Article

ISSN 0006-2979, Biochemistry (Moscow), 2025, Vol. 90, No. 11, pp. 1698-1710 © Pleiades Publishing, Ltd., 2025.

Russian Text © The Author(s), 2025, published in Biokhimiya, 2025, Vol. 90, No. 11, pp. 1816-1829.

1698

Comparative Analysis

of RNA-Chromatin Interactome Data:

Resolution, Completeness, and Specificity

Grigory K. Ryabykh

1,2,a

*, Arina I. Nikolskaya

1,2

, Lidia D. Garkul

and Andrey A. Mironov

1,2

Faculty of Bioengineering and Bioinformatics, Lomonosov Moscow State University, 119234 Moscow, Russia

Vavilov Institute of General Genetics, Russian Academy of Sciences, 119991 Moscow, Russia

e-mail: ryabykhgrigory@gmail.com

Received July 3, 2025

Revised October 25, 2025

Accepted October 28, 2025

Abstract— Two types of experiments are used to study RNA-chromatin interactions: the interactome search

for individual RNAs (“one-to-all” or OTA) and genome-wide contact mapping for all RNAs (“all-to-all” or ATA).

Comparative analysis of ATA and OTA data revealed fundamental differences in resolution, completeness,

and specificity. OTA data exhibit high resolution (~1000  bp) and reproducibility (>90%), serving as a “gold

standard”. ATA data, however, have lower resolution (~5000  bp), and their reproducibility (<10%) is criti-

cally dependent on the protocol, with two-step fixation using disuccinimidyl glutarate and formaldehyde

(GRID-seq) showing a clear advantage over formaldehyde alone. The introduced “chromatin potential” metric

and BaRDIC peak filtering effectively isolate the specific signal. This study proposes a strategy for reliable in-

teractome analysis: combining RNA selection based on chromatin potential with the use of concordant contacts

from peaks.

DOI: 10.1134/S0006297925601923

Keywords: RNA-chromatin interactome, non-coding RNAs, RNA-Chrom database, RNA-chromatin interactome

data concordance

* To whom correspondence should be addressed.

INTRODUCTION

Non-coding RNAs (ncRNAs) in animals and plants

are involved in a wide range of biological process-

es, including cell differentiation, gene expression

regulation, chromatin remodeling, chromatin struc-

ture maintenance, splicing, RNA processing, and bio-

molecular condensate formation. Disruptions in the

ncRNA-mediated regulatory pathways are associated

with the development of various diseases, emphasiz-

ing importance of understanding their mechanisms of

action [1]. A significant portion of ncRNA functions is

realized in the cell nucleus, necessitating a detailed

study of the RNA-chromatin interactome.

RNA molecules interact with numerous proteins,

chromatin, and other RNAs. Experimental methods

to identify DNA loci in contact with ncRNAs can be

divided into two groups: “one-to-all” (OTA) and “all-

to-all” (ATA). The first group (RAP  [2], CHART-seq  [3],

ChIRP-seq  [4], dChIRP-seq  [5], ChOP-seq  [6], CHIRT-seq

[7]) identifies contacts of a known RNA with chro-

matin, while the second group (MARGI  [8], GRID-seq

[9], ChAR-seq  [10,  11], iMARGI  [12], RADICL-seq  [13],

Red-C  [14]) aims to determine all possible RNA-DNA

contacts in the cell [15].

Both groups of methods are actively used in re-

search, but typically not in combination thus prevent-

ing development of the unified standards to enhance

reliability and significance of the RNA-chromatin in-

teractome data analysis. To date, there has been no

systematic comparison of ATA and OTA data in terms

of key characteristics such as accuracy, completeness,

and specificity.

Despite the rapid development of these tech-

nologies, the resulting data are characterized by

COMPARATIVE ANALYSIS OF RNA-CHROMATIN INTERACTOME DATA 1699

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

Fig. 1. Accuracy of determining position of the real contact differs in ATA and OTA protocols. a)Source of position bias in

ATA data – chromatin structure. b) In OTA, the observed contact position offset from the real position is determined only

by the size of DNA fragments. c) Possible source of non-specific interactions.

significant methodological issues and systematic bi-

ases. First, the density of RNA contacts depends on

the distance between the RNA source gene and the

target DNA loci on the same chromosome [9, 11-14,

16]. This bias, termed “RNA-DNA scaling” (RD-scaling),

is analogous to scaling in the DNA-DNA interactome

data (Hi-C method) [17]. Second, chromatin accessi-

bility significantly influences the data, referred to

as “background.” Background is assessed using “in-

put” data in OTA experiments or contacts of the pro-

tein-coding RNAs (mRNAs) in the ATA experiments [9].

Additionally, these experiments inherently have limit-

ed precision in determining contacts. In the ATA ex-

periments, RNA crosslinking with chromatin can occur

at a distance from the real contact (Fig. 1a), whereas

in the OTA experiments, precision of the contact po-

sition determination depends only on the size of DNA

fragments (Fig. 1b). Presence of non-specific interac-

tions poses a particular problem. A significant portion

of the observed contacts could be explained by elec-

trostatic attraction between the negatively charged

RNA and positively charged histone tails, as well as

by preferential crosslinking of amino groups present

on the lysines and arginines of histones by formalde-

hyde [18]. Although affinity of such non-specific inter-

actions is relatively low, their cumulative contribution

is substantial due to the large number of potential

binding sites (Fig.  1c). On the other hand, technical

limitations of the existing experimental methods re-

sult in the loss of some true contacts. These factors

collectively raise questions about the specificity ofthe

detected interactions and accuracy, completeness, and

specificity of the RNA-chromatin interactome data.

The aim of this study is a systematic comparative

analysis of data obtained by OTA and ATA methods

to assess their accuracy, completeness, and specificity.

The following objectives were addressed:

• development of metrics for assessing interaction

specificity (chromatin potential, chP) and data re-

producibility (concordance);

• comparative analysis of replicate consistency

within each method;

• cross-validation of data obtained by different

methods;

• development of recommendations for improv-

ing the reliability of RNA-chromatin interactome

analysis.

