A conserved eIF1A⁺ luminal cell-centered hypoxic and “cold” tumor microenvironment promotes pan-subtype prostate cancer progression

Summary

Prostate cancer (PCa) is a malignancy with high heterogeneity arising from tumor microenvironment and histological subtypes. Identifying conserved progression drivers within such heterogeneity is essential for improving clinical outcomes. Using imaging mass cytometry, this study analyzes 38 proteins across paracancerous tissue and four histological subtypes: low-grade prostate acinar adenocarcinoma (LgPAC), high-grade PAC (HgPAC), intraductal carcinoma (IDC), and ductal adenocarcinoma (DAC). Results reveal that eIF1A is overexpressed in high-risk subtypes including HgPAC, IDC, and DAC and correlates with poor prognosis. In luminal cells, EIF1A knockdown and the translation inhibitor homoharringtonine (HHT) both suppress HIF-1α translation and tumor growth, while promoting infiltration of anticancer immune cells including PD-1⁻ T cells and CD163⁻ macrophages. Clinically, neoadjuvant HHT combined with androgen deprivation therapy reduces hypoxia and enhances immune cell infiltration, as shown by single-cell RNA sequencing. Collectively, this work defines conserved molecular features across PCa subtypes, providing promising insights for clinical management. This study was registered at Clinicaltrials.gov (NCT06834321).

Graphical abstract

Highlights

• •In situ proteomic profiling delineates spatial landscapes in PCa histological subtypes
• •Activation of eIF1A-translation-HIF-1α axis is conserved across high-risk PCa subtypes
• •eIF1A⁺ luminal cells reside in an immunologically “cold” microenvironment
• •Targeting eIF1A and translation holds potential for clinical intervention in PCa

Cheng et al. describe the conserved features across different histological subtypes of prostate cancer at proteomic level using imaging mass cytometry. Spatial analyses indicate that eIF1A⁺ luminal cells are associated with hypoxia response and immunologically “cold” microenvironment. Targeting eIF1A and translation holds potential for clinical intervention in prostate cancer.

Introduction

Prostate cancer (PCa) is the most frequently diagnosed malignancy and the second most common cause of cancer-related death in men worldwide.¹ Although androgen deprivation therapy (ADT), androgen receptor antagonist, and chemotherapy have delayed clinical progression to some degree, treatment options remain limited for advanced-stage PCa. Immune checkpoint inhibitors (ICIs) exhibit significant efficacy against some tumors, but owing to the immunosuppressive characteristics,^2,3 ICI efficacy is unsatisfactory in PCa.^4,5 Hence, there is an urgent need to explore therapeutic directions for advanced PCa.

PCa is a malignancy with high heterogeneity arising from both the tumor microenvironment (TME) and histological subtypes. To achieve superior clinical efficacy, it is imperative to identify conserved progression drivers within such heterogeneity in PCa, thereby enabling the development of more consistent and effective therapeutic strategies. PCa histological subtypes include prostate acinar adenocarcinoma (PAC), intraductal carcinoma (IDC), and ductal adenocarcinoma (DAC), which have different clinical characteristics. IDC is usually associated with higher Gleason scores, larger tumor volumes, neuroendocrine differentiation, and faster disease progression^6,7,8 and is less responsive to ADT, ultimately leading to lower recurrence-free and overall survival (OS) rates.^9,10,11 DAC is a subtype frequently associated with biochemical recurrence and distant metastasis.^12,13,14,15 Unfortunately, technical limitations in isolating relevant tissues from PCa samples have restricted the characterization of rare subtypes (IDC and DAC) to in situ analyses. Hence, understanding the spatial pattern of distinct histological entities in advanced PCa is crucial for revealing the common underlying progression mechanisms, thus providing insights for clinical therapeutic strategies.

Imaging mass cytometry (IMC) is a tissue section-based method for in situ cellular analysis using metal-labeled antibodies to provide information on cellular proteomic phenotypes and spatial distribution features at single-cell resolution.¹⁶ In this study, by applying IMC to paracancerous tissues (PCTs) and PCa tissues covering different histological subtypes, we aimed to delineate the consistent molecular features across high-risk subtypes and provide transformative insights for clinical management.

Results

IMC analysis of paracancerous prostate and PCa tissues

First, we investigated the heterogeneity of the proteomic microenvironment in prostate tissues through IMC. A tissue microarray (TMA) was generated from tumor and PCTs from 47 PCa patients, with corresponding follow-up information up to at least April 2025 (Table S1). Resultantly, a total of 71 regions of interest (ROIs), including 25 paracancerous, 17 low-grade PAC (LgPAC; PAC with a Gleason score of 3 + 3), 7 HgPAC (PAC with a Gleason score ≥3 + 4), 11 IDC, and 11 DAC ROIs of high quality were used for IMC analyses. In this study, we assigned HgPAC, IDC, and DAC to a high-risk, while LgPAC to a low-risk PCa group. The histological type of each ROI was determined through a combination of hematoxylin-eosin staining and immunohistochemical staining for CKH and p63 (Figure S1A).

The TMA was stained with 38 well-designed metal-labeled antibodies against established cell type markers and key functional markers (Figures S1B and S1C; Table S2). We then performed IMC analysis and quantified marker expression in the resulting highly multiplexed images at single-cell resolution, which generated a total of 345,233 cells and formed the “IMC cohort” (Figure 1A). Through dimensionality reduction and clustering analysis on the basis of protein expression profiles, we identified distinct cell subtypes. The marker panCK was used to identify 156,404 epithelial cells, which were classified as basal epithelia ([BE], CK5^hi) and luminal epithelia ([LE], CK5^lo). The other microenvironment cells included 14,995 T cells (CD3^hi), 2,548 B cells (CD20^hi), 4,670 CD163^- macrophages (CD68^hi and CD163^lo), 12,565 CD163⁺ macrophages (CD68^hi and CD163^hi), 8,767 other antigen-presenting cells (APCs, HLA-DR^int), 4,060 other immune cells (CD45^int and low expression of other immune markers), 23,537 endothelial cells (CD31^hi), 50,966 fibroblasts (collagen-I^hi and α-SMA^lo), 20,851 myofibroblasts (collagen-I^hi and α-SMA^hi), 18,089 smooth muscle cells ([SMCs], collagen-I^lo and α-SMA^hi) and 27,781 other stromal cells (undefined and clustered together with stromal cells) (Figures 1B and S1D). LE constituted the predominant component within the TMA, accounting for 38.5% of the total population (Figures 1C–1E).

Figure 1.

IMC analysis of paracancerous prostate and PCa tissues

(A) Flowchart of IMC analysis.

(B) Scaled protein expression levels in different cell types.

(C) The distribution of different cell types among 71 IMC ROIs.

(D) Representative IMC images showing the staining of cell type-specific proteins and the distribution of different types of cells in PCT, LgPAC, HgPAC, IDC, and DAC. Scale bars, 200 μm.

(E) Boxplots showing the proportion of each type of cell. The data points represent individual ROIs, the boxes represent the interquartile range (IQR) with the median line inside, and the whiskers show the values within 1.5×IQR.

(F–K) Violin plots comparing the proportions of LE (F), BE (G), lymphoid cells (H; T and B cells), myeloid cells (I; macrophages and other APCs), endothelial cells (J), and stromal cells (K; fibroblasts, myofibroblasts, SMCs, and other stroma) between different types of tissues (PCT, PCa, low-risk PCa, high-risk PCa, LgPAC, HgPAC, IDC, and DAC). The data points represent individual ROIs; the width of violins corresponds to the proportion of data points at that size, with the median line inside (two-sided Wilcoxon tests).

We observed significant enrichment of LE in PCa ROIs, whereas BE were predominantly enriched in PCT, which is consistent with the pathological features of PCa (Figures 1F and 1G). However, the proportions of LE and BE were not significantly associated with different grades or subtypes of PCa (Figures 1F and 1G). Furthermore, across PCT and PCa ROIs with varying grades or histological subtypes, no significant differences were detected in the proportions of lymphocytes, myeloid cells, endothelial cells or stromal cells, except for slightly greater lymphocyte infiltration in IDC than in LgPAC (Figures 1H–1K). In summary, the abundance of LE is insufficient to predict progression in PCa, highlighting the necessity for further annotations to delineate more precise functional subtypes at the proteomic level. On the other hand, as a type of “cold” tumor, the differences in the immune microenvironment among PCa subtypes may be mediated by spatial heterogeneity in cellular distribution rather than variations in infiltration levels.

eIF1A⁺ LE is associated with progression and a poor prognosis of PCa

To identify crucial epithelial subtypes, we assessed the relationships between the expression of key functional markers in epithelia and clinical features. Compared with the well-established PCa procancer factors, such as androgen receptor, prostate-specific membrane antigen, and breast cancer 1 (BRCA1), eukaryotic initiation factor 1A (eIF1A) was significantly associated with more and the most progression-linked features (including greater tumor quantification, greater pathological grade, progression stage, and periphery invasion), advanced histological types (HgPAC, IDC and DAC), and poorer OS and progression-free intervals (PFIs) in PCa patients (Figures 2A–2C). The Cancer Genome Atlas (TCGA) database also revealed that EIF1A (the key gene encoding eIF1A) expression was significantly higher in PCa than in PCT and was higher in patients with greater Gleason scores (Figures S2A and 2D). Moreover, EIF1A expression also suggested poorer OS and relapse-free survival according to the TCGA database and a large-scale PCa cohort (GEO: GSE116918, n = 248) (Figures 2E and S2B). In contrast, the expression of previously identified PCa-associated genes showed relative inferior predictive value for poor prognosis, suggesting that eIF1A is a promising target that may participate in facilitating PCa progression (Figure S2B).

