https://orcid.org/0000-0002-3214-897X
Figures
Abstract
Introduction
Glioblastoma (GBM) invasion studies have focused on coding genes, while few studies evaluate long non-coding RNAs (lncRNAs), transcripts without protein-coding potential, for role in GBM invasion. We leveraged CRISPR-interference (CRISPRi) to evaluate invasive function of GBM-associated lncRNAs in an unbiased functional screen, characterizing and exploring the mechanism of identified candidates.
Methods
We implemented a CRISPRi lncRNA loss-of-function screen evaluating association of lncRNA knockdown (KD) with invasion capacity in Matrigel. Top screen candidates were validated using CRISPRi and oligonucleotide(ASO)-mediated knockdown in three tumor lines. Clinical relevance of candidates was assessed via The Cancer Genome Atlas(TCGA) and Genotype-Tissue Expression(GTEx) survival analysis. Mediators of lncRNA effect were identified via differential expression analysis following lncRNA KD and assessed for tumor invasion using knockdown and rescue experiments.
Results
Forty-eight lncRNAs were significantly associated with 33–83% decrease in invasion (p<0.01) upon knockdown. The top candidate, LINC03045, identified from effect size and p-value, demonstrated 82.7% decrease in tumor cell invasion upon knockdown, while LINC03045 expression was significantly associated with patient survival and tumor grade(p<0.0001). RNAseq analysis of LINC03045 knockdown revealed that WASF3, previously implicated in tumor invasion studies, was highly correlated with lncRNA expression, while WASF3 KD was associated with significant decrease in invasion. Finally, WASF3 overexpression demonstrated rescue of invasive function lost with LINC03045 KD.
Author summary
Glioblastoma is the most common malignant brain tumor to originate from brain tissue in adults. Most research has focused on the role of genes that code for protein, but genes that do not code for protein can still produce long non-coding RNAs (lncRNAs) that have been demonstrated to have important functions in cancer processes. We utilized a method called CRISPRi to screen for lncRNAs that are important in glioblastoma invasion, which is the capacity of cancer cells to invade into surrounding tissue. With this CRISPRi invasion screen, we identified LINC03045 as a gene that affects glioblastoma invasion. We validated this finding using cultured cancer stem cells and found that knocking down this gene decreased invasion. We also analyzed the public databases for clinical relevance, and found that LINC03045 was associated with patient survival and tumor grade. We also identified the gene WASF3, previously implicated in multiple tumor invasion studies (an element of the JAK-STAT pathway) as a downstream element of LINC03045. Thus, here, we have identified LINC03045 as a critical non-coding gene in glioblastoma invasion, and further characterizing this gene could potentially make it a therapeutic target for glioblastoma therapy in the future.
Citation: Tsung K, Liu KQ, Han JS, Deshpande K, Doan T, Loh Y-HE, et al. (2024) CRISPRi screen of long non-coding RNAs identifies LINC03045 regulating glioblastoma invasion. PLoS Genet 20(6): e1011314. https://doi.org/10.1371/journal.pgen.1011314
Editor: Timothy E. Reddy, Duke University, UNITED STATES
Received: May 10, 2023; Accepted: May 21, 2024; Published: June 10, 2024
Copyright: © 2024 Tsung et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The data that support the findings of this study are openly available in the NCBI Gene Expression Omnibus (GEO) database, accession number GSE268342. Other data utilized in this study are available publicly in the Genotype-Tissue Expression Project, Gene Ontology, The Cancer Genome Atlas, and the Kyoto Encyclopedia of Genes and Genomes databases.
Funding: FJA was funded by the Margaret E. Early Medical Research Trust (https://rii.usc.edu/limited-submissions/margaret-e-early/), NIH K08 Grant 5K08NS114172-02 (https://www.nia.nih.gov/research/training/k08-mentored-clinical-scientist-development-awards), NIH KL2 Grant 5KL2TR001854-04, and the Keck School of Medicine Dean’s Pilot Funding Program (https://keck.usc.edu/research-funding/programs-administered-by-ksom/). LD was supported by Grant UL1TR001855 from the National Center for Advancing Translational Science (NCATS) of the U.S. National Institutes of Health (https://ncats.nih.gov/). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The authors did not receive a salary from any of the funders.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Glioblastoma (GBM) is the most common and aggressive adult primary malignant brain tumor, with an estimated annual incidence of 17,000 new cases per year in the USA [1]. Standard of care treatment includes surgical resection, chemotherapy with temozolomide (TMZ), and radiation therapy, but despite the available therapeutic options, mean survival remains at about 12 to 15 months and the five-year survival rate at 5.5% [2–5]. The majority of GBM patients also invariably experience tumor recurrence with increased chemo- and radio-resistance, and there is no standardized second line of management after initial adjuvant treatment. Further evidence suggests that microscopic tumor invasion beyond gross radiographic and intraoperatively visualized tumor specimen contributes to continued tumor growth despite gross total radiographic resection [6–8]. The limited efficacy of treatment options necessitates novel therapeutic strategies and molecular targets that can limit this tumor invasion and subsequent tumor recurrence.
While current genetic studies in glioma research have primarily focused on targeting coding genes, about 80% of the genome is composed of non-coding RNAs [9,10]. Long non-coding RNAs (lncRNAs) are transcripts greater than 200 nucleotides that do not code for proteins but have demonstrated involvement in transcriptional and post-transcriptional gene regulation. LncRNAs have been shown to play roles in the pathogenesis of multiple malignancies. Notably, because they exhibit highly cell type-specific expression and function, they have become attractive novel targets of gene therapy [9,11–18]. To identify novel genetic therapeutic targets in cancer treatment, systematic functional screens have recently been successfully employed to rapidly and efficiently identify transcripts could sensitize cancer cells to treatment. Despite the rapidly emerging use of these functional screens to identify novel genetic targets, few studies have specifically identified novel candidate lncRNAs targeting glioma.
Clustered regularly interspaced short palindromic repeats (CRISPR) is a technology that has allowed for genome-wide screens of gene function [19–26]. This technology has been valuable in the identification of both coding and non-coding genes that are involved in various cellular phenotypes. Due to the nature of non-coding RNAs, standard CRISPR knockdown may not effectively limit lncRNA function, as small deletions do not typically reduce lncRNA-induced effects [20,23,27]. CRISPRi (CRISPR interference) circumvents this challenge, repressing transcription of the lncRNA at the nucleus via fusion of a catalytically inactive Cas9 (dCas9) with a repressive KRAB domain [28]. We have previously successfully utilized these CRISPRi screens to assess in vitro and in vivo GBM proliferation [19,29]. CRISPRi screens simultaneously repress multiple genetic targets, with each cell receiving specific individual lncRNA repression. These pools of lncRNA cellular perturbation are then selected for those cells displaying a phenotype of interest. Association of the specific lncRNAs knocked down in remaining selected cells then identifies the lncRNAs that are most closely associated with the selected phenotype.
Here, we developed a CRISPRi-based invasion screen to identify lncRNAs that affect the in vitro invasive capacity of glioma cells. In this screen of 2,307 candidate lncRNAs, we sought to identify specific lncRNAs associated with glioma cell invasion. We further assess the clinical association of candidate lncRNAs with tumor grade and patient outcomes while also evaluating potential mechanisms by finding associations of candidate lncRNAs with potential coding genes affecting GBM invasion.
Results
CRISPRi screen identifies lncRNAs that affect invasion
To screen for and identify lncRNAs affecting in vitro glioma cell invasion, we utilized an invasion screen with CRISPRi as a tool for gene knockdown. Repression of transcription occurred when CRISPRi recruited the Cas9 protein fused to a KRAB repressor (dCas9-KRAB), which is guided and targeted to transcriptional start sites (TSS) by single guide RNAs (sgRNAs) [19]. A GBM cell line (U87) was successfully tranduced with a custom plasmid to express dCas9-KRAB (U87-dCas9-KRAB) fused to blue fluorescent protein (U87-dCas9-KRAB-BFP), and sorted by FACS to a pure population of U87-dCas9-KRAB. Transduced cells were successfullly infected with lentivirus that contained a U87-specific CRISPRi library derived from CRinCL (Addgene #86542, a gift from Jonathan Weissman). This library was comprised of sgRNAs targeting 2,307 lncRNA, with 10 sgRNAs per lncRNA, and 247 negative control sgRNAs, for a total of 23,564 sgRNAs. U87 cells stably expressing dCas9-KRAB were infected such that there was 1,000X coverage per sgRNA, and we selected for infected cells using puromycin (Fig 1A). A low infection efficiency (MOI < 0.3) was utilized to decrease chances of cells being infected by more than one sgRNA. Cells were then seeded in Matrigel-coated Boyden chambers for 24 hours at a density of 5 x 105 cells per chamber, with serum-free medium in the top chamber and 10% fetal bovine serum medium in the bottom chamber as the chemoattractant for 24 hours. At the completion of this invasion period, screen population cells remaining in the top chamber were successfully isolated and collected. DNA was extracted from cells, followed by PCR amplification and isolation of lentivirally incorporated sgRNA screen fragments, prior to analysis by deep sequencing.
