Integrated Bioinformatic Analyses Reveal Thioredoxin as a Putative Marker of Cancer Stem Cells and Prognosis in Prostate Cancer

Abstract

Objectives:

Prostate cancer stem cells (CSCs) play an important role in cancer cell survival, proliferation, metastasis, and recurrence; thus, removing CSCs is important for complete cancer removal. However, the mechanisms underlying CSC functions remain largely unknown, making it difficult to develop new anticancer drugs targeting CSCs. Herein, we aimed to identify novel factors that regulate stemness and predict prognosis.

Methods:

We reanalyzed 2 single-cell RNA sequencing data of prostate cancer (PCa) tissues using Seurat. We used gene set enrichment analysis (GSEA) to estimate CSCs and identified common upregulated genes in CSCs between these datasets. To investigate whether its expression levels change over CSC differentiation, we performed a trajectory analysis using monocle 3. In addition, GSEA helped us understand how the identified genes regulate stemness. Finally, to assess their clinical significance, we used the Cancer Genome Atlas database to evaluate their impact on prognosis.

Results:

The expression of thioredoxin (TXN), a redox enzyme, was approximately 1.2 times higher in prostate CSCs than in PCa cells (P < 1 × 10⁻¹⁰), and TXN expression decreased over CSC differentiation. In addition, GSEA suggested that intracellular signaling pathways, including MYC, may be involved in stemness regulation by TXN. Furthermore, TXN expression correlated with poor prognosis (P < .05) in PCa patients with high stemness.

Conclusions:

Despite the limited sample size in our study and the need for further in vitro and in vivo experiments to demonstrate whether TXN functionally regulates prostate CSCs, our findings suggest that TXN may serve as a novel therapeutic target against CSCs. Moreover, TXN expression in CSCs could be a useful marker for predicting the prognosis of PCa patients.

Introduction

In 2022, prostate cancer (PCa) was the second most frequent cancer and the fifth leading cause of death in men. ¹ Primarily diagnosed through prostate biopsy and measurement of prostate-specific antigen blood levels, ² primary treatments include surgery, radiation, and androgen deprivation. PCa cells can metastasize to various organs, including the lymph nodes and bone. The 5-year survival rate for localized PCa patients is nearly 100%, whereas that of metastatic PCa patients is extremely poor, approximately 40%. ³

The “cancer stem cell (CSCs) hypothesis” has been proposed as a potential mechanism of tumor metastasis and recurrence. CSCs, similar to normal stem cells (eg, hematopoietic stem cells [HSCs] and mesenchymal stem cells [MSCs]), have the capacity for self-renewal and differentiation. They can reconstruct the tumor microenvironment (called “niche”) to promote their survival, resulting in recurrence and metastasis. ⁴ Recent research suggests that CSCs may be a new therapeutic target for cancer treatment; for example, Fukasawa et al showed that suppressing the function of cyclin-dependent kinase 8 in glioma stem cells suppresses tumorigenicity, drastically improving survival rates. ⁵ Therefore, uncovering factor(s) controlling CSC function may help improve treatment outcomes.

Many genetic mutations have been reported in PCa. NKX3-1 loss, structural variants of PTEN, and MYC gain of function are frequently observed in localized PCa, and known to regulate stemness (eg, by inducing reprograming, activating PI3K/Akt signaling).^6-9 These genetic alterations can be accurately detected using next-generation sequencing technology, ¹⁰ which allows not only evaluations at the genomic level but also at the transcriptomic level through RNA sequencing (RNA-seq). However, bulk RNA-seq, which analyzes the total RNA extracted from tumor tissues, provides only an averaged gene expression profile, masking the heterogeneity of cellular subpopulations within the tumor. This cellular heterogeneity, particularly the presence of CSCs, is a critical factor in drug resistance and recurrence, resulting in poor treatment outcomes. ¹¹ Therefore, advanced techniques with single-cell resolution are essential for dissecting the cellular diversity within tumors and identifying CSC-specific mechanisms.

