Hairpins I-XII were synthesized as previously described [61]. Oligonucleotides XIII and 7–9 were synthesized via solid phase synthesis using standard phosphoramidite methods (see Supplementary Methods for details).

Samples containing the required strands were heated to 90°C and slowly cooled down to allow triplex formation in suitable buffers (see Supplementary Methods for details). Melting experiments were performed by heating from low temperature to 100°C at 0.5°C/min, monitoring absorbance at 260 nm every 0.5°C. Experiments were repeated for 5, 8, 12, 18, and 22 μM oligonucleotide concentration to derive melting thermodynamic parameters from Van’t Hoff analysis (see Supplementary Methods for details).

NMR spectra of hairpins were first recorded at a range of temperatures (5–45°C). Later, the TFO was added and the mixture was heated (95°C) and cooled down slowly, collecting spectra in the same temperature range. Spectra were acquired in Bruker spectrometers operating at 600 and 800 Mhz, equipped with cryoprobe and processed with the TOPSPIN software. Water suppression was achieved by including an excitation sculpting module in the pulse sequence [62], see Supplementary Methods for additional details.

25 pmoles of DNA oligonucleotide XIII was labeled with ³²P ATP using T4 PNK following the manufacturer protocol. The duplex XIII was formed by heating 32P-labeled DNA XIII at 75°C during 10 min and then quickly cooled on ice. RNA TFO 7 was heated at 75°C for 10 min to prevent self-aggregation and then quickly cooled on ice. Triplex formation was initiated by mixing 2 pmoles of duplex XIII with 2, 20, or 200 pmoles of RNA TFO 7 in RNaseH buffer [20 mM Tris–HCl (pH 7.8), 40 mM KCl, 8 mM MgCl₂, and 1 mM DTT) in a final volume of 40 μl. To equilibrate triplex formation, the reaction mixture was incubated at 37°C for 6 h. Then, 0.5 pmoles of triplex was digested with 0.5 units of RNaseH (Thermo Scientific EN0201) or 1 μg of RNaseA (Thermo Scientific EN0531) during 1 h at 37°C. 4 μl of 6× loading buffer (30% glycerol, 0.05% bromophenol, and 0.025% xylene cyanol) was added, and the sample was directly loaded onto a 12% native polyacrylamide gel, prepared in 50 mM Tris-acetate (pH 7.5) and 10 mM MgCl₂ buffer. Electrophoresis was performed at 6 V/cm for 14.5 h at 4°C in 50 mM Tris-acetate (pH 7.5) and 10 mM MgCl₂ buffer. The gel was dried for 2 h at 64°C and then analyzed by phosphorimaging.

HeLa cells were grown to 90% confluency in T75 flask with Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. Cells were trypsinized and nuclei were isolated using standard procedures (see Supplementary Methods) and lysed to obtain chromatin, which was then subjected to treatment with Proteinase K and DNAse I to yield fragments with an average size >10 kb (Supplementary Fig. S10A) followed by phenol/chloroform extraction and ethanol precipitation (see Supplementary Methods for details).

Purified genomic DNA was sheared into 200–300 bp fragments by sonicating using a Bioruptor Pico (see Supplementary Methods for details). The resulting DNA mixtures were incubated with biotinylated TFO (8 or 9) at two different pH values (pH 7.4 and 5.5). TFO-associated DNA was captured by incubation with MyOne Streptavidin DynaBeads followed by treatment with RNase H and elution with suitable buffer (see Supplementary Methods for details).

Following Roberts and Crothers [58], we determined the enthalpy of triplex formation by (eq. 1):

where (XX) refers to the number or dinucleotide steps of the type XpX in the TFO (CC, UC, CU, or UU), and αs are fitted parameters.

The ΔG is determined as a function of the nucleotide content and the pH (eq. 2):

where all symbols in Greek letters are fitted parameters.

From ΔH and ΔG, we can extract the T_m using (eq. 3) [63].

where R is the ideal gas constant, and C_t is the concentration of the (hairpin) duplex and RNA strands.

The model was parametrized by nonlinear fitting using ΔH and ΔG values obtained from our training set (Supplementary Fig. S8) at different pH and concentrations (see Supplementary Methods for additional details).

