Spatial analysis of microRNA regulation at defined tumor hypoxia levels reveals biological traits of aggressive prostate cancer

Abstract

Mechanisms regulating the gene expression program at different hypoxia severity levels in patient tumors are not understood. We aimed to determine microRNA (miRNA) regulation of this program at defined hypoxia levels from moderate to severe in prostate cancer. Biopsies from 95 patients were used, where 83 patients received the hypoxia marker pimonidazole before prostatectomy. Forty hypoxia levels were extracted from pimonidazole-stained histological sections and correlated with miRNA and gene expression profiles determined by RNA sequencing and Illumina bead arrays. This identified miRNAs associated with moderate (n = 7) and severe (n = 28) hypoxia and predicted their target genes. The scores of miRNAs or target genes showed prognostic significance, as validated in an external cohort of 417 patients. The target genes showed enrichment of gene sets for cell proliferation and MYC activation at all hypoxia levels and PTEN inactivation at severe hypoxia. This was confirmed by RT-qPCR for MYC and PTEN, by Ki67 immunohistochemistry, and by gene set analysis in an external cohort. To assess whether miRNA regulation occurred within the predicted hypoxic regions, a method to quantify co-localization of multiple histopathology parameters at defined hypoxia levels was applied. A high Ki67 proliferation index co-localized significantly with hypoxia at all levels. The co-localization index was strongly associated with poor prognosis. Absence of PTEN staining co-localized significantly with severe hypoxia. The scores for miRNAs correlated with the co-localization index for Ki67 staining and hypoxia, consistent with miRNA regulation within the overlapping regions. This was confirmed by showing miR-210-3p expression within severe hypoxia by in situ hybridization. Cell line experiments (22Rv1, PC3) were conducted to determine whether miRNAs and target genes were regulated directly by hypoxia. Most of them were hypoxia-unresponsive, and probably regulated by other mechanisms such as MYC activation. In conclusion, in aggressive, hypoxic prostate tumors, cancer cells exhibit different proliferative gene expression programs that is regulated by miRNAs and depend on whether the cells reside in moderate or severe hypoxic regions. © 2024 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Introduction

Tumor hypoxia is an adverse factor in prostate cancer and is associated with an increased risk of recurrence after radiotherapy and prostatectomy [1-3]. The gene expression program of hypoxic prostate tumors has been shown to reflect disease aggressiveness [2, 4, 5]. However, mechanisms regulating this program are not understood. Such knowledge may facilitate the development of novel treatment strategies and is urgently needed [6]. MicroRNAs (miRNAs) are short, non-coding RNA molecules that control cellular responses to environmental stresses like hypoxia by repressing gene expression [7]. This can happen through a direct regulation of miRNAs when cells are exposed to hypoxia. In addition, miRNAs that are constitutively regulated in the tissue, as a consequence of selection pressures during tumor evolution, can provide a survival benefit under hypoxic conditions [8]. miRNAs downregulate gene expression by degrading mRNAs or inhibiting their translation into proteins. Each miRNA can target many genes in large regulatory networks, often in cross-talk with transcription factors such as the androgen receptor (AR) and the MYC oncoprotein [9, 10]. miRNAs have been studied extensively in prostate cancer and seem to play a role in disease progression [11]. However, only a few selected candidates have been addressed in relation to hypoxia in patient tumors [12].

During tumor evolution, hypoxia exerts a selection pressure on cancer cells, allowing only cells adapted to the deprived condition to survive [13]. The adaptation results from changes in gene expression, partly controlled by miRNAs. A key mediator of this adaptation is the hypoxia-inducible transcription factor HIF1A [14]. This protein regulates numerous miRNAs, including miR-210, which is upregulated under hypoxia and has been associated with prostate cancer progression [15]. The adaptation is likely to depend on the hypoxia level, which ranges from mild, almost non-hypoxic, to severe and anoxic levels within tumors [16]. Exposure of cells to mild and moderate hypoxia leads to enhanced transcription of selected genes that increases glucose uptake and sustains proliferation. At more severe levels, the activities are suppressed to conserve energy and protect cells against lethal effects, often leading to a quiescent state [17]. This knowledge largely comes from studies in monolayer cell cultures, which have been adjusted to an ambient oxygen concentration of air and exposed to acute hypoxic stress for 1–3 days [16]. Such experiments have identified several hypoxia-responsive miRNAs [18], but failed to discover constitutively regulated miRNAs with a pro-survival function. Clinical verification, where cancer cells are examined in their true tissue context and have adapted to a harsh hypoxic microenvironment over time, is lacking. Studies in patient tumors are, however, hampered by the lack of technologies to assess hypoxia levels and identify biological processes within the individual levels.

