Cell type-specific intronic RNAs shape genome architecture during neuronal lineage specification

Cell type-specific intronic RNAs shape genome architecture during neuronal lineage specification | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Cell type-specific intronic RNAs shape genome architecture during neuronal lineage specification Wenjing Kang , View ORCID Profile Wing Hin Yip , View ORCID Profile Quentin Verron , View ORCID Profile Britta A.M. Bouwman , Xiaoze Li-Wang , Andrea Abou Yaghi , Mitsuyoshi Murata , Merula Stout , View ORCID Profile Lorenzo Salviati , Xufeng Shu , Kayoko Yasuzawa , View ORCID Profile Marco Gaviraghi , View ORCID Profile Rodrigo Pracana , Johan Lord , View ORCID Profile Roberto Ballarino , View ORCID Profile Anna Falk , Jay W. Shin , View ORCID Profile Takeya Kasukawa , View ORCID Profile Chi Wai Yip , View ORCID Profile Masaki Kato , View ORCID Profile Hazuki Takahashi , View ORCID Profile Nicola Crosetto , View ORCID Profile Piero Carninci , View ORCID Profile Magda Bienko doi: https://doi.org/10.1101/2025.03.13.641826 Wenjing Kang 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Wing Hin Yip 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Wing Hin Yip Quentin Verron 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Quentin Verron Britta A.M. Bouwman 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Britta A.M. Bouwman Xiaoze Li-Wang 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Andrea Abou Yaghi 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Mitsuyoshi Murata 3 RIKEN Center for Integrative Medical Sciences , Yokohama, Kanagawa 230-0045 Japan Find this author on Google Scholar Find this author on PubMed Search for this author on this site Merula Stout 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Lorenzo Salviati 4 Human Technopole , Milan, Italy Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Lorenzo Salviati Xufeng Shu 3 RIKEN Center for Integrative Medical Sciences , Yokohama, Kanagawa 230-0045 Japan Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kayoko Yasuzawa 3 RIKEN Center for Integrative Medical Sciences , Yokohama, Kanagawa 230-0045 Japan Find this author on Google Scholar Find this author on PubMed Search for this author on this site Marco Gaviraghi 4 Human Technopole , Milan, Italy Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Marco Gaviraghi Rodrigo Pracana 4 Human Technopole , Milan, Italy Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Rodrigo Pracana Johan Lord 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Roberto Ballarino 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Roberto Ballarino Anna Falk 5 Department of Neuroscience, Karolinska Institutet , Stockholm, Sweden 6 Lund Stem Cell Center, Department of Experimental Medical Science, Science for Life Laboratory, Lund University , Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Anna Falk Jay W. Shin 3 RIKEN Center for Integrative Medical Sciences , Yokohama, Kanagawa 230-0045 Japan 7 Genome Institute of Singapore , A*STAR Singapore 138672, Singapore Find this author on Google Scholar Find this author on PubMed Search for this author on this site Takeya Kasukawa 3 RIKEN Center for Integrative Medical Sciences , Yokohama, Kanagawa 230-0045 Japan Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Takeya Kasukawa Chi Wai Yip 3 RIKEN Center for Integrative Medical Sciences , Yokohama, Kanagawa 230-0045 Japan Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Chi Wai Yip Masaki Kato 3 RIKEN Center for Integrative Medical Sciences , Yokohama, Kanagawa 230-0045 Japan Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Masaki Kato Hazuki Takahashi 3 RIKEN Center for Integrative Medical Sciences , Yokohama, Kanagawa 230-0045 Japan Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Hazuki Takahashi Nicola Crosetto 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden 4 Human Technopole , Milan, Italy Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Nicola Crosetto Piero Carninci 3 RIKEN Center for Integrative Medical Sciences , Yokohama, Kanagawa 230-0045 Japan 4 Human Technopole , Milan, Italy Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Piero Carninci For correspondence: piero.carninci{at}fht.org magda.bienko{at}ki.se Magda Bienko 1 Department of Microbiology , Tumor and Cell Biology, Karolinska Institutet , Stockholm, Sweden 2 Science for Life Laboratory , Stockholm, Sweden 4 Human Technopole , Milan, Italy Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Magda Bienko For correspondence: piero.carninci{at}fht.org magda.bienko{at}ki.se Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract Cell differentiation towards neurons is accompanied by widespread changes in three-dimensional (3D) genome organization and gene expression. Nuclear RNAs (nucRNAs) have been proposed to be important regulators of such changes; however, the type, abundance, and functions of nucRNAs during neurodifferentiation remain largely unexplored. Here, we integrate multi-omic data generated in the frame of the Functional ANnoTation Of the Mammalian genome (FANTOM6) to chart 3D genome, RNA-DNA contactome, and transcriptome changes during in vitro differentiation of human induced pluripotent stem cells to neural stem cells and neurons. We find that most RNA-DNA contacts form between transcripts and their source gene; however, a group of intronic RNAs engages in chromatin contacts with distal loci on the same or on different chromosomes. We detect such trans-contacting intronic RNAs (TIRs) in all cell types profiled by FANTOM6, but most prominently in neurons, where TIRs are produced from highly expressed, neuron-specific, long (mean length: 400 kilobases, kb) protein-coding genes. In neurons, TIRs accumulate in the nucleus forming large clouds around their source loci, and they occasionally spread across the nucleus. TIRs engage in local and distal contacts with a set of genomic regions (named TIR-contacted regions or TIRCs) carrying much shorter (mean length: 30 kb) neuron-specific genes that become upregulated and move towards the nuclear center during neurodifferentiation, forming high-connectivity hubs. TIR source genes, and especially their introns, are strongly enriched in risk loci for neurodevelopmental disorders (NDDs). Notably, knockdown of a single TIR by antisense oligonucleotides leads to downregulation of multiple genes implicated in NDDs and of source genes of most other TIRs. We propose that TIRs orchestrate cell type-specific gene expression during neurodifferentiation and might be pathogenically linked to NDDs. Introduction The three-dimensional (3D) spatial organization of the genome in the nucleus provides a multi-scale framework for spatio-temporal control of key genomic functions such as DNA replication, transcription, and repair 1 . Substantial 3D genome re-organization occurs as cells differentiate during embryonic development and in adulthood. While pluripotent stem cells harbor an open 3D genome, permissive to low-level genome-wide transcription, differentiation encompasses global reduction of chromatin plasticity and progressive multi-scale re-organization of the 3D genome as cells exit pluripotency and undergo lineage commitment 2 , 3 . This multi-scale rewiring is particularly evident during brain development; for example, during neurodifferentiation in vitro and in vivo, chromatin loops undergo cell type-specific remodeling and the number of topologically associating domains (TAD) boundaries decreases 4 , 5 . At the same time, novel TAD boundaries form at promoters of developmentally upregulated genes and, contextually, the contact frequency between the transcription start site and the gene body of several activated long neuronal genes increases 5 – 7 . 