We analyzed the triplex potential of annotated lncRNA and microRNA (miRNA) sequences from GENCODE [64] and miRbase [65], respectively. All annotated sequences were scanned with our stability predictor defining potential TFOs with a minimum length of 10 and a maximum of 30 bps. The pH value was set at a default value of 7.0, and the was set at a value of 12 μM. A T_m of 30°C was set as a threshold to classify stable fragments, while T_m of 45°C was considered to detect highly stable triplexes. In order to detect the formation of potential parallel triplex cores (see Supplementary Methods for additional details), we defined our TTSs as polypurine segments with perfect parallel alignment to the previously found TFOs. The extension of the core was evaluated by the melting predictor (see above) with a penalty equal to 10°C decrease in T_m per mismatch. This is about twice the average decrease of melting when one nucleotide is eliminated from a triplex based on our melting experiments, which means that we follow a conservative approach to triplex stability, designed to reduce the number of false positives. A more detailed evaluation of mismatches’ effects would require a different set-up as that considered in previous studies and a careful calibration of the effect of the mismatch in different triplex environments [66].

The population of potential RNA–DNA·DNA triplexes in the human genome was compared with a random background model. In this model we randomly generated 1 million sequences which followed the base distributions found in the human genome. This allowed us to get a large enough sample for scanning candidate TFOs. We then obtained the target sites from our randomly generated TFOs and used them in the downstream analysis. The potential formation of triplexes in the human genome with our predicted TFOs was validated using two published datasets. First, it was validated with the DNA-associated RNA dataset published by Grummt et al. [53] and available in GEO repository under the accession number GSE120849. The data used correspond to the DNA-associated RNA isolated in HeLa cells. The analysis combined the sequences from both the DNA-associated RNA from DNA-IP and SPRI-size selection, which contain 50% and 63% polypyrimidine sequences, respectively. In order to get the percentages of both miRNAs and lncRNAs, we calculated the total number of our candidate TFOs that were found in the isolated RNAs from both samples.

Aiming to find the enriched regions where the RNA candidate TFOs would bind, we re-aligned the complementary sequences of our TFOs against the human genome. We used STAR (version 2.5.3a [67]) mapping the candidate TTSs to the hg38 assembly of the human genome. The aligned reads were mapped to the corresponding annotation files and classified accordingly. The annotations for genes, exons, transcripts, and UTR regions were obtained from GENCODE (Release 35 (GRCh38.p13)). Promoters were defined as the regions from the transcription start sites up until 1 kb upstream. Counts for different features at the gene, promoter regions, exon, transcript, and UTR-level resulting from mapping based on the reads were determined separately using the Bioconductor package Rsubread’s function featureCounts (version 2.0.1) [68]. To further investigate the role of our TTSs, we mapped the obtained promoter sites to their associated genes and performed a Gene Ontology (GO) Analysis using g:Profiler [69]. A Benjamin–Hochberg False Discovery Rate (FDR) index < 0.05 was set to assess significance corrected from multiple test biases. The GO biological processes of the annotated genes were investigated for terms with size >15 and <2700 in order to avoid mappings to large pathways that are of limited value and increase statistical value when removing small pathways [70].

The nucleosome map in human lymphoblastoid cell lines was obtained analyzing MNase-seq data (accession number: GSE36979) from Gaffney et al. [71]. The reads were processed with the nucleR package [72] as follows: mapped fragments were trimmed to 50 bp maintaining the original center and transformed to reads per million. Noise was filtered through Fast Fourier Transform, keeping 2% of the principal components, and peak calling was performed using the following parameters: peak width: 147 bp, peak detection threshold: 35%, maximum overlap of 45 bp, dyad length: 60 bp. The lncRNA and miRNAs expressed in lymphoblastic cells were obtained from RNA-seq and small RNA-seq data (accession numbers: E-MATB-8300 and E-MTAB-8301) [73].

