were compared using
incubation, the cells were stained and counted. unpaired Student’s t-test or Wilcoxon rank sum test. For
comparisons between more than two groups, differences
Colony formation assay were compared using one-way ANOVA or Kruskal–
MG-63 and U2OS cells with knocked down or overex- Wallis test. The correlation between two groups was
pressed ST3GAL4 were seeded in 6-well plates at a den- calculated using Sperman’s correlation analysis. Unless
sity of 1000 cells per well. After 10 days of cultivation, the otherwise indicated, statistical significance was set at
cells were fixed with 4% paraformaldehyde at room tem- two-sided p < 0.05. All statistical calculations were per-
perature for 30 min. Subsequently, the cells were stained formed using R software (Version 4.2.1).
with 0.1% crystal violet, and the number of colonies in
each well was counted. Results
Identification of key metabolic pathways in osteosarcoma
Seahorse assays Metabolic heterogeneity may lead to differences in clini-
10,000 tumor cells were seeded in a Seahorse 96-well cal outcomes, and we are committed to exploring the key
assay plate and incubated overnight. The probe plates metabolic pathways associated with clinical outcomes.
were pretreated and the calibration solution was pre- After quantifying the activity of the seven metabolic
pared following the manufacturer’s protocol. Subse- super-pathways, Kaplan–Meier curve analysis identi-
quently, the probe plates were placed in a ­ CO2-free fied four key metabolic pathways that were significantly
incubator overnight. After the overnight incubation, the associated with prognosis. Higher levels of carbohydrate
detection solution was prepared as per the instructions (p = 0.038), energy (p = 0.017), lipid (p = 0.010), and vita-
of the Glycolysis Stress Test kit (Agilent Technologies, min & cofactor metabolism (p = 0.009) were associ-
#103020-100) and the reagents were added sequentially. ated with better OS in osteosarcoma (Fig. 1A). Among
Real-time metabolic changes in cells were detected using them, vitamin & cofactor metabolism has the highest
the Agilent Seahorse XFe96 (Agilent Technologies). significance. These four key metabolic pathways were
first explored in relation to the overall level of immune
Co‑culture experiment and flow cytometry assay infiltration. As shown in Fig. 1B, only vitamin & cofac-
Co-culture was performed using the Boyden chamber, tor metabolism is significantly positively correlated with
M0 macrophages were seeded at upper chamber and overall immune and stromal infiltration and negatively
tumor cells were seeded at lower chamber. After 48 h, correlated with tumor purity. Further analysis of immune
cells from the upper chamber were collected. For flow cell infiltration revealed a potential positive correlation
cytometry assay, cells were prepared for single cell sus- between vitamin & cofactor metabolism and infiltra-
pension and were fixed with 2% paraformaldehyde solu- tion of most immune cells, including activated CD8 T
tion in PBS. cells and activated dendritic cell (Fig. 1C). Consistently,
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 6 of 26

in the subgroup analysis, only the high vitamin & cofac- of which contains many APO family genes (Fig. 2E).
tor metabolism group showed higher levels of immune Furthermore, hub genes were identified in three other
cell infiltration compared to the low vitamin & cofactor prognostic-related metabolic super-pathways (Additional
metabolism group (Additional file 1: Figure S5). Due to file 1: Figure S8-S10). Among them, the top 10 hub genes
the positive correlation with vitamin & cofactor metabo- in the carbohydrate metabolism pathway are not associ-
lism, carbohydrate and lipid metabolism were also posi- ated with prognosis. The top 10 hub genes in the energy
tively correlated with infiltration of some immune cells. metabolism pathway are mainly composed of the G pro-
Immune checkpoint analysis only found a potential tein family, and GNG4 and GNG10 have prognostic sig-
positive correlation between hepatitis A virus cellular nificance. The top 10 hub genes in the lipid metabolism
receptor 2 (HAVCR2) and metabolic pathways (Fig. 1D). pathway are mainly composed of the mediator complex
Moreover, a robust positive correlation prevails among family, and only CD36 among the top 10 genes has prog-
diverse immune characteristics, encompassing immune nostic significance. Moreover, we have also constructed
checkpoints and the infiltration of immune cells (Addi- interaction networks and identified hub genes in three
tional file 1: Figure S6). Core biological pathway analysis additional non-prognostic metabolic super-pathways
revealed that vitamin & cofactor metabolism was poten- (Additional file 1: Figure S11-S13).
tially associated with immune-related biological path-
ways, such as CD8 T effector, antigen processing and Identification of metabolic pathway‑related clusters
immune checkpoint (Fig. 1E). The results of the subgroup and the relationship between clusters and TIME
analysis further confirmed these findings (Additional in osteosarcoma
file 1: Figure S7). Lipid metabolism was associated with To systematically assess the metabolic patterns of osteo-
some immune pathways, but also with cell cycle and mis- sarcoma, the four key metabolic pathways were analyzed
match repair. Overall, vitamin & cofactor metabolism is using unsupervised clustering. As shown in Fig. 3A, oste-
the important metabolic pathway affecting prognosis and osarcoma samples can be clearly divided into two distinct
TIME in osteosarcoma. metabolic pathway-related clusters (C1 and C2). C1 has a
higher level of energy metabolism compared to C2, while
Identification of hub genes in the vitamin & cofactor C2 has a higher lipid and vitamin & cofactor metabolism
metabolic pathway (Fig. 3B, C). Survival analysis showed that C1 patients
Given the importance of the vitamin & cofactor meta- had relatively better long-term OS and RFS, but it did not
bolic pathway, we constructed a PPI network of path- reach statistical significance, which could be attributed to
way genes and found extensive interactions (Fig. 2A). In the limitation in sample size (Fig. 3D, E).
addition, the top 10 hub genes in the vitamin & cofactor We also explored the relationship between metabolic
metabolic pathway were identified according to the PPI clusters and osteosarcoma TIME. Immune checkpoint
network (Fig. 2B). Notably, most of the hub genes belongs analysis showed that C2 had higher CD274 (PD-L1) and
to the apolipoprotein (APO) family, indicating the central HAVCR2 expression than C1 (Fig. 3F), suggesting higher
role of APO family genes in vitamin & cofactor metabo- immunosuppression in C2. Immune cell infiltration anal-
lism. Eight hub genes were subsequently matched in the ysis showed that C2 had a higher infiltration of activated
RNA-Seq data of the TARGET cohort, and correlation CD4 and CD8 T cells, as well as a higher infiltration of
analysis showed some positive correlations among the regulatory T cells that exerted immunosuppressive
APO family genes in the eight hub genes (Fig. 2C). Uni- effects (Fig. 3G, H). However, there was no significant
variate Cox regression analysis showed that APOB and difference in the overall level of immune infiltration
APOE were significantly associated with OS in patients between C1 and C2 samples (Fig. 3I). Core biological
with osteosarcoma (Fig. 2D). Furthermore, two gene pathway analysis revealed that C2 had higher levels of
modules were also identified from the PPI network, one antigen processing, CD8 T effector and mismatch repair,

(See figure on next page.)

Fig. 1 Identification of key metabolic pathways in osteosarcoma. A Kaplan–Meier curves depict the overall survival difference between pathway
activity-high and pathway activity-low groups in the TARGET cohort. Red representing the pathway activity-high group and blue representing
the pathway activity-low group. B Correlations between four key metabolic pathways and immune infiltration scores, only correlations that are
significant are shown. C Correlations between four key metabolic pathways and abundance of 28 immune cells, only correlations that are significant
are shown. D Correlations between four key metabolic pathways and expression of immune checkpoint genes. E Correlations between four key
metabolic pathways and known core biological pathway scores. Correlation coefficients are calculated by Spearman’s correlation analysis, with red
representing negative correlations and blue representing positive correlations. Blank represents a correlation P-value > 0.05
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 7 of 26

Fig. 1 (See legend on previous page.)

Wu et al. Journal of Biomedical Science (2024) 31:4 Page 8 of 26

Fig. 2 PPI network and hub genes in the vitamin & cofactor metabolic pathway. A PPI network of vitamin & cofactor metabolic pathway genes
according to the STRING database. B The top 10 hub genes of vitamin & cofactor metabolic pathway genes. C Correlations among eight matched
hub genes in the TARGET cohort. Red representing negative correlations and blue representing positive correlations. Blank represents a correlation
P-value > 0.05. D Univariate Cox regression analysis of overall survival for eight hub genes. E The two gene modules identified from the PPI network
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 9 of 26

