Abstract

   Gliomas are the most prevalent and aggressive neoplasms of the central
   nervous system, representing a major challenge for effective treatment
   and patient prognosis. This study identifies the proteasome subunit
   beta type-8 (PSMB8/LMP7) as a promising prognostic biomarker for
   glioma. Using a multiparametric radiomic model derived from
   preoperative magnetic resonance imaging (MRI), we accurately predicted
   PSMB8 expression levels. Notably, radiomic prediction of poor prognosis
   was highly consistent with elevated PSMB8 expression. Our findings
   demonstrate that PSMB8 depletion not only suppressed glioma cell
   proliferation and migration but also induced apoptosis via activation
   of the transforming growth factor beta (TGF-β) signaling pathway. This
   was supported by downregulation of key receptors (TGFBR1 and TGFBR2).
   Furthermore, interference with PSMB8 expression impaired
   phosphorylation and nuclear translocation of SMAD2/3, critical
   mediators of TGF-β signaling. Consequently, these molecular alterations
   resulted in reduced tumor progression and enhanced sensitivity to
   temozolomide (TMZ), a standard chemotherapeutic agent. Overall, our
   findings highlight PSMB8's pivotal role in glioma pathophysiology and
   its potential as a prognostic marker. This study also demonstrates the
   clinical utility of MRI radiomics for preoperative risk stratification
   and pre-diagnosis. Targeted inhibition of PSMB8 may represent a
   therapeutic strategy to overcome TMZ resistance and improve glioma
   patient outcomes.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s43556-025-00268-5.

   Keywords: Glioma, Proteasome subunit beta type-8 (PSMB8),
   Proliferation, Apoptosis, Radiomics

Introduction

   Gliomas, which account for over 80% of malignant brain tumors, are the
   most prevalent primary CNS tumors, with prominent recurrence propensity
   posing significant clinical challenges [[42]1]. According to the World
   Health Organization (WHO) classification, the majority of gliomas are
   diffuse and stratified into three molecular subtypes: (1) Isocitrate
   dehydrogenase (IDH) mutant-1p/19q co-deleted (oligodendroglioma), (2)
   IDH-mutant-1p/19q retained (astrocytoma), and (3) IDH-wildtype
   glioblastoma (GBM) [[43]2]. Although significant progress has been made
   in therapeutic modalities including surgical resection, radiation
   therapy, and systemic chemotherapy, the median overall survival period
   of patients with GBM remains less than 16 months [[44]3]. In recent
   years, immunotherapy has come to the fore as a potentially effective
   treatment method for patients suffering from malignant tumors,
   including glioma [[45]4, [46]5]. However, only a subset of glioma
   patients responds to immunotherapy. GBM, the most aggressive subtype,
   demonstrates extreme invasiveness and poor prognosis, with a mere 5.6%
   5-year survival rate post-diagnosis [[47]2, [48]6].

   The 26S proteasome consists of the 20S proteasome, 19S (PA700) and 11S
   (PA28) [[49]7]. The 20S proteasome represents the catalytic core of
   this enzymatic system, comprising two α-rings and two β-rings
   (α₁₋₇β₁₋₇β₁₋₇α₁₋₇), with each ring comprising seven subunits. The
   external α-rings play a critical role in controlling substrate entry
   into the internal catalytic domain, which is constituted by β-subunits
   [[50]8]. When the PA700 and PA28 regulatory complexes associate with
   the α-rings of the core particle, they give rise to the 26S proteasome
   and immunoproteasome, respectively. Notably, the 20S proteasome is
   recognized for its ability to degrade oxidized proteins, a process that
   operates independently of ubiquitination [[51]7]. The proteasome
   mediates the degradation of ubiquitinated proteins, enabling rapid
   intracellular turnover and generating the primary source of peptide
   antigens for major histocompatibility complex class I (MHC-I)
   presentation [[52]9]. Proteasome subunit beta type-8 (PSMB8/LMP7), a
   catalytic β subunit (β5i) of the 20S proteasome, is transcriptionally
   upregulated by interferon-gamma (IFN-γ) [[53]10]. A recent analysis
   showed that high PSMB8 methylation is associated with an inhibitory
   response to immunotherapy [[54]11]. A previous study demonstrated that
   PSMB8 optimized the antigen presentation of MHC-I-restricted epitopes
   to CD8^+ T cells by cleaving proteins after their basic and hydrophobic
   residues [[55]12]. Moreover, during the development of
   colitis-associated colonic cancer, the pro-tumoral capacity of
   immunoproteasomes appears to contribute to carcinogenesis by triggering
   the upregulation of diverse pro-inflammatory factors, like CXCL1,
   CXCL2, CXCL3, IL17 A, IL1β, and IL6 [[56]13, [57]14]. In gastric
   cancer, the high nuclear expression level of PSMB8 has been linked to
   the depth of tumor invasion, lymph node metastasis, and an unfavorable
   prognosis [[58]15]. However, overexpression of PSMB8 increased
   cytotoxicity and apoptosis, and promoted radiation-induced tumor
   necrosis in rectal cancer [[59]16]. Previous research in glioma
   demonstrated that PSMB8 promotes tumor cell proliferation and migration
   by downregulating cyclin family proteins, N-cadherin, and vimentin,
   while upregulating E-cadherin via activation of the ERK1/2 and PI3K/AKT
   signaling pathways [[60]17]. Moreover, another study indicated that in
   glioblastoma, high expression of PSMB8 could increase tumor
   angiogenesis by enhancing the expression of vascular endothelial growth
   factor (VEGF)-A, vascular endothelial growth factor receptor (VEGFR),
   and CD31 [[61]18].

