Tumour mutations in long noncoding RNAs enhance cell fitness

Abstract

Long noncoding RNAs (lncRNAs) are linked to cancer via pathogenic changes in their expression levels. Yet, it remains unclear whether lncRNAs can also impact tumour cell fitness via function-altering somatic "driver" mutations. To search for such driver-lncRNAs, we here perform a genome-wide analysis of fitness-altering single nucleotide variants (SNVs) across a cohort of 2583 primary and 3527 metastatic tumours. The resulting 54 mutated and positively-selected lncRNAs are significantly enriched for previously-reported cancer genes and a range of clinical and genomic features. A number of these lncRNAs promote tumour cell proliferation when overexpressed in in vitro models. Our results also highlight a dense SNV hotspot in the widely-studied NEAT1 oncogene. To directly evaluate the functional significance of NEAT1 SNVs, we use in cellulo mutagenesis to introduce tumour-like mutations in the gene and observe a significant and reproducible increase in cell fitness, both in vitro and in a mouse model. Mechanistic studies reveal that SNVs remodel the NEAT1 ribonucleoprotein and boost subnuclear paraspeckles. In summary, this work demonstrates the utility of driver analysis for mapping cancer-promoting lncRNAs, and provides experimental evidence that somatic mutations can act through lncRNAs to enhance pathological cancer cell fitness.

Fig. 1. Driver lncRNA discovery with ExInAtor2.

a ExInAtor2 accepts input in the form of maps of single-nucleotide variants (SNVs) from cohorts of tumour genomes. Two signatures of positive selection are evaluated, and compared to simulated local background distributions to evaluate statistical significance. The two significance estimates are combined using Fisher’s method. b Summary of the primary tumour datasets used here, obtained from Pancancer Analysis of Whole Genomes (PCAWG) project. c A filtered lncRNA gene annotation was prepared, and combined with a set of curated cancer lncRNAs from the Cancer LncRNA Census.

Fig. 2. ExlnAtor2 accurately identifies driver genes.

a The list of driver discovery methods to which ExInAtor2 was compared. The signatures of positive selection employed by each method are indicated to the right. PCAWGc indicates the combined driver prediction method from Pan-Cancer Analysis of Whole Genomes (PCAWG), which integrates all ten methods. b Benchmark gene sets. LncRNAs (blue) were divided in positives and negatives according to their presence or not in the Cancer LncRNA Census, respectively, and similarly for protein-coding genes in the Cancer Gene Census. c Comparing performance in terms of precision in identifying true positive known cancer lncRNAs from the CLC dataset, using PCAWG Pancancer cohort. x axis: genes sorted by increasing P value (uncorrected for multiple hypothesis testing) from each correspondent method, as described in ref. . y axis: precision, being the percentage of true positives amongst cumulative set of candidates at increasing P value cutoffs. The horizontal line shows the baseline, being the percentage of positives in the whole list of tested genes. Coloured dots represent the precision at cutoff of q ≤ 0.1 (Benjamini–Hochberg method). Inset: Performance statistics for cutoff of q ≤ 0.1. d Driver prediction performance for all methods in all PCAWG cohorts. Cells show the F1-score of each driver method (x axis) in each cohort (y axis). Grey cells correspond to cohorts where the method was not run. The bar plot on the top indicates the total, non-redundant number of True Positives (TP) and False Positives (FP) calls by each method. Driver methods are sorted from left to right according to the F1-score of unique candidates. e Evaluation of P value distributions for driver lncRNA predictions. Quantile–quantile plot (QQ-plot) shows the distribution of observed vs expected –log10 P values (uncorrected for multiple hypothesis testing) for each method run on the PCAWG Pancancer cohort (uncorrected for multiple hypothesis testing; as described in ref. ). The Mean Log-Fold Change (MLFC) quantifies the difference between observed and expected values (“Methods”).

