MicroRNA Signatures of Prostate Cancer Spheroids in Microfluidic Culture Under Hormone-Deprivation Conditions

TL;DR

This study uses a microfluidic model to identify microRNA changes in prostate cancer cells under hormone-deprivation conditions, revealing early signs of aggressive cancer progression.

Contribution

A novel microfluidic PCa model is introduced to capture early molecular events in the transition to castration-resistant prostate cancer.

Findings

Sustained hormone deprivation alters microRNA expression, reducing tumor-suppressive miRNAs.

Downregulated miRNAs target genes involved in neuronal differentiation pathways linked to neuroendocrine-like features.

A refined gene set from miRNA targets was used to identify shared regulatory pathways during disease progression.

Abstract

Background: Prostate cancer (PCa) is prevalent in men over 65 and requires effective clinical management. Standard PCa therapies often offer positive outcomes; however, its castration-resistant form (CRPC) is aggressive and associated with poor prognosis. The objective of this study is to characterize the microRNA profiles associated with the PCa to CRPC transition using a microfluidic PCa model. Methods: LNCaP-derived hormone-sensitive PCa spheroids were cultured for 30 days under recirculating flow conditions mimicking hormone deprivation. Total RNA was isolated from the spheroids and perfusate at Day 5 and Day 30. Exosomal microRNAs were profiled by miRNA-seq. Differentially expressed miRNAs were used for target prediction across multiple databases, and gene set enrichment analysis (GSEA) was performed to identify pathways affected during prolonged hormone deprivation. Results: Sustained hormone deprivation induced a shift in microRNA expression. Tumor-suppressive miRNAs were broadly reduced. To evaluate functional consequences, predicted targets were compiled for all regulated miRNAs. For the 33 intracellular miRNAs downregulated on Day 30, 430 genes were predicted as targets for at least 16 of these miRNAs, revealing strong convergence on shared regulatory pathways. Thirty-five genes overlapped with predicted targets of the single upregulated miRNA and were removed, yielding a refined set of 395 unique genes used for GSEA. Overall, the neuronal differentiation pathways observed reflect early features of a neuroendocrine-like phenotype. Conclusions: This microfluidic PCa model captures early molecular events associated with progression toward CRPC. It provides a controlled system for studying disease evolution and supports the development of more precise therapeutic and diagnostic strategies.

Figures

Full text

1. Introduction

Prostate cancer (PCa) remains one of the most common malignancies among men worldwide [1]. Disease progression is characterized by distinct pathophysiological transitions, from benign prostatic hyperplasia (BPH) to hormone-sensitive prostate cancer (HSPC) and ultimately to castration-resistant prostate cancer (CRPC), with profound changes in underlying cellular mechanisms [2,3]. Among the molecular regulators implicated in PCa survival and proliferation, microRNAs (miRNAs) have emerged as key pathway regulators involved in gene expression [4], intercellular communication [5], cell proliferation and survival, epithelial–mesenchymal transition (EMT) [6], angiogenesis and immunogenicity [7]. PCa cells are prolific in miRNAs secretion, and several studies have identified individual miRNA associated with disease stage, aggressiveness, and therapy resistance. Considering that miRNAs can be detected in blood or urine, they have been long considered as promising biomarkers for disease diagnosis, prognosis, and therapeutic monitoring, using minimally invasive procedures [8].

The heterogeneity of miR expression across cancers and other pathologies presents a significant challenge to their clinical translation. Non-specific or overlapping expression patterns increase the risk of false positives and complicate the interpretation of miR-based assays [9]. Despite these challenges, a catalog of over 2000 identified human miRNAs, together with advances in isolation and analysis, has enabled the comprehensive characterization of cancer-associated miRNA increasingly feasible [10]. In PCa and other malignancies, miR secretion serves multiple biological functions [11]. Circulating miRNA can promote metastasis by targeting and disrupting tight junctions in vascular endothelium, facilitating tumor cell extravasation. They can modulate the tumor microenvironment by reprogramming surrounding stromal cells, such as cancer-associated fibroblasts (CAFs), to adopt tumor-supportive phenotypes [12]. Secreted miRNAs also enable immune evasion, suppressing cytotoxic responses or reprogramming T cells to adopt immunosuppressive activity and can reprogram the metabolism of host tissue [13,14]. In PCa, miRs have been associated with cell invasion, neuroendocrine trans-differentiation and castration-resistance progression [15,16]. Extracellular miRNAs can be found encapsulated within exosomes, which protect them from degradation in the extracellular space and mediate delivery to target cells [17]. This packaging confers multiple advantages, protects from enzymatic breakdown, mediates efficient intercellular communication and biocompatibility [18]. Exosomal membranes shield miR from immune detection and facilitate their extravasation through vessels and capillaries, mediating metastatic dissemination [19].