   Transforming growth factor beta (TGF-β), a multifunctional polypeptide
   cytokine, has emerged as a promising therapeutic target due to its role
   in regulating cell growth and differentiation [[62]19]. The mouse
   models have been utilized to explore the functional significance of the
   TGF-β pathway in mammary gland physiology and carcinogenesis, and these
   studies are achieved by making adjustments to the principal components
   of this signaling cascade [[63]20]. Cancer cells manipulate multiple
   pathways triggered by TGF-β for their own advantage, transforming TGF-β
   into an oncogenic factor. This factor then promotes angiogenesis,
   enables cancer cell invasion, leads to immunosuppression, and
   facilitates the self-renewal of cancer-initiating cells [[64]21]. The
   cellular response to TGF-β is triggered when ligands (TGF-β1, TGF-β2,
   or TGF-β3) bind to TGFBR2, a serine-threonine kinase. Subsequently,
   TGFBR2 phosphorylates and activates TGFBR1, the primary mediator of
   downstream signaling transduction [[65]22]. TGFBR1 then phosphorylates
   SMAD2/3, activating these transcription factors to drive
   TGF-β-dependent gene expression [[66]22].

   In the current research, we explored the biological function of PSMB8
   during the proliferation, migration and apoptosis processes of glioma
   cells. We further investigated whether the elevated expression of PSMB8
   has a connection with the activation of the TGF-β pathway, thereby
   accelerating glioma progression. Moreover, radiomics was used to
   preoperatively predict PSMB8 expression and it highlighted that PSMB8
   is a prognostic marker and treatment target for temozolomide
   (TMZ)-resistant glioma.

Results

Image prediction and regulating mechanism for selected genes

   Figure [67]1 shows an overview of the radiomics analysis. To begin
   with, for the purpose of normalizing the intensity and geometry, all
   MRI images underwent preprocessing. Specifically, the bias field
   distortion was rectified by making use of the N4ITK algorithm.
   Trilinear interpolation was used to resample isotropic voxels into
   1×1×1 mm^3 voxels. Subsequently, MRI images were rigidly registered
   using the axial resampled T1c images as a reference. This registration
   process was based on a mutual information similarity metric, ultimately
   generating the registered images. To accomplish intensity
   normalization, histogram matching was employed. The workflow included:
   (1) tumor delineation, (2) image preprocessing, (3) radiomics feature
   extraction, (4) feature selection, and (5) model construction and
   assessment. This research emphasized the utility of MRI radiomics in
   preoperative risk stratification and pre-diagnosis for targeting genes.
   We used our own cohort of 333 glioma cases to analyze the genes that
   may play crucial regulatory roles during the progression of glioma. For
   the selected genes, we then investigated the regulating mechanism using
   bioinformatics analysis and biological experimental techniques in vitro
   and in vivo.

Fig. 1.

   [68]Fig. 1
   [69]Open in a new tab

   The workflow of this study. LGG: Low-grade glioma; GBM: Glioblastoma
   multiforme; TCGA: The Cancer Genome Atlas; CGGA: the Chinese Glioma
   Genome Atlas; CPTAC: Clinical Proteomic Tumor Analysis Consortium; SMC:
   Samsung Medical Center

Increased PSMB8 expression in GBM was associated with poor clinical outcome

   Recent investigations have revealed that the abnormal gene expression
   of the 26S proteasome subunits family has been described in various
   oncogenic situations [[70]23–[71]26]. The volcano plot analysis of TCGA
   cohort ([72]https://www.cancer.gov/ccg/research/genome-sequencing/tcga)
   and CGGA cohort ([73]https://www.cgga.org.cn/) showed that PSMB8 was
   the most up-regulated in GBM, compared to LGG (Fig. [74]2a). A heat map
   of proteasome-related genes expression from our pre-existing mRNA
   sequencing dataset of 333 glioma tissues [[75]27–[76]29] demonstrated
   that the expression level of PSMB8 in WHO grade 4 gliomas was
   significantly elevated in contrast to that in lower-grade gliomas (WHO
   grades 2 and 3) (Fig. [77]2b), which was in line with the outcome of
   PSMB8 mRNA expression in the [78]GSE16011 cohort
   ([79]https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE16011) and
   REM cohort
   ([80]https://www.cancerimagingarchive.net/collection/rembrandt/)
   (Fig. [81]2c). PSMB8 mRNA expression in 32 tumor types was compared
   using the Gene Expression Profiling Interactive Analysis database
   ([82]http://gepia.cancer-pku.cn/). GBM and LGG were among the top six
   tumor types expressing PSMB8 (Fig. S1a). To gain insight into PSMB8 at
   the proteomic level, we investigated global proteomic data from the
   Samsung Medical Center cohort [[83]30] and Clinical Proteomic Tumor
   Analysis Consortium protein dataset [[84]31]. PSMB8 protein expression
   profiles in patients of different subtypes showed that PSMB8 was more
   highly expressed in tumor tissues. Moreover, patients with the
   malignant mesenchymal subtype and the highest PSMB8 had a worse
   prognosis than those in the other two subtypes (Fig. [85]2d). To
   further investigate PSMB8 expression in gliomas, we explored its
   expression level in gliomas and normal brain tissues using
   immunohistochemistry. Analysis of 110 glioma and 30 normal brain tissue
   arrays indicated that PSMB8 levels were significantly higher in tumor
   tissues, and elevated PSMB8 protein expression was associated with
   higher grades (Fig. [86]2e). Additionally, analysis of pre-existing RNA
   sequencing data sourced from the CGGA revealed that there was a
   negative correlation between PSMB8 and prognosis in both primary and
   recurrent gliomas (Fig. [87]2f).

Fig. 2.