MATERIALS AND METHODS

Data. Human and mouse RNA-chromatin interac-

tome data were obtained from the RNA-Chrom data-

base [19]. Only ATA data with corresponding RNA-seq

data from the same cell line were used. When more

than two replicates were available in the ATA data,

the two most complete replicates were selected. RNA-

seq data were obtained from the GEO database and

processed similarly to the ATA data processing pro-

cedure described in the RNA-Chrom. List of the used

data is provided in Tables  S1 and S2 in the Online

Resource  1. Only RNAs demonstrating more than 1000

contacts with chromatin in each replicate were includ-

ed in analysis to ensure sufficient statistical power for

identifying “peaks” (genomic regions enriched with

RNA-chromatin contacts) using the BaRDIC program

[20]. Ribosomal RNAs were excluded from the analy-

sis. For example, applying this filter in the “RADICL,

ES  (NPM)” and “RADICL, ES  (ActD)” experiments left

fewer than 1000 RNAs and less than 50% of contacts

from the initial size of the selected replicates (Fig.  S1,

a  and  b in the Online Resource 1). Considering sig-

nificant overrepresentation of the proximal contacts

(RD-scaling), interactions located within 1  Mb of the

RYABYKH et al.1700

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

genes encoding the corresponding RNAs were exclud-

ed from further analysis in this study.

Use of BaRDIC and threshold selection. Like

most genome-wide data, RNA-chromatin interactome

data are characterized by high level of non-specific

signals (“noise”). Specialized peak-calling algorithms

are used to identify significant interactions by de-

tecting statistically significant clusters of interactions

in the specific genomic loci.

In this study, we used the BaRDIC algorithm

[20], which accounts for RD-scaling and chromatin

openness. This algorithm uses a probabilistic esti-

mate of the likelihood that the contacts in a chroma-

tin locus belong to a peak or noise. The Benjamini–

Hochberg multiple testing correction (FDR, false dis-

covery rate) is next applied to control proportion of

false positives based on the background distribution.

However, in our case, significant overlap of signal

and noise distributions leads to the loss of a sub-

stantial proportion of true interactions when using

a strict FDR threshold. To avoid this problem, we

used a flexible selection criterion: for each RNA, we

selected top 10% of the peaks with the lowest FDR.

Since peak sizes could reach tens of kilobases due

to the data sparsity, all comparisons were conducted

at the level of individual contacts intersecting these

peaks.

For the ATA data analysis, BaRDIC was run with

default parameters. For the OTA experiments, which

have better contact coverage, the following parame-

ters were set: --trans_min 400 bp; --cis_start 100 bp;

--trans_step 50 bp. Background was calculated using

input data converted to BedGraph, with a window

size of 1000 bp.

Chromatin potential. In nearly all studies in-

volving ATA experiments, it has been noted that the

number of RNA contacts with chromatin linearly de-

pends on the expression level of the corresponding

RNA [9-11, 13, 14]. Normalization by expression level

allows us to identify RNAs that demonstrate an in-

creased tendency to interact with chromatin, i.e. the

molecules with contact frequency significantly ex-

ceeding what would be expected at a given expres-

sion level.

To assess tendency of RNAs to contact chroma-

tin, we introduce the concept of “chromatin poten-

tial.” Let us consider RNAs that have more than 1000

contacts in each replicate. Let N

be the total number

of contacts of selected RNAs in the ATA experiment

accounting for the RD-scaling filter; N

be the total

number of uniquely mapped and gene-annotated

reads in the RNA-seq experiment; n

be the number

of contacts accounting for the RD-scaling filter of a

specific i-th RNA in the ATA experiment; and n

the number of reads of a specific i-th RNA in the

RNA-seq experiment. To compare these observations,

we apply a Z-test for proportions. For each i-th RNA,

we calculate the Z-statistic (Z

) (1):

(1)

The Z-statistic follows a standard normal dis-

tribution, allowing us to estimate p-value and the

Benjamini–Hochberg FDR. We refer to the Z-statistic

value as the chromatin potential. Chromatin poten-

tial surpasses the simple ratio of the number of con-

tacts to the expression level because it accounts for

the statistical significance of the deviation. The ratio

of the number of contacts to the expression level is

heavily biased toward RNAs with low coverage in the

RNA-seq data, where the denominator (expression

level) is estimated with significant error, leading to a

wide scatter of values (up to six orders of magnitude;

see Fig.  S2 in the Online Resource 1).

However, the following circumstances must be

considered. First, strand-specific total RNA sequenc-

ing with rRNA depletion is required for such analy-

sis. Second, this analysis is applicable only to long

RNAs, as standard RNA-seq data do not allow for an

adequate assessment of the expression level of RNAs

shorter than 100 nucleotides [21].

RESULTS

Chromatin potential. In all genome-wide stud-

ies of RNA-chromatin interactions (ATA experiments),

there is a significant predominance of mRNA contacts.

This is because mRNAs generally have higher expres-

sion levels compared to ncRNAs. Chromatin potential

(chP) addresses the question of whether the propor-

tion of contacts for a given RNA is statistically signifi-

cantly different from what would be expected if all

RNAs contacted chromatin non-specifically and pro-

portionally to their expression levels. If most mRNA

contacts with chromatin are non-specific, it can be

expected that ncRNAs will demonstrate higher affinity

for chromatin. As expected, most ncRNAs exhibited a

chromatin potential greater than zero (Fig.  2a; Fig.  S3

in the Online Resource 1), but a large number of

mRNAs also had a positive chromatin potential. Asthe

chP threshold increased, the proportion of mRNAs

among the RNAs passing the threshold decreased

(Fig.  2b; Table  S3 in the Online Resource  1) with a

sharp drop in almost all experiments at chP ≥ 20.

The fact that even at high chromatin poten-

tial thresholds, a significant number of protein-cod-

ing RNAs remain may be due to several factors.

COMPARATIVE ANALYSIS OF RNA-CHROMATIN INTERACTOME DATA 1701

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

Fig. 2. Characteristics of RNA chromatin potential. a) Dependence of chromatin potential on the number of RNA contacts

in the Red-C K562 experiment. Blue – protein-coding RNAs, orange – non-coding RNAs. b)Proportion of mRNAs depending

on the chromatin potential threshold (chP > x) for different ATA experiments. The proportion of ncRNAs corresponds to

1 minus the proportion of mRNAs. ActD – actinomycin D treatment; NPM – proteinase K treatment; 1% FA – fixation with

1% formaldehyde; 2% FA – fixation with 2% formaldehyde.

For example, some protein-coding genes contain func-

tional ncRNAs in their intronic regions [22, 23], among

which a significant number of unannotated ncRNAs

can be expected. Positive chromatin potential of some

mRNAs may be associated with these ncRNAs. On the

other hand, non-coding isoforms of mRNAs may them-

selves play a role in chromatin regulation [24].

