Genes linked to cell stress involved in the development of osteoporosis

Three datasets related to osteoporosis, specifically GSE56815 [8], GSE230665 [9], and GSE7429 [10], were extracted from the GEO database using the R package GEO query [11]. Further detailed information can be found in Supporting Information 1: Table S1. All samples were thoroughly evaluated in the analysis.

A study on ERSRGs [12] was carried out by utilizing the GeneCards database. The search term "Endoplasmic reticulum stress" was employed, and only ERSRGs categorized as "Protein Coding" were included.

The R package "sva" [13] was used to remove batch effects from the datasets GSE56815 and GSE230665 to obtain the integrated GEO dataset. Following this, the integrated combined dataset underwent standardization using the R package limma [14], with probes being annotated before standardization and normalization. Principal component analysis (PCA) [15] was conducted on the expression matrix to evaluate the impact of batch effect removal.

The R limma package was used to compare osteoporosis samples with control samples and the gene expression variance was calculated within the combined GEO dataset.

GSE7429 was then differentially analyzed using the same procedure and screening criteria as the GEO combined dataset. Co-DEGs were obtained by intersecting DEGs in the GEO combined dataset with DEGs in GSE7429.

The intersection of Co-DEGs and ERSRGs was used to create a Venn diagram to identify ERSRDEGs.

The GSEA method [16] is employed to evaluate the pattern of gene distribution in a defined set of genes in a gene table organized by phenotypic correlation. In this study, genes in the integrated GEO dataset (pooled dataset) were initially sorted by logFC value, followed by GSEA using the R package clusterProfiler [17]. The significance criterion for GSEA settings consisted of the seed (2022), the number of iterations (1000), the minimum gene count per group (10), and the maximum gene count (500). The gene collection c2.cp.all.v2022.1.Hs.symbols.gmt (All Canonical Pathways; 3050) was obtained from the Molecular Signatures Database (MSigDB) [18] for GSEA analysis. GSEA set at p < 0.05.

GSVA [19] was employed to quantitatively evaluate gene set enrichment results in the context of a microarray-based transcriptome. In examining datasets related to osteoporosis, GSVA was conducted on all genes using the gene set c2.cp.all.v2022.1.Hs.symbols.gmt from the MSigDB. GSVA set at p < 0.05.

The analysis of GO [20] is frequently used in broad functional enrichment investigations, encompassing the study of biological process (BP), cellular component (CC), and molecular function (MF). The KEGG [21] is a widely accessed database containing information on genetic material, biological pathways, diseases, and drugs. Using the R package clusterProfiler, ERSRDEGs were subjected to enrichment analysis based on both GO and KEGG, with a significance threshold of p < 0.05 for inclusion. To visualize the results of the KEGG enrichment analysis, the R package Pathview [22] was employed.

Using the combined GEO dataset, a model was developed for the diagnosis of osteoporosis. Logistic regression was performed on the ERSRDEGs to investigate the correlation between independent variables and a binary dependent variable, distinguishing osteoporosis samples from control samples. By setting a significance level of p < 0.1, the ERSRDEGs were identified and incorporated into the logistic regression model.

Afterward, we employed the Mann–Whitney U test to assess the ERSRG expression levels across the combined dataset and the GSE7429 dataset. Differential expression of genes related to ERS was compared based on expression differences and visually represented in group comparison plots. Subsequently, only the differentially expressed ERSRGs that showed significant differences between the osteoporosis and normal groups in both datasets were retained for further analysis.

Next, the support vector machine (SVM) algorithm [23] was utilized to establish an SVM model for the ERSRDEGs identified in the group comparison plot.

At the same time, we utilized the R package glmnet [24] to perform least absolute shrinkage and selection operator (LASSO) analysis on the ERSRDEGs maintained in the group comparison chart. LASSO regression is rooted in linear regression, incorporating a penalty term (lambda × absolutevalue of the coefficient) to mitigate overfitting and enhance model generalization. The outcomes of the LASSO regression analysis can be displayed visually using diagnostic model plots and variable trajectory graphs.

The identified ERSRDEGs in both the LASSO model and SVM were compared using a Venn diagram to establish the model genes. Next, the coefficients of these genes from the LASSO model and the dataset expression were merged to create the diagnostic model for osteoporosis, outlined as follows.