We aimed to determine the regulation of miRNAs at different hypoxia levels in prostate tumors. In recent work, we developed methodology to extract hypoxia levels from high-resolution, digitized images of histological sections stained for the hypoxia marker pimonidazole [19, 20]. In the present study, we used this approach on tumor biopsies of 83 prostate cancer patients receiving pimonidazole prior to prostatectomy to retrieve a distribution of hypoxia levels for each patient. This distribution was combined with large-scale miRNA and gene expression data to identify miRNAs and predict their target genes at individual hypoxia levels. Cell line experiments were conducted to determine whether the patient-derived miRNAs were regulated directly by exposure to hypoxia. We developed technology to co-register multiple biopsy sections and quantify the co-localization of histopathology parameters with individual hypoxia levels. Based on this technology and predictions of miRNA function, we identified biological traits of cancer cells residing within hypoxic regions and showed that these traits depend on the hypoxia level.

Materials and methods

More details are provided in Supplementary materials and methods.

Patient cohort

The study included 95 prostate cancer patients referred for radical prostatectomy (supplementary material, Table S1). Pimonidazole hydrochloride (Hypoxyprobe Inc., Burlington, MA, USA; 500 mg/m²) was administered intravenously or orally to 83 patients 13–24 h prior to surgery [2]. Diagnostic procedures, grossing, biopsy sampling, treatment, and follow-up for biochemical recurrence (BCR) were performed according to standard procedures [2, 5, 20, 21]. An overview of the data generated from all biopsies is provided in supplementary material, Figure S1. Written informed consent was obtained for all patients. The study was conducted in accordance with the Declaration of Helsinki and approved by the institutional review board and the Regional Committee for Medical Research Ethics South East Norway (2010/1656).

Cell line experiments

The human prostate cancer cell lines 22Rv1 and PC-3 were used. These cell lines were selected to cover a broad range of tumor features, including origin (primary, metastatic) [22], HIF1A expression under normoxia [23], and oxygen consumption and extracellular acidification rate (supplementary material, Figure S2).

miRNA sequencing

Small RNA sequencing was performed for 91 patient biopsies and cell lines grown under hypoxia or normoxia, using the Illumina HiSeq 2500 platform (Illumina Inc., San Diego, CA, USA) for single-end read sequencing of cDNA, as described elsewhere [24]. Raw counts were normalized using trimmed mean of M in edgeR [25, 26]. For patients, miRNAs with a read count/million (rcpm) less than 10 across all patients were removed, yielding 378 candidates for analysis. In cell lines, miRNAs with an average ratio of at least 1.5 between hypoxia- and normoxia-treated samples were selected, according to the range of 1.5- to 4.0-fold variation expected for miRNAs [27]. Upregulated miRNAs with less than 10 rcpm in the hypoxia-treated samples, and downregulated miRNAs with less than 10 rcpm in the normoxia samples were excluded. Log₂-transformed data were used in the analyses. The data are available in NCBI's Gene Expression Omnibus (GEO) under GEO Series accession numbers GSE256390 and GSE256391.

Gene expression

Gene expression profiling was performed for all 95 patient biopsies and cell lines grown under hypoxia or normoxia, using the Illumina BeadArray HT-12 v4 platform (Illumina Inc., GSE178631, GSE259218). RT-qPCR was performed for MYC and PTEN in biopsies from 88 patients, and for miR-210-3p in cell lines.

Digital histopathology

Immunohistochemistry for pimonidazole, Ki67, and PTEN, and in situ hybridization of miR-210-3p were performed on adjacent biopsy sections. Sections were analyzed using a digital histopathology platform in MATLAB (MathWorks Inc., Natick, MA, USA). Hypoxic fraction and PTEN-negative fraction were quantified relative to the parenchymal region, while the Ki67 proliferation index (labeling index) was calculated as the number of Ki67-positive cells relative to the total number of cells (supplementary material, Figure S3) [20]. The hypoxic fraction (HF_Pimo) for each of 40 hypoxia levels was extracted from the digitized pimonidazole-based images of each biopsy, as previously described [20]. This data set was used to identify biological features associated with individual hypoxia levels.

miRNA target genes

Target genes of each miRNA (miRNA–mRNA pairs) were predicted by selecting genes with a significant correlation (Spearman p < 0.05) between their expression and hypoxic fraction at the pimonidazole staining intensity level where the miRNA was regulated. Among these genes, candidate miRNA–mRNA pairs were identified based on an inverse correlation between their expression levels, using Spearman correlation analysis with a false discovery rate (FDR) < 0.1. Those candidate pairs that were also predicted in the miRGate database [28] were considered as miRNA–mRNA pairs. In miRGate, both validated and computationally predicted targets from several databases were considered.