Figure 2.

eIF1A⁺ LE is associated with PCa progression and poor prognosis

(A) Dot plot showing the associations between the functional marker expression levels of epithelial cells and clinical features. Size of the dots corresponds to significant levels, while the color indicates in which group the marker expression levels were elevated (corresponding to the group colors shown on the right).

(B–D) (B and C) Kaplan-Meier [K‒M] plots showing the differences in OS (B) and PFI (C) durations between eIF1A-high and eIF1A-low patients (divided by optimal cutoff) in the IMC cohort. (D) Violin plots comparing the expression level of EIF1A between tumors with Gleason scores of 8–10 and 6–7 in the TCGA-PRAD database.

(E) K‒M plots showing the differences in OS between eIF1A-high and eIF1A-low patients in the TCGA-PRAD database.

(F) Representative IMC images showing the staining of panCK, CK5, CK8, and eIF1A (upper), as well as the distribution of different types of epithelial cells (lower) in PCT, LgPAC, HgPAC, IDC, and DAC. Scale bars, 200 μm.

(G and H) K‒M plots showing the differences in OS (G) and PFI (H) durations between patients with high and low proportions of eIF1A⁺ LE in the IMC cohort.

(I) Representative mIF images (upper) and gray value distribution plots (lower) showing the colocalization of CK8 and eIF1A in different types of PCa. Scale bars, 50 μm.

(J and K) Bar plots comparing the mean gray value of eIF1A staining (J) and the proportion of eIF1A⁺ LE in LE (K) between different types of PCa. Data are displayed as mean ± SD, with individual data points representing each sample (n = 30, 17, 17, and 13 for LgPAC, HgPAC, IDC and DAC, respectively; two-sided Wilcoxon tests).

eIF1A participates in the scanning process of translation initiation and is a key rate-limiting step in translation that ensures protein synthesis accuracy. We further categorized the epithelia as eIF1A⁺ LE (CK5^lo/eIF1A^hi), eIF1A⁻ LE (CK5^lo/eIF1A^lo), and BE (CK5^hi) (Figure S2C). We found notable enrichment of eIF1A⁺ LE in the tumors, especially in HgPAC, IDC, and DAC (Figure 2F). As expected, a greater percentage of eIF1A⁺ LE was associated with poorer OS and PFI durations (Figures 2G and 2H). We then constructed a tissue validation cohort consisting of 30 additional cases of PCa to validate the above IMC results, in which 17 cases were histologically confirmed to contain IDC, 13 cases contained DAC, 17 cases contained HgPAC, and 21 cases contained LgPAC. The results of multiplex immunofluorescence (mIF) staining indicated that CK8 and eIF1A were more strongly colocalized in HgPAC, IDC, and DAC than in LgPAC (Figure 2I). Further quantitative comparative analysis revealed that both eIF1A expression and eIF1A⁺ LE abundances were elevated in high-risk subtypes (Figures 2J and 2K). These findings suggest that eIF1A may play an important role in the progression of PCa and that this phenomenon is consistent across different histological subtypes, including PAC, IDC, and DAC.

eIF1A⁺ LE exhibits a hypoxic phenotype

To deeply investigate the characteristics of eIF1A⁺ LE, we compared the expression levels of functional markers among epithelial subtypes. The results demonstrated that, in addition to the PCa-associated proteins, eIF1A⁺ LE exhibited a distinct hypoxic (HIF-1α^hi) and metabolically reprogrammed feature (Figure 3A). Hypoxia is a hallmark of tumor progression and is responsible for angiogenesis, tumor stemness, and alterations in glucose, lipid, and energy metabolism.^17,18 Consistently, we found simultaneous overexpression of VEGFA and CD29 in eIF1A⁺ LE, indicating increased angiogenesis. Additionally, key rate-limiting enzyme of gluconeogenesis (FBP1) was significantly upregulated in eIF1A⁺ LE than in eIF1A⁻ LE. The levels of key enzymes involved in fatty acid synthesis (acetyl-CoA carboxylase [ACC]) and cholesterol synthesis (3-hydroxy-3-methylglutaryl-CoA Reductase [HMGCR]) were also elevated, although the increase was not statistically significant. Elevated expressions of isocitrate dehydrogenase 1 (IDH1) and IDH2 further suggested enhanced tumor energy metabolism (Figure 3A). Thus, we further focused on the hypoxic characteristics and HIF-1α expression of eIF1A⁺ LE. As expected, in eIF1A⁺ LE, HIF-1α was significantly overexpressed in high-risk PCa compared with that in low-risk PCa (Figure 3B). Further, the expression levels of eIF1A and HIF-1α were significantly correlated, implying an underlying regulatory mechanism between them (Figure 3C).

Figure 3.

eIF1A⁺ LE exhibits a hypoxic phenotype

(A) Boxplots comparing the expression levels of functional markers between different types of epithelial cells. The data points represent individual ROIs (n = 71); the expression values represent the median of the expression levels for markers derived from eIF1A + LE, eIF1A- LE, or BE cells across all ROIs; the boxes represent the IQR with the median line inside; and the whiskers show the values within 1.5×IQR (two-sided Wilcoxon tests).

(B) Violin plots comparing the expression level of HIF-1α in eIF1A⁺ LE between different types of tissues. The data points represent individual ROIs; the width of violins corresponds to the proportion of data points at that size, with the median line inside (two-sided Wilcoxon tests).

(C) Correlation analysis between the median expression levels of eIF1A and HIF-1α in eIF1A⁺ LE among the 71 ROIs in the IMC cohort.

(D) GSEA of EIF1A^hi vs. EIF1A^lo epithelial cells in the scRNA-seq validation cohort based on the BIOCARTA, PID, and HALLMARK databases.

(E) Correlation analysis between the AUCell scores of the HIF-1 TF pathway and the EIF pathway in the scRNA-seq validation cohort and TCGA-PRAD database.

(F) Representative IMC images showing the staining of CK8, eIF1A, and HIF-1α in different types of PCa. Scale bar, 200 μm.

(G) Representative mIF images showing the colocalization of CK8, eIF1A, and HIF-1α in different types of PCa (left); bar plots comparing the mean gray value of HIF-1α staining and the proportion of HIF-1α⁺ eIF1A⁺ LE in eIF1A⁺ LE between different types of PCa (right; n = 30, 17, 17, and 13 for LgPAC, HgPAC, IDC, and DAC, respectively; two-sided Wilcoxon tests; also see Figure S5A). Scale bars, 50 μm.

To better validate the hypoxic signature of eIF1A⁺ LE, we utilized a single-cell RNA-seq (scRNA-seq) validation cohort comprising eight treatment-naive PAC samples (Table S3; Figures S3A–S3E). We found that classical carcinogenic signaling pathways (MYC signaling, E2F targets, G2M checkpoint, and epithelial-mesenchymal transition [EMT] pathways) were enriched in EIF1A^hi malignant LE. Notably, translation- (especially translation initiation) and hypoxia-related pathways (HIF1 TF pathway, hypoxia, angiogenesis, and cellular response to starvation) were also activated in the EIF1A^hi epithelia. O-linked glycosylation, which was found to interact with HIF-1α and hypoxia, was also found upregulated in EIF1A^hi epithelia^19,20 (Figure 3D). Expression of EIF1A was significantly positively correlated with the AUCell score of the EIF pathway both in the TCGA database and the scRNA-seq validation cohort, indicating that high expression of eIF1A represented activation of the EIF and translation initiation (Figure S3F). EIF activity was also associated with a poor prognosis of PCa (Figures S4A and S4B). Additionally, EIF1A was significantly positively correlated with almost all the other key EIF genes (Figure S4C). Furthermore, the activation level of the EIF pathway in epithelial cells was significantly positively correlated with that of the HIF-1 TF pathway; both EIF1A expression and EIF activation levels were also positively associated with HIF-1 TF pathway in bulk transcriptomic data from TCGA (Figures 3E, S4D, and S4E). These results reflected a potential regulatory relationship between EIF-induced translation and the HIF-1α-mediated hypoxia response.

Spatially, we observed greater HIF-1α expression in eIF1A⁺ LE in high-risk PCa subtypes (Figures 3F and S3G). mIF staining verified greater mean gray levels of HIF-1α and higher greater HIF-1α⁺/eIF1A⁺ LE abundance in HgPAC, IDC, and DAC than in LgPAC (Figures 3G and S5A). Paired tests in which each sample was used as a unit (e.g., samples containing both IDC and LgPAC components were included in paired comparisons between IDC and LgPAC) yielded consistent results (Figure S5B).

eIF1A⁺ LE-centered “cold” and hypoxic immune microenvironment

When we compared the expression levels of functional markers among epithelial subcell types, we also detected increased expression of PD-L1 in eIF1A⁺ LE. IDO-1, which is the key enzyme of tryptophan metabolism and a factor that mediates hypoxia-induced immunosuppression, was also overexpressed in eIF1A⁺ LE (Figure 3A). Therefore, we further investigated the heterogeneity of the immune microenvironment surrounding eIF1A⁺ LE.

