In order to identify the highest number of putative pol III transcription units possibly accounting for the synthesis of small nuclear (sn)RNA-like transcripts we took advantage of COMPASS, a public software that identifies the simultaneous occurrence of some superimposed conditions in a relevant number of sequences [12]. The conditions chosen defined a canonical pol III type 3 promoter with a PSE (Proximal Sequence Element) consensus sequence, a TATA box and a transcription termination signal at appropriate distances from one another (Table 1). Using these conditions, we identified 3909 putative transcription units (see supplementary data S1) of which a subset of 1473 (37.68%) were maps within a protein-coding gene portion. The analysis of these sequences shows that the averaged PSE/TATA box distance is 34.19 bp whereas the PSE/PolyT distance is 407.8 bp. Since, according to the literature [1], PSE maps 50 bp upstream from the transcription start site, the average size of the transcript would be 358 nucleotides, a size compatible with that of canonical pol III type 3 ncRNA. Although no structural features that might make the possible transcription of the extragenic putative units unlikely were identified, we focused our analysis on the intragenic transcription units due to our interest in their hypothetic role in the regulation of protein-coding gene expression.

In order to assess whether the putative transcription units map preferentially in introns or rather in exons, their distribution was compared to a theoretical one based on the known structure of the human genome using the χ² test [13]. We found that the number of putative transcripts mapping in introns was significantly higher than expected (p = 0.00878), which is in accordance with previous results showing that the introns were enriched in ncRNAs, which mildly regulate gene expression [14].

Next, we analyzed the orientation of the putative ncRNAs inside each gene, assuming the number of elements in 5′-3′ to be statistically equivalent to that in 3′-5′. Interestingly, we found that the vast majority of the transcripts are in 3′-5′ orientation with respect to the coding gene in which they map (p = 0.00664). Therefore, altogether these data indicate that the novel putative transcriptional units identified in the human genome map preferentially within intronic portions of pol II coding genes in an antisense configuration.

In order to predict in silico the possible function of the novel putative transcripts, we took advantage of the David algorithm [15] using all the human genes as background to perform the comparisons. We used the “Functional Annotation Chart” tool to identify the most representative functions among these protein-coding genes.

We found that the most reported function is “alternative splicing”, as 646 of 1075 (60.1%) of the genes belong to this category (Table 2). Since the p-value resulting from the modified Fisher exact test is 7.70 × 10⁻⁴⁸ this test suggests that this function is enriched in our group of transcriptional units. Next, analyzing the false discovery rate of our findings with the Benjamini test, we obtained a p-value of 3.90 × 10⁻⁴⁵ supporting a functional association between protein-coding genes involved in alternative splicing and our collection of transcriptional units.

This result is particularly intriguing in light of our recent works showing that antisense RNAs mapping in introns of protein-coding genes can regulate their splicing pattern in different biological contexts [6,7,14]. It is tantalizing to postulate that putative ncRNAs of our collection mapping in introns of protein-coding genes constitute a new class of splicing regulators across the board, capable of regulating the splicing of splicing regulators. In addition, the “functional annotation clustering” tool of the same algorithm showed that genes grouped into Cluster 1 (enrichment score: 9.19) play a role in the synapse and in the cell junction (Table 3), whereas the “tissue expression” tool showed that their expression is preferentially associated with the hippocampus, epithelium and amygdala (Table 4).

Altogether these results suggest that the function of these novel ncRNAs might be associated with the regulation of expression of protein-coding genes involved in alternative splicing and that they play a role in the central nervous system. In agreement with these results, recent studies show that ncRNAs are indeed enriched in the central nervous system and that their coordinated expression might amplify brain complexity [16]. In addition, several recent reports document a deregulation of ncRNAs in different human neuropathologies such as Alzheimer’s disease, Parkinson’s disease and Fragile X mental retardation, suggesting a possible unprecedented relevance of these small RNAs in physiopathology [8,16–20].

