Multiplexed functional genomic analysis of 5' untranslated region mutations across the spectrum of prostate cancer

Abstract

The functional consequences of genetic variants within 5' untranslated regions (UTRs) on a genome-wide scale are poorly understood in disease. Here we develop a high-throughput multi-layer functional genomics method called PLUMAGE (Pooled full-length UTR Multiplex Assay on Gene Expression) to quantify the molecular consequences of somatic 5' UTR mutations in human prostate cancer. We show that 5' UTR mutations can control transcript levels and mRNA translation rates through the creation of DNA binding elements or RNA-based cis-regulatory motifs. We discover that point mutations can simultaneously impact transcript and translation levels of the same gene. We provide evidence that functional 5' UTR mutations in the MAP kinase signaling pathway can upregulate pathway-specific gene expression and are associated with clinical outcomes. Our study reveals the diverse mechanisms by which the mutational landscape of 5' UTRs can co-opt gene expression and demonstrates that single nucleotide alterations within 5' UTRs are functional in cancer.

Fig. 1. Development of a massively parallel reporter assay to quantify the impact of 5′-UTR mutations on transcript levels and mRNA translation.

a Histogram of the genomic distribution of all somatic single-nucleotide 5′-UTR mutations in five prostate cancer patient-derived xenografts (PDX) from the LuCaP series. b Percentage of 5′-UTR mutations in each LuCaP PDX that significantly alter transcript or mRNA translation efficiency (TE) levels (FDR < 0.1, represented in pink). Across five xenografts representing a spectrum of advanced human prostate cancer (adenocarcinoma, neuroendocrine prostate cancer, hypermutated prostate cancer), 13 mutations exhibited a decrease in transcript levels, while 35 mutations exhibited an increase. At the level of translation, 31 5′-UTR mutations decreased ribosome occupancy (decreased translation efficiency [TE]), while 42 had the opposite effect, independent of changes at the mRNA level. c Volcano plot showing TE fold changes of all 5′-UTR mutations in these LuCaP PDXs. Each dot represents TE fold change of a 5′-UTR mutation; turquoise dots are 5′-UTR mutations that significantly downregulate TE of its specific mRNA (FDR < 0.1), yellow dots are 5′-UTR mutations that significantly upregulate TE of its specific mRNA (FDR < 0.1). Thirty-one 5′-UTR mutations decreased ribosome occupancy (decreased translation efficiency [TE]), while 42 mutations increased TE. Mutations selected for orthogonal validation are demarcated in pink and labeled with the gene name. d Luciferase assays validating potentially functional 5′-UTR mutations (pink) identified by ribosome profiling including ADAM32 (chr8: 38965236, C -> T) and COMT (chr22: 19939057, G -> A), as well as the negative control ZCCHC7 (chr9: 37120713, C -> T). Normalization was performed by taking the ratio of the relative luminescence unit (RLU) to the amount of luciferase transcript determined by qPCR (n = 4–8 biological replicates, one-sided Student’s t test). Data are presented as median by the center line, and the first and third quartiles as the upper and lower edges of the box. All minimum and maximum data points are indicated by error bars. e Simplified schematic of the Pooled full-length UTR Multiplex Assay on Gene Expression (PLUMAGE). f All 30 unique 8-bp barcodes were detected and linked with their respective WT and mutant 5′-UTR by PacBio long-read sequencing (average of 39.4–254.2 read counts per 5′-UTR–barcode pair). Each colored circle represents a unique 8-bp barcode. Data are presented as median by the center line. g Comparison of mRNA translation efficiency between WT and mutant ADAM32, COMT, and ZCCHC7 5′-UTRs by PLUMAGE. Mutations are represented in pink. Results are concordant with ribosome profiling and luciferase assay findings in  (c, d). Normalized polysome read counts for each barcode per construct were taken as a ratio over normalized total RNA read counts for the same barcode (n = 4–5 biological replicates, one-sided Student’s t test). Data are presented as median by the center line, and the first and third quartiles as the upper and lower edges of the box. All minimum and maximum data points are indicated by error bars. n.s. not statistically significant. Source data are provided as a Source data file.