To identify spatial functional units, we defined a total of 20 cellular neighborhoods (CNs) by identifying the 10 nearest spatial neighbors for each cell (Figures 4A and S6A). CN10, named the eIF1A⁺ LE separation CN, was abundant in tumors, especially high-risk PCa, and it was associated with multiple malignant pathological features and a poor prognosis (Figure 4B, Box 1). CN7, which consisted of immune cells and eIF1A⁺ LE, represented an eIF1A⁺ LE immune response CN and was less enriched in IDC than in LgPAC. CN7 was also associated with an unfavorable prognosis, suggesting the absence of tumor-killing CNs in high-risk PCa (Figure 4B, Box 2). CN15 was an eIF1A⁻ LE immune response CN enriched in LgPAC and significantly reduced in IDC and DAC, suggesting that the reduced malignant potential of LgPAC might be attributed to effective immune recognition and elimination of eIF1A⁻ LE (Figure 4B, Box 3). Interestingly, as a neighborhood composed of T, CD163⁻ macrophages, and other immune cells, the immune separation CN1 was highly enriched in IDC, suggesting an immune exclusion feature in IDC (Figure 4B, Box 1). The above results could be spatially illustrated on the IMC staining images (Figure 4C).

Figure 4.

eIF1A⁺ LE-centered “cold” immune microenvironment

(A) Heatmap showing the components of 20 identified CNs.

(B) Dot plot showing the associations between the proportion of CNs and clinical features. Size of the dots corresponds to significant levels, while the color indicates in which group the CN proportions were elevated (corresponding to the group colors shown on the right).

(C) Representative IMC images showing the distributions of CN1, CN7, CN10, and CN15 in different types of PCa. Scale bars, 200 μm.

(D) Dot plot showing the associations between the distance from eIF1A⁺ LE to immune cells and clinical features. Size of the dots corresponds to significant levels, while the color indicates in which group the distances were elevated (corresponding to the group colors shown on the right).

(E) Representative IMC images showing the distance from eIF1A⁺ LE to PD-1^- T cells and CD163⁻ macrophages in different types of PCa. Scale bar, 200 μm.

(F) Representative mIF images showing the colocalization of eIF1A⁺ LE (CK8, eIF1A), T cells (CD3, CD8), and macrophages (CD68) in different types of PCa (n = 30, 17, 17, and 13 for LgPAC, HgPAC, IDC, and DAC, respectively). Scale bars, 200 μm.

(G) The differences in the ratio of CD8⁺ T cells and macrophages in a 350 × 200-μm window among different types of PCa (two-sided Wilcoxon tests).

(H) Representative IMC images showing the locations of HIF-1α⁺ patches, as well as the eIF1A⁺ LE, PD-1⁻ T cells and CD163⁻ macrophages in HIF-1α⁺ patches among different types of PCa. Scale bar, 200 μm. eIF1A⁺ LE, PD-1⁻ T cells, CD163⁻ macrophages, and other cells within HIF-1α⁺ patches were labeled blue, red, orange, and cyan, respectively. The cells outside HIF-1α⁺ patches were labeled gray.

(I) Boxplots comparing the proportions of eIF1A⁺ LE, CD163⁻ macrophages, and PD-1^- T cells inside and outside HIF-1α⁺ patches. The data points represent individual ROIs, the boxes represent the IQR with the median line inside, and the whiskers show the values within 1.5×IQR (two-sided paired Wilcoxon tests).

We subsequently assessed the cellular interactions among the different subtypes. We found that avoidance between immune cells and eIF1A⁺ LE was significantly greater in high-risk PCa than in low-risk PCa (Figure S6B). Therefore, we further evaluated the distances between different immune cell types and eIF1A⁺ LE and assessed the associations of the distances with different clinical features (Figure 4D). The results revealed that for antitumor cells such as PD-1⁻ T cells and CD163⁻ macrophages, their distances from eIF1A⁺ LE were positively correlated with malignant features, including higher prostate specific antigen (PSA) levels, greater pathological grades, progression, and malignant histologic types (IDC and DAC, Figures 4D and S6C). Additionally, their distances from eIF1A⁺ LE did not differ among HgPAC, IDC, and DAC, indicating a unified immune TME among high-risk histological subtypes (Figure 4D). We then demonstrated the distance of eIF1A⁺ LE to key immune cells in different histological subtypes by cell spatial distribution mapping. It was found that HgPAC exhibited relatively sparse immune cell infiltration; in IDC, eIF1A⁺ LE were separated by a BE-stromal barrier, while in DAC, immune cells failed to infiltrate into eIF1A⁺ LE regions (Figure 4E). Collectively, these advanced subtypes all developed “cold” immune microenvironments centered on eIF1A⁺ LE populations.

The above results were further validated by mIF staining detecting eIF1A⁺ LE (CK8⁺eIF1A⁺), CD8⁺ T cells (CD3⁺CD8⁺), and macrophages (CD68⁺). Although our initial IMC analysis based on 1-mm ROIs revealed a non-significant trend (Figures 1H and 1I), a focused analysis within a 350 × 200-μm window centered on eIF1A⁺ LE in the tissue validation cohort showed a significant reduction in the infiltration of CD8⁺ T cells and macrophages in high-risk PCa subtypes (Figures 4F and 4G). The results also demonstrated that HgPAC exhibited the lowest CD8⁺ T cell and macrophage proportion, indicating an immune desert phenotype (Figures 4F and 4G). In particular, T cells and macrophages were enriched mostly in stromal regions and segregated from eIF1A⁺ LE by the basal barrier in IDC (Figures 4E, 4F, and S6D–S6F). Quantitatively, the distance between eIF1A⁺ LE and the BE is significantly shorter in IDC than in other subtypes (Figure S6G). Furthermore, within IDC, eIF1A⁺ LE are located closer to the BE than to PD-1⁻ T cells or CD163⁻ macrophages (Figures S6H and 1I).

Given that eIF1A⁺ LE exhibited a hypoxic phenotype, we further investigated whether the aforementioned “cold” TME existed in a hypoxic niche through HIF-1α⁺ patch analysis. By designating cells with high HIF-1α expression as hub cells, contiguous clusters of at least 10 HIF-1α⁺ cells with cells within their surrounding 10-μm-radius microenvironment were designated as HIF-1α⁺ patches (Figure 4H). In agreement with the hypothesis, eIF1A⁺ LE was significantly enriched in HIF-1α⁺ patches (Figures 4H and 4I). The HIF-1α⁺ patches also presented fewer PD-1⁻ T cells and CD163⁻ macrophages, in accordance with the cellular characteristics of the above “cold” TME (Figures 4H and 4I). These findings demonstrate the emergence of a consistent eIF1A⁺ LE-centered “cold” and hypoxic immune microenvironment across different high-risk subtypes of PCa that cooperates with eIF1A + LE to drive disease progression.

Homoharringtonine suppresses PCa growth by alleviating hypoxia and enhancing the immune response

To determine whether eIF1A influences HIF-1α expression and thereby mediates the hypoxic response, we knocked down EIF1A gene expression in the LNCaP and Myc-CaP cell lines. Additionally, the eIF1A pathway and its associated EIF pathway identified in the present study are critical regulators of mRNA translation. Thus, we also applied homoharringtonine (HHT), the Food and Drug Administration-approved translation inhibitor, to examine the effect of translation on HIF-1α expression. EIF1A knockdown and HHT treatment did not alter the transcription of HIF1A but significantly decreased the HIF-1α protein level (Figures S7A–S7D, 5A, and 5B). This phenomenon was also validated by overexpression of EIF1A (Figures S7E–S7G). Additionally, we treated LNCaP cells with MG132 and demonstrated that EIF1A did not affect the stability and degradation of HIF-1α (Figures 5C and S7H). Polysome profiling analysis revealed that EIF1A knockdown led to an increase in free mRNA and monosome peaks, accompanied by a decrease in polysome peaks, suggesting global translational suppression (Figure 5D). Furthermore, quantitative real-time polymerase chain reaction analysis revealed reduced HIF1A mRNA binding with polysomes upon EIF1A depletion, indicating that HIF-1α translation is regulated by eIF1A (Figure 5E). Through ribosome profiling (Ribo-seq), we observed that the RNA levels of most genes remained unchanged or slightly changed in EIF1A-knockdown cells, while the ribosome-protected fragments (RPFs) of 171 genes, including HIF1A, exhibited significant alterations, indicating a translational activation by eIF1A (Figure 5F). Additionally, gene set enrichment analysis (GSEA) on gene translational efficiency (characterized as RPF/RNA) revealed that hypoxia-related and translation-related pathways were significantly downregulated upon EIF1A knockdown. In particular, the hypoxia pathway was identified as the most prominently inhibited signaling pathway (Figure 5G). Our results indicated that the regulation of eIF1A on HIF-1α might be dependent on the hypoxic environment (Figure 5B), so we further determine whether HIF-1α, as a transcriptional factor, could reversely affect the expression of EIF1A. We interrogated the ChIP-Atlas database to identify potential targets of HIF-1α. Figure S8A displays the model-based analysis of ChIP-seq 2 (MACS2) scores (−10×log₁₀[Q value]) for HIF-1α chromatin immunoprecipitation sequencing peaks within the transcription start site (TSS) ±5 kb region, listing the top 30 genes alongside EIF1A. Notably, in contrast to the established HIF-1α target genes P4HA1, the EIF1A TSS showed a negligible HIF-1α binding signal (Figure S8B). Furthermore, neither overexpression nor knockdown of HIF-1α under hypoxic conditions substantially altered EIF1A mRNA levels (Figure S8C). These findings ruled out the possibility of a transcriptional regulation of HIF-1α on eIF1A.