   [88]Fig. 2
   [89]Open in a new tab

   Increased PSMB8 expression in GBM was associated with poor clinical
   outcome. a Volcano plots acquired from the CGGA and TCGA data analysis,
   displaying the expression of the 26S proteasome subunits genes among
   LGG and GBM. Ns., not significant; FC, fold change. b A significant
   upregulation of PSMB8 with increasing WHO grade was determined based on
   mRNA-sequencing of our 333 gliomas cohort. c PSMB8 mRNA expression
   profiles in different-grade glioblastoma patients were collected from
   the [90]GSE16011 and REM cohorts to compare PSMB8 expression in
   glioblastoma (WHO grade 4), low-grade glioma (WHO grades 2 and 3), and
   normal brain tissues (Control). d The protein levels of PSMB8 between
   three mRNA subtypes and normal brain tissues were detected via the
   MS/MC proteomic data from SMC cohort, and CPTAC cohort. e Left:
   Representative Immunohistochemistry (IHC) images of PSMB8 expression in
   our 110 glioma and 30 normal tissue arrays. Scar bar = 100 µm. Right:
   The frequency distribution of PSMB8 expression in low-stage glioma
   tissues (WHO grade 2, n = 30. WHO grade 3, n = 45), high-grade
   glioblastoma (WHO grade 4, n = 35), and normal brain tissues (n = 30).
   f Kaplan–Meier plot of overall survival of glioma patients with low and
   high PSMB8 expression from CGGA primary and recurrent glioma cohort.
   *p < 0.05. Data are analyzed of three independent experiments and
   presented as mean ± SD (c, d, and e right). Statistical analyses are
   performed using two-tailed Student’s t-test (c, d, and e right) and
   log-rank test (f)

Development of the multiparametric radiomic model based on preoperative MRI

   All 162 patients with information on the expression of PSMB8 were
   separated into a training set (n = 121) and a validation set (n = 41).
   And the training set (age, 52.4 ± 13.8 years) included 82 (67.8%) high
   expression, and 39 (32.2%) low expression PSMB8 gliomas (cutoff value
   = 52.48). Concurrently, in the validation set (age, 49.9 ± 11.6 years),
   there were 34 (82.9%) high PSMB8 expression and 7 (17.1%) low PSMB8
   expression gliomas. Clinical factors, including sex (P = 1.00), age
   (P = 0.92), WHO grade (P = 0.56) and PSMB8 expression (P = 0.10)
   between the training and validation sets were not significantly
   different (P < 0.05, considered statistically significant), as shown
   in Supplementary Table S1.

   For each patient, a quantity of 5929 radiomic features were extracted
   out of multiparametric MRI, including T1-weighted (T1), T2-weighted
   (T2), fluid-attenuated inversion recovery (FLAIR), T1-weighted
   gadolinium contrast-enhanced (T1c), diffusion-weighted (DWI) imaging,
   and apparent diffusion coefficient (ADC) maps. Subsequently, 1402
   features remained after the stage of intraclass correlation coefficient
   (ICC) and redundancy reduction. Heat maps representing the correlation
   coefficients of all nonredundant feature sets are shown in Fig. S2a.
   Ten highly relevant radiomic features, including six texture features
   and four intensity features, were selected after Boruta selection (Fig.
   S2b, Supplementary Table S2. A significant correlation was found
   between the 10 selected features and the expression level of PSMB8
   (false discovery rate-adjusted P < 0.01). The classification
   performance of the multiparametric radiomic model, which was used to
   predict the expression of PSMB8, in the training and validation sets is
   presented in Supplementary Table S3. The univariate contribution of
   each of the ten selected features for the expression of PSMB8
   prediction can be described using the Mean-Decrease-Gini index. A heat
   map of the Mean-Decrease-Gini index is shown in Fig. [91]3a.
   Figure [92]3b displays the receiver operating characteristic (ROC)
   curves for both the training set and the validation set. The area under
   the curve (AUC) values of these two sets were 0.88 and 0.81,
   respectively. And the decision curves for predicting the expression of
   PSMB8 are shown in Fig. [93]3c, which indicates a substantial overall
   net benefit throughout the majority of the ranges of threshold
   probabilities. Simultaneously, the calibration curve of the
   multiparametric radiomic model showed a striking level of concordance
   between the predicted values and the actual observations, as shown in
   Fig. [94]3d. A greater Mean-Decrease-Gini index value indicates that a
   feature exhibits superior classification performance in predicting the
   expression of PSMB8. The detailed definitions of the 10 selected
   features are described in Supplementary Table S4.

Fig. 3.

   [95]Fig. 3
   [96]Open in a new tab

   A graphic illustration of classification performance of the
   multiparametric radiomic model for the expression of PSMB8. a The heat
   map of the univariate contribution of each 10 selected features for the
   prediction of the expression of PSMB8; a higher Mean-Decrease-Gini
   index value indicated better classification performance of a feature. b
   ROC curves of both the training and validation sets; AUC values are
   shown in this part. c The decision curves for the prediction of the
   expression of PSMB8; the multiparametric radiomic model showed a great
   overall net benefit in most ranges of the threshold probabilities. d
   The calibration curve of the multiparametric radiomic model; there was
   a high agreement between prediction and observation. e Left: Immunoblot
   (IB) of the protein expression levels of PSMB8 in normal tissues and
   glioma tissues with different WHO grades. Right: Quantitative analysis
   of the expression levels (relative band density compared to β-actin) in
   normal tissues (Normal) and glioma tissues with different WHO grades
   (grades 2, 3, and 4). f RT-qPCR analysis of normal tissues and glioma
   tissues with different WHO grades. g The overall survival analysis of
   another 138 patients with glioma in terms of radiomics. *p < 0.05. Data
   are analyzed of at least three independent experiments and presented as
   mean ± SD (e right and f). Statistical analyses are performed using
   two-tailed Student’s t-test (e right, and f) and log-rank test (g)

   Then, another 138 glioma patients with multiparametric MRI data were
   encompassed in the present study for validating. And the results of
   radiomic model showed 80 in high groups and 58 in low groups. For the
   samples of these 138 patients, high and low groups were subjected to
   IB. Compared with the low group, PSMB8 was significantly more
   abundantly present in the high group (Fig. [97]3e), consistent with the
   results of real-time reverse transcription PCR (RT-qPCR) (Fig. [98]3f).
   Analysis of the 138 glioma cohort sequencing data indicated that high
   subtype of PSMB8, predicted via radiomics was negatively correlated
   with prognosis, compared with low subtype (Fig. [99]3g). All the
   results showed that radiomics has potential to preoperatively predict
   the expression of PSMB8 and its relation with clinical prognosis
   (Fig. [100]3e-g).