and C1 had higher levels of EMT and fibroblast growth clusters. More detailed analysis of immune cell infiltra-
factor receptor 3 (FGFR3)-related genes (Fig. 3J). It is tion showed that GC2 and GC3 had higher infiltration of
noteworthy that we also identified two clusters based on immune activating and immunosuppressive cells (Fig. 5B,
seven metabolic pathways (7MC1 and 7MC2). However, C). In addition, GC2 and GC3 also had higher expression
there were no obvious differences observed in terms of of immune checkpoint genes (Fig. 5D). Although in the
prognosis, immune checkpoint expression, immune infil- core biological pathway analysis both GC2 and GC3 had
tration, and core biological pathways between 7MC1 and higher activity of immune-related pathways such as anti-
7MC2 (Additional file 1: Figure S14). gen processing, immune checkpoint and CD8 T effector
and lower EMT and FGFR3-related genes. GC2 had sig-
Identification of MRG‑related gene clusters nificantly lower Wnt signaling pathway activity than GC3
To further characterize and understand the biological (Fig. 5E), which may be one of the reasons of the prog-
features of metabolic pathway-related clusters and to nostic differences between the two. Further GSEA anal-
find a more effective classification, we identified 1218 ysis showed that GC3 had the worst prognosis despite
differentially expressed MRGs between C1 and C2. Not high immune infiltration probably due to the highly acti-
surprisingly, GO enrichment analysis showed that MRGs vated MYC and MTOR pathways (Fig. 5F).
were mainly enriched in biological processes related to
antigen presentation and cellular respiration as well as Construction of a metabolism‑related risk model
mitochondria-related cellular components (Fig. 4A). and the relationship between the risk model and TIME
KEGG pathway analysis also showed that MRGs were To further construct a metabolism-related risk predic-
enriched in pathways represented by antigen processing tion tool, 114 prognosis-related MRGs were first down-
and oxidative phosphorylation (Fig. 4B). Then 114 repre- scaled and 25 hub prognosis MRGs were identified using
sentative prognosis-related MRGs were identified by Cox LASSO regression analysis (Additional file 1: Figure
regression analysis (Additional file 4: Table S3). Based S15). The optimal risk model containing 17 core MRGs
on these representative MRGs, osteosarcoma patients was subsequently constructed by stepwise multivariate
were divided into 3 distinct patient groups, termed regression analysis. There are some correlations among
MRG-associated gene clusters 1–3 (GC1-3, Fig. 4C). the expression of the 17 MRGs (Additional file 1: Fig-
GC1 had the highest level of energy metabolism among ure S16). Figure 6A illustrates the coefficients of each
the three clusters, while GC2 and 3 had higher level of MRGs in the risk model. After dividing the TARGET
vitamin & cofactor metabolism (Fig. 4D, E). Survival cohort into two groups by the median risk score, it was
analysis showed significant prognostic differences among seen that patients with osteosarcoma in the low-risk
gene clusters, with GC2 having the best OS (p = 0.0002) group had significantly better OS (Fig. 6B, p < 0.0001).
and the best RFS (p < 0.0001), while GC3 had the worst Using the GSE16091 cohort as an independent validation
(Fig. 4F, G). cohort, although only 15 risk MRGs could be matched,
patients with low risk score in this cohort still had a bet-
The relationship between MRG‑related gene clusters ter prognosis than patients with high risk scores (Fig. 6C,
and TIME in osteosarcoma p = 0.05). The ROC curve showed that the risk score was a
To understand whether significant prognostic differ- good predictor of OS in the TARGET cohort, with AUCs
ences among gene clusters were associated with TIME, of 0.987, 0.979, and 0.985 at 1, 3, and 5 years, respectively
we first assessed the overall immune infiltration differ- (Fig. 6D). In addition, patients with high risk scores had
ences across GC1-3. As shown in Fig. 5A, GC2 had the significantly worse RFS (Fig. 6E, p < 0.0001), and the risk
highest ImmuneScore and the lowest tumor purity, rep- score also had good efficiency in predicting RFS (Fig. 6F).
resenting a better immune response, while there was To understand the differences in TIME across risk
no significant difference in StromalScore among gene groups, ESTIMATE analysis was performed. It found

(See figure on next page.)

Fig. 3 Metabolic pathway-related clusters and the relationship between clusters and TIME in osteosarcoma. A Consensus heatmap based on four
key metabolic pathways in the TARGET cohort. B The heatmap of four key metabolic pathways between C1 and C2. C Differences of four key
metabolic pathways between C1 and C2. D, E Kaplan–Meier curves depict the OS (D) and RFS (E) difference between C1 and C2. Red representing
the C1 patients and blue representing the C2 patients. F Differences of immune checkpoint genes expression between C1 and C2. G The heatmap
of 28 immune cells between the two clusters and the correlations of the clusters and clinical parameters. H Differences of the abundance of 28
immune cells between C1 and C2. I Differences of ImmuneScore, StromalScore and tumor purity between C1 and C2. J Differences of core
biological pathway activity between C1 and C2. *P < 0.05, **P < 0.01, ****P < 0.0001
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 10 of 26

Fig. 3 (See legend on previous page.)

Wu et al. Journal of Biomedical Science (2024) 31:4 Page 11 of 26

that low-risk patients had higher ImmuneScore and Stro- no significant difference in ICB response between C1 and
malScore and lower tumor purity than high-risk patients C2 and between high and low risk groups, but GC3 had
(Fig. 6G), implying that low-risk patients had higher the highest proportion of ICB responders (Fig. 7D), sug-
overall immune and stromal infiltration. The heatmap gesting that the MRG-related gene cluster may be a good
demonstrated the relationship between risk score and predictor of ICB response.
immune cell infiltration and clinical parameters such as Three guideline-based used chemotherapeutic agents
survival status (Additional file 1: Figure S17). As shown (cisplatin, cyclophosphamide and gemcitabine) for osteo-
in Fig. 6H, high-risk patients had significantly lower lev- sarcoma and one targeted agent (sorafenib) with clinical
els of immune cell infiltration, such as activated CD8 T application value were retrieved from the GDSC data-
cells, than low-risk patients. In addition, low-risk patients base. Then, the relationship of these drugs with different
also had higher immune checkpoint gene expression clusters as well as the risk score was explored. As shown
(Fig. 6I). In the correlation analysis, consistent with these in Fig. 7E, C2 patients had lower half-maximal inhibitory
findings, the risk score was negatively correlated with concentration (IC50) values for cisplatin and gemcit-
the level of immune cells infiltration and the expression abine than C1 patients, indicating that C2 patients were
of immune checkpoint genes (Additional file 1: Figure more sensitive to cisplatin and gemcitabine. Among the
S18A-C). In the core biological pathway analysis, the risk MRG-related gene clusters, the IC50 values of cisplatin,
score potentially exhibited a negative correlation with gemcitabine, and sorafenib were sequentially lower in
antigen processing, CD8 T effector, and immune check- GC1-GC3, indicating the different sensitivity of the three
point (Fig. 6J). Differential analysis revealed that patients clusters to these three drugs (Fig. 7F). Although the risk
with high-risk scores had lower enrichment levels of score did not correlate with IC50 values for these drugs
CD8 T effector and immune checkpoints (Additional (Fig. 7G), we screened for 18 drugs that were significantly
file 1: Figure S18D). The alluvial diagram in Fig. 6K dem- associated with the risk score (Additional file 5: Table S4).
onstrated the relationship among metabolism clusters, Therefore, the risk score may serve as the predictor of
MRG-related gene clusters, and risk levels. GC1 patients sensitivity to these drugs.
were mainly from C1 and GC3 patients were mainly from
C2. In addition, most GC2 patients were assigned to the ST3GAL4 is highly expressed in malignant cells
low-risk group and most GC3 patients were assigned to and is closely associated with the TIME of osteosarcoma
the high-risk group. In the above results we identified 17 core MRGs to con-
struct a risk model. The advent of scRNA-seq has ena-
Immunotherapy response and drug sensitivity analysis bled researchers to investigate the activity of genes across
It is well known that higher TIDE score is associated with diverse cell types. The activation of genes in malignant
lower response to ICB treatment [39].The TIDE score cells can significantly impact their biological behav-
helps in the clinical selection of patients who may be suit- ior, consequently influencing tumor progression. To
able for ICB treatment and in identifying responders. delve deeper into the expression patterns of the 17 core
This study explored the relationship between metabo- MRGs across distinct cell types, we initially identified
lism clusters, MRG-related gene clusters, and risk score 11 major cell types using characteristic gene expression
with ICB treatment response by TIDE score. There is no in the scRNA-seq cohort GSE152048 (Fig. 8A, B). Sub-
difference in TIDE score between the different metabo- sequently, ST3GAL4 was found to not only have a high
lism clusters (Fig. 7A). Among the different MRG-related positive coefficient in the risk model, but also to be pre-
gene clusters, GC3 had the lowest CTL dysfunction, dominantly expressed in malignant cells (osteoblastic and
exclusion, and TIDE scores (Fig. 7B), indicating a good chondroblastic osteosarcoma cells) compared to other
response to ICB treatment. The risk score is potentially core MRGs (Fig. 8C, Additional file 1: Figure S19 A, C).
negatively correlated with the CTL dysfunction score, Notably, the violin plot showed that ST3GAL4, rather
but potentially positively correlated with the MDSC than other MRGs, was specifically highly expressed in
score (Fig. 7C). Consistent with these results, there was proliferating osteoblastic osteosarcoma cells (Fig. 8D,

(See figure on next page.)

Fig. 4 Enrichment analysis of MRGs and identification of MRG-related gene clusters. A The top eight enriched terms in GO enrichment analysis
of MRGs. B The KEGG pathway analysis networks of MRGs. C Consensus heatmap based on MRGs in the TARGET cohort. D The heatmap of four
key metabolic pathways among GC1-3. E Differences of four key metabolic pathways among GC1-3. F, G Kaplan–Meier curves depict the OS (F)
and RFS (G) difference among GC1-3. Red representing the GC1 patients, blue representing the GC2 patients, and yellow representing the GC3
patients. *P < 0.05, **P < 0.01
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 12 of 26

Fig. 4 (See legend on previous page.)