MiRNA signatures are defined as combinations of specific miRNAs that collectively indicate a disease state and are emerging as valuable diagnostic and prognostic tools. These signatures function as molecular fingerprints, capturing the pattern of altered miRNA expression characteristic of specific pathological processes [20,21]. Their prognostic potential is exemplified in a recent study using stool-based miRNA profiles for the early detection of colorectal cancer. A panel comprising miR-21-5p and miR-199a-5p, combined with patient age, achieved an 88% sensitivity for colorectal cancer detection. An expanded panel including miR-451a improved discrimination of high-grade dysplasia to 91% sensitivity. When integrated with fecal occult blood testing, these panels achieved 96% sensitivity for detecting high-grade lesions [22]. This approach underscores the diagnostic power of miR profiling relative to single-miRNA analysis. Identifying complex miR expression profiles in PCa rather than individual miRNAs can potentially characterize disease stage and predict the transition from hormone-sensitive to hormone-insensitive (castration-resistant) states [23,24]. Despite significant advances in the characterization of PCa-associated miRNAs, the mechanistic relationships behind their regulation and function remain elusive.

We previously developed a micro-physiology PCa model where cells are cultured under dynamic conditions, mimicking a microenvironment with active fluid circulation. In our system, PCa cells developed a robust epithelial phenotype and displayed prolific miRNA secretion [25]. In the present study, we introduce a new iteration of our microfluidic PCa model in which LNCaP spheroids are cultured for 30 days under androgen-depleted conditions. In this study, we aim to capture early phenotypic changes associated with the transition from hormone-sensitive to androgen-insensitive prostate cancer, in order to investigate microRNA signatures linked to the onset of castration resistance in a physiologically relevant and dynamic culture environment.

2. Materials and Methods

2.1. Cell and Microfluidic Culture

For maintenance and expansion, LNCaP cells (American Type Culture Collection, Manassas, VA, USA) were cultured under standard conditions using RPMI-1640 medium supplemented with sodium pyruvate, 10% (v/v) fetal calf serum (FCS), and 5% (v/v) penicillin–streptomycin (10,000 U/mL). To generate LNCaP spheroids, cells were harvested, resuspended as a single-cell suspension, and seeded at a density of 10,000 cells per well into individual wells of a Biofloat™ plate, allowing for uniform spheroid formation under non-adherent conditions. Spheroids were maintained in low-adhesion plates for three days before further processing. The hydrogel used for spheroid embedding consisted of 4% (w/v) agar mixed with rat tail collagen I (3 mg/mL) and culture medium at a ratio of 100 μL agar, 25 μL collagen, and 75 μL medium, resulting in a final mixture of 200 μL containing 2% agar and 1.1 mg/mL collagen. Eight LNCaP spheroids were collected into a 1.5 mL tube and gently mixed with the hydrogel solution, following the procedure described by Padmyastuti et al. [25]. The hydrogels were allowed to polymerize at 37 °C for 1 h, after which they were removed from the tubes and transferred to 24-well plates (one gel per well) containing 1 mL of standard culture medium.

For microfluidic culture, the Humimic Chip 2 24-well platform (TissUse GmbH, Berlin, Germany) was used. Each experimental condition included three biological replicates, each consisting of three hydrogels (containing a total of 24 spheroids) placed within the 24-well-sized compartment of the chip, with 1 mL of culture media. In this configuration, the adjacent 96-well-sized compartment of the chip was used for medium exchange and sample collection, with a volume of 0.5 mL. The hydrogels were perfused for 30 days under recirculating flow at a frequency of 25 Hz. The overall culture layout and experimental configuration are illustrated in Figure 1. To simulate androgen deprivation, charcoal-stripped FCS (csFCS) was used. For static reference cultures, hydrogels were maintained in conventional 24-well plates, with three hydrogels per well, under identical conditions (Table 1).