Figure 5.

eIF1A knockdown and HHT suppresses PCa growth by alleviating hypoxia and enhancing the immune response

(A) Quantitative reverse-transcription PCR analysis of the relative expression level of HIF1A in the si-NC, si-eIF1A, sh-NC, sh-EIF1A, PBS, and HHT groups. Data were presented as the mean ± SD (3 biological replicates for each group and three technical replicates for each biological replicate; two-sided t tests).

(B) Western blot analysis to compare the protein levels of HIF-1α and eIF1A among the si-NC, si-eIF1A, sh-NC, sh-EIF1A, PBS, and HHT groups (3 biological replicates for each group).

(C) Western blot analysis to compare the protein levels of HIF-1α and eIF1A upon MG132 treatment and eIF1A knockdown (3 biological replicates for each group).

(D) Changes in polysome profiles and polysome-to-monosome (P/M) ratios upon EIF1A knockdown. Data were presented as the mean ± SD (3 biological replicates for each group; two-sided t tests).

(E) qPCR analysis comparing the HIF1A mRNA level in polysome and non-polysome fractions upon EIF1A knockdown. Data were presented as the mean ± SD (3 biological replicates for each group and three technical replicates for each biological replicate; two-sided t tests).

(F) Quadrant plot illustrating the levels of RPFs and RNA expression levels derived from Ribo-seq, where the larger pink dots represent genes with significantly differential RPF levels (2 biological replicates for each group).

(G) GSEA of the differentially activated pathways between the EIF1A knockdown and negative control groups based on translational efficiency derived from Ribo-seq.

(H) Comparisons of mouse tumor growth curves and tumor volumes between the HHT + ADT and vehicle groups. Data are presented as mean ± SEM (5 mice for each group). Statistical significance was determined by assessing the group × time interaction via a linear mixed-effects model (restricted maximum likelihood).

(I) GSEA of the differentially activated pathways between the HHT + ADT and vehicle groups based on RNA-seq (3 mice for each group).

(J) Immune cell inference based on RNA-seq data comparing the tumor purity, immune score, and T cell and macrophage ratio between the HHT + ADT and vehicle groups based on RNA-seq (3 mice for each group). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

(K and L) Representative mIF images showing the colocalization of eIF1A⁺ LE (CK8, eIF1A) and hypoxia markers (HIF-1α, CA9), T cells (CD3), and macrophages (CD68) in subcutaneous Myc-CaP tumors in the HHT+ADT and vehicle groups (4 mice for each group; two-sided t test; also see Figures S9F and G). Scale bars, 200 μm.

We demonstrated that the proliferative capacity of LNCaP cells was significantly reduced in the si-EIF1A group compared to the control under a hypoxic condition (Figure S9A). In vivo validation confirmed a reduced volume of subcutaneous tumors originated from EIF1A knockdown Myc-CaP cells (Figure S9B). Corresponding RNA sequencing (RNA-seq) revealed a decrease in eukaryotic translation initiation and hypoxia-related pathways, while an elevation of immune response pathways (Figure S9C). Cellular composition inference suggested that the abundance of CD8⁺ T cells and M1 cells was enriched upon EIF1A knockdown (Figure S9D). We further investigated the treatment efficacy of HHT. Similar to the findings in our previous study,²¹ the mouse body weights were not affected by treatment and the growth of subcutaneous Myc-CaP tumors was significantly inhibited, demonstrating the safety and anti-tumor efficacy of HHT + ADT therapy (Figures 5H and S9E). RNA-seq analysis suggested that the induced differentially expressed genes were associated with the decrease of PCa malignant signaling (EMT and androgen response pathway), transcription process (eukaryotic translation initiation and translation) and HIF-1 TF pathway, and the activation of immune response pathways (B cell receptor [BCR] pathway, T cell receptor [TCR] pathways, and TNF signaling and interleukin pathways; Figure 5I). Cellular composition inference also revealed a significant elevation of immune infiltration following HHT + ADT, accompanied by a marked increase in the proportions of CD8⁺ T cells and M1-differentiated macrophages, further suggesting that HHT treatment might enhance antitumor immunity (Figure 5J). mIF staining revealed a significant decrease in hypoxia in the tumors after HHT treatment (Figures 5K and S9F). The infiltration of macrophages and CD8⁺ T cells was also significantly increased after HHT treatment (Figures 5L and S9G). Through T cell depletion using in vivo CD3ε antibody, we found that tumor suppression was more pronounced in immunocompetent mice than in T-cell-depleted mice (fold change: 2.45 vs. 1.75; Figure S9H), indicating that HHT + ADT exerts both direct tumoricidal activity and indirect immunomodulatory effects.

Furthermore, we conducted an investigator-initiated clinical study to evaluate the efficacy of HHT as a neoadjuvant therapy for PCa (NCT06834321). Based on the surgically resected samples collected from patients with high-/very high-risk localized or locally advanced PCa (defined as PSA >20 ng/mL, Gleason score ≥3 + 4, T stage ≥ T3a, or N1; M0) in the early stage of the study, we employed scRNA-seq analyses to validate the molecular signatures identified in this study. We integrated the scRNA-seq data of the surgical tissues from four patients with the data from the aforementioned eight treatment-naive patients and identified a notable decrease in malignant LE in the HHT + ADT group (Figures 6A–6C). Compared with treatment-naive samples, HHT + ADT effectively inhibited pathways related to translation, hypoxia, or HIF-1α- and PCa-associated processes while enhancing antitumor immune processes, such as TCR, BCR, interleukin, and innate immune pathways (Figure 6D). In terms of immune cells, we detected a significant increase in the abundance of effector T cells (CD8⁺ effector memory T cells, Tem), including GZMK⁺ Tem and GZMB⁺ Tem, in the HHT + ADT group (Figures 6E and S10). Moreover, macrophages from the HHT + ADT group presented a consistently greater M1 phenotype score (extracted from Li et al.²²) compared with those from the treatment-naive group (Figure 6F), and the M1 vs. M2 score difference (M1 – M2) was also elevated in the treatment group (Figures S11A and S11B). The alterations in the hypoxic phenotype and CD8⁺ T cell and macrophage infiltration were additionally validated by mIF staining of paired pretreatment biopsy sections and postsurgical sections (Figures 6G, 6H, S12A, and S12B). Additionally, the staining of CD31, CK8, and CA9 demonstrated that pretreatment tumors exhibited significant hypoxia and tumoral angiogenesis, characterized by the proliferation of large, irregular, tortuous, and thin-walled blood vessels. Notably, these features were substantially alleviated following HHT + ADT therapy (Figure S12C).

Figure 6.

Molecular alteration of HHT in patients with progressive PCa

(A and B) The single-cell landscape of 100,912 cells from 10 PCa samples, including samples from four patients who underwent neoadjuvant HHT+ADT and eight treatment-naive patients (A), and the dot plot showing the expression of established markers for different cell types (B).

(C) The difference of epithelial cell distributions between treatment-naive and HHT+ADT group, and the dot plot showing the expression of established markers for different epithelial cell types.

(D) GSEA of malignant luminal cells from the HHT+ADT group versus the treatment-naive group based on BIOCARTA, PID, HALLMARK, and REACTOME databases.

(E) Boxplot comparing the ratio of effector T cells among T and NK cells between the HHT+ADT group (n = 4) and the treatment-naive group (n = 8) (two-sided Wilcoxon tests).

(F) Boxplot comparing the M1 AUCell score of macrophages between the HHT+ADT group and the treatment-naive group (two-sided Wilcoxon tests).

(G and H) Representative mIF images showing the colocalization of eIF1A⁺ LE (CK8, eIF1A), T cells (CD3), and macrophages (CD68) in paired pretreatment biopsy sections and postsurgical sections from PCa patients (n = 4; two-sided t tests; also see Figures S12A and S12B). Scale bars, 200 μm.

(I) Boxplot comparing the ratio of effector T cells among T and NK cells between the ADT group (n = 4) and the treatment-naive group (n = 4) (two-sided Wilcoxon tests).

(J) Boxplot comparing the M1 AUCell score of macrophages between the ADT group and the treatment-naive group (two-sided Wilcoxon tests).

In (E and I), the data points represent individual samples, the boxes represent IQR with the median line inside, and the whiskers show the values within 1.5×IQR. In (F and J), the boxes represent IQR with the median line inside, the whiskers show the values within 1.5×IQR, and outliers are plotted as individual points representing single cells.

To rule out the effects of ADT on the TME alterations, we also employed samples from four patients receiving neoadjuvant ADT + androgen receptor antagonist (ADT group) and merged the scRNA-seq data with the eight treatment ones (Figures S13A–S13D). We found that ADT significantly suppressed androgen response and related pathways (Figure S13E). Notably, ADT alone failed to inhibit hypoxia-related HIF-1α and other signaling molecules (Figure S13F). Although ADT also enhanced immune responses, the NES values of the corresponding pathway were all lower than those of HHT+ADT (2.01 vs. 2.68 for PID CD8 TCR downstream pathway, 2.00 vs. 2.30 for BIOCARTA TCR pathway, and 2.16 vs. 2.43 for REACTOME IL-10 signaling; Figure S13G). In terms of immune cells, ADT monotherapy did not significantly increase the proportion of Tem cells (Figures 6I and S14A–S14C), and although the M1 score and M1 − M2 score of macrophages showed an increase, it was relatively modest (Figures 6J, S11C, and S11D). These results collectively demonstrate the therapeutic effects of HHT independent of ADT.