Wu et al. Journal of Biomedical Science (2024) 31:4 Page 13 of 26

Fig. 5 The relationship between MRG-related gene clusters and TIME in osteosarcoma. A Differences of ImmuneScore, StromalScore and tumor
purity among GC1-3. B The heatmap of 28 immune cells among the three gene clusters and the correlations of the gene clusters and clinical
parameters. C Differences of the abundance of 28 immune cells among GC1-3. D Differences of immune checkpoint genes expression
among GC1-3. E Differences of core biological pathway activity among GC1-3. F GSEA enrichment plot based on the HALLMARK gene set showing
the relatively significantly enriched pathways in GC3 patients. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 14 of 26

Additional file 1: Figure S19 B, D), suggesting the poten- shorter OS and RFS in osteosarcoma, but its prognos-
tially important role of ST3GAL4 in the proliferation of tic value was not as significant as ST3GAL4 (Additional
osteosarcoma cells. Importantly, it was also verified in file 1: Figure S22C-G). Immune-related analysis found
the scRNA-seq cohort GSE162454 that ST3GAL4 was that only ST3GAL2 was associated with overall immune
predominantly expressed in malignant cells (Fig. 8E). infiltration in osteosarcoma, but immune checkpoint
Further, osteosarcoma patients with high ST3GAL4 analysis failed to find a widespread correlation (Addi-
were found to have significantly worse OS and RFS tional file 1: Figure S22H, I). Furthermore, it was found
(Fig. 8F, all p < 0.001). C2 samples had significantly higher that ST3GAL5 and ST3GAL6 were negatively correlated
ST3GAL4 expression compared to C1 samples (Fig. 8G), with TIDE score, MDSC score, CAF score, exclusion
and, although not statistically significant, GC2 samples of CTLs, and some core biological pathways includ-
had relatively lower ST3GAL4 expression compared to ing EMT (Additional file 1: Figure S22J, K). In sum-
GC1 and GC3 samples (Fig. 8H). Immune checkpoint mary, only ST3GAL4 in the ST3GAL family is associated
analysis revealed that samples with high ST3GAL4 with both the prognosis and immune characteristics of
had significantly lower expression of CD274, cytotoxic osteosarcoma.
T-lymphocyte associated protein 4 (CTLA4), HAVCR2
and programmed cell death 1 ligand 2 (PDCD1LG2) ST3GAL4 is a potential prognostic marker and associated
(Fig. 8I). Samples with high expression of ST3GAL4 also with tumor progression, glycolysis and the M2 polarization
had lower ImmuneScore, StromalScore and higher tumor of macrophages in osteosarcoma
purity (Fig. 8J). Analysis based on the TIDE algorithm To verify the clinical application, IHC staining was per-
revealed that ST3GAL4 expression was positively corre- formed on osteosarcoma and normal tissue samples.
lated with MDSC score and TAM-M2 score (Fig. 8K), but The protein expression of ST3GAL4 was found to be
not correlated with CAF score, TIDE score, and dysfunc- significantly higher in tumor tissue than in normal tis-
tion and exclusion of CTLs (Additional file 1: Figure S20). sue (Fig. 9A). Survival analysis indicated that patients
Taken together, samples with high ST3GAL4 may have with high ST3GAL4 protein expression had shorter RFS
difficulty responding to ICB treatment. The response pre- (Fig. 9B, p = 0.0014). We knocked down and overex-
diction based on the TIDE algorithm also demonstrated pressed ST3GAL4 in the osteosarcoma cell lines MG-63
that patients with high ST3GAL4 had a relatively low ICB and U2OS to explore its effect on the malignant phe-
response rate (Fig. 8L). However, there was no signifi- notype of osteosarcoma cells. After the knockdown of
cant associations between ST3GAL4 and sensitivities to ST3GAL4, the proliferation, invasion, migration and the
cisplatin, cyclophosphamide, gemcitabine and sorafenib ability of colony formation of MG-63 and U2OS were all
(Additional file 1: Figure S21). ST3GAL4 expression was inhibited (Fig. 9C–F). After overexpressing ST3GAL4,
potentially positively correlated with cell cycle and mis- the malignant phenotypes mentioned above were all
match repair and potentially negatively correlated with enhanced.
immune checkpoint and Pan-F-TBRS (Fig. 8M). Although no correlation was found between ST3GAL4
The ST3GAL family consists of six members and the four metabolic super-pathways (Additional
(ST3GAL1-6), and it is necessary to further analyze file 1: Figure S23), based on previous studies [44–48], we
the other members of this family. ScRNA-seq analy- speculate that ST3GAL4 may have a potential associa-
sis showed that other ST3GAL members were not spe- tion with glycolysis. To further explore the relationship
cifically highly expressed in proliferating malignant cells between ST3GAL4 and glycolysis, seahorse assay was
(Additional file 1: Figure S22A, B). In addition, survival conducted. As expected, the knock down of ST3GAL4
analysis showed that only ST3GAL1 was associated with reduced the basal glycolysis level and maximal glycolysis

(See figure on next page.)

Fig. 6 Construction of a metabolism-related risk model and the relationship between the risk model and TIME. A Coefficients for the 17 genes
in the risk model. B Kaplan–Meier curve depicts the OS difference between high-risk and low-risk groups in the TARGET cohort. Red representing
the high-risk group and blue representing the low-risk group. C Kaplan–Meier curve depicts the OS difference between high-risk and low-risk
groups in the GSE16091 cohort. Red representing the high-risk group and blue representing the low-risk group. D ROC curve analysis of the risk
score for OS in the TARGET cohort. E Kaplan–Meier curve depicts the RFS difference between high-risk and low-risk groups in the TARGET cohort.
Red representing the high-risk group and blue representing the low-risk group. F ROC curve analysis of the risk score for RFS in the TARGET cohort.
G Differences of ImmuneScore, StromalScore and tumor purity between high and low risk groups. H Differences of the abundance of 28 immune
cells between high and low risk groups. I Differences of immune checkpoint genes expression between high and low risk groups. J Differences
of core biological pathway activity between high and low risk groups. K Alluvial diagram of metabolism clusters, MRG-related gene clusters, and risk
levels. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 15 of 26

Fig. 6 (See legend on previous page.)

Wu et al. Journal of Biomedical Science (2024) 31:4 Page 16 of 26

level in osteosarcoma cells (Fig. 10A, B). Furthermore, [56]. However, in agreement with our study, Peng et al.
through the co-culture system, we explored the impact of found that lipid metabolism was associated with a better
ST3GAL4 on macrophage polarization. RT-PCR analy- prognosis for a variety of tumors in a pan-cancer analysis,
sis showed that the knock down of ST3GAL4 signifi- and energy metabolism showed a heterogeneous prog-
cantly decreased the expression of the M2 macrophage nostic correlation [26]. Notably, they found that carbohy-
marker CD206 (Fig. 10C). Interestingly, the expres- drate and vitamin & cofactor metabolism were associated
sion of PD-L1 in macrophages was also reduced in the with worse prognosis in tumors, which is different to our
ST3GAL4 knockdown group (Fig. 10C). Flow cytometry results in osteosarcoma. They are highly heterogeneous
analysis further confirmed a lower proportion of M2 tumors containing multiple subtypes including osteo-
macrophages in the ST3GAL4 knockdown group com- blastic and chondroblastic osteosarcoma [57]. The exact
pared to the control group, confirming the regulation of characteristics of osteosarcoma metabolism remain to
ST3GAL4 on macrophage polarization in osteosarcoma be elucidated, which may result in a different metabolic
(Fig. 10D). pattern from other tumors as well as clinical relevance.
It should not be overlooked that a large number of pre-
Discussion vious studies have focused on carbohydrate metabolism
Metabolic reprogramming is considered to be one of the in osteosarcoma [57–60]. In this study, the most signifi-
hallmarks of cancer [2, 49]. The metabolic activity of can- cant difference was found between OS of osteosarcoma
cer is extremely complex and needs to be systematically with different levels of vitamin & cofactor metabolism,
characterized. However, several previous studies have implying that vitamin & cofactor metabolism may largely
demonstrated considerable heterogeneity in the expres- influence the prognosis of osteosarcoma. Importantly,
sion of genes involved in various metabolic pathways vitamin & cofactor metabolism was strongly correlated
[50–54], and thus, metabolic gene expression alone can- with immune and stromal cell infiltration in the TIME of
not accurately reflect the metabolic changes in tumors. osteosarcoma, and carbohydrate and vitamin & cofactor
Based on studies with parallel metabolomics data as well metabolism were also correlated with infiltration levels
as transcriptomics data, Peng et al. proved that meta- of various antitumor immune cells such as effector mem-
bolic pathway-based expression patterns reflect the true ory CD8 T cells. Previous studies have demonstrated
metabolic status well [26]. Therefore, the investigation of that higher immune cell infiltration in osteosarcoma is
osteosarcoma metabolism from metabolic pathways is associated with better prognosis [9], which may par-
equally promising. Current studies on the metabolic pro- tially explain why osteosarcoma patients with high lev-
file of osteosarcoma tend to focus only on a specific met- els of carbohydrate and vitamin & cofactor metabolism
abolic pathway [23, 24, 55], and the authors are not aware have better clinical outcomes. Given the importance of
that any studies have yet examined the impact of differ- vitamin & cofactor metabolism in osteosarcoma prog-
ent metabolic pathways on osteosarcoma from a holistic nosis and TIME, further in-depth study of its mecha-
perspective. In addition, it is well known that metabolic nism in osteosarcoma and development of therapeutic
activity greatly influences the formation of TIME [15– strategies targeting the vitamin & cofactor metabolism
17], therefore it is necessary to further reveal the relation- may be promising. Further, we identified the hub genes
ship between metabolism and TIME in osteosarcoma. in the vitamin & cofactor metabolic pathway. Remark-
In this study, we first explored the impact of enrichment ably, most of the hub genes belonged to the APO family.
levels of the seven most prominent metabolic super-path- APOs are proteins that bind to lipids (such as cholesterol
ways on the prognosis of osteosarcoma. Unexpectedly, and triglycerides) in the blood, forming lipoproteins and
four of the seven metabolic super-pathways (carbohy- transporting them through the bloodstream to cells and
drate, lipid, energy, and vitamin & cofactor) were all asso- tissues. Lipoproteins play an important role in vitamin
ciated with better OS in osteosarcoma. This appears to metabolism. For example, APOA is the main component
be a departure from previous knowledge that cancer cells of high-density lipoprotein (HDL) and is directly corre-
have an increased need for glucose and energy uptake lated with the level of vitamin E in the blood, promoting

(See figure on next page.)