3D genome changes occurring during neurodifferentiation have been linked with cognitive dysfunction, neuropsychiatric diseases, and neurodevelopmental disorders (NDDs) including autism spectrum disorder (ASD) and schizophrenia 3 , 8 – 12 . For example, chromatin loops that are acquired during human cortex development have been linked to risk loci for ASD identified in genome-wide association studies (GWAS) 6 . In other studies, human neural progenitor cells were shown to gain connectivity between loci harboring common risk variants associated with schizophrenia and bipolar disorder, while differentiating to neurons 13 , 14 . Intriguingly, risk variants show preference for different genomic elements and brain cell types; for instance, variants associated with schizophrenia and bipolar disorder were predominantly found in enhancers and promoters active in neurons 13 , 14 , although the two diseases have distinct etiologies 11 . Interestingly, risk loci for schizophrenia are enriched in genes that regulate neuronal connectivity and chromatin remodeling and that are part of complex ‘connectomes’ involving DNA, RNAs, and proteins and display highly coordinated gene expression in neuronal cells 13 . A growing body of evidence indicates that nucRNAs play important roles in shaping the 3D genome and in the regulation of genomic functions. At nuclear scale, nucRNAs exert an important structural role, independent of their sequence, by competing with DNA for binding to histone tails to reduce the electrostatic compaction of DNA 15 . nucRNAs also act as scaffolds of chromosome territories (CTs) 16 by binding to different types of RNA-binding proteins (RBPs) and forming an insoluble mesh that favors the formation of transcriptionally active compartments within CTs 17 , 18 . Furthermore, recent evidence suggests that nucRNAs might shape the 3D genome by spatially segregating active (A) and inactive (B) chromatin compartments 19 . Besides playing a structural role, nucRNAs also seem important for regulating genomic functions such as gene expression regulation, RNA processing, and heterochromatin assembly, by forming biomolecular condensates through phase separation 20 . For example, repeat-rich, long nascent RNAs as well as long non-coding RNAs (lncRNAs) can trigger phase separation through a combination of electrostatic forces and by acting as potent scaffolds for the condensation of RBPs 20 . Although the contribution of specific nucRNAs to shaping the 3D genome has been investigated, we currently lack a genome-wide perspective on how contacts between DNA and nucRNAs influence both genome structure and functions remains. To fill in this knowledge gap, here we perform multi-omic profiling of human induced pluripotent stem cells differentiated to neural stem cells and cortical neurons and apply integrative genomic analyses to map RNA-DNA contactome, 3D genome, and transcriptome changes during neurodifferentiation, within the sixth edition of the Functional ANnoTation Of the Mammalian genome (FANTOM6) project. We identify a group of trans-contacting intronic RNAs (TIRs) that form large nuclear ‘clouds’ that resemble molecular condensates, which are strongly enriched in risk loci for NDDs. We first comprehensively characterize the TIR source loci and TIR contact regions together with their chromatin neighborhood, and then functionally assess the most abundant TIR-producing gene in neurons ( NALF1 ) by leveraging antisense oligonucleotide technology. Our study provides the first comprehensive characterization of the RNA-DNA contactome during human neurodifferentiation and identifies TIRs as a novel class of 3D genome shapers and potential pathogenic targets. Results Multi-omic profiling of human iPSC in vitro differentiated to cortical neurons One of the goals of the FANTOM6 consortium is to unravel the interplay between coding and non-coding RNA and genome structure-function during cell differentiation. As part of this effort, we performed multi-omic profiling of human induced pluripotent stem cells (iPSC), iPSC-derived neural stem cells (NSC), and cortical neurons (NEU), using the following assays: RNA and DNA Interacting Complexes Ligated and sequenced (RADICL-seq) 21 to profile RNA-DNA contacts; high-throughput chromosome conformation capture (Hi-C) 22 to measure DNA-DNA contacts; Genomic Loci Positioning by Sequencing (GPSeq) 23 to map radial positions in the nucleus, genome wide (using a matched model system, see Methods ); pseudo-bulk single-cell Assay for Transposase-Accessible Chromatin using sequencing (scATAC-seq) 24 for measuring chromatin accessibility; total RNA sequencing (RNA-seq); and Cap Analysis of Gene Expression (CAGE) 25 for promoter identification and full-length transcript profiling ( Fig. 1a and Supplementary Table 1 ). Download figure Open in new tab Figure 1. Long-range RNA-DNA contacts mediated by RNAs produced from protein-coding genes form during neuronal lineage specification. ( a ) Scheme of the model system (left box) and datasets generated from them (right box) within the sixth edition of the Functional ANnoTation Of the Mammalian genome project (FANTOM6), which were analyzed in this study. ( b ) Proportion of protein-coding genes producing RNAs engaging in RNA-DNA contacts as detected by RADICL-seq, for different genomic distances between the RNA source locus and the contacted loci. n , number of protein-coding genes. ( c ) Representative cis and trans RNA contact profiles (trans defined as any contact ≥ 5 Mb from the source gene). Red segments mark the position of the RNA source gene. RPM, reads per million. ( d-f ) RNA-DNA contact matrices (100 kb resolution) for iPSC (d), NES (e) and NEU (f). Only trans contacts are shown. ( g ) Top track: profile of inter-chromosomal DNA-DNA contacts measured by Hi-C along chromosome (chr) 10 (top track) using the NALF1 gene locus on chr13 as viewpoint. Middle track: trans RNA-DNA contacts of NALF1 RNA detected by RADICL-seq along chr10. Bottom track: A/B chromatin compartments identified based on Hi-C data along chr10. For all the tracks the resolution is 100 kb. ICE, iterative correction and eigenvector decomposition. RPM, reads per million. HEV, first Hi-C eigenvector. ( h-j ) Relationship between the length (in base pairs) and expression level of all human protein-coding genes, in iPSC (h), NSC (i) and NEU (j). Each dot represents a gene. Dots are colored based on whether the produced RNA was detected by RADICL-seq and on the farthest distance between the source gene and the contacted DNA loci. rlog, regularized log transform calculated by DESeq2. ( k ) Venn diagram showing the number ( n ) of protein-coding genes producing trans-contacting RNAs in the three cell types profiled by RADICL-seq. ( l ) Gene expression levels measured by RNA-seq and ( m ) significant gene ontology terms, for cell type-specific and common protein-coding gene sets producing trans-contacting RNA. Each column in the RNA-seq heatmap panel represents one gene. Gene expression was scaled gene-by-gene across the three differentiation stages. ( n ) Proportion