A) Schematic of the screening protocol. The CRISPRi library was packaged into lentivirus and transduced into U87 cells stably expressing dCas9-KRAB, selected for to obtain 1000 fold coverage/sgRNA, and seeded onto Matrigel Boyden chambers. After 24 hours, cells in the top chamber were harvested and processed for deep sequencing. Created by Biorender.com. B) Volcano plot of invasion phenotype and negative log10(Mann-Whitney p-value). Screen replicates were averaged and the top 3 sgRNAs for each lncRNA compared to non-targeting controls were used to determine screen hits. The dashed lines represent thresholds to determine screen hits.
The process was repeated to attain two independent replicates, which were averaged together for downstream analysis (S2 Fig). To determine the lncRNAs with the most notable effects on invasion, screen scores, which were defined as the average phenotype of the top three sgRNAs for a given gene multiplied by the negative log10(Mann-Whitney p-value), were compared to non-targeting controls and plotted on a volcano plot, with the dashed lines representing the threshold to determine screen hits (Figs 1B and S1). These dashed lines were chosen based on threshold cutoffs of prior screens, with the threshold for the y-axis cutoff was -log(0.05). This chosen threshold selects lncRNAs with a minimum effect size of greater than approximately 25% at the p-values approach zero (at the highest y-axis -log[p-value] points). Our screen score thus incorporates both p-value and effect size, as previously described [19]. In the non-invading cells, 45 gene targets were significantly enriched in the library, which represents genes that are necessary for glioma cell invasion. Positive hits were identified through this process, and the top candidate was identified as LINC03045 (annotation LH02236) on chromosome 10. Another candidate, ZNF215-AS1 (annotation LH02727), was also identified on chromosome 11, but this study will focus on LINC03045 since subsequent investigation demonstrated that it had a stronger phenotype. Preliminary results, including clinical correlations with tumor grade, patient survival, and effects of in vitro knockdown on invasion and growth for ZNF215-AS1 are shown in S3 Fig.
Patient outcome analysis via TCGA and GTEx analysis
Since LINC03045 was previously unannotated in genome maps, we utilized position coordinates to assess expression levels of LINC03045, measured in counts per million (CPM). Both GBM (3.6±0.27 CPM) and LGG (2.06±0.30 CPM) showed higher average LINC03045 gene expression than normal brain cortex (0.67±0.06 CPM) (mean ± SEM) (Fig 2A). Pairwise comparisons of LINC03045 in tumors vs control showed a significant fold change in expression (GBM vs control: 6.70, p≤0.001; LGG vs control: 6.25, p≤0.001).
A) Boxplots of pairwise comparisons of LINC03045 expression in normal human brain cortex to low grade glioma (LGG) and glioblastoma (GBM) patient samples. Pairwise comparisons of LINC03045 expression in patient samples (GBM and LGG) and normal human brain cortex show significantly higher expression in patient tissue than normal tissue (p≤0.001). Expression levels are normalized to standardized genes on the same chromosome (chr10). B) Kaplan-Meier survival curves of LINC03045 for all gliomas (stages 2–4) with expression quantified by normalized counts per million (CPM). Cohorts divided into quartiles (left) show significant decrease in survival with increased expression levels (p<0.0001). Cohorts divided into high vs low risk groups (right) are defined at 2.5 LINC03045 CPM. The high-risk group (>2.5 CPM) has significantly lower survival (p<0001). Per 1 CPM increase, hazards regression (HR) is 1.99 from 0 to 2.5 CPM. HR meets Cox regression assumption. C) Kaplan-Meier survival curves of LINC03045 for GBM (stage 4, left) and LGG (stages 2–3, middle) adjusted for IDH-status with expression quantified by normalized counts per million (CPM). Cohorts divided into quartiles show a significant decrease in survival with increased expression levels in LGG (p = 0.002), but not GBM (p = 0.40). Survival analysis of IDH-mutant grade 2 alone does not show a significant decrease in survival with increased expression levels (p = 0.18) (right). Position coordinates were used to identify LINC03045 in RNA-seq data. Expression was normalized to standardized genes on the same chromosome (chr 10). Shading demonstrates the confidence interval. Censored data indicates patients lost to follow-up. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.
Kaplan-Meier (KM) survival analysis of patients with all gliomas (stages 2–4), was conducted based on LINC03045 counts. The median expression level was 1.65 CPM (IQR 0.72–3.09) and the median survival time was 460.5 days (IQR 231.3–854.3). Log-rank tests for this cohort showed that patient cohorts with elevated expression of LINC03045 were significantly associated with decreased survival relative to patient cohorts with lower LINC03045 expression (p<0.0001) (Fig 2B, left). A Cox regression analysis of this cohort showed a dose-response between LINC03045 expression and survival when the expression level was < 2.5 CPM. The risk maximized and remained constant when the expression level was > = 2.5 CPM. The hazard ratio (HR) is 1.99 (95% CI: 1.61, 2.45, p<0.0001) per 1 CPM increase when the expression level is 0–2.5 CPM. HR is 0.99 (95% CI 0.95–1.03, p = 0.61) when the expression level is > = 2.5 CPM (Fig 2B, right).
We also analyzed survival adjusting both for glioma grade and IDH status. In the GBM cohort (stage 4), there was no significant difference in survival with changes in LINC03045 expression (p = 0.40) (Fig 2C, left). The median expression was 2.77 CPM (IQR 1.46–4.65) median survival was 360 days (IQR 160–532) for the GBM cohort. Because the TCGA stratifies lower grade gliomas (grade 2 and 3) together due to nonspecific grading data on 20 tumors that could be either grade 2 or 3, we conducted an IDH-adjusted survival analysis of this cohort. In the LGG cohort (grades 2–3) including the 20 patients with uncertain grade 2 or 3 tumors, elevated expression of LINC03045 was significantly associated with a decrease in survival (p = 0.002) (Fig 2C, middle). The median expression was 0.92 CPM (IQR 0.50–1.77) and median survival time was 656 days IQR (417–1294) for this cohort. Notably, there was no significant difference in LINC0304 expression beetween grade 2 and 3 gliomas within this cohort. A separate survival analysis of IDH-mutant grade 2 gliomas alone (excluding the 20 samples with nonspecific grading) showed that LINC03045 expression was not associated with survival (p = 0.18) (Fig 2, right). Notably, all analyses were also adjusted for IDH-status, MGMT methylation and 1p/19q codeletion when data was available, though inclusion of variables did not significantly alter effect sizes or p-values of LINC03045 in Cox regression (not changing from significant to non-significant or vice versa).
Screen-identified lncRNA phenotypes are reproducible with individual sgRNA knockdown
To validate that the top screen candidates had reproducible phenotypes upon individual lncRNA repression, CRISPRi knockdown (KD) of these gene hits was conducted by transducing U87 Cas9 cells with the top two sgRNAs for each candidate. Effect size and reproducibility were most apparent with LINC03045 and the following primary results will focus on LINC03045.
Transcriptional expression was evaluated after LINC03045 CRISPRi KD with qPCR, demonstrating that U87 CRISPRi KD cells had decreased transcriptional expression of LINC03045 (sgRNA1: 33.4% decrease, p-value = 0.04; sgRNA2: 44.1% decrease, p = 0.02) compared to control (Fig 3A). To validate changes in invasion following CRISPRi KD, invasion assays with LINC03045 CRISPRi KD were conducted. CRISPRi KD cells demonstrated decreased invasion (sgRNA1: 82.7% decrease, p < 0.0001; sgRNA2: 79.7%, p < 0.0001) (Fig 3C). To ensure that the decreased cell density in Matrigel chambers resulted from chamber invasion, rather than cell death, MTT assays were conducted following LINC03045 lncRNA KD and demonstrated no change in proliferation in LINC03045 KD relative to control sgRNA (sgRNA1: p = 0.5, sgRNA2: p = 0.4) (Fig 3B).
A) qPCR of CRISPRi knockdown (KD) of the top two sgRNAs for LINC03045 in U87 cells. All expression levels were normalized to the housekeeping gene RPLP0. KD cells were compared to the non-targeting sgRNA control. Data are expressed as mean ± SD for replicates. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 B) An MTT assay showing no change in proliferation after KD of LINC03045 by CRISPRi in U87 cells. KD cells were compared to the non-targeting sgRNA control. Data are expressed as mean ± SD. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001. C) Representative images of invasion assay through a Matrigel-coated Boyden chamber after KD of LINC03045 by CRISPRi in U87 cells. The bar graph represents the number of invaded cells per field counted, with KD cells compared to the non-targeting sgRNA control. Data are expressed as mean ± SEM of three independent experiments with 9 fields imaged per experiment. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.
Antisense oligonucleotide mediated KD of LINC03045 in multiple cell lines replicates CRISPRi results
To assess a whether a second KD method for LINC03045 is associated with decreased malignant glioma cell invasion and to utilize a drug-based method with translational potential, antisense oligonucleotides (ASOs) were utilized for LINC03045 KD in three glioma cell lines: a patient-derived tumor line, USC02, and two commercial lines, U87 and U251. Transcriptional expression via qPCR after ASO KD showed decreased transcriptional expression of LINC03045 in USC02s (33.4% decrease, p = 0.04), U87s (43.9% decrease, p = 0.04), and U251s (51.6% decrease, p = 0.02) (Fig 4A). Likewise, evaluation of in vitro invasion via Matrigel assays after ASO-mediated LINC03045 KD showed decreased invasion in USC02s (44.9% decrease, p = 0.005, Fig 4C), U87s (45.8% decrease, p<0.001, Fig 4D), and U251s (58.4% decrease, p = 0.0017, Fig 4E). Correspondingly, 3D invasion assays in U87 spheroids showed reduced invasion into the surrounding matrix after ASO-mediated LINC03045 KD, compared to control untransfected U87 spheroids (ASO 1: 34.2% decrease, p = 0.0193, ASO 2: 37% decrease, p = 0.0158, S4C Fig). MTT assays showed no change in cell proliferation after KD of LINC03045 in USC02s (p = 0.73), U87s (p = 0.15), and U251s (p = 0.51) (Fig 4B).