External patient data

The Cancer Genome Atlas data for prostate cancer (TCGA-PRAD) [29] were used for validation of the results. The data included miRNA and gene expression levels, assessed using Illumina HiSeq RNA-seq v2, from the primary prostate adenocarcinomas of 490 and 497 patients, respectively. Biochemical relapse indicator was used as the outcome parameter and was available for 417 (miRNA expression) and 423 (gene expression) patients. The data of all but one of our hypoxia-related miRNAs and all their target genes were available. A 32-gene expression signature by Ragnum et al [5] was used to reflect tumor hypoxia in this cohort. The Ragnum hypoxia signature has been shown to correlate with pimonidazole-defined hypoxic fraction in the FuncProst cohort [2, 5].

Statistics

Spearman rank and Pearson bivariate correlation analyses were used to evaluate associations between parameters. The Mann–Whitney U-test was employed to compare data across groups. Scores were calculated for miRNAs or genes by averaging their median-centered, log₂-transformed expression level. Gene set enrichment analysis (GSEA) was performed by the Molecular Signatures Database (MSigDB) tool [30], with 50 cancer hallmark gene sets and the PTEN inactivation gene set from Markert et al [31]. FDR was controlled by Benjamini–Hochberg's algorithm [32]. Cox univariate analysis was performed to evaluate prognostic significance, and Kaplan–Meier curves were compared using log-rank tests.

Results

Pimonidazole-based images reveal spatial heterogeneity in tumor hypoxia levels

Pimonidazole-stained histological sections were examined to define a proper range of staining intensities for the analyses. Necrotic regions were used as the morphological endpoint of severe hypoxia. Gradients in the brown-colored intensities were observed, with a strong intensity close to necrotic regions that decreased with increasing distance from necrosis (Figure 1A). An intensity range from 0.43 (weak) to 1.53 (strong) was identified as appropriate. An intensity of 0.43 represented the detection limit of ~1.3% O₂ for pimonidazole binding [33, 34] and was termed moderate hypoxia, consistent with a definition proposed by others [35]. Intensities around 1.00 and higher appeared dark brown and were defined as severe hypoxia. Only a few patients’ tissues showed intensities above 1.53.

Figure 1

Construction of high-resolution, digitized images of hypoxia levels. (A) Pimonidazole staining intensity in a histological subsection from a prostate cancer biopsy versus distance from necrosis. Mean intensity along the vertical axis of the above subsection is plotted. The corresponding color-coded, digitized image of the subsection is also shown above. (B) Pimonidazole-stained histological section of a prostate cancer biopsy (left) and corresponding color-coded, digitized image of hypoxia levels (HL_Pimo) (right). (C) Binary pimonidazole-based image of the section in B visualizing pixels in hypoxic regions (white) for four different HL_Pimo thresholds. (D) Pie charts showing the fractions of pixels with HL_Pimo within the indicated intervals for four biopsies with different distributions of hypoxia levels.

Pimonidazole-based images of hypoxia levels (HL_Pimo) were constructed for all 83 patients who had received pimonidazole (Figure 1B). By varying the intensity threshold from 0.43 to 1.53 (Figure 1C), a set of 40 hypoxic fractions was generated for each biopsy. Each fraction represented the amount of hypoxia above the intensity threshold (% of parenchyma > HL_Pimo). Considerable differences in the hypoxia levels were observed within and across tumors. Pie charts of each biopsy were generated, visualizing these differences for seven HL_Pimo intervals from non-hypoxic to severe hypoxia (Figure 1D and supplementary material, Figure S4). Most patients (80 out of 83) had more than 1% hypoxia (HL_Pimo > 0.43; range 0.1–84% hypoxia), while severe hypoxia (HL_Pimo > 1.00) of more than 1% was observed in 53 patients (range 0–62% severe hypoxia).

miRNAs with prognostic impact are associated with distinct hypoxia levels

miRNAs associated with hypoxia at defined pimonidazole-based levels were identified by correlating their expression profiles with the above set of 40 hypoxic fractions across all biopsies. Using a strict significance level of FDR < 0.1 across five or more consecutive hypoxia levels, we identified 27 negatively correlated miRNAs (downregulated in hypoxic tumors) and eight positively correlated miRNAs (upregulated in hypoxic tumors). Each miRNA was assigned to the hypoxia level with its strongest correlation (Figure 2A and supplementary material, Figure S5), reflecting the level where upregulation was the highest or downregulation the lowest. Two distinct groups appeared: a group of seven miRNAs assigned to moderate hypoxia (0.46 < HL_Pimo < 0.71) and another group of 28 miRNAs assigned to more severe levels (0.98 < HL_Pimo < 1.53; Figure 2B and supplementary material, Table S2). All eight upregulated miRNAs were assigned to severe hypoxia.