PSMB8 downregulation reduced the malignant behaviors of glioma cell lines in
vitro

   Based on the aforementioned findings, we put forward the hypothesis
   that a reduction in the expression of PSMB8 could potentially impede
   the progression of GBM. Thus, we silenced PSMB8 in LN229 and U87 MG GBM
   cell lines via lentiviral infection. Transfection efficiency was
   evaluated using fluorescence (Fig. S3a), RT-qPCR (Fig. S3b), and IB
   (Fig. S3c), suggesting that PSMB8 knockdown cell models were
   successfully constructed in both cell lines. We then explored the
   effects of PSMB8 downregulation on GBM malignancy based on CCK-8, wound
   healing, and flow cytometric assays. PSMB8 suppression resulted in a
   substantial reduction in the proliferation rate, self-renewal ability,
   and migratory capacity of GBM cells (Fig. [101]4a–d). Moreover, flow
   cytometry analysis suggested that PSMB8 knockdown arrested the cell
   cycle (Fig. [102]4c). Simultaneously, fluorescence-activated cell
   sorting analysis revealed that the apoptosis rate in PSMB8-silenced GBM
   cells was remarkably higher compared to that in control cells (Co.sh)
   (Fig. [103]4d).

Fig. 4.

   [104]Fig. 4
   [105]Open in a new tab

   PSMB8 downregulation reduced the malignant behaviors of glioma cell
   lines in vitro. a Cell proliferation rate was evaluated in GBM cell
   lines after transfection using the CCK-8 assay. b Wound healing assay
   of Co.sh, sh-M #1, and sh-M #2 in LN229 cells after 24 h. c The effects
   of PSMB8 knockdown on the cell cycle were determined using flow
   cytometry. d The effects of PSMB8 knockdown on apoptosis were examined
   using flow cytometry. *p < 0.05. Data are analyzed of three independent
   experiments and presented as mean ± SD (a, b right, c and d down).
   Statistical analyses are performed using two-tailed Student’s t-test
   (a, b right, c down and d down)

PSMB8 expression correlates with TGF-β signaling pathway activity in glioma

   To uncover the mechanism by which PSMB8 facilitates malignant glioma
   cell behavior, we applied RNA sequencing analysis to Co.sh, sh-M#1, and
   sh-M#2 LN229 cells to determine transcriptional changes. Among the
   significantly differentially regulated genes (DRGs), transcripts of
   2319 (sh-M#1, 17.8%) and 1263 (sh-M#2, 9.75%) genes were downregulated,
   whereas transcripts of 1958 (sh-M#1, 15.03%) and 974 (sh-M#2, 7.52%)
   were upregulated in PSMB8 knockdown cells as compared to the control
   LN229 cells (Fig. [106]5a, b, and S4a-b). Furthermore, gene ontology
   (GO) enrichment analysis conducted on the DRGs disclosed there were
   significantly impacted gene categories. These genes were either
   downregulated or upregulated as a response to PSMB8 deficiency. The
   downregulated genes of Co.sh vs sh-M#2 were associated with
   multicellular organismal processes, responses to endogenous stimuli,
   and cell adhesion (Fig. [107]5c, Supplementary Table S5), suggesting
   that PSMB8 was closely associated with cellular biological processes
   and migration in cancer. Upregulated GO terms indicated anatomical
   structure development and cellular response to chemical stimuli (Fig.
   S4c, Supplementary Table S5). Furthermore, KEGG pathway analysis of
   downregulated genes of Co.sh vs sh-M#2 showed that the genes were
   enriched within the ECM-receptor interaction, focal adhesion, and the
   TGF-β signaling pathway (Fig. [108]5d, Supplementary Table S6), whereas
   those enriched in ribosome and oxidative phosphorylation were
   upregulated in PSMB8-knockdown glioma cells (Fig. S4d, Supplementary
   Table S6). Moreover, many downregulated DEGs were crucial for the TGF-β
   signaling pathway (Fig. [109]5e). Additionally, KEGG pathway enrichment
   analysis of all genes selected from the differentially expressed genes
   showed that the TGF-β signaling pathway might exert crucial functions
   in glioma progression (Fig. [110]5f). We validated the differentially
   altered targets of the TGF-β pathway using RT-qPCR (Fig. [111]5g). In
   line with the RNA-seq data, RT-qPCR verified that UBE2M and STUB1 were
   upregulated in PSMB8-deficient cells, whereas TGFBR1, TGFBR2, ITGB8,
   and ITGAV were downregulated (Fig. [112]5g). We further used IB to
   determine the differentially altered protein levels of the key protein
   targets (Fig. [113]5h). The biochemical results showed that TGFBR1 and
   TGFBR2 levels were significantly reduced based on IB and RT-qPCR
   (Fig. [114]5g, h). Here, we combined knock down of PSMB8 with a
   short-term course of treatment with galunisertib (LY2157299), which is
   a small molecule inhibitor targeting TGFBR1 (hereafter referred to as
   TGFBRi) (Fig. [115]5i, j). After silencing PSMB8, reduced expression of
   TGFBR1 and TGFBR2 led to decreased phosphorylation of SMAD2/3
   (Fig. [116]5h), which inhibited the malignant behavior of glioma cells
   (Fig. [117]5j). Moreover, TGFBRi treatment further inhibited the
   phosphorylation of SMAD2/3 and tumor proliferation. However, the
   opposite result was observed in PSMB8-silenced cells overexpressing
   TGFBR1 or TGFBR2 (Fig. [118]5i, j).