A) qPCR of antisense oligonucleotide (ASO) KD of LINC03045 in USC02, U87, and U251 cells. All expression levels were normalized to the housekeeping gene RPLP0. KD cells were compared to the non-targeting ASO control. Data are expressed as mean ± SD for replicates. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 B) An MTT assay showing no change in proliferation after KD of LINC03045 by ASO in USC02, U87, and U251 cells. KD cells were compared to the non-targeting ASO control. Data are expressed as mean ± SD. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 C—E) Representative images of invasion assay through a Matrigel-coated Boyden chamber after KD of LINC03045 in USC02 (C), U87 (D), and U251 (E) cells by ASO. The bar graph represents the number of invaded cells per field counted, with KD cells compared to the non-targeting ASO control. Data are expressed as mean ± SEM of three independent experiments with 9 fields imaged per experiment. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.
Genome-wide differential gene expression via TCGA and GTEx analysis
TCGA and GTEx patient data was further explored for the association of LINC03045 with genome-wide gene expression changes. The distribution of LINC03045 expression was first evaluated, noting a clear distinction between the highest and lowest 25 percent of patients (Fig 5A). Subsequently, patients in TCGA-GBM were divided into two groups (high LINC03045 and low LINC03045) to investigate the functional implications of the LINC03045 based on its expression value. Evaluation of the cohort of patients with low expression of LINC03045 revealed significant global alterations in gene expression, as represented in the top 100 and 500 genes most altered in association with low lncRNA expression (Fig 5B and 5C). Widespread up- and down-regulation of genes was seen during upregulation of LINC03045 among patient cohorts as well (Fig 5D). KEGG pathway analysis revealed a significant association between LINC03045 expression and oxidative phosphorylation, as well as ribosomal and spliceosomal activity (Fig 5E). Gene ontology analysis revealed widespread association of RNA transcription, RNA polymerase II activity, RNA splicing, and chromatin organization with increased LINC03045 expression (Fig 5F). Among implicated gene ontology pathways, WASF3 was significantly associated with an enriched pathway involving cytoskeleton modulation (GO:0007010). Further, we evaluated the intersection of differential gene expression profile associated with LINC03045 KD in vitro and LINC03045 expression changes in vivo (Fig 5G). For the TCGA-associated genes, we specifically included protein-coding genes that showed a significant association with LINC03045 (p< = 0.05). Of 36 differentially expressed genes following in vitro linc03045 KD, 16 genes were differentially expressed in the TCGA. Notably, the overlapping genes (e.g. WASF3, EDNRB, SOX10, BAMBI) were found to be involved in several pathways, such as regulation of cell shape and neural crest cell migration (Fig 5 and S1 Table).
A) Distribution of LINC03045 expression utilized to determine high and low LINC03045 expression groups. B) Heat map showing differential expression of the top 100 genes in high vs. low LINC03045 expression groups. C) Heat map showing differential expression of the top 500 genes in high vs. low LINC03045 expression groups. D) Volcano plot showing upregulated and downregulated genes in high vs. low LINC03045 expression groups. E) KEGG pathway analysis. F) Gene ontology analysis. G) Venn diagram illustrating the intersection of differentially expressed genes in our in vitro LINC03045 KD and in vivo LINC03045 TCGA gene expression profiles.
WASF3 is a downstream mediator of LINC03045
To investigate downstream effects of LINC03045 and begin elucidating potential mechanisms mediating the LINC03045 activity, we conducted ASO-mediated KD of LINC03045 in U87 cells, evaluating gene expression via RNAseq. Results from RNAseq analysis of cells with LINC03045 knockdown relative to control are shown in the form of a heat map in Fig 6A and a volcano plot in Fig 6B. Several genes were found to be significantly up or downregulated with LINC03045 knockdown. To further improve the reliability and comparability of detected differential expressed genes, only 4 protein coding genes (ABCC1, DCT, RPS17, TBCCD1, and WASF3) with RPKM values ≥1 in either control or treatment group were identified. These genes were individually validated for gene expression via qPCR following LINC03045 knockdown. In addition, when candidate genes were compared to genes associated with LINC03045 expression in TCGA patient data, we found that WASF3, a gene that has previously been found to be associated with the JAK-STAT pathway, was significantly positively correlated with the expression of LINC03045 (Fig 6C and S1 Table). Finally, the candidates were evaluated for individual knockdown/overexpression and assessment of in vitro invasion via Boyden chamber assay.
A) Heat map showing differential expression of genes with LINC03045 KD (right) compared to control (left). B) Volcano plot showing differential gene expression with LINC03045 KD. C) Analysis of WASF3 correlation with LINC03045.
WASF3 was uniquely noted to be reproducibly decreased upon LINC03045 KD, as well as associated with decreased tumor cell invasion upon WASF3 KD. Furthermore, WASF3 was the only one of these gene candidates noted to be significantly downregulated among patient cohorts with low LINC03045 expression in our analysis above (Fig 6B).
Specifically, upon ASO-mediated LINC03045 KD, we noted a decrease in transcriptional expression of WASF3 in USC02s (25.9% decrease, p = 0.03), U87s (42.7% decrease, p = 0.002), and U251s (57.3% decrease, p = 0.01) (Fig 7A). Furthermore, upon siRNA-mediated KD of WASF3, qPCR confirmed a decrease in WASF3 transcriptional expression in all USC02s (siRNA1: 47.9% decrease, p = 0.004; siRNA2: 87.3% decrease, p = 0.03), U87s (siRNA1: 64.2% decrease, p = 0.002; siRNA2: 85.9% decrease, p = 0.05), and U251s (siRNA1: 74.7% decrease, p = 0.02; siRNA2: 94.2% decrease, p = 0.007) (Fig 7B). In addition, WASF3 KD followed by assessment of cell invasion via Matrigel assays demonstrated a statistically significant decrease in invasion relative to control in USC02s (25.7% decrease, p<0.001, Fig 7C), U87s (16.4% decrease, p<0.001, Fig 7D), and U251s (27.3% decrease, p<0.001, Fig 7E). Additionally, 3D invasion assays demonstrated reduced invasion of U87 spheroids into the surrounding matrix after siRNA-mediated WASF3 KD, compared to control untransfected U87 spheroids (siRNA1: 41.1% decrease, p = 0.0113; siRNA2: 49.4% decrease, p = 0.0071), S4D Fig), Further analysis demonstrated that WASF3 has a statistically significant positive correlation with LINC03045 (p = 0.016) (Fig 6C).
A) qPCR of WASF3 after antisense oligonucleotide (ASO) KD of LINC03045 in USC02, U87, and U251 cells. All expression levels are normalized to the housekeeping gene RPLP0. KD cells were compared to the non-targeting ASO control. Data are expressed as mean ± SD for replicates. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 B) qPCR of dicer-substrate short interfering RNA (siRNA) KD of WASF3 in USC02, U87, and U251 cells. All expression levels are normalized to the housekeeping gene RPLP0. KD cells were compared to the non-targeting siRNA control. Data are expressed as mean ± SD for replicates. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 C–E) Representative images of invasion assay through a Matrigel-coated Boyden chamber after KD of WASF3 in USC02 (C), U87 (D), or U251 (E) cells by siRNA. The bar graph represents the number of invaded cells per field counted, with KD cells compared to the non-targeting siRNA. Data are expressed as mean ± SEM of three independent experiments with 9 fields imaged per experiment. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001.
To further reinforce the relationship between LINC03045 and WASF3, we also investigated if WASF3 overexpression could rescue invasive capacity after LINC03045 KD in the patient-derived cell line, USC02. Upon ASO-mediated LINC03045 KD, we once again noted a transcriptional decrease in LINC03045 (82.3% decrease, p = 0.001) and WASF3 (68.8% decrease, p = 0.02) (Fig 8A). Concurrent ASO-mediated LINC03045 KD and WASF3 overexpression demonstrated a transcriptional decrease in LINC03045 (51.9% decrease, p = 0.001) as well as a dramatic transcriptional increase of WASF3 (14,816.4% increase, p = 0.008) (Fig 8A). Assessment of cell invasion via Matrigel assays reproducibly demonstrated the statistically significant decrease in invasion relative to control for ASO-mediated LINC03045 KD only (54.8% decrease, p = 0.002), and when “rescuing” LINC03045 KD tumor cells with concurrent WASF3 overexpression, cells demonstrated a statistically significant increase in invasive capacity compared to LINC03045 KD only cells (234.5% increase, p = 0.002) (Fig 8B and 8C). However, when compared to WASF3 overexpression only, the rescue cells showed no statistically significant difference in invasion (30.8% decrease, p = 0.08) (Fig 8B and 8C).