Overall, HHT treatment could alleviate hypoxia and enhance the immune response in PCa, thus exerting anticancer effects, which further validated the functions of eIF1A⁺ LE and the “cold” TME in advanced PCa.

Discussion

Previous research on PCa has encountered significant difficulties in accurately delineating the spatial distribution and interactions among various histological subtypes.²³ Our work represents a meaningful application of IMC technology to analyze the spatial microenvironmental characteristics of PCa and its subtypes. This work not only emphasized the key role of translation process and its therapeutic implications for PCa but also highlights the advantages of spatial proteomic technology.

eIF1A is a eukaryotic translation initiation factor that is integral to the process of mRNA translation.²⁴ Mutations or aberrant expression of its encoding gene, EIF1A, have been closely linked to the initiation, progression, and prognosis of various malignancies.^25,26 Recently, the role of translation initiation has received increasing attention. Based on the significant association between eIF4A1 and PCa patient prognosis, Kuzuoglu-Ozturk et al. evaluated the efficacy of a clinical small molecule, zotatifin, and discovered promising anticancer effects.²⁷ Furthermore, Zhou et al. reported that targeting eIF3F could decelerate the progression of hepatocellular carcinoma and enhance the efficacy of anti-programmed cell death-1 immunotherapy.²⁸ Our study revealed the critical role of eIF1A in PCa, providing further insight into the key role of translation initiation in tumors. Additionally, our analysis linked eIF1A and translation processes to the hypoxic response. We would like to emphasize that we used eIF1A as a key representative to investigate the broader biological role of translation initiation and did not rule out the possibility that other eIFs might play significant role in PCa progression.

There exists considerable variability in clinical survival durations among distinct pathological subtypes of PCa,²³ which may be linked to differences in the TME. Previous studies have identified variations in immune infiltration in localized PCa and that metastatic PCa and castration-resistant PCa (CRPC) are typically regarded as existing within an immunosuppressive environment.^29,30,31,32 Nonetheless, prior studies have been insufficient in delineating the immune microenvironment characteristics of different histological subtypes. Although investigations have applied RNA-seq and single-cell omics,³³ these techniques present notable limitations. RNA-seq does not elucidate the spatial distribution of cells within the TME. Recent studies utilizing scRNA-seq also faced difficulties in detecting focally distributed rare subtypes, such as IDC and DAC.³⁴ The focal distribution of these subtypes complicates the accurate identification of pertinent tumor and microenvironmental cells from a large cellular population, potentially resulting in unreliable findings. To overcome these challenges, we employed IMC technology to construct an in situ single-cell protein atlas of various histological subtypes of PCa, enabling the precise identification of small ROIs that encompass diverse pathological types.³⁵ Through this approach, more important than uncovering heterogeneity is our discovery of the unified hypoxia and “cold” tumor characteristics across high-risk subtypes, which is crucial for suggesting promising therapeutic strategies.

Our previous investigations have established a significant correlation between the aggressive phenotype of small cell-like PCa (SCLPC) and the activation of translation.²¹ HHT, the sole translation inhibitor currently approved for clinical use, has exhibited antitumor efficacy across various malignancies.³⁶ Our previous studies demonstrated that HHT effectively hinders the progression of CRPC.²¹ Building upon these findings using comprehensive animal studies, we launched an investigator-initiated clinical trial. scRNA-seq analyses revealed both alleviation of the hypoxia response and enhancement of the immune response. This study confirms the therapeutic value of HHT in treatment-naive (hormone-naive) PCa (HNPC). Together with our previous finding that HHT suppresses SCLPC, these results collectively suggest the therapeutic potential of HHT across multiple stages of PCa, including both HNPC and CRPC. Consequently, we posit that our findings offered a promising strategy for the precision treatment of PCa and present opportunities for further translational research in this area.

In summary, this study underscores the conserved interplay between translation and hypoxia in PCa subtypes and their contributions to the development of an immunosuppressive microenvironment, offering insights for the precision treatment of PCa.

Limitations of the study

The sample size of this study is relatively small, and despite using external data for validation, there may still be insufficient statistical power in the analysis of clinical relevance. Additionally, the IMC panel was constructed based on key PCa-related proteins identified through a literature review, which may have overlooked other important functional molecules. Furthermore, this study did not elucidate the specific mechanisms by which the eIF1A-HIF-1α axis regulates immune infiltration or immune cell function in high-risk PCa. Further investigation into this aspect will be conducted in our future work.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Bin Xu (njxbseu@seu.edu.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •All matrices generated from this study, including human IMC, mouse bulk RNA-seq, human scRNA-seq, and human Ribo-seq are available via Dryad at https://doi.org/10.5061/dryad.tqjq2bwb3. The raw data of mouse bulk RNA-seq, human scRNA-seq, and human Ribo-seq data are available via NCBI SRA with BioProject ID PRJNA1390428.
• •This paper does not report original code. The code used to conduct bioinformatic analysis is available via Zenodo at https://doi.org/10.5281/zenodo.18015231.
• •Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

Acknowledgments

This work was funded by National Natural Science Foundation of China (92359304; S.J.), Noncommunicable Chronic Diseases-National Science and Technology Major Project (2024ZD0535900; Y.C.); General sSocial Development Project of Jiangsu Science and Technology Department (BE2023771; B.X.) and Jiangsu Province High-Level Hospital Construction Funds of Zhongda Hospital (GSPJCYJ05; B.X.). We are grateful to the staff of the Public Scientific Research Platform of Zhongda Hospital, affiliated with Southeast University, and Infinity Scope company for technical assistance.