2.2. Cell Viability and PSA Secretion

To evaluate cell viability during the androgen-deprivation culture period, lactate dehydrogenase (LDH) release, caspase-3/7 activity, and metabolic activity were assessed. The Promega assay kits used for these analyses are listed in Appendix A, Table A1 For each assay, 25 μL of perfusate (effluent) was collected from the microfluidic circuit and processed according to the manufacturer’s specifications. Prostate-specific antigen (PSA) secretion was quantified using the DuoSet ELISA system (R&D Systems) following the protocol previously reported for PCa microfluidic culture [25].

2.3. Morphological Characterization

Morphological characterization of LNCaP spheroids was performed using immunofluorescent markers associated with prostate cancer phenotype, polarity, adhesion, and cellular activity. These markers provide insight into the maintenance of epithelial organization and differentiation status under androgen-deprivation conditions. Spheroids cultured under static androgen-deprivation conditions were fixed with 4% paraformaldehyde (PFA) and permeabilized with 0.1% Triton X-100 (TXT) in Phosphate Saline buffer (PBS) overnight at 4 °C. Staining buffer consisted of 0.1%TXT in PBS with 2% bovine serum albumin (w/v), as a blocking agent. Washing buffer consisted of 0.1%TXT in PBS. Immunofluorescent staining was carried out according to the conditions listed in Table A2. All markers were co-stained with Phalloidin-488 (f-actin) and Hoechst33342 (nuclei). All primary antibodies were incubated overnight at 4 °C and secondary antibodies the following day for 3 h at room temperature.

2.4. MicroRNA Isolation and Sequencing

For microRNA analysis, perfusate and hydrogel samples were collected from both 3D static and microfluidic cultures. Three hydrogels were combined to form a single sample, which was lysed in 750 µL TRIzol per 250 µL sample volume with 2% Triton X-100. Lysates were homogenized by repeated resuspension through a 20G needle and incubated at room temperature for 20 min. One volume of 100% ethanol was added to the lysate mixture, and total RNA, including microRNAs, was isolated using the Qiagen miRNeasy Mini Kit (Qiagen, Hilden, Germany) following the manufacturer’s specifications [25]. Exosomal small RNA sequencing (sRNA-seq) was performed by Novogene Europe (Munich, Germany), which included library construction with adapter ligation, reverse transcription, amplification, and size selection, followed by single-end 50 bp high-throughput sequencing in the Illumina NovaSeq 6000 high-throughput next-generation sequencing platform. Sequencing reads underwent quality control, mapping to the reference genome, and alignment for quantification of known microRNAs.

2.5. MicroRNA Expression Analysis

Pre-normalized transcript per million (TPM) expression data [26,27] for intracellular and extracellular (liquid phase) microRNAs were analyzed in R (v4.4.3). After import of the TPM matrices, non-numeric entries were converted to missing values and set to zero to ensure consistent data structure. TPM-normalized miRNA expression data from circulating samples (circulating miRNAs) collected at Day 5 (control, n = 3) and Day 30 (n = 2) were compared. Differential expression was assessed using two-sided t-tests for each miRNA (10.2202/1544-6115.1027), and miRNAs with p < 0.05 were considered differentially expressed. TPM-normalized intracellular miRNA expression profiles from Day 5 and Day 30 samples (each n = 2) were compared using two-sided t-tests for each miRNA. miRNAs with p < 0.05 were considered differentially expressed. Fold changes (FCs) were calculated as the ratio of mean expression from Day 30 to Day 5. miRNAs with FC > 1 were classified as upregulated at Day 30, and those with FC < 1 as downregulated. To support the sequencing analyses, the expression of the top deregulated miRNAs was examined in The Cancer Genome Atlas Prostate Adenocarcinoma (TCGA-PRAD) miRNA-Seq dataset: https://portal.gdc.cancer.gov/projects/TCGA-PRAD (accessed on 29 January 2026), with expression levels evaluated relative to the reference miRNA miR-16 [28].