of trans contacts mediated by RNAs derived from protein-coding genes, for different segments of the gene body to which the RNA sequence aligns. UTR, untranslated region, either at the 5′ or 3′ end of the gene. CDS, coding DNA sequence. n , number of trans RNA-DNA contacts detected by RADICL-seq. ( o ) Ratio between intronic and exonic RNA coverage for RNAs that are involved in RNA-DNA contacts in trans or cis and are expressed at comparable levels in NEU, as calculated from RNA-seq (left) or RADICL-seq (right) data. n , number of genes. Violins extend from minimum to maximum, boxplots extend from the 25 th to the 75 th percentile, white dots represent the median, whiskers extend from –1.5×IQR to +1.5×IQR from the closest quartile. IQR, inter-quartile range. A link to the Source Data and code to regenerate the plots displayed in this figure is provided in the Data Availability and Code Availability statements. We generated libraries from two biological replicates per cell type (iPSC, NSC, NEU) and assay (except for scATAC-seq, 1 replicate), and sequenced them on Illumina platforms ( Supplementary Table 1 and Methods ). We analyzed RADICL-seq, GPSeq, and CAGE data using custom pipelines, while we leveraged state-of-the-art pipelines for Hi-C, RNA-seq, and scATAC-seq ( Computational Methods ). We confirmed that biological replicates were highly correlated ( Supplementary Fig. 1a-e ); hence we merged them for all subsequent analyses. To confirm that iPSC were successfully differentiated to cortical neurons, we identified differentially expressed genes (DEGs) and performed gene ontology (GO) analysis on the RNA-seq datasets ( Computational Methods ). Among genes upregulated in NEU compared to iPSC, we found many genes involved in synapse organization, regulation of neurotransmitter transport, and other neuron-specific processes ( Supplementary Fig. 1f ). Conversely, genes downregulated in NEU compared to iPSC included genes involved in cell division and chromatin organization ( Supplementary Fig. 1g ), as expected, since cells exit the cell cycle during differentiation to mature neurons. These results confirmed the validity of our model system, giving us clearance to proceed to in-depth analyses of all the other datasets. RADICL-seq uncovers multiple RNAs that contact DNA far from their locus of origin We first analyzed RADICL-seq data by binning the identified RNA-DNA contacts in 100 kilobase (kb) genomic windows (bins) and generating RNA-DNA contact matrices ( Computational Methods ). We classified each RNA-DNA contact as cis or trans, depending on whether the RNA fragment contacted a DNA locus within 5 megabase (Mb) (on either side) from the RNA-transcribing locus (cis contacts), or farther along the same chromosomes or on other chromosomes (collectively named ‘trans contacts’). We chose 5 Mb as threshold to exceed the typical size of TADs (1–2 Mb) and thus avoid calling trans contacts within the TAD of the RNA-producing locus (hereafter, ‘source’). In all three cell types, ∼99% of all RADICL-seq reads originated from RNAs transcribed from protein-coding (pc) genes or from lncRNAs ( Supplementary Fig. 2a-j ). Most RNA-DNA contacts (excluding contacts involving the lncRNA MALAT1 ) occurred in cis, in the immediate vicinity of the source ( Fig. 1b, c ). Among trans contacts, more than 60% involved MALAT1 , which formed genome-wide contacts in all three cell types ( Supplementary Fig. 2f, h, j ). A few other lncRNAs also formed genome-wide contacts, predominantly in iPSC, including RNVU1-7 , GAS5, U8 , RMRP and SNHG14 ( Supplementary Fig. 2f, h, j ). The majority of the remaining trans contacts involved RNA transcribed from pc genes ( Fig. 1d-f and Supplementary Fig. 2e, g, i ). Whereas only a small fraction of RNAs transcribed from pc genes engaged in trans contacts in iPSC (1.4%, n =205), the proportion significantly increased in NSC (4.5%, n =646) and in NEU (5.5%, n =790). Among pc genes producing trans-contacting RNAs in NEU, we identified dozens of genes selectively expressed in the brain, including NALF1 , LSAMP , CNTNAP2 , PCDH9 , and NRG3 ( Supplementary Fig. 3a-e ). Notably, trans contacts did not simply recapitulate long-range Hi-C contacts beyond 5 Mb but rather displayed a unique pattern with more evident peaks and valleys compared to Hi-C profiles, and trans contacts were enriched in the A chromatin compartment ( Fig. 1g and Computational Methods ). Trans-contacting RNAs originate from highly expressed, cell type-specific, long pc genes We then focused on characterizing trans-contacting RNAs originating from pc genes. In all three cell types, these were transcribed from long and highly expressed genes with cell type-specific expression and function, as revealed by DEG and GO analyses ( Fig. 1h-m , Supplementary Fig. 3f-i, Supplementary Fig. 4a ). Other cell types profiled by RADICL-seq within FANTOM6 could also be separated in distinct clusters based on their repertoire of trans-contacting RNAs ( Supplementary Fig. 4b, c ). In all cell types analyzed, the source genes of trans-contacting RNAs were significantly longer and more highly expressed than pc genes, on average ( Supplementary Fig. 4d, e ). However, only neurons displayed a widespread pattern of trans contacts ( Supplementary Fig. 4f ). Notably, pc genes with comparably high expression levels, but not producing trans-contacting RNAs, were associated with housekeeping functions based on GO analysis ( Supplementary Fig. 5a-e ). These results indicate that, in all the cell types that we profiled by RADICL-seq—but most prominently in NEU—a set of cell type-specific, highly expressed, and longer-than-average pc genes generate RNAs contacting multiple regions on different chromosomes. Trans-contacting RNAs are mainly composed of intronic sequences To further investigate the nature of the identified trans-contacting RNAs, we annotated their sequence from RADICL-seq data and found that more than 95% of the RADICL-seq reads from NSC and NEU aligned to introns ( Fig. 1n ). To check whether these reads originate from spliced-out introns or from intronic segments still embedded in the nascent RNA or pre-mRNA, we aligned all the RADICL-seq and RNA-seq reads to the corresponding source genes and computed the intron-over-exon coverage ratio ( Supplementary Fig. 6a-p and Computational Methods ). Despite the presence of many intronic reads in our RNA-seq data (in line with previous human brain RNA-seq data 26 ), the coverage was clearly skewed towards exon. In contrast, for RADICL-seq the coverage was slightly higher for intronic sequences ( Fig. 1o ). In both RADICL-seq and RNA-seq tracks, we observed a decay of the RNA coverage from the 5′ end of many intronic RNA segments towards their 3′ end ( Supplementary Fig. 6a-p ). This pattern—which likely reflects co-transcriptional splicing as previously reported 26 —was consistently observed along almost all the introns of the source genes of the trans-contacting RNAs, whereas the same pattern was not seen in a set of control pc genes expressed at similar levels in NEU but not producing trans-contacting RNAs (hereafter, ‘TIR control genes’) ( Supplementary Fig. 6q-t ). Importantly, we did not observe strong enrichment of CAGE peaks in the introns of trans-contacting RNA source genes ( Supplementary Fig. 6r, s ), indicating that this pattern is unlikely caused by alternative transcription start sites. For four of the