Author contributions

Conceptualization, L.W., E.H., D.Z., Y.C., and B.X.; methodology, L.W., E.H., S.T., W.L., C.C., Y.C., and B.X.; investigation, L.W., B.Z., X.D., G.M., Z.C., T.W., S.C., and J.Z.; visualization, L.W., E.H., B.Z., X.D., S.T., G.M., W.L., C.C., Y.C., and B.X.; funding acquisition, S.J., D.Z., Y.C., and B.X.; project administration, M.C., G.T., S.J., D.Z., Y.C., and B.X.; supervision, R.N., M.C., S.J., D.Z., Y.C., and B.X.; writing – original draft, L.W., E.H., B.Z., X.D., and Y.C.; writing − review & editing, R.N., M.C., G.T., S.J., S.J., D.Z., Y.C., and B.X.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibody
Anti-CD44 (IMC)	Abcam	Cat# ab6124; RRID: AB_305297
Anti-AR (IMC)	Boster	Cat# A00542; RRID: AB_3081165
Anti-IDH1 (IMC)	Abcam	Cat# ab242078
Anti-HLA I (IMC)	Abcam	Cat# ab239788; RRID: AB_3665873
Anti-TP53 (IMC)	Boster	Cat# PB9008; RRID: AB_3082963
Anti-CK5 (IMC)	Boster	Cat# M00398-7; RRID: AB_3081774
Anti-Collagen I (IMC)	abcam	Cat# ab215969; RRID: AB_2909621
Anti-CASP3 (IMC)	Cell Signaling Technology	Cat# 94530; RRID: AB_3076239
Anti-CD31 (IMC)	Cell Signaling Technology	Cat# 85873SF
Anti-E-cadherin (IMC)	Abcam	Cat# ab256580
Anti-HIF-1a (IMC)	Abcam	Cat# ab210073
Anti-CD45 (IMC)	Cell Signaling Technology	Cat# 47937; RRID: AB_2922773
Anti-IDH2 (IMC)	Abcam	Cat# ab230796
Anti-CD163 (IMC)	Abcam	Cat# ab215976; RRID: AB_3717832
Anti-IDO-1 (IMC)	Proteintech	Cat# 66528-1-Ig; RRID: AB_2881891
Anti-PD-L1 (IMC)	Abcam	Cat# ab226766; RRID: AB_3073663
Anti-VEGFA (IMC)	Abcam	Cat# ab185238
Anti-CD68 (IMC)	Biolegend	Cat# 916104; RRID: AB_2616797
Anti-PSMA (IMC)	Sino biological	Cat# 201427-T10
Anti-CD20 (IMC)	Abcam	Cat# ab213033
Anti-PTEN (IMC)	Sino biological	Cat# 100729-T08
Anti-CK8 (IMC)	Boster	Cat# M01421-3
Anti-SF3B1 (IMC)	Abcam	Cat# ab249538
Anti-PD-1 (IMC)	Cell Signaling Technology	Cat# 63815; RRID: AB_3675993
Anti-Ki-67 (IMC)	BD	Cat# 550609; RRID: AB_393778
Anti-FBP1 (IMC)	Abcam	Cat# ab247749
Anti-HLA-DR (IMC)	Abcam	Cat# NB600-989
Anti-eIF1A (IMC)	Abcam	Cat# ab243919
Anti-CD3 (IMC)	Cell Signaling Technology	Cat# 24581SF
Anti-FAP (IMC)	Abcam	Cat# ab271976; RRID: AB_3676372
Anti-c-MYC (IMC)	Abcam	Cat# ab264603
Anti-CD29/Integrin beta 1 (IMC)	Abcam	Cat# ab271931
Anti-HMGCR (IMC)	Boster	Cat# A00643-3
Anti-ACC (IMC)	Abcam	Cat# ab173584
Anti-Pan-cytokeratin (IMC)	Biolegend	Cat# 914204, RRID: AB_2616960
Anti-aSMA (IMC)	Biolegend	Cat# 904601, RRID: AB_2565041
Anti-BRCA1 (IMC)	Boster	Cat# PB9015, RRID: AB_3082969
Anti-Vimentin (IMC)	Abcam	Cat# ab193555, RRID: AB_2814713
Anti-eIF1A (WB and IF)	Proteintech	Cat# 11649-2-AP; RRID: AB_2097803
Anti-HIF-1α (WB and IF)	Cell Signaling Technology	Cat# 14179, RRID: AB_2622225
Anti-β-actin (WB)	Proteintech	Cat# 66009-1-Ig; RRID: AB_2687938
Anti-CA9 (IF)	Proteintech	Cat# 11071-1-AP; RRID: AB_2066528
Anti-CD3 (IF)	Abcam	Cat# ab16669; RRID: AB_443425
Anti-CD68 (IF)	Affinity biosciences	Cat# DF7518; RRID: AB_2841017
Anti-human CD8A (IF)	Proteintech	Cat# 66868-1-Ig; RRID: AB_2882205
Anti-human CD31 (IF)	Proteintech	Cat# 11265-1-AP; RRID: AB_2299349
Anti-p63 (IF)	Proteintech	Cat# 12143-1-AP, RRID:AB_10597397
Anti-collagen-I (IF)	Proteintech	Cat# 67288-1-Ig, RRID:AB_2882554
InVivoMAb mouse IgG1 isotype control	Bio X cell	Cat# BE0083
InVivoPlus Anti-mouse CD3ε	Bio X cell	Cat# BP0001-1
Gene primer	Sense	Antisense
human-EIF1A	CCGCTACCCGGAAAGAAGT	AGCTTTGTTATCCTGGTAGTCTCG
human-HIF1A	CTGAGGGGACAGGAGGA	CACACGCGGAGAAGAGA
human-β actin	CATGTACGTTGCTATCCAGGC	CTCCTTAATGTCACGCACGAT
mouse-EIF1A	TGCGGATTATTACCCGATTCAGT	GAAGATCCACAGGCAGCAAACA
mouse-HIF1A	AACCCATTCCTCATCCGTCAA	CCGGCTCATAACCCATCAACT
mouse-β actin	GTGACGTTGACATCCGTAAAGA	GTAACAGTCCGCCTAGAAGCAC
siRNA and shRNA
si-EIF1A	GAGGUUAUGUCACAUCAGATT	UCUGAUGUGACAUAACCUCTT
si-HIF1A	GGGAUUAACUCAGUUUGAATT	UUCAAACUGAGUUAAUCCCTT
si-NC	UUCUCCGAACGUGUCACGUTT	ACGUGACACGUUCGGAGAATT
sh-EIF1A	GATCCGATGTTGAGTTATCATCTTAACTCGAG TTAAGATGATAACTCAACATCTTTTTT	AATTAAAAAAGATGTTGAGTTAT CATCTTAACTCGAGTTAAGATG ATAACTCAACATCG
sh-NC	GATCTGTTCTCCGAACGTGTCACGTTTCAAGA GAACGTGACACGTTCGGAGAATTTTTTC	AATTGAAAAAATTCTCCGAACG TGTCACGTTCTCTTGAAACGTG ACACGTTCGGAGAACA
Critical Reagent
RNA extraction kit	Takara Kusatsu	Cat# 9767
Hiscript II First-Strand cDNA Synthesis Kit	Vazyme Biotech	Cat# R211
MonAmp SYBR Green qPCR Mix	Monad Biotech	Cat# MQ10101S
RIPA lysis buffer	KeyGene Biotech	Cat# KGB5203-100
BCA Protein Assay Kit	Thermo Scientific	Cat# 23225
SYBR Green	Thermo Scientific	Cat# 4368577
RPMI 1640 medium	Thermo Scientific	Cat# C11875500BT
DMEM medium	Thermo Scientific	Cat# C11995500BT
Fetal bovine serum	Suzhou Shuangru Biotechnology CO.td	Cat# LONSERA/S711-001S
MACS Tissue Storage Solution	Miltenyi	Cat# 130-100-008
Phosphate buffered saline (PBS)	Hyclone	Cat# SH30256.01
Collagenase Type II	Sigma	Cat# V900892
Collagenase Type IV	Sigma	Cat# C5138
Dispase II	Roche	Cat# 4942078001
Fetal bovine serum (FBS)	Gibco	Cat# 10100147C
6-Colour IHC kit for human	Absin	Cat# abs50014
6-Colour IHC kit for mouse	Absin	Cat# abs50030
CoCl2	Macklin	Cat# C885204
Cell Counting Kit-8	TargetMol	Cat# C0005
Degarelix Acetate	TargetMol	Cat# 934016-19-0
Homoharringtonine	TargetMol	Cat# 26833-87-4
Experimental models
293T cell	American Type Culture Collection (ATCC, USA)	N/A
LNCaP cell	ATCC	N/A
MyC-CaP cell	ATCC	N/A
Male FVB/NJGpt mice	GemPharmatechCo., Ltd.	N000026
Biological samples
Tissue microarray of prostate cancer and paracancerous tissues	Zhongda Hospital Southeast University	Ethics: 2024ZDSYLL382-P01
Pre- and post-treatment PCa samples	Zhongda Hospital Southeast University First People’s Hospital of Nantong	Ethics: 2024ZDSYLL510-P01, 2025KT076
Deposited data
Human IMC matrix	This paper (Deposited on DRYAD)	DRYAD Data: https://doi.org/10.5061/dryad.tqjq2bwb3
Human single-cell RNA-seq matrix	This paper (Deposited on DRYAD)	DRYAD Data: https://doi.org/10.5061/dryad.tqjq2bwb3
Human single-cell RNA-seq raw data	This paper (Deposited on SRA)	SRA: PRJNA1390428; https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1390428
Mouse bulk RNA-seq matrix	This paper (Deposited on DRYAD)	DRYAD Data: https://doi.org/10.5061/dryad.tqjq2bwb3
Mouse bulk RNA-seq raw data	This paper (Deposited on SRA)	SRA: PRJNA1390428; https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1390428
Human Ribo-seq matrix	This paper (Deposited on DRYAD)	DRYAD Data: https://doi.org/10.5061/dryad.tqjq2bwb3
Human Ribo-seq raw data	This paper (Deposited on SRA)	SRA: PRJNA1390428; https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1390428
Human single-cell RNA-seq data	DRYAD	DRYAD Data: https://doi.org/10.5061/dryad.4b8gthtqb
Human bulk RNA-seq data (GEO)	GEO (https://www.ncbi.nlm.nih.gov/gds)	GEO: GSE116918
Human bulk RNA-seq data (TCGA)	TCGA (https://portal.gdc.cancer.gov)	TCGA: PRAD dataset (FPKM)
Human HIF-1α ChIP-seq data	GEO	GEO: GSE200205
Software and packages
R	https://www.r-project.org/	v.4.0.5
Python	https://www.python.org/	v.3.7.12
Harmony	R package (CRAN)	v.1.2.0
Rphenograph	R package (github)	v.0.99.1
imcRtools	R package (github)	v.1.0.2
Cell Ranger	10x Genomics (https://www.10xgenomics.com/)	v.3.0.3
Seurat	R package (CRAN)	v.5.0.1
AUCell	R package (Bioconductor)	v.1.24.0
clusterProfiler	R package (github)	v.4.2.2
GSVA	R package (Bioconductor)	v.1.42.0
survminer	R package (CRAN)	v.0.4.9.999
limma	R package (Bioconductor)	v.3.54.0
xCell	R package (github)	v.1.1.0
estimate	R package (github)	v.1.0.13
mMCPcounter	R package (github)	v.1.1.0
Zen	ZEISS (https://www.zeiss.com/)	v.2.3
CaseViewer	3DHISTECH (https://www.3dhistech.com)	v.2.4.0.119028
ImageJ	ImageJ (https://imagej.net/)	v.1.53t

Experimental models and study participant details

Ethics statement

The use of PCa tissues in this study was approved by the Clinical Research Ethics Committee of Zhongda Hospital Southeast University (Approval No. 2024ZDSYLL382-P01). The investigator-initiated clinical trial evaluating the therapeutic efficacy of HHT in combination with ADT was approved by the Clinical Research Ethics Committee of Zhongda Hospital Southeast University (Approval No. 2024ZDSYLL510-P01) and the Ethics Committee of the First People’s Hospital of Nantong (Approval No. 2025KT076), with a registration number NCT06834321 on ClinicalTrials.gov. Written informed consent was obtained from all participants for both the use of specimen and the participation of the clinical trial. All animal experiments in this study were conducted at the Animal Experiment Center of Southeast University, strictly adhering to the protocol approved by the Southeast University Animal Research Committee (Approval No. SEU-IACUC-20250401001), in compliance with the “Guide for the Care and Use of Laboratory Animals”.

