MicroRNA patterns in luminal B breast cancer cells grown in a 3D environment

To establish the 3D cultures, we sourced BT-474 cell lines from ATCC (Manassas, Virginia) and cultured them in HybriCare medium supplemented with 10% fetal bovine serum (FBS). In 24-well plates, we added 120 µL of Geltrex (Thermo Fisher Scientific) and allowed it to solidify at 37 °C for 30 min. Subsequently, we introduced 3.1 × 10⁴ cells from the 2D BT-474 culture and 250 µL of HybriCare medium and incubated them at 37 °C for an additional 30 min. Finally, we added 250 µL of HybriCare medium mixed with 5% Matrigel. Over the following days, the 3D cultures were refreshed with 500 µL of complete medium every three days.

To analyze miRNA expression, total RNA was extracted from BT-474 cells cultured in both 2D monolayer and 3D cultures. For this, we started with 3.1 × 10⁴ cells, which were incubated for 6 days. We added 1 mL of Trizol to the 2D and 3D cultured cells and mixed them by pipetting. After allowing the cell lysate to incubate at room temperature for 5 min, 200 µL of chloroform per mL of Trizol reagent was added and the tube was vigorously shaken for 15 s. The tube was then incubated at room temperature for 3 min before being centrifuged at 12,500 rpm for 25 min at 4 °C. After centrifugation, we carefully collected the aqueous phase without disturbing the other layers and transferred it to a new Eppendorf tube. 500 µL of isopropanol was added to precipitate the RNA, and the pellet was incubated on ice for 20 min. The tubes were then centrifuged at 12,500 rpm for 25 min at 4 °C. After discarding the supernatant, we washed the pellet with 1 mL of 75% ethanol and allowed it to air-dry for 5 min. Finally, the pellet was resuspended in 20 µL of nuclease-free water. The extracted RNA was quantified using spectrophotometry and visualized by 1% agarose/TAE 1× electrophoresis.

We employed the GeneChip miRNA Array 4.0 (Affymetrix GeneChip, Santa Clara, CA, USA) to analyze miRNA expression profiles with biological triplicates for 2D and 3D culture conditions (n = 3). Following the manufacturer's protocol, we used 100 ng of total RNA to synthesize the first-strand cDNA. RNA was degraded, and second-strand cDNA was synthesized using DNA polymerase I and RNase H. The sense-strand cDNA was subsequently synthesized and used for microarray experiments. A key step, reverse transcription, was carried out to generate the sense-strand cDNA from the RNA samples. This cDNA was loaded onto the GeneChip Array provided by Thermo Fisher Scientific (Waltham, MA, USA). The arrays were incubated at a constant temperature of 45 °C for 16 h in the GeneChip Hybridization Oven 645 (Affymetrix Inc., Santa Clara, CA, USA). Following the incubation, hybridization was performed using the GeneChip Fluidics Station 450. We scanned the arrays using the GeneChip Scanner 3000 7G (Affymetrix Inc., Santa Clara, CA, USA) to analyze gene expression profiles. Raw.CEL files were imported into Transcriptome Analysis Console (TAC; Thermo Fisher Scientific, v [4.x]). Array quality was assessed with TAC's QC panel (intensity distributions, array-to-array correlations, PCA). Data were normalized using SST-RMA (Signal Space Transformation–RMA) with background correction, quantile normalization, and probe-set summarization. Differential expression (3D vs. 2D) was computed on the log₂ SST-RMA values using TAC's limma-based statistical framework (moderated t/empirical Bayes). For each feature we report log₂ fold change (log₂FC) as the difference of group means; linear FC was obtained as log₂FC. P-values were adjusted for multiple testing by Benjamini–Hochberg FDR; significance thresholds (FDR < 0.05 and log₂FC ≥ 1).

