2.1 Enzalutamide-sensitive and enzalutamide-resistant cells have distinct m6A profiles

We undertook a broad discovery approach comparing enzalutamide-sensitive C4-2B PC cells (IC₅₀: 50uM enzalutamide) to established enzalutamide-resistant MDV-R PC cells (IC₅₀: 200uM enzalutamide) that were derived from C4-2B by serial passage [29]. (Figure 1A). Resistant MDV-R cells have a 2.5-fold increase in ARV7 mRNA expression compared to sensitive C4-2B (Figure 1A). We used a low-input methyl-RNA-immunoprecipitation and sequencing (MeRIP-seq) protocol [30] and a stringent differential analysis pipeline incorporating all recommended best practices published to date [31], including 4 replicates per condition and filtering criteria to mitigate the confounding effect of differential expression on differential methylation (Figure 1B).

Figure 1. Enzalutamide-sensitive and Enzalutamide-resistant cells have distinct m⁶A profiles. (A) Profiling of C4-2B and MDV-R PC cells including cell viability curves with enzalutamide treatment and mRNA expression of AR-FL and ARV7. (B) MeRIP-seq informatics filtering pipeline for differential methylation analysis with summarized peak number for different filtering steps. Numbers in parentheses refer to: peaks hypermethylated in C4-2B, peaks hypermethylated in MDV-R. (C) PCA plot of MeRIP samples with 4 replicates per cell line using all 36,393 peaks. (D) Summary of filtering results comparing C4-2B to MDV-R, with 74 peaks hypermethylated in C4-2B and 72 peaks hypermethylated in MDV-R. (E) Volcano Plot of statistically significant differential MeRIP-results (p < 0.05 by DESEQ2, edgeR, and QNB). C4-2B m⁶A represents the 74 peaks hypermethylated in C4-2B cells with log2FC < –1 and MDV-R m⁶A represents the 72 peaks hypermethylated in MDV-R with log2FC > 1. (F) Heatmap of top 20 differentially methylated transcripts ranked by DESeq2padj. MDV-R replicates are labeled M5-M8. C4-2B replicates are labeled C1-C4.

As a first pass quality control measure, we applied principal component analysis (PCA) to all 36,393 peaks called by MACS2 based on pull down versus input in all 8 replicates (4 sensitive and 4 resistant). The sensitive and resistant replicates clustered tightly into two distinct groups, with PC1 and PC2 representing 76% and 5% variation, respectively, reflecting the power of the study design to accurately discriminate the two conditions (Figure 1C). Encouraged by these preliminary results, we proceeded with differential methylation analysis using three statistical models (DESeq2, EdgeR and QNB) to evaluate statistical significance (p<0.05). We identified 633 differentially methylated peaks between enzalutamide-sensitive and resistant cells (Figure 1E), of which 487 remained after filtering for differential expression and 146 remained after selecting those with at least 10 reads. 74 of these peaks were hypermethylated in resistant cells, and 72 were hypermethylated in sensitive cells (Figure 1D). The volcano plot represents the 633 statistically significant peaks, highlighting the 487 peaks that meet all the filtering criteria and are unique to either sensitive or resistant cells (Figure 1E). The top 20 differentially methylated transcripts based on DESeq2 p-adj were visualized using a heatmap (Figure 1F).