Comparison of replicates in ATA data. To as-

sess consistency of the RNA-chromatin interactions

between replicates, we evaluated proportion of the

reproducible contacts. Since the exact contact coordi-

nate in the ATA methods could be shifted due to the

protocol specifics, we introduced a genomic distance

parameter(L), within which contacts belonging to the

same RNA but detected in different replicates were

considered concordant. To determine the L threshold

that adequately reflects the method’s resolution, we

calculated, for each RNA, proportion of its contacts

for which at least one contact of the same RNA was

detected in another replicate within the specified dis-

tance L. Analysis of this proportions dependence on

L for the “GRID, ES, Mus musculus” data showed that

the median proportion of concordant contacts ceased

to increase significantly at L  ≥  5000  bp (Fig.  S4 in the

Online Resource  1), reaching a plateau. This indicates

that 5000  bp is an empirical estimate of contact po-

sitioning accuracy in the ATA methods. Based on this

result, for subsequent analysis, we divided the genome

into non-overlapping fragments (bins) of a fixed size

(bin bp). The main analysis was conducted with the

bin size of 5000  bp, corresponding to the empirically

estimated positioning accuracy. To test robustness of

the results and simulate a “high-resolution” scenario,

a bin size of 1000  bp was also used. A bin was con-

sidered concordant for a given RNA if at least one

contact in that bin was detected in both replicates,

and discordant if contacts were present in only one

replicate. This approach allows data aggregation and

quantitative assessment of interaction reproducibility

at the level of genomic loci.

To assess randomness of the matches, we used

a simple model. Assuming that, all RNAs contact ge-

nomic DNA uniformly, the probability of at least one

contact falling into a bin in one experiment can be

estimated as p

bin

(i)  =  n

bin

, where i is the RNA

index; e is the experiment (replicate) number; n

the number of bins with contacts of the i-th RNA;

bin

is the total number of bins into which the cor-

responding genome was divided. Here, we neglect

biases in the data, particularly chromatin accessibili-

ty, and assume that the bin size is sufficiently small.

To avoid the influence of RD-scaling, we selected

bins located more than 1 Mb away from the gene

source of the i-th RNA. The probability that contacts

of the i-th RNA from two experiments (a and b) fall

into the same bin is p

bin

(i)  =  p

bin

(i)∙p

bin

(i). A rough

estimate of the probability of observing k matching

bins can be made using the Bernoulli distribution(2):

. (2)

This allows for a probabilistic assessment of

the correspondence between the replicates or ex-

periments. We define λ(i)  =  (n

bin

). For λ  ≥  10,

RYABYKH et al.1702

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

a normal approximation can be used to estimate the

probability of such event assuming that replicates

have independent contacts (3):

. (3)

For λ < 10, a Poisson approximation is used (4):

. (4)

To assess consistency of the replicates in the ATA

experiments, we analyzed a set of RNAs with more

than 1000 contacts with chromatin in each replicate.

When selecting RNAs, the filter to exclude RD-scaling

regions (within 1  Mb of the RNA source gene) was

not applied; it was applied when selecting contacts.

Importantly, we assessed not just presence of at least

one concordant bin but statistical significance of the

overall level of concordance for each RNA as a whole.

For this, we counted total number of the concordant

and discordant bins for each RNA and applied a sta-

tistical criterion to test the hypothesis of non-ran-

domness of the observed level of matches (see above

for details). An RNA was considered concordant if its

calculated FDR was less than 0.05. As seen in Fig.  S5

in the Online Resource  1, presence of the individual

concordant bins does not guarantee passing this strict

significance threshold.

Table  1 and Fig.  S6 in the Online Resource  1 show

the number of RNAs with concordant bins between

the replicates with FDR < 0.05. The analysis was con-

ducted under four conditions to assess the influence

of two factors: genome bin size (1000  bp vs. 5000  bp)

and contact filtering (all contacts vs. contacts from

BaRDIC peaks). For the GRID experiments, neither of

these conditions had an effect: the number of concor-

dant RNAs remained unchanged and almost always

equal to the initial number of selected RNAs (discus-

sion below).

For other ATA data, as expected, when using

all contacts, the number of statistically significantly

Table 1. Number of RNAs with concordant bins in the replicates (FDR < 0.05)

Experiment

Initial number

of mRNAs

(ncRNAs)

Number of concordant mRNAs

(ncRNAs), all contacts

Number of concordant mRNAs

(ncRNAs), contacts from peaks

Bin 1000 bp Bin 5000 bp Bin 1000 bp Bin 5000 bp

Red-C, K562,

H. sapiens

3230 (636) 1571 (341) 2418 (486) 2779 (556) 3188 (628)

GRID, MM.1S,

H. sapiens

3771 (413) 3771 (413) 3771 (413) 3771 (413) 3771 (413)

GRID, MDA_MB_231,

H. sapiens

4844 (653) 4844 (653) 4844 (653) 4844 (653) 4844 (653)

GRID, ES,

M. musculus

4706 (436) 4706 (429) 4706 (427) 4706 (432) 4706 (435)

RADICL (2% FA), ES,

M. musculus

2758 (162) 1829 (87) 2552 (124) 2226 (131) 2704 (158)

RADICL, OPC,

M. musculus

2580 (197) 1954 (136) 2484 (175) 2203 (161) 2555 (191)

RADICL (ActD), ES,

M. musculus

657 (87) 345 (42) 576 (76) 512 (74) 646 (86)

RADICL (NPM), OPC,

M. musculus

3734 (275) 504 (40) 1464 (103) 2128 (136) 2839 (200)

RADICL (1% FA), ES,

M. musculus

2079 (117) 1533 (66) 1986 (80) 1811 (102) 2056 (115)

RADICL (NPM), ES,

M. musculus

643 (149) 643 (149) 643 (149) 643 (148) 643 (148)

Note. Only RNAs with more than 1000 contacts in each replicate were selected. A filter was applied to exclude contacts

within 1 Mb of the RNA source gene.

COMPARATIVE ANALYSIS OF RNA-CHROMATIN INTERACTOME DATA 1703

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

concordant RNAs was substantially lower under the

strict condition (bin size  =  1000  bp) compared to the

condition corresponding to the method’s resolution

(bin size  =  5000  bp). This confirms that a larger bin

better aggregates technical variations and more ac-

curately reflects interaction reproducibility. The most

important observation was that preliminary selection

of the contacts belonging to BaRDIC peaks significant-

ly increased replicate consistency. This filtering either

increased the number of concordant RNAs or allowed

achieving a comparable level of concordance even

when using a strict bin size of 1000 bp compared to

analyzing all contacts with the bin size of 5000  bp.

Thus, identifying RNA-chromatin interaction peaks

using BaRDIC effectively filters out random interac-

tions and highlights the most reliable, reproducible

RNA-chromatin contacts, significantly increasing con-

sistency between the replicates.