Fig. 2. The mutational landscape of prostate cancer 5′-UTRs.

a Comparison of 5′-UTR mutation rate (5′-UTR mutation/Mb) in localized prostate cancer (PCa) patients (n = 149, each patient represented by a blue dot) and metastatic castration-resistant prostate cancer (mCRPC) patients (n = 80, each patient represented by a green dot). Each dot represents the mutation rate per patient (***p = 0.0001, two-tailed Mann–Whitney U test). Data are presented as mean values ± s.e.m. b KEGG and Reactome pathway analyses of all genes with 5′-UTR and protein-coding sequence (CDS) mutations across 229 prostate cancer patients. Genes with 5′-UTR mutations can cluster with or independent of genes with CDS mutations (Fisher’s hypergeometric test, FDR < 0.05). c The absolute genomic distance of somatic single-nucleotide 5′-UTR mutations within recurrently mutated genes; 38.7% of recurrently mutated 5′-UTRs have alterations located less than 50-bp apart. d Predicted enrichment of observed 5′-UTR mutations in our patient cohort across known DNA- and RNA-binding regulatory elements. Validated DNA (Homer) and RNA protein-binding motifs (Hughes) were analyzed. To generate the background (null) distribution of mutations, permutations of all 5′-UTR mutation locations (n = 2200 mutations) found in our dataset were performed ~10,000 times taking into account covariates such as trinucleotide context. The total number of observed mutations (represented by pink dots) impacting each regulatory element type was compared to the background distribution of the permutation data (represented by gray square) and the p value was computed from a distribution obtained from a Monte Carlo simulation (**p = 0.001, ***p = 0.0001). Data are presented as mean values ± s.d. e Predicted enrichment of observed 5′-UTR mutations in our patient cohort across cis-regulatory elements known to affect translation such as upstream open-reading frames (uORFs), terminal oligo pyrimidine (TOP)-like or pyrimidine-rich translational elements (PRTEs), G-quadruplexes, and 5′-TOP elements. To generate the background (null) distribution of mutations, permutations of all 5′-UTR mutation locations (n = 2200 mutations, represented by pink dots) found in our dataset were performed ~10,000 times taking into account covariates such as trinucleotide context. The total number of observed mutations impacting each regulatory element type was compared to the background distribution of the permutation data (represented by gray square) and the p value was computed from a distribution obtained from a Monte Carlo simulation (***p = 0.0001). Data are presented as mean values ± s.d. n.s. not statistically significant. Source data are provided as a Source data file.

Fig. 3. Thirty-five percent of 5′-UTR mutations across the spectrum of human prostate cancer functionally impact transcription or translation.

a Per-gene percentages of distinct barcodes associated with an exact match to an expected 5′-UTR sequence by PacBio long-read sequencing. All distinct 5′-UTR sequences were observed by long-read sequencing and linked to an average of 236 distinct 30-bp barcodes. For each gene, the percentage of barcodes associated with an exactly matching 5′-UTR are plotted as black vertical bars (correctly synthesized). Genes are ordered by increasing 5′-UTR length from left to right, and the average rate of exactly matching barcodes is marked by a horizontal dashed line at 85%. A smoothed fit, using loess regression, of percentage matching vs rank order of length is shown in red. b Correlation of normalized read counts per WT and mutated 5′-UTR in each technical and biological replicate for each PLUMAGE DNA sample. Three biological replicates were analyzed for each cell line (293T and PC3). Pearson correlation coefficient was calculated to determine significance and was found to be r > 0.99 for all samples (all p values = 0.00001). c Correlation of normalized read counts per WT and mutated 5′-UTR in each technical and biological replicate for each PLUMAGE total mRNA sample. Three biological replicates were analyzed for each cell line (293T and PC3). Pearson correlation coefficient was calculated to determine significance and was found to be r > 0.8 for all samples (all p values = 0.00001). d Correlation of normalized read counts per WT and mutated 5′-UTR in each technical and biological replicate for each PLUMAGE polysome-bound mRNA sample. Three biological replicates were analyzed for each cell line (293T and PC3). Pearson correlation coefficient was calculated to determine significance and was found to be r > 0.89 for all samples (all p values < 0.0001). e Proportion of all 5′-UTR mutations assayed by PLUMAGE that showed a significant (FDR < 0.1, in pink) change in mRNA transcript or translation levels. f 5′-UTR mutations (190 mutations) that significantly change gene expression affect important cancer-related pathways by KEGG pathway analysis (Fisher’s hypergeometric test, FDR < 0.05). Source data are provided as a Source data file.