Single-cell RNA-seq (scRNA-seq), a recently developed technique, not only allows investigating gene expression patterns specific to CSCs, but also to determine the types and proportions of cells within tumor tissues, investigate how the expression of specific genes changes with cellular differentiation, and explore cell–cell interactions. ¹² Thus, scRNA-seq could unveil CSC characteristics that conventional methods cannot resolve, and which have been only analyzed in whole tumor tissue. Further, these data could help accelerate the development of new therapeutic drugs targeting CSCs. Although recent studies have reported the existence of prostate CSCs, ¹³ their properties remain unclear; given the lack of comprehensive single-cell analysis of CSCs in PCa patients. Thus, this study aimed to identify new markers of prostate CSCs using bioinformatic methods.

Thioredoxin (coded by TXN) is a ubiquitously expressed oxidoreductase that catalyzes a unique reducing effect by cleaving disulfide bonds resulting from excessive oxidative stress and acts together with NADPH and thioredoxin reductase (TrxR1) to induce a redox reaction.^14,15 TXN plays an important role in some cancers; its increased activity protects cancer cells from oxidative stress, thereby promoting survival.^16-19 However, little is known about its function in prostate CSCs. Here, we reanalyzed available scRNA-seq data of PCa tissues as well as clinical data from the Cancer Genome Atlas (TCGA) database and identified TXN as a potential marker for prostate CSCs and prognosis of PCa patients.

Materials and Methods

Sequencing data used in this study

First, we obtained scRNA-seq data from the Gene Expression Omnibus (GEO) database (GSE181294). ²⁰ Hirz et al collected normal prostate tissues from 4 patients as well as PCa and surrounding normal tissue from 19 untreated patients (46-73 years old; Figure 1A). Then, live cells from enzymatically dissociated tissues, except erythrocytes, were obtained using fluorescence-activated cell sorting and scRNA-seq libraries were prepared using the 10X Chromium technology.

Figure 1.

Identifying and validating Stem^High prostate cancer stem-like cells in the GSE181294 dataset. (A) Summary of the analysis. Expression data of normal prostate and prostate cancer (PCa) tissues were downloaded from the GSE181294 dataset. We obtained 13 405 epithelial cells, including 1156 tumor cells, and identified 462 Stem^High cells and 694 Stem^Low cells. (B) Heatmap showing ssGSEA results. (C) Violin plots of representative stem cell marker genes. (D) Box plot showing the stemness score. (E) Enrichment plot associated with MYC signaling. ^*P < .05, ^***P < .001. Abbreviation: NES, normalized enrichment score.

Second, we analyzed scRNA-seq data (EGAS00001005787) of Tuong et al’s, ²¹ which included PCa and adjacent normal tissue data from 10 patients (50-74 years old). They collected single cells by enzymatic dissociation followed by the Percoll gradient procedure. scRNA-seq libraries were constructed as described above.

Third, we reanalyzed the bulk RNA-seq data provided by Peña-Hernández et al. (GSE131268). ²² To harvest prostate CSCs, they seeded PC3 cells (a human prostate cell line) in a low-attachment plate and cultured them in serum-free medium. Then, total RNA was collected and RNA-seq was performed.

Processing of scRNA-seq data

For analyzing the GSE181294 dataset, scRNA-seq data were imported into the Seurat R package (version 4.3.0). ²³ To avoid erroneous results in subsequent analyses (ie, to remove low-quality data), we filtered out cells with nCount_RNA ⩽ 600 as specified in the original paper, followed by doublet removal as estimated by the doubletFinder_v3 function in the DoubletFinder R package (version 2.0.4). ²⁴ Next, we performed SCTransform, and expression data were integrated using the Harmony R package (version 1.2.1). These approaches aimed to improve downstream analysis and computational efficiency, respectively.^25,26 Uniform manifold approximation and projection (UMAP) was used for dimensionality reduction of the integrated data, and a cluster analysis was performed. Based on the original report, we used the first 15 principal components.

The EGAS00001005787 dataset was analyzed in the same way, except for the following parameters: filtering out low-quality data (nFeature_RNA ⩽ 200, nFeature_RNA ⩾ 2,500, or percent.mt (percentage of mitochondrial genes) ⩾30) and number of principal components (1-50). These parameters are the same as those in the original paper.