Figure 2

Correlation of pimonidazole-based hypoxia levels with miRNA and gene expression. (A) Spearman correlation coefficient (𝜌) from analysis of hypoxic fraction versus miRNA expression, plotted as a function of HL_Pimo for a negatively correlated (left) and a positively correlated (right) miRNA (80 patients). The white circle indicates the hypoxia level with the strongest correlation; the stippled line shows the limit for a significant correlation (p = 0.05). (B) Spearman correlation coefficient (𝜌) versus HL_Pimo from the analysis in A for all 378 miRNAs in the data set. Significant miRNAs are indicated in red (eight upregulated) and blue (27 downregulated). (C) Overlap of predicted target genes for downregulated (blue) and upregulated (red) miRNAs at moderate and severe hypoxia. (D) Word clouds visualizing the number of target genes for each downregulated (blue) and upregulated (red) miRNA, clustered based on assigned hypoxia level. The number of target genes is reflected by the size of the miRNA name. Each word cloud includes seven (moderate, down), 20 (severe, down), and eight (severe, up) miRNAs. (E) Representative western blots for HIF1A in 22Rv1 (upper) and PC-3 (lower) cells exposed to the indicated hypoxia levels, confirming exposure of cells to decreasing O₂ concentration by increasing HIF1A stability. (F) Overlap of patient-derived miRNAs (n = 35) and miRNAs responsive to at least one of the hypoxia levels in E in either 22Rv1 or PC-3 cells (n = 146). Up- and down-regulated miRNAs are indicated in red and blue, respectively. (G) miR-210-3p expression in 22Rv1 (left) and PC-3 (right) cells at the same hypoxia levels as in E measured by sequencing or RT-qPCR (average of two replicates, log₂-transformed ratios of hypoxic versus normoxic expression). The stippled line shows the limit for a 1.5-fold upregulation. (H) Ragnum hypoxia score versus scores based on the downregulated and upregulated miRNAs (490 patients, TCGA-PRAD). (I) BCR-free survival for patients with high (green) and low (black) miRNA scores (417 patients, TCGA-PRAD), stratified by 71st percentile (miRNAs down) and median (miRNAs up). p value in log-rank test and the number of patients at risk are indicated.

Target genes were predicted for all miRNAs, yielding 147 unique genes for the downregulated miRNAs and 96 unique genes for the upregulated miRNAs (supplementary material, Table S3). The strong inverse correlation between miRNA and target gene expression supported the hypothesis that the miRNAs had a functional role in prostate cancer. The number of target genes per miRNA ranged from 2 to 45. Eleven genes were unique at moderate hypoxia, while 199 genes were unique at severe hypoxia (Figure 2C). Consequently, miRNA regulation seemed to be more extensive in tumors with severe hypoxia, affecting genes not targeted at moderate hypoxia. This difference was clearly visible in word cloud plots, where the number of target genes of each miRNA is represented by the font size of its name (Figure 2D). Only four miRNAs (miR-210-3p, miR-21-3p, miR-128-3p, miR-30c-5p) overlapped with miRNAs regulated in cell lines exposed to a range of hypoxia levels (Figure 2E,F), and only six out of in total 243 patient-derived target genes were responsive to hypoxia (supplementary material, Figure S6 and Table S4). The cell line studies confirmed a strong upregulation of miR-210-3p at severe hypoxia, both by miRNA sequencing and by RT-qPCR (Figure 2G).

Scores based on miRNAs, or their target genes, correlated significantly with BCR in the COX univariate analysis of continuous data (supplementary material, Table S5) and in the log-rank test comparing dichotomous classification of patients (supplementary material, Figure S7). In the TCGA-PRAD cohort, a negative correlation between miRNA and target gene scores was confirmed (supplementary material, Figure S8A). Both scores showed a significant correlation with the Ragnum hypoxia score (Figure 2H and supplementary material, Figure S8B), validating their association with hypoxia. Moreover, the scores correlated with clinical characteristics such as International Society of Urological Pathology (ISUP) grade group and with BCR (Figure 2I and supplementary material, Figure S8C,D and Table S5). miRNA regulation, therefore, reflected aggressive characteristics of hypoxic tumors.

miRNA expression is associated with increased cell proliferation and MYC expression in hypoxic tumors

Biological processes regulated by the miRNAs were predicted by subjecting their target genes to GSEA. To enhance the statistical power, all target genes of the 27 downregulated miRNAs were collectively analyzed, irrespective of their assigned hypoxia level. The most significant enrichment was found for gene sets related to cell proliferation, such as E2F targets, glycolysis, MYC activation, G2M checkpoint, and PTEN inactivation (Figure 3A). Target genes of the eight upregulated miRNAs showed enrichment of UV response down, androgen response, and adipogenesis (Figure 3A). Inspection of the target genes within each gene set indicated an association between miRNA regulation and high proliferation activity (supplementary material, Table S6).