Human specimens and clinical data collection

The tissue microarray (TMA) was constructed using tumoral and paracancerous (PCT) tissues from 47 treatment-naïve male patients aged 56–86 who were diagnosed with PCa from October 2019 to September 2023, with a follow-up until April 2025. PCT samples were obtained from patients with low-risk (Gleason score 3 + 3) PAC, and were preferentially collected from the contralateral, non-cancerous prostate lobe. In cases of bilateral involvement, PCT was harvested at a minimum distance of 1 cm from the tumor. All demographic (age and BMI), clinical (PSA, tumor stage, progression status and survival status), and pathological (Gleason score, tumor quantification, histological subtypes, bladder neck invasion, spermatic glands invasion, vascular invasion and neural invasion) data were extracted from the electronic medical record system or follow-up records, with pathological findings further reviewed and confirmed by professional pathologists. Detailed information for all enrolled patients is provided in Table S1.

To comprehensively assess cellular heterogeneity and validate IMC findings, we conducted single-cell RNA-seq (scRNA-seq) analysis on surgically resected tumor specimens from 16 male patients age 48–76 with high/very high-risk localized or locally advanced PCa (defined as PSA >20ng/mL, Gleason score ≥3 + 4, T stage ≥ T3a, or N1; M0; clinical characteristics in Table S3). The study generated scRNA-seq data from ten newly-collected cases (two treatment-naïve, four receiving HHT+ADT neoadjuvant therapy and four receiving neoadjuvant ADT), supplemented with six additional high/very high-risk localized or locally advanced cases from the validation cohort of our previous study (excluding discovery cohort samples due to platform differences and four intermediate-risk cases in the validation cohort).^32,37

Cell lines and culture

Human (LNCaP and 293T) and mouse (MycCap) PCa cell lines were purchased from the American Type Culture Collection (ATCC, USA). Cells were cultured in an RPMI 1640 or DMEM medium (Gibco, Thermo Fisher Scientific, Waltham, MA, USA) containing 10% fetal bovine serum (FBS; Suzhou Shuangru Biotechnology CO.td), 1% penicillin G, and streptomycin sodium (Gibco) in 95% humidified air at 37°C and 5% CO2. For chemical hypoxic conditions, cells were cultured at 37°C with 5% CO2, adding cobalt chloride (CoCl₂) into the medium.

Mouse models

The experiment utilized 6 to 8-week-old male FVB mice, purchased from the Animal Experiment Center of Southeast University, and acclimatized for one week. The mice were housed under standard conditions, including a 12-h light/dark cycle, with free access to food and water. All animal experiments were conducted at the Animal Experiment Center of Southeast University, strictly following the protocol approved by the Animal Research Committee of Southeast University. To construct the Myc-CaP cell (sh-NC, sh-EIF1A) subcutaneous tumor model in mice, cells in the logarithmic growth phase were digested, washed, centrifuged, and then diluted to a cell suspension. Using a 1 mL syringe, 1.0×10⁶ cells were injected subcutaneously into the right dorsal region of the mice. After injection, the mice were randomly divided into an experimental group and a control group, with each group containing 5 mice. To evaluate the therapeutic efficacy and investigate the mechanism of HHT, mice in the treatment group received a combination therapy of HHT and ADT. Treatment was initiated when tumor volumes reached 0.2–0.3 cm³. The dosing regimen consisted of a subcutaneous injection of degarelix (25 mg/kg) at the beginning of treatment, along with intraperitoneal injections of HHT (2 mg/kg) administered three times per week. The vehicle group received PBS injections via the corresponding routes. During the treatment period, body weight was monitored regularly to assess treatment-related toxicity. Tumor volume was also measured to evaluate the preliminary therapeutic efficacy. The mice were sacrificed at the third day following the fourth HHT treatment.

To validate the function of immune components in the response to HHT, in vivo anti-mouse CD3ε antibody was applied to construct T cell depletion mouse models. Prior to HHT + ADT treatment, mice received intravenous injections of either 50 μg anti-CD3ε antibody or 50 μg of a non-reactive isotype-matched control (mouse IgG1) for five consecutive days to achieve T cell depletion. Following the initiation of HHT+ADT therapy, intravenous injections were administered every two days to maintain sustained T cell depletion. The anti-tumor effects of HHT+ADT were then evaluated and compared between the T-cell-depleted group and the IgG1 control group.

Method details

Tissue microarray preparation

TMA is constructed by selecting representative cores with a diameter of 1 mm from surgical tumor specimens. All cores were jointly selected by two pathologists, who confirmed the histological subtypes and prepared them into tissue microarrays. The specific sampling locations were determined according to the research requirements.

Imaging mass cytometry (IMC) data processing and analysis

IMC panel design and antibody selection

The selection of the IMC staining panel is based on an analysis of target cell types and their biological characteristics. Antibodies undergo rigorous screening and validation, including dual verification through immunofluorescence (IF) and IMC, to ensure their specificity and signal intensity. The panel encompasses a variety of cell type markers and immune or cancer-related functional markers. The antibody dilution is optimized through a series of dilution experiments to achieve the optimal staining effect. Comprehensive antibody information is detailed in Table S2.

IMC sample preparation

The TMA was serially sectioned to a thickness of 5 μm. The sectioning process proceeded as follows: Sections were dewaxed at 70°C, followed by antigen retrieval at 95°C. Non-specific binding sites were blocked using Dako’s protein-free blocking solution and incubated for 45 min. A mixture of metal-labeled antibodies, diluted in 1% BSA/PBS, was applied to the sections and incubated overnight at 4°C. The sections were then washed with 0.2% Triton X-100 and 1× PBS, followed by nuclear counterstaining, which was performed at room temperature for 30 min. Subsequently, the sections were washed with ultrapure water and air-dried naturally. IMC images were acquired using the Hyperion imaging system.

IMC antibody labeling and validation

All antibodies were optimized and validated across various tissues, including the spleen, tonsil, appendix, placenta, thymus, normal lung tissue, and PCa tissue, to ensure their specificity and signal strength. The validation process involved assessing expected staining specificity, which included co-staining, mutual exclusivity, and the signal-to-noise ratio. Nuclear labeling was achieved with an Intercalator-Ir (201192B, Fluidigm) solution in PBS-TB (1.25 μM) for 30 min at room temperature, followed by two PBS-TB washes and a final rinse with ddH2O.

IMC imaging and data acquisition

Following the acquisition of IMC images, each TMA core is individually imaged using the Fluidigm Hyperion imaging system. The tissue is scanned through laser ablation at a pixel size of 1 μm², where the ablated tissue plume is ionized by a plasma source. Signals from metal-tagged antibodies are then detected via time-of-flight mass spectrometry (TOF-MS). Ultimately, images corresponding to each antibody are reconstructed based on the metal abundance of each pixel. The staining results are reviewed by professional pathologists. IMC image preprocessing and cell segmentation involve a comprehensive data analysis pipeline that includes spillover signal compensation, image denoising, image contrast enhancement, and cell segmentation. The spillover signals in each channel are corrected using a previously defined spillover matrix.³⁸ Median filtering is applied to suppress noise, followed by intensity adjustment using the MATLAB function imadjust to optimize the intensity distribution. Finally, connectivity-aware segmentation methods are employed to accurately segment individual cells or components in the IMC images.³⁹ An imaging mass cytometer (Fluidigm, Hyperion) was employed to scan the prepared sections to generate multiplexed images.

Image processing and spatial information analysis

Overflow signals are filtered out, and median filtering is applied to reduce noise in the signals from each channel.³⁸ The contrast between the signals and the background is enhanced through linear adjustment. During cell segmentation, membrane protein signals located more than 20 μm away from the nucleus cnter are excluded. Marker expression data undergo hyperbolic sine transformation and are subsequently normalized using the min-max method. Batch effects are corrected using the R software Harmony,⁴⁰ and cells are clustered using Rphenograph (k = 100).⁴¹ The clustering results are employed for cell type annotation. The cellular neighborhood (CN) of each cell consists of its 10 nearest neighboring cells. The neighborhoods are clustered and annotated using the k-means clustering method (k = 20) and validated through Voronoi diagrams. Permutation tests are conducted using imcRtools to analyze the interactions between cell types within the selected region of interest (ROI). Differences in interactions between various groups are compared using Student’s t test, with p < 0.05 indicating significant spatial interactions between cells across different groups.

Single-cell RNA sequencing

The fresh tumor tissues were stored in the MACS Tissue Storage Solution and transported on ice, which were then washed in ice-cold PBS and dissociated using Collagenase Ⅱ, Collagenase Ⅳand Dispase II. Finally fresh cells were washed twice in RPMI1640 and then resuspended at 1 × 10⁶ cells per mL in RPMI1640 and 2% FBS. Single-cell RNA-Seq libraries were prepared using SeekOne DD Single Cell 3′ library preparation kit (SeekGene Catalog No. K00202) and sequenced on an Illumina NovaSeq X Plus platform with PE150 read length.

scRNA-seq data processing and analysis

Data preprocessing and dimensionality reduction

Raw reads were processed to generate gene expression profiles using Cell Ranger (v.3.0.3, 10x Genomics). Reads were mapped to GRCh38 with ensemble version 92 gene annotation. Cells with gene counts less than 500 or more than 550, UMI counts less than 1,000 or more than 30,000, or more than 10% expression on mitochondrial genes were filtered out. Genes expressed in less than 10 cells were also excluded from analysis. Initially, the data were normalized using the Seurat package with the LogNormalize method (scale.factor = 10,000). Subsequently, the top 2,000 highly variable feature genes were identified using the FindVariableFeatures function with the vst method. Following this, all genes were standardized using the ScaleData function, and PCA dimensionality reduction was performed with the RunPCA function.