2.6. Target Prediction Analysis

For the differentially expressed miRNAs, target gene prediction was performed using multiple established databases to ensure robust coverage of both conserved and non-conserved interactions [26]. The following resources were included: MicroCosm [29], microRNA.org (conserved and non-conserved targets) [30], miRDB v5 [31], miRecords [32], PicTar [33], PITA (including PITA_all) [34], and TargetScan [35]. Only genes predicted by at least two of these databases were retained for subsequent downstream analyses. Pathway enrichment analysis was performed using the predicted target gene lists derived from miRNAs upregulated or downregulated at Day 30. To identify pathways potentially activated at Day 30, enrichment was calculated for genes targeted by miRNAs that were downregulated at Day 30, as these targets are expected to be transcriptionally upregulated. Before analysis, genes that also appeared in the target list of upregulated miRNAs were excluded to avoid confounding effects. Only miRNA–target interactions supported by at least two independent databases were considered. To validate the expression of key target genes identified in our study, miRNA expression and corresponding mRNA target expression were analyzed in the TCGA-PRAD dataset, with analyses restricted to experimentally validated targets of the three core miRNAs as defined by miRTarBase analysis described in Table 2. Progression-free survival (PFS) analysis was performed using the Gene Expression Profiling Interactive Analysis (GEPIA3) web-based platform to evaluate associations between mRNA expression levels and clinical outcome in the TCGA-PRAD dataset. The analyses were restricted to predicted and experimentally validated miRNA target genes identified and listed in Table 2. A summary of the TCGA-PRAD dataset analysis is described in Appendix C.

3. Results

3.1. PCa Spheroids Exhibit a Differentiated Epithelial Morphology

Morphological analysis using immunofluorescence staining of epithelial markers revealed a high degree of polarization and epithelial organization in LNCaP spheroids. Localization of F-actin almost exclusively to the cell boundaries (Figure 2A) supports this observation. EGFR was primarily localized to the periphery of the spheroids (Figure 2(B1,B2), and cells displayed pronounced expression of ZO-1 (Figure 2(C1,C2)), tubulin (Figure 2D), and E-cadherin (Figure 2E) along their cellular boundaries. Brightfield images already suggested the presence of two distinct structural regions within the spheroids—an outer and an inner area. The spheroids displayed an oblique morphology, and the inner region did not form a compact core, as it was not fully enclosed by the outer layer but remained partially exposed to the external environment. The outer region is estimated to display a thickness between 100 and 150 μm, representing about 60–70% of the spheroid area. F-actin organization clearly delineates these two regions (Figure 2G), while ZO-1 and E-cadherin expression appeared more structured and continuous in the outer region, forming a distinct epithelial boundary most evident in the E-cadherin staining. Fibronectin expression (Figure 2H) was predominantly observed in the outer region, indicating that cell–cell and cell–matrix adhesion in the inner region is comparatively weaker. Live/dead staining (Figure 2I) demonstrated that both regions contained viable cells, with no significant evidence of cell death in the inner compartment.

3.2. Spheroid Viability Is Maintained Under Prolonged Microfluidic and Static Conditions

LNCaP spheroids maintained their viability under both microfluidic and static culture conditions over the 30-day experimental period, as indicated by stable LDH leakage throughout(Figure 3A). Caspase-3/7 activity gradually declined over time in both culture systems, with overall extracellular caspase activity remaining at baseline levels, suggesting no significant induction of apoptosis (Figure 3B). Assessment of dehydrogenase activity using the PrestoBlue assay further confirmed that viability was preserved in static cultures, while microfluidic cultures displayed a marked increase in metabolic activity over time (Figure 3C). Additionally, PSA secretion was elevated in microfluidic cultures relative to static conditions, indicating maintained functional activity (Figure 3D). Collectively, these results demonstrate that the experimental culture conditions employed do not compromise LNCaP spheroid viability or functional integrity.

3.3. Hormone-Deprivation Conditions Alter MicroRNAs Expression Profile

To assess intracellular miRNA expression dynamics under hormone-deprivation conditions, TPM-normalized miRNA profiles from Day 5 and Day 30 samples were compared. One miRNA, hsa-miR-125b-5p, was found to be upregulated on Day 30 (p < 0.05). In contrast, 33 miRNAs and one novel miRNA (novel_262) were downregulated at Day 30 compared to Day 5 (Figure 4A).