Using logistic regression analysis outcomes in the R package rms, a nomogram [25] is generated as a graphical representation in a rectangular coordinate system. The R package ggDCA is capable of producing a DCA graph for the model genes identified in the combined dataset. This graph provides valuable information regarding the clinical significance and overall advantages of the predictive model across different probability thresholds [26].

Following this, subsequently, the R package pROC was used to generate receiver operating characteristic (ROC) curves and calculate area under the curve (AUC) values, evaluating the diagnostic performance of the LASSO risk score in osteoporosis.

Using the formula for computing the risk score in the osteoporosis diagnosticmodel, the dataset GSE7429 underwent computation of the risk score. Utilizing the R package (pROC), an ROC curve was generated to calculate the AUC for the dataset. The samples from the combined dataset were then divided into high-risk and low-risk groups using the median risk score from the diagnostic model for osteoporosis.

The evaluation of gene sets for individual samples (ssGSEA) [27] evaluates the proportion of infiltrating immune cells. Initially, the different types of infiltrating immune cells were identified, the enrichment score derived from ssGSEA was utilized to indicate the proportion of immune cell infiltration in each individual sample. In order to compare the level of immune cell infiltration between samples from high-risk and low-risk groups in the merged dataset, group comparison diagrams were utilized, with the Spearman statistical method employed for computation.

CIBERSORT [28] utilizes linear support vector regression to analyze the transcriptome expression matrix, assessing the composition and presence of immune cells within mixed cell populations. Subsequent analysis involves using the ggplot2 R package to generate comparative charts that highlight differences in LM22 immune cell expressions between osteoporosis and control groups in the combined dataset.

Data sourced from the ChIPBase database [29] made the evaluation of how these transcription factors (TFs) affect their target genes possible. The interactions were then depicted in the mRNA–TF regulatory network, which was mapped using Cytoscape software.

The Comparative Toxicogenomics Database (CTD) [30] was employed to identify both direct and indirect drug targets associated with specific genes, exploring how these genes interact with various drugs.

The relationships of these miRNAs with specific genes were explored using the ENCORI database. Subsequently, the interactions within the mRNA–miRNA regulatory network were depicted through Cytoscape software.

The identification of specific RNA-binding proteins (RBPs) associated with target genes was facilitated using the the StarBase v3.0 database [31] and the resulting mRNA–RBP regulatory network was mapped using Cytoscape software.

The AlphaFold Database [32] (https://www.alphafold.ebi.ac.uk/↗) contains around 350,000 protein structures predicted by the AlphaFold AI system. Researchers utilized the AlphaFold platform to analyze the protein structures of specific genes and showcase the results.

Analyses were conducted using version 4.1.2 of the R software. Comparisons between groups utilized the Student's t-test and the Mann–Whitney U test. Additionally, correlation coefficients were calculated through Spearman's method. A p-value of less than 0.05 was considered to indicate statistical significance.

Osteoporosis Mouse Model: Female C57BL/6 mice (8-weeks old) were randomly divided into control (CON, sham surgery) and osteoporosis (OVX, bilateral ovariectomy) groups (n = 5 per group). Surgery was performed under inhaled anesthesia (3% isoflurane induction and 1.5% maintenance). Postoperative care included single housing, standard diet, and controlled environmental conditions (22 ± 2°C, 12-h light/dark cycle). After 2 months, micro-CT confirmed successful osteoporosis induction. Bone tissues were harvested, with left limbs used for histology and right limbs for mRNA extraction. All procedures were approved by the Institutional Animal Ethics Committee (Approval No. IACUC-FPH-SL-20250210 [0638]).

Immunohistochemistry: Paraffin-embedded bone sections were stained with antibodies against C1QBP, CES2, CYB5R4, RAB1B, SERP1, and UFSP2. Staining intensity was quantified using image analysis software under uniform conditions.

RNA Extraction and qRT-PCR: Total RNA was extracted using TRIzol and reverse-transcribed with PrimeScript RT Reagent Kit (TaKaRa). qRT-PCR was performed using SYBR Premix Ex Taq II (Takara) with primers for target genes and 18S rRNA (internal control). Data were analyzed via ΔΔCt method.