Differentially expressed genes were identified by Wilcoxon’s rank-sum test with Benjamini–Hochberg adjustment using the presto R package (version 1.0.0), which provides fast calculations. Gene ontology (GO) analysis and GSEA were performed using the clusterProfiler R package (version 4.10.1). ²⁷ To perform GO analysis, we used the enrichGO function with the following parameters: ont = “MF,” pAdjustMethod = “BH,” pvalueCutoff = 0.01. For performing GSEA, the GSEA function with the following parameters was used: minGSSize = 5, maxGSSize = 500, eps = 0, pAdjustMethod = “BH,” pvalueCutoff = 0.05. A heatmap was drawn using the pheatmap (version 1.0.12) or ComplexHeatmap (version 2.18.0) R packages, and a Venn diagram was generated using the VennDiagram R package (version 1.7.3).^28-30 Spearman’s correlation coefficients were calculated for correlation analyses.

Classifications based on stemness signatures

To identify cells with high stemness, single-sample GSEA (ssGSEA) was conducted using the GSVA R package (version 1.50.5). ³¹ The results were Z-scored for clustering and heatmap generation using the ComplexHeatmap R package. To divide cells into 2 groups using k-means partitioning, the value of column_km was set to 2. We used stem cell-related gene sets deposited in the Molecular Signatures Database (MSigDB). In addition, we calculated a stemness score based on a previously reported gene set,^32,33 and statistical significance was determined using Wilcoxon’s rank-sum test.

Trajectory analysis

A trajectory analysis was performed using the monocle 3 R package (version 1.3.7). ³⁴ The expression matrix was imported and preprocessed (num_dim = 8 was specified based on the result of the plot_pc_variance_explained function). For batch removal, the align_cds function was used. Next, nonlinear dimensionality reduction was performed using UMAP and clustered cells. The root node was selected based on the stemness score and the cells were ordered in pseudotime.

Processing of bulk RNA-seq data

Raw fastq files were downloaded from the GEO database (GSE131268) using the sra-toolkit (version 3.0.3). A quality check and trimming were performed using FastQC (version 0.12.1) and Trim Galore (version 0.6.10), respectively. Clean reads were then aligned against a human reference sequence (GRCh38.p13) using STAR (version 2.7.10b), and the transcripts per million (TPM) were quantified using RSEM (version 1.3.1). A heatmap was generated based on the log₂(TPM + 1) value.

Survival analysis

mRNA expression and clinical data for TCGA-PRAD were obtained from the cBioPortal database. We extracted patients with high stemness tumors as described above. mRNA expression was normalized using a Z-score, and survival analysis was performed based on TXN expression levels. The cutoff value was set to the mean TXN expression, and a log-rank test was used for statistical analysis. A Kaplan–Meier curve was generated using the survminer R package (version 0.5.0).

Statistical analysis

Analyses were performed using R (version 4.3.2). In this study, statistical analysis was conducted using a nonparametric test, which does not require the assumption that data follows a specific distribution; the statistical methods are described above. A P-value of <.05 was considered to indicate statistical significance.

Results

Identifying prostate cancer stem-like cells using two independent scRNA-seq datasets

Although several reports have examined prostate CSCs,^22,35,36 no detailed transcriptome analyses used multiple scRNA-seq data from human PCa tissues. In the present study, we reanalyzed publicly available scRNA-seq data from 2 independent groups (GSE181294 and EGAS00001005787) to identify genes specifically upregulated in CSCs.

Defining cancer cells with high stemness

Using the Seurat R package, ²³ we extracted 1156 and 6139 cancer cells from 13 405 and 10 551 epithelial cells in GSE181294 and EGAS00001005787 datasets, respectively (Figures 1A and 2A; Supplemental Figure S1A and B). To identify cancer cells with high stemness, an enrichment analysis at a single-cell resolution was performed. We used 6 different gene sets associated with stem/progenitor cells and classified cancer cells into 2 populations based on the ssGSEA score. These included cell populations with a high and low score, designated as stemness-high prostate cancer stem-like cells (Stem^High cells) and stemness-low PCa cells (Stem^Low cells), respectively (Figures 1B and 2B).

Figure 2.

Identifying and validating Stem^High prostate cancer stem-like cells in the EGAS00001005787 dataset. (A) Summary of the analysis. Expression data of normal prostate and prostate cancer (PCa) tissues were downloaded from the EGAS00001005787 dataset. We obtained 10 551 epithelial cells, including 6139 tumor cells, and identified 1783 Stem^High cells and 4356 Stem^Low cells. (B) Heatmap showing the ssGSEA results. (C) Violin plots of representative stem cell marker genes. (D) Box plot showing the stemness score. (E) Enrichment plot associated with MYC signaling. ^***P < .001. Abbreviation: NES, normalized enrichment score.

