Tissue samples were obtained from 19 patients who attended the Neurosurgery Department of Beijing Tiantan Hospital for resection of pituitary adenoma. The protocol of this study was approved by the Ethics Committee of Beijing Tiantan Hospital, and all patients signed the informed consent form. Postoperative tissue samples were examined at the Department of Pathology, Beijing Tiantan Hospital, and all histological assessments confirmed pituitary adenoma. Invasive and non-invasive adenomas were identified using imaging and pathology reports. Ten invasive samples were silent corticotroph adenomas, and 9 non-invasive samples were follicle-stimulating hormone (FSH)-secreting adenomas. All postoperative samples were immediately frozen in liquid nitrogen and stored at -80° C in the refrigerator.

A total of 19 tissue samples from patients with early-stage pituitary adenomas were lysed with UA buffer (8 M urea in 0.1 M Tris-HCl, pH 8.5), and sonicated on ice (180 W, 1 second on and 2 seconds off, for 99 cycles). The lysate was centrifuged at 14,000 × g for 15 min. The supernatant was collected and quantified using the Bradford method. The protein was lysed by filter-aided sample preparation (FASP) following the protocol described in references (9, 10).

The peptides used for DDA analysis were pre-fractionated by high-pH reversed-phase chromatography (Hp-RP). Two hundred micrograms of peptides were loaded and separated with a 45-min gradient, pH 10.45. Fractions were collected, and 12 fractions were combined, heat-dried, and stored at -80°C after centrifugation. The LC-MS/MS detection system consisted of a nanoflow high-performance liquid chromatography (HPLC) instrument (Easy-nLC 1000 System; Thermo Fisher, Waltham, MA, USA) coupled to a Q-Exactive HF mass spectrometer (Thermo Fisher). A home-packing column (150-μm inner diameter, ReproSil-Pur C18-AQ, 1.9 μm; Dr. Maisch) with a length of 20 cm was used for peptide separation at 60°C. The flow rate was 600 nL/min over a 90-min gradient (0–8 min, 3–8% B; 8–68 min, 8–20% B; 68–83 min, 20–30% B; 83–84min, 30–90% B; 84–90 min, 90% B). The full MS scan range was 400–1200 Da. MS/MS was operated in the top 20 modes. iRT (Biognosys, Schlieren, Switzerland) was added as an internal standard based on the manufacturer’s instructions. Peptides from 293 cell lysis were used in all experiments. The LC-MS/MS system for the DIA analysis consisted of a nanoflow HPLC (Easy-nLC 1000 System; Thermo Fisher) coupled to a Q-Exactive HF mass spectrometer (Thermo Fisher). The flow rate was 600 nL/min over different gradients. The range of the MS1 survey scan was 400–1200 m/z, followed by 29 MS2 scans of overlapping sequential precursor isolation windows (20 m/z isolation width, 40 m/z isolation width, 50 m/z isolation width, and 1 m/z overlap). The accumulation time was set according to the experimental requirements. The chromatograph peak width was 18 s. iRT was added to ensure calibration on difficult matrices, allowing for detailed quality control. All data were obtained on a 20-cm analytical column. DDA raw data files were searched against the human UniProt database (20190617 with sequences of iRT peptides) using MaxQuant (version 1.6.5.0) with its default settings. The false discovery rate (FDR) was set to 0.01 for both peptides and proteins. The DDA data search results were imported into Spectronaut (version 14) with its default settings. The FDR threshold was set to less than 0.01. The DIA raw data files were directly imported into Spectronaut with its default settings, and the results for protein identification and quantitation were exported using protein and peptide FDRs of less than 0.01.

