RETRACTED: Transcriptome-scale RNA-targeting CRISPR screens reveal essential lncRNAs in human cells

Abstract

This article has been retracted: please see Elsevier policy on Article Withdrawal (https://www.elsevier.com/about/policies-and-standards/article-withdrawal). This article has been retracted at the request of the authors. In Liang et al., the authors recently identified that the CRISPR library used in the pooled screens targeting long noncoding RNAs (lncRNAs) and protein-coding transcripts was not filtered to remove potential off-targets elsewhere in the human transcriptome (as described in the “Method details”). The authors are grateful to the lab of Dr. Roland Rad for bringing this to their attention and apologize for this error. After proper filtering for potential off-targets by removing CRISPR guide RNAs (gRNAs) with up to 2 mismatches to protein-coding transcripts or lncRNAs, the authors have re-analyzed the pooled CRISPR screens. Also, they have updated experiments involving lncRNAs that were targeted using gRNAs with potential off-targets. The updated analysis and experiments are available as a preprint (http://doi.org/10.6084/m9.figshare.30370171). Given the changes in essential lncRNAs identified after proper gRNA filtering, the authors wish to retract the paper and apologize for any inconvenience this may have caused. While the authors work to publish the corrected study, they hope that the updated preprint will be helpful and encourage the community to reach out with any questions.

Keywords: CRISPR-Cas13; MALAT1; MIR17HG; RNA targeting; SLC16A1-AS1; essential genes; human development; lncRNA; noncoding RNA; transcriptome scale; tumor biomarkers.

Figure 1.. Transcriptome-scale RNA-targeting CRISPR screens to identify essential lncRNAs in human cells.

(A) Overview of the Cas13-based loss-of-function screens to identify essential long noncoding RNAs (lncRNAs). lincRNA – long intergenic noncoding RNA, asRNA – antisense RNA, gRNA – guide RNA. (B) Fold-change (FC) of gRNAs targeting lncRNAs in two independent biological replicate pooled screens in HAP1 cells at 14 days after Cas13 induction. Color denotes the number of Cas13 gRNAs. (C) Fold-change (Day 14 vs. Day 0) of five individual gRNAs (pink lines) targeting the indicated genes. The shaded region indicates the 95% confidence interval computed using the distribution of non-targeting (NT) gRNAs. The diamond denotes the mean fold-change of the five gRNAs in HAP1 cells. (D) Ranking of lncRNAs via robust rank aggregation (RRA) in HAP1 screens, based on consistent depletion of five individual guide RNA (gRNA) targeting the same gene. (E) Essential (RRA P < 0.05) and non-essential lncRNAs from the Cas13 screens in the five cell lines. In panels B and C, the dashed lines indicate the 95% confidence interval for NT gRNAs.

Figure 2.. Distinct and common essential lncRNAs across five cell lines.

(A) Enrichment of essential lncRNAs over non-essential lncRNAs for genomic position to nearest protein-coding gene (left) and evolutionary age (right). For evolutionary age, mya denotes million years ago. The odds ratio is determined by a Fisher’s exact test with the significance given by the dot size (dark outline indicates P < 0.05). (B) Distribution of shared, partially shared and cell-type specific essential lncRNAs (left) and protein-coding genes (PCGs) (right) across all five cell lines. Shared genes are essential in all cell lines, and partially shared genes are essential in two to four cell lines. Numbers in parentheses indicate the number of essential genes in each cell line. (C) The proportion of essential lncRNAs and PCGs in each cell line. Fisher’s exact test for essential lncRNAs compared to essential PCGs for each essentiality category (see categories in B). (D) Fold-change (Day 14 vs. Day 0) of cell-type specific, partially shared, and shared essential lncRNAs in HAP1 cells after Cas13 induction. (E) Expression of cell-type specific, partially shared, and shared essential lncRNAs in HAP1 cells. (F) Pearson correlation of essential (upper) and non-essential lncRNAs (lower) expression across five cell lines. (G) Fold-change (Day 14 vs. Day 0, x-axis) and RNA-seq reads/expression (y-axis) for two cell-specific essential and one non-essential lncRNAs. (H) A GFP-labeled competition assay to quantify the impact of knock-down of essential lncRNAs. (I) Representative images of HAP1 (left) and flow cytometry of THP1 (right) cells transduced with individual gRNAs targeting highly expressed lncRNAs indicated in panel G six days after Cas13 induction. Survival of GFP⁺ cells transduced with three non-overlapping gRNAs per gene normalized to the non-targeting (NT) gRNAs (right). Each green circle denotes a single gRNA and single transduction replicate. The diamonds denote the mean survival (n = 6 experiments with three gRNAs from two independent transductions). The dashed lines indicate the 95% confidence interval (CI) for NT gRNAs. Statistical significance was determined by a Student’s t-test. Scale bar for HAP1 images: 200 μm. In panels D and E, box plots indicate the median 25th and 75th percentiles, while whiskers are 1.5 times the interquartile range and statistical significance was determined by a two-sided Mann-Whitney U test.