Fig. 7 Immunotherapy response and drug sensitivity analysis. A Differences of TIDE-related scores between C1 and C2. B Differences of TIDE-related
scores among GC1-3. C Correlations of the risk score with TIDE-related scores. D Rate of predicted clinical response to ICB immunotherapy
in different clusters and risk levels. E Differences in IC50 values of cisplatin, cyclophosphamide, gemcitabine and sorafenib between C1 and C2.
F Differences in IC50 values of cisplatin, cyclophosphamide, gemcitabine and sorafenib among GC1-3. G Correlations of the risk score with IC50
values of cisplatin, cyclophosphamide, gemcitabine and sorafenib. *P < 0.05, **P < 0.01, ****P < 0.0001
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 17 of 26

Fig. 7 (See legend on previous page.)

Wu et al. Journal of Biomedical Science (2024) 31:4 Page 18 of 26

its absorption in the intestines [61]. In colon cancer cells, of the metabolism in the immune regulation of osteosar-
APOB also participates in the transport of vitamin E [62]. coma. Previous studies have shown that many metabolic
In addition, APOE largely affects the concentration of genes are also important immunomodulatory genes.
fat-soluble vitamins in plasma [63]. Recent studies have For instance, Wolf et al. found that hexokinase, which is
found that APOC1 promotes osteosarcoma progres- involved in the process of glycolysis, is an important reg-
sion through binding to MTCH2 [64], and preoperative ulator of innate immunity [69], demonstrating the over-
APOB/APOA1 has been identified as an independent lap between cellular energy metabolism and the immune
prognostic factor for osteosarcoma in children and ado- system. PDCD1 (also known as PD-1), a well-known
lescents [65]. In addition, APOD induced the osteoblastic immune checkpoint gene, has been identified to inter-
differentiation of the osteosarcoma cell line Sao-2 [66]. act with arginine biosynthesis or fatty acid degradation
This suggests that the APO family may regulate osteosar- and elongation [70]. According to the identified MRGs,
coma vitamin & cofactor metabolism and affect the prog- we defined MRGs-related gene clusters (GC1-3), which
nosis of osteosarcoma. may have more important clinical translational implica-
Based on the activity levels of the four clinically relevant tions than metabolic pathway-related clusters. Firstly,
key metabolic super-pathways, two distinct metabolic the significantly different clinical outcomes among GCs
pathway-related clusters (C1 and C2) were identified in suggest that these GCs reflect essential aspects of tumor
a cluster analysis. C1 is mainly characterized by energy development and can be used as potential prognostic
metabolism, while C2 is characterized by lipid and vita- predictors. Secondly, different GCs had significantly dif-
min metabolism. Although C1 and C2 do have distinct ferent TIME, that is, GC2 and GC3 had higher immune
metabolic characteristics, there was no significant dif- infiltration and were more likely to respond to ICB treat-
ference in survival between them due to the limitation ment, suggesting that this classification approach may
in sample size. It is necessary to explore the prognostic facilitate the development of personalized ICB treatment
differences between them in larger cohorts in the future. strategies for osteosarcoma. In addition, differences in
It is noteworthy that C2 patients were more sensitive to sensitivity to cisplatin, gemcitabine and sorafenib among
cisplatin and gemcitabine. Therefore, this classification GCs also suggest their potential to guide clinical dos-
may be appropriate to identify osteosarcoma patients ing. From a mechanistic perspective, GC3 with the worst
with different metabolic profiles and to guide the dosing prognosis exhibits higher activity in the MYC and mTOR
of cisplatin and gemcitabine. To further explore the clini- pathways. MYC is one of the most frequently dysregu-
cal significance of the metabolic profile of osteosarcoma, lated oncogenes known so far, highly expressed in various
we identified MRGs based on metabolic pathway-related tumors including osteosarcoma [71]. Its expression pro-
clusters. It should not be overlooked that MRGs were motes tumor progression by providing sufficient energy
enriched not only in metabolism-related pathways, such and metabolic substrates for uncontrolled cell prolifera-
as oxidative phosphorylation, but also in immune-related tion [72, 73]. The mTOR pathway is also abnormally acti-
pathways, including antigen processing and presenta- vated in many cancers, including human osteosarcoma
tion. Previous studies have demonstrated that metabo- [74]. In osteosarcoma, mTOR promotes cell growth and
lites in TIME affect the differentiation and function of proliferation, induces cell metastasis, inhibits apoptosis,
immune cells, thereby modulating the immune response and suppresses autophagy [74]. Therefore, the activa-
[67, 68]. The results further emphasize the importance tion of the MYC and mTOR pathways may be one of the

(See figure on next page.)

Fig. 8 ST3GAL4 is highly expressed in malignant cells and is closely associated with the TIME of osteosarcoma. A The dot plot shows the expression
of 41 characteristic genes in 11 cell clusters. The size of the dots indicates the proportion of cells expressing a specific marker, and the color
indicates the average expression level of the markers. MSC, mesenchymal stem cell; TIL, tumor-infiltrating lymphocyte. B The t-SNE plot
of the 11 main cell types in osteosarcoma. C Feature plot for ST3GAL4. The color legend shows the normalized expression levels of the genes.
D Violin plot showing the normalized expression levels of ST3GAL4 across the 11 cell types. E Expression of ST3GAL4 among different cell types
in the GSE162454 cohort. The color indicates the average expression level of ST3GAL4. F Kaplan–Meier curves depict the OS and RFS difference
between high-ST3GAL4 and low-ST3GAL4 groups in the TARGET cohort. Red representing the high-ST3GAL4 group and blue representing
the low-ST3GAL4 group. G Differences of the expression of ST3GAL4 between C1 and C2. H Differences of the expression of ST3GAL4 among GC1-3.
I Differences of immune checkpoint genes expression between high-ST3GAL4 and low-ST3GAL4 groups. *P < 0.05. J Differences of ImmuneScore,
StromalScore and tumor purity between high-ST3GAL4 and low-ST3GAL4 groups. K Correlations of the expression of ST3GAL4 with MDSC
score and TAM-M2 score. L Rate of predicted clinical response to ICB immunotherapy in high-ST3GAL4 and low-ST3GAL4 groups. M Correlations
between the expression of ST3GAL4 and known core biological pathway scores. Correlation coefficients are calculated by Spearman’s correlation
analysis, with red representing negative correlations and blue representing positive correlations. Blank represents a correlation P-value > 0.05
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 19 of 26

Fig. 8 (See legend on previous page.)