A) qPCR of LINC03045 and WASF3 after antisense oligonucleotide (ASO) KD of LINC03045 and WASF3 overexpression in USC02 cells. All expression levels are normalized to the housekeeping gene RPLP0. KD cells were compared to the non-targeting ASO control. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 B) Bar graph represents the number of invaded cells per field, with KD cells compared to the non-targeting ASO and concurrent KD and WASF3 overexpression cells compared to WASF3 overexpression only. Data are expressed as mean ± SEM of three independent experiments with 9 fields imaged per experiment. **p ≤ 0.01 C) Representative images of invasion assay through a Matrigel-coated Boyden chamber after ASO KD of LINC03045 and WASF3 overexpression in USC02 cells.
Discussion
To efficiently identify novel lncRNAs in GBM that affect the invasive ability of malignant cells and determine possible therapeutic targets, we developed a CRISPRi-based invasion screen. Here we report the first large-scale CRISPRi screen on lncRNAs implicated in glioblastoma invasion. Our results identified LINC03045 as significantly associated with in vitro glioma cell invasion across 3 tumor lines. Multiple methods of lncRNA knockdown further validated the functional role of LINC03045 via CRISPRi-based and ASO-based knockdown in all three lines. Patient gene expression analysis of TCGA, and GTEx raw transcriptional data revealed that LINC03045 expression was associated with tumor grade and patient survival, as well as widespread changes in global gene expression data, including oxidative phosphorylation and RNA processing. Finally, gene expression analysis following LINC03045 KD revealed that WASF3, a factor evaluated for tumor cell invasion in multiple studies, could be a possible mediator of LINC03045 function [30,31]. Mechanistic studies supported the role of WASF3 as a lncRNA mediator, with WASF3 KD paralleling invasion loss seen with LINC03045 KD in both 2D Matrigel and 3-dimensional invasion modles. Finally, WASF3 rescue experiments confirmed that WASF3 overexpression restores loss-of-invasion from lncRNA KD.
The CRISPRi invasion screen of 2,307 loci identified 45 hits that affect GBM invasion. Upon evaluation of these candidate hits, LINC03045, a previously unannotated and poorly studied lncRNA, was noted to be significantly associated with invasion. Functional genomic methods such as CRISPR (and CRISPRi) screens have gained significant popularity in recent years due to the ability demonstrated here to quickly and efficiently identify potential therapeutic targets [14,19]. While prior manuscripts have notably focused on the role of coding genes affecting glioma invasion [32], the body of literature characterizing the role of lncRNAs in GBM pathogenesis has only just begun to grow in recent years and have primarily focused on the role of lncRNAs in GBM proliferation and progression [33–37]. Prior studies on lncRNAs in GBM invasion, however, remain scarce, and our study uniquely addresses the role of lncRNAs specifically in GBM invasion, suggesting that noncoding RNAs play a significant functional role in this phenotype. Knockdown of our candidate lncRNA via multiple methods, including via both CRISPRi and ASO, revealed similar decreases in tumor invasion while not affecting tumor proliferation. While our knockdown models demonstrated 40–70% decrease in lncRNA expression, we did notice relatively dramatic phenotypic changes in invasion of over 80% in the Matrigel invasion after a knockdown of less than 40%. Part of the observed effect is likely variability. However, essential genes in certain processes have been previously noted to show exaggerated phenotypes with small changes in gene expression, which could play a role in the observed changes in our results [38]. In addition, because lncRNAs have been commonly characterized as extremely abundant and specific to a small subset of cells [39], another possible explanation is that gene expression was particularly abundant in an aggressive, invading, lncRNA-rich subpopulation is capable of demonstrating invasion at 24 hours during the assay. Under this assumption, it is possible a more modest 40% knockdown of the highest expressing cells can more significantly affect the 24 hour invasion phenotype.
LncRNAs represent challenging experimental and therapeutic targets. CRISPR/Cas9 or CRISPRi-based drugs carry inherent clinical challenges of non cell-specific effects for treating diseases. ASOs are single-stranded nucleotide strands that are taken up freely by cells to form heteroduplexes with the complementary target RNA, designating them for degradation, and can offer a potential treatment option. since they are single-stranded nucleotide strands that are taken up freely by cells to form heteroduplexes with the complementary RNA. Clinically, ASO-mediated therapies are currently approved and being utilized for treating diseases like spinal muscular atrophy [40]. Thus, we also successfully utilized ASOs as a confirmatory second knockdown method and therapeutic medium to knock down LINC03045 and evaluate subsequent effects.
Notably, our experimental data was also confirmed to be clinically relevant, as TCGA and GTEx analysis showed that LINC03045 expression is higher in GBM (grade 4) than in LGG (grades 2–3), and higher in LGG when compared to normal cortical tissue. In addition, Kaplan-Meier survival curves for all gliomas (LGG and GBM) showed patients with higher LINC03045 expression were associated with decreased survival. After adjusting for grade and IDH-status, higher LINC03045 expression was significantly associated with lower survival only in the LGG cohort that included grades 2 and 3. The survival analysis for this LGG cohort included both grades 2 and 3 because the TCGA defines LGG as these two grades, and a portion of the LGG dataset did not further distinguish grade between them [41]. Notably, grade 2 and 3 gliomas within this cohort did not significantly differ in linc03045 expression.
However, we also conducted analysis of Grade 2 gliomas IDH-mut tumors alone when excluding the 20 patients with uncertain grade 2/3 pathology. Notably, in this cohort, the sample size was significantly reduced (possibly underpowered), though the trends in the survival curves remained similar between Grade 2 IDH-mut curves and Grade 2/3 curves. However, LINC03045 expression was not associated with survival in this cohort (p = 0.18, Fig 2).[42]. Further, LINC03045 also did not appear to be significantly associated with survival when strictly in the GBM cohort (stage 4) with adjustment for IDH-status. Notably, it appears that in grade 4 gliomas, the vast majority of samples express linc03045 >1.46, which is greater than the 75th percentile of linc03045 expression in all other tumor grades. It may be that above a significant threshold of gene effect, or with a the more limited survival of grade 4 tumors, the gene expression in Grade 4 tumors is less significant to survival. Nonetheless, the enhanced expression of LINC03045 in GBM in comparison to normal brain cortex and LGG suggests that it has an association with GBM aggressiveness and higher grade. Prior literature has shown that lncRNA expression has been associated with GBM outcome in multiple studies and has been explored in other lncRNAs as a biomarker for prognosis [11,43]. Thus, further study of these findings may reveal the role of LINC03045 expression in resected patient tumors as a prognostic biomarker.
While lncRNA-based prognostic and diagnostic markers and lncRNA-based therapeutics described above are still in their experimental stages, the study of lncRNAs often offers insight into mechanisms regulating tumor phenotypes such as, in our study, invasion. When evaluating the intersection between differential gene expression in LINC03045 KD in vitro and LINC03045 expression changes in vivo (in the TCGA), we noted an overlap of 16 out of the 36 differentially expressed genes in vitro, with significant alterations in GO pathways in cell shape and neural crest cell migration (S1 Table). In prior literature, implicated overlapping genes, including WASF3, EDNRB, SOX10, BAMBI have been implicated in processes including tumor progression, migration, and invasion [44,45]. Of particular interest in our study was WASF3. Here, our RNA-seq analysis following LINC03045 KD revealed that WASF3 is a gene that was reproducibly downregulated with LINC03045 knockdown, while a correlation analysis of patient TCGA/GTEx sequencing data revealed a significant positive correlation between LINC03045 and WASF3. Additionally, the GO term for cytoskeleton organization (GO: 0007010) process was enriched for WASF3, which is consistent with prior literature that describes cytoskeletal reorganization as commonly associated with tumor invasion [46]. WASF3, a Wiskott-Aldrich syndrome protein family member, has been implicated in multiple studies as a regulator of the actin cytoskeleton and in cancer invasion, which is again consistent with our results [47]. Studies have shown that it is a scaffolding protein that is essential in coordinating metastatic signaling complexes, promoting invasion and metastasis through activation of leading edge membrane structures. WASF3 has been shown to be associated with invasion in several cancers, including breast and ovarian cancer, and is inversely correlated with overall progression-free survival [48]. Notably, WASF3 is a critical component of the JAK-STAT3 pathway. STAT3 promotes WASF3 expression while JAK2 independently activates it, and this process is necessary for invasion [30,31]. WASF3 overexpression also activates NFkB and ZEB1 and also promotes invasion by regulating genes involved in metastasis. While we have noted that WASF3 could be a mediator of LINC03045 function, lncRNAs typically regulate transcriptional activity of target genes by acting as scaffold or cofactor for transcription factors. Future studies are needed to characterize the mechanistic role by which LINC03045 regulates WASF3 gene expression in the context of its known role in the JAK-STAT3 pathway.
LINC03045 knockdown was futher associated with widespread gene expression changes potentially affecting tumor invasion, as noted in KEGG pathway analysis. Pathway analysis found that genes involved in oxidative phosphorylation pathways were most strongly associated with LINC03045. Prior studies have found that aberrations in glycolysis and oxidative phosphorylation could contribute to GBM aggressiveness and invasion and that an acidic and hypoxic tumor microenvironment can promote tumor invasion by changing glycolytic flux, thus promoting local protumorigenic inflammatory responses [49].