2.2 Six cancer-relevant candidate transcripts were validated as differentially expressed and differentially methylated

To evaluate whether m⁶A alterations are associated with changes in gene expression, we overlapped the MeRIP-seq differential methylation results with RNA-seq differential expression from the same enzalutamide-sensitive and resistant cell lines (Figure S1A–C). We then applied the filtering pipeline outlined in Figure 2A in order to home in on transcripts relevant to cancer progression, as follows: We identified 46 candidate transcripts (52 out of 146 m⁶A peaks) that were both differentially expressed and differentially methylated (Figure 2B). We used Gene Ontology (GO) and Gene Set Enrichment Analysis (GSEA) to analyze the 46 transcripts. We identified 9 genes with membership in 10 GSEA Hallmarks of Cancer (p-adj<0.05, Figure 2C) and 20 genes with relevance to 27 GO terms like vasculogenesis and response to radiation (p-adj<0.05, Figure 2D). We further ranked the resulting transcripts based on log fold change and relevance to prostate cancer in existing literature and clinical databases such as The Cancer Genome Atlas (TCGA), the Oncology Research Information Exchange Network (ORIEN) Avatar and the Catalog of Somatic Mutations in Cancer (COSMIC). After ranking, we selected the top 12 transcripts for in vitro validation by qPCR and by targeted immunoprecipitation and PCR (MeRIP-PCR). 9 of the 12 differentially expressed transcripts by RNA-seq were also differentially expressed by qPCR (p<0.05), however only 8 were statistically significant (Figure 3A). Of the 8 transcripts (10 peaks) whose differential expression was validated by qPCR, 6 transcripts (8 peaks) were also validated as differentially methylated by MeRIP-qPCR (p<0.05, Figure 3B).

Figure 2. Identification of cancer-relevant transcripts that are both differentially methylated and differentially expressed in enzalutamide-resistant cells. (A) Discovery and filtering pipeline. (B) RNA expression Log2FC vs. RNA Methylation Log2FC of the 46 transcripts (52 peaks) that are both differentially methylated and differentially expressed between C4-2B and MDV-R. Peaks are colored based on filtering criteria: GO, GSEA, both or neither. (C) Differentially methylated and expressed transcripts statistically significantly associated with Gene Ontology (GO) Terms (p-adj <0.05). (D) Differentially methylated and expressed transcripts statistically significantly associated with Gene Set Enrichment Analysis (GSEA) Cancer Hallmarks (p-adj < 0.05).

Figure 3. Validation of top 12 differentially methylated and expressed candidate transcripts. (A) 8 of 12 candidate transcripts (boxed in red) were validated as differentially expressed using qPCR, corroborating RNA-seq results. (B) 6 of the 8 candidate transcripts (boxed in red) were validated by MeRIP-qPCR, corroborating MeRIP-seq results. Red dotted lines in (A) and (B) represent the validation threshold.

2.3 Depletion of THBS1 induces resistance to enzalutamide

We determined if any of the 6 differentially methylated and expressed genes identified in enzalutamide-resistant MDV-R cells also played a role in the transition to treatment resistance in enzalutamide-sensitive cell lines. To do this, each of the 6 genes (GRHL2, THBS1, FN1, PARP10, PAK4, SYT4) was siRNA- depleted in enzalutamide-sensitive cell lines (Table S1). Subsequently, the cells were treated with enzalutamide and assessed for viability after 72 hrs. One candidate transcript, THBS1, was functionally validated in this manner. Analysis of the MeRIP-seq data revealed several exonic peaks in THBS1 in MDV-R only, of which the peak at 9–10 K (boxed) reached statistical significance after all normalizations described above and was validated by MeRIP-PCR. Another peak at 3’UTR was not statistically different between enzalutamide-sensitive and enzalutamide-resistant cells (Figure 4A). Analysis of the RNA-seq data showed that THBS1 mRNA expression was decreased 41-fold in enzalutamide-resistant cells (p-adj<E-100, Figure 4B). TCGA analysis comparing THBS1 expression in prostate tumor and normal prostate samples showed significantly lower THBS1 expression in tumor samples (Figure S2). The reduction in THBS1 mRNA and protein levels in MDV-R enzalutamide-resistant cells relative to C4-2B enzalutamide-sensitive cells was further confirmed by RT-qPCR, which showed a 92% reduction (Figure 4C), and western blot (Figure S3). MDV-R enzalutamide-resistant cells also had higher mRNA levels of ARV7 (9.2-fold), AR-FL (1.8-fold), and MALAT1 (1.3-fold) compared to C4-2B enzalutamide-sensitive cells (Figure 4C). Short-term enzalutamide treatment of enzalutamide-sensitive C4-2B and LNCaP PC cells led to a significant reduction in THBS1 transcript levels, with a 66% and 82% decreases in C4-2B and LNCaP cells, respectively, and significantly increased expression of AR-FL (2.6-fold in C4-2B; 2.2-fold in LNCaP) and MALAT1 (3.0-fold in C4-2B; 2.0-fold in LNCaP), however, ARV7 was only upregulated in LNCaP cells (Figure 4D). THBS1 depletion by siRNA knockdown significantly increased relative enzalutamide resistance in the two enzalutamide-sensitive PC cell lines, C4-2B and LNCaP, but was not sufficient to recapitulate the increases in AR, ARV7, or MALAT1 seen with enzalutamide treatment (Figure 4C, Figure S4,S5).