Fig. 3. The landscape of driver lncRNAs in primary tumours.

a “Oncoplot” overview of driver lncRNA analysis in PCAWG primary tumours. Rows: 17 candidate driver lncRNAs at cutoff of FDR ≤ 0.1. Columns: 2580 tumours. b LncRNA candidates across all cohorts. Rows: Cohorts where hits were identified. Columns: 17 candidate driver lncRNAs. “Known” lncRNAs are part of the literature-curated Cancer LncRNA Census (CLC2) dataset. Functional labels (oncogene/tumour suppressor/both) were also obtained from the same source. c Intersection of candidate driver lncRNAs identified in PCAWG primary tumours, Hartwig Medical Foundation (HMF) metastatic tumours and the CLC2 published cancer-lncRNA set. Statistical significance was estimated by Fisher’s exact test. d Genomic features of driver lncRNAs. Each plot displays the values of indicated features for 17 candidate driver lncRNAs (blue) and all remaining tested lncRNAs (non-candidates, grey). Significance was calculated using two-sided Wilcoxon test, (uncorrected for multiple hypothesis testing). For each comparison, the ratio of means was calculated as (mean of candidate values/mean of non-candidate values). Centre = medians (Line), bottom and top boundaries of the box = 25 and 75th percentiles of the data, minima and maxima = lowest and highest data points. See “Methods” for more details. e Clinical features of driver lncRNAs. Each point represents the indicated feature. y axis: log2-transformed ratio of the mean candidate value and mean non-candidate value. x axis: The statistical significance of candidate vs non-candidate values, as estimated by a two-sided Wilcoxon test and corrected for multiple testing with Benjamini–Hochberg method. See “Methods” for more details. Source data are provided as a Source Data file.

Fig. 4. Functional effects of driver lncRNAs in cell viability.

a Plasmid-transfected cells were measured at indicated timepoints. Statistical significance was estimated by two-sided Student’s t test based on n = 3 independent replicates. Mean value +/− SD is plotted. Replicates were performed at different times (experimental replicates). b Overexpression of LOHAN1&2 in HN5 cells. RNA levels were measured by qRT-PCR; n = 3 (experimental replicates performed at different times). c Results of colony formation assay in HN5 cells. Data are presented as mean values −/+ SD of the percent of well area covered from 18 culture wells. Statistical significance was estimated using one-way ANOVA. Replicates were performed at the same time (technical replicates). d The genomic locus of the lncRNA LOLI1. Also shown are SNVs from PCAWG and HMF cohorts. The SNVs included in the mutated plasmid are indicated in the grey boxes. e ASOs were transfected to knock down LOLI1 expression and f RNA levels measured in HuH7 cells. Statistical significance was estimated using one-sided Student’s t test; n = 3 (experimental replicates). g ASO-transfected cells were measured at indicated timepoints; n = 3. Statistical significance was estimated by linear regression model on log2 value (experimental replicates). h CRISPRa targeting LOLI1. On the right, qRT-PCR measurements of LOLI1 with indicated sgRNAs; n = 3 (experimental replicates). i The effect of CRISPRa on HeLa cells’ viability; n = 6 (experimental replicates). Statistical significance was estimated by one-sided paired t test at 48 h. j Plasmids expressing spliced LOLI1 sequence, in wild-type (WT) or mutated form (Mut) were transfected into HuH7 cells. RNA levels were measured by qRT-PCR; n = 3 (experimental replicates). k Populations of plasmid-transfected cells were measured at indicated timepoints. Statistical significance was estimated by one-sided Student’s t test based; n = 3 (experimental replicates). l LOLI1 overexpression in immortalised human hepatocytes (IHH). RNA levels were measured by qRT-PCR; n = 3 (experimental replicates). m The viability of IHH was measured at indicated timepoints. Statistical significance was calculated by one-sided Student’s t test; n = 3 (experimental replicates). n Primary human hepatocytes were transduced to overexpress LOLI1 mutant transcript (left panel). Transduction was monitored by EGFP marker gene (left panel). The change in proliferation-associated cytokine was measured by qRT-PCR (right panel); n = 6 experimental replicates. n indicates the number of independent experiments. Data show the mean value +/− SD in (a–c, f–n). Source data are provided as a Source Data file.

Fig. 5. Mutations in NEAT1 promote cell fitness and correlate with survival.