In order to assess whether the predicted novel putative transcription units are indeed transcribed, we selected five elements, mapped in five protein-coding genes of interest (NF1, Neurofibromin 1 NM_001042492; Park2, parkin NM_004562; TNC, Tenascin C NM_002160; Runx2, Runt-related transcription factor 2 NM_004348; NDUFS4, NADH dehydrogenase Fe-S protein 4 NM_002495) as experimental models to test the transcriptional activity of their pol III type 3 promoters in vitro, in different cell lines. To detect quantitatively the transcription rate of the novel elements we fused their pol III type 3 promoters to a luciferase silencer hairpin (pSHAG-NF1/-Park2/-TNC/-Runx2/- NDUFS4, hereafter referred to as pS-NF1/-Park2/-TNC/-Runx2/-NDUFS4). In this condition, if the promoter is active in a specific cell line, the transcription of the hairpin drives the post-transcriptional silencing of a co-transfected luciferase cDNA and, ultimately, leads to the decrease of luciferase signal; on the contrary, an unaltered luminescent signal would indicate that the luciferase is not silenced and thus the promoter of the specific ncRNA is not actively transcribed in the cell line tested. Although a different experimental approach such as in vitro transcription and/or primer extension is needed to unambiguously validate the transcriptional activity of these putative promoters, as shown in Figure 1, they affect luciferase activity in a cell type-specific manner, suggesting specific biological roles associated with their corresponding protein-coding genes. Accordingly, the Runx2-associated ncRNA is actively transcribed in Murine Liver NCTC and in osteosarcoma Osteosarcoma U2OS cell lines (respectively 80% and 72% decrease of luciferase emission in cells transfected with pShag-Runx2), in which the Runx2 protein-coding gene (that encodes a nuclear protein essential for osteoblastic differentiation and skeletal morphogenesis) is expressed at high levels. Similarly, in neuroblastoma SH-SY5Y cells we detected the highest expression level of the park2-associated ncRNA (77% decrease of luciferase emission in cells SH-SY5Y transfected with pShag-Park2) that maps in the Park2 protein (parkin, NM_004562) [20].

The COMPASS (Complex Patterns Search Software, v.1.0.1.1.) software, which allows the use of degenerate sequences [12], was used to search the whole human genome for new possible non coding RNAs transcribed by the pol III promoter. The software searches a motif consisting of sequence and spacer in the whole sequence. As a motif we used the sequence of the pol III type 3 promoter, the PSE (proximal sequence element) and the TATA box, and the poly-T end as a termination signal. The consensus sequence used for the PSE was TYACCNTAAC, for the TATA-Box it was TATA, and for the termination signal a run of four Ts. The spacer between the PSE and TATA-Box was 35 ± 25 bp and the spacer between the TATA box and the poly T was 350 ± 200 bp (Table 1). This distance was chosen because the known transcripts made by pol III have a range from 69 bp (tRNAs, 70–90 bp) to a maximum of 400 bp (SINEs). A hit is only found if all three elements are found in the right distance to each other. To increase the velocity of the search COMPASS searches not only in 5′-3′ direction but also in 3′-5′ direction, converting the search string into 3′-5′. The FastA sequences of the human genome were taken from the University of California, Santa Cruz Genome Browser website [21].

Putative pol III type 3 promoters identified by Compass software lay inside a gene sequence (intron, exon or spanning both). We analyzed this gene list using the DAVID algorithm [15] to extract different biological features and highlight the most over-represented biological annotation of all the genes. We used the “functional annotation card” tool to identify the most representative functions. As a background for the comparison we used all the human genome genes, and, if a biological process was enriched in our group, the p-value of the modified Fisher exact test would be lower than the cut off (0.05). To test the false discovery rate of the findings the Benjamini correction was applied.

The “functional annotation clustering” tool was used to group the similar annotation terms into clusters, and a high enrichment score indicates that the annotation term members in the groups play more important roles. The enrichment score is the geometric mean of all enrichment p values of each annotation term in the group. The minus log transformation is applied to this value so more attention should be given to groups with scores >1.3.

Finally, we used the tool “tissue-expression” to discover whether the genes that lie within these new putative pol III transcription units, are preferably expressed in specific tissues.

HeLa, U2OS and SH5YSY cells were maintained on DMEM medium as described elsewhere [22]. NCTC cells were maintained on MEM medium, 10% FBS and 2 mM L-Glutamine (EuroClone). Subconfluent cells were transfected with Effectene (Quiagen) according to the manufacturer’s instructions.

We selected five new putative pol III type 3 promoters lying inside intronic sequences to investigate their in vitro activity. In Table 2 these genes and the specific introns are reported. To characterize their activity, transient luciferase reporter assays were carried out. Briefly, each promoter sequence (sequence in Supplementary Material S2) was cloned in a pShag plasmid, upstream of a luciferase silencer hairpin. The promoter that activates the hairpin transcription will lead to an inhibition of the luciferase. On the contrary, an unaltered luciferase activity indicates that the putative promoter is not active. As positive and negative controls, a construct with a well-assessed type 3 promoter (U6, component of U6 small nuclear ribonucleoprotein) and a promoterless hairpin, respectively, were used. All cell lines used in this work were co-transfected with two plasmids: pShag, containing the promoter sequence, and pEGFP-N1 plasmid encoding for GFP gene. The experiment was performed in a 96-well plate, and 48 h after the transfection the medium was replaced with 150 μL of PBS to read the fluorescence (emission 535 nm, excitation 485 nm). Then the PBS was replaced with 150 μL of solution of medium and luciferine. The luminescence was read (integration time 1500 ms) and was normalized to the fluorescence as internal control of transfection efficiency. The readings were performed with the Genius pro instrument (TECAN).