The technology roadmap is detailed in the Supporting Information: Figure. 2 S1

The analysis commenced with the normalization of the GSE7429 data using the limma R package. The effectiveness of this preprocessing was assessed by comparing the datasets before and after normalization through box plots (Supporting Information: FigureA,B). The results from the box plots indicated that the expression levels in GSE7429 were consistently uniform after processing, demonstrating successful normalization and improved comparability of the samples. 3 S2

Batch effects in the osteoporosis datasets GSE56815 and GSE230665 were addressed using the sva R package. Subsequently, normalization of the aggregated data was conducted with limma. The impact of these adjustments was evaluated through both box plots and PCA (Supporting Information: FigureC–F). The consistent expression levels across the combined dataset after processing, as indicated by the box plots and PCA results, suggest that the batch effects were effectively mitigated. 3 S2

To compare gene expression between osteoporosis and control groups within the integrated dataset, a differential analysis was performed with limma. This analysis revealed 1161 DEGs that met the threshold of |log fold change (log FC)| > 0 and p < 0.05. Of these, 614 genes showed increased expression while 547 exhibited decreased levels. The dataset's analysis outcomes were illustrated in a volcano plot (Figure 1A).

To evaluate gene expression differences in osteoporosis and control samples within the GSE7429 dataset, we employed limma for differential analysis. This identified 960 genes that surpassed the thresholds of |log FC| > 0 and p < 0.05. Of these, 461 genes showed increased expression and 499 displayed decreased levels. The outcomes, including a volcano plot (Figure 1B), effectively illustrate the differential analysis results from the GSE7429 dataset.

To pinpoint genes that significantly vary Co-DEGs between osteoporosis and control groups within both the GSE7429 and the merged datasets, 381 Co-DEGs were identified. This was achieved by finding the intersection of DEGs from both the GSE7429 and the combined datasets (Figure 1C).

In order to obtain the ERSRDEGs, the intersection of all Co-DEGs and the ERSRGs was taken and a Venn diagram was drawn (Figure 1D). A total of 56 ERSRDEGs were obtained, namely, MAPK3, STK11, CYB5R4, PIGK, ROR2, CHI3L1, ACSL3, SLC22A13, SERP1, ABT1, PPP2CA, RRAGC, TNFRSF10B, UFSP2, TUBA1B, RPS6KA3, C1QBP, RAB1B, CACYBP, MRPL2, SUCLG2, EHD4, RNF13, CCDC6, G6PC3, ASGR2, CES2, COPS8, SRC, PPP1R13B, H6PD, IFNG, CCT2, GNAI3, NUPR1, GALNT2, CHRNE, EIF2B4, NPHS1, PTP4A1, SCARB1, USP9X, GNE, UBE2I, STAU2, RPL15, TRAF2, GABARAPL1, TNFRSF12A, FDFT1, CYP1A2, NGF, PLD3, STARD5, NR3C1, and MEF2A.

All gene expression levels in the combined GEO dataset on osteoporosis, GSEA was used to study the disease control group in the combined GEO dataset. There is a relationship between the expression of all genes and the BP they participate in, the CC they influence, and the MF they exert (Figure 2A). The specific results are shown in Supporting Information 1: Table S3. Genes in the combined dataset were significantly enriched in pathways including Type I collagen synthesis (Figure 2B), mitochondrial gene expression (Figure 2C), RHOBTB1 Gtpase cycle (Figure 2D), Il-17 pathway (Figure 2E), IL-5 pathway (Figure 2F), IL-2 pathway (Figure 2G), and other biologically relevant functions and signaling pathways.

GSVA analysis was conducted on all the genes in the combined dataset comparing disease control groups. Specific information can be found in Supporting Information 1: Table S4. Following this, pathways that showed statistical significance (p < 0.05) and had top 10 positively enriched and top 10 negatively enriched rankings of log FC were identified and visualized in group comparison diagrams (Figure 3A). The GSVA results indicated enrichment in various pathways. These findings were statistically significant (p < 0.05) in samples from individuals with osteoporosis and control samples. Last, based on the GSVA results, the expression differences in 20 pathways between the osteoporosis samples and the control sample groups were analyzed and presented visually using heat maps (Figure 3B).