Characterization of Stem^High cells

In Stem^High cells, CD44 and ITGA2 expression, representative CSC markers, was significantly increased ³⁷ (Figures 1C and 2C). Further, the stemness score used by Liu et al ³² was significantly higher in the Stem^High group (Figures 1D and 2D). MYC signaling is important for stemness maintenance in prostate CSCs ³⁸ ; a gene set associated with MYC signaling was significantly enriched in the Stem^High group (Figures 1E and 2E). Accordingly, we could identify stem-like cells (Stem^High cells) from 2 independent scRNA-seq datasets.

TXN is a putative marker for prostate CSCs

Characterization of genes commonly upregulated in the 2 datasets

Differential gene expression analysis revealed that 491 and 745 upregulated genes in Stem^High cells in the GSE181294 and EGAS00001005787 datasets, respectively, of which 158 were commonly upregulated (Figure 3A). GO analysis was performed to determine the biological functions of these 158 genes; interestingly, they were primarily associated with redox activity (Figure 3B).

Figure 3.

Relationship between stemness and redox activity. (A) Venn diagram representing upregulated genes (Padj < 1 × 10, ⁵ logFC > 0.25, and pct_in > 30) in each dataset. A total of 158 genes were co-upregulated in both datasets. (B) Dot plot of GO analysis results. (C) Enrichment plot. Gene sets #1–#4 were “GOCC_OXIDOREDUCTASE_COMPLEX” (P < .001, NES = 1.37), “GOMF_OXIDOREDUCTASE_ACTIVITY_ACTING_ON_NAD_P_H” (P = .030, NES = 1.34), “GOMF_OXIDOREDUCTASE_ACTIVITY_ACTING_ON_NAD_P_H_QUINONE_OR_SIMILAR_COMPOUND_AS_ACCEPTOR” (P = .035, NES = 1.38), and “GOMF_OXIDOREDUCTION_DRIVEN_ACTIVE_TRANSMEMBRANE_TRANSPORTER_ACTIVITY” (P = .015, NES = 1.39), respectively. (D) Scatter plot showing the correlation between oxidoreductase activity (y-axis) and stemness score (x-axis) with a 95% confidence interval and a regression curve (blue dotted line). R represents the Spearman’s correlation coefficient. (E) Violin plots of TXN expression in each dataset. ^***P < .001. Abbreviation: NES, normalized enrichment score.

Relationship between stemness and oxidoreductase activity

Next, we performed a GSEA using the GSE181294 dataset and identified gene sets associated with oxidoreductase activity significantly enriched in Stem^High cells (Figure 3C). In addition, we observed a significant positive correlation between the stemness score and oxidoreductase activity in all cancer cells (Stem^High and Stem^Low cells; Figure 3D).

Increased expression of TXN in Stem^High cells

Although several proteins involved in redox reactions have been identified, ³⁹ TXN, which encodes the oxidoreductase enzyme thioredoxin, was highly expressed in Stem^High cells among the 158 co-upregulated genes shown in Figure 3A (Figure 3E). Moreover, TXN expression was significantly higher in tumor cells than in normal luminal cells, considered the origin of PCa cells ⁴⁰ (Supplemental Figure S2A).

Activation of NRF2 signaling in Stem^High cells

Previous reports have shown that NRF2 signaling promotes TXN transcription in response to oxidative stress ⁴¹ ; Stem^High cells exhibited a significant enrichment (GSE181294) or enriched tendency (EGAS00001005787) in the NRF2 signaling-related gene set (Supplemental Figures S2B and C).

Changes in TXN expression during differentiation

To determine how TXN expression changes during cell differentiation, we performed a trajectory analysis using the monocle 3 R package ³⁴ (Figure 4A-C). The expression of ALDH1A1 and ABCG2, representative CSC markers, markedly decreased with pseudotime progression (Supplemental Figure S3 and Figure 4D). Similarly, TXN expression decreased over pseudotime, suggesting that TXN expression decreases with CSC differentiation (Figure 4E).

Figure 4.

TXN as putative CSC marker. (A-C) UMAP plots of the results of the trajectory analysis. Each point is colored according to (A) group, (B) stemness score, and (C) pseudotime. The trajectory graph is represented by a gray line. (D and E) Pseudotime kinetics of (D) representative CSC markers and (E) TXN.