After identifying statistically significant concor-

dant RNAs, we assessed completeness of the ATA data

by calculating the median proportion of contacts that

fall into concordant 5000-bp bins, which corresponds

to the estimated resolution of the ATA methods. This

metric reflects proportion of interactions reproduc-

ible between the replicates out of the total number

of detected contacts (Tables  S4 and S5 in the Online

Resource 1).

Fundamental differences between the methods

were identified. Completeness of the data for the

Red-C and RADICL-seq did not exceed 2% when an-

alyzing all contacts and 5% for the contacts filtered

by the BaRDIC peaks. In contrast, completeness of

the GRID-seq data was significantly higher, reaching

29% and 82% for all contacts and contacts from the

peaks, respectively.

High reproducibility of the GRID data could likely

be explained by the specifics of the fixation proto-

col. Unlike the methods that use only formaldehyde

(such as Red-C and RADICL-seq), the GRID-seq proto-

col employs a two-step fixation with disuccinimidyl

glutarate (DSG) and formaldehyde. DSG is a crosslink-

ing agent with a long spacer (7.7  Å), which effective-

ly crosslinks protein-protein interactions, stabilizing

protein complexes before the chromatin structure is

fixed with formaldehyde [25]. This allows for more

effective “sealing” of protein-mediated RNA-chromatin

interactions, which constitute majority of the specific

contacts.

This hypothesis is quantitatively supported. De-

spite the comparability of the medians of the total

number of contacts of concordant RNAs across all

ATA data (Fig.  S7a in the Online Resource  1), the me-

dian number of reproducible (concordant) contacts in

the GRID-seq data was an order of magnitude higher

than the corresponding indicators for the Red-C and

RADICL- seq data (Fig. S7b in the Online Resource 1).

This indicates that the DSG protocol not only increas-

es the volume of data but fundamentally enhances

proportion of the specific, reproducible signals in

the overall dataset. Thus, the protocol with addition-

al DSG treatment ensures more complete and stable

capture of multiprotein complexes, leading to the sig-

nificant reduction in technical noise and increased

reproducibility between the replicates. Meanwhile,

fixation with formaldehyde alone may inadequately

stabilize large supramolecular complexes, which, in

turn, increases the proportion of random, unstable

interactions and reduces overall concordance.

Despite the fundamental differences in the abso-

lute level of concordance between the methods, we

found a common pattern for all ATA experiments: re-

producibility of contacts positively correlates with the

total number of RNA interactions and its chromatin

potential (Fig.  3; Fig.  S8 in the Online Resource  1). The

discovered dependency allows us to draw two import-

ant conclusions about the nature of RNA-chromatin

interactome data:

1. Completeness of the data is a function of se-

quencing depth for a specific RNA. Low reproducibility

of RNAs with a small number of contacts (<10,000) in-

dicates that for such molecules, the data are substan-

tially incomplete and contain a high level of noise.

Sufficient completeness is achieved only with large

number of interactions, indicating the need for deep

sequencing to reliably identify interactome of the in-

dividual RNAs.

2. Reproducibility is a marker of biological sig-

nificance. Positive correlation between the chromatin

potential and the proportion of concordant contacts

suggests that the more specific is the interaction

(higher chromatin potential), the more stable and re-

producible it is between replicates. This strengthens

the position of chromatin potential not only as a mea-

sure of specificity but also as a predictor of reliability

and reproducibility of interactions.

It is also worth noting that the median propor-

tions of concordance of contacts between mRNAs and

ncRNAs were practically indistinguishable (Tables  S4

and S5, columns 1 and 4 in the Online Resource  1),

indicating that reproducibility does not depend on the

RNA biotype. To interpret the unexpectedly high lev-

el of concordance of the mRNA contacts, comparable

tothat of ncRNAs, two non-exclusive hypotheses have

been proposed:

1. Existence of non-specific but statistically re-

producible interactions, where electrostatic or other

weak forces may lead to massive yet stable binding

of RNA to chromatin.

2. Presence of unknown specific functions in

some mRNAs related to direct interaction with chro-

matin (e.g., mediated by non-coding isoforms or un-

annotated intronic ncRNAs).

RYABYKH et al.1704

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

Fig. 3. Dependence of replicate concordance on the number of contacts and chromatin potential. a and b) Concordance

calculated for all contacts. c and d) Concordance calculated for contacts from BaRDIC peaks. Data from Red-C on K-562

cells, bin size = 5000 bp. MALAT1 is not shown in the graph because this RNA has extreme values of chromatin potential

and proportion of concordant contacts: 991 and 58.2% – panels (a) and (b); 740 and 71.9% – panels (c) and (d).

Comparison of replicates in the OTA data. To

assess reproducibility of the experiments with indi-

vidual RNAs, we analyzed consistency of the repli-

cates in the corresponding datasets (Table  2; Fig.  S9

in the Online Resource 1). First, the expected high

level of reproducibility was confirmed: in the com-

plete OTA dataset, proportion of the concordant con-

tacts between the replicates exceeded 90% even at a

bin size of 1000  bp. This indicates that the OTA data

have resolution of 1000  bp and high completeness.

Second, a critically important aspect of signal spec-

ificity was identified. When the analysis was limited

to only those contacts that fall into peaks identified

by the BaRDIC program (which filters out rare, single

contacts in favor of statistically significant clusters),

the level of concordance dropped by nearly half.

COMPARATIVE ANALYSIS OF RNA-CHROMATIN INTERACTOME DATA 1705

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

Table 2. Proportion of concordant contacts in the OTA replicates (%)

RNA Experiment

Bin = 1000 bp Bin = 5000 bp

All contacts,

Contacts from

peaks, %

All contacts,

Contacts

from peaks, %

JPX

CHART, ES d0

(GSM4278791, GSM4278795)

99.5 53.3 100.0 78.2

JPX

CHART, ES d3

(GSM4278799, GSM4278803)

99.3 36.8 100.0 70.6

JPX

CHART, ES d7

(GSM4278807, GSM4278811)

99.2 44.5 100.0 75.0

MALAT1

ChIRP, ES, genotype: Ythdc1-cKO

(conditional); treatment: DMSO,

(GSM4669091, GSM4669092)

79.9 26.9 99.6 50.6

MALAT1

ChIRP, ES, genotype: Mettl3-WT,

(GSM4875651, GSM4875652)

92.4 40.9 99.9 66.3

Note. d0, d3, and d7 correspond to 0, 3, and 7 days of cell differentiation, respectively; p < 0.05.

This sharp decline suggests that the significant pro-

portion (more than half) of all detected contacts, in-

cluding the concordant ones, in the OTA experiments

are likely non-specific.