Quantitative difference analysis was performed using the quantitative protein values. For a two-by-two comparison between invasive group B and non-invasive group A, the mean value of the signals in each group was calculated, from which the ratio between groups was calculated. The comparison was done using the student’s t-test. Proteins that met the following conditions were screened as differentially expressed proteins (DEPs) (1): Ratio between groups >=2 or <=0.5; (2) P-value <0.05. The Volcano plot was drawn using DEPs to show the difference between the groups. In the volcano plot, the vertical axis shows -log10(p-value) and the horizontal axis shows log2(Fold change). Upregulated DEPs with significant fold change and p-value are shown by red dots. Downregulated DEPs are shown by green dots, and other proteins are shown by black dots. Unsupervised hierarchical cluster analysis was performed for DEPs to show the expression of DEPs in each sample in the two groups. Finally, we performed GO (Biological Process, Molecular Function, Cellular Component) enrichment analysis and KEGG pathway enrichment analysis on the DEPs. We performed bioinformatic annotation and analysis at two levels to analyze the functional properties of these proteins and the corresponding signaling pathways to select key proteins for further research.

We screened EMT-related proteins using the GeneCards database (http://www.genecards.org/), which integrates genomic, transcriptomic, proteomic, clinical, and other relevant information related to genes. We collected and collated data from over 100 sites. The screening criteria were correlation score ≥ 3. Then, the Venn program (http://bioinformatics.psb.ugent.be/webtools/Venn/) was used to identify the intersection between EMT-related proteins and DEPs, as EMT-DEPs. To explore the interactions between EMT-DEPs, we entered EMT-DEPs into the STRING database (https://string-db.org/) to obtain the protein interaction network. To find the most important pathways, we performed GO functional annotation and KEGG pathway enrichment analysis of EMT-DEPs using R software to obtain their potential protein functions and key pathways.

To identify hub proteins in EMT-DEPs. We imported the information on EMT-DEPs interaction into Cytoscape software and screened hub proteins using the cytoHubba plugin. The cytoHubba plugin can predict and explore the key nodes and sub-networks in a given network by several topological algorithms. We calculated the rankings using the MCC algorithm and selected the top 20 EMT-DEPs as the hub EMT-DEPs. The protein interaction network of hub EMT-DEPs was visualized by Cytoscape software to show their potential interactions.

To assess the correlation between EMT-DEPs, we performed a correlation analysis between EMT-DEPs to determine the paired proteins of EMT-DEPs (Cor>0.7, P<0.05). The EMT-related DEPs were ranked by the number of paired proteins, and the top 20 were identified as core EMT-related DEPs. We then used the GeneMANIA database to map the protein network of core EMT-related DEPs and their associated proteins. The GeneMANIA database (http://genemania.org/) was used to generate hypotheses about gene function and to find genomic and proteomic data regarding the function of target proteins.

We identified the intersection of hub EMT-related DEPs and core EMT-related DEPs through the Venn online program as the key EMT-related DEPs. The R software was used to analyze the correlation between the quantitative and clinical features of key EMT-related DEPs. The ROC curve was used to evaluate the sensitivity and accuracy of key EMT-related DEPs in predicting the invasion of pituitary adenomas. We also used the GSE169498 gene expression profile to measure clinical correlation. ROC curve analysis was used to assess the efficacy of key EMT-related DEPs expression in predicting the invasive behavior of pituitary adenoma. Wilcoxon signed-rank test was used to evaluate the relationship between the groups. P<0.05 was considered statistically significant.

We identified the target proteins using proteomics analysis and GSE169498 chip. Then, the paired proteins of the target protein were screened (Cor>0.7, P<0.05), and the correlation between them was measured by R software. We input the target protein and paired proteins into the GeneMANIA database to explore the potential functions of the target proteins and related key pathways. GO functional annotation and KEGG pathway enrichment of target and paired proteins were performed to identify the role of the target proteins in the invasion and EMT of pituitary adenomas and their related key pathways. To verify the relevance of the target proteins to EMT, we verified the co-expression of target proteins and classical EMT markers (CDH1, CDH2, DSP, FN1, ITGB6, MMP9, TJP1, and VIM) by quantitative proteomic correlation analysis. We found which markers of EMT are related to target proteins.