Figure 4. Validation of THBS1 m⁶A methylation, mRNA expression, and effect of enzalutamide treatment. (A) m⁶A coverage plot for THBS1 in C4-2B and MDV-R cells depicting several exonic peaks, of which the peak at 9-10K (boxed) reached statistical significance after all normalizations and filtering and was validated by MeRIP-qPCR. 3’UTR peak is unchanged between the two cell lines. (B) Normalized RNA-seq count of THBS1 in C4-2B and MDV-R cells. (C) RT-qPCR analysis of gene expression in enzalutamide-sensitive C4-2B and enzalutamide-resistant MDV-R PC cell lines. MDV-R cells show reduced THBS1 expression and elevated ARV7, AR-FL, and MALAT1 levels compared to C4-2B. Gene expression levels were normalized to the housekeeping gene GAPDH and further normalized to the expression levels in C4-2B cells. Graph displays mean ± SEM values in n=3 replicates. (D) RT-qPCR analysis of gene expression in C4-2B and LNCaP cells untreated or treated with short-term 5uM enzalutamide(ENZ) treatment. ENZ reduced THBS1 expression and increased AR-FL and MALAT1 levels in both cell lines. Data normalized to GAPDH and untreated controls. Graph displays mean ± SEM values in n=3 replicates; p values comparing treated cells to respective untreated cells. Significant codes: ‘****’: p<0.0001, ‘***’: p<0.001. (E) Functional validation of THBS1 evaluating its effect on enzalutamide resistance via siTHBS1 knockdown followed by 5uM enzalutamide treatment in C4-2B cells (CDI: 1.30) and LNCaP cells (CDI: 1.09). CDI < 0.7 indicates statistically significant synergy. CDI > 1.0 indicates statistically significant antagonism. Experiments were performed using three biological replicates. Significant codes: ‘*’: p< 0.05, ‘***’: p<0.001.

We evaluated THBS1 differential expression from read counts data from previously published preclinical datasets [32] comparing androgen-dependent xenografted LAPC9 and LnCAP PC cell lines before and after surgical castration of the host (Figure 5A and Figure S6). Indeed, we saw a consistent trend of decreased THBS1 expression in castration-resistant cell lines compared to androgen-dependent cell lines. In particular, we observed a statistically significant (p=0.0286) decrease in THBS1 expression between the two castration-resistant LNCaP cell lines, LNCaP_PRC1 and LNCaP_PRC2, the latter of which was derived from the first following enzalutamide treatment. Finally, we compared THBS1 expression from a previously published clinical dataset [33] comparing 21 matched samples from PC patients at baseline and following progression on enzalutamide. 13 of the 21 matched pairs had a decrease in THBS1 expression (Figure 5B). Per previous analysis, three of the 21 matched pairs were deemed “converters” demonstrating lineage plasticity following enzalutamide treatment. Interestingly, all three of these converters (bolded) had a decrease in THBS1 expression, with log₂ fold change of –2.12 and p-adj= 0.011 from previously published analysis [33].