Through GO and KEGG enrichment analysis, the BP, CC, and MF of 56 ERSRDEGs were further explored and the relationship between biological pathways and osteoporosis. The specific results are shown in Supporting Information 1: Table S5. The results show that 56 ERSRDEGs are mainly enriched in positive regulation of protein kinase activity, positive of kinase activity, activation of protein kinase activity, and extrinsic apoptotic signaling pathway, intrinsic apoptotic signaling pathway and other BP; membrane raft, membrane microdomain, microtubule, cytoplasmic side of plasma membrane, cytoplasmic side of membrane and other CC; GTP binding, guanyl nucleotide binding, guanyl ribonucleotide binding, ubiquitin protein ligase binding, GABA receptor binding, and other MF. At the same time, it is also enriched in apoptosis, autophagy-animal, hepatitis C, pathogenic Escherichia coli infection, chemical carcinogenesis-receptor activation, and other KEGG. The results of GO and KEGG enrichment analysis were visualized through histograms (Figure 4A).

A circular diagram representing the network of BP, CC, MF, and biological pathways was created based on the results of GO and KEGG enrichment analysis (Figure 4). The connecting line displays the associated molecule along with the annotation of its respective entry. The size of each node within the diagram corresponds to the number of molecules contained in the entry.

Finally, pathway KEGG enrichment analysis of Chemical carcinogenesis-receptor activation (Supporting Information 4: Figure S3A), pathogenic Escherichia coli infection (Supporting Information 4: Figure S3B), hepatitis C (Supporting Information 4: Figure S3C), autophagy-animal (Supporting Information 4: Figure S3D), and apoptosis (Supporting Information 4: Figure S3E) was performedvia the R package Pathview Perform a visual display of the pathway diagram.

In order to determine the diagnostic value of 56 ERSRDEGs in osteoporosis, a logical regression was conducted, constructing a logistic regression model and displaying it visually through forest plot (Figure 5A). The results showed that 13 ERSRDEGs were statistically significant (p < 0.1) in the logistic regression model, namely, C1QBP, CES2, CYB5R4, G6PC3, GALNT2, PPP1R13B, PPP2CA, RAB1B, RNF13, SERP1, SUCLG2, TNFRSF10B, and UFSP2.

Thirteen distinct ERSRDEGs were identified and comparative group diagrams were generated for visualization purposes (Figure 5B,C). Utilizing the group comparison data, only those ERSRDEGs displaying noteworthy variances between the osteoporosis and normal groups across both sets of data were included. Ultimately, 10 ERSRDEGs met this criterion: CYB5R4, RAB1B, UFSP2, GALNT2, RNF13, SERP1, CES2, G6PC3, PPP1R13B, and C1QBP.

A SVM model was created using 10 ERSRDEGs and the SVM algorithm. The model determined the genes with the lowest error rate (Figure 5D) and the highest accuracy rate (Figure 5E). The findings indicate that with 10 genes, the SVM model achieves peak accuracy.

A LASSO regression model was constructed based on 10 ERSRDEGs through LASSO regression analysis. The LASSO regression model diagram (Figure 5F) and the LASSO variable trajectory diagram (Figure 5G) were plotted for visual representation. The results showed that the LASSO regression model included seven ERSRDEGs: CYB5R4, RAB1B, UFSP2, RNF13, SERP1, CES2, and C1QBP.

The genes selected for the LASSO regression model overlapped with those chosen for the SVM, resulting in the identification of model genes through the creation of a Venn diagram. The coefficients of the model genes in the LASSO regression model were then combined with the expression data from the combined dataset to obtain the osteoporosis diagnostic model and the corresponding risk score. The risk score was calculated as follows: Risk score CYB5R4 SERP1 C1QBP CES2 RAB1B RNF13 UFSP2 = − + × + × + × + × − + × + × + × 92.92863372 2.289953309 1.035150235 0.296414099 1.177135117 4.94123259 1.499424138 1.77301333.

In conclusion, we utilized the R software package GOSemSim to assess the semantic similarities among the GO terms, collections of GO terms, gene products, and gene sets of the seven model genes. Subsequently, a boxplot was generated to visualize the results. The results indicated that SERP1 exhibited the highest functional similarity score among the model genes.