Fig. 5.

   [119]Fig. 5
   [120]Open in a new tab

   PSMB8 expression correlates with TGF-β signaling pathway activity in
   glioma. a A volcano plot illustrating differentially regulated gene
   expression from RNA-sequencing analysis between the Co.sh group and
   sh-M#1 group. Genes upregulated and downregulated in the sh-M#1 group
   compared with those in the control group are shown in red and green,
   respectively. b RNA-seq comparison of Co.sh and sh-M#2 revealed a total
   of 12,958 genes expressed, of which 974 genes were upregulated and 1263
   genes were downregulated. c Gene ontology functional clustering of
   genes that were downregulated in biological processes. d KEGG pathway
   analysis of genes that were downregulated in PSMB8-deficient
   transcriptome. e Heatmap of representative TGF-β signaling
   pathway-related genes. f TGF-β signaling pathway in KEGG; The red or
   blue targets represent the upregulated or downregulated KEGG targets,
   respectively. (g, h) The expression of several TGF-β signaling
   pathway-related genes identified by RT-qPCR (g) and IB (h). i TGFBRi
   treatment or TGFBR1/2 overexpression (OE) in PSMB8-silenced cells and
   immunoblot of the phosphorylation of SMAD2/3, TGFBR1, and TGFBR2. j
   Growth of PSMB8-silenced cells treated with or without TGFBRi. *p <
   0.05; ns., not significant. Data are analyzed of three independent
   experiments and presented as mean ± SD (g, j). Statistical analyses are
   performed using two-tailed Student’s t-test (g, j)

PSMB8 suppression restrains GBM growth and enhances the sensibility of TMZ in
mice

   The development of TMZ resistance remains a major obstacle to GBM
   treatment [[121]32]. To verify that the knockdown of PSMB8 was able to
   suppress the growth of GBM tumors and improve the effectiveness of TMZ
   in vivo, U87MG (5 × 10^6) cells underwent subcutaneous injection into
   the armpits of 6-week-old experimental BALB/c nude mice. Subsequently,
   these nude mice were thereafter grouped into four groups according to
   PSMB8 silencing and presence or absence of TMZ treatment. Nine days
   after the injection, the mice underwent intraperitoneal injection
   treatments with TMZ or the same volume of phosphate buffer saline (PBS)
   every 3 days (Fig. [122]6a). Compared with the control group, tumors in
   the sh-M#2 group or the TMZ treatment group had smaller sizes, weights,
   and volumes. Moreover, in the group receiving TMZ treatment, PSMB8
   knockdown led to an additional decrease in these parameters
   (Fig. [123]6b-d). After that, immunofluorescence staining was employed
   to examine the expression of PSMB8, p-SMAD2, and Ki-67 in the
   transplanted tumors of the Co.sh and sh-M#2 groups. The results
   indicated that, when compared to Co.sh-transduced xenografts, the
   expression magnitudes of PSMB8, p-SMAD2, and Ki-67 in the
   sh-M#2-transduced GBM xenografts were significantly reduced
   (Fig. [124]6f). In line with this, RT-qPCR analysis demonstrated that
   knockdown of PSMB8 or TMZ treatment dramatically reduced the expression
   of c-Myc and cyclin D1 (tumor growth markers) and enhanced caspase- 3
   (tumor apoptosis marker) (Fig. [125]6e). Depletion of PSMB8 or TMZ
   combination inhibited GBM tumor growth by reducing the downstream
   protein expression and SMAD2 phosphorylation of the TGF-β pathway
   (Fig. [126]6g). Taken together, our results showed that PSMB8 could
   stabilize or enhance the expression of TGFBR1/2, and the subsequent
   increase in the phosphorylation of SMAD2/3 promoted nuclear
   translocation and downstream target genes of the TGF-β pathway
   (Fig. [127]6h).

Fig. 6.

   [128]Fig. 6
   [129]Open in a new tab

   PSMB8 suppression restrains GBM growth and enhances the sensibility of
   TMZ in mice. a Diagram of the experimental procedure. U87MG cells (5
   × 10^6 cells) transfected with Co.sh and sh-M#2 were subcutaneously
   injected into BALB/c nude mice (n = 5 mice per group). Nine days after
   injection, mice were injected with PBS or TMZ every 3 days. b
   Representative images xenograft tumors of the nude mice. c Curves of
   tumor growth in the PSMB8 knockdown group and control group with or
   without TMZ treatment were analyzed. The tumor volumes were measured
   every 9 days. d Tumor weights of each group were measured after being
   surgically dissected. e RT-qPCR analysis of c-Myc, cyclin D1, and
   caspase-3 mRNA expression in tumors from four different treatment
   groups. f PSMB8, p-SMAD2, and Ki67 expression in tumor tissues of nude
   mice was determined via immunofluorescence staining. Scar bar = 100 µm.
   g Immunoblotting analysis of tumor tissues. h Schematic illustration of
   radiomic prediction of PSMB8 expression, which enhanced the efficacy of
   temozolomide via the TGF-β signaling pathway. *p < 0.05; Data are
   representative of three independent experiments (e) and are shown as
   mean ± SD. Statistical analyses were performed using a two-tailed
   Student’s t-test (c-e)

Discussion

   Gliomas rank among the most prevalent and severe types of human
   cancers. The diagnosis of glioma relies on histopathological criteria
   and molecular genetics, and tumors are categorized into different
   grades in accordance with the WHO classification system. This
   categorization is intimately associated with the prognosis and
   treatment approaches for glioma patients [[130]33]. Glioblastoma (WHO
   grade 4) represents the most common glioma among adults and is
   characterized by an unfavorable prognosis [[131]33–[132]35]. Even with
   the existing standard treatment, involving the maximum safe resection,
   and then radiotherapy in combination with TMZ [[133]36], and a range of
   salvage therapies being available once tumor progression takes place,
   the vast majority of patients still succumb to the disease within two
   years after being diagnosed [[134]3, [135]37].