These results indicate a predominant downregulation of intracellular miRNAs after prolonged hormone deprivation, suggesting a reduction in post-transcriptional regulatory activity at later stages of adaptation. For the extracellular (liquid phase) samples, TPM-normalized miRNA expression levels from Day 5 and Day 30 were compared. In total, the following eight miRNAs were significantly increased at Day 30 (p < 0.05): hsa-miR-1323, hsa-miR-181a-2-3p, hsa-miR-192-5p, hsa-miR-205-5p, hsa-miR-302b-3p, hsa-miR-372-5p, hsa-miR-516a-5p, and hsa-miR-516b-5p (Figure 4B). No overlap was observed between the significantly deregulated extracellular and intracellular miRNAs. However, several of the extracellularly increased miRNAs showed a similar trend toward higher expression intracellularly, although not reaching statistical significance. The expression of the top downregulated miRNAs in our experimental (miR-139, miR-186, and miR-9) was also found to be significantly lower, compared with the reference miRNA, in the TCGA-PRAD database cohort. Conversely, the upregulated miRNA miR-125b showed significantly higher expression relative to the same reference miRNA.

3.4. Deregulated microRNAs Profiles Are Associated with PCa Cancer Progression

To investigate the functional impact of the observed miRNA expression changes, we performed a comprehensive target prediction and pathway enrichment analysis for the intracellular miRNAs regulated at Day 30. For each differentially expressed miRNA, predicted target genes were retrieved from multiple databases, with the supporting sources documented for every interaction. Target lists were generated separately for miRNAs upregulated and downregulated on Day 30, and additional filtered gene sets were created based on increasing evidence thresholds (no cutoff, ≥2 databases, and ≥3 databases). We focused the pathway analysis on the predicted targets of the 33 intracellularly downregulated miRNAs on Day 30. Across these 33 miRNAs, 430 genes were predicted as targets for at least 16 of the 33 downregulated miRNAs, indicating substantial convergence on shared regulatory nodes. Of these, 35 genes overlapped with the predicted targets of the single upregulated miRNA (hsa-miR-125b-5p). To isolate pathways specifically influenced by the downregulated miRNAs, these 35 shared genes were removed, resulting in a final set of 395 unique target genes. This gene set was used as input for gene set enrichment analysis (GSEA). Experimentally validated targets for the three core miRNAs were identified using miRTarBase (Figure 5A,B). Among predicted targets, experimental validation was available for 4 of 39 targets for miR-9-5p, 10 of 55 targets for miR-186-5p, and 17 of 29 targets for miR-139-5p (Figure 6A,B). Correlation analysis between miRNA and target mRNA expression in PRAD samples revealed multiple significant negative correlations, consistent with functional miRNA–target regulation (Figure 6C). This effect was most pronounced for miR-186-5p, for which the majority of experimentally validated targets also exhibited significant negative correlations. Across all predicted targets listed in Table 2, negative correlations were observed for a substantial proportion of targets for each miRNA. Notably, miR-139-5p also showed strong positive correlations with several targets, including experimentally validated genes such as PDE4D, PPP3CB, ZEB2, FOXP2, DNM3, and FAT3, suggesting the presence of context-dependent regulatory mechanisms or feedback regulation (Appendix C, Table A3).

Pathway enrichment revealed that the predicted derepressed target genes clustered in distinct biological processes and signaling pathways, suggesting that coordinated downregulation of miRNAs at Day 30 may contribute to specific transcriptional reprogramming events during adaptation to hormone deprivation. The identification of pathways associated with vesicle-mediated transport and membrane trafficking is expected, given that our profiling focused on exosomal microRNAs, which are selectively packaged into exosomes. Interestingly, additional pathways related to chemical synapses, neurotransmitter signaling, and axon guidance suggest the emergence of neuronal differentiation features, highlighting a potential link between exosomal signaling and neuroendocrine-like phenotypes. Several predicted and experimentally validated target genes listed in Table 2 showed significant associations with progression-free survival (PFS). Specifically, lower expression of NRP1, GATAD2B, RUFY3, KIF3A, and ZBTB34 was associated with longer PFS. As these genes are predicted targets of the downregulated miRNAs identified in our dataset, these associations are consistent with a model in which increased miRNA expression is linked to improved clinical outcome through suppression of specific downstream target genes.