Wu et al. Journal of Biomedical Science (2024) 31:4 Page 20 of 26

intrinsic factors contributing to the poor prognosis of no association was found between ST3GAL4 and four
GC3. It is worth noting that the Wnt signaling pathway is metabolic super-pathways, this study confirmed the
highly active in both GC1 and GC3. This signaling path- involvement of ST3GAL4 in glycolysis in osteosarcoma.
way is associated with tumorigenesis and can regulate Similarly, previous studies indicated that ST6GAL1
the metastasis of osteosarcoma cells through autocrine regulates glycolysis in ovarian cancer [45]. Liu et al.
or paracrine mechanisms, thus reducing patient survival identified ST3GAL4 as a hypoxia-related gene [47], and
rate [75]. This may be one of the reasons for the poor it is well-known that a hypoxic microenvironment can
prognosis of GC3. Admittedly, in GC1, the activation induce glycolysis in tumor cells [48]. These findings
of immune response contributes to a better prognosis support our results. Additionally, the hyperactivation
even with higher Wnt pathway activity. This reflects the of glycolysis in tumors sustains and promotes various
importance of immune response in the survival of GC1. malignant behaviors in osteosarcoma cells [80], which
Furthermore, we downscaled the MRGs by mul- may also be a potential mechanism by which ST3GAL4
tiple algorithms and identified 17 core MRGs and influences the malignant phenotype of osteosarcoma
constructed a risk model. It has shown that risk strati- cells. In addition, it was found that samples with high
fication could significantly improve the treatment out- expression of ST3GAL4 were mainly enriched in cell
come of many tumors including osteosarcoma [9, 76, cycle and DNA repair-related pathways, which may
77]. This risk model has good efficacy and was vali- be a potential mechanism by which ST3GAL4 pro-
dated in an independent cohort, suggesting its poten- motes malignant phenotypes (Additional file 1: Figure
tial value of clinical application. In addition, the risk S24A). We further provided GSEA results of other core
model reflects the different immune status for osteo- MRGs (Additional file 1: Figure S24B) to explore their
sarcoma, whereby patients with higher risk score have relationship with tumor-related pathways and suggest
lower immune infiltration, which is consistent with potential therapeutic targets.
previous studies [9]. Remarkably, scRNA-seq-based ST3GAL4 was also associated with the immune
analysis revealed that ST3GAL4, one of the 17 core response in osteosarcoma and may be an important regu-
MRGs, was highly expressed in proliferating malignant lator of the TIME of osteosarcoma. This study confirmed
cells. Mechanistically, ST3GAL4 was also associated the regulatory effect of ST3GAL4 on macrophage polari-
with cell cycle and mismatch repair, further suggest- zation in osteosarcoma using a co-culture system. Studies
ing that ST3GAL4 influences the development of oste- have demonstrated that lactate, a metabolite generated
osarcoma. The ST3GAL4 gene encodes the enzyme during the glycolytic process in tumor cells, plays a role in
Galβ1-4GlcNAc α2,3-sialyltransferase. This enzyme inducing M2 polarization in macrophages, thereby exert-
is involved in protein glycosylation and the synthe- ing direct immune-suppressive effects [81]. Our find-
sis of the sialyl Lewis x antigen [78]. Previous studies ings provide evidence for the involvement of ST3GAL4
have shown that ST3GAL4 affects several biological in promoting glycolysis, which could partially explain
behaviors in tumors such as proliferation, invasion, and its role in regulating macrophage polarization. A recent
migration in non-small cell lung cancer and pancreatic study has shown that ST3GAL4 is not only involved in
cancer cells [42, 79]. The present study demonstrated protein glycosylation processes, but also affects the sign-
for the first time that knockdown of ST3GAL4 in cell aling pathways of Siglec-7 and Siglec-9 by promoting the
lines (MG-63 and U2OS) suppressed the malignant synthesis of ligands in tumor cells, thereby promoting
phenotype of osteosarcoma. More importantly, we con- macrophage polarization [82]. This further validates our
firmed the clinical feasibility of ST3GAL4 as a prognos- findings and potentially unveils additional mechanisms
tic marker in an independent clinical cohort. Although through which ST3GAL4 facilitates macrophage M2

(See figure on next page.)

Fig. 9 Protein expression of ST3GAL4 in osteosarcoma tissues and its effects on proliferation, invasion and migration of osteosarcoma cells.
A IHC staining images of ST3GAL4 in osteosarcoma tissues (#6, n = 14) and corresponding normal tissues (#2, n = 5). The IHC scores indicated
that the protein expression of ST3GAL4 was higher in tumor tissues. B Kaplan–Meier curve depicts the RFS difference between high and low
ST3GAL4 protein group in the Xiangya cohort. Red representing the high ST3GAL4 protein group and blue representing the low ST3GAL4 protein
group. C Folded line plots showing the effect of ST3GAL4 knockdown and overexpression on the proliferation of MG-63 and U2OS cells. The blue
line represents the control group and the yellow line represents the knockdown/overexpression group. D Transwell chamber experiments showing
the effect of ST3GAL4 knockdown and overexpression on the invasion of MG-63 and U2OS cells. Scale bar: 100 μm. E Scratch assays showing
the effect of ST3GAL4 knockdown and overexpression on the migration of MG-63 and U2OS cells. F Colony formation assays showing the effect
of ST3GAL4 knockdown and overexpression on the ability of colony formation of MG-63 and U2OS cells. Data are represented as mean ± SEM.
*P < 0.05, **P < 0.01, ***P < 0.001
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 21 of 26

Fig. 9 (See legend on previous page.)

Wu et al. Journal of Biomedical Science (2024) 31:4 Page 22 of 26

Fig. 10 ST3GAL4 regulates the glycolysis of tumor cells and the M2 polarization of macrophages in osteosarcoma. A, B Seahorse assays indicated
that the knock down of ST3GAL4 inhibited glycolysis in osteosarcoma cells. Left, representative curve; Right, quantification of basal ECAR and maxi
ECAR. ECAR, extracellular acidification rate. C RT-qPCR analysis is shown for PD-L1 and M2 marker CD206 in macrophages. D Flow cytometry analysis
is shown for expression of CD206 in macrophages cultured with si-NC or si-ST3GAL4 tumor cells. Shown are representative plots and quantification
of the percentage of CD206 positive cells in total macrophages

polarization. These findings provide support for consid- it is reasonable that there are some differences between
ering ST3GAL4 as a promising and innovative target for ST3GAL4 expression and risk score or C1/C2 in response
cancer immunotherapy. to ICB, because the population with high/low ST3GAL4
However, further investigation is required to further expression does not completely overlap with the popula-
elucidate the role of ST3GAL4 in the TIME of osteosar- tion represented by high/low risk group or C1/C2. Over-
coma. Notably, while patients exhibiting high ST3GAL4 all, our findings support the potential utility of ST3GAL4
expression demonstrated a relatively low response rate as a prognostic marker and a new therapeutic target.
to ICB, there was minimal disparity in ICB response There are still some limitations in this study. Firstly,
rates between the high-risk and risk groups, as well as due to the rarity of osteosarcoma, it is difficult to obtain
between subtypes C1 and C2. This is because there is no a large sample cohort to validate the results. Secondly,
absolute linear relationship between ST3GAL4 expres- using an osteosarcoma cohort with parallel metabo-
sion and risk score or C1/C2. There are a large number lomics and transcriptomics data would increase the value
of differentially expressed genes between high and low of this study. In addition, the effects and mechanisms of
risk groups or C1 and C2, not just ST3GAL4. Therefore, vitamin & cofactor metabolism and ST3GAL4 on the
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 23 of 26

TIME of osteosarcoma require further in-depth in vivo cells, M0 macrophages, and M2 macrophages. Figure S3. The efficiency
and in vitro studies. of ST3GAL4 overexpression plasmid in this study. Figure S4. The efficiency
of ST3GAL4 siRNAs in this study. Figure S5. Differences of the overall
immune infiltration (A) and 28 immune cells infiltration (B) between high
and low metabolism groups. Figure S6. The correlation between various
Conclusion immune features in osteosarcoma. Figure S7. Differences of the expres-
Vitamin & cofactor metabolism plays an important role sion of immune checkpoints (A) and core biological pathway activity
(B) between high and low metabolism groups. Figure S8. PPI network
in the prognosis and TIME of osteosarcoma. MRG- and hub genes in the carbohydrate metabolic pathway. Figure S9. PPI
related gene clusters can reflect the immune status of network and hub genes in the energy metabolic pathway. Figure S10.
osteosarcoma and facilitate the development of person- PPI network and hub genes in the lipid metabolic pathway. Figure S11.
PPI network and hub genes in the amino acid metabolic pathway. Figure
alized immunotherapy and chemotherapy strategies. S12. PPI network and hub genes in the nucleotide metabolic path-
The metabolism-related risk model may serve as a useful way. Figure S13. PPI network and hub genes in the TCA cycle metabolic
prognostic predictor. ST3GAL4 plays a critical role in the pathway. Figure S14. Metabolic pathway-related clusters based on seven
metabolic super-pathways and the relationship between clusters and
progression, glycolysis, and TIME of osteosarcoma and TIME in osteosarcoma. Figure S15. LASSO regression of 114 prognosis-
affects the prognosis. related MRGs. Figure S16. Correlations among 17 MRGs of the risk model
in the TARGET cohort. Figure S17. The heatmap of 28 immune cells
between high and low risk groups. Figure S18. Correlations between
Abbreviations risk score and overall immune infiltration (A), 28 immune cells infiltra-
TIME Tumor immune microenvironment tion (B), and the expression of immune checkpoints (C) and differences
MRG Metabolism-related gene of core biological pathway activity between high and low risk score
TCA​ Tricarboxylic acid (D). Figure S19. Feature plots and violin plots for the core MRGs in the risk
GSVA Gene set variation analysis model. Figure S20. The relationships between ST3GAL4 and CAF score,
GSEA Gene set enrichment analysis TIDE score, and dysfunction and exclusion of CTLs. Figure S21. The rela-
ssGSEA Single sample gene set enrichment analysis tionships between ST3GAL4 and sensitivities to cisplatin, cyclophospha-
OS Overall survival mide, gemcitabine and sorafenib. Figure S22. The prognosis and immu-
RFS Relapse-free survival nological features of other members in the ST3GAL family. Figure S23.
PPI Protein–protein interaction Correlations of the expression of ST3GAL4 with metabolic super-pathways
GO Gene ontology (A) and differences of metabolic super-pathways activity between high-
KEGG Kyoto encyclopedia of genes and genomes ST3GAL4 and low-ST3GAL4 groups (B). Figure S24. The GSEA results of
GDSC Genomics of drug sensitivity in cancer ST3GAL4 (A) and other core MRGs (B) based on KEGG gene set.
ICB Immune checkpoint blockade Additional file 2. Table S1. Gene sets for seven metabolic
CTL Tumor-infiltrating cytotoxic T lymphocyte super-pathways.
CAF Tumor-associated fibroblast
MDSC Myeloid-derived suppressor cell Additional file 3. Table S2. The primers used in the PCR reaction.
TAM_M2 M2 tumor-associated macrophage Additional file 4. Table S3. The 114 representative prognosis-related
scRNA-seq Single-cell RNA-sequencing MRGs were identified by Cox regression analysis.
IHC Immunohistochemistry
ST3GAL4 ST3 beta-galactoside alpha-2,3-sialyltransferase 4 Additional file 5. Table S4. Correlations between risk score and IC50 of
TISCH Tumor immune single-cell hub drugs.
APO Apolipoprotein
FGFR3 Fibroblast growth factor receptor 3
Acknowledgements
EMT Epithelial mesenchymal transition
Not applicable.
ROC Receiver operating characteristic
AUC​ Area under the time-dependent ROC curve
Author contributions
IC50 Half-maximal inhibitory concentration
Conception and design: CW, HS, NS. Collection and assembly of data: CW.
HIF-1α Hypoxia inducible factor 1 subunit alpha
Data analysis and interpretation: CW, JT. Conducting experiments: CW, CD.
LASSO Least absolute shrinkage and selection operator
Manuscript writing and revisions: CW, JT, NS, CK, GO. Final approval of manu-
HAVCR2 Hepatitis A virus cellular receptor 2
script: All authors.
CTLA4 Cytotoxic T-lymphocyte associated protein 4
PDCD1LG2 Programmed cell death 1 ligand 2
Funding
IDO1 Indoleamine 2,3-dioxygenase 1
This study was supported by two Grants from National Natural Science
PDCD1 Programmed cell death 1
Foundation of China (No. 82203397 and No. 82303253), one Grant from the
LAG3 Lymphocyte activating 3
Nature Science Foundation of Hunan Province (No. 2022JJ40814), one Grant
HSPG2 Heparan sulfate proteoglycan 2
from the Nature Science Foundation of Changsha (No. kq2202375), one Grant
AGRN Agrin
from the Youth Foundation of Xiangya Hospital (NO. 2021Q06), and two grants
SDC1 Syndecan 1
from the China Postdoctoral Science Foundation (No. 2022M723560 and No.
MTCH2 Mitochondrial carrier 2
2023M733960).
ECAR​ Extracellular acidification rate
Availability of data and materials
Supplementary Information The datasets analyzed during the current study are available in the TARGET
database (https://​ocg.​cancer.​gov/​progr​ams/​target) and GEO database
The online version contains supplementary material available at https://​doi.​
(https://​www.​ncbi.​nlm.​nih.​gov/​geo/). Further information is available from the
org/​10.​1186/​s12929-​024-​00999-7.
corresponding author on reasonable request.