Comparative analysis of reproducibility be-

tween the ATA and OTA methods. Comparative

analysis of reproducibility between the ATA and OTA

methods allows us to draw the following conclusions:

1. ATA data (except GRID) are characterized by

low reproducibility between the replicates (median

proportion of concordant contacts <5%), indicating

their substantial incompleteness.

2. OTA data, on the other hand, demonstrate high

reproducibility (>90%), confirming their completeness

and allowing them to be considered as a reliable ref-

erence (“gold standard”) for validating interactions

identified in the genome-wide approaches (ATA data).

Comparison of ATA and OTA experiments.

High reproducibility of the OTA data, demonstrated

in the previous section, allows their use as a ref-

erence to assess the degree of consistency between

the genome-wide approach data (ATA) and this ref-

erence. Conducting such comparative analysis comes

with the significant limitations, as it requires avail-

ability of both types of data for the same RNAs in

the identical cell lines and under similar cultivation

conditions, as well as sufficient number of contacts in

the ATA data to ensure statistical power. The publicly

available OTA data matching the conditions of ATA

experiments were found only for two RNAs: MALAT1

and JPX.

For the ncRNAs MALAT1 and JPX, we conduct-

ed comparison using bins with 5000-bp size. As a

measure of consistency, we calculated proportion of

the contacts from the ATA data that fell into bins en-

riched with the contacts from the BaRDIC peaks of

the corresponding OTA experiment. The analysis was

performed for all ATA contacts as well as for the sub-

set filtered by the BaRDIC peaks. The sets of contacts

from the ATA replicates were combined to increase

data power. The results for the ncRNA MALAT1 are

presented in Table 3 and Fig. S10 in the Online Re-

source 1, and for the ncRNA JPX in Table 4.

For the ncRNA MALAT1, which exhibits an ex-

tremely high level of interactions in the ATA data,

a significant proportion (~50%) of overlaps with the

OTA data was identified, indicating good consistency

between the methods. However, approximately half

of the MALAT1 contacts detected solely by the ATA

method are not confirmed by the independent OTA

method. This allows us to estimate proportion of

non-specific signals in the ATA data for this RNA as

~50%. Importantly, in this case, we do not observe a

significant advantage of the GRID-seq method, which

was so evident in the analysis of the ATA replicate

consistency. This is likely because the total number

of contacts and their reproducibility for MALAT1 are

so high in all ATA experiments that the effect of the

more specific fixation protocol is overshadowed by

the dominant signal.

The situation for the ncRNA JPX, characterized

by low level of contacts in the ATA data, is funda-

mentally different. Overlap with the OTA data was

~60%, allowing us to roughly estimate proportion

of non-specific contacts as ~40%. The low absolute

number of contacts makes this estimate less reli-

able. As expected, based on the known association of

JPX with XIST [26], most of its contacts are localized

RYABYKH et al.1706

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

Table  3. Percentage of the consistent MALAT1 RNA contacts with chromatin in the ATA data when compared

with the MALAT1 RNA contacts from the OTA experiments (contacts from BaRDIC peaks) in mouse embryonic

stem cells

Experiment

Number

of contacts

in ATA data

RAP ChIRP

pSM33 ES,

DMSO

1 hour, %

V6.5 ES, %

ES, Ythdc1-cKO;

DMSO, %

ES, Mettl3-WT,

E14 ES, %

GRID, ES,

M. musculus

522,741

(109,371)

38.0 (46.5) 55.8 (61.6) 50.6 (51.0) 58.8 (58.3) 53.8 (57.0)

RADICL (1% FA), ES,

M. musculus

636,802

(138,422)

42.9 (56.2) 61.5 (74.1) 50.4 (49.6) 58.9 (57.2) 55.1 (58.6)

RADICL (2% FA), ES,

M. musculus

484,878

(99,985)

41.5 (51.0) 59.7 (69.0) 50.7 (49.6) 59.2 (58.2) 54.9 (56.8)

Note. Values in parentheses represent results for the ATA contacts from BaRDIC peaks. Bin size = 5000 bp.

Table  4. Percentage of the consistent JPX RNA contacts with chromatin in the ATA data (all contacts) when

compared with the JPX RNA contacts from OTA experiments (contacts from BaRDIC peaks) in mouse embryonic

stem cells

Experiment

Number of JPX contacts

in ATA data

CHART, ES

d0 d3 d7

GRID, ES,

M. musculus

459 57.1 (0.05) 61.9 (0.22) 63.6 (0.002)

RADICL (1% FA), ES,

M. musculus

341 57.2 (0.09) 62.8 (0.15) 61.0 (0.03)

RADICL (2% FA), ES,

M. musculus

332 54.5 (0.24) 56.6 (0.65) 63.9 (0.005)

Note. Values in parentheses represent the p-value of concordance. Bin size = 5000 bp; d0, d3, and d7 correspond to 0, 3,

and 7 days of cell differentiation, respectively.

on the X chromosome. The low number of contacts

did not allow application of the BaRDIC peak filter-

ing, which would likely have increased specificity

of the analysis.

The conducted analysis demonstrates fundamen-

tal possibility of cross-validation but also highlights its

limitations. Unfortunately, for the RNAs with average

level of interactions – which are of the greatest in-

terest for assessing specificity of the ATA methods –

comparative analysis with OTA was not possible due

to the lack of paired data under consistent biolog-

ical conditions. Thus, the OTA data serve as a reli-

able reference primarily for the highly interacting

RNAs, while assessing specificity of ATA for the rest

of interactome requires development of alternative

approaches.

Comparison of the OTA experiments. To as-

sess consistency of the OTA data, we conducted com-

parative analysis of the RNA-chromatin interaction

maps for various ncRNAs in human cells (Fig. 4) and

mouse cells (Fig. S11 in the Online Resource 1). As a

measure of similarity, we used the ratio of concor-

dant contacts to the total number of detected inter-

actions in the compared OTA experiments (Jaccard

index).

The analysis revealed clusters of high function-

al consistency, as well as overlaps, likely related to

the common principles of chromatin organization. In

the heatmap for human cells, distinct clusters were

observed, corresponding to the specific RNAs such

as MALAT1, NEAT1, and HOTAIR. The most striking

example of the expected similarity were the profiles

of MALAT1 and NEAT1. High concordance of their

chromatin contacts aligns well with their known co-

localization in the nucleus: NEAT1 serves as a struc-

tural basis of paraspeckles, while MALAT1 is a key

COMPARATIVE ANALYSIS OF RNA-CHROMATIN INTERACTOME DATA 1707

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

Fig. 4. Heatmap reflecting proportion of the concordant contacts (from BaRDIC peaks, FDR<0.05) from “one-to-all” experi-

ments for the human cell lines. Non-significant enrichments (p>0.05) are set to zero. Clustering is performed by cell types

and RNAs used in the experiment. Bin size = 1000 bp.

component of nuclear speckles [27, 28]. Both RNAs

are associated with active genes and are involved in

splicing regulation [16], which explains similarity in

their chromatin landscape.