4. Discussion

Available in vitro tools to study prostate cancer are restricted to a handful of cell lines derived from metastatic lesions. Because prostate tissue is less terminally differentiated than other organs of the urogenital tract, prostate cells tend to de-differentiate once isolated, drifting away from their native phenotype [36]. Among the most well characterized and widely used, Pca lines are LNCaP cells which were selected for our study. LNCaP cells are hormone-sensitive, PSA-secreting, androgen-receptor-positive, with a plastic phenotype [37]. These cells were shown to acquire CRPC-like features after culture in androgen-deprivation conditions, namely the expression of neuroendocrine (NE) and stem cell marker including neuron-specific enolase, neurotensin, CD133 and ALDH1A1, as well as androgen-independent proliferation [38,39,40]. The culture conditions employed induce a hormone-insensitive phenotype in LNCaP include the use of depleted or serum-free media, over extended periods of time, that can range from 3 months to 2 years [16,40].

In recent years, advanced in vitro systems for modeling prostate cancer have gained momentum. Growing PCa cells as spheroids or organoids, embedding them in 3D extracellular matrices, and using dynamic fluidic culture has enabled higher-fidelity recreation of in vivo biology under controlled experimental conditions [41,42]. In this study we employed an iteration of our PCa-microfluidic model [25], using LNCaP spheroids that display a robust epithelial phenotype, forming polarized structures with distinct regions, demarcated by differential expression of adhesion proteins, and enriched EGFR expression at the surface (Figure 2), prior to hormone-deprivation culture. After 30 days of culture under decreasing androgen levels, the spheroids exhibited a substantial shift in the expression profile of exosome-associated miRNAs. Exosomes serve as vehicles for cell-to-cell communication and are believed to play a critical role in PCa progression [43]. Shear stress generated under microfluidic culture conditions, combined with continuous fluid renewal at the cellular surface, has been reported to facilitate extracellular vesicle biogenesis and secretion [44,45]. These vesicles are released into the surrounding microenvironment and have been implicated in the transition of androgen-dependent PCa cells to androgen-independent states [46]. Malignant cells utilize exosomes to modulate the transcriptional programs of neighboring cells, which is particularly relevant given the multifocal nature of PCa.

The hormone-deprivation culture conditions used in our study did not compromise spheroid viability. In fact, the resazurin-based assay (PrestoBlue) reported higher activity under dynamic perfusion, which is more likely linked to altered metabolic activity rather than increased proliferation, given that the dye is reduced by mitochondrial dehydrogenases. The stable PSA secretion observed over 30 days in microfluidic culture further supports the long-term robustness of the spheroids. The differences in PSA levels between static and dynamic culture may be due to flow-driven stimulation of PSA release; although recirculation does not alter the overall PSA content in the closed system, it likely enhances its diffusion out of the agar matrix in which the spheroids are embedded, improving measurable recovery.

