VlincRNAs controlled by retroviral elements are a hallmark of pluripotency and cancer

Abstract

Background: The function of the non-coding portion of the human genome remains one of the most important questions of our time. Its vast complexity is exemplified by the recent identification of an unusual and notable component of the transcriptome - very long intergenic non-coding RNAs, termed vlincRNAs.

Results: Here we identify 2,147 vlincRNAs covering 10 percent of our genome. We show they are present not only in cancerous cells, but also in primary cells and normal human tissues, and are controlled by canonical promoters. Furthermore, vlincRNA promoters frequently originate from within endogenous retroviral sequences. Strikingly, the number of vlincRNAs expressed from endogenous retroviral promoters strongly correlates with pluripotency or the degree of malignant transformation. These results suggest a previously unknown connection between the pluripotent state and cancer via retroviral repeat-driven expression of vlincRNAs. Finally, we show that vlincRNAs can be syntenically conserved in humans and mouse and their depletion using RNAi can cause apoptosis in cancerous cells.

Conclusions: These intriguing observations suggest that vlincRNAs could create a framework that combines many existing short ESTs and lincRNAs into a landscape of very long transcripts functioning in the regulation of gene expression in the nucleus. Certain types of vlincRNAs participate at specific stages of normal development and, based on analysis of a limited set of cancerous and primary cell lines, they appear to be co-opted by cancer-associated transcriptional programs. This provides additional understanding of transcriptome regulation during the malignant state, and could lead to additional targets and options for its reversal.

Figure 1

Properties of vlincRNA-associated promoters. VlincRNAs tend to have promoters at one of their bounds (A, B) belonging to either the Active or All category of promoters. This tendency was common to all vlincRNAs (A) or the stand alone vlincRNAs (B) defined based on being at least 50 kb away from the nearest gene on either side. Panels A and B show the density of ENCODE promoter coverage (Y-axis) in all 9 ENCODE cell lines (see text) summed up in 1 kb bins +/-50 kb around boundary of each vlincRNA of the 2,147 vlincRNA (see Additional File 1, Table S1). The coordinates on the X-axis refer to middle position of each bin. VlincRNAs assigned to LTR-containing promoters from any of the 9 cell lines tend to be more tissue specific (C) and have a higher RNAseq signal (D).

Figure 2

Examples of two stand-alone K562-specific vlincRNAs that are distal from annotations. VlincRNAs supported by EST evidence (A & B) and with no EST evidence (C & D) are shown. Zoom-in on the corresponding promoter regions marked by arrows in A& C are shown in the panels B & D. The original 580 vlincRNAs[12] (black, non-strand specific) and the ENCODE vlincRNAs (green, strand-specific) are shown. The vlincRNA designations (vlinc_500 and vlinc_21) refer to Supplementary Table S3 of Kapranov et al [12]. Vlinc_500 (A & B) represents an example of a novel region of ~350 kb whose left bound corresponds fairly well to a cluster of spliced ESTs that share the same 5' ends. Consistent with the K562-specific nature of this region, 9 out of 13 of these ESTs have been isolated from a chronic myelogenous leukemia (CML), the same cancer type as K562, with one of the 9 ESTs spanning almost the entire length of vlincRNA (A). The common 5' end of these ESTs falls within the annotated K562-specific promoter that has within its core an LTR sequence (B). Most of the RNAseq signal was in the intronic regions of these ESTs, consistent with the overall observation that majority of the "dark matter" RNA signal is intronic [12]. Unlike vlinc_500 which is supported by the EST evidence, Vlinc_21 (~121 kb; panels C & D) is located within a totally un-annotated region of chromosome 1 with no spliced EST, ~1.4 Mb from the closest UCSC gene annotation. It also has an LTR repeat at its core (D). RNAseq data shows density of informative reads normalized by 10M informative reads. The Y-axis of the alignability track (Materials and Methods) is on the scale of 0 to 1. Coordinates: hg18.