top-10 trans-contacting RNAs identified by RADICL-seq in NEU (i.e., RNAs displaying the most abundant and widespread trans contacts), RNA-seq revealed alternative isoform usage during development, with shorter isoforms typically expressed in iPSC and full-length isoforms detected only in NSC and NEU ( Supplementary Fig. 7 ). In sum, these results indicate that trans-contacting RNAs are mainly composed of intronic sequences, likely including both spliced-out introns and intronic sequences in nascent transcripts. Therefore, from now on we refer to them as trans-contacting intronic RNAs or TIRs. TIRs form large clouds around their source gene in neurons To further explore whether TIRs represent spliced-out introns and visualize their spatial distribution in single cells, we performed RNA fluorescence in situ hybridization (RNA FISH) leveraging an updated version of the iFISH pipeline that we previously developed 27 . We designed RNA FISH probes targeting either intronic or exonic regions of 11 out of the top-55 TIRs displaying the most abundant and widespread trans contacts in NEU, as measured by RADICL-seq ( Supplementary Fig. 8a, b, Supplementary Table 2, and Methods ). As control, we designed RNA FISH probes targeting 11 pc genes with comparable expression levels but whose transcripts did not engage in trans contacts ( Supplementary Table 2 ). When probing for TIRs in NEU, we observed many RNA FISH signals (fluorescence dots) in the nucleus, typically forming 1–4 large clusters (‘clouds’) per nucleus, most likely around the source loci ( Fig. 2a , b , Supplementary Fig. 9a-d, Supplementary Fig. 10a, b, Supplementary Fig 11a-c, and Supplementary Fig. 12a-e ). In contrast, RNA FISH with probes targeting the intronic sequence of control genes typically yielded 2–3 single FISH dots per nucleus, as expected ( Supplementary Fig. 13a-e ). Furthermore, RNA FISH with probes targeting the exonic sequence of the same TIR source or TIR control genes yielded numerous dots predominantly localized outside of the nucleus, representing mature RNAs ( Fig. 2c and Supplementary Fig. 9-13 ). Notably, TIRs specific to NEU did not form clouds in NSC, despite some of the TIR source genes being expressed in NSC ( Supplementary Fig. 14a-c ). For some of the TIRs analyzed, in addition to the clouds, we observed multiple intronic FISH dots scattered throughout the nucleus, potentially representing spliced-out intronic fragments ( Fig. 2b , c , and Supplementary Fig. 9-12 ). Indeed, when staining for both exons and introns of the same source gene, we observed almost no colocalization of the sparse intronic dots with exonic signals ( Fig. 2c , and Supplementary Fig. 9-12 ), suggesting that the sparse intronic dots correspond to spliced-out introns. Of note, when staining for two different introns of the same TIR source gene, we observed substantial co-localization within the clouds, while the sparse dots did not seem to colocalize ( Fig. 2d and Supplementary Fig. 9-12 ). Download figure Open in new tab Figure 2. Trans-contacting Intronic RNAs (TIRs) form large clouds around the source locus and re-organize the local chromatin architecture. ( a ) Scheme of RNA FISH probes (blue rectangles) targeting the indicated intronic or exonic regions of the NALF1 gene. ( b ) Maximum intensity projection of a z-stack widefield microscopy image exemplifying the nuclear pattern of TIRs originating from NALF1 intron-3 (green), as detected by RNA FISH using probes 1 and 2 shown in (a) labeled with the same color. Gray, DNA stained by Hoechst 33342. Scale bar, 5 μm. ( c ) As in (b) showing NALF1 exons and introns detected by the indicated RNA FISH probes. ( d ) As in (b) using two RNA FISH probes labeled in different colors. ( e ) Correlation between the number of discrete intronic RNA FISH signals (dots) per cell and the total number of DNA contacts formed by the RNA produced by the same gene in NEU, as measured by RADICL-seq, for different TIR source and TIR control genes. Dashed black line: linear regression. Yellow band, 90% confidence interval. PCC, Pearson’s correlation coefficient. ( f ) Quantification of the number of intronic and exonic FISH dots per cell, for both TIR source and TIR control genes, in NEU cells. In (e) and (f), FISH dot counts represent averages calculated from 335–943 cells across 8–12 fields of view (FOV) per condition, and from 1–2 biological replicates for each gene. All images in (c) and (h) were deconvolved using Deconwolf ( https://deconwolf.fht.org/ ). A link to the Source Data and code to regenerate the plots displayed in this figure is provided in the Data Availability and Code Availability statements. To gain further insight into the observed intronic RNA clouds, we performed quantitative analyses of our FISH data ( Methods ). For the top TIRs visualized by FISH, we found a strong positive correlation (Pearson’s correlation coefficient, PCC: 0.8) between the number of intronic RNA FISH signals per cell and the corresponding number of RADICL-seq contacts ( Fig. 2e ). Of note, the numbers of intronic and exonic FISH dots per cell were not correlated among TIR source genes (PCC: –0.015) ( Fig. 2f ). These results demonstrate that, in neurons, TIRs typically form large clouds around their source genes, but they can also diffuse away from the source locus, appearing as FISH signals scattered throughout the nucleus. The intronic RNA clouds possibly represent transcription or splicing intermediates, including nascent and partially spliced RNA as well as fully spliced-out introns, whereas the sparse intronic FISH dots are most likely spliced-out introns. TIR clouds do not overlap with splicing speckles Since the TIR source genes represent the most highly expressed genes in neurons, composed of long introns that require heavy splicing including alternative splicing to create isoform variation, we hypothesized that the observed TIR clouds might overlap with nuclear speckles, which represent the main splicing hubs in the nucleus 28 , 29 . To test this, we first mined our RADICL-seq data for contacts between the lncRNA MALAT1 —a speckle marker 30 , 31 —and TIR source genes. We computed a speckle proximity score based on the MALAT1 RNA contact frequency calculated from RADICL-seq data ( Methods ). TIR source genes were marked by low score values ( Fig. 3a , b and Supplementary Fig. 15 ), suggesting they do not frequently associate with speckles. To corroborate this finding, we first performed dual RNA FISH targeting MALAT1 lncRNA and selected TIRs identified in NEU, which showed absence of immediate co-localization between MALAT1 and TIR clouds ( Fig. 3c , d and Supplementary Fig. 16a ). Notably, we observed that different TIR clouds did not colocalize but rather formed individual clouds ( Fig. 3d and Supplementary Fig. 16b ). Furthermore, when analyzing the radial localization of TIR source genes using our GPSeq data, we found that many TIR source genes were located in the periphery of the nucleus in both NSC and NEU, even though in the latter TIR source genes shifted slightly inwards ( Fig. 3e-g and Supplementary Fig. 17a, and Methods ). Interestingly, the GPSeq score varied along the gene body of the top TIRs, with some parts of the gene body appearing more central in NEU compared to NSC ( Supplementary Fig. 17b-d ). Consistently, the GPSeq score range was significantly broader in NEU compared to NSC ( Supplementary Fig. 17e ). Altogether, these