To validate the value of the osteoporosis diagnostic model, a nomogram was created using the model genes to illustrate their relative importance in the combined dataset (Supporting Information: FigureA). The results indicate that the expression level of the CYB5R4 gene has a notably higher influence on the osteoporosis diagnostic model compared to other factors, while the expression level of the CES2 gene has a lower impact on the osteoporosis diagnostic model relative to other factors. 5 S4

The clinical utility of the osteoporosis diagnostic model was assessed using DCA on the combined dataset (Supporting Information: FigureB). The results indicate that the model consistently outperforms both positive and negative controls within a specific range, resulting in higher net benefit and improved clinical utility. Furthermore, the ROC curve generated using the R package pROC for the combined dataset (Supporting Information: FigureC) demonstrates high accuracy of the risk score (AUC > 0.9) across various groups. 5 S4 5 S4

Based on the expression levels of the seven model genes in the GSE7429 dataset, a calculation formula was applied to determine the risk score of the osteoporosis diagnostic model. The risk score was then used to evaluate the accuracy of the model in predicting osteoporosis within the GSE7429 dataset. By utilizing the R package pROC, a ROC curve was generated based on the risk score, demonstrating high accuracy (AUC > 0.9) in distinguishing between different groups within the GSE7429 dataset (Supporting Information: FigureD). This indicates that the osteoporosis diagnostic model is effective in accurately identifying individuals at risk for osteoporosis within this specific dataset. 5 S4

Initially, only samples with osteoporosis were selected, and the median risk score expression from the osteoporosis diagnostic model based on the combined dataset was used to divide the osteoporosis samples into two groups: high-risk (high) and low-risk (low).

We performed a differential gene expression analysis on all genes in the combined dataset using the limma package. The comparison was carried out between the high-risk (high) and low-risk (low) groups. The results obtained from this analysis were then used in GSEA to investigate the association between gene expression in the high and low-risk groups and their significance in BP, CC, and MF. By examining these associations, we can better understand the mechanisms underlying the observed differences between the high and low-risk groups. The specific results are presented in Supporting Information: Table. The results show that genes within the combined dataset are significantly enriched in biologically relevant functions and signaling pathways, including the IL-2–PI3K pathway (Supporting Information: FigureC), photodynamic therapy-induced unfolded protein response (UPR; Supporting Information: FigureD), IL-18 signaling pathway (Supporting Information: FigureE), UPR (Supporting Information: FigureF), intra-golgi and retrograde golgi-to-ER traffic (Supporting Information: FigureG), golgi-to-ER retrograde transport (Supporting Information: FigureH), and others. 1 S6 6 S5 6 S5 6 S5 6 S5 6 S5 6 S5

The combined dataset was analyzed using GSVA to compare gene expression levels between high and low risk groups. See Supporting Information 1: Table S7 for specific information. Subsequently, the top 10 positively and negatively correlated enriched pathways were identified using a significance level of p < 0.05 and log FC ranking. The findings were then visualized using a comparative chart (Supporting Information 7: Figure S6A). The GSVA results revealed significant enrichment (p < 0.05) of the following pathways in both high-risk and low-risk sample groups: Samols targets of KHSV miRNAs up, Tomida lung cancer poor survival, Wang esophagus cancer progression up, Nikolsky breast cancer 1q32 amplicon, Yamanaka glioblastoma survival up, Bierie inflammatory response TGFB1, attachment of GPI anchor to UPAR, CREB3 factors activate genes, Jones TCOF1 targets, Turjanski MAPK11 targets, Serotonin and melatonin biosynthesis, Tesar ALK and JAK targets mouse ES d4 dn, Fructose catabolism, Biocarta PELP1 pathway, Schlesinger methylated in colon cancer, Eicosanoid metabolism via cytochrome P450 monooxygenases pathway, Lopez mesothelioma survival worst vs. best dn, Tonks targets of RUNX1 RUNX1T1 fusion sustained in granulocyte dn, amino acid conjugation, and conjugation of benzoate with glycine. Following the GSVA analysis, a differential expression analysis of the 20 pathways was conducted between osteoporosis and control samples. The results were then visualized using heat maps (Supporting Information 7: Figure S6B).

Initially, only osteoporosis samples were selected, and the median risk score expression from the osteoporosis diagnostic model in the combined dataset was used to classify the osteoporosis samples into high-risk (high) and low-risk (low) categories.

The abundance of immune cell infiltration for 28 different types of immune cells was calculated using the combined GEO dataset and the ssGSEA algorithm. First, the expression matrix was used to identify variations in the abundance of immune cell infiltration across different groups. The group comparison charts show significant differences in the infiltration levels of eight types of immune cells in the bone, including activated B cells, activated CD8⁺ T cells, central memory CD8⁺ T cells, effector memory CD8⁺ T cells, macrophages, monocytes, NK cells, and T follicular helper cells. The group comparison chart (Figure 6A) demonstrates a statistically significant difference between the control and experimental groups (p < 0.05).