This study was approved by the Institution Review Board of the Beijing Tiantan Hospital. Between December 2020 and November 2021, specimens were obtained from patients who underwent surgery at Beijing Tiantan Hospital, Beijing, China. Informed consent was obtained from all individuals. All patients provided written informed consent. Each sample was allocated to the invasive or non-invasive group based on the Knosp grade. We collected the clinical information of patients. All samples were fixed in 10% formalin for 24 h, paraffin-embedded, and then cut into 5 μm thick sections. Glass slide was inserted, and the samples were dewaxed and rehydrated. After antigen repair and endogenous peroxidase blocking, goat serum was used to seal. Then, diluted antibodies were added and incubated at 4°C overnight. After washing with PBS, enzyme-labeled IgG polymer was added and incubated at room temperature for 20 min. Finally, diaminobenzidine (DAB) color development solution and hematoxylin were used as the double stain to visualize the antibodies. The negative control was stained with PBS buffer instead of antibodies, and the known positive tissue (breast) was used as positive control. Images were taken using a slide scanner (Leica, Germany). Two experienced neuropathologists independently assessed the samples. The percentage of positive cells was calculated under high magnification (×400).

In total, 5598 proteins were identified in the quantitative analysis of proteomics data, and 833 DEPs were obtained after differential analysis. Among them, 638 were upregulated and 195 were downregulated in invasive pituitary adenomas ( Figure 2A ). Cluster analysis of DEPs revealed that there were significantly more upregulated proteins in the invasive group and more downregulated proteins in the non-invasive group ( Figure 2B ). The GO function and KEGG pathway enrichment analysis of DEPs showed that: the BP functions enriched with DEPs included cellular localization, establishment of localization in cell, and CC functions included organelle membrane, vesicle, MF functions included nucleoside phosphate binding, nucleotide binding, and small-molecule binding ( Figure 2C ). Metabolic pathways, citrate cycle (TCA cycle), and ECM−receptor interaction were the major enrichment pathways of DEPs ( Figure 2D ). GSEA enrichment analysis showed that the main pathways affected by DEPs were citrate cycle (TCA cycle), Parkinson’s disease, and carbon metabolism ( Figure 2E ).

There were 46 intersections between EMT-related proteins and DEPs in the GeneCards database ( Figure 3A ). The STRING database visualized potential protein interactions between EMT-DEPs ( Figure 3B ). Potential GO functions of EMT-DEPs at BP level included: cell junction organization, extracellular structure organization, cell junction assembly, and extracellular matrix organization. Potential GO functions of EMT-DEPs at CC level included: cell−cell junction, membrane raft, membrane microdomain, and membrane region. Potential GO functions of EMT-DEPs at MF level included: cell adhesion molecule binding, protease binding, cadherin binding, and scaffold protein binding ( Figure 3C ). KEGG pathway enrichment analysis showed that EMT-related DEPs were mainly enriched in human papillomavirus infection, PI3K/Akt signaling pathway, hepatitis C infection, human T−cell leukemia virus 1 infection, chemical carcinogenesis−reactive oxygen species, microRNAs in cancer, and insulin signaling pathway ( Figure 3D ).

The hub EMT-related DEPs were identified in Cytoscape using the cytoHubba plugin. The MCC scores were ranked. The top 20 positions were CDH1, MMP9, STAT3, TIMP1, S100A4, SOD2, HMOX1, ACTA2, ITGB3, IKBKB, DSP, CLDN4, KRT5, CLDN3, SLC2A1, COL4A1, MAP2K1, ITGA3, LAMC2, and IDH1 ( Table 1 ). Their potential protein interactions are shown in Figure 4A . Based on the pairwise correlation analysis of 46 EMT-related DEPs by R software, the top 20 core EMT-DEPs with the highest number of correlated paired proteins, were FLOT1, FLOT2, CD36, PPP2CB, PTPN12, CA2, COL4A1, NRP1, MCM5, CD82, SLC2A1, CLDN4 CXADR, FAM3C, ITGB3, PRKCI, JAK1, LAMC2, S100A4, and CDC16, respectively ( Table 2 ). The GeneMANIA database exhibits their interactions and related genes ( Figure 4B ).