results demonstrate that the top TIR source genes (i) form large intronic RNA clouds occupying discreet and separate volumes in the nucleus; (ii) are not typically engaged with nuclear speckles; (iii) occupy mostly peripheral locations in the nucleus in NSC; and (iv) tend to move inwards in the nucleus during neurodifferentiation. These observations are consistent with prior reports according to which several highly expressed genes with long introns and low GC-content do not associate with speckles 32 , 33 . Download figure Open in new tab Figure 3. TIR source genes do not associate with nuclear speckles and reorganize during neurodifferentiation. ( a ) Profiles of RNA-DNA contacts (100 kb resolution) detected by RADICL-seq along chromosome (chr) 7, for MALAT1 (top track, showing trans contacts only) and for 5 TIRs along chromosome (chr) 7 (bottom track, showing cis contacts only) from the top-55 TIRs identified in NEU. ( b ) Distribution of the nuclear speckle proximity score inferred from the normalized (based on reads per million) MALAT1 trans contacts for the top-55 TIR source genes and an equal number of TIR control genes. ( c ) Maximum intensity projection of a z-stack widefield microscopy image exemplifying the pattern of MALAT1 RNA (marking nuclear speckles, in magenta) and of TIRs (green) originating from NALF1 intron-3, in one NEU cell nucleus. Gray, DNA stained by Hoechst 33342. Scale bar, 5 μm. ( d ) As in (c) showing TIRs originating from different TIR source genes and introns. Scale bars, 5 μm. ( e, f ) ‘Pizza plots’ showing the radial placement (100 kb resolution) of the top-55 TIR source genes in NSC (f) and NEU (g), as measured by GPSeq. The outermost and innermost dashed circle of each pizza plot represent the center and periphery of the cell nucleus, respectively. Dashed circles from center to periphery represent GPSeq deciles. Radii separate individual chromosomes. Each dot represents a 100 kb bin. Dots occupying the same radial position within the same chromosomal sector are jittered for visualization. Four representative TIR source genes are labeled in red. ( g ) Distribution of the GPSeq score percentile (100 kb resolution) of the top-55 TIR source genes, in either NES or NEU. P, paired t-test, two-tailed. ( h ) As in (c) showing TIRs originating from NALF1 intron-3 (green) together with the corresponding chromosome territory (CT) (chr13 in red) as well as an unrelated CT (chr12 in yellow), in four representative nuclei. Scale bars: 5 μm. All the images in (c) and (h) were deconvolved using Deconwolf ( https://deconwolf.fht.org/ ). For a scheme of the RNA FISH probes used in (c, d) and (h), refer to Fig. 2a . In (b) and (g), violins extend from minimum to maximum, boxplots extend from the 25 th to the 75 th percentile, white dots represent the median, whiskers extend from –1.5×IQR to +1.5×IQR from the closest quartile. A link to the Source Data and code to regenerate the plots displayed in this figure is provided in the Data Availability and Code Availability statements. TIR source genes undergo local structure reorganization during neurodifferentiation Prompted by these observations, we then sought to further characterize the chromatin neighborhood of TIR clouds. Given their confined shape, we wondered whether TIR clouds overlap with the corresponding chromosome territory (CT) and decorate it similarly to what XIST RNA does during chromosome X inactivation, or whether they extend beyond the territory. To explore this, we performed DNA-RNA FISH using probes against NALF1 intronic RNA, combined with chromosome spotting probes 27 targeting chr13 harboring the NALF1 locus, as well as unrelated chr12 for comparison ( Supplementary Table 2 and Methods ). NALF1 TIR clouds most often extended beyond the chr13 CT and only minimally overlapped with its territory ( Fig. 3h and Supplementary Fig. 18 ). In some nuclei, we found that the intron cloud was located substantially away from the bulk of chr13 CT and, interestingly, some of the sparse NALF1 TIR signals occasionally colocalized with chr12 ( Fig. 3h and Supplementary Fig. 18 ). To further assess the chromatin environment of TIR clouds, we examined Hi-C maps from iPSC, NSC, and NEU and assessed the chromatin contacts of TIR genes across differentiation ( Methods ). We observed that TIR source genes (but not TIR control genes) lost a substantial portion of their contacts with DNA in a 30 Mb region around the locus during differentiation ( Supplementary Fig. 19a-d and Supplementary Fig. 20a-h ). Moreover, the Hi-C eigenvector values of TIR source genes tended to increase ( Supplementary Fig. 20i, j ), indicating either an enhanced connectivity with the A compartment, a shift from B to A, or a loss of connectivity with the surrounding B compartment. We then wondered whether TIR source genes are subjected to the ‘gene melting’ phenomenon previously described based on Immuno-genome Architecture Mapping (Immuno-GAM) on neuronal cells 34 . This comprises a strong depletion of cis contacts along the gene body of long neuronally expressed genes and is thought to reflect the adoption of an extended, open configuration upon activation of these genes in neuronal cells 34 . However, in our differentiation model system, we observed well-defined TAD structures around TIR source loci, which became even more pronounced during differentiation to neurons ( Supplementary Fig. 19e, f and Supplementary Fig. 21a-f ). These findings are consistent with previous reports 5 – 7 but exclude a strong gene melting effect in our model system. When comparing RADICL-seq and Hi-C contact maps, we found that intra-chromosomal Hi-C profiles and RADICL-seq profiles were similar in the vicinity of many TIR source genes, while RADICL-seq profiles of TIR control genes were mostly restricted to their local TAD ( Supplementary Fig. 19g-k and Supplementary Fig. 22a-f ). To quantitatively compare our data, we calculated neuronal RADICL-seq residues and correlated those with the Hi-C observed-over-expected (O/E) values for different cell types ( Methods ). As expected, in NEU, RADICL-seq residues were mostly correlated with Hi-C O/E values from the same cell type ( Supplementary Fig. 19l and Supplementary Fig. 22a, b ), indicating that the chromatin structure in the vicinity of TIR source genes demarcates the RNA-DNA connectivity detected by RADICL-seq at those loci. Interestingly, for many TIR source genes, we found that RNA-DNA contacts extended beyond multiple TAD distances, which was not the case for TIR control genes ( Supplementary Fig. 19m ). Altogether, these results indicate that the formation of intronic RNA clouds around TIR source genes is accompanied by a profound re-organization of both the local chromatin topology and the long-range contactome of those genes, as they become highly expressed during neurodifferentiation. We speculate that the long introns of TIR source genes might play a role in this process. TIRs contact short neuronally expressed genes across the genome Having thoroughly characterized TIR source genes, we then turned our attention to the DNA loci contacted by TIRs. Firstly, we found the genome-wide contact patterns of the top-10 TIRs to be overall fairly similar (PCC: 0.54–0.89 at 1 Mb; 0.29–0.75 at 100 kb), forming broad peaks common to all the top TIRs ( Fig. 4a and Supplementary Fig. 23a-c ). This indicates that different TIRs contact common genomic