   PSMB8 is a major component of the immunoproteasome [[136]38], which is
   found mainly in monocytes and lymphocytes and is known to process MHC-I
   [[137]39]. PSMB8 also has a significant impact on myatrophy,
   arthrogryposis, panniculitis-associated lipodystrophy syndrome, and
   microcytic anemia [[138]38]. Earlier research has indicated that PSMB8
   exhibits high expression levels in tumor tissues. Moreover, its
   expression level rises as gliomas progress [[139]17, [140]18]. In other
   words, the expression level of PSMB8 is significantly higher in
   high-grade gliomas as opposed to low-grade gliomas. Additionally, it
   was discovered that the ERK1/2 and PI3 K/AKT signaling pathways
   triggered by PSMB8 played a role in the migration and proliferation
   processes of glioma cells [[141]17]. However, our study focused on
   uncovering the role of PSMB8-mediated malignant glioma phenotypes
   through the activation of the TGFBR1/2-SMAD2/3 axis. This new direction
   of our investigation explored a different molecular mechanism compared
   to the previously identified pathways, which may offer potential
   therapeutic targets that have not been considered before in the context
   of glioma treatment related to PSMB8's actions. TGF-β signaling plays
   important roles in inducing epithelial-mesenchymal transition, cell
   growth, differentiation, and morphogenesis [[142]40–[143]42]. Some
   studies have shown that TGF-β can induce migration and invasion of
   various cancer cell types, including gliomas [[144]43]. Previous
   research also shown that aggressive and highly proliferative gliomas
   exhibit high TGFβ-Smad activity, resulting in a poor prognosis for
   patients [[145]44]. TGF-β1, along with TGFBR1/2, shows expression only
   in malignant gliomas and is not expressed in normal brain tissues,
   gliosis, or low-grade astrocytomas [[146]45, [147]46]. Our study also
   showed that blocking the key proteins of TGF-β signaling pathway can be
   a new therapeutic target.

   An escalating number of investigations imply that radiomics has
   multiple applications in glioma, including early diagnosis and
   classification, as well as prediction of biological behavior and
   prognosis [[148]47–[149]49]. After incorporation of molecular pathology
   into the traditional histological analysis used for classifying
   gliomas, a steadily increasing number of studies have centered on
   machine-learning-based radiomics analysis with the aim of uncovering
   molecular markers such as IDH mutations, 1p/19q codeletion, and various
   other markers [[150]50–[151]52]. A non-invasive approach for
   preoperatively forecasting key molecular markers in gliomas holds
   clinical importance, and we combined radiomics with genomics from a
   novel perspective to explore the image recognition ability of new gene
   markers.

   To date, few studies have clarified the association between PSMB8 and
   prognostic value prediction via MRI. In this study, we constructed a
   multiparametric radiomic model based on preoperative MRI, which
   achieved high AUC values for predicting high and low PSMB8 expression.
   Moreover, we investigated the downstream signaling pathway via
   knockdown of PSMB8 in glioma cell lines.

   The radiomic model attained high AUC values for PSMB8 based on
   single-center data. Nevertheless, international multi-institutional
   datasets are required for further validation. While we showed the
   silenced PSMB8 could reduce the expression of TGFBR1/2, how PSMB8
   regulated directedly and why PSMB8 effects the expression of them
   together. It required further research that the mechanism of PSMB8
   enhanced the sensitivity of TMZ in GBM patients.

Conclusion

   Collectively, our study revealed that PSMB8 can not only be a useful
   prognostic factor in glioma patients, but also a predictive marker of
   prognostic factors and novel therapeutic target by blocking key
   proteins of TGF-β signaling.

Materials and methods

Development of the radiomic model

   A total of 162 patients with the expression of PSMB8 were included in
   the current MRI-based radiomics study. The standards for patient
   inclusion were (1) age ≥ 18 years; (2) pathologically confirmed WHO
   grade 2,3,4 diffuse gliomas; (3) availability of multiparametric MRI,
   including T1, T2, FLAIR, T1c, and ADC; (4) each image without any
   artifacts, confirmed by a neurosurgeon (ZYZ); (5) absence of
   preoperative medication treatment, chemotherapy, and radiation therapy;
   and (6) availability of information on the expression of PSMB8.

   MRI images were acquired using 1.5 T or 3.0 T clinical MR scanners with
   different imaging parameters, and the detailed information on the MR
   machines and imaging parameters was summarized in Supplementary Methods
   A, and Table S7. To standardize the intensity and geometry, all MRI
   images were preprocessed. The bias field distortion was rectified by
   making use of the N4ITK algorithm [[152]53]. The detailed process of
   image normalization and the delineation of the tumors was described in
   Supplementary Methods B. The radiomic features were extracted using the
   Python tool PyRadiomics [[153]54]. A summary of the detailed procedure
   for radiomic feature extraction can be found in Supplementary Methods
   C.