Fig. 4. PLUMAGE reveals how somatic 5′-UTR mutations affect mRNA transcript levels.

a 5′-UTR mutations that significantly affect mRNA transcript levels (Mann–Whitney U test, FDR < 0.1) and magnitude fold change compared to unmutated 5′-UTR (Supplementary Data 6d). A proportion of mutations also impact a known DNA-binding element, indicated by black bars. b qPCR validation of the FOS (chr14: 75745674, C -> G) and FGF7 (chr15: 49715462, C -> T) 5′-UTR mutations identified by PLUMAGE. WT (represented by gray dots) and mutant (represented by pink dots) 5′-UTRs were cloned into a luciferase reporter construct and transduced into PC3 prostate cancer cells. Luciferase transcript levels were then normalized to luciferase DNA (n = 9 biological replicates for FOS WT and mutant, mean ± s.e.m. Student’s t test; n = 6 biological replicates for FGF7 WT and mutant, data are presented as mean ± s.e.m., one-sided Student’s t test). c RNAseq volcano plot of all significantly up- and down-regulated mRNAs in the human prostate cancer PDX LuCaP 81 (FDR < 0.1). Each dot represents the transcript fold change of a 5′-UTR mutation in LuCaP 81; turquoise dots are 5′-UTR mutations that significantly downregulate transcript expression of its specific mRNA (FDR < 0.1), yellow dots are 5′-UTR mutations that significantly upregulate transcript expression of its specific mRNA (FDR < 0.1). Within this PDX, FGF7 exhibits a 5′-UTR mutation (indicated by pink dot) at chr15: 49715462, C -> T that is associated with an increase in FGF7 transcript levels. d The FGF7 5′-UTR mutation introduces a thymidine at position chr15: 49715462, which transforms the CACGCG sequence into an E-box motif (CACGTG). Mutated nucleotide is represented in red. e Representative EMSA using the WT vs mutant FGF7 5′-UTR. Labeled probe sequences (33-bp) containing the E-box sequence generated by the mutation in the 5′-UTR of FGF7 and the wild-type sequence are shown. Mutated nucleotide is represented in red. Binding of MYC:MAX heterodimer protein complex is observed only with the mutated oligonucleotide probe containing the E-box sequence. Binding of the labeled oligo can be abolished using an unlabeled competitor. Source data are provided as a Source data file.

Fig. 5. PLUMAGE uncovers somatic 5′-UTR mutations that affect mRNA translation efficiency and multiple layers of gene expression.

a 5′-UTR mutations that significantly affect mRNA translation efficiency (Mann–Whitney U test, FDR < 0.1) and magnitude fold change compared to unmutated 5′-UTRs (Supplementary Data 6e). A proportion of mutations impact known RNA-binding protein-binding motifs, indicated by black bars. b Validation of 5′-UTR mutations in AKT3 (chr1: 244006547, C -> T) and NUMA1 (chr11: 71780891, C -> A) by luciferase assay. Each WT (represented in gray) or mutant (represented in pink) construct was separately transfected into PC3 cells and assayed for luciferase activity and luciferase mRNA expression after 24 h. Luciferase activity (RLU) was normalized to the amount of luciferase transcript in each transfection to determine translation efficiency (n = 4 biological replicates for AKT3 WT and mutant, mean ± s.e.m. Student’s t test; n = 3 biological replicates for NUMA1 WT and mutant, data are presented as mean ± s.e.m., one-sided Student’s t test). c The C -> A 5′-UTR mutation in NUMA1 at position chr11: 71780891 abrogates an existing SRSF9 RNA-binding protein motif. Mutated nucleotide is represented in red. d The 5′-UTR mutation in QARS (chr3: 49142179, G -> A) shows significant changes in both transcript levels and translation efficiency, not attributable to the amount of DNA transfected. Each QARS 5′-UTR WT (represented in gray) and mutant (represented in pink) plasmid was transfected individually into PC3 cells (n = 5–6 biological replicates), followed by luciferase assay, luciferase RNA PCR, and luciferase DNA qPCR. RLU values were normalized to luciferase mRNA to determine translation efficiency. Luciferase mRNA was normalized to luciferase DNA to determine the effects on transcript levels (n = 6 biological replicates, data are presented as mean ± s.e.m., one-sided Student’s t test). n.s. not statistically significant. Source data are provided as a Source data file.