Figure 3.. Knockdown of shared essential lncRNAs reduces cell survival.

(A) Essentiality, expression levels and genomic classification of shared essential lncRNAs ordered by their median essentiality (n = 5 cell lines). For comparison, six non-essential lncRNAs are included at the bottom of the heatmap. For some lncRNAs, a prior CRISPRi screen examined essentiality and we have indicated those lncRNAs found to be essential in any cell line screened (n = 7 cell lines screened with CRISPRi). (B) Representative images of HAP1 cells transduced with individual gRNAs targeting shared essential lncRNAs indicated in panel A four days after Cas13 induction (left). Survival of GFP⁺ cells transduced with three non-overlapping gRNAs per gene normalized to the non-targeting (NT) gRNAs in HAP1, MDA-MB-231 and HEK293FT (right). Each green circle denotes a single gRNA and single transduction replicate. The diamonds denote the mean survival (n = 6 experiments with three gRNAs from two independent transductions). The gray shaded area indicate the 95% confidence interval (CI) for NT gRNAs. (C) Fluorescence ubiquitination cell cycle indicator (FUCCI) assay to measure cell fraction in different cell cycle phases. (D) Median distribution of MDA-MB-231 cells transduced with indicated gRNAs in cell cycle phases G1 (red), S (green) and G2-M (yellow) 24, 48 and 72 hours after Cas13 induction (n = 54 images per perturbation and timepoint with 9 images per biological replicate and 6 biological replicates per perturbation). P values from the predominantly enriched cell cycle phase (determined for each lncRNA individually) were computed by a Mann-Whitney U test to test for differences from cells transduced with non-targeting gRNAs. (E) Apoptosis assay using Annexin V staining. (F) Representative images of MDA-MB-231 cells transduced with gRNAs targeting shared essential lncRNAs at 72 hours after Cas13 induction (left). Annexin V⁺ cells were quantified and normalized to the total cell area (right). Each pink circle denotes a single gRNA and single transduction replicate. The diamonds denote the mean survival (n = 54 images per perturbation with 9 images per biological replicate and 6 biological replicates per perturbation). In panels B and F, statistical significance was determined by a Student’s t-test. Scale bar: 200 μm.

Figure 4.. Nearest protein-coding genes of essential lncRNAs are often not essential.

(A) Number of essential lncRNAs and protein-coding genes (PCGs) across five cell lines. (B) Alluvial diagram of lncRNA-PCG pairs, depicting pairs where only the lncRNA is essential, where only the PCG is essential and where both the lncRNA and nearest PCG are essential. Numbers in parentheses indicates lncRNA-PCG pairs with at least one essential gene in each cell line. (C) Fold-change (Day 14 vs. Day 0) of lncRNAs and PCGs in each lncRNA-PCG pair in HAP1 cells after Cas13 induction. The pairs are separated by those pairs where only the lncRNA is essential (left), where only the PCG is essential (middle), and where both the lncRNA and nearest PCG are essential (right). (D) Examples of lncRNA-PCG pairs where one or both genes are essential. Fold change of five individual gRNAs targeting the indicated genes with the 95% confidence interval (range) of non-targeting gRNAs (gray). The diamond denotes the mean of the five gRNAs. (E) Representative images of HAP1 cells transduced with individual gRNAs targeting indicated genes five days after Cas13 induction (left). Survival of GFP⁺ cells transduced with three non-overlapping gRNAs per gene normalized to the median of non-targeting (NT) gRNAs (right). Each green circle denotes a single gRNA and single transduction replicate. The diamonds denote the mean survival (n = 6 experiments with three gRNAs from two independent transductions). The dashed lines indicate the 95% confidence interval (CI) for NT gRNAs. Statistical significance was determined by a Student’s t-test. Scale bar: 200 μm. (F) Essentiality of the closest PCGs (left) and the distance between lncRNAs and closest PCGs (right) for the shared essential lncRNAs. Orange boxes indicate that the closest PCG is essential. For distances, pink lines denote a distance of less than 1 kb between the lncRNA and PCG and blue lines denote a distance greater than 1 kb. (G) Key mechanistic differences in knock-down of lncRNAs and nearby genes with DNA-targeting CRISPRi or RNA-targeting Cas13 (upper). The proportion of essential closest PCGs for lncRNAs identified in this study and a prior lncRNA pooled CRISPRi screen (lower). Dot size corresponds to the number of essential lncRNAs identified. Common cell lines used in both studies (HEK293FT, K562, MDA-MB-231) are labeled in pink, and the study-specific cell lines are labeled in yellow. (H) The fraction of essential lncRNAs identified in the CRISPRi study and the present study, categorized by the distance to their nearest PCG and whether they were identified as essential in the DNA-targeting (CRISPRi) study, this RNA-targeting (Cas13) study or both studies. (I) The fraction of essential lncRNAs identified in both DNA- and RNA-targeting studies, categorized by essentiality level (left) and all essential lncRNAs (right) from this (RNA-targeting) study.