Public data for mRNAs in BT-474 3D cultures (GSE206836), miRNAs in luminal breast cancer tumors (GSE225292), and miRNAs from 2D cultures of BT-474 cells (GSE163490) were collected from the Gene Expression Omnibus (GEO) database. We analyzed miRNAs from luminal breast cancer tumors and 2D cultures of BT-474 cells using the authors' normalized log₂ expression provided in the GEO Series Matrix files: luminal breast tumors (platform GPL18402) and BT-474 cells grown in 2D (platform GPL16770). To avoid cross-platform artifacts, the two cohorts were analyzed individually on their native Agilent log₂ scales with no cross-platform merging or additional normalization/batch correction. miRNAs were matched by standard miRNA name. We did not re-annotate probes or collapse probe signals; each row corresponds to the miRNA as reported in the respective matrix. For limits of detection and inclusion criteria, we use Agilent's sentinel value − 9.969 ("not detected") was handled as missing/off-scale in the primary analysis, and we restricted interpretation to miRNAs with RMA >1 across samples. For each miRNA, we computed log2FC = log2(tumor) − log2(BT-474), and group differences were tested on the log₂ scale using a two-sided Welch's t-test (full per-miRNA outputs in Supplementary Data S4). For differential expression of mRNA in 3D vs. 2D culture, the analysis steps were analogous: CEL files were imported into TAC, v4.x, normalized with SST-RMA, and differential expression between 3D and 2D was assessed on the log₂ scale using TAC's limma-based framework (moderated t/empirical Bayes). We then performed a biological integration using miRNet and ENCORI databases [26, 27]. Specifically, results were matched only at the interaction level, retaining pairs where (i) the miRNA was down-regulated (3D vs. 2D), (ii) its mRNA target was up-regulated (3D vs. 2D), and (iii) the pair appeared in miRNet or ENCORI (predicted/curated evidence). This strategy focuses on the directional concordance expected for regulatory pairs (miRNA↓/target mRNA↑). Furthermore, we assessed the performance of these miRNAs using a metric called the area under the curve (AUC) values to predict if miRNA expression could discriminate between normal and breast cancer samples. The AUC values were retrieved from CancerMIRNome [29], and the TCGA expression analyses were obtained from the UALCAN database [30].

All statistical analyses were performed using GraphPad Prism version 8. Data are presented as mean ± standard error of the mean (SEM) unless otherwise indicated. For experiments involving more than two groups, one-way or two-way analysis of variance (ANOVA) was used, followed by Tukey's post hoc test, to assess group differences. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.

To explore the impact of 3D culturing on BT-474 luminal B breast cancer cells morphology, we conducted a characterization using optical microscopy to examine the cellular architecture. Cells cultured in 3D formed compact spheroids with smooth surfaces and a homogeneous, mass-like morphology that obscured individual cell boundaries (Fig. 1A, right panel). In contrast, cells grown under traditional 2D monolayer conditions exhibited distinct, patchy arrangements with clearly defined cellular contours (Fig. 1A, left panel). These structural differences highlight the profound changes in cell organization and microenvironmental interactions that occur when cells are cultured in 3D. Based on these morphological findings, we then examined whether the 3D culture environment also affected the regulation of gene expression at the post-transcriptional level. To achieve this, we conducted a comparative analysis of global miRNA expression using the GeneChip miRNA 4.0 microarray platform, which includes 5,214 probes targeting mature human miRNAs. Our results revealed a substantial alteration in miRNA expression when cells were cultured in 3D compared to conventional 2D conditions. Specifically, 75 miRNAs were significantly upregulated, while 82 miRNAs were notably downregulated in 3D cultures (fold change > 1.5, adjusted p-value < 0.05) (Fig. 1B and Supplementary data). These differentially expressed miRNAs segregated into two distinct clusters as illustrated in the heatmap (Fig. 1C). The inset table in Fig. 1D lists the ten most prominently downregulated miRNAs in 3D conditions, including hsa-miR-664b-3p, hsa-miR-4454, hsa-miR-5100, hsa-miR-1260b, and hsa-miR-2115-5p. The top upregulated miRNAs comprised hsa-miR-455-3p, hsa-miR-29b-1-5p, hsa-miR-1244, hsa-miR-4669, and hsa-miR-1246. Functional pathway enrichment analysis indicates that the downregulation of these miRNAs likely leads to the activation of critical signaling pathways associated with sustaining proliferative signaling, estrogen receptor (ESR1)/ERBB-positive luminal breast cancer, ESR1 signaling, and JNK/MAPK signaling (Fig. 1E, upper panel). This observation aligns with previous reports indicating enhanced activation of the JNK/MAPK pathways in 3D cultures of breast cancer cell lines, such as T47D, suggesting that miRNA downregulation may facilitate these proliferative and survival signaling mechanisms during 3D growth. On the other hand, the upregulated miRNAs in 3D cultures appear to target tumor suppressor pathways, including the TGF-β, MAPK signaling, and TP53 signaling pathways (Fig. 1E, lower panel).