The results from Figures 1 to 4 indicate that TXN is a novel prostate CSC biomarker.

TXN is associated with MYC signaling in prostate CSCs

Characterization of high TXN-expressing Stem^High cells

To identify the intracellular signals linked to TXN in prostate CSCs, we divided Stem^High cells into 2 groups based on TXN expression (Figure 5A). Among Stem^High cells, cells with higher TXN expression (TXN^High Stem^High cells) exhibited significantly higher stemness scores (Figure 5B). Subsequently, we performed GSEA between TXN^High Stem^High and TXN^Low Stem^High cells. Interestingly, MYC signaling was significantly enriched in the former (Figure 5C). We also observed a significant positive correlation between oxidoreductase activity and MYC signaling in all Stem^High cells (Figure 5D).

Figure 5.

Relationship between TXN and MYC signaling in Stem^High cells. (A) Summary of the analysis. Stem^High tumor cells extracted from each scRNA-seq data were divided into 2 groups based on the mean expression level of TXN (TXN^High Stem^High and TXN^Low Stem^High cells). (B) Box plots of the stemness scores of (left) GSE181294 (TXN^Low Stem^High: n = 204; TXN^High Stem^High: n = 258) and (right) EGAS00001005787 (TXN^Low Stem^High: n = 591; TXN^High Stem^High: n = 1192). (C) Dot plots of GSEA results. (D) Scatter plots showing the correlation between oxidoreductase activity (y-axis) and MYC signaling (x-axis) with a 95% confidence interval and a regression curve (blue dotted line). R represents the Spearman’s correlation coefficient. (E) Heatmap showing TXN and MYC expression in cancer stem cells (CSC) and prostate cancer cells (PCa). Color represents the log₂-transformed transcripts per million (TPM) value. ^***P < .001.

TXN expression in PC3 cell-derived prostate CSCs

Peña-Hernández et al. performed RNA-seq of PC3 human PCa cells and prostate CSCs obtained from serum-free culture of PC3 cells. ²² Reanalyzing these data indicated that not only MYC but also TXN expression was markedly upregulated in CSCs (Figure 5E). These results suggest that TXN is probably involved in the functional regulation of CSCs via MYC signaling.

TXN is related to stemness-related signaling

In addition to MYC signaling, GSEA showed significant enrichment of some gene sets related to stemness regulation in TXN^High Stem^High cells.

Proteasome activity

Previous reports have shown high proteasome activity in several stem/progenitor cells (eg, embryonic stem (ES) cells, MSCs and neural progenitor cells) and that the proteasome is extremely important for maintaining stem cells function.^42-44 We found that TXN^High Stem^High cells had high proteasome activity, suggesting that proteostasis is enhanced in TXN^High Stem^High cells (Fig S4A).

Nonsense-mediated mRNA decay

A mechanism for maintaining cellular homeostasis is nonsense-mediated mRNA decay (NMD), which induces the degradation of abnormal mRNA with nonsense mutations. ⁴⁵ NMD is important for maintaining stem cell properties, and its suppression is necessary for stem cell differentiation. ⁴⁶ GSEA showed increased NMD activity in TXN^High Stem^High cells (Fig S4B). As NMD plays a pivotal role in preventing the production of abnormal proteins, NMD activation may also contribute to proteostasis.

Translation and metabolism

Recent research has shown that refeeding after fasting enhances the stemness of intestinal stem cells due to increased mammalian/mechanistic target of rapamycin 1 (mTORC1)-mediated protein synthesis and polyamine metabolism. ⁴⁷ GSEA revealed that both translation and polyamine metabolism were enhanced in TXN^High Stem^High cells (Figs S4C and S4D), suggesting that pathways other than MYC are involved in the regulation of prostate CSC function by TXN.

TXN^High Stem^High cells may be involved in major clinical issues

Metastasis

As CSCs are involved in metastasis, we investigated whether TXN is related to their metastatic ability. GSEA showed that genes associated with metastasis were significantly enriched in TXN^High Stem^High cells (Figure 6A). This suggests that reducing TXN expression can suppress PCa metastasis.

Figure 6.