Although circulating microRNAs are of considerable interest as potential diagnostic biomarkers, our analysis revealed that the expression of exosomal miRNAs in perfusate samples was highly variable, with substantial disparities observed between biological replicates (Figure A1). This inherent variability in secreted microRNAs is a well-recognized limitation that has hindered their reliable exploitation as clinical biomarkers [47,48]. In our analysis, significantly deregulated miRNAs in the perfusate did not overlap with intracellularly deregulated transcripts, although substantial miRNAs were present both intra- and extracellularly regardless of expression level. Interestingly, intracellularly expressed miRNAs are downregulated, while circulating transcripts are upregulated at Day 30 compared to Day 5, which may reflect accumulation of secreted exosomes in the perfusate over 30 days of closed culture. Exosomal sorting, which governs the selective packaging of extracellular vesicle cargo, may contribute to the discrepancies observed [49]. Tumor cells may selectively secrete oncogenic miRNAs as an adaptive mechanism to enhance their survival by modulating the behavior of neighboring cells. On the other hand, intracellular miRNA expression remained consistent across biological replicates. Downregulated miRNAs were linked to neuronal activity and differentiation, suggesting that the culture conditions promoted a shift in the LNCaP spheroids toward a NE-hybrid phenotype. The suppression of these transcripts is expected to positively regulate the expression of their target genes and associated cellular activity. Identified genes are associated with the PI3K-AKT-mTOR signaling pathway, cell cycle progression, epithelial-to-mesenchymal transition, and metastasis, all cellular processes central to PCa progression and the onset of CRPC [50,51]. Notably, the three most downregulated miRNAs, identified based on their number of interaction nodes, have previously been reported to exhibit tumor-suppressive activity in PCa. These miRNAs are also found to be downregulated in a clinical cohort of PCa samples (PRAD). MiR-9-5p has been shown to enhance the growth and metastatic potential of CD44^+^ prostate cancer stem cells [52], miR-186 inhibits prostate cancer cell proliferation and tumor growth by targeting YY1 and CDK6 [53], while miR-139-5p suppresses proliferation by targeting Notch1 [54]. Our findings underscore the functional significance of coordinated miRNA activity in the transition to hormone-insensitive PCa. On the other hand, mir-125b was the sole significantly upregulated miRNA identified. This transcript is associated with genes regulating the activity of p53 (Figure A2), which deregulated activity is well documented in PCa progression [55,56]. These synergistic effects between reduced tumor-suppressive activity and enhanced genetic instability play an important role in progression to advanced PCa. The identified targets in our analysis (Table 2) include genes with experimentally validated miRNA–target interactions and established clinical associations (Figure 7). Neuropilin-1 (NRP1) has been implicated in PCa aggressiveness and progression through its roles in angiogenesis, growth factor signaling, and therapy resistance [57]. Phosphodiesterase 4D (PDE4D) regulates cAMP-dependent signaling pathways that promote tumor cell proliferation and survival, and its dysregulation has been linked to advanced CRPC phenotypes [58]. Notably, PDE4D has been proposed as a potential therapeutic target in prostate cancer, given its druggability and contribution to oncogenic signaling [59]. The GATA zinc finger domain-containing 2B (GATAD2B) and Kinesin family member 3A (KIF3A) may contribute to epigenetic and signaling rewiring during disease progression [60,61]. The Zinc finger and BTB domain-containing 34 (ZBTB34) and RUN and FYVE domain-containing 3 (RUFY3) may contribute to regulatory roles downstream of miRNA control [62,63]. Despite their clinical association with PCa progression, the functional characterization of these genes is still limited.

Our microfluidic platform has demonstrated robust performance for modeling PCa progression and the transition to castration-resistant, despite being based on a single cell type. Contemporary studies using complex PCa models have incorporated endothelial cells or stromal cells to better mimic the tumor microenvironment [64,65]. Future work can extend dynamic culture beyond 30 days to determine whether the induced phenotypes are terminally differentiated or can be reverted after restoring androgens in culture. Additional cell types can also be included to capture immune interactions (e.g., macrophage population in the hydrogel). Long-term experiments will require active monitoring to prevent contamination and to track nutrient consumption, particularly glucose, to avoid metabolic starvation, an issue mitigated in the present study by replenishing culture media every five days during perfusate collection. Detection of circulating exosomal miRNAs could be improved by isolating vesicles from the perfusate, using enrichment methods (e.g., ultracentrifugation). Functional characterization of LNCaP spheroids post-perfusion is warranted to evaluate neuroendocrine trans-differentiation and epithelial–mesenchymal transition. Metabolic analyses will help define the energetic demands of dynamic culture and further validate the phenotypic shifts observed in our model. Overall, the robust long-term PCa spheroid culture shows that our model can support studies using patient-derived micro-tissues and organoid for personalized medicine applications, drug testing and biomarker discovery.

5. Conclusions

The current study demonstrates that microfluidic culture of PCa spheroids under hormone-deprived conditions can induce pronounced phenotypic changes that recapitulate early steps of disease progression. This system provides a physiologically relevant and controlled environment in which the dynamic adaptation of cancer cells can be monitored over time, offering valuable insights into the transition from hormone-sensitive PCa to CRPC. LNCaP spheroids acquire a hybrid phenotype displaying NE features, where downregulated intracellular miRNAs target genes associated with PCa progression and poor clinical outcomes. Our model enables systematic characterization of microRNA signatures and their functional roles in CRPC development. Prospective studies will take into account the technical constraints associated with implementing miRNA-seq pipelines for microfluidic-derived samples, with the aim of generating robust biological replicates. In addition, future work will include paired miRNA and mRNA sequencing of the same samples under identical experimental conditions to enable direct matching of miRNAs with their targets. By combining spheroid culture with microRNA profiling, it is possible to identify regulatory pathways that drive aggressive phenotypes, test therapeutic interventions, and explore potential diagnostic markers.