To provide a more biologically relevant context to the enriched processes, we integrated publicly available mRNA expression data from 3D cultures of BT-474 cells (GSE206836), which we previously reported [31]. Specifically, the mRNA targets of the differentially expressed miRNAs were predicted using Encori and miRNET software and compared with the mRNA expression dataset obtained from 3D cultures. We focused on miRNA-mRNA pairs where the fold change values were in opposite directions. This analysis revealed 1,175 mRNAs were downregulated in 3D cultures, which were predicted to be targets of 74 upregulated miRNAs (Fig. 2A and Supplementary data S3). On the other hand, 872 mRNAs upregulated in BT-474 3D cell cultures and predicted to be targets of 79 downregulated miRNAs mapped here (Fig. 2B and Supplementary data S2). Subsequently, we used these mRNA–miRNA pairs to perform an enrichment analysis to identify the associated biological processes. The set of 74 upregulated miRNAs was significantly enriched in pathways related to TGF-β signaling and G1/S DNA damage checkpoint regulation (Fig. 2C). In contrast, 872 upregulated mRNAs—predicted targets of 79 downregulated miRNAs—were associated with cell cycle progression, DNA replication, and focal adhesion pathways. Notably, we observed enrichment in signaling pathways such as MAPK and AKT signaling, and ESR1-mediated signaling, as well as biological processes related to epithelial-to-mesenchymal transition (Fig. 2C). We found a striking consistency when comparing these results with the predictive pathway analysis. Both analyses revealed enrichment in the estrogen-related ESR1 and the MAPK/JNK signaling pathways, further strengthening the hypothesis that these pathways play a key role in the biological processes associated with 3D cell culture. The convergence of these findings highlights the robustness of ESR1 and MAPK/JNK signaling as central processes in the 3D culture system, providing more substantial evidence that these pathways are critical in modulating the behavior of breast cancer cells in three-dimensional environments.

Integrating data from tumor samples significantly may enhance experimental findings' robustness and biological relevance. In this analysis, we compared miRNA expression profiles from two distinct sources: (i) the dataset generated in this study, which includes miRNAs differentially expressed between 3D and 2D cultures of BT-474 cells, and (ii) publicly available miRNA data from luminal breast tumors (GSE225292) and 2D cultures of BT-474 cells (GSE163490). By cross-referencing miRNAs with differential expression across these datasets, we aimed to identify miRNAs consistently regulated across distinct tumor microenvironments, particularly those showing similar expression in 3D cultures and luminal breast tumors, which could help reduce the gap between in vitro and in vivo data. Our analysis revealed that, in luminal breast cancer tumors, 43 miRNAs were upregulated compared to the 2D culture, and 186 miRNAs were downregulated (FC ≥ 1.5 and a p-value < 0.05) (Fig. 3A and Supplementary data S4). Following this, we conducted an overlapping miRNA analysis, identifying 8 upregulated miRNAs common to both the 3D culture and luminal tumor samples, compared to the 2D culture system. Additionally, we identified 6 downregulated miRNAs that were shared between the 3D cultures and luminal tumors (Fig. 3, panels B-C). Table 1 summarizes the fold change values of the identified shared miRNAs, their established biological functions, and experimentally confirmed targets. This approach strengthens confidence that these miRNAs play a pivotal role in cancer biology, rather than coincidental findings specific to a single model. Identifying overlapping miRNAs between these different contexts—3D cultures, 2D cultures, and luminal tumor samples—emphasizes the relevance of these molecules as potential biomarkers for breast cancer.