The infiltration abundance of eight immune cell types in samples from both high-risk and low-risk groups was analyzed using correlation heat maps (Figure 6B,C). The results revealed predominantly positive correlations among the immune cells. These findings suggest that changes in the abundance of one immune cell type are associated with changes in the abundance of other immune cell types. This information may be valuable for understanding the interactions and dynamics of immune cell populations in different risk groups.

Subsequently, an analysis was performed to investigate the relationship between seven model genes and eight immune cells in both the high-risk and low-risk group samples within the combined dataset. The findings were then visualized using a correlation bubble chart (Figure 6D,E). The results show that among the samples from the low-risk group, the ERSRDEG UFSP2 exhibited the strongest positive correlation with activated B cells (r = −0.818181818). Among the samples from the high-risk group, the strongest correlation was observed between the ERSRDEG CYB5R4 and effector memory CD8⁺ T cells (r = −0.863636364).

The abundance of 22 distinct immune cell types was estimated using the CIBERSORT algorithm on the combined dataset. After analyzing the immune cell infiltration, a bar graph was generated to visualize the distribution of immune cells in the combined dataset (Figure 7A). Subsequently, a diagram comparing the differences in immune cell infiltration abundance between high-risk and low-risk groups in the combined dataset was presented (Figure 7B). The results revealed significant differences in the abundance of three distinct immune cell types between high-risk and low-risk group samples in the combined GEO dataset (p < 0.05). The immune cell types with significant differences were M0 macrophages, activated NK cells, and resting memory CD4⁺ T cells.

Bubble charts were used to visualize the correlation between the abundance of immune cell infiltration and model genes in the high-risk and low-risk groups, utilizing data from the combined GEO dataset (Figure 7C,D). The findings revealed that among the samples from the low-risk group, the ERSRDEG UFSP2 showed the strongest positive correlation with activated NK cells (r = 0.642283206). In the high-risk group samples, the ERSRDEG CYB5R4 exhibited the strongest correlation with resting NK cells (r = 0.770674636).

TFs influence gene regulation at the post-transcriptional level by binding to specific genes. Model genes and their associated TFs were obtained from the ChIPBase database. Using Cytoscape software (Figure 8A), a visual representation of the mRNA–TF regulatory network was constructed. The resulting network consists of 52 TFs and seven model genes (Supporting Information 1: Table S8).

The miRNAs significantly impact the regulation of gene expression, playing a key role across biological development and evolutionary processes. It is common for several miRNAs to control a single gene. The miRNAs associated with the model genes were obtained from the TarBase database, and an mRNA–miRNA interaction network was constructed and visualized using the Cytoscape software (Figure 8C). The network consists of five model genes and 42 miRNAs. Detailed information is provided in Supporting Information 1: Table S10.

The CTD was used to identify potential drugs or molecular compounds associated with the model genes. The mRNA–drug regulatory network was first constructed and then visualized using Cytoscape software (Figure 8B). The network includes five model genes and 27 drugs or molecular compounds. Detailed information can be found in Supporting Information 1: Table S9.

RBPs [33] were pivotal in gene regulation by influencing essential BPs such as RNA synthesis, editing, modification, transport, and decoding. The StarBase database was used to predict RBPs associated with the model genes. The mRNA–RBP regulatory network was then constructed and visualized using the Cytoscape software (Figure 8D). The network consists of seven model genes and 32 RBPs. Detailed information is provided in Supporting Information 1: Table S11.

Finally, the protein structures of the seven model genes were analyzed using the AlphaFold website, and the results were displayed (Supporting Information: FigureA–G). 8 S7

Immunohistochemistry revealed significant upregulation of C1QBP, CYB5R4, RAB1B, SERP1, and UFSP2 proteins in OVX mice compared to controls(p < 0.01), while CES2 showed no significant change (Figure 9A,B). qRT-PCR demonstrated concordant mRNA upregulation for C1QBP, CYB5R4, RAB1B, and UFSP2 (p < 0.01). Notably, CES2 mRNA was elevated in OVX mice (p < 0.001), contrasting with unchanged protein levels, whereas SERP1 mRNA remained unaltered despite increased protein expression (Figure 9C).