Starting conformations for the six triplexes considered here: r(Py)-d(Pu)·r(Py), r(Py)-d(Pu)·d(Py), d(Py)-d(Pu)·d(Py), d(Py)-d(Pu)·r(Py), r(Py)-r(Pu)·r(Py), and r(Py)-r(Pu)·d(Py) were built from DNA triplex structures [17, 18, 74, 75]. In order to maximize potential unfolding events, most simulations were done using 8-mer triplexes, but control simulations were performed with 18-mer triplexes. Systems were solvated with waters, neutralized with Na⁺ adding 100 mM additional NaCl. The size of the final triclinic box was ∼75 Å × 75 Å × 75 Å for 8-mer and 100 Å × 100 Å × 100 Å for 18-mer, respectively. Simulation systems were optimized and slowly heated and equilibrated for 50 ns prior to production replicas 3 × 500 ns in the isothermal isobaric ensemble (NPT; T = 35 °C and P= 1 atm). In order to improve sampling temperature, replica exchange simulations (RexMD) were done (8 replicas in the range 30–65°C) with individual trajectories extending for at least 400 ns in the NVT ensemble. For well known technical problems, no Mg²⁺ was introduced in the calculations, which should decrease the stability of the triplexes, favoring again the existence of unfolding events. Finally a rough estimate of the impact of crowding effect of chromatin, one of the triplexes (the 8-mer r(Py)-d(Pu)·d(Py) one) was simulated for 0.5 μs in a box containing c.a. 20 mM of DNA duplex (8 mer) with Na⁺ neutralizing the system plus 100 mM NaCl. A control simulation using identical conditions was done using the pure DNA d(Py)-d(Pu)·d(Py) triplex.

Long-range electrostatic interactions were calculated with the particle mesh Ewald method (PME) with a real space cut-off of 12 Å and periodic boundary conditions in the three directions of Cartesian space were used [76]. Constant temperature was imposed using Langevin dynamics [77] with a damping coefficient of 1 ps, while pressure was maintained with Langevin-Piston dynamics [78] with a 200 fs decay period and a 50 fs time constant. LINCS [79] was used to maintain covalent bonds at equilibrium distance, allowing the use of 2 fs integration step. Parmbsc1 was used to describe DNA interactions [80–82], while RNA was described using the chi OL3 force-field [83]. Water molecules were represented by the TIP3P [84] model, while ions were modeled by Dang’s parameters [85].

All MD simulations were performed using AMBER24 code [86]. Coordinates of the systems were collected every 5 ps of the production trajectory. Analyses were carried out using AMBER analysis tools, PyMOL [87], Curves+ [88] and NAFlex [89], and BIGNAsim analysis tools [90, 91]. Trajectories were stored in our BIGNAsim database [90, 91] following FAIR data standards as described elsewhere [91].

Melting experiments were first performed using different homopyrimidine triplexes as TFO and homopurine–homopyrimidine hairpins as TTSs. The use of hairpins (polyethylene glycol was used as loop) has the advantage to minimize the formation of other competing structures in the TTS, such as reverse Watson–Crick [92, 93], Hoogsteen duplexes [94–98], quadruplexes, and others [99–101], which will introduce noise in the T_m estimates. We consider three compositions of the hairpin: (100% A·T/U (I–IV), 70% A·T/U (V–VIII), and 50% A·T/U (IX–XII)), we do not consider higher percentages of guanines as this will increase the risk of quadruplex formation. With these compositions, we create all the combinations of DNA and RNA in the hairpin: d(Pu)·d(Py) (I, V, and IX), r(Pu)·d(Py) (II, VI, and X), d(Pu)·r(Py) (III, VII, and XI), and r(Pu)·r(Py) (IV, VIII, and XII) and incubate them with the corresponding homopyrimidine TFO (DNA with 100%, 70%, or 50% T [1, 3, 5]; RNA with 100%, 70%, or 50% U [2, 4, 6]. Combination of all TFOs with all TTS leads to 24 potential triplexes whose stability was measured by the corresponding melting curves (recorded in all cases at pH 6.0; see “Materials and methods” section). Results (Fig. 2) show melting temperatures in the range T < 15°C (not detectable) to 52°C. Triplexes with 100% A·T/U show in general a poor stability with a decreasing order of stability d(Py)-r(Pu)·r(Py)> d(Py)-d(Pu)·d(Py)> r(Py)-r(Pu)·r(Py)> r(Py)-r(Pu)·d(Py), the rest being not detectable (Fig. 2). When the ratio of G·C increases, the triplexes become more stable, showing a quite well-defined order of stability (Fig. 2): r(Py)-d(Pu)·r(Py)> r(Py)-d(Pu)·d(Py)> d(Py)-d(Pu)·d(Py)> d(Py)-d(Pu)·r(Py)> r(Py)-r(Pu)·d(Py) ≈ r(Py)-r(Pu)·r(Py). At the studied pH, the increase in the ratio G·C/A·T implies an increase in the stability of the triplex. For a given G·C/A·T ratio, the two most stable triplexes are those with RNA in the TFO with d(Pu)·r(Py) preferred over d(Pu)·d(Py) in the TTS, the difference being reduced as the percentage of G/C in the TTS increases. Interestingly, the triplex formed by a single RNA recognizing the genomic duplex (r(Py)-d(Pu)·d(Py)) is very stable, even more than the canonical d(Py)-d(Pu)·d(Py) one.