Six intersecting key EMT-related DEPs, including SLC2A1, CLDN4, LAMC2, S100A4, ITGB3, and COL4A1, were identified from the hub EMT-related DEPs and core EMT-related DEPs using Venn online program ( Figure 4C ). A significant positive correlation was found between all of them according to the quantitative protein co-expression correlation ( Figure 4D , cor>0.3). SLC2A1, CLDN4, LAMC2, S100A4, and ITGB3 were significantly correlated with the invasiveness of pituitary adenoma (P<0.05, Figures 5A–E ). COL4A1 was not significantly correlated with the invasiveness of pituitary adenoma (P>0.05, Figure 5F ). The areas under the ROC curves of the six intersection proteins for predicting invasion were greater than 0.7 ( Figures 5G–L ). SLC2A1 was significantly associated with the invasiveness of pituitary adenoma in the GSE169498 microarray. Consistent with the results of protein quantification (P<0.05, Figure 6A ), SLC2A1 was significantly upregulated in the invasive group. Contrary to the results of proteomic quantification, COL4A1 was downregulated in invasive pituitary adenomas (P<0.05, Figure 6F ). CLDN4, LAMC2, S100A4, and ITGB3 were not significantly associated with the invasiveness of pituitary adenoma (P>0.05, Figures 6B–E ). The area under the ROC curve for predicting invasion by SLC2A1 was 0.697 ( Figure 6G ), and the AUC values for the remaining genes were <0.65 ( Figures 6H–L ).

As SLC2A1 was similarly correlated with invasion in proteomics and GSE169498 microarrays, it may be a potential marker for invasion in pituitary adenoma. We demonstrated the co-expression correlation of SLC2A1 with its paired proteins (CA2, MCM5, NRP1, CD82, COL4A1, FLOT1, FLOT2, PTPN12, and CD36) ( Figure 7A ). Because of their high correlation, we mapped the protein network of SLC2A1 with its paired proteins. SLC2A1 and its paired proteins were related to GRIN2B, GRIN1, KIF9, SLC26A6, PGF, SCARB1, DISP3, GRIN2A, SCARB2, DXO, ARG1, DMTN, ANKRD17, SESN3, SNX27, TEX10, COL16A1, ZFP2, RBM19, and BCAR1 ( Figure 7B ). GO and KEGG enrichment analysis showed that SLC2A1 and its paired proteins (CA2, MCM5, NRP1, CD82, COL4A1, FLOT1, FLOT2, PTPN12, and CD36) were mainly enriched in positive regulation of cell adhesion, positive regulation of NF−kappa B transcription factor activity, regulation of protein binding (BP function), basolateral plasma membrane, membrane raft, membrane microdomain, membrane region, cell−cell contact zone (CC function), growth factor binding, ionotropic glutamate receptor binding, and glutamate receptor binding (MF function) ( Figure 7C ). Their main enriched pathways included adipocytokine signaling pathway, ECM−receptor interaction, bile secretion, and insulin resistance ( Figure 7D ).

We analyzed the key factors affecting the association of SLC2A1 with EMT. We performed co-expression correlation analysis between SLC2A1 and classical EMT markers (CDH1, CDH2, DSP, FN1, ITGB6, MMP9, TJP1, and VIM) using proteomic quantification. We found a significant positive expression correlation of SLC2A1 with CDH2, DSP, FN1, ITGB6, and TJP1 (Cor>0.3, P<0.05, Figures 8B–E, G ). In contrast, there was no significant correlation between SLC2A1 and CDH1, MMP9, or VIM (P>0.05, Figures 8A, F, H ).

We used immunohistochemical staining to detect SLC2A1 expression in non-invasive and invasive pituitary adenomas. The results showed that SLC2A1 was mainly expressed in the cytoplasm. SLC2A1 expression was significantly higher in invasive pituitary adenomas than in non-invasive pituitary adenomas, the difference was statistically significant (P<0.05) ( Figure 9 ). This was consistent with our quantitative proteomics and GSE169498 analysis, suggesting that SLC2A1 may be a potential biomarker of invasion for pituitary adenoma.