Figure 5. THBS1 Expression in samples from existing preclinical and clinical datasets. (A) THBS1 expression (counts) decreased in androgen-dependent (AD) xenografted LAPC9 and LNCaP PC cell lines after surgical castration of the host (PMID: 30190514)[32]. LNCaP_CRPC1 refer to the cell lines derived after surgical castration; LNCaP_CRPC2 refer to the castration-derived cell line after treatment with enzalutamide. P-values presented are from Wilcoxon rank sum test for LAPC9_AD vs LAPC9_CR (p=0.0556), LNCaP_AD vs. LNCaP_CRPC1 (p=0.0571), LNCaP_AD vs. LNCaP_CRPC2 (p=0.0571) LNCaP_CRPC1 vs. LNCaP_CRPC2 (p=0.0286). One outlier was removed from LNCaP_AD replicates based on Dixon-Q test. (B) THBS1 expression (TPM) at baseline and progression following enzalutamide treatment in 21 matched patient samples from exisiting clinical dataset (PMID: 36109521)[33]. Bolded patient samples (DTB_080, DTB_135, DTB_210) refer to patient samples termed “converters” having undergone enzalutamide-induced lineage plasticity as defined in the original paper. Colored-tiles refer to direction of THBS1 expression change from baseline to progression: increasing (green), decreasing (red), no change (grey).

4.1 Cell culture

Human prostate cancer cell lines, C4-2B and LNCaP, were cultured in RPMI 1640 and DMEM (Mediatech Inc., Manassas, VA), respectively, supplemented with 10% heat-inactivated fetal bovine serum (Omega) and 1% penicillin/streptomycin (100 units/mL, Invitrogen), at 37 °C and 5% CO₂. Prior to conducting all experiments, the cell lines were authenticated using 9-marker short tandem repeat (STR) profiling and testing interspecies and mycoplasma contamination (CellCheck 9 Plus, IDEXX BioAnalytics, Columbia, MO). We maintain stringent, good cell culture practice (GCCP), and keep extensive records of cell lines’ culture and harvesting condition, including passage count and cell density. We performed all experiments below passage 20, after which cultures were restarted using thawed cells from earlier passages. MDV-R cells were generated by serial desensitization of C4-2B cells with enzalutamide and were a generous gift from the Gao Lab (UC Davis) [29]. MDV-R cell lines were maintained in RPMI 1640 media with 55M enzalutamide (Catalog#MDV3100, SelleckChem, Houston, TX).

4.2 RNA isolation

Total RNA was extracted from cell lines using Trizol reagent and Direct-zol RNA extraction kit (R2071, Zymo Research), which included treatment with DNase I for 20min at 37 °C. Total RNA was extracted from organoids using RNEasy Micro Kit (Catalog# 74004, Qiagen, Germany). The concentration of total RNA was measured by Qubit RNA HS Assay Kit (Catalog# Q32855, Thermo Fisher Scientific) or via nanodrop, depending on the starting amount.

4.3 RNA fragmentation and MeRIP

MeRIP was performed on RNA chemically fragmented to ~100nt fragments as previously described [17,30]. Briefly, 3-5 μg (18μL volume) of purified RNA was incubated with 2μL 10X RNA Fragmentation Buffer (100 mM Tris-HCl, 100 mM ZnCl₂ in nuclease free H₂O) in a preheated thermal cycler for 4 ~ 5 min at 70 °C. The reaction was stopped with 0.5M 2μL EDTA, followed by 178μL of H₂O, 20μL of sodium acetate (3 M, pH 5.2, S7899, Sigma-Aldrich, St. Louis, MO), 14.4μL of glycogen (5 mg/ml, Catalog#AM9510, Thermo Fisher Scientific) and 500μL of 100% ethanol and incubated at -80 °C overnight. Fragmented RNA was pelleted (centrifuged 30min at 12,000g at 4 °C), washed once with 75% ethanol and resuspended in RNAse-free water (10 μL H₂O per 1 μg human total RNA). Size distribution was assessed using RNA 6000 Pico Kit on BioAnalyzer (Catalog# 50671513, Agilent Technologies, Santa Clara, CA).