   The ICC values corresponding to each feature of the 30 selected
   patients were calculated with R software (version
   4.2.1, [154]www.Rproject.org) to measure the repeatability. Features
   with ICC values higher than 0.7 were retained. Subsequently, the
   patients were randomly partitioned into a training set and a validation
   set according to a 7:3 ratio. Detailed information on the radiomic
   feature selection was presented in the Supplementary Methods D. The
   multiparametric radiomic model for classifying the expression of PSMB8
   (high and low expression, cutoff value = 52.48) was built by applying a
   probability algorithm based on random forest. For the purpose of
   optimizing the multiparametric radiomic model, tenfold cross-validation
   was applied.

Patients

   The First Affiliated Hospital of Zhengzhou University's Human
   Scientific Ethics Committee approved this study (Approval No.
   2019-KY-176). All the steps of procedures were implemented in
   conformity with the relevant guidelines and regulatory requirements set
   forth by the First Affiliated Hospital of Zhengzhou University. For the
   patients whose fresh tumor specimens were utilized for mRNA-sequencing,
   quantitative reverse transcription PCR, IHC, and western blotting,
   informed consent was duly acquired. And informed consent was also
   acquired from the patients who consented to undergo MRI scans.

Cell culture and reagents

   Human glioblastoma cell lines LN229 and U87MG were procured from the
   Cell Resource Center of Science (Shanghai, China). The cultivation of
   these cells took place in Dulbecco’s modified Eagle’s medium (DMEM)
   which had 10% FBS (Cat No. FSP500, ExCell Bio) added to it. All the
   cells were maintained in a humid environment at a temperature of 37℃
   with a 5% carbon dioxide atmosphere. Moreover, they were confirmed to
   be free from mycoplasma contamination.

Lentivirus transduction

   By transfecting HEK293 T cells with plKO-1 plasmids (designed to knock
   down PSMB8 shRNA) along with the helper plasmids pCMV-VSVG, pMDLg-RRE
   (gag/pol), and pRSV-REV, lentiviruses were generated. Forty-eight hours
   post-transfection, the cell supernatants were collected. These
   supernatants were either utilized to infect cells or stored at -80°C.
   To establish stable cell lines, GBM cells at a low confluence level of
   20% were infected for 24 h with lentiviral supernatants that were
   diluted 1:1 in normal culture medium. Forty-eight hours following the
   infection, the cells underwent puromycin selection for a duration of
   one week. Subsequently, they were subcultured before being used. A
   concentration of 1 µg/ml of puromycin was employed to sustain the
   selection pressure on the stably transfected GBM cells. The lentiviral
   shRNAs were identified and tested, and the shRNA with the highest
   efficacy was selected for use:
     * TRCN0000007268 (sh-M#1): CCACGTTAAGTCCAAGGAGAA;
     * TRCN0000007269 (sh-M#2): CCACTCACAGAGACAGCTATT;

Cell proliferation detection

   The LN229 and U87MG cells infected with shRNAs were seeded into 96-well
   plates at a density of 5.0 × 10^3 cells per well. After that, the
   plates were placed in an incubator maintained at 37 °C with a 5% CO₂
   atmosphere and cultured overnight. At 24, 48, 72, 96, and 120 h after
   changing the medium, the old solution was discarded. Each well received
   10 μL of the CCK-8 solution, and the plates were then kept in an
   incubator at 37 °C for 4 h for the incubation process. The optical
   density (OD) value of each experimental well at a wavelength of 450 nm
   was measured using a microplate reader. The purpose of this measurement
   was to monitor the variations in the cell proliferation capacity among
   different groups.

Wound healing

   Initially, the cell concentration was set at 3 × 10^5/ml and then
   seeded into a 6-well plate. Following incubation, once it was observed
   that the cells had achieved 100% confluence, a 200 µl sterile pipette
   tip was employed to draw three vertical lines. Subsequently, the cells
   that had detached were cleaned using sterile PBS. After that, 2 ml of
   DMEM was added to each well. Eventually, the scratches were examined
   under a microscope and captured at 0, and 24 h.

Apoptosis assays

   An Annexin V-fluorescein isothiocyanate (FITC)/propidium iodide (PI)
   Apoptosis Detection kit (Cat#C1062L, Beyotime) was used to evaluate
   apoptosis of LN229 and U87MG cells infected with shRNAs. The cells were
   gathered and rinsed on two occasions using precooled PBS. Subsequently,
   they were resuspended in 1 × binding buffer. After that, 5 µl of FITC
   and 5 µl of PI were added. After a 15-min incubation at room
   temperature without light, apoptosis was analyzed using a flow
   cytometer.

   Meanwhile, precooled 70% ethanol was used to fix the cells at 4 °C
   throughout the night. The following day, the fixed cells underwent
   centrifugation and were washed once with precooled PBS. Afterward, 500
   µl of the prepared PI staining buffer was gently added to each tube to
   fully resuspend the cells. Subsequently, the cells underwent incubation
   in a 37 °C warm bath in the dark for 30 min. Finally, a flow cytometer
   was used to measure the DNA content of the cells.

RNA isolation and quantitative real-time PCR (RT-qPCR)

   The NucleoSpin RNA II kit (BIOKE, Netherlands) was utilized to extract
   RNAs from the entire population of cells. Next, the process of reverse
   transcribing cDNA was accomplished by utilizing the RevertAid First
   Strand cDNA Synthesis Kit (Fermentas). Real-time PCR was conducted
   using a StepOnePlus real-time PCR system (Applied Bioscience 7500) with
   SYBR Green (Applied Bioscience). Each sample was analyzed in at least
   three replicates, and GAPDH functioned as an internal calibration
   standard. The primer sequences employed are listed in Supplementary
   Table S8.