Figure 3

GSEA of target genes. (A) Enriched gene sets (FDR < 0.1) in the list of 147 and 96 unique predicted target genes of the downregulated and upregulated miRNAs, respectively (left), and the number of miRNAs with targets in each gene set (right). Down- and up-regulated miRNAs are indicated in red and blue, respectively. (B) Biopsy section stained for the proliferation marker Ki67. (C) p value in Spearman correlation of Ki67 proliferation index versus HF_Pimo (83 patients), plotted for increasing values of HL_Pimo (increasing hypoxia severity level). The dotted line indicates p = 0.05. (D) miRNA scores versus Ki67 proliferation index (91 patients). (E) miRNA scores versus a proliferation score based on E2F target and G2M checkpoint gene sets (490 patients, TCGA-PRAD). (F) MYC expression by RT-qPCR versus miRNA and target scores of the MYC target V1&V2 gene sets (88 patients). (G) MYC expression by RT-qPCR versus Ki67 proliferation index (88 patients). (H) p value in Spearman correlation of MYC expression by RT-qPCR versus HF_Pimo (77 patients), plotted for increasing values of HL_Pimo. The dotted line indicates p = 0.05. (F, G) Log₂-transformed data for MYC expression are used. (D–G) Pearson correlation coefficient and p value are indicated.

This hypothesis was further examined in Ki67-stained histological sections (Figure 3B). Analysis of the Ki67 proliferation index against the set of 40 hypoxic fractions utilized in Figure 2 showed a strong correlation for HL_Pimo > 0.49, and the correlation strengthened with increasing hypoxia severity level (Figure 3C). Tumors showing a high Ki67 proliferation index were therefore likely to contain hypoxic regions. The Ki67 proliferation index also correlated with the miRNA and target gene scores and with the expression of most individual miRNAs (Figure 3D and supplementary material, Figure S9A,B). The relationship between miRNA expression and proliferation was validated in TCGA-PRAD, using a proliferation score based on 13 unique target genes in the E2F targets and G2M checkpoint gene sets (Figure 3E and supplementary material, Table S6). Altogether, increased cell proliferation seemed to be a dominant biological process associated with miRNA regulation at all hypoxia levels.

The enriched MYC activation gene sets (MYC targets V1&V2) involved miRNAs that were downregulated at both moderate and severe hypoxia (Figure 3A and supplementary material, Table S6). There was a significant correlation between MYC expression by RT-qPCR and the scores of miRNAs or target genes in these gene sets (Figure 3F). MYC expression also correlated with Ki67 proliferation index and with hypoxic fraction at all hypoxia levels (Figure 3G,H). Taken together, these findings validated the GSEA results based on bead array and sequencing technologies and support a role for MYC in the increased proliferation activity of tumors with miRNA downregulation.

Tile analysis identifies overlap between hypoxia and cell proliferation in aggressive tumors

To determine whether increased cell proliferation not only was associated with hypoxia-related miRNAs but also occurred within hypoxic regions, we developed technology to quantify the co-localization of histopathology parameters (Supplementary materials and methods). This technology assessed the spatial localization of histopathology parameters in relation to individual hypoxia levels. An algorithm for accurate co-registration of sections was constructed, ensuring exact spatial overlap of the sections (supplementary material, Figure S10). For calculation of co-localization indices, hypoxia and histopathology parameters were quantified in tile representation (supplementary material, Figure S11) with microregions of 0.16 × 0.16 mm² (350 × 350 pixels²). A co-localization index was defined for each biopsy as the Pearson correlation coefficient in a tile-wise correlation analysis of the parameters (supplementary material, Figure S12).

Proper co-registration of pimonidazole- and Ki67-stained sections was obtained for 76 biopsies (supplementary material, Figure S4), and tile representation of the hypoxic fraction and Ki67 proliferation index was generated for each of these biopsies (Figure 4A). The co-localization index was calculated for all hypoxia levels combined (${\mathit{HL}}_{\text{Pimo}}=\left[0.43,\to ⟩\right)$ and for moderate (${\mathit{HL}}_{\text{Pimo}}=\left[0.43,1.00⟩\right)$ and severe (${\mathit{HL}}_{\text{Pimo}}=[1.00,\to ⟩$) hypoxia, corresponding roughly to the levels assigned to miRNAs. Significant co-localization was found for 38 biopsies (50%) across all levels (Figure 4B) and for 36 and 27 biopsies at moderate and severe hypoxia, respectively (supplementary material, Figure S13A,B). Biopsies with co-localization had a higher hypoxic fraction, both when considering all hypoxia levels combined and when considering severe hypoxia alone (Figure 4C and supplementary material, Figure S13C). There was no difference in the Ki67 proliferation index. The co-localization index was, therefore, not only an indicator of increased cell proliferation but rather identified biopsies where proliferating cells were located within hypoxic regions. Visual inspection confirmed overlap between pimonidazole- and Ki67-staining at moderate and severe hypoxia in histological sections (Figure 4D). Moreover, patients with a co-localization index higher than 0.15 (p = 0.001) had a higher risk of BCR compared with the other patients (Figure 4E). No prognostic significance was found for the hypoxic fraction at any hypoxia level (data not shown).