a Experimental strategy to simulate tumour-like mutations in the NEAT1 gene by Cas9 protein. b Detailed map of the six NEAT1 target regions and 15 sgRNAs. Paired gRNAs used for the deletion of NEAT1_1 are indicated as KO- sgRNA1 and KO- sgRNA2. Previously described NEAT1 functional regions are indicated below. c Analysis of mutations created by Cas9 recruitment. The frequency, size and nature of resulting DNA mutations are plotted. d Competition assay to evaluate fitness effects of mutations. Above: Rationale for the assay. Below: Red/green ratios for indicated mutations. “Control1/2” indicate sgRNAs targeting AAVS1 region. “KO” indicates paired sgRNAs designed to delete NEAT1_1. N = 4 experiments were performed, and statistical significance was estimated by linear regression model on log2 values. The mean value −/+ SD is plotted. Replicates were performed at different times (experimental replicates). e Upper panel: Set-up of mini CRISPR fitness screen. HeLa cells are infected with lentivirus-carrying mixtures of sgRNAs. The sgRNA sequences are amplified and sequenced at defined timepoints. Lower panel: Abundances of displayed sgRNAs, normalised to the Control2 negative control. Statistical significance was estimated by linear regression model; n = 4 (experimental replicates). The mean value with SD is represented. f HCT116 cells were cultured as spheroids and their population measured. Data show the mean value −/+ SD of n = 4 (experimental replicates). Statistical significance was estimated using Student’s one-sided t test. g As for (d), but with non-transformed MRC5 lung fibroblast cells at timepoint Day 14. Statistical significance was estimated by one-sided Student’s t test. Data show the mean value -/+ SD of n = 3 (experimental replicates). h MRC5 cells were seeded in soft agar, and the area of colonies was calculated. The mean and SD of n = 3 experiments is shown (experimental replicates). i NEAT1 mutations in Reg2 enhances cell growth in NSG mice. HeLa cells were mutated and then implanted subcutaneously. Resulting tumour weight is shown at 4 weeks post-transplantation. Statistical significance was estimated by one-sided Student’s t test based on n = 9 animals. Experiments were pooled from two groups of animals studied at different times. Data show the mean value -/+ SD. j The survival time of 184 lymphoid cancer patients from PCAWG is displayed. Patients were stratified according to whether they have ≥1 SNVs in the NEAT1 gene. Two-sided Gehan Breslow rank test with confidence interval style set to dotted lines (95% CI). Source data are provided as a Source Data file.

Fig. 6. Mutations at the 5′ end of NEAT1 increase paraspeckle formation and alter its protein interactome.

a Counts of paraspeckles and b paraspeckle size in HeLa cells treated with indicated sgRNAs. Values were obtained from 80 to 100 cells per replicate, with n = 5 (experimental replicates performed at different times). Statistical significance was estimated using one-tailed paired t test. Data show the mean value +/− SD. c Representative images from fluorescence in situ hybridisation (FISH) visualisation of NEAT1 in HeLa cells expressing sgRNAs for Control2 and NEAT1 Region 2. n > 3 biological replicates. Scale bar = 10 µM. d Sequences of biotinylated probes used for the mass-spectrometry analysis of NEAT1-interacting proteins. e Proteins detected by wild-type (WT) NEAT1 probe, filtered for nuclear proteins only, are ranked by intensity and labelled when intersecting previously detected NEAT1-interacting proteins (green) and paraspeckle proteins (orange). Statistical significance was calculated by one-tailed hypergeometric test (to background of nuclear proteins n = 6758). f Histogram shows differential detection of proteins comparing mutated (Mut) and wild-type (WT) probes. Dotted lines indicate log2 fold-change cutoffs of −1/+ 1. g STRING interaction network based on a subset of the proteins lost upon mutation (grey borders) interacting with the RNA polymerase II core complex. h Validations by RNA immunoprecipitation using antibodies for PQBP1 and SREK1. i FISH using NEAT1 probes (green) in HeLa cells treated with indicated siRNAs. Cell nuclei in blue (DAPI). Values were obtained from 80 to 100 cells per replicate, with n = 3 replicates (experimental replicates). Scale bar = 10 µM. j qRT-PCR measurement of RNA levels in HeLa cells after transfection of siRNAs targeting U2SURP and SREK1 genes. Data show the mean value +/− SD of n = 3 independent experiments. k Paraspeckles area in HeLa cells treated with two independent siRNAs targeting SREK1. Measurements from 80 to 100 cells per replicate. Statistical significance was estimated using one-sided paired t test. Data shows the mean value +/− SD of n = 3 replicates (experimental replicates). l Cell viability in siRNA-transfected cells was measured at indicated timepoints. Data shows the mean value +/− SD of n = 3 replicates (experimental replicates). Statistical significance was estimated by using tow-tailed t test. m Proposed model by which somatic mutations in NEAT1 gene can alter protein interactome, increase paraspeckle numbers and boost cell proliferation. Source data are provided as a Source Data file.