Figure 5.. Single-cell transcriptomics after Cas13 perturbation (CaRPool-seq) of essential lncRNAs identifies shared cellular pathways for proliferation.

(A) Schematic of Cas13 RNA Perturb-seq (CaRPool-seq) using guide RNA (gRNA) arrays that encode two gRNAs that target the same gene (lncRNA or protein-coding). Each array also contains a barcode gRNA (bcgRNA) to enable identification of the gRNA array using single-cell sequencing. (B) Correlation between fold-change (FC) from the transcriptome-scale pooled screen (Day 14 vs. Day 0) and the CaRPool-seq pooled screen (Day 12 vs. Day 0) for 50 essential lncRNAs (purple) and 21 protein-coding genes (orange and green) in MDA-MB-231 cells. (C) Single-cell mRNA expression heatmap with the 25 most differentially downregulated genes for each lncRNA perturbation in MDA-MB-231 cells (P_adj < 0.05). For each lncRNA, transcripts with the lowest median DepMap scores are labeled (n = 3 most essential transcripts per lncRNA and median over 1,095 DepMap cell lines). (D) Normalized enrichment scores (NES) from gene set enrichment analysis (GSEA) for 21 perturbed protein-coding genes (PCGs) (left) and 50 perturbed essential lncRNAs (middle) in MDA-MB-231 cells. The number of lncRNAs with the indicated pathway (MSigDB Hallmark pathways) enriched or depleted (P_adj < 0.05, black dots) in CaRPool-seq from MDA-MB-231 and HAP1 cells (right). Pathways categorized as proliferation or apoptosis are labeled (far left column). The fold-change in gRNA array abundance from the pooled readout of the MDA-MB-231 CaRPool-seq is shown at the bottom.

Figure 6.. Essential lncRNAs are expressed broadly across different tissues and at early stages of development.

(A) Transcriptome profiles for each tissue at various developmental time points from recent developmental atlases of lncRNA and protein-coding gene (PCG) expression (n = 182 tissue samples)^,. (B) Empirical cumulative distributions of tissue-specificity (left) and time-specificity (right) indices for essential and non-essential lncRNAs (two-sided Mann-Whitney U test). (C) The proportion of dynamic lncRNAs for essential and non-essential lncRNAs (Fisher’s exact test in comparison to non-essential lncRNAs). (D) The median expression of dynamic shared essential (purple) and non-essential (turquoise) lncRNAs at different developmental time points in each tissue. The heatmaps below provide annotations for the relative abundance of proliferation markers PCNA and MKI67. (E) Gene set enrichment analysis (GSEA) of co-expressed PCGs for essential and non-essential lncRNAs across tissues, represented by the median normalized enrichment scores (NES) across 50 GSEA hallmark pathways. (F) NES of proliferation- or apoptosis-associated pathways (a subset of MSigDB Hallmark pathways) for each of the shared essential lncRNAs in brain (left). The fraction of shared essential lncRNAs with proliferation- or apoptosis-associated pathways enriched or depleted (P_adj < 0.05, black dots) across different tissues (right).

Figure 7.. Essential lncRNAs are differentially expressed in tumors and correlate with survival.

(A) Computational workflow to identify functional lncRNAs across 29 cancer types from The Cancer Genome Atlas (TCGA). (B) Median expression of essential and non-essential lncRNAs in primary tumors from 29 cancer types (two-sided Mann-Whitney U test). (C) The fraction of differentially expressed lncRNAs in primary tumors compared to normal tissue across 22 cancer types (bottom). The pie chart (top) separates differentially expressed essential lncRNAs by expression level: High (increased expression in tumor), Low (decreased expression in tumor), and Mix (varying expression in different cancer types). The number of cancer types with significant differential expression for the example lncRNAs is indicated in parentheses (Fisher's exact test). (D) Gene set enrichment analysis (GSEA) of co-expressed PCGs for essential and non-essential lncRNAs, represented by the median normalized enrichment scores (NES) across 50 MSigDB Hallmark pathways. Marker size represents the fraction of lncRNAs within each respective group. Only enriched pathways in the top quartile (by NES) are plotted. (E) The proportion of lncRNAs associated with better or worse overall survival or progression-free survival. (F) Survival analysis coefficients for TCGA tumor types for the 46 shared essential lncRNAs. LncRNAs with negative coefficients (pink) are associated with worse survival outcomes (increased hazard). LncRNAs with positive coefficients (green) are associated with better survival outcomes (decreased hazard). Rectangles denote P < 0.05 and black dots denote P_adj < 0.05 (logrank test). (G) Kaplan-Meier progression-free (left) or overall survival (right) estimates for patients with prostate adenocarcinoma (PRAD), low-grade glioma (LGG), kidney renal clear cell carcinoma (KIRC), and uterine corpus endometrial carcinoma (UCEC) stratified by the expression of SLC16A1-AS1 (BH-adjusted logrank test). Sample sizes for high and low expression groups: LGG: n = 257 (high), n = 259 (low). KIRC: n = 265 (high), n = 265 (low). UCEC: n = 272 (high), n = 263 (low).