Our melting experiments can be qualitatively compared by stability estimates obtained by Brown’s group using gel electrophoresis using U/A rich duplexes as TTS and shorter pyrimidine strands as TFO. Due to the used experimental set-up they were not able to detect all possible RNA/DNA topologies in the triplex [66, 102], but their results agree well with ours: r(Py)-r(Pu)·r(Py) is very unstable, and both r(Py)-d(Pu)·d(Py) and d(Py)-d(Pu)·d(Py) show a similar stability. This is consistent with our results from a triplex with c.a. 9/1 ratio between U and C in the TFO.

In order to confirm that melting experiments were really analyzing triplex→duplex transitions instead of other processes related to strand invasion with R-loop formation, we repeated melting experiments for triplexes d(Py)-d(Pu)·d(Py), r(Py)-d(Pu)·d(Py), d(Py)-d(Pu)·r(Py), and r(Py)-d(Pu)·r(Py) for the case of 70% A·T/U in the TTS at higher pH values (from 6.0 to 8.0; see Fig. 3A–H) finding a reduction of T_m, consistent with triplex formation. The triplex nature of the structures was further confirmed by means of ¹H NMR spectroscopy. Thus, ¹H-NMR spectra of several hairpins and their equimolar mixes with their corresponding Hoogsteen strands were recorded at different temperatures (see Fig. 3I and Supplementary Figs S1A–4A). In all the hairpin spectra at T = 5 °C, the number of imino signals in the 12–13 ppm region and ∼14 ppm is consistent with the formation of the expected number of GC and AT/U WC base pairs. As temperature increases, the imino signal intensity decreases. Upon addition of the third strands, the number of exchangeable proton signals increases drastically. At low temperature, NMR signals exhibit a notable broadening, an effect that is more pronounced in triplexes containing more RNA strands. Despite this, the formation of GC Hoogsteen base pairs is demonstrated by the observation of protonated cytosine imino signals ∼15 ppm and their corresponding amino signals ∼10 ppm. Additional imino signals can be observed in the 13–14 ppm region, supporting the formation of extra AU base pairs. Many of these signals persist at 45°C, indicating the formation of a very stable structure (Fig. 3J and Supplementary Figs S1B–4B). Overall, the NMR spectra of all the mixes are consistent with formation of the expected parallel triplexes depicted in the figures’ inserts. Overall, our experiments explain apparently contradictory previous data, agreeing with Crother’s estimates [103] obtained for 66% GC triplexes and with Dervan’s data [60, 103] collected for 81% AT triplexes.

Very interestingly, hybrid triplexes d(Py)-d(Pu)·r(Py) are quite unstable compared to the r(Py)-d(Pu)·d(Py) ones, which suggests that triplexes with a 2:1 (DNA:RNA) stoichiometry show a topology r(Py)-d(Pu)·d(Py). Furthermore, triplexes r(Py)-d(Pu)·r(Py), which (to our knowledge) were never explored before are very stable. However, as its formation requires strand invasion, untwisting of the DNA, and the generation of an unpaired d(Py) strand, we expect them to be unlikely in the cell, except in the context of very active transcription, where the formation of R-loop can facilitate them. Further work is required to verify this exciting possibility. At this time, our results suggest that r(Py)-d(Pu)·d(Py), with TTS being the genomic DNA and the TFO being expressed RNAs will be the most stable triplex topology in the cell.