Figure 4

Tile analysis of co-registered pimonidazole- and Ki67-stained biopsy sections. (A) Pimonidazole- and Ki67-stained co-registered histological sections from a selected biopsy (left) and hypoxic fraction and Ki67 proliferation index in tile representation of the same sections (right). Tile size is 0.16 × 0.16 mm² (350 × 350 pixels²). (B) p value in tile-wise Pearson correlation analysis of hypoxic fraction versus Ki67 proliferation index as a function of the co-localization index at ${\mathit{HL}}_{\text{Pimo}}=[0.43,\to ⟩$ (76 patients). The dotted lines indicate significant co-localization (p < 0.05). The number of biopsies with significant and non-significant co-localization is indicated. (C) Hypoxic fraction at ${\mathit{HL}}_{\text{Pimo}}=[0.43,\to ⟩$ (upper) and Ki67 proliferation index (lower) for biopsies stratified by the co-localization index in B. The boxes extend from the first to the third quartile, with median value indicated. (D) Subsection of pimonidazole staining (left), Ki67 staining (middle), and the overlap between the two parameters (right) at moderate hypoxia (upper) and severe hypoxia (lower). The sections are from two different biopsies. (E) BCR-free survival for 76 patients stratified by a co-localization index of 0.19 (p = 0.0002 in B), which provided the strongest prognostic value. p value in log-rank test and the number of patients at risk are indicated.

PTEN is downregulated in regions with severe hypoxia and increased cell proliferation

The enriched PTEN inactivation gene set exclusively involved miRNAs with their strongest downregulation at severe hypoxia (Figure 3A and supplementary material, Table S6). Moreover, PTEN was part of the UV response down gene set and predicted the target gene of three miRNAs upregulated at severe hypoxia (miR-5680, miR-708-3p, miR-95-3p) (supplementary material, Table S3). A weak, but significant, correlation was found between PTEN expression assessed using RT-qPCR and the scores for miRNAs or target genes in the PTEN inactivation gene set (Figure 5A and supplementary material, Figure S14A). Moreover, low PTEN expression was associated with a high Ki67 proliferation index (supplementary material, Figure S14A). In the analysis of PTEN expression against the set of 40 hypoxic fractions utilized in Figure 2, an association between PTEN downregulation and severe hypoxia (HL_Pimo > 1.31) was confirmed (Figure 5B). Moreover, the PTEN-negative fraction by immunohistochemistry correlated with the scores of miRNAs assigned to severe hypoxia (supplementary material, Figure S14B). These findings supported the GSEA results and suggested that a miRNA-related PTEN depletion primarily occurred in tumors with severe hypoxia.

Figure 5

Tile analysis of co-registered pimonidazole- and PTEN-stained sections. (A) PTEN expression by RT-qPCR versus a score based on miRNAs in the PTEN inactivation gene set (88 patients). Pearson correlation coefficient and p value are indicated. (B) p value in Spearman correlation analysis of PTEN expression by RT-qPCR versus HF_Pimo (77 patients), plotted for increasing values of HL_Pimo. The dotted line indicates p = 0.05. (C) Pimonidazole- and PTEN-stained co-registered histological sections from a selected biopsy (left) and hypoxic fraction and PTEN-negative fraction in tile representations (right). Tile size is 0.16 × 0.16 mm² (350 × 350 pixels²). (D) p value in tile-wise Pearson correlation analysis of hypoxic fraction versus PTEN-negative fraction as a function of the co-localization index at ${\mathit{HL}}_{\text{Pimo}}=[1.00,\to ⟩$ (44 patients). The dotted lines indicate significant co-localization (p < 0.05). The number of biopsies with significant and non-significant co-localization is indicated. (E) Hypoxic fraction for biopsies stratified by the co-localization index at ${\mathit{HL}}_{\text{Pimo}}=[1.00,\to ⟩$. The boxes extend from the first to the third quartile, with median value indicated. (F) p value in tile-wise Pearson correlation analysis of Ki67 proliferation index versus PTEN-negative fraction as a function of the co-localization index for the two parameters (39 patients). (G) Pimonidazole-stained (upper) and PTEN-stained (lower) subsection of the biopsy in C with visualization of hypoxia levels and segmented PTEN-negative regions.