30 μL of protein-A magnetic beads (Catalog# 10002D, Thermo Fisher Scientific) and 30 μL of Protein-G magnetic beads (Catalog# 10004D, Thermo Fisher Scientific) were washed twice with IP buffer (150 mM NaCl, 10 mM Tris-HCl, pH 7.5, 0.1% IGEPAL CA-630 in nuclease free H₂O), and incubated with 5μg anti-m⁶A antibody (Catalog# E1610, NEB, Ipswich, MA) at 4 °C overnight. The bead-antibody mixture was washed twice with IP buffer and resuspended in 500 μL IP buffer containing the fragmented RNA, 100 μL of 5X IP buffer and 5 μL RNasin Plus RNAse Inhibitor (Catalog# N2611, Promega, Madison, WI), and incubated for 2 hours at 4 °C.

The RNA reaction mixture was washed twice with 1000μL IP buffer, twice with 1000 μL low-salt IP buffer (50 mM NaCl, 10 mM Tris-HCl, pH 7.5, 0.1% IGEPAL CA-630 in nuclease free H₂O), and twice with 1000μL high-salt IP buffer (500 mM NaCl, 10 mM Tris-HCl, pH 7.5, 0.1% IGEPAL CA-630 in nuclease free H₂O) for 10min each at 4 °C. After the washes, the m⁶A-containing fragments was eluted from the beads in 200 μL of RLT Buffer supplied in the RNeasy Micro Kit (Catalog# 74004, Qiagen, Germany) for 2 min at room temperature. Using magnetic separation rack (Catalog# 1231D, Thermo Fisher Scientific) to pull beads to the side of the tube, supernatant was collected and combined with 400 μL 100% ethanol. The mixture was transferred to an RNeasy MicroElute spin column (RNeasy Micro Kit) and centrifuged at >12,000rpm at 4 °C for 1 min. The column membrane was washed with 500μL RPE Buffer (RNeasy Micro Kit) once, and with 500 μL 80% ethanol once. The column was centrifuged at full speed for 5min at 4 °C to remove residual ethanol. The m⁶A-containing fragments were eluted with 14 μL nuclease-free water.

4.4 RT-qPCR and MeRIP-qPCR

cDNA synthesis was performed using qScript cDNA SuperMix (Catalog# 95408-500, QuantaBio, Beverly, MA). Real Time PCR was performed using PerfeCTa SYBR Green FastMix (Catalog# 95071-250, QuantaBio) using a BioRad CFX96 Real Time PCR Detection System. Primer sequences can be found in Table S2. All experiments were performed in biological triplicates, and additional technical triplicates were used for all RT-qPCR experiments.

4.5 RNA-seq

RNA-seq libraries were prepared using NEBNExt Ultra II RNA Library Prep Kit for Illumina (Catalog# E7420, NEB) according to manufacturer’s protocol. RNA-seq reads were aligned to human genome hg38 with reference annotation GENCODE v39 and counted using STAR (version 2.7.0) [39,40]. Only uniquely mapped reads without duplicates were selected using samtools (version 1.10) [41]. Read counts were assigned to genes using Subread featureCounts [42]. Read counts were normalized using DESeq2 package in R (version 4.1.3). To generate more accurate log2fold change estimates, shrinkage of the LFC estimates towards zero was applied using DESeq2 [43]. Differentially expressed transcripts with absolute |log2FC| > 0.5 and adjusted-p value < 0.05 were retained. Further filtering (|log2FC| > 0.5, FDR<0.00001) was later performed for gene selection for downstream qPCR validation. Gene set enrichment analysis and Gene Ontology analysis were implemented and visualized using clusterProfiler package [44-46].