To gain structural and mechanistic insights on the structure and stability of hybrid triplexes, we performed a set of extensive MD simulations (see “Materials and methods” section) of six triplexes in aqueous solution: r(Py)-d(Pu)·r(Py), r(Py)-d(Pu)·d(Py), d(Py)-d(Pu)·d(Py), d(Py)-d(Pu)·r(Py), r(Py)-r(Pu)·r(Py), and r(Py)-r(Pu)·d(Py). In the stable portion of the trajectories, all hybrid triplexes show similar structures, close to the canonical DNA triplex [10–17, 61], with the DNA and RNA strands showing South and North puckerings respectively. As expected (see “Materials and methods” section) by the length of the triplexes, the temperature, and the absence of Mg²⁺, significant unfolding is detected, which becomes very evident in the disruption of hydrogen bonds, especially the Hoogsteen ones (see Fig. 4). Better conservation of H-bonds is found for the Watson–Crick pairings (see Supplementary Fig. S5), in good agreement with the expected three state unfolding mechanism of triplexes (see Fig. 3).

Very interestingly, the loss of H-bond is not the same for all the triplexes, which help us to obtain qualitative information about their relative stability. Thus, the r(Py)-r(Pu)·d(Py), and specially the r(Py)-r(Pu)·r(Py) triplex shows a massive distortion of the geometry of the Hoogsteen strand, which in fact, remains attached to the duplex just by a few G–C⁺ interactions (see Fig. 4). This agrees with the very poor stability found for these triplexes in our experimental measurements (Fig. 2), as well as with previous experimental data from Brown’s group [102], supporting the reliability of our simulations [61, 80–82]. The remaining triplexes show a similar stability with the triplex with potential impact in gene regulation r(Py)-d(Pu)·d(Py) appearing quite stable (Fig. 4), again in good agreement with our experimental measurements and with previous data by Brown’s group [102].

The use of short triplexes offered us the possibility to detect significant distortions in short simulations but raises concerns about the value of the results for longer oligos. Thus, we decided to explore 18-mer triplexes, using RexMD to increase sampling, and analyzing data at one of the highest temperatures (65°C) explored, which allowed us to enrich the simulation of partially unfolding events. Results in Supplementary Fig. S6 confirms those obtained for shorter oligos, with r(Py)-r(Pu)·d(Py) and r(Py)-r(Pu)·r(Py) triplexes showing poor stability, as expected the differences are narrower with respect to those found for shorter oligos. Again, the triplex of interest r(Py)-d(Pu)·d(Py) shows a stability similar to that of the reference d(Py)-d(Pu)·d(Py) one and much higher than that of the r(Py)-r(Pu)·r(Py) one.

Final concerns arise on the applicability of our dilute simulations in the context of chromatin, where the triplex should face a very crowded environment, which is experimentally suggested to stabilize it [104]. It is very difficult to define a realistic model of the nuclei environment, as inert crowders might be a bad mimic of cellular environments [105], so we decide to create a dense nucleic acid environment (compatible with that existing in human nuclei) by surrounding the triplex by 8-mer duplexes with their associated cationic environment (see “Materials and methods” section). Results suggest that the r(Py)-d(Pu)·d(Py) triplex remains stable in crowded environments, with a maintenance of the structure that is even larger than that found in dilute conditions (see Supplementary Fig. S7).

As discussed above, the most stable triplex (r(Py)-d(Pu)·r(Py)) is not expected to have a large prevalence in the cell out of R-loop constructs. However, the second most stable triplex: r(Py)-d(Pu)·d(Py) can be easily formed by pairing an RNA segment with genomic DNA. Following Robert and Crothers approach [58], we trained a simple nearest neighbor model for r(Py)-d(Pu)·d(Py) to reproduce experimental data in a variety of triplexes (see “Materials and methods” section and Supplementary Figs S8 and S9) in several conditions. The refined method predicts experimental melting observables with root mean square errors around: 4.8 degrees (T_m), and 0.7 kcal/mol (melting free energy), improving dramatically the accuracy obtained by transferring Roberts−Crothers DNA triplex method (see Fig. 5A). Our predictions also outperform the widely used PATO and Triplexator softwares [106, 107] which are unable to detect all stable triplexes at a given temperature; see green bar at Fig. 5B.