Additional file 1. Figure S1. The gene expression data distribution of the
TARGET (A) and GEO (B) databases. Figure S2. Typical images of THP-1
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 24 of 26

Declarations 14. Ali S, Kjeken R, Niederlaender C, Markey G, Saunders TS, Opsata M, et al.
The European Medicines Agency Review of Kymriah (Tisagenlecleucel)
Ethics approval and consent to participate for the treatment of acute lymphoblastic leukemia and diffuse large
This study was approved by the Medical Ethics Committee of Xiangya Hospital B-cell lymphoma. Oncologist. 2020;25:e321–7. https://​doi.​org/​10.​1634/​
of Central South University (Approval number: 202303046). THEON​COLOG​IST.​2019-​0233.
15. Vander Heiden MG, DeBerardinis RJ. Understanding the Intersections
Consent for publication between Metabolism and Cancer Biology. Cell. 2017;168:657–69. https://​
Not applicable. doi.​org/​10.​1016/J.​CELL.​2016.​12.​039.
16. DeBerardinis RJ, Lum JJ, Hatzivassiliou G, Thompson CB. The biology of
Competing interests cancer: metabolic reprogramming fuels cell growth and proliferation. Cell
The authors declared no potential conflicts of interest. Metab. 2008;7:11–20. https://​doi.​org/​10.​1016/J.​CMET.​2007.​10.​002.
17. Vander Heiden MG. Targeting cancer metabolism: a therapeutic window
Author details opens. Nat Rev Drug Discov. 2011;10:671–84. https://​doi.​org/​10.​1038/​
1
Department of Neurosurgery, Xiangya Hospital, Central South University, NRD35​04.
Changsha, Hunan, China. 2 National Clinical Research Center for Geriatric 18. Ippolito L, Morandi A, Giannoni E, Chiarugi P. Lactate: a metabolic driver
Disorders, Xiangya Hospital, Central South University, Changsha, Hunan, in the tumour landscape. Trends Biochem Sci. 2019;44:153–66. https://​
China. 3 Department of Oncology, Xiangya Hospital, Central South University, doi.​org/​10.​1016/J.​TIBS.​2018.​10.​011.
Changsha, Hunan, China. 4 Key Laboratory for Molecular Radiation Oncology 19. Reznik E, Luna A, Aksoy BA, Liu EM, La K, Ostrovnaya I, et al. A landscape
of Hunan Province, Xiangya Hospital, Central South University, Changsha, of metabolic variation across tumor types. Cell Syst. 2018;6:301-313.e3.
Hunan, China. 5 Department of Orthopedics, Xiangya Hospital, Central South https://​doi.​org/​10.​1016/J.​CELS.​2017.​12.​014.
University, Changsha, Hunan, China. 6 Sarcoma Center, Department of Ortho- 20. Hensley CT, Faubert B, Yuan Q, Lev-Cohain N, Jin E, Kim J, et al. Metabolic
pedics, Trauma and Plastic Surgery, University Hospital Leipzig, Leipzig, heterogeneity in human lung tumors. Cell. 2016;164:681–94. https://​doi.​
Germany. org/​10.​1016/J.​CELL.​2015.​12.​034.
21. Kim J, DeBerardinis RJ. Mechanisms and implications of metabolic hetero-
Received: 10 July 2023 Accepted: 2 January 2024 geneity in cancer. Cell Metab. 2019;30:434–46. https://​doi.​org/​10.​1016/J.​
CMET.​2019.​08.​013.
22. Qi F, Li J, Qi Z, Zhang J, Zhou B, Yang B, et al. Comprehensive metabolic
profiling and genome-wide analysis reveal therapeutic modalities for
hepatocellular carcinoma. Research (Washington DC). 2023;6:36. https://​
References doi.​org/​10.​34133/​resea​rch.​0036.
1. Bielack SS, Kempf-Bielack B, Delling G, Exner GU, Flege S, Helmke K, 23. Xu F, Yan J, Peng Z, Liu J, Li Z. Comprehensive analysis of a glycolysis and
et al. Prognostic factors in high-grade osteosarcoma of the extremities cholesterol synthesis-related genes signature for predicting prognosis
or trunk: an analysis of 1,702 patients treated on neoadjuvant coopera- and immune landscape in osteosarcoma. Front Immunol. 2022. https://​
tive osteosarcoma study group protocols. J Clin Oncol. 2002;20:776–90. doi.​org/​10.​3389/​FIMMU.​2022.​10960​09/​FULL.
https://​doi.​org/​10.​1200/​jco.​2002.​20.3.​776. 24. Li Z, Jin C, Lu X, Zhang Y, Zhang Y, Wen J, et al. Studying the mechanism
2. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. underlying lipid metabolism in osteosarcoma based on transcriptomic
Cell. 2011;144:646–74. https://​doi.​org/​10.​1016/J.​CELL.​2011.​02.​013. RNA sequencing and single-cell data. J Gene Med. 2023. https://​doi.​org/​
3. Heiden MGV, Cantley LC, Thompson CB. Understanding the War- 10.​1002/​JGM.​3491.
burg effect: the metabolic requirements of cell proliferation. Science. 25. Paoloni M, Davis S, Lana S, Withrow S, Sangiorgi L, Picci P, et al. Canine
2009;324:1029–33. https://​doi.​org/​10.​1126/​SCIEN​CE.​11608​09. tumor cross-species genomics uncovers targets linked to osteosarcoma
4. Feng Z, Ou Y, Hao L. The roles of glycolysis in osteosarcoma. Front Phar- progression. BMC Genomics. 2009;10:625. https://​doi.​org/​10.​1186/​
macol. 2022. https://​doi.​org/​10.​3389/​fphar.​2022.​950886. 1471-​2164-​10-​625.
5. Zhang J, Pavlova NN, Thompson CB. Cancer cell metabolism: the essential 26. Peng X, Chen Z, Farshidfar F, Xu X, Lorenzi PL, Wang Y, et al. Molecular
role of the nonessential amino acid, glutamine. EMBO J. 2017;36:1302–15. characterization and clinical relevance of metabolic expression subtypes
https://​doi.​org/​10.​15252/​EMBJ.​20169​6151. in human cancers. Cell Rep. 2018;23:255-269.e4. https://​doi.​org/​10.​1016/j.​
6. Heymann MF, Lézot F, Heymann D. The contribution of immune infiltrates celrep.​2018.​03.​077.
and the local microenvironment in the pathogenesis of osteosarcoma. 27. Mariathasan S, Turley SJ, Nickles D, Castiglioni A, Yuen K, Wang Y, et al.
Cell Immunol. 2019. https://​doi.​org/​10.​1016/j.​celli​mm.​2017.​10.​011. TGFβ attenuates tumour response to PD-L1 blockade by contributing to
7. Cortini M, Avnet S, Baldini N. Mesenchymal stroma: role in osteosarcoma exclusion of T cells. Nature. 2018;554:544–8. https://​doi.​org/​10.​1038/​natur​
progression. Cancer Lett. 2017;405:90–9. https://​doi.​org/​10.​1016/j.​canlet.​ e25501.
2017.​07.​024. 28. Shen W, Song Z, Zhong X, Huang M, Shen D, Gao P, et al. Sangerbox:
8. McGranahan N, Swanton C. Clonal heterogeneity and tumor evolution: a comprehensive, interaction-friendly clinical bioinformatics analysis
past, present, and the future. Cell. 2017;168:613–28. https://​doi.​org/​10.​ platform. IMeta. 2022;1: e36. https://​doi.​org/​10.​1002/​imt2.​36.
1016/j.​cell.​2017.​01.​018. 29. Chin CH, Chen SH, Wu HH, Ho CW, Ko MT, Lin CY. cytoHubba: identifying
9. Wu C, Gong S, Duan Y, Deng C, Kallendrusch S, Berninghausen L, et al. A hub objects and sub-networks from complex interactome. BMC Syst Biol.
tumor microenvironment-based prognostic index for osteosarcoma. J 2014. https://​doi.​org/​10.​1186/​1752-​0509-8-​S4-​S11.
Biomed Sci. 2023;30:1–22. https://​doi.​org/​10.​1186/​S12929-​023-​00917-3. 30. Wu C, Qin C, Long W, Wang X, Xiao K, Liu Q. Tumor antigens and immune
10. Zhang T, Guo S, Li F, Lan X, Jia Y, Zhang J, et al. Imaging-guided/improved subtypes of glioblastoma: the fundamentals of mRNA vaccine and indi-
diseases management for immune-strategies and beyond. Adv Drug vidualized immunotherapy development. J Big Data. 2022. https://​doi.​
Deliv Rev. 2022. https://​doi.​org/​10.​1016/J.​ADDR.​2022.​114446. org/​10.​1186/​s40537-​022-​00643-x.
11. Zheng B, Ren T, Huang Y, Sun K, Wang S, Bao X, et al. PD-1 axis expression 31. Wu C, Long W, Qin C, Wang X, Li Y, Xiao K, et al. Liquid biopsy-based
in musculoskeletal tumors and antitumor effect of nivolumab in osteo- identification of prognostic and immunotherapeutically relevant gene
sarcoma model of humanized mouse. J Hematol Oncol. 2018. https://​doi.​ signatures in lower grade glioma. J Big Data. 2023;10:19. https://​doi.​org/​
org/​10.​1186/​S13045-​018-​0560-1. 10.​1186/​s40537-​023-​00686-8.
12. Brohl AS, Sindiri S, Wei JS, Milewski D, Chou HC, Song YK, et al. Immuno- 32. Wu C, Duan Y, Gong S, Osterhoff G, Kallendrusch S, Schopow N. Identi-
transcriptomic profiling of extracranial pediatric solid malignancies. Cell fication of tumor antigens and immune subtypes for the development
Rep. 2021. https://​doi.​org/​10.​1016/J.​CELREP.​2021.​110047. of mRNA vaccines and individualized immunotherapy in soft tissue
13. Wen Y, Tang F, Tu C, Hornicek F, Duan Z, Min L. Immune checkpoints in sarcoma. Cancers (Basel). 2022. https://​doi.​org/​10.​3390/​cance​rs140​20448.
osteosarcoma: recent advances and therapeutic potential. Cancer Lett. 33. Yoshihara K, Shahmoradgoli M, Martínez E, Vegesna R, Kim H, Torres-
2022. https://​doi.​org/​10.​1016/J.​CANLET.​2022.​215887. Garcia W, et al. Inferring tumour purity and stromal and immune cell
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 25 of 26