Tile analysis of PTEN- and pimonidazole-stained sections was performed in 44 patients with satisfactory co-registration (Figure 5C). Significant co-localization of PTEN depletion and severe hypoxia (${\mathit{HL}}_{\text{Pimo}}=[1.00,\to ⟩$) was found for 11 biopsies (25%), while this number was lower for moderate hypoxia or for all levels combined (Figure 5D and supplementary material, Figure S15A,B). Biopsies with co-localization had a higher hypoxic fraction, and the difference was most pronounced at severe hypoxia (${\mathit{HL}}_{\text{Pimo}}=[1.00,\to ⟩$) (Figure 5E and supplementary material, Figure S15C). Moreover, a high Ki67 proliferation index co-localized with PTEN depletion in 5 out of 39 biopsies (15%) with satisfactory co-registration of these parameters (Figure 5F). Visual inspection confirmed PTEN depletion within regions of severe hypoxia in histological sections (Figure 5G).

miRNA expression co-localizes with hypoxia and increased cell proliferation

To strengthen our hypothesis of a role of miRNAs in the overlap between hypoxia and cell proliferation, miRNA scores were correlated with the co-localization indices presented in supplementary material, Figure S13A. A significant correlation was found at moderate and severe hypoxia levels for the corresponding miRNA scores (Figure 6A). Therefore, miRNA regulation occurred primarily in tumors with overlap between hypoxia and cell proliferation. As a proof of concept, we performed in situ hybridization to localize miR-210-3p in a tumor with pronounced upregulation of this miRNA. Heterogeneous expression of miR-210-3p was observed within the biopsy with notable co-localization in some regions of severe hypoxia (Figure 6B,C), consistent with the assigned hypoxia level of this miRNA. Proliferating cells were also seen in these regions.

Figure 6

Overlap between hypoxia, cell proliferation, and miRNA expression. (A) Co-localization index for Ki67 proliferation index versus hypoxic fraction plotted against scores based on seven downregulated miRNAs at moderate hypoxia (left), 20 downregulated miRNAs at severe hypoxia (middle), and eight upregulated miRNAs at severe hypoxia (right; 73 patients). Data are shown for ${\mathit{HL}}_{\text{Pimo}}=[0.43,\to ⟩$ (right) and ${\mathit{HL}}_{\text{Pimo}}=[1.00,\to ⟩$ (middle, left) to match the assigned level for each group of miRNAs. Pearson correlation coefficient and p value are indicated. (B) Subsection of pimonidazole-stained (left), Ki67-stained (middle), and miR-210-3p-stained (right) histological section. The insets below show part of the section at higher magnification. (C) Overlap between pimonidazole and Ki67 staining (upper) and pimonidazole and miR-210-3p staining (lower). Ki67-positive cells within miR-210-3p-positive region is shown at high magnification in the circular inset.

Discussion

The present work reports a spatial, histopathology approach to uncover the regulation of miRNAs at individual hypoxia levels in prostate tumors. The distributions of hypoxia levels generated at high resolution constitute a unique data set and new insight into the inter- and intra-tumor heterogeneity in prostate cancer. Hence, hypoxia is almost exclusively considered as a binary metric in work on patient tumors, presented as the absence or presence of hypoxia [16]. By using novel technology for quantitative co-localization analysis of histopathology parameters in tile representation, we verified that the regulation of miRNAs not only was associated with hypoxia but occurred within hypoxic regions at the predicted severity level. This is an important, but largely lacking, documentation for patient tumors, where conclusions are often based on correlation analyses. Here, we utilized correlations to generate hypotheses that were verified in co-localization studies. This showed that miRNAs are regulated differentially within hypoxic regions, depending on the severity level.

miRNA regulation was associated with a proliferative hypoxia phenotype, where increased proliferation was observed in both regions of moderate and severe hypoxia. Downregulation of let-7c and miR-30 family members and upregulation of miR-21 have led to increased proliferation in cell lines of different cancer types [36-38], supporting our finding. However, cell proliferation is reduced in hypoxic regions of bladder, head and neck, and uterine cervical tumors [39-42]. This suggests that the biology of hypoxic tumors differs, depending on the cancer type. High cell proliferation may be directly linked to the development of hypoxia in prostate cancer, through an elevated oxygen demand of proliferating cancer cells. This is likely, due to the frequent co-localization of cell proliferation with hypoxia at all severity levels.