Other notable processes identified to be associated with LINC03045 from KEGG analysis were ribosomal processes and spliceosomal processes. A prior study found that ribosomal proteins are associated with enzymes that promote proliferation and invasion of GBM cells as well as apoptotic processes [50,51]. Another previous study found that spliceosome assembly and alternative splicing could contribute to GBM invasion and the GBM mesenchymal phenotype [52].
Our study does carry limitations. Although CRISPR/Cas9 is a robust tool to target lncRNAs, major limitations include risk of affecting adjacent genes, including those that overlap with the lncRNA gene or elements in the lncRNA locus that may regulate other genes [33]. Notably, our gene expression analysis following LINC03045 KD in patient cell lines revealed that no coding genes, with the closest one residing over 1kB from LINC03045, were significantly changed following lncRNA KD. Another limitation lies in the challenge of screening a complex phenotype such as invasion. We attempted to circumvent this limitation by performing screens over the course of 24 hours, a time period during which pilot studies (and prior studies [14,19,32]) showed adequate invasion. Furthermore, our MTT assays suggested no significant alterations in tumor proliferation during this screening time period. In addition, our study initially screened CRISPRi lncRNA knockdowns within a commercial tumor cell line, U87. The 24-hour screen of a patient-derived cell line proved challenging due to difficulties achieving stable transfection of both the dCas9 and sgRNA plasmids. While we demonstrated consistent phenotype and downstream changes in multiple cell lines, including a recently derived patient tumor line, and included a 3D invasion assay as well, further comprehensive screens may include screening of more patient lines. Additionally, given the technical challenges of conducting long-term lentiviral knockdown of lncRNAs in vivo, including target specificity and the instability of lentiviral transduction of the LINC03045 knockdown, we have chosen to utilize 3D invasion assays as an approximation of in vivo behavior, which allowed for stable, continuous knockdown of LINC03045 via ASO [53–55]. However, a true in vivo knockdown study of LINC03045 is needed in further studies to corroborate our results, since in vivo models are still the most accurate method for studying tumor behavior and environmental interactions [56]. Furthermore, though LINC03045 appears to modulate WASF3 to impact GBM invasion, we did not perform formal mechanistic studies involving co-localization or RNA pulldown. Potential hypotheses for a possible mechanism include functioning as a microRNA sponge, lncRNA interaction with transcriptional regulators, or functioning as a guide or scaffold for transcription factors of either WASF3 or its upstream regulators. However, the transcriptional regulation of WASF3 still remains relatively understudied, and elucidating the mechanism behind these relationships would require extensive further research.
After utilizing a large-scale functional CRISPRi invasion screen, LINC03045 was identified to regulate in vitro invasion of multiple GBM cell lines. Subsequent evaluation of patient clinical data reveals significant associations between LINC03045 and tumor grade, as well as patient survival. Analysis of patient data, tumor cell lncRNA knockdown, and WASF3 knockdown reveal that WASF3 mediates the LINC03045 tumor effect. While future studies must characterize the specific mechanism by which LINC03045 modulates WASF3 expression, as well as the role of LINC03045 using in vivo models, our results show that LINC03045 has an important role in GBM invasion.
Methods
Cell culture
U87 cells (ATCC, HTB-14; RRID:CVCL_0022), HEK293T cells (ATCC, CRL-3216; RRID:CVCL_0063), and U251 cells (Sigma-Aldrich, 09063001-1VL; RRID:CVCL_0021) were maintained in D10 medium, which contained high glucose Dulbecco’s modified Eagle’s medium (DMEM) (Gibco, 11995–081) supplemented with 10% fetal bovine serum (FBS) (Seradigm, 97068–085) and 1X antibiotic-antimycotic (Gibco, 15240–062). USC02 cells, a primary glioma stem cell (GSC) line, were a gift from Thomas C. Chen [29] and maintained in cancer stem cell (CSC) medium containing DMEM/F-12 (Gibco, 11320–033) supplemented with 1% B-27 (Life Technologies, 12587–010), 1X antibiotic-antimycotic, and 20 ng/mL epidermal growth factor (EGF) (Peprotech, AF-100-15) and basic fibroblast growth factor (bFGF) (Peprotech, 100-18B). All cells were kept in a humidified incubator at 37°C with 5% CO2 and confirmed to be mycoplasma free with the LookOut Mycoplasma PCR Detection Kit (Sigma-Aldrich, MP0035) immediately prior to a large-scale screen and at least once every two months.
CRISPRi screens
U87 cells expressing dCas9-KRAB-BFP were generated by lentiviral infection. pHR-SFFV-dCas9-BFP-KRAB was a gift from Stanley Qi & Jonathan Weissman (Addgene plasmid #46911; http://n2t.net/addgene:46911; RRID:Addgene_46911) [57]. Fluorescence associated cell sorting (FACS) was used 48 hours post-transduction to isolate cells stably expressing BFP (U87+dCas9-KRAB).
The U87-specific CRISPR inhibition (CRISPRi) library, derived from the human CRISPRi non-coding library (CRinCL) (Addgene #86542, a gift from Jonathan Weissman) by selecting sub-libraries that targeted all expressed lncRNAs in U87 [19], was used to generate the sgRNA pool used for screening. TransIT-Lenti (Mirus, 6600) was used to package the plasmids for transfection in HEK293Ts and viral production was allowed to occur for 48 hours before harvesting. Filtered viral supernatant was then added to U87+dCas9-KRAB cells with polybrene (Sigma-Aldrich, TR-1003-G) at a final concentration of 4 μg/mL at an MOI < 0.3. Two days after infection, cells were treated with 0.75 μg/mL puromycin (Gibco, A1113803) for 3 days, allowed to recover for one day without puromycin, and used for assays the next day (T0), with cells being maintained at a coverage of 1000X. Two independent screen replicates were conducted, with cells collected at T0 and at the endpoint, and processed for sequencing on Illumina HiSeq 4000 and analyzed as previously described [19]. A scatterplot of invasion phenotype of sgRNA of two independent replicates was generated, with the x and y-axes representing different screen replicates (S2 Fig). Analysis output denoted p-value and effect sizes for lncRNAs targeted in the screen. A specific screen score, incorporating these values, was calculated via x -log10 p value relative to negative internal control sgRNA targets as previously described [19,22]. These were plotted via volcano plot for selection of highest screen scores.
Cell invasion assay
Boyden chambers comprised of polyethylene terephthalate with 8 μm pore size and coated in a layer of Matrigel (Corning, 354481) were used to test invasion of cells. U87 screen cells, prepared as described above, were seeded at a density of 5 x 105 cells. For individual knockdown validation, U87 cells (5 x 105), U251 cells (1.25 x 105), or USC02 cells (2 x 105) were seeded in the top chamber with serum-free medium while the lower chamber contained D10 medium as the chemoattractant. Chambers were then incubated for 24 hours in a humidified incubator at 37°C with 5% CO2. For deep sequencing following screen, cells were collected from the top chamber by trypsinization.
To count cells that had invaded when evaluating individual knockdown in validation studies, a cotton swab was used to remove cells remaining in the upper chamber and cells on the underside were fixed and stained with Hemacolor Stain Set (Sigma-Aldrich, 65044). Photos were taken on a Keyence, BZ-X800, with 9 fields per chamber, and cells counted with ImageJ Software. Experiments were performed in triplicates and independent t-test used to evaluate statistical significance.
3D spheroid invasion assay
Cultrex spheroid invasion assay (3500-096-K) was used to model 3D invasion of gliomaspheres into surrounding invasion matrix. Control (untransfected) U87) cells, and U87 cells transfected with LINC03045 ASO or WASF3 siRNA were plated in 50ul of 1X spheroid-forming matrix diluted in serum-free DMEM/F12 medium in a 96-well round-bottom ultra-low attachment plate on ice. Plated cells were immediately centrifuged at 200g for 3 min at room temperature to concentrate the cells in the center of the welland enhance gliomasphere formation. After incubation at 37°C for 72h, the spheroid-containing plate was equilibriated on ice for 15 mins, followed by addition of 50ul 1X invasion matrix per well, and the plate was centrifuged at 300g for 5min at 4°C. After incubation for 1 hour at 37°C, 100ul of chemoattractant medium (10%FBS final concentration) was added to each well, and cells were allowed to invade into the surrounding matrix. Serum-mediated invasion was monitored over 48 hours, and wells were imaged at 2x magnification on the Keyence, BZ-X800. Invasion area was manually outlined, and measured using the ImageJ Software. Invasion from ASO and siRNA-transfected spheroids was compared to invasion from untransfected control U87 spheroids. Experiments were performed in triplicates, and statistical significance between groups was evaluated using independent t-tests.
Validation of screen candidates via knockdown
To clone the sgRNAs for individual validation, oligo pairs containing the sgRNA protospacer and adjoining BstXI and BlpI cloning sites were annealed and ligated into the sgRNA plasmid backbone pU6-sgRNA EF1Alpha-puro-T2A-BFP (Addgene plasmid #60955; http://n2t.net/addgene:60955; RRID:Addgene_60955, a gift from Jonathan Weissman) [22]. A non-targeting control (GAL4) was also subcloned into the aforementioned backbone. The final sgRNA sequences used were: GAL4 5’-GAACGACTAGTTAGGCGTGTA-3’; LINC03045 sgRNA 1 5’-GTTTAACCAACGGGTTATTT-3’; LINC03045 sgRNA 2 5’-GTTTGTTTTATTAGTAAAAG-3’.