Figure 3

Examples of two stand-alone K562-specific vlincRNAs that are adjacent (A & B) or overlap (C & D) annotations. Panels A & B show a vlincRNA potentially produced by bi-directional transcription from a promoter upstream of a known gene. Panels C & D show a vlincRNA antisense to known genes. Zoom-in on the corresponding promoter regions marked by arrows in A& C are shown in the panels B & D. The original 580 vlincRNAs[12] (black, non-strand specific) and the ENCODE vlincRNAs (green, strand-specific) are shown. The vlincRNA designations (vlinc_377 and vlinc_185) refer to Supplementary Table S3 of Kapranov et al[12]. Vlinc_377 (~122 kb, panels A & B) is adjacent to an annotated gene, ARL15. However, it does not appear to be connected to ARL15, but rather represent a stand-alone independently regulated unit. Expression of vlinc_377 was restricted to K562, while ARL15 was constitutively expressed and the direction of vlinc_377 transcription was opposite from ARL15 as shown by the vlincRNA from the ENCODE/CSHL long RNAseq dataset [56] produced with a protocol that reduced spurious second strand formation [57](Materials and Methods). Consistent with this, a K562-specific promoter element could be identified at the boundary of vlincRNA_377, upstream of the constitutive promoter controlling the ARL15 expression. Vlinc_185 (C & D) represents an interesting example of a very long antisense transcript. Due to the original constraint of annotating vlincRNAs as intergenic regions, the vlincRNA bounds did not extend over the KCNJ16 gene even though the RNAseq signal clearly extends into that gene (C). Furthermore, a spliced EST BE777672 connects vlinc_185 to an intron of KCNJ16 and the 5' end of this EST corresponds quite well with the end of the RNAseq signal and also with a K562-specific promoter (C & D). Consistent with this, BE777672 was also isolated from a CML source. The entire length of this vlincRNA would be ~130 kb. Finally, the ENCODE K562 vlincRNA shows the recapitulates the strand and the start site of the previously published vlincRNA. RNAseq data shows density of informative reads normalized by 10M informative reads. The Y-axis of the alignability track (Materials and Methods) is on the scale of 0 to 1. Coordinates: hg18.

Figure 4

Luciferase reporter assays. Diagrams of the regions used for the promoter assays (A & B) and the actual results (C) (Materials and Methods) are shown. The original 580 vlincRNAs (black, non-strand specific) and the ENCODE vlincRNAs (green, strand-specific) are shown (A&B). Also shown are the positions of the LTR repeats as annotated by the RepeatMasker, ENCODE K562 promoters (see text), ESTs and the actual regions used for the reporter gene assays (A &B). Error bars show standard deviations of different biological replicas.

Figure 5

Results of vlincRNA RNAi inhibition study. Percent of apoptotic GFP-positive cells (Y-axis) is shown for each indicated treatment (X-axis) for original set of 20 vlincRNAs with 1 siRNA per vlincRNA (A) and for the validation set where 2 additional independent siRNAs per vlincRNA were designed (C). Depletion of selected vlincRNAs was confirmed using real-time RT-PCR (B). "EGFP only" is transfection with EGFP-expressing plasmid only, all other designations imply co-transfection of EGFP-expressing plasmid with the indicated siRNA. Treatments highlighted in yellow are significant at p-value of < 0.05 (Student one-sided t-test). Error bars show standard deviations of 3 independent transfections (A, C) or RT-PCR reaction (B). Results from the negative control siRNA, positive control siRNA against BCR-ABL and AllStars Hs Cell Death positive control siRNA blend are also shown. The vlincRNA data is sorted by the K562-specificity index (see text) with the most K562-specific vlincRNAs on the left (A). Additional details in text and Materials and Methods. The original data is shown in Additional File 3, Table S3.

Figure 6

Schematic diagram of two-dimensions of endogenous LTR promoter activity: gradient of activation in malignant/pluripotent state and preferential association with tissue-specific vlincRNAs.