Batch effect correction and cell clustering

To address potential batch effects, the Harmony package was utilized for dimensionality reduction and integration. The number of principal components was determined using the Elbow plot, with components accounting for over 90% of the cumulative variance selected for subsequent analysis. Based on the dimensionality reduction results from Harmony, cell clustering analysis was conducted using Seurat’s FindNeighbors and FindClusters functions, evaluating clustering effects at various resolutions (0.05, 0.1, 0.2, 0.3, 0.5, 0.8, 1.0, 1.5 and 2.0). An appropriate resolution was ultimately chosen for UMAP dimensionality reduction visualization (RunUMAP), and cell types were annotated according to the clustering results.

Gene set scoring

The AUCell algorithm from AUCell package, was employed to calculate the transcriptional signature scores for individual cells and samples.

Identification of differentially expressed genes and functional enrichment analysis

To investigate the potential mechanisms underlying the high and low expression of EIF1A in epithelial cell subpopulations, as well as the effects of different therapies, the FindMarkers function in Seurat was utilized to identify differentially expressed genes. Following this, gene set enrichment analysis (GSEA) analyses were performed (clusterProfiler) based on the differentially expressed genes, which included the HALLMARK, BIOCARTA, REACTOME, and PID databases.

Mouse bulk RNA-seq and data analysis

Total RNA was extracted with TRIzol reagent and quality-controlled using NanoDrop and Bioanalyzer (RIN >7.0). Following poly(A) RNA selection and fragmentation, double-stranded cDNA was synthesized and ligated with adapters after end repair and A-tailing. After UDG treatment and PCR amplification, a final cDNA library with an average insert size of 300 ± 50 bp was constructed.

We used the DESeq2 package to identify differentially expressed genes (DEGs) based on the criteria of adjusted p value <0.05 and |logFC| > 1. Subsequently, we employed the clusterProfiler R package to conduct GSEA of the differentially expressed genes based on the HALLMARK, BIOCARTA, REACTOME, and PID databases, aiming to explore the biological functional differences between different groups. Immune cell inference methods including ESTIMATE, MCPcounter and xCell were employed to measure the cell components based on RNA-seq data.

Small interfering RNA, short hairpin RNA, and overexpression lentivirus

Small interfering RNAs (siRNAs) targeting EIF1A was designed and synthesized by GenePharma Co. (Shanghai, China). Short hairpin RNA (shRNA) targeting EIF1A was synthesized into the GZ404 lentivirus, GX402 lentivirus, GW401 lentivirus (Genomeditech, Shanghai, China), and an empty lentivirus was used as the negative control. EIF1A overexpressed lentivirus (oe-EIF1A) was synthesized and subcloned into the pGBKT7 plasmid. HIF1A overexpressed lentivirus (oe-HIF1A) were synthesized and subcloned into the pcDNA3.1-EGFP plasmid, and an empty plasmid was used as the negative control.

RNA extraction and RT-PCR

RNA was extracted using an RNA extraction kit (Takara Kusatsu, Shiga, Japan) according to the manufacturer’s protocol, and purified RNA was used to synthesize cDNA by reverse transcription PCR (RT-PCR) using a Hiscript II First-Strand cDNA Synthesis Kit (Vazyme Biotech). Quantitative real-time PCR was performed using the MonAmp SYBR Green qPCR Mix (Monad Biotech, Zhuhai, China). Relative gene expression levels were normalized to the internal control β-actin.

Western blotting

Total protein was extracted using the RIPA lysis buffer (1:1000) (KeyGene Biotech, Nanjing, China), and the supernatant was collected. Protein quantification was performed using the bicinchoninic acid assay (KeyGene Biotech). The extracted total protein was separated into a 10% sodium dodecyl sulfate-polyacrylamide (SDS) gel electrophoresis and transferred onto a polyvinylidene difluoride membrane (Merck Millipore, Burlington, MA, USA). The membrane was then blocked for 1 hour in Tris-buffered saline with Tween 20 containing 5% non-fat milk. Next, we incubated the membrane with the following primary antibodies overnight at 4°C: eIF1A (1:1000), HIF-1α (1:1000) and β-actin (1:1000).

Polysome profiling

Polysome profiling separates translated mRNAs on a sucrose gradient according to the number of bound ribosomes. Firstly, cells are lysed and loaded on top of a 10–45% sucrose gradient. After ultracentrifugation, the gradient is monitored at A260 using a flow cell coupled to a spectrophotometer and then fractionated into equal fractions: untranslated mRNAs (top fractions) are separated from polysome-associated mRNAs (bottom fractions). Fractions are then processed for RNA extraction manually by acid phenol–chloroform extraction, simultaneously handling several gradients. Extracted RNA were used for qPCR analysis of HIF1A mRNA. The ratio of mRNA abundance in the polysome fractions (fractions 11–18) to that in the non-polysome fractions (fractions 1–10) was used to represent the translational activity of the corresponding gene.

Ribo-seq

To perform Ribo-seq, cells or tissues are treated with translation inhibitors to arrest ribosomes on the mRNA, freezing the translation process. The samples are then lysed, and ribonuclease is added to digest the unprotected mRNA regions, leaving behind RPFs. Subsequently, the ribosome - mRNA complexes are isolated through sucrose cushion. After separating the ribosomes from the RPFs, the RPFs are purified. rRNA contaminants are removed, and the purified RPFs are used to construct a sequencing library. This library is then subjected to high - throughput sequencing. Reads are mapped to the reference genome. The differentially enriched RPFs and differentially expressed RNA were evaluated using edgeR package (v.4.6.3) in R. The translation efficiency is calculated as the ratio of FPKM of RPFs (Fragments Per Kilobase of exon per Million reads mapped) in Ribo-seq to FPKM of RNA in RNA-seq.

Cell proliferation analysis

The cell proliferation capacity was tested with Cell Counting Kit-8 (CCK-8) assay. A total of 2,000 cells were seeded into 96-well plates for 24h-120 h, and 10 μL of the CCK-8 solution was added per well. After a 2 h incubation at 37°C, optical density at 450 nm (OD 450 nm) was measured on a microplate reader (Bio-Tek, USA).

Multiplex immunofluorescence staining

The multiplex immunofluorescence (mIF) staining protocol was performed using a Sive-color multi-labeled immunofluorescence staining kit (abs50014, Absin) according to the instructions. Briefly, sections were deparaffinized, rehydrated, subjected to antigen retrieval, and then blocked with PBS containing 10% goat serum for 20 min at room temperature, followed by incubation with the primary antibody overnight at 4°C. After washing, sections were incubated with a secondary antibody for 1 h at room temperature, and amide signal amplification was performed. The primary antibodies included CK8 (1:100), eIF1A (1:400), HIF-1α (1:200), CA9 (1:200), CD3 (1:800), CD8A (1:200), CD68 (1:400), p63 (1:400), Collagen-I (1:400), CD31 (1:200). Cells were labeled with antibodies and then stained with DAPI. Images were acquired by laser scanning confocal microscopy (LSM700, Zeiss) and processed by Zen software (v.2.3, Zeiss). ImageJ software (v.1.53t) was used to obtain the fluorescence intensity represented by mean gray value through setting threshold for signal capturing.

Quantification and statistical analysis

In this study, all statistical analyses were conducted using R (v.4.0.5). Specifically, for omics data, Wilcoxon rank-sum test was employed to compare differences between two groups, which is suitable for non-normally distributed data. Simultaneously, the Kruskal-Wallis test was used to compare differences among multiple groups, also applicable to non-normally distributed data. Additionally, for in vivo and in vitro experimental data, assuming that the data followed a normal distribution, the t test was utilized to compare differences between two groups, while one-way analysis of variance (ANOVA) was applied to compare differences among multiple groups. For correlation analysis, we employed Spearman analyses to explore the relationships between factors. We utilized the survminer package (v.0.4.9.999) to conduct survival analysis via Kaplan-Meier curves, dividing individuals into high-expression and low-expression groups based on the determined optimal cutoff value. The corresponding p-values were obtained from the log rank test. For the comparison of time-relied curves including tumor volume, mouse body weight and CCK-8 assays, the statistical significance was determined by assessing the Group × Time interaction via a linear mixed-effects model (Restricted Maximum Likelihood [REML]).

Specifically, for the variables (excluding OS and PFS) tested in the dot plots of Figures 2A, 4B, and 4D, the 71 ROIs were categorized into the groups indicated in the figures (corresponding to the group colors shown on the right). For comparisons of marker expression levels (Figure 2A) or CN proportions (Figure 4B) or distances (Figure 4D), the median value within each ROI was calculated, and Wilcoxon sum-rank tests were used to assess differences between groups.

All statistical details, including the specific tests used, the number of replicates, and the measures of center and dispersion, were provided in the corresponding figure legends. All statistical tests were two-sided.

Additional resources

The associated clinical trial has been registered on https://clinicaltrials.gov (NCT06834321).

Published: February 17, 2026