RT-qPCR
TRIzol Reagent (Invitrogen, 15596–018) or the RNeasy Plus Mini Kit (Qiagen, 74134) was used to extract total RNA 24 or 48 hours following ASO or WASF3 overexpression plasmid (WASF3 NM_006646) Human Tagged ORF Clone (Origene, RG216156) transfection, or at T0 for CRISPRi. Genomic DNA removal and cDNA synthesis were completed in the same tube with the iScript gDNA Clear cDNA Synthesis Kit (Bio-Rad, 1725034). qPCR was performed with the iTaq Universal SYBR Green Supermix (Bio-Rad, 1725121) on a Roche LightCycler 96 instrument. Experiments were performed in triplicates and independent t-test used to evaluate statistical significance. Primers used for qPCR were: RPLP0 forward 5’- CATATCCGGGGGAATGTGGG-3’; RPLP0 reverse 5’- AGCAGCTGGCACCTTATTGG-3’; LINC03045 forward 5’- CCCATCCAAGAAGTTTGCAG-3’; LINC03045 reverse 5’- GCCACTGTCTTTCTGCTTCC-3’; WASF3 forward 5’-TCAACAGTGGAAGAGGTCTCAC-3’; and WASF3 reverse 5’- TTCAGAGGCGGTGGCTTATC-3’.
Cell proliferation assay
3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assays were used to measure cell proliferation during the time points cells would be undergoing invasion assays. Cells (USC02, U87, and U251) were seeded in 96-well plates (5 x 103 cells/well) in 100 μL of complete media and the next day incubated with 10 μL of 5 mg/mL tetrazolium dye (Invitrogen, M6494) at 37°C with 5% CO2 for 2 hours. The solution was then removed, and the cells incubated for 10 minutes at room temperature with 100 μL of dimethyl sulfoxide (Sigma-Aldrich, D8418). The optical density values at 570 nm and 650 nm were detected with a Thermo Scientific Varioskan LUX microplate reader. Experiments were performed in triplicates and independent t-test used to evaluate statistical significance.
Antisense oligonucleotides
2’-O-methoxyethyl (2’-MOE) modified antisense oligonucleotide (ASO) gapmers were designed against linc03045 by Integrated DNA Technologies. ASOs were transfected at a final concentration of 25 nm with TransIT-siQUEST Transfection Reagent (Mirus, MIR 2110) or the TransIT-X2 Dynamic Delivery System (Mirus, MIR 6003) following the manufacturer’s protocol. U87 and U251 cells were transfected with a either a non-targeting ASO or a linc03045 ASO for 24 hours while USC02 cells were transfected for 48 hours. ASO target sequences were: non-targeting 5’-/52MOErG/*/i2MOErC/*/i2MOErG/*/i2MOErA/*/i2MOErC/*T*A*T*A*C*G*C*G*C*A*/i2MOErA/*/i2MOErT/*/i2MOErA/*/i2MOErT/*/32MOErG/-3’; LINC03045 ASO 1 5’-/52MOErA/*/i2MOErT/*/i2MOErC/*/i2MOErG/*/i2MOErT/*T*T*C*T*C*G*G*C*C*T*/i2MOErT/*/i2MOErT/*/i2MOErT/*/i2MOErG/*/32MOErG/-3’; and LINC03045 ASO 2 5’-/52MOErT/*/i2MOErA/*/i2MOErG/*/i2MOErG/*/i2MOErC/*A*C*A*A*G*C*C*A*C*C*/i2MOErG/*/i2MOErT/*/i2MOErT/*/i2MOErT/*/32MOErA/-3’.
RNA-seq sample preparation and data analysis
U87 cells were transfected with either a non-targeting or LINC03045 ASO for 24 hours as described above, spun down into a pellet, and flash frozen. Cell pellets underwent RNA extraction, library generation by Poly(A) selection, and sequencing on an Illumina HiSeq with the 2x150 bp configuration (GENEWIZ, Azenta Life Sciences).
Pair-end raw sequencing reads were processed as follows: Firstly, we adapted Trim Galore [58] to remove adaptors and low-quality reads. Cleaned reads with length greater than 20 were mapped to the human hg19 reference genome by using tophat [59] with default parameters. To quantify the expression of LINC03045, GENCODE [60] release 19 was downloaded from the ensemble database with subtle modifications. We added the coordinates of LINC03045 to the GENCODE annotation database. Then, featureCount [61] was employed to count reads in individual genes. A matrix containing the read counts was subjected to the DEseq2 [62] package for differential expression genes analysis. In addition, Reads Per Kilobase Million (RPKM) was used to quantify the expression level of each genes. Genes with an absolute value of log2 fold change greater than 1 and a Benjanmini-Hochberge adjusted p-value less than 0.05 were considered differentially expressed for further validation.
Dicer-substrate short interfering RNA
Dicer-substrate short interfering RNAs (siRNAs) were designed against WASF3 by Integrated DNA Technologies. siRNAs were transfected at a final concentration of 10 nm with TransIT-siQUEST Transfection Reagent following the manufacturer’s protocol. U87, U251, and USC02 cells were transfected with either a non-targeting ssiRNA or a WASF3-targeting siRNA.
TCGA and GTEx analysis
RNA-seq alignment data (BAM files aligned to GRCh38) of patient samples and normal brain cortical tissue were acquired through the Database of Genotypes and Phenotypes (dbGaP), specifically Genotype-Tissue Expression (GTEx) for normal tissue (n = 104) and The Cancer Genome Atlas (TCGA) for GBM (n = 169) and lower grade glioma (LGG) (n = 169). The Samtools package was used to filter the BAM files by flag status to retain read alignments that mapped in proper pairs, after which expression read counts were tabulated for the positional coordinates of LINC03045 (chr10:92658250–92661763). The expression counts were normalized to total read alignments on chromosome 10 to obtain counts per million (CPM) values. Gene expression distribution shifts amongst the three cohorts were observed using histograms. DESeq2 was used to calculate to calculate the significance of LINC03045 differential expression between the glioma datasets and control, against a backdrop distribution of all other chromosome 10 genes. Information on WHO grade and molecular subtype (e.g. IDH status) of the GBM and LGG cohorts from TCGA were obtained from a prior study[63].
Kaplan-Meier (KM) survival curve and log-rank test were performed for all gliomas (LGG and GBM), LGG and GBM. Cox regression was conducted using expression levels as a continuous variable. Variables of IDH status, MGMT methylation, and 1p/19q codeletion, when available, were utilized in multivariate Cox regression analysis. None of the variables were acting as confounders for the gene expression and survival. To avoid overadjustment we present results using gene expression alone. Linearity assumption and proportional hazard assumption was evaluated using Martingale residuals. For LINC03045, linearity was violated, and the model was built using two linear splines with the knot at 2.5 CPM. The knot position was tested by modeling 2.0 to 3.0 by 0.1 increment. 2.5 CPM was chosen as the lowest Akaike information criterion (AIC) value. Significance level was 0.05, 2-sided. All analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, NC, USA.).
For GBM patiens RNA-seq data from TCGA, raw reads were processed as previously described in RNA-seq analysis section, with modifications. Briefly, reads were trimmed using Trim Gare [58] and then mapped to the hg19 reference genome with STAR [64]. The gene expression was quantified in TPM at gene level using GENCODE [60] version 19 gene annotation, including the postion of LINC03045. To obtain the genes that are highly correlated with LINC03045, the Pearson correlation coefficients were calculated between the target lncRNA and any other genes. A threshold of p-value less than 0.05 was reported as remarked genes. Significant positively correlated genes (correlation coefficient >0) and negatively correlated genes (correlation coefficient <0) were submitted to the online DAVID [65] database for Gene Ontology (GO) enrichment analysis, repectively. In addition, patients were divided into high and low groups, according to the LINC03045 expression value at the upper 25 percent and lower 25 percent. These two groups were loaded into the Limma [66] package for differential expressed genes detection. Differential expression analyses (DEGs) were also carried out for enriched GO analysis. Gene Set Enrichment Analysis (GSEA) was performed to enrich pathways using the clusterProfiler [67] package for statistical analysis and visualization.
Supporting information
S1 Fig. Annotated volcano plot of invasion phenotype and negative log10(Mann-Whitney p-value).
Screen replicates were averaged and the top 3 sgRNAs for each lncRNA compared to non-targeting controls were used to determine screen hits. The dashed lines represent thresholds to determine screen hits.
https://doi.org/10.1371/journal.pgen.1011314.s001
(TIF)
S2 Fig. Scatterplot of invasion phenotype of sgRNA of two independent replicates.
A positive phenotype indicates knockdown of a lncRNA caused a decrease in invasion, whil a negative phenotype indicates knockdown of a lncRNA caused an increase in invasion. The X and Y-axes represent different screen replicates.
https://doi.org/10.1371/journal.pgen.1011314.s002
(TIFF)
S3 Fig. Preliminary results of the ZNF215-AS1 candidate.