To further validate the predictive power of our model, we designed a 50 nt polypyrimidine TFO (TFO 7; Fig. 6A) which, according to our method, should form stable triplexes (T_m values = 57°C at pH 6.5 and 48°C at pH 7.0) in the promoter region of the BRD7 gene. As shown in Fig. 6A, synthetic TFO 7 interacts with a radiolabeled synthetic double-stranded DNA hairpin (XIII) comprising the target BRD7 polypurine sequence, forming a low-mobility complex resistant to RNaseH but sensitive to RNaseA. To validate the triplex nature of this complex, chromatin extracted from HeLa cells (see “Materials and methods” section for details) was treated with DNase I and proteinase K and sonicated to yield fragments of 200–300 nt (see Supplementary Fig. S10). Two aliquots of this purified DNA were incubated with TFO 8, a biotinylated version of TFO 7, at pH 5.5 and 7.0 in the presence of RNase H to digest putative R-loops, or r(Py)-d(Pu)·r(Py) triplexes. The streptavidin-retained DNA was eluted and identified by qPCR amplification, finding significant DNA recovery when using BRD7 promoter-specific primers amplifying a 92 nt region just around the target triplex region (92Nt primers; Fig. 6B, red panel; Supplementary Fig. S11 and Supplementary Table S1). Interestingly, a decrease in the pH led to an increase in DNA recovery, which is consistent with pH-dependent stability in C-C·G triplex formation as captured by our predictor. Similar results were obtained when using promoter-specific primers amplifying a larger region (217 nt) around the target triplex region (217Nt primers; Fig. 6B, yellow panel). On the contrary, no recovery was observed when using intronic-specific primers (Fig. 6B, green panel), or when genomic DNA fragments were incubated with a TFO lacking the sequence matching the target region (TFO 9), confirming the specificity of the above-described results and the triplex nature of the complex predicted by our model.

We used our predictor to screen for potential TFOs amongst annotated human lncRNAs and miRNAs from the gencode and miRbase [64, 65] databases respectively (see “Materials and methods” section). We found a strong enrichment of TFO candidates (triplex T_m > 30°) in both dataset in comparison to the population of expected TTSs from a random distribution (see random model in “Materials and methods” section) (Fig. 7, and Supplementary Tables S2 and S3), suggesting potential r(Py)-d(Pu)·d(Py) triplex formation in vivo. In order to further validate our results, we decided to compare our predicted candidates with two published studies [53, 108]. When compared with the DNA-associated RNA isolated by Grummt et al. [53], we found that 44% or our predicted TFOs from lncRNAs and 51% from miRNAs were indeed found associated with DNA in a triplex structure. In addition, when we intersected our predicted TFOs with those found by Maldonado et al. [108], 35% of our predicted miRNA candidates and 31% of the lncRNA candidates were found in this published dataset. The overall distribution of distances to an in vivo triplex from this study is clustered at 0bp with distances mostly <100 bps (see Supplementary Fig. S12).

Both sets of TTS (from miRNA and lncRNA TFOs) were mapped to the human genome, and we observed an over-representation in promoters when analyzing stable triplexes (T_m > 30°C), and in 5′UTR (in the case of miRNA’s TFO) when analyzing very stable triplexes (T_m > 45°C), suggesting that parallel r(Py)-d(Pu)·d(Py) triplexes form preferentially in regions important for the control of gene expression.

GO analysis of the genes potentially controlled by RNA–DNA·DNA triplex formation with miRNAs and lncRNAs showed that these genes are frequently related to complex processes, such as development (see Supplementary Fig. S13), with very significant hits in the development of the nervous system. It is tempting to speculate that stable triplexes generated by the binding of transcribed RNAs with genomic DNA can be involved in a fine-tuning regulatory mechanism, which was inherited from an ancient triplex-mediated DNA–RNA regulatory network. Note that this finding agrees well with the work from Pasquier et al. in Drosophila that showed that the genes targeted by TFOs were involved in development and morphogenesis [109].

To investigate triplex formation in the context of chromatin, we predicted the putative TFOs from lncRNA and miRNAs expressed in lymphoblastic cells [73] and compared the location of their target sites along the human genome with a genome-wide map of nucleosome occupancy in human lymphoblastoid cell line [71]. We observed a local minimum which coincides with the nucleosome dyads (Fig. 8), suggesting a correlation between chromatin accessibility and triplex formation. These results which agree with previous findings by Maldonado et al. [108] showed that triplexes could form away from the dyad, at the entry–exit site of the nucleosome, helping to fix the nucleosome array and in the case of very long lncRNA helping to approach in the space distant regions as suggested by Marti et al. [55].