Immunoblotting

   Cells were subjected to lysis using 1 ml of lysis buffer (comprising 20
   mM Tris–HCl at pH 7.4, 2 mM EDTA, 25 mM NaF, and 1% Triton X-100)
   supplemented with protease inhibitors (Cocktail, Bimake). This lysis
   process was carried out at 4 °C for 10 min. After that, the samples
   were centrifuged at 12 × 10^3 g for 10 min. Subsequently, the
   concentrations of the proteins were ascertained. Next, equivalent
   quantities of the proteins within the lysate were segregated via
   SDS–polyacrylamide gel electrophoresis (SDS–PAGE). Following SDS–PAGE,
   a transfer operation was carried out to move the proteins onto a
   nitrocellulose membrane. The membranes were first blocked in a blocking
   buffer (5% skim milk in TBS-T) for 1 h at room temperature. Then,
   immunoblotting was executed. And in this process, specific antibodies
   were employed, followed by the use of secondary anti-mouse or
   anti-rabbit antibodies that had been conjugated to horseradish
   peroxidase (Amersham Biosciences). Chemiluminescence was employed for
   visualization. The antibodies that were employed in IB procedure were
   as follows: PSMB8 at 1:1000 (IB; Cat#PTM-5783, PTM-biolab), β-actin at
   1:20,000 (IB; Cat#AC004, Abclonal), SMAD2 at 1:1,000 (IB; Cat#A-11,
   Santa Cruz), SMAD3 at 1:5,000 (IB; Cat#9513, Cell Signaling), SMAD4 at
   1:1,000 (IB; Cat#B8, Santa Cruz), TGFBR1 at 1:1,000 (IB; Cat#C-12,
   Santa Cruz), TGFBR2 at 1:1,000 (IB; Cat#E-6, Santa Cruz), and
   phospho-SMAD2/3 at 1:1,000 (IB; Cat#AP0548, Abclonal).

Immunohistochemical analysis

   The normal brains and glioma tissues were immersed in 4%
   Paraformaldehyde (PFA) for fixation, then encased in paraffin, and
   finally sliced into coronal sections with a thickness of 8 µm. Next,
   the sections underwent a process where they were first deparaffinized
   by being immersed in xylene. Then, they were rehydrated by passing them
   through a series of ethanol solutions with gradually decreasing
   concentrations. This was followed by an antigen retrieval step. After
   that, the sections were treated with 0.3% H₂O₂ for 20 min to suppress
   endogenous peroxidase activity. Then, they were incubated with 5% goat
   serum for 30 min. Next, the sections were incubated with the specified
   primary antibodies against PSMB8 (Cat#PTM-5783, PTM-biolab), p-SMAD2
   (Cat#A-11, Santa Cruz) or Ki 67 (Cat#A20018, Abclonal) at 4 °C for 12
   h. Following the incubation process, the sections underwent a washing
   step using PBS. Then, they were placed in an environment of 37 °C for a
   60-min incubation period with biotinylated secondary antibodies. Once
   the sections were washed again with PBS, they were treated with
   diaminobenzidine (DAB) for staining and then with hematoxylin for
   counterstaining.

Subcutaneous tumor model

   Mice experiments were approved by the First affiliated Hospital of
   Zhengzhou University Animal Welfare Committee. Female BALB/c nude mice
   that were six weeks old were obtained from the animal husbandry center
   of the Shanghai Institute of Cell Biology, Academia Sinica, situated in
   Shanghai, China. Prior to the commencement of the experiment, these
   mice were maintained in a specific pathogen-free setting for a period
   of three days. The mice were distributed randomly into two groups,
   where each group contained 10 mice. Subcutaneous injections of Co.sh or
   shPSMB8 cells were administered into the flanks of nude mice, at a rate
   of 5 million (5 × 10⁶) cells per mouse. Nine days after the injection
   of tumor cells, Co.sh or shPSMB8 mice were randomly partitioned into
   two distinct groups (n = 5 per subgroup). For one subgroup,
   intraperitoneal administration of TMZ was carried out at a dose of
   5 mg/kg, with the injection being administered every third day, while
   the other subgroup received an equal volume (100 µl) of PBS. This
   treatment continued until the 36 th day. After that, the mice were
   subjected to euthanized. The tumors were removed and weighed.
   Subsequently, tumors tissues were fixed with optimal cutting
   temperature compound (OTC) for immunofluorescence staining, and the
   tissues mRNA or proteins was extracted for RT-qPCR or immunoblotting.

Statistical analysis

   R software was utilized to carry out the statistical analyses. To
   assess the equilibrium in the distribution of clinical information such
   as sex, age, WHO grades, and PSMB8 expression between the training and
   validation sets, either the Wilcoxon test or the chi-square (× 2) test
   was employed. The classification capabilities of the multiparametric
   model were put to the test and evaluated based on metrics including
   accuracy, sensitivity, specificity values, and AUC value of the ROC
   curve. The optimal threshold of the AUC value was calculated based on
   the maximum Youden index (sensitivity + specificity—1). Subsequently,
   employing the identical threshold that was chosen from the training
   set, the multiparametric radiomic model underwent testing in the
   validation set. Unpaired two-tailed Student’s t-tests were used to
   examine the differences in values between two groups, while ANOVA was
   applied for multivariate analysis. Utilizing the Kaplan–Meier approach,
   survival curves were formulated, and the log-rank test was carried out
   to ascertain the P values. GraphPad Prism 7 (GraphPad Inc., La Jolla,
   CA, USA) was also employed for performing statistical analysis and
   graphing. In the Figures, P values were denoted by asterisks: *p <
   0.05. P value smaller than 0.05 was considered to be statistically
   significant.

Supplementary Information

   [155]Supplementary Material 1.^ (2.1MB, docx)

Acknowledgements