A) qPCR of CRISPRi KD of the top two sgRNAs for ZNF215-AS1 in U87 cells. All expression levels normalized to the housekeeping gene RPLP0. KD cells compared to the non-targeting sgRNA. Data expressed as mean ± SD for replicates. B) Representative images of invasion assay through a Matrigel-coated Boyden chamber after KD of ZNF215-AS1 by CRISPRi KD in U87 cells. The bar graph represents the number of invaded cells per field counted, with KD cells compared to the non-targeting sgRNA. Data are expressed as mean ± SEM of three independent experiments with 9 fields imaged per experiment. C) An MTT assay showing no change in proliferation after KD of ZNF215-AS1 or LH02236 by CRISPRi in U87 cells. KD cells compared to the non-targeting sgRNA. Data are expressed as mean ± SD. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 D) Boxplots of pairwise comparisons of ZNF215-AS1 expression in normal human brain cortex to low grade glioma (LGG) and glioblastoma (GBM) patient samples. E) Kaplan-Meier survival curves of ZNF215-AS1 for gliomas (stages 1–4) with expression quantified by normalized counts per million (CPM). Position coordinates were used to identify both genes in RNA-seq data. Expression was normalized to standardized genes on the same chromosome. Censored data indicates patients lost to follow-up. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001
https://doi.org/10.1371/journal.pgen.1011314.s003
(TIFF)
S4 Fig.
qPCR confirmation of LINC03045 KD (A) and WASF3 KD (B) in U87 cells relative to control untransfected U87 cells, after transfection with LINC03045 ASO or WASF3 siRNA respectively. All data were normalized to RPLPO housekeeping gene. Data are expressed as mean ± SEM for replicates. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001. (C) Quantification and representative images of U87 spheroid invasion into surrounding invasion matrix after ASO-mediated knockdown of LINC03045 compared to control untransfected U87 spheroids. Data are expressed as mean ± SEM for replicates. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001. (D) Quantification and representative images of U87 spheroid invasion into surrounding invasion matrix after siRNA-mediated knockdown of WASF3 compared to control untransfected U87 spheroids. Data are expressed as mean ± SEM for replicates. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001. Spheroid images taken at 2x magnification.
https://doi.org/10.1371/journal.pgen.1011314.s004
(TIF)
S1 Table. Gene Ontology analysis of the relationship between WASF3 and enriched pathways in in vivo gene expression.
https://doi.org/10.1371/journal.pgen.1011314.s005
(TIFF)
References
- 1. Omuro A, DeAngelis LM. Glioblastoma and other malignant gliomas: a clinical review. JAMA. 2013;310(17):1842–50. pmid:24193082
- 2. Fabian D, Guillermo Prieto Eibl MDP, Alnahhas I, Sebastian N, Giglio P, Puduvalli V, et al. Treatment of Glioblastoma (GBM) with the Addition of Tumor-Treating Fields (TTF): A Review. Cancers (Basel). 2019;11(2). pmid:30717372
- 3. Davis ME. Glioblastoma: Overview of Disease and Treatment. Clin J Oncol Nurs. 2016;20(5 Suppl):S2–8. pmid:27668386
- 4. Tykocki T, Eltayeb M. Ten-year survival in glioblastoma. A systematic review. J Clin Neurosci. 2018;54:7–13. pmid:29801989
- 5. Thakkar JP, Dolecek TA, Horbinski C, Ostrom QT, Lightner DD, Barnholtz-Sloan JS, et al. Epidemiologic and molecular prognostic review of glioblastoma. Cancer Epidemiol Biomarkers Prev. 2014;23(10):1985–96. pmid:25053711
- 6. Sahm F, Capper D, Jeibmann A, Habel A, Paulus W, Troost D, et al. Addressing diffuse glioma as a systemic brain disease with single-cell analysis. Arch Neurol. 2012;69(4):523–6. pmid:22158715
- 7. Debus J, Abdollahi A. For the next trick: new discoveries in radiobiology applied to glioblastoma. Am Soc Clin Oncol Educ Book. 2014:e95–9. pmid:24857153
- 8. Tang Z, Dokic I, Knoll M, Ciamarone F, Schwager C, Klein C, et al. Radioresistance and Transcriptional Reprograming of Invasive Glioblastoma Cells. Int J Radiat Oncol Biol Phys. 2022;112(2):499–513. pmid:34534627
- 9. Zhang X, Sun S, Pu JK, Tsang AC, Lee D, Man VO, et al. Long non-coding RNA expression profiles predict clinical phenotypes in glioma. Neurobiol Dis. 2012;48(1):1–8. pmid:22709987
- 10. Mercer TR, Dinger ME, Mattick JS. Long non-coding RNAs: insights into functions. Nat Rev Genet. 2009;10(3):155–9. pmid:19188922
- 11. Peng Z, Liu C, Wu M. New insights into long noncoding RNAs and their roles in glioma. Mol Cancer. 2018;17(1):61. pmid:29458374
- 12. Han L, Zhang K, Shi Z, Zhang J, Zhu J, Zhu S, et al. LncRNA pro fi le of glioblastoma reveals the potential role of lncRNAs in contributing to glioblastoma pathogenesis. Int J Oncol. 2012;40(6):2004–12.
- 13. Paul Y, Thomas S, Patil V, Kumar N, Mondal B, Hegde AS, et al. Genetic landscape of long noncoding RNA (lncRNAs) in glioblastoma: identification of complex lncRNA regulatory networks and clinically relevant lncRNAs in glioblastoma. Oncotarget. 2018;9(51):29548–64. pmid:30038703
- 14. Liu SJ, Malatesta M, Lien BV, Saha P, Thombare SS, Hong SJ, et al. CRISPRi-based radiation modifier screen identifies long non-coding RNA therapeutic targets in glioma. Genome Biol. 2020;21(1):83. pmid:32234056
- 15. Gupta RA, Shah N, Wang KC, Kim J, Horlings HM, Wong DJ, et al. Long non-coding RNA HOTAIR reprograms chromatin state to promote cancer metastasis. Nature. 2010;464(7291):1071–6. pmid:20393566
- 16. Ji P, Diederichs S, Wang W, Boing S, Metzger R, Schneider PM, et al. MALAT-1, a novel noncoding RNA, and thymosin beta4 predict metastasis and survival in early-stage non-small cell lung cancer. Oncogene. 2003;22(39):8031–41. pmid:12970751
- 17. Rinn JL, Chang HY. Genome regulation by long noncoding RNAs. Annu Rev Biochem. 2012;81:145–66. pmid:22663078
- 18. Schmitt AM, Chang HY. Long Noncoding RNAs in Cancer Pathways. Cancer Cell. 2016;29(4):452–63. pmid:27070700
- 19. Liu SJ, Horlbeck MA, Cho SW, Birk HS, Malatesta M, He D, et al. CRISPRi-based genome-scale identification of functional long noncoding RNA loci in human cells. Science. 2017;355(6320). pmid:27980086
- 20. Wang T, Wei JJ, Sabatini DM, Lander ES. Genetic screens in human cells using the CRISPR-Cas9 system. Science. 2014;343(6166):80–4. pmid:24336569
- 21. Behan FM, Iorio F, Picco G, Goncalves E, Beaver CM, Migliardi G, et al. Prioritization of cancer therapeutic targets using CRISPR-Cas9 screens. Nature. 2019;568(7753):511–6. pmid:30971826
- 22. Gilbert LA, Horlbeck MA, Adamson B, Villalta JE, Chen Y, Whitehead EH, et al. Genome-Scale CRISPR-Mediated Control of Gene Repression and Activation. Cell. 2014;159(3):647–61. pmid:25307932
- 23. Shalem O, Sanjana NE, Hartenian E, Shi X, Scott DA, Mikkelson T, et al. Genome-scale CRISPR-Cas9 knockout screening in human cells. Science. 2014;343(6166):84–7. pmid:24336571
- 24. Konermann S, Brigham MD, Trevino AE, Joung J, Abudayyeh OO, Barcena C, et al. Genome-scale transcriptional activation by an engineered CRISPR-Cas9 complex. Nature. 2015;517(7536):583–8. pmid:25494202
- 25. Yeo NC, Chavez A, Lance-Byrne A, Chan Y, Menn D, Milanova D, et al. An enhanced CRISPR repressor for targeted mammalian gene regulation. Nat Methods. 2018;15(8):611–6. pmid:30013045
- 26. Hart T, Chandrashekhar M, Aregger M, Steinhart Z, Brown KR, MacLeod G, et al. High-Resolution CRISPR Screens Reveal Fitness Genes and Genotype-Specific Cancer Liabilities. Cell. 2015;163(6):1515–26. pmid:26627737
- 27. Shi J, Wang E, Milazzo JP, Wang Z, Kinney JB, Vakoc CR. Discovery of cancer drug targets by CRISPR-Cas9 screening of protein domains. Nat Biotechnol. 2015;33(6):661–7. pmid:25961408