admixture from expression data. Nat Commun. 2013. https://​doi.​org/​10.​ metabolic network. Nat Biotechnol. 2013;31:522–9. https://​doi.​org/​10.​1038/​
1038/​ncomm​s3612. nbt.​2530.
34. Gong S, Wu C, Köhler F, Meixensberger J, Schopow N, Kallendrusch S. 53. Haider S, McIntyre A, van Stiphout RGPM, Winchester LM, Wigfield S, Harris
Procollagen-lysine, 2-oxoglutarate 5-dioxygenase family: novel prognos- AL, et al. Genomic alterations underlie a pan-cancer metabolic shift associ-
tic biomarkers and tumor microenvironment regulators for lower-grade ated with tumour hypoxia. Genome Biol. 2016. https://​doi.​org/​10.​1186/​
glioma. Front Cell Neurosci. 2022. https://​doi.​org/​10.​3389/​fncel.​2022.​ s13059-​016-​0999-8.
838548. 54. Gaude E, Frezza C. Tissue-specific and convergent metabolic transformation
35. Charoentong P, Finotello F, Angelova M, Mayer C, Efremova M, Rieder of cancer correlates with metastatic potential and patient survival. Nat Com-
D, et al. Pan-cancer immunogenomic analyses reveal genotype-immu- mun. 2016. https://​doi.​org/​10.​1038/​ncomm​s13041.
nophenotype relationships and predictors of response to checkpoint 55. Zhong Y, Zhang Y, Wei S, Chen J, Zhong C, Cai W, et al. Dissecting the effect
blockade. Cell Rep. 2017;18:248–62. https://​doi.​org/​10.​1016/j.​celrep.​2016.​ of sphingolipid metabolism gene in progression and microenvironment of
12.​019. osteosarcoma to develop a prognostic signature. Front Endocrinol (Laus-
36. Blanche P, Dartigues J-F, Jacqmin-Gadda H. Estimating and comparing anne). 2022. https://​doi.​org/​10.​3389/​FENDO.​2022.​10306​55.
time-dependent areas under receiver operating characteristic curves for 56. Pavlova NN, Thompson CB. The emerging hallmarks of cancer metabolism.
censored event times with competing risks. Stat Med. 2013;32:5381–97. Cell Metab. 2016;23:27–47. https://​doi.​org/​10.​1016/j.​cmet.​2015.​12.​006.
https://​doi.​org/​10.​1002/​sim.​5958. 57. Leite TC, Watters RJ, Weiss KR, Intini G. Avenues of research in dietary inter-
37. Wu C, Tan J, Wang X, Qin C, Long W, Pan Y, et al. Pan-cancer analyses ventions to target tumor metabolism in osteosarcoma. J Transl Med. 2021.
reveal molecular and clinical characteristics of cuproptosis regulators. https://​doi.​org/​10.​1186/​s12967-​021-​03122-8.
IMeta. 2023;2: e68. https://​doi.​org/​10.​1002/​IMT2.​68. 58. Chen W, Lin Y, Huang J, Yan Z, Cao H. A novel risk score model based on
38. Maeser D, Gruener RF, Huang RS. oncoPredict: an R package for predict- glycolysis-related genes and a prognostic model for predicting overall
ing in vivo or cancer patient drug response and biomarkers from cell line survival of osteosarcoma patients. J Orthop Res. 2022;40:2372–81. https://​
screening data. Brief Bioinform. 2021;22:1–7. https://​doi.​org/​10.​1093/​BIB/​ doi.​org/​10.​1002/​jor.​25259.
BBAB2​60. 59. Yang G, Jiang J, Yin R, Li Z, Li L, Gao F, et al. Two novel predictive biomarkers
39. Jiang P, Gu S, Pan D, Fu J, Sahu A, Hu X, et al. Signatures of T cell dysfunc- for osteosarcoma and glycolysis pathways: a profiling study on HS2ST1 and
tion and exclusion predict cancer immunotherapy response. Nat Med. SDC3. Med (United States). 2022;101:E30192. https://​doi.​org/​10.​1097/​MD.​
2018;24:1550–8. https://​doi.​org/​10.​1038/​s41591-​018-​0136-1. 00000​00000​030192.
40. Zhou Y, Yang D, Yang Q, Lv X, Huang W, Zhou Z, et al. Single-cell RNA 60. Huang W, Xiao Y, Wang H, Chen G, Li K. Identification of risk model based on
landscape of intratumoral heterogeneity and immunosuppressive micro- glycolysis-related genes in the metastasis of osteosarcoma. Front Endocrinol
environment in advanced osteosarcoma. Nat Commun. 2020;11:6322. (Lausanne). 2022;13:1047433. https://​doi.​org/​10.​3389/​fendo.​2022.​10474​33.
https://​doi.​org/​10.​1038/​s41467-​020-​20059-6. 61. Reboul E. Vitamin E intestinal absorption: Regulation of membrane trans-
41. Sun D, Wang J, Han Y, Dong X, Ge J, Zheng R, et al. TISCH: a comprehen- port across the enterocyte. IUBMB Life. 2019;71:416–23. https://​doi.​org/​10.​
sive web resource enabling interactive single-cell transcriptome visualiza- 1002/​iub.​1955.
tion of tumor microenvironment. Nucleic Acids Res. 2021;49:D1420–30. 62. Anwar K, Iqbal J, Hussain MM. Mechanisms involved in vitamin e transport
https://​doi.​org/​10.​1093/​nar/​gkaa1​020. by primary enterocytes and in vivo absorption. J Lipid Res. 2007;48:2028–38.
42. Guerrero PE, Miró L, Wong BS, Massaguer A, Martínez-Bosch N, de Llorens https://​doi.​org/​10.​1194/​jlr.​M7002​07-​JLR200.
R, et al. Knockdown of α2,3-sialyltransferases impairs pancreatic cancer 63. Ortega H, Castilla P, Gómez-Coronado D, Garcés C, Benavente M, Rodríguez-
cell migration, invasion and e-selectin-dependent adhesion. Int J Mol Sci. Artalejo F, et al. Influence of apolipoprotein E genotype on fat-soluble
2020;21:1–24. https://​doi.​org/​10.​3390/​IJMS2​11762​39. plasma antioxidants in Spanish children. Am J Clin Nutr. 2005;81:624–32.
43. Wu C, Su J, Wang X, Wang J, Xiao K, Li Y, et al. Overexpression of the https://​doi.​org/​10.​1093/​ajcn/​81.3.​624.
phospholipase A2 group V gene in glioma tumors is associated with poor 64. Li R, He H, He X. APOC1 promotes the progression of osteosarcoma by
patient prognosis. Cancer Manag Res. 2019;11:3139–52. https://​doi.​org/​10.​ binding to MTCH2. Exp Ther Med. 2023. https://​doi.​org/​10.​3892/​ETM.​2023.​
2147/​CMAR.​S1992​07. 11862.
44. Zhang Y, Wang R, Feng Y, Ma F. The role of sialyltransferases in gynecological 65. Ma C, Li R, Yu R, Guo J, Xu J, Yuan X, et al. Predictive value of preoperative
malignant tumors. Life Sci. 2020;263: 118670. https://​doi.​org/​10.​1016/j.​lfs.​ platelet-to-albumin ratio and apolipoprotein B-to-apolipoprotein A1 ratio
2020.​118670. for osteosarcoma in children and adolescents: a retrospective study of 118
45. Jones RB, Silva AD, Ankenbauer KE, Britain CM, Chakraborty A, Brown JA, cases. BMC Cancer. 2022. https://​doi.​org/​10.​1186/​S12885-​022-​09223-X.
et al. Role of the ST6GAL1 sialyltransferase in regulating ovarian cancer cell 66. Sasaki Y, Negishi H, Koyama R, Anbo N, Ohori K, Idogawa M, et al. p53 family
metabolism. Glycobiology. 2023;33:626–36. https://​doi.​org/​10.​1093/​glycob/​ members regulate the expression of the apolipoprotein D gene. J Biol
cwad0​51. Chem. 2009;284:872–83. https://​doi.​org/​10.​1074/​JBC.​M8071​85200.
46. Hugonnet M, Singh P, Haas Q, von Gunten S. The distinct roles of sialyltrans- 67. Daneshmandi S, Wegiel B, Seth P. Blockade of lactate dehydrogenase-A
ferases in cancer biology and onco-immunology. Front Immunol. 2021;12: (LDH-A) improves efficacy of anti-programmed cell death-1 (PD-1) therapy
799861. https://​doi.​org/​10.​3389/​fimmu.​2021.​799861. in melanoma. Cancers (Basel). 2019;11:450. https://​doi.​org/​10.​3390/​cance​
47. Liu D, Hu Z, Jiang J, Zhang J, Hu C, Huang J, et al. Five hypoxia and immunity rs110​40450.
related genes as potential biomarkers for the prognosis of osteosarcoma. 68. Patel CH, Leone RD, Horton MR, Powell JD. Targeting metabolism to regulate
Sci Rep. 2022;12:1617. https://​doi.​org/​10.​1038/​s41598-​022-​05103-3. immune responses in autoimmunity and cancer. Nat Rev Drug Discov.
48. Zheng X, Fan H, Liu Y, Wei Z, Li X, Wang A, et al. Hypoxia boosts aerobic gly- 2019;18:669–88. https://​doi.​org/​10.​1038/​s41573-​019-​0032-5.
colysis in carcinoma: a complex process for tumour development. Curr Mol 69. Wolf AJ, Reyes CN, Liang W, Becker C, Shimada K, Wheeler ML, et al.
Pharmacol. 2022;15:487–501. https://​doi.​org/​10.​2174/​18744​67214​66621​ Hexokinase is an innate immune receptor for the detection of bacterial pep-
08111​45752. tidoglycan. Cell. 2016;166:624–36. https://​doi.​org/​10.​1016/j.​cell.​2016.​05.​076.
49. Ward PS, Thompson CB. Metabolic reprogramming: a cancer hallmark even 70. Chen D, Zhang Y, Wang W, Chen H, Ling T, Yang R, et al. Identification and
Warburg did not anticipate. Cancer Cell. 2012;21:297. https://​doi.​org/​10.​ characterization of robust hepatocellular carcinoma prognostic subtypes
1016/J.​CCR.​2012.​02.​014. based on an integrative metabolite-protein interaction network. Adv Sci.
50. Nilsson R, Jain M, Madhusudhan N, Sheppard NG, Strittmatter L, Kampf C, 2021. https://​doi.​org/​10.​1002/​ADVS.​20210​0311.
et al. Metabolic enzyme expression highlights a key role for MTHFD2 and 71. Chen BJ, Wu YL, Tanaka Y, Zhang W. Small molecules targeting c-Myc onco-
the mitochondrial folate pathway in cancer. Nat Commun. 2014;5:3128. gene: promising anti-cancer therapeutics. Int J Biol Sci. 2014;10:1084–96.
https://​doi.​org/​10.​1038/​ncomm​s4128. https://​doi.​org/​10.​7150/​IJBS.​10190.
51. Reznik E, Sander C. Extensive decoupling of metabolic genes in cancer. PLoS 72. Dang CV. MYC on the path to cancer. Cell. 2012;149:22. https://​doi.​org/​10.​
Comput Biol. 2015. https://​doi.​org/​10.​1371/​journ​al.​pcbi.​10041​76. 1016/J.​CELL.​2012.​03.​003.
52. Hu J, Locasale JW, Bielas JH, O’Sullivan J, Sheahan K, Cantley LC, et al. 73. Wu C, Su J, Long W, Qin C, Wang X, Xiao K, et al. LINC00470 promotes
Heterogeneity of tumor-induced gene expression changes in the human tumour proliferation and invasion, and attenuates chemosensitivity
Wu et al. Journal of Biomedical Science (2024) 31:4 Page 26 of 26