Relationship between TXN and clinical relevance. (A) Enrichment plots associated with tumor metastasis (“ALONSO_METASTASIS_UP”). (B) Enrichment plots associated with the ABC transporter (“REACTOME_ABC_TRANSPORTER_DISORDERS”). (C) Enrichment plots associated with DNA repair (“HALLMARK_DNA_REPAIR”). TXN^Low Stem^High: n = 204; TXN^High Stem^High: n = 258 (GSE181294) and TXN^Low Stem^High: n = 591; TXN^High Stem^High: n = 1192 (EGAS00001005787).

Drug efflux

CSCs have a high ability to efflux drugs, which limits the development of drugs targeting them; in particular, ATP-binding cassette (ABC) transporters are involved in drug excretion from cells. ⁴⁸ In this dataset, a gene set related to ABC transporter function was significantly enriched in TXN^High Stem^High cells (Figure 6B), indicating that TXN may be involved in the drug excretion ability of CSCs and the acquisition of drug resistance.

DNA repair

CSCs have a high capacity for repairing damaged DNA, which can be due to activation of DNA repair mechanisms by homologous recombination, the improvement of splicing fidelity, strong activation of cell cycle checkpoints, etc. ⁴⁹ Interestingly, TXN^High Stem^High cells showed activated DNA repair-related signaling (Figure 6C), suggesting that targeting TXN in CSCs may solve clinical issues such as drug resistance, metastasis.

TXN may serve as prognostic marker for patients with prostate cancer harboring high stemness tumor tissue

Classification of prostate cancer patients based on the stemness

Finally, we performed a survival analysis to determine whether TXN is a clinically useful biomarker. We defined Stem^High and Stem^Low patients by applying ssGSEA to PCa patients in the TCGA database (TCGA-PRAD), further subdividing the former group into 2 cohorts based on TXN expression (TXN^High Stem^High patients and TXN^Low Stem^High patients; Figure 7A and B). Our results showed a significantly higher stemness score in Stem^High patients compared with Stem^Low patients, indicating that ssGSEA could successfully identify PCa patients with high stemness (Figure 7C).

Figure 7.

Relationship between TXN and prognosis in Stem^High prostate cancer patients. (A and B) Summary of the analysis (A). Prostate cancer patients in the TCGA-PRAD cohort were divided into the Stem^High (n = 202) and Stem^Low (n = 291) categories based on the ssGSEA score (B), and the former subdivided into 2 groups, designated TXN^High Stem^High and TXN^Low Stem^High, based on TXN expression levels. (C) Box plot showing the stemness score. (D) Kaplan–Meier curve for disease-free survival (DFS) comparing TXN^High Stem^High (n = 64) and TXN^Low Stem^High (n = 74) patients. (E) Kaplan–Meier curve for progression-free survival (PFS) comparing TXN^High Stem^High (n = 104) and TXN^Low Stem^High (n = 98) patients. ^***P < .001.

Relationship between TXN expression and prognosis

Disease-free survival (DFS) was significantly shorter in TXN^High Stem^High patients than in TXN^Low Stem^High patients; a similar result was observed for progression-free survival (PFS; Figure 7D and E). In contrast, there was no significant difference between TXN expression and prognosis in Stem^Low patients (Supplemental Figures S5A and B).

Taken together, these results indicate that TXN is a promising marker of prostate CSCs and a potentially important factor in determining prognosis.

Discussion

Study uniqueness

To date, many studies have characterized CSCs and developed novel therapeutic strategies targeting them.^5,50-54 Despite several reports describing transcriptome analyses of prostate CSCs using PCa tissues and cell lines,^22,55,56 to the best of our knowledge, this is the first study to explore novel markers of prostate CSCs integrating scRNA-seq data from multiple human PCa tissues. Using 2 independent scRNA-seq datasets allowed cross-validation and reduced bias.

We also refined the CSC estimation method. Numerous in vivo and in vitro studies used a limited number of stem cell markers to identify CSCs.^57,58 However, the characteristics of CSCs differ depending on the markers used. ³⁷ In addition, scRNA-seq cannot appropriately detect low-expression genes, making it challenging to use relatively low-expression markers for CSC identification. Therefore, an approach that analyzes the whole transcriptome may more accurately identify CSCs than a method that uses a very limited number of CSC markers. ⁵⁹ In this study, we used ssGSEA to score multiple sets of stem/progenitor cell-related genes and used non-hierarchical clustering with k-means to classify them into Stem^High and Stem^Low cells. Using this method, we could extract high stemness cells without relying on specific genes. Our success distinguishing stem and differentiated cells is supported by the expression of representative marker genes and the stemness score.