regions. Accordingly, we found that the total number of RADICL-seq contacts per genomic bin formed collectively by all the TIRs identified correlated strongly with the number of distinct TIRs (produced by distinct source genes) contacting each bin (PCC: 0.93 at 100 kb). We then specifically focused on regions contacted by many different TIRs and identified 557 bins (100 kb) contacted by at least 16 different TIRs in NEU ( Fig. 4a and Computational Methods ). Hereafter, we refer to these regions as TIR contact regions or TIRCs. In NEU, TIRCs were strongly enriched in GO terms for genes related to neuronal functions ( Fig. 4b ). Notably, the enrichment was higher for TIRCs than for MALAT1 contact regions in NEU or for genes within the A compartment ( Supplementary Fig. 23d, e ). Among genes related to neuronal functions, TIRCs contained genes specifically upregulated in NEU, similarly to TIR source genes ( Fig. 4c ). In contrast, housekeeping genes within TIRCs were not upregulated during differentiation ( Fig. 4c ). However, compared to TIR source genes, neuronally expressed TIRC genes were exceptionally short ( Fig. 4d ). These results indicate that, in neurons, TIRs tend to engage with multiple shared regions along the genome that are enriched in short active genes associated with neuronal functions. Download figure Open in new tab Figure 4. TIR contact regions (TIRCs) form hubs that spatially rearrange during differentiation. ( a ). Top track (green): binned RADICL-seq trans contact counts (100 kb resolution) along chromosome (chr) 1 and 12, for the top-10 TIR source genes identified in NEU, whose individual tracks are visualized in gray below. RPM, reads per million. Orange boxes, examples of common TIR contact regions (TIRCs). ( b ) Enriched gene ontology (GO) terms in biological processes, based on expressed genes ( n ) inside TIRCs. ( c ) Distribution of the fold change (FC) between NEU and iPSC in the expression of neuronal genes (left) and housekeeping genes (right) within TIRCs or TIRC control set 1. n , number of genes. ( d ) Distribution of the length of the top-55 TIR source genes shown in (a), expressed protein-coding (pc) genes within TIRCs, and human pc genes. n , number of genes. ( e-g ) ‘Pizza plots’ showing the number of TIRs (distinct source genes) contacting 100 kb genomic bins radially placed based on GPSeq data, for iPSC (e), NSC (f) and NEU (g). ( h, i ) Pizza plots showing the radial placement of all 567 TIRCs in NSC and NEU, as assessed by GPSeq. ( j) Distribution of the GPSeq percentile of TIRCs in NSC and NEU. P, Welch’s two-sample, two-tailed t-test. ( k, l ) As in (h, i) for TIRC control set 1. ( m ) As in (j) for TIRC control set 1. ( n ) Enriched GO terms in biological processes, based on the expressed genes ( n ) within TIRC control set 1. In all the pizza plots in the figure, the outermost and innermost dashed circle of each pizza plot represent the center and periphery of the cell nucleus, respectively. Dashed circles from center to periphery represent GPSeq deciles. Radii separate individual chromosomes. Each dot represents a 100 kb bin. Dots occupying the same radial position within the same chromosomal sector are jittered for visualization. In all the violin plots in the figure, violins extend from minimum to maximum, boxplots extend from the 25 th to the 75 th percentile, white dots represent the median, whiskers extend from –1.5×IQR to +1.5×IQR from the closest quartile. IQR, inter-quartile range. A link to the Source Data and code to regenerate the plots displayed in this figure is provided in the Data Availability and Code Availability statements. TIRCs move towards the nuclear center during differentiation to neurons Next, we sought to understand where RNA trans contacts form in the nucleus. To this end, we leveraged our GPSeq data and assessed how the genome is radially reorganized during neurodifferentiation, first at genome-wide level and then focusing on TIRCs. On a global scale, we found that the correlation between the GPSeq score and the GC-content distribution along the genome increased from NSC to NEU ( Supplementary Fig. 24a-d and Methods ). This indicates that many genomic regions change their radial position as cells differentiate to neurons, leading to the formation of a steeper GC-gradient from the nuclear periphery inwards. Indeed, we identified multiple regions that moved towards the center during the transition from NSC to NEU, as well as regions that showed the opposite trend ( Supplementary Fig. 24e ). GO term analysis of the inward-moving regions revealed enrichment in genes specifying neuronal functions, whereas regions moving outwards were enriched in genes associated with chromatin organization and gene silencing ( Supplementary Fig. 24f, g ). These radial changes were reflected in gene expression changes: genes within regions moving inwards were upregulated in NEU compared to NSC, whereas genes in regions that moved towards the nuclear periphery were slightly downregulated ( Supplementary Fig. 24h ). Notably, the nuclear periphery was enriched in genes involved in synapse organization, the mid radial layers were enriched in genes involved in RNA metabolism, while the innermost portion of the nucleus was enriched in genes involved in protein metabolism ( Supplementary Fig. 25a-f ). Next, we focused on the radial arrangement of TIRCs in the nucleus and observed that RADICL-seq contacts involving TIRs were enriched in the most central DNA loci in NEU ( Fig. 4e-g ). Contrary to our expectation, TIRCs did not all occupy the most central radial location in NSC despite their GC-rich sequences ( Fig. 4h and Supplementary Fig. 26a ). Instead, only upon differentiation to neurons, more TIRCs moved inwards to occupy more central positions, as expected based on their GC-content ( Fig. 4i , j ). In contrast, a control set of DNA loci sharing the same radial location as TIRCs in NSC (TIRC control 1) moved outwards during differentiation ( Fig. 4k-m ). GO-term analysis of these TIRC control regions revealed that they were enriched in genes involved in cell cycle progression and histone demethylation ( Fig. 4n ), in line with their downregulation in NEU ( Supplementary Fig. 26b ). Of note, genes within regions with similarly high GC-content as TIRCs (TIRC control 2) occupied the most central locations in both NSC and NEU, and their expression did not change during differentiation ( Supplementary Fig. 26b, c ). These results indicate that TIRCs are enriched in short neuron-specific genes that become activated and move towards the nuclear center during neurodifferentiation. TIRCs colocalize with speckles and show increased connectivity in neurons We then further analyzed the identified TIRCs by comparing them with various genomic features ( Fig. 5a-c and Supplementary Fig. 27a ). Contrary to TIRs, the genome-wide profiles of the trans contacts formed by TIRs and MALAT1 in NEU were overall similar (Spearman’s correlation coefficient, SCC: 0.65), although the profile of MALAT1 trans contacts displayed less distinct peaks ( Fig. 5d and Supplementary Fig. 27b ). In NSC and even more in NEU, but not in iPSC, TIRCs had a significantly higher speckle proximity score (P value < 2.2 × 10 − 16 , Welch Two Sample t-test, two-tailed) compared to the TIR source genes and TIRC control regions, and TIRCs were enriched in the A compartment ( Fig. 5e-g , and Supplementary Fig. 27a ). To further investigate the relationship between TIRCs and speckles, we performed combined DNA-RNA FISH in NEU using RNA FISH probes targeting MALAT1 lncRNA (as proxy of speckles) and DNA FISH probes targeting 16 selected TIRCs ( Supplementary Table 2 ). We then measured the extent of co-localization between DNA and RNA FISH signals and found that TIRCs were slightly closer to speckles in comparison to the control regions ( Fig. 5h-j ). Of note, the DNA accessibility, as measured by scATAC-seq, was significantly lower inside TIRCs compared to TIRC control regions ( Supplementary Fig. 27c ). These findings indicate that TIR-TIRC contacts are not driven by a particularly elevated accessibility of TIRCs. Download figure Open in new tab Figure 5. TIRCs reside in active compartments and form high-connectivity hubs. ( a-c ) Pairwise correlation (Spearman’s correlation coefficient) between multiple genomic features (100 kb resolution) in iPSC (a), NSC (b), and NEU (c). PC1, Hi-C matrix principal component 1. Note that, for iPSC, a different GPSeq dataset derived from a different neurodifferentiation model system was used (see Experimental Methods ). ( d ) Chromosome-wide profiles of MALAT1 RNA trans contacts detected by RADICL-seq (top track); number of trans contacts made by any of the top-55 TIRs with each 100 kb genomic bin (middle track); and number of TIR source genes producing RNAs contacting that bin, for each 100 kb bin (bottom track). ( e-g ) Distribution of the speckle proximity score for the top-55 TIR source genes and TIRCs in NEU, and their respective control groups. ( h ) Maximum intensity projection of a z-stack widefield microscopy image exemplifying the spatial proximity between MALAT1 RNA (marking speckles) and selected TIRC regions in one representative NEU cell nucleus, as detected by RNA and DNA FISH, respectively. Gray, DNA stained by Hoechst 33342. Scale bar, 5 μm. ( i ) As in (h) for selected TIRC control genes. ( j ) Quantification of the spatial proximity between MALAT1 RNA and selected TIRC or TIRC control regions, as exemplified in (h, i). ( k ) Percentage of intra- and inter-chromosomal DNA-DNA contacts identified by Hi-C in iPSC, NSC, and NEU. n , total number of Hi-C contacts. ( l-n ) Observed-over-expected (O/E) ratio of DNA-DNA inter-chromosomal contacts (1 Mb resolution) between chr11 and 12 detected by Hi-C. The bar plots on the left and bottom of each Hi-C matrix show the total number of TIRs (from distinct source genes) that engage in trans contacts with each 100 kb genomic bin, for chr12 and 11, respectively. The TIR source genes from the top-55 TIRs list that are located on chr11 or 12 are shown in purple. ( o ) Mean Log2 of the O/E ratio of DNA-DNA inter-chromosomal contacts within all the members of each genomic region group shown. ( p ) Difference, between NEU and iPSC, in the mean O/E ratio of DNA-DNA inter-chromosomal contacts. ( q, r ) As in (n) and (m), respectively, for intra-chromosomal contacts. In all the violin plots in the figure, violins extend from minimum to maximum, boxplots extend from the 25 th to the 75 th percentile, white dots represent the median, whiskers extend from – 1.5×IQR to +1.5×IQR from the closest quartile. IQR, inter-quartile range. A link to the Source Data and code to regenerate the plots displayed in this figure is provided in the Data Availability and Code Availability statements. The 3D contacts of TIRC regions, as assessed by Hi-C, also changed dramatically during cell differentiation. Overall, the ratio between intra-and inter-chromosomal contacts substantially increased from iPSC to NEU ( Fig. 5k ). However, despite a global decrease of inter-chromosomal contacts in NEU, inter-chromosomal contacts between different TIRCs gradually intensified during differentiation ( Fig. 5l-p ). Moreover, TIRCs located on the same chromosome also showed increased inter-TIRC connectivity in NEU ( Fig. 5q , r ). Overall, these results indicate that, in neurons, TIRCs are part of a high-connectivity chromatin neighborhood in close spatial proximity to speckles. Expression of TIR source genes and TIRC genes strongly co-varies between single cells Having found that TIRCs are part of high-connectivity hubs in proximity of speckles, we hypothesized that this spatial configuration might favor co-regulation of the expression of genes contained within these regions, therefore resulting in higher-than-expected co-variation in their expression. To test this hypothesis, we leveraged a method for calculating gene expression co-variation from scRNA-seq data 35 ( Computational Methods ) and applied it to publically available scRNA-seq datasets from motor neurons derived from SOD1 A4V/+ iPSC 36 . The co-variation enrichment score (CES) calculated for TIRCs identified in NEU progressively increased with the number of shared TIR source genes, reaching the highest values for TIRCs contacted by the same set of 7–8 different TIRs ( Fig. 6a ). Moreover, although TIR source genes did not cluster strongly in Hi-C maps, their CES scores were as high as those of TIRC genes contacted by the same sets of 7 or more TIRs ( Fig. 6a , b ). The expression of TIR source genes as well as of genes contained within TIRCs also showed significantly higher CES values compared to TIRC control 1 and 2 regions ( Fig. 6b ). Altogether, these results indicate that the expression of both TIR source genes and TIRCs greatly co-vary across single neuronal cells. Download figure Open in new tab Figure 6. Functional characterization of TIRs and TIRCs. ( a ) Distribution of the gene covariation enrichment score (CES) for different sets of expressed genes located inside the genomic regions (100 kb resolution) co-contacted by RNAs produced from the indicated number of top-10 TIR source genes in NEU. n , number of genes. ( b ) As in (a) for covariation within or between the indicated gene groups. n , number of genes. ( c ) Fold change (FC) in gene expression (from RNA-seq) between NEU cells treated with a Gapmer Antisense Oligonucleotide (ASO) targeting NALF1 intron 3 or with siRNA targeting NALF1 mRNA, and cells treated with scrambled ASO or siRNA sequences (control). ( d, e ) Ratio of intron vs. exon per-base RNA-seq coverage across the NALF1 gene body in NEU cells treated with NALF1 ASO or siRNA or control cells. Rep, biological replicate. ( f ) Correlation between the expression levels (from RNA-seq) of all protein-coding genes in NEU cells treated with NALF1 ASO vs. control cells. ( g ) Significance of gene expression fold changes (FC) between NALF1 ASO and control cells. Dashed line, P value equal to –Log10(0.01). ( h, i ) As (f, g) for NALF1 siRNA. ( j ) Fold change (FC) in gene expression (from RNA-seq) between NEU cells treated with NALF1 ASO and control cells, for the indicated gene groups. n , number of genes. (k ) Length distribution of significantly differentially expressed genes (DEGs) (P value < 0.01, Wilcoxon rank-sum test) upon ASO treatment ( n = 2826) and in NEU compared to iPSCs ( n = 7641). ( l ) Disease ontology enrichment analysis of protein-coding genes ( n = 1558) downregulated upon ASO treatment. ( m ) Heritability enrichment (HE) of single-nucleotide polymorphisms (SNPs) identified across multiple genome-wide association studies (GWAS) grouped by trait analyzed (x-axis) within the annotation groups listed on the y-axis. Significant HE is shown (P < 0.00001). Traits are based on the S-LDSC