through the LINC00470/miR-134/Myc/ABCC1 axis in glioma. J Cell Mol Med.

2020;24:12094–106. https://​doi.​org/​10.​1111/​jcmm.​15846.
74. Ding L, Congwei L, Bei Q, Tao Y, Ruiguo W, Heze Y, et al. mTOR: an attractive
therapeutic target for osteosarcoma? Oncotarget. 2016;7:50805–13. https://​
doi.​org/​10.​18632/​oncot​arget.​9305.
75. Lin YC, Haas A, Bufe A, Parbin S, Hennecke M, Voloshanenko O, et al.
Wnt10b-GSK3β–dependent Wnt/STOP signaling prevents aneuploidy in
human somatic cells. Life Sci Alliance. 2021. https://​doi.​org/​10.​26508/​LSA.​
20200​0855.
76. Casanova M, Ferrari A. Pharmacotherapy for pediatric soft-tissue sarcomas.
Expert Opin Pharmacother. 2011;12:517–31. https://​doi.​org/​10.​1517/​14656​
566.​2011.​524926.
77. Wu C, Gong S, Osterhoff G, Schopow N. A novel four-gene prognostic sig-
nature for prediction of survival in patients with soft tissue sarcoma. Cancers
(Basel). 2021. https://​doi.​org/​10.​3390/​cance​rs132​25837.
78. Martínez-Morales PL, Ortiz-Mateos CA, Reyes-Pineda J, Reyes-Vallejo T,
Aguilar-Lemarroy A, Jave-Suárez LF, et al. Identification and characterization
of the V3 promoter of the ST3GAL4 gene. Biosci Trends. 2020;14:144–50.
https://​doi.​org/​10.​5582/​bst.​2019.​01331.
79. Singh P, Joon A, Kumari M, Singh T, Bal A, Maan P, et al. Role of a disease-
associated ST3Gal-4 in non-small cell lung cancer. Cell Biochem Biophys.
2022;80:781–93. https://​doi.​org/​10.​1007/​S12013-​022-​01091-3.
80. Feng Z, Ou Y, Hao L. The roles of glycolysis in osteosarcoma. Front Pharma-
col. 2022;13: 950886. https://​doi.​org/​10.​3389/​fphar.​2022.​950886.
81. Colegio OR, Chu NQ, Szabo AL, Chu T, Rhebergen AM, Jairam V, et al. Func-
tional polarization of tumour-associated macrophages by tumour-derived
lactic acid. Nature. 2014;513:559–63. https://​doi.​org/​10.​1038/​natur​e13490.
82. Rodriguez E, Boelaars K, Brown K, Eveline Li RJ, Kruijssen L, Bruijns SCM, et al.
Sialic acids in pancreatic cancer cells drive tumour-associated macrophage
differentiation via the Siglec receptors Siglec-7 and Siglec-9. Nat Commun.
2021. https://​doi.​org/​10.​1038/​S41467-​021-​21550-4.

Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.