To further explore the molecular interactions within the ceRNA model, we performed an integrated analysis of miRNAs and mRNAs, focusing on miRNAs with differential expression between 3D cultures of BT-474 cells and luminal breast cancer tumors. In addition, we performed in silico prediction of long non-coding RNAs (lncRNAs) that sponge the miRNAs modulated in 3D. According to the ceRNA concept, upregulated lncRNAs can act as "sponges" for downregulated miRNAs, thereby promoting the activation of upregulated mRNAs in the network. This hypothesis is biologically plausible, as the lncRNAs predicted to interact with these miRNAs should be downregulated in expression within the ceRNA framework due to their role in sequestering miRNAs, ultimately enabling the overexpression of mRNAs that these miRNAs would usually suppress. Previously, we identified six downregulated miRNAs that were common to both the cell and tumor datasets and were selected for further analysis (Fig. 3C). Using the Encori platform, we identified 12 lncRNAs (OIP5-AS1, TBX5-AS1, MALAT1, SNAI3-AS1, PITPNA-AS1, FGD5-AS1, H19, NEAT1, MIR17HG, MAGI2-AS3, CTD-2566J3.1, and LINC01087) that were predicted to interact with four downregulated miRNAs (hsa-miR-539-3p, hsa-miR-92a-3p, hsa-miR-18a-5p, and hsa-miR-130a-3p). These miRNAs were found to regulate 58 upregulated mRNAs in 3D cultures of BT-474 cells (Fig. 4). Subsequently, we conducted an enrichment analysis of these interaction networks to identify the biological processes in which this ceRNA network may be involved. Notably, we observed potential involvement in key processes related to the tumor microenvironment of cells in 3D culture, such as oncogenic MAPK signaling, fibronectin matrix formation, estrogen receptor transcription, genes involved in cell migration, and cellular responses to mechanical stimuli (Fig. 4B). These processes suggest that the lncRNAs and miRNAs within the network may be regulating crucial aspects of tumor biology, such as MAPK signaling, cell migration, extracellular matrix formation, and responses to both hormonal and mechanical stimuli, providing a more relevant context to tumor conditions. This model not only emphasizes the critical role of miRNAs as key regulators but also underscores how interactions between lncRNAs, miRNAs, and mRNAs can contribute to cancer progression. These interactions may act as compensatory and regulatory mechanisms in 3D environments, replicating in vivo tumor conditions more accurately.

To further investigate the expression characteristics and potential prognostic value of the four miRNAs identified in the ceRNA network, we analyzed their expression patterns across different breast cancer subtypes and normal tissue from the TCGA dataset. Notably, miR-92a-3p, miR-130a-3p, and miR-539-5p were upregulated in normal tissue compared to luminal A and HER2 subtypes (Fig. 5, panels A-C). Interestingly, the expression of these miRNAs in 3D cultures of BT-474 cells was downregulated, consistent with their upregulation in normal tissue compared to tumor tissue in vivo. In contrast, miR-18a-5p exhibited higher expression levels in tumor tissue compared to normal tissue (Fig. 5D). This observation suggests that the regulatory dynamics of these miRNAs might be influenced by the tumor microenvironment, where their downregulation in 3D culture conditions may reflect their potential involvement in tumor progression. Additionally, we evaluated the diagnostic potential of these miRNAs by assessing their ability to distinguish between breast cancer and normal tissue. The area under the curve (AUC) values were promising, with miR-92a-3p showing an AUC of 0.68, miR-539-5p an AUC of 0.72, miR-18a-5p an AUC of 0.74, and miR-130a-3p exhibiting the highest AUC of 0.81 (Fig. 5E). While these results suggest that these miRNAs could serve as potential biomarkers for breast cancer, further studies would be needed to confirm their clinical utility and to understand their precise role in cancer biology. The fact that these miRNAs show significant differential expression between normal and tumor tissue, along with their distinct regulation in 3D culture, supports the hypothesis that they may play essential roles in cancer development and progression.