TXN function in adult stem cells

TXN has been reported to contribute to MSC differentiation and proliferation. Mechanistically, TXN regulates osteoblast differentiation and myogenesis via autophagy and PI3K/Akt signaling, respectively.^60,61 Furthermore, Txn deletion in the HSCs of Txn-knockout mice not only induced the activation of p53 signaling and cellular senescence but also reduced the ability to reconstitute hematopoiesis. ⁶² Moreover, Tian et al demonstrated that treatment of neural stem cells (NSCs) with recombinant TXN promotes NSC proliferation and differentiation, which is mediated by ERK1/2 signaling. ⁶³ Notably, these intracellular signaling pathways are also implicated in the regulation of CSC function,^9,64,65 suggesting that TXN plays a crucial role in maintaining the function of prostate CSCs as well as (normal) stem cells. In this study, we mainly analyzed the relationship between TXN and MYC signaling; however, as these signals are also related to MYC signaling,^66,67 future research should aim to investigate how these signals affect the TNX/MYC-mediated regulation of the stemness of prostate CSCs.

Stem cell regulation by TXN/TrxR1 and MYC

We observed MYC signaling activation in TXN^High Stem^High cells (Figure 5C). To the best of our knowledge, no reports showed a direct TXN-MYC interaction in the field of oncology, but Raninga et al found that MYC-high high-grade serous ovarian carcinoma (HGSOC) tissues exhibit increased TrxR1 activity. ⁶⁸ As MYC is involved in stemness regulation in many CSC types,^5,67,69 TXN-mediated maintenance of stemness may occur through this pathway. In addition, the TrxR1 inhibitor auranofin inhibited proliferation and induced apoptosis of MYC-high HGSOC cells. ⁶⁸

GSEA results showed a positive correlation between oxidoreductase activity and MYC signaling (Figure 5D), which was supported by a previous report showing that reactive oxygen species (ROS) activate MYC through ERK signaling. ⁷⁰ Furthermore, TXN expression reportedly increases in response to hydrogen peroxide stimulation. ⁷¹ Further research is needed to determine how TXN is related to MYC signaling.

NRF2 signaling as an upstream factor of TXN

As mentioned previously, NRF2 signaling increases the expression of TXN. ⁴¹ Previous reports have shown that activation of NRF2 signaling enhances the self-renewal ability and suppresses differentiation of human bone marrow-derived MSCs by increasing sirtuin 1 expression via p53 expression suppression. ⁷² In addition, NRF2 activation suppresses ROS accumulation and maintains the self-renewal activity of breast CSCs through the FoxO3/BMI-1 axis. ⁷³ Furthermore, NRF2 increases the expression of the ABC transporter ABCF2, possibly related to the acquisition of drug resistance. ⁷⁴ Based on our GSEA results showing enhanced NRF2 signaling in prostate CSCs (Supplemental Figure S2), NRF2 inhibitors can induce differentiation of CSCs and suppress the expression of ABC transporters, thereby reducing the CSC pool and inhibiting the acquisition of drug resistance, respectively. ⁷⁵

Alternative mechanisms of CSC maintenance

Although the present study focused on stemness regulation via MYC signaling, other signaling pathways are involved in stem cell maintenance. Wang et al reported that TXN increases stemness in a redox-independent manner by stabilizing BTB and CNC homology 1 (BACH1) and activating mTOR signaling in hepatocellular carcinoma CSCs. ⁷⁶

Our enrichment analysis showed that TXN^High Stem^High cells had high proteasome activity (Fig S4A). Intriguingly, Seo et al reported that the combination of TrxR1 knockdown and a proteasome inhibitor, but not TrxR1 knockdown alone, induce programed cell death in breast cancer. ⁷⁷ Mechanistically, misfolding of proteins containing thiols causes cytotoxicity. CSCs are resistant to apoptosis, ⁷⁸ but dual inhibition of TrxR1/TXN and the proteasome may lead to CSCs elimination via proteotoxic stress.