4.6 Differential MeRIP-seq analysis

MeRIP libraries were prepared using SMARTer Stranded Total RNA-Seq Kit v2- Pico Input Mammalian (Catalog # 634413). MeRIP-seq reads were aligned to human genome hg38 with reference annotation GENCODE v39, and counted using STAR (version 2.7.0) [39,40]. Only uniquely mapped reads without duplicates were selected using samtools (version 1.10)[41]. IP over input peaks were called using MACS2 callpeak using the parameters “-nomodel -extsize100 -gsize300e6” [47]. Differential m⁶A analysis was performed using DEQ package in R as previously published [31]. Briefly, DEQ runs statistical analysis using DESeq2, edgeR and QNB packages [31,43,48,49]. Results that were statistically significant with adjusted p-value < 0.05 using all three packages were considered significant. Gene and peak expression changes were estimated as log2 fold changes from DESeq2. Additional filtering to mitigate the confounding effect of differential gene expression on determining differential methylation was applied using |peak IP log2FC- gene input log2FC| ≥ 1. Peaks with fewer than 10 read counts were also removed to mitigate the confounding effect of differential gene expression.

4.7 siRNA knockdown

Dicer-substrate short interfering RNA (DsiRNAs) targeting transcripts of interest and corresponding negative control were purchased from Integrated DNA Technologies (IDT, Coralville, IA). DsiRNA sequences, concentrations and timepoints used can be found in Table S3. C4-2B and LNCaP cells, transfection was performed 24 hours after seeding. DharmaFECT #3 (Catalog# T-2003-02, Horizon Discovery, Waterbeach, UK) and OPTI-MEM I reagents (Catalog# 31985062, ThermoFisher) were used to transfect cells at 70% confluence according to manufacturer’s protocol. Cells were harvested and total RNA was extracted at 24 or 48 hours depending on the transcript. Knockdown efficiency was evaluated by RT-qPCR. In subsequent experiments, total RNA was extracted at 48 or 72 hours post-transfection followed by RT-qPCR of the downstream target genes.

4.8 Enzalutamide resistance assay

C4-2B or LNCaP cells were seeded and transfected the following day with corresponding siRNA or negative control for 24-48 hours prior to treating reseeding cells with 5uM enzalutamide (Catalog# S1166, Selleck Chemicals) for 72 hours. Enzalutamide was dissolved in DMSO and stored in single use aliquots at -80 °C. MTS proliferation assay was performed using CellTiter 96 AQueous One Solution Cell Proliferation Assay (Catalog#G3582, Promega) according to manufacturer’s protocol. Cell viability was calculated using: Percentage Viability= (Absorbance[Sample]/Absorbance[NC - ENZ])x100%. Coefficient of drug interaction (CDI) to evaluate synergy was calculated using: CDI= Percentage Viability[Condition + ENZ]/(Percentage Viability [Condition – ENZ] x Percentage Viability [NC – ENZ]). CDI > 1.0 indicates antagonism and CDI < 0.7 indicates significant synergy [50]. Experiments were repeated in 3-4 biological replicates, using 6 technical replicates.

4.9 THBS1 expression analysis in preclinical and clinical datasets

Using previously published preclinical dataset (PMID: 30190514) comparing androgen-dependent (AD) and castration-resistant (CR) LAPC9 and LNCaP PC xenografted cell lines [32]. Briefly, AD xenograft tumors were routinely maintained in intact mice; while CR lines LAPC9_CR and LNCaP_CRPC1) were established by serially passaging AD tumor cells in surgically castrated mice. CR tumors that became enzalutamide-resistant following enzalutamide treatment were deemed secondary CRPC (LNCaP_CRPC2).Using read count data from 4-5 replicates for each cell line, DeSEQ2 was used to evaluate differential expression of THBS1 between the samples. Dixon-Q test was used to identify outliers. Analysis was repeated after removal of one outlier replicate. Both analyses (with and without outliers) are presented in Supp. Figure 5A and Figure 5A, respectively. Wilcoxon rank sum test was used to calculate p-values.

Using previously published clinical dataset (PMID: 36109521) [33] comparing 21 matched pairs of prostate cancer patient samples at baseline and at progression on enzalutamide. Using TPM values, we evaluated THBS1 differential expression between matched samples. Three of the matched pairs were deemed “converters” meaning they underwent lineage plasticity following enzalutamide treatment.