Overlaps were also detected, for example, be-

tween the contacts of LINC02085 and DACOR1.

LINC02085 is involved in the NF-κB-dependent regu-

lation [29], while DACOR1 is implicated in maintain-

ing DNA methylation patterns [30], which may reflect

their joint involvement in epigenetic control.

At the same time, the analysis identified simi-

larity clusters lacking an obvious functional expla-

nation. For instance, profile of the telomerase RNA

TERC showed significant concordance with the RNAs

such as SRA1, SNHG1, and KCNQ1OT1, for which di-

rect functional links are unknown. This result sug-

gests presence of the background noise. If we assume

that most of the detected RNA-chromatin interactions

are protein-mediated, low specificity of these contacts

could be attributed not to the experimental methods

themselves but to the relatively low specificity of the

RNA-binding domains in the proteins [31, 32]. This

leads to similar association patterns for the function-

ally unrelated RNAs.

Unlike the human data, the mouse OTA data pri-

marily focus on the study of XIST. The observed high

concordance of the XIST profiles with the RNAs such

as its known activator JPX [26] serves as an addition-

al internal quality control for the data and confirms

specificity of the method for the functionally related

pairs (Fig. S11).

CONCLUSION

This study conducted comparative analysis of the

RNA-chromatin interactome data obtained using “all-

to-all” (ATA) and “one-to-all” (OTA) methods, focusing

RYABYKH et al.1708

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

on evaluating their accuracy, completeness, and spec-

ificity.

We compared the genome-wide RNA-chromatin

interactome data (ATA) with the RNA-seq data and

introduced the concept of chromatin potential – a

numerical characteristic of individual RNAs that in-

dicates the extent to which the number of contacts

of a given RNA exceeds the expected number based

on the RNA-seq data. This metric allows filtering out

RNAs with predominantly non-specific interactions

due to the high expression levels. Setting a threshold

for chP significantly reduces proportion of mRNAs in

the interactome, effectively isolating RNAs with high-

er affinity for chromatin. It is important to note that

some mRNAs exhibit high chromatin potential, which

could indicate presence of unknown specific functions

or expression of the non-coding isoforms and unan-

notated intronic ncRNAs. Positive correlation of chP

with the contact reproducibility confirms that chro-

matin potential is not only a measure of specificity

but also a predictor of interaction reliability.

Comparison of the methods revealed fundamen-

tal differences in resolution (~5000 bp for ATA vs.

~1000bp for OTA) and reproducibility. The developed

metric for replicate consistency showed that the OTA

data have high reproducibility (>90%), allowing them

to be used as a “gold standard.” In contrast, the ATA

data (except for GRID-seq) were characterized by low

concordance (<5-10%), indicating substantial incom-

pleteness.

It was found that completeness of the ATA data

is a function of sequencing depth for each specific

RNA. To achieve statistically significant reproducibil-

ity, the number of RNA contacts must exceed 10,000,

indicating the need for exceptionally deep sequencing

in the genome-wide experiments to reliably identify

interactome of the individual RNAs.

Critical influence of the fixation protocol was

demonstrated. It was shown that the use of the two-

step fixation with DSG/formaldehyde (GRID-seq) com-

pared to the fixation with formaldehyde alone (Red-C,

RADICL-seq) leads to the significant increase in the

proportion of reproducible signals. This suggests crit-

ical role of the protein complex stabilization in the

quality of ATA data.

In all types of experiments, preliminary selection

of contacts belonging to the peaks identified using

BaRDIC significantly increased the data consistency.

This proves that identifying statistically significant in-

teraction clusters is a powerful tool for separating the

biologically significant signals from the background

noise.

Based on these results, we recommend the follow-

ing approach to enhance reliability and significance

of conclusions when working with the RNA-chromatin

interactome data:

• When analyzing the OTA data, focus on the con-

tacts that have passed peak filtering (e.g., using

BaRDIC), as they demonstrate significantly high-

er specificity. The high overall reproducibility of

OTA data confirms their reliability as a reference.

• Note that chromatin potential selects promis-

ing RNAs, while concordance analysis and peak

searching select significant RNA-chromatin con-

tacts. Therefore, when analyzing the ATA data,

the strategy should be two-level:

1. First, select RNAs with high chromatin poten-

tial (chP > 20), focusing on the molecules with in-

creased probability of specific interactions with chro-

matin.

2. Second, select RNAs with more than 10,000

contacts and use only those contacts that both fall

into the BaRDIC peaks and are reproducible between

the replicates.

Thus, the combined use of chromatin potential

(for RNA selection) and concordant contacts from the

peaks (for genomic locus selection) maximizes filter-

ing of non-specific noise and highlights the most re-

liable interactions. The proposed approach enhances

reliability of bioinformatics analysis and interpreta-

tion of the RNA-chromatin interactome data, which

is particularly important for identifying functionally

significant associations.

Supplementary information

The online version contains supplementary material

available at https://doi.org/10.1134/S0006297925601923.

Contributions

G. K. Ryabykh – analysis of chromatin potential, concor-

dance analysis of replicates, and comparison of “all-to-

all” (ATA) and “one-to-all” (OTA) data; A. I. Nikolskaya–

calculation of BaRDIC peaks and concordance analysis

of “one-to-all” (OTA) data; L. D. Garkul – processing

of RNA sequencing data; A. A. Mironov – conceptual-

ization and supervision of the study; G. K. Ryabykh,

A. I. Nikolskaya, L. D. Garkul, and A. A. Mironov– writ-

ing the manuscript; G. K. Ryabykh, A. I. Nikolskaya,

L. D. Garkul, and A. A. Mironov – editing the manu-

script.

Funding

This work was financially supported by the Russian

Science Foundation (grant no.23-14-00136).

Ethics approval and consent to participate

This work does not contain any studies involving hu-

man and animal subjects.

Conflict of interest

The authors of this work declare that they have

noconflicts of interest.