High MYC expression and depletion of PTEN were characteristics of the proliferative hypoxia phenotype. These aberrations promote oncogenic signaling and are frequently observed in prostate cancer [43]. In agreement with our result, work at the gene level has shown enrichment of MYC gain and PTEN loss in hypoxic prostate tumors [12]. PTEN was a predicted target of miR-5680, miR-708, and miR-95, which were upregulated at severe hypoxia. Moreover, miR-21, which was upregulated, has been shown to inhibit PTEN [44]. In addition to genetic loss, miRNA-mediated regulation thus seems to be a mechanism of PTEN depletion in hypoxic tumors, being most prominent within regions of severe hypoxia. MYC can also be targeted by miRNAs identified in our work, such as let-7c, miR-135, and miR-200c [45], and miRNA repression by MYC has been reported for miR-30e, miR-30c, let-7c, miR-29b, and miR-99a [46]. This is consistent with MYC activation in tumors with downregulation of these miRNAs and suggests a role of this aberration to sustain proliferation in hypoxic regions at all severity levels.

Most of the miRNAs identified in our work have been associated with prostate cancer in previous studies [11, 47, 48], strengthening the reliability of the results [27]. Associations with hypoxia in patient tumors have, however, not been reported, except for a few of these miRNAs [12, 49]. Our cell line experiments identified only four of the miRNAs as hypoxia-responsive, including miR-210-3p and miR-21-3p, which have been found previously in prostate cancer cell lines [49-52], and miR-128-3p and miR-30c-5p, which appear to be novel hypoxia-responsive miRNAs in this disease. Work on cell lines from different cancer types has demonstrated a direct response to hypoxia for only a few of the remaining miRNAs (miR-29b, miR-128, miR-224, miR-424, let-7c) [18, 53]. Most target genes were also hypoxia-unresponsive, consistent with observations in other prostate cancer cell lines [5]. The majority of our miRNAs therefore seemed to be regulated by hypoxia-independent mechanisms, such as MYC- or AR-mediated suppression [46, 54-57]. These observations support a hypothesis where constitutively regulated miRNAs are involved in the cell's adaptation to hypoxia, in addition to hypoxia-responsive ones. This strategy is widely used by microbes for adjustment of gene expression to ensure growth in a changing environment [58, 59]. The strategy has been suggested for miRNA regulation of stress responses in cancer diseases and may be mediated through, for example, oncogenic activation selected for during cancer development [8]. Our analysis, combining clinical and cell line data, suggests that this strategy could be important for the cell's adaptation to hypoxia in prostate cancer. Moreover, it demonstrates the limitation of cell lines in work to understand stress responses in patient tumors.

miRNA and target gene scores and co-localization of hypoxia and cell proliferation showed a stronger prognostic significance than hypoxic fraction alone. The proliferative hypoxia phenotype controlled by our miRNAs is therefore a more aggressive tumor trait than hypoxia by pimonidazole staining. Our previous work, comparing pimonidazole-based hypoxic fraction and the Ragnum hypoxia gene score, led to the same conclusion [2]. Moreover, a study in glioblastoma showed worse outcome for patients with proliferating cells within hypoxic regions [60]. The strong prognostic significance might reflect a combined effect of several risk factors, as suggested for the combination of hypoxia and genomic instability [61]. Upregulation of MYC at all hypoxia levels and depletion of PTEN in severe hypoxia likely play a role in this aggressive tumor trait. Both MYC gain-of-function and PTEN loss are considered to be key genetic events in the progression of prostate cancer from castration-sensitive to castration-resistant disease [43], consistent with this hypothesis.

Our study may have implications for the choice of intensified therapy for patients with hypoxic tumors. Drugs reducing cell proliferation seem to be more relevant than strategies to increase oxygen supply. Promising approaches include inhibition of the upregulated miRNAs or restoring the expression of downregulated miRNAs by oligonucleotides or virus-based constructs [62, 63]. miRNA regulation may also be inhibited indirectly by targeting their downstream signaling [63]. Altogether, this could be the basis for novel drugs to overcome hypoxia-related tumor aggressiveness in prostate cancer.

Acknowledgements

This study was supported by grants from The South-Eastern Norway Regional Health Authority (grant number 2020044) and The Norwegian Cancer Society (grant number 182451). Technical assistance from D Trinh and M Nguyen, Department of Pathology, is highly appreciated.

Author contributions statement

VES and HL conceived and designed the study. VES, UBS, TH, CSF, KAMF, HH, HBR, LV, KHH and TS developed the methodology and analyzed data. HL supervised the study. VES, UBS and HL drafted the manuscript. All authors revised the manuscript and gave their approval of the final version.

Open Research

Data availability statement

The large-scale data are publicly available in GEO under GSE256390, GSE256391 (miRNA expression; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE256390, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE256391) and GSE178631, GSE259218 (gene expression; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE178631, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE259218).