4.10 Western blot

Whole cell lysates were extracted from PC cells using RIPA lysis buffer (Sigma-Aldrich, Cat# R0278). Total protein concentration was determined by Lowry Assay using the BCA Protein Assay Kit (Bio-Rad, Hercules, CA, USA). 100 μg protein lysate samples were boiled in loading buffer for 10 minutes before running on 3-8% Tris-Acetate gradient gels (Invitrogen, Cat# EA0375BOX) with PageRuler Plus Prestained Protein Ladder (10-250 kDa) (ThermoFisher, Catalog#26619). Western blot transfer was performed using the iBlot2 Dry Blotting System (ThermoFisher Scientific) and PVDF mini iBlot 2 Transfer Stack (Invitrogen, Catalog #IB24002). Membranes were then blocked in Odyssey blocking buffer (LI-COR, Lincoln, NE, USA) and incubated with primary antibodies overnight at 4 °C. Membranes were then washed 3 times in 1X TBST + 0.10% Tween20 Buffer, and then incubated in respective secondary anti-mouse (goat anti-mouse IRDye 680RD, 1:5000, LI-COR, Cat# 926-68070) and secondary anti-rabbit (goat anti-rabbit IRDye 800CW, 1:5000, LI-COR, Cat# 926-68070) for 45 minutes at room temperature. After 3 more washes in 1X TBST + 0.10% Tween20 Buffer, membranes were visualized using an Odyssey DLx Imaging System (LI-COR). The primary antibodies used were as follows: GAPDH (Cat#437000, 1:5000; Invitrogen), THBS1 (Cat#37879, 1:1000; Cell Signaling Technology).

4.11 Statistical analysis

P-values were calculated using two-tailed Student’s t-test or Wilcoxon rank sum testbetween two groups, and Wilcoxon signed rank test between matched groups. Data are presented as means +/- SEMs from at least three independent experiments. Significant codes: ‘****’: p< 0.00001, ‘***’: p< 0.001, ‘**’: p<0.01, ‘*’: p <0.05, ‘.’: p<0.1, and ‘ns’: not significant GraphPad Prism version 9, Excel version 16.68, or R version 4.1.3 were used for statistical analysis.

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

This work was supported by the National Institutes of Health [National Cancer Institute 1F30 CA257401-01A1, 5P30CA014089-45].

Data Availability Statement

The datasets generated for this study can be found in the GEO Database: GSE249750. All other relevant data are within the manuscript.

Link: https://urldefense.com/v3/__https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE249750__;!!LIr3w8kk_Xxm!rDJbDcCVPWv-VrddOabaP8rFUqPTHQ_N64hIA5-G26mP-PclaTKEkNtJp1cFShgZrLi0zfX52Ne1g_VXUQ$

Reviewer Token: cbsbicwidxqftkr

Author Contributions

E.H. conceived and designed the study, conducted all experiments, interpreted the results and wrote the manuscript. L.S. conducted knockdown and RT-qPCR experiments, interpreted the results, and helped write and prepare the manuscript. A.M. conducted the qPCR and MeRIP-qPCR validation experiments, reviewed and edited the manuscript. D.B. participated in bioinformatics analysis and data visualization. T.X. participated in troubleshooting and study design. P.J.F. and S.K.R. provided critical feedback for data analysis and reviewed the manuscript. A.G. designed the study, analyzed the data and wrote the manuscript.

Note on Rigor and Reproducibility

To esure our approach was scientifically rigorous and subsequent results are robust and unbiased, we used 3-5 biological replicates, 3-6 technical replicates, and repeated experiments in multiple PC cell lines and PDOs.

Acknowledgement

We thank Dr. Hansen He and Dr. Shiyan Wang at Princess Margaret Medical Center, and Dr. Gideon Rechavi and Dr. Sharon Moshitch-Moshkovitz at Sheba Medical Center for their guidance in establishing MeRIP-seq protocols in our lab. We thank Dr. Allen Gao at UC Davis for providing MDV-R and parental C4-2B cell lines. We also thank Stephanie Tring, Helen Truong and Tatei Torres from the USC Norris Molecular Genomics Core. Illustrations were created using Biorender: Hodara, E. (2025) https://BioRender.com/20f15rg.