COMPARATIVE ANALYSIS OF RNA-CHROMATIN INTERACTOME DATA 1709

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

REFERENCES

1. Mattick, J.S., Amaral, P. P., Carninci,P., Carpenter,S.,

Chang, H. Y., Chen, L.-L., Chen, R., Dean, C., Dinger,

M. E., Fitzgerald, K. A., Gingeras, T. R., Guttman, M.,

Hirose, T., Huarte, M., Johnson, R., Kanduri, C.,

Kapranov,P., Lawrence, J.B., Lee, J. T., Mendell, J. T.,

Mercer, T. R., Moore, K. J., Nakagawa, S., Rinn, J. L.,

Spector, D. L., et al. (2023) Long non-coding RNAs:

definitions, functions, challenges and recommenda-

tions, Nat. Rev. Mol. Cell Biol., 24, 430-447, https://

doi.org/10.1038/s41580-022-00566-8.

2. Engreitz, J. M., Pandya-Jones, A., McDonel, P.,

Shishkin,A., Sirokman,K., Surka,C., Kadri,S., Xing,j.,

Goren, A., Lander, E. S., Plath, K., and Guttman, M.

(2013) The Xist lncRNA exploits three-dimensional

genome architecture to spread across the Xchromo-

some, Science, 341, 1237973, https://doi.org/10.1126/

science.1237973.

3. Simon, M.D., Wang, C.I., Kharchenko, P.V., West, J.A.,

Chapman, B. A., Alekseyenko, A. A., Borowsky, M. L.,

Kuroda, M.I., and Kingston, R.E. (2011) Thegenomic

binding sites of a noncoding RNA, Proc. Natl. Acad.

Sci. USA, 108, 20497-20502, https://doi.org/10.1073/

pnas.1113536108.

4. Chu,C., Qu,K., Zhong, F.L., Artandi, S.E., and Chang,

H. Y. (2011) Genomic maps of long noncoding RNA

occupancy reveal principles of RNA-chromatin inter-

actions, Mol. Cell, 44, 667-678, https://doi.org/10.1016/j.

molcel.2011.08.027.

5. Quinn, J. J., Ilik, I. A., Qu, K., Georgiev, P., Chu, C.,

Akhtar, A., and Chang, H. Y. (2014) Revealing long

noncoding RNA architecture and functions using do-

main-specific chromatin isolation by RNA purification,

Nat. Biotechnol., 32, 933-940, https://doi.org/10.1038/

nbt.2943.

6. Mondal,T., Subhash, S., Vaid, R., Enroth, S., Uday, S.,

Reinius, B., Mitra, S., Mohammed, A., James, A. R.,

Hoberg, E., Moustakas, A., Gyllensten, U., Jones,

S. J., Gustafsson, C. M., Sims, A. H., Westerlund, F.,

Gorab,E., and Kanduri,C. (2015) MEG3 long noncod-

ing RNA regulates the TGF-β pathway genes through

formation of RNA-DNA triplex structures, Nat. Com-

mun., 6, 7743, https://doi.org/10.1038/ncomms8743.

7. Chu, H. P., Cifuentes-Rojas, C., Kesner, B., Aeby, E.,

Lee, H.-G., Wei, C., Oh, H. J., Boukhali, M., Haas, W.,

and Lee, J. T. (2017) TERRA RNA antagonizes ATRX

and protects telomeres, Cell, 170, 86-101, https://

doi.org/10.1016/j.cell.2017.06.017.

8. Sridhar,B., Rivas-Astroza,M., Nguyen, T.C., Chen,W.,

Yan, Z., Cao, X., Hebert, L., and Zhong, S. (2017) Sys-

tematic mapping of RNA-chromatin interactions

invivo, Curr. Biol., 27, 602-609, https://doi.org/10.1016/

j.cub.2017.01.011.

9. Li, X., Zhou, B., Chen, L., Gou, L. T., Li, H., and Fu,

X.D. (2017) GRID-seq reveals the global RNA-chroma-

tin interactome, Nat. Biotechnol., 35, 940-950, https://

doi.org/10.1038/nbt.3968.

10. Bell, J. C., Jukam, D., Teran, N. A., Risca, V. I., Smith,

O.K., Johnson, W.L., Skotheim, J.M., Greenleaf, W.J.,

and Straight, A. F. (2018) Chromatin-associated RNA

sequencing (ChAR-seq) maps genome-wide RNA-to-

DNA contacts, Elife, 7, e27024, https://doi.org/10.7554/

eLife.27024.

11. Limouse, C., Smith, O. K., Jukam, D., Fryer, K. A.,

Greenleaf, W. J., and Straight, A. F. (2023) Global

mapping of RNA-chromatin contacts reveals a prox-

imity-dominated connectivity model for ncRNA-

gene interactions, Nat. Commun., 14, 6073, https://

doi.org/10.1038/s41467-023-41848-9.

12. Yan,Z., Huang,N., Wu,W., Chen,W., Jiang,Y., Chen,J.,

Huang,X., Wen,X., Xu, j., Jin, Q., Zhang,K., Chen,Z.,

Chien, S., and Zhong, S. (2019) Genome-wide colo-

calization of RNA–DNA interactions and fusion RNA

pairs, Proc. Natl. Acad. Sci. USA, 116, 3328-3337,

https://doi.org/10.1073/pnas.1819788116.

13. Bonetti, A., Agostini,F., Suzuki, A. M., Hashimoto, K.,

Pascarella, G., Gimenez, J., Roos, L., Nash, A. J.,

Ghilotti,M., Cameron, C.J.F., Valentine,M., Medvedeva,

Y. A., Noguchi, S., Agirre, E., Kashi, K., Samudyata,

Luginbühl,J., Cazzoli,R., Agrawal,S., Luscombe, N.M.,

Blanchette, M., Kasukawa, T., Hoon, M., Arner, E.,

Lenhard, B., et al. (2020) RADICL-seq identifies gen-

eral and cell type-specific principles of genome-wide

RNA-chromatin interactions, Nat. Commun., 11, 1018,

https://doi.org/10.1038/s41467-020-14337-6.

14. Gavrilov, A. A., Zharikova, A. A., Galitsyna, A. A.,

Luzhin, A.V., Rubanova, N.M., Golov, A. K., Petrova,

N.V., Logacheva, M.D., Kantidze, O.L., Ulianov, S.V.,

Magnitov, M.D., Mironov, A.A., and Razin, S.V. (2020)

Studying RNA-DNA interactome by Red-C identifies

noncoding RNAs associated with various chromatin

types and reveals transcription dynamics, Nucleic

Acids Res., 48, 6699-6714, https://doi.org/10.1093/nar/

gkaa457.

15. Ryabykh, G.K., Mylarshchikov, D.E., Kuznetsov, S.V.,

Sigorskikh, A. I., Ponomareva, T. Y., Zharikova, A. A.,

and Mironov, A. A. (2022) RNA-chromatin interac-

tome: What? Where? When? Mol. Biol., 56, 210-228,

https://doi.org/10.31857/S002689842202015X.