- 28. Joung J, Konermann S, Gootenberg JS, Abudayyeh OO, Platt RJ, Brigham MD, et al. Genome-scale CRISPR-Cas9 knockout and transcriptional activation screening. Nat Protoc. 2017;12(4):828–63. pmid:28333914
- 29. Attenello FJ, Tsung K, Bishara I, Loh YE, Chen TC. In vivo CRISPR screening for novel noncoding RNA functional targets in glioblastoma models. J Neurosci Res. 2021;99(9):2029–45. pmid:33969526
- 30. Teng Y, Pi W, Wang Y, Cowell JK. WASF3 provides the conduit to facilitate invasion and metastasis in breast cancer cells through HER2/HER3 signaling. Oncogene. 2016;35(35):4633–40. pmid:26804171
- 31. Teng Y, Ross JL, Cowell JK. The involvement of JAK-STAT3 in cell motility, invasion, and metastasis. JAKSTAT. 2014;3(1):e28086. pmid:24778926
- 32. Prolo LM, Li A, Owen SF, Parker JJ, Foshay K, Nitta RT, et al. Targeted genomic CRISPR-Cas9 screen identifies MAP4K4 as essential for glioblastoma invasion. Sci Rep. 2019;9(1):14020. pmid:31570734
- 33. Yuan E, Liu K, Lee J, Tsung K, Chow F, Attenello FJ. Modulating glioblastoma chemotherapy response: Evaluating long non-coding RNA effects on DNA damage response, glioma stem cell function, and hypoxic processes. Neurooncol Adv. 2022;4(1):vdac119. pmid:36105389
- 34. Liang J, Liu C, Xu D, Xie K, Li A. LncRNA NEAT1 facilitates glioma progression via stabilizing PGK1. J Transl Med. 2022;20(1):80. pmid:35123484
- 35. Qi J, Pan L, Yu Z, Ni W. The lncRNA RP3-439F8.1 promotes GBM cell proliferation and progression by sponging miR-139-5p to upregulate NR5A2. Pathol Res Pract. 2021;223:153319. pmid:33991848
- 36. Wang P, Bao W, Liu X, Xi W. LncRNA miR143HG inhibits the proliferation of glioblastoma cells by sponging miR-504. Int J Neurosci. 2022;132(11):1137–42. pmid:33461388
- 37. Zhou J, Xu N, Liu B, Wang C, He Z, Lenahan C, et al. lncRNA XLOC013218 promotes cell proliferation and TMZ resistance by targeting the PIK3R2-mediated PI3K/AKT pathway in glioma. Cancer Sci. 2022;113(8):2681–92. pmid:35637600
- 38. Mukherjee V, Lind U, St Onge RP, Blomberg A, Nygard Y. A CRISPR Interference Screen of Essential Genes Reveals that Proteasome Regulation Dictates Acetic Acid Tolerance in Saccharomyces cerevisiae. mSystems. 2021;6(4):e0041821. pmid:34313457
- 39. Liu SJ, Nowakowski TJ, Pollen AA, Lui JH, Horlbeck MA, Attenello FJ, et al. Single-cell analysis of long non-coding RNAs in the developing human neocortex. Genome Biol. 2016;17:67. pmid:27081004
- 40. Chiriboga CA. Nusinersen for the treatment of spinal muscular atrophy. Expert Rev Neurother. 2017;17(10):955–62. pmid:28884620
- 41.
TCGA Cancers Selected for Study: Lower Grade Glioma Study NIH National Cancer Institute, Center for Cancer Genomics: National Institute of Health (NIH); [Available from: https://www.cancer.gov/ccg/research/genome-sequencing/tcga/studied-cancers/lower-grade-glioma-study.
- 42. Louis DN, Perry A, Reifenberger G, von Deimling A, Figarella-Branger D, Cavenee WK, et al. The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary. Acta Neuropathol. 2016;131(6):803–20. pmid:27157931
- 43. Yu W, Ma Y, Hou W, Wang F, Cheng W, Qiu F, et al. Identification of Immune-Related lncRNA Prognostic Signature and Molecular Subtypes for Glioblastoma. Front Immunol. 2021;12:706936. pmid:34899682
- 44. Marwitz S, Depner S, Dvornikov D, Merkle R, Szczygiel M, Muller-Decker K, et al. Downregulation of the TGFbeta Pseudoreceptor BAMBI in Non-Small Cell Lung Cancer Enhances TGFbeta Signaling and Invasion. Cancer Res. 2016;76(13):3785–801.
- 45. Wu Y, Fletcher M, Gu Z, Wang Q, Costa B, Bertoni A, et al. Glioblastoma epigenome profiling identifies SOX10 as a master regulator of molecular tumour subtype. Nat Commun. 2020;11(1):6434. pmid:33339831
- 46. Akdas EY, Temizci B, Karabay A. miR96- and miR182-driven regulation of cytoskeleton results in inhibition of glioblastoma motility. Cytoskeleton (Hoboken). 2023;80(9–10):367–81. pmid:36961307
- 47. Loveless R, Teng Y. Targeting WASF3 Signaling in Metastatic Cancer. Int J Mol Sci. 2021;22(2). pmid:33467681
- 48. Limaye AJ, Whittaker MK, Bendzunas GN, Cowell JK, Kennedy EJ. Targeting the WASF3 complex to suppress metastasis. Pharmacol Res. 2022;182:106302. pmid:35691539
- 49. Morrow D, Minami J, Nathanson DA. Metabolic Vulnerabilities in Brain Cancer. Neurosurg Clin N Am. 2021;32(2):159–69. pmid:33781499
- 50. Sulzmaier FJ, Young-Robbins S, Jiang P, Geerts D, Prechtl AM, Matter ML, et al. RSK2 activity mediates glioblastoma invasiveness and is a potential target for new therapeutics. Oncotarget. 2016;7(48):79869–84. pmid:27829215
- 51. Tao X, Wu X, Zhou P, Yu X, Zhao C, Peng X, et al. UBE2T promotes glioblastoma malignancy through ubiquitination-mediated degradation of RPL6. Cancer Sci. 2022. pmid:36156329
- 52. Li Y, Wang X, Qi S, Gao L, Huang G, Ren Z, et al. Spliceosome-regulated RSRP1-dependent NF-kappaB activation promotes the glioblastoma mesenchymal phenotype. Neuro Oncol. 2021;23(10):1693–708.
- 53. Srinivas T, Siqueira E, Guil S. Techniques for investigating lncRNA transcript functions in neurodevelopment. Mol Psychiatry. 2023. pmid:38145986
- 54. Hira VV, Breznik B, Van Noorden CJ, Lah T, Molenaar RJ. 2D and 3D in vitro assays to quantify the invasive behavior of glioblastoma stem cells in response to SDF-1alpha. Biotechniques. 2020;69(5):339–46.
- 55. Corcoran A, De Ridder LI, Del Duca D, Kalala OJ, Lah T, Pilkington GJ, et al. Evolution of the brain tumour spheroid model: transcending current model limitations. Acta Neurochir (Wien). 2003;145(9):819–24. pmid:14505114
- 56. Hubert CG, Rivera M, Spangler LC, Wu Q, Mack SC, Prager BC, et al. A Three-Dimensional Organoid Culture System Derived from Human Glioblastomas Recapitulates the Hypoxic Gradients and Cancer Stem Cell Heterogeneity of Tumors Found In Vivo. Cancer Res. 2016;76(8):2465–77. pmid:26896279
- 57. Gilbert LA, Larson MH, Morsut L, Liu Z, Brar GA, Torres SE, et al. CRISPR-mediated modular RNA-guided regulation of transcription in eukaryotes. Cell. 2013;154(2):442–51. pmid:23849981
- 58. Krueger F. Trim Galore: a wrapper tool around Cutadapt and FastQC to consistently apply quality and adapter trimming to FastQ files, with some extra functionality for MspI-digested RRBS-type (Reduced Representation Bisufite-Seq) libraries. URL http://www bioinformatics.babraham.ac.uk/projects/trim_galore/ (Date of access: 28/04/2016). 2012.
- 59. Trapnell C, Pachter L, Salzberg SL. TopHat: discovering splice junctions with RNA-Seq. Bioinformatics. 2009;25(9):1105–11. pmid:19289445
- 60. Frankish A, Diekhans M, Ferreira A-M, Johnson R, Jungreis I, Loveland J, et al. GENCODE reference annotation for the human and mouse genomes. Nucleic acids research. 2019;47(D1):D766–D73. pmid:30357393
- 61. Liao Y, Smyth GK, Shi W. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics. 2014;30(7):923–30. pmid:24227677
- 62. Love M, Anders S, Huber W. Differential analysis of count data–the DESeq2 package. Genome Biol. 2014;15(550):10.1186.
- 63. Ceccarelli M, Barthel FP, Malta TM, Sabedot TS, Salama SR, Murray BA, et al. Molecular Profiling Reveals Biologically Discrete Subsets and Pathways of Progression in Diffuse Glioma. Cell. 2016;164(3):550–63. pmid:26824661
- 64. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29(1):15–21. pmid:23104886
- 65. Dennis G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, et al. DAVID: database for annotation, visualization, and integrated discovery. Genome biology. 2003;4(9):1–11. pmid:12734009
- 66.
Smyth GK. Limma: linear models for microarray data. Bioinformatics and computational biology solutions using R and Bioconductor: Springer; 2005. p. 397–420.
- 67. Yu G, Wang L-G, Han Y, He Q-Y. clusterProfiler: an R package for comparing biological themes among gene clusters. Omics: a journal of integrative biology. 2012;16(5):284–7. pmid:22455463