Fig. 6. CRISPR-Cas9 base editing of a point mutation in the CKS2 5′-UTR increased translation efficiency in its endogenous context.

a Schematic showing wild-type (WT, top) and mutant (bottom) versions of CKS2 transcript, including 5′-UTR, normal coding sequence (CDS), and mutant N terminally extended CDS. The C to T mutation (chr9: 91926143) within the 5′-UTR of CKS2 generates a start codon that extends the coding sequence of CKS2. The mutated nucleotide is represented in red. b Method of CRISPR-Cas9 base editing using evoAPOBEC1-BE4max-NG, which is composed of APOBEC1, a Cas9-nickase domain, and uracil-DNA glycosylase inhibitor (UGI). This base editor deaminates target cytosines to uracil, which changes the original G–C base pair into an A–T base pair after DNA repair. c Sanger sequencing traces from polyclonal population of CRISPR-transfected 293T cells and six individual single-cell clones selected from this pool for further study. The target C (blue) -> T (red) mutation in the 5′-UTR of CKS2 is shown within the dashed box. d Western blot of the three WT and three CKS2 mutant clonal cell lines created by CRISPR base editing with antibodies against CKS2 and β-actin. The graph shows these results quantified using ImageJ, where each CKS2 band intensity was measured and normalized to the intensity of the corresponding β-actin loading control. Statistics show two-sided Student’s t test with multiple comparisons correction using the three WT (gray) vs three CKS2 mutant (pink) biological replicates (total p value = 0.00001 and 14 kDa p value = 0.00023). Data are presented as mean values ± s.d. Full immunoblots are provided in the Source data file. e CKS2 qPCR shows no change in mRNA levels between three WT (represented in gray) and three mutant (represented in pink) clonal cell lines created from CRISPR base editing. CKS2 mRNA levels in each sample were normalized to β-actin as a loading control (n = 6 biological replicates). Two-sided Student’s t test, data are presented as mean values ± s.e.m. n.s. not statistically significant. Source data are provided as a Source data file.

Fig. 7. 5′-UTR mutations in MAP kinase signaling pathway genes associated with increased pathway activity and metastases in prostate cancer patients.

a Genes with 5′-UTR mutations in localized and advanced prostate cancer cluster into distinct functional categories as determined by KEGG pathway analysis (Fisher’s hypergeometric test, FDR < 0.05). b Heatmap of a MAP kinase pathway activity signature demonstrating that patients with functional 5′-UTR mutations to MAP kinase regulators (PLUMAGE FDR < 0.1) exhibit increased pathway activation compared to nonfunctional mutations (PLUMAGE FDR > 0.1). The heatmap color key represents normalized log₂(CPM + 1) values for each gene. c Metastatic castration-resistant prostate cancer patients harbor 5′-UTR mutations within genes found in the MAP kinase signaling pathway. Gene names in green represent those with 5′-UTR mutations in mCRPC patients. Gene names in gray are MAP kinase signaling pathway components and downstream effectors that are not mutated in mCRPC patients. d mCRPC patients with mutated MAP kinase pathway genes are significantly more prone to bone metastases at diagnosis (black) compared to patients who do not harbor these mutations (green) (p = 0.045, two-sided Student’s t test). e The difference in bone metastasis at diagnosis between the two patient groups is independent of any differences in 5′-UTR tumor mutational burden (n.s. not statistically significant, two-sided Student’s t test). Source data are provided as a Source data file.