16. West, J. A., Davis, C. P., Sunwoo, H., Simon, M. D.,

Sadreyev, R. I., Wang, P. I., Tolstorukov, M. Y., and

Kingston, R. E. (2014) The long noncoding RNAs

NEAT1 and MALAT1 bind active chromatin sites,

Mol. Cell, 55, 791-802, https://doi.org/10.1016/j.molcel.

2014.07.012.

17. Lieberman-Aiden,E., van Berkum, N.L., Williams,L.,

Imakaev, M., Ragoczy, T., Telling, A., Amit, I., Lajoie,

B. R., Sabo, P. J., Dorschner, M. O., Sandstrom, R.,

Bernstein,B., Bender, M.A., Groudine,M., Gnirke,A.,

Stamatoyannopoulos, J., Mirny, L. A., Lander, E. S.,

and Dekker, J. (2009) Comprehensive mapping of

RYABYKH et al.1710

BIOCHEMISTRY (Moscow) Vol. 90 No. 11 2025

long-range interactions reveals folding principles of

the human genome, Science, 326, 289-293, https://

doi.org/10.1126/science.1181369.

18. Hoffman, E. A., Frey, B. L., Smith, L. M., and Auble,

D. T. (2015) Formaldehyde crosslinking: a tool for

the study of chromatin complexes, J.Biol. Chem.,

44, 26404-26411, https://doi.org/10.1074/jbc.R115.

651679.

19. Ryabykh, G. K., Kuznetsov, S. V., Korostelev, Y. D.,

Sigorskikh, A. I., Zharikova, A. A., and Mironov,

A. A. (2023) RNA-Chrom: a manually curated an-

alytical database of RNA–chromatin interactome,

Database, 2023, baad025, https://doi.org/10.1093/

database/baad025.

20. Mylarshchikov, D.E., Nikolskaya, A.I., Bogomaz, O.D.,

Zharikova, A.A., and Mironov, A.A. (2024) BaRDIC: ro-

bust peak calling for RNA-DNA interaction data, NAR

Genom. Bioinform., 6, lqae054, https://doi.org/10.1093/

nargab/lqae054.

21. Alberti, A., Belser, C., Engelen, S., Bertrand, L., Or-

vain, C., Brinas, L., Cruaud, C., Giraut, L., Silva, C.,

Firmo, C., Aury, J.-M., and Wincker, P. (2014) Com-

parison of library preparation methods reveals their

impact on interpretation of metatranscriptomic

data, BMC Genomics, 15, 912, https://doi.org/10.1186/

1471-2164-15-912.

22. Lin, S.-L., Miller, J. D., and Ying, S.-Y. (2006) Intron-

ic microRNA (miRNA), J.Biomed. Biotechnol., 2006,

026818, https://doi.org/10.1155/JBB/2006/26818.

23. Bergeron, D., Faucher-Giguère, L., Emmerichs, A.-K.,

Choquet, K., Song, K. S., Deschamps-Francoeur, G.,

Fafard-Couture, E., Rivera, A., Couture, S., Church-

man, L.S., Heyd,F., Elela, S.A., and Scott, M.S. (2023)

Intronic small nucleolar RNAs regulate host gene

splicing through base pairing with their adjacent

intronic sequences, Genome. Biol., 24, 160, https://

doi.org/10.1186/s13059-023-03002-y.

24. Nam, J.-W., Choi, S.-W., and You, B.-H. (2016) Incred-

ible RNA: dual functions of coding and noncod-

ing, Mol. Cells, 39, 367-374, https://doi.org/10.14348/

molcells.2016.0039.

25. Machyna, M., and Simon, M. D. (2018) Catching

RNAs on chromatin using hybridization capture

methods, Brief. Funct. Genomics, 17, 96-103, https://

doi.org/10.1093/bfgp/elx038.

26. Oh, H. J., Aguilar,R., Kesner, B., Lee, H.-G., Kriz, A. J.,

Chu, H.-P., and Lee, J.T. (2021) Jpx RNA regulates CTCF

anchor site selection and formation of chromosome

loops, Cell, 184, 6157-6173, https://doi.org/10.1016/

j.cell.2021.11.012.

27. Clemson, C. M., Hutchinson, J. N., Sara, S. A., Ens-

minger, A. W., Fox, A. H., Chess, A., and Lawrence,

J. B. (2009) An architectural role for a nuclear non-

coding RNA: NEAT1 RNA is essential for the struc-

ture of paraspeckles, Mol. Cell, 33, 717-726, https://

doi.org/10.1016/j.molcel.2009.01.026.

28. Tripathi, V., Ellis, J. D., Shen, Z., Song, D. Y., Pan, Q.,

Watt, A. T., Freier, S. M., Bennett, C. F., Sharma, A.,

Bubulya, P. A., Blencowe, B. J., Prasanth, S. G., and

Prasanth, K. V. (2010) The nuclear-retained noncod-

ing RNA MALAT1 regulates alternative splicing by

modulating SR splicing factor phosphorylation, Mol.

Cell., 39, 925-938, https://doi.org/10.1016/j.molcel.

2010.08.011.

29. Cai, D., and Han, J.-D. J. (2021) Aging-associated

lncRNAs are evolutionarily conserved and partici-

pate in NFκB signaling, Nat. Aging, 1, 438-453, https://

doi.org/10.1038/s43587-021-00056-0.

30. Merry, C. R., Forrest, M. E., Sabers, J. N., Beard, L.,

Gao, X.-H., Hatzoglou, M., Jackson, M. W., Wang, Z.,

Markowitz, S. D., and Khalil, A. M. (2015) DNMT1-

associated long non-coding RNAs regulate global

gene expression and DNA methylation in colon can-

cer, Hum. Mol. Genet., 24, 6240-6253, https://doi.org/

10.1093/hmg/ddv343.

31. Stitzinger, S. H., Sohrabi-Jahromi, S., and Söding, J.

(2023) Cooperativity boosts affinity and specifici-

ty of proteins with multiple RNA-binding domains,

NAR Genom. Bioinform., 5, lqad057, https://doi.org/

10.1093/nargab/lqad057.

32. Khlebnikov, D. A., Nikolskaya, A. I., Zharikova, A.A.,

and Mironov, A. A. (2025) Comprehensive analysis

of RNA-chromatin, RNA-, and DNA-protein interac-

tions, NAR Genom. Bioinform., 7, lqaf010, https://

doi.org/10.1093/nargab/lqaf010.

Publisher’s Note. Pleiades Publishing remains

neutral with regard to jurisdictional claims in published

maps and institutional affiliations. AI tools may have

been used in the translation or editing of this article.