In addition to genome stability (Figure 6C), TXN^High Stem^High cells had activated NMD (Fig S4B). As ES cells rely on high NMD activity to regulate Wnt signaling and differentiation (and Wnt signaling is critical for CSC regulation), the enhanced stemness observed in TXN^High Stem^High cells may be attributed to NMD’s modulation. ⁴⁶

Moreover, stemness is influenced by nutritional status. For instance, Taya et al reported that valine and cysteine, which form the active sites of TXN, are important for maintaining function of mouse and human HSCs. ⁷⁹ Furthermore, the mechanism underlying polyamine-mediated stem cell regulation has been recently clarified. ⁴⁷ Interestingly, difluoromethylornithine, an ornithine decarboxylase inhibitor, not only depletes polyamines but also reduces TXN levels.^80,81 Accordingly, in addition to chemotherapy, dietary restriction may be an effective strategy for eliminating prostate CSCs.

Clinical implications of TXN in Stem^High cells/patients

TCGA dataset analysis revealed that patients with high TXN expression in Stem^High patient group had significantly shorter PFS and DFS (Figure 7). However, in Stem^Low patient group, TXN expression did not correlate with prognosis (Supplemental Figure S5). One possible reason for the poor prognosis of TXN^High Stem^High patients is that TXN contributes to the treatment resistance of CSCs. As shown in Figure 6, TXN has been implicated in metastasis, drug efflux, and DNA repair, phenomena that make cancer treatment difficult. Therefore, future research should investigate the relationship between clinical events such as recurrence and/or metastasis and TXN expression. Overall, we consider TXN a promising predictive and prognostic marker for predicting prostate CSCs; however, prior to clinical use, accurately measuring TXN expression in CSCs by single-cell techniques, such as histological analysis or flow cytometry is needed.

Limitations of the current study

First, the sequencing data used in this study is limited. To evaluate the reliability of the data obtained in this study, we analyzed tumor tissues from several patients, but considering the intra-tumor and inter-tumor complexity, the number of cases remains insufficient. Therefore, it is important to recruit more patients with a wider range of backgrounds (eg, sex, race, age).

Second, technical issues specific to scRNA-seq. For example, we used the Seurat R package to perform scRNA-seq analysis. This package is a widely used analysis tool, and scran and python-based Scanpy have recently been used for this purpose.^82,83 Although excellent results have been obtained, they can vary depending on the tool used, even when analyzing the same data. ⁸⁴ In addition, scRNA-seq can hardly detect low expressed genes, and given the processes used to remove batch effects and doublets, important differentially expressed genes and intracellular signals may be overlooked.

Third, sample preparation. Although not limited to scRNA-seq, there is the concern that solid tumor processing to single-cell suspension and the time lag between sampling and sequencing can drastically change the transcriptome. ⁵⁹ In addition, using trajectory analysis, this study showed that TXN expression decreases with differentiation. However, the scRNA-seq data used in this study was not sampled over time, it only evaluates the transcriptome at a single time point, so the results should be interpreted with caution.

Finally, and most importantly, the present results were obtained using existing databases, being only ‘estimates’; in other words, whether TXN is a marker for prostate CSCs useful for predicting prognosis needs to be validated in vivo and in vitro. Future research in cultured cells and human specimens is needed to clarify the role of TXN in prostate CSCs from both a basic and clinical perspective.

Conclusions

Based on the present findings, TXN likely inhibits CSC differentiation. Therefore, one possible treatment strategy would be to administer a drug that suppresses TXN system to induce CSC differentiation before conventional anticancer drugs to shrink the tumor. According to the results of the survival analysis, stratifying PCa patients with high stemness as high-risk patients is crucial to maximize the efficacy of TXN-targeted therapies.

Supplemental Material

sj-pdf-1-cix-10.1177_11769351251319872 – Supplemental material for Integrated Bioinformatic Analyses Reveal Thioredoxin as a Putative Marker of Cancer Stem Cells and Prognosis in Prostate Cancer

Supplemental material, sj-pdf-1-cix-10.1177_11769351251319872 for Integrated Bioinformatic Analyses Reveal Thioredoxin as a Putative Marker of Cancer Stem Cells and Prognosis in Prostate Cancer by Shigeru Sugiki, Tetsuhiro Horie, Kenshiro Kunii, Takuya Sakamoto, Yuka Nakamura, Ippei Chikazawa, Nobuyo Morita, Yasuhito Ishigaki and Katsuhito Miyazawa in Cancer Informatics