results from 37 GWAS ( Supplementary Table 4 ). In all the violin plots in the figure, violins extend from minimum to maximum, boxplots extend from the 25 th to the 75 th percentile, white dots represent the median, whiskers extend from – 1.5×IQR to +1.5×IQR from the closest quartile. IQR, inter-quartile range. A link to the Source Data and code to regenerate the plots displayed in this figure is provided in the Data Availability and Code Availability statements. Silencing of one TIR has a strong effect on the expression of other TIRs The expression co-variation of TIR source genes and genes contained within TIRCs prompted us to further investigate the functional relevance of TIRs by reducing their levels using either antisense oligonucleotides (ASOs) or double-stranded silencing RNAs (siRNAs) ( Methods ). ASOs predominantly target RNA inside the nucleus, whereas siRNAs act on mRNA in the cytoplasm 37 . We reasoned that by comparing the effects of ASOs and siRNAs targeting the same TIR on global gene expression, we would be able to functionally dissect the role played by TIR intronic RNAs (either in the form of nascent or spliced-out introns) versus the corresponding mature transcripts. As proof-of-principle, we targeted NALF1 RNA, which we identified as the TIR forming the highest number of genome-wide trans contacts in NEU ( Supplementary Fig. 8a ). We designed 10 locked nucleic acids (LNA)-Gapmer ASOs targeting NALF1 intron 3 ( NALF1.Int3 ) and 5 siRNAs against NALF1 mRNA and transfected them in NEU cells ( Supplementary Table 3 and Methods ). To confirm the knock-down, we first performed quantitative PCR using different primer pairs targeting either NALF1 intronic or exonic regions or exon-intron junctions and identified LINC21_2 (targeting NALF1.Int3 ) and siRNA5 as the most efficient ASO and siRNA, respectively ( Supplementary Fig. 28a-f ). Next, we performed RNA-seq on NEU cells transfected with either LINC21_2 or siRNA5 and found that LINC21_2 reduced both intronic and exonic NALF1 RNA at comparable levels, whereas siRNA5 mainly reduced NALF1 exonic RNA, resulting in significantly higher intron-to-exon coverage ratio ( Fig. 6c-e and Supplementary Fig. 28g-i ). Unexpectedly, we found that almost all of the top-55 TIR source genes identified by RADICL-seq in NEU were significantly downregulated in NEU cells treated with LINC21_2, while their levels were unchanged in cells treated with siRNA5 ( Fig. 6f-k and Supplementary Fig. 29a-m ). Importantly, the extent of downregulation of TIR source genes correlated with the knockdown efficiency by different ASOs ( Supplementary Fig. 29n-z ). TIR source genes were significantly more enriched among downregulated genes ( P < 2.2×10 − 16 , Chi-square test, two-tailed) compared to all the genes associated with neuronal functions ( Fig. 6g, i ). Unlike TIR source genes, TIRC genes and TIRC control genes were both up- and down-regulated upon LINC21_2 or siRNA5 ( Fig. 6j ). Notably, the effect of LINC21_2 was not a mere reverse of what unveils during neurodifferentiation in terms of gene expression changes: only ∼10% of the genes upregulated upon differentiation to NEU were downregulated upon ASO treatment, and ∼45% of the genes downregulated upon LINC21_2 were upregulated during differentiation ( Supplementary Fig. 30a ). This difference was particularly pronounced when we grouped genes by their length, with ASO treatment showing a strong bias towards ultra-long neuronal genes, reminiscent of what was previously reported for genes downregulated in Rett Syndrome 38 , a severe NDD associated with pathogenic variants in the MECP2 gene 39 ( Fig. 6k and Supplementary Fig. 30b ). GO-term analysis of genes downregulated upon ASO treatment yielded terms linked to neuronal development, similar to those found when performing the analysis on the top 55 TIR-source genes identified in NEU ( Supplementary Fig. 30c, d ). Lastly, we performed a disease ontology (DO) analysis on genes downregulated upon LINC21_2, on the top-55 TIR-source genes, as well as on gene upregulated during neurodifferentiation. All the three gene groups were enriched in terms related to various mental disorders, including NDDs such as ASD and schizophrenia ( Fig. 6l and Supplementary Fig. 31a-c ). Collectively, these results indicate that NALF1 intronic RNA, but not NALF1 exonic RNA or NALF1 protein, regulate the expression of TIR source genes in neurons and suggest a functional role of TIRs in neurodevelopment. TIR intronic regions are enriched in NDD risk loci To further explore the link between TIRs and NDDs, we applied stratified linkage disequilibrium score regression (S-LDSC) 40 to test whether the heritability of single-nucleotide polymorphisms (SNPs) associated with NDD traits is enriched in TIR source genes, TIRCs or 96 annotated genomic regions from the Baseline model 41 , which includes regulatory regions, conserved regions, pc regions, untranslated regions, etc. ( Supplementary Table 4 and Computational Methods ). The heritability enrichment score (HES) indicates whether a region of interest is associated with a greater proportion of the heritability of a given trait, as compared to the proportion of heritability expected based on the SNPs contained in the region. As control, we also applied S-LDSC to partition SNP heritability of 19 non-NDD disease traits, including 7 neurodegenerative disorder traits, 6 immune-related traits and 6 other traits ( Supplementary Table 4 ). We found that the introns of TIR source genes and regions with high neutral rate score based on Genomic Evolutionary Rate Profiling (GERP.NS) 42 had the most significant and consistent SNP heritability enrichment for NDD traits, but not for control traits ( Supplementary Fig. 32 ). Specifically, we observed a significant enrichment for heritability in 11 out of 18 NDD traits versus 0 out of 19 control traits inside the introns of TIR source genes ( Fig. 6m ). In contrast, we did not observe a similar enrichment across TIRCs ( Supplementary Fig. 32 ). Altogether, these results indicate that TIR source genes are significantly enriched for risk loci associated with NDDs, potentially implicating TIRs in neurodevelopmental functions. Discussion Our study provides a comprehensive portrait of the RNA-DNA connectome, 3D genome, and transcriptome and of their changes during in vitro differentiation of human iPSCs to neurons. Through integrative genomic analyses, we identified a novel class of nuclear RNAs, which we named Trans-contacting Intronic RNAs (TIRs), that are enriched in intronic sequences and form widespread contacts with DNA, extending far from the producing locus on the same chromosome or on different chromosomes. We detected TIRs across multiple cell types, including three cancer lines (MCF7, K562, HCT116) profiled by RADICL-seq 21 in FANTOM6. However, neurons displayed by far the highest number of TIRs, which increased gradually during neurodifferentiation. This is possibly due to the fact that TIRs derive from long genes (which typically carry long introns), which are among the most highly expressed genes in neurons 43 . Contrary to our expectation, the majority of TIRs are not derived from lncRNAs, with the exception of MALAT1 , which is well-known to be part of nuclear speckles and form widespread genomic contacts 31 , 44 – 46 . Instead, across all the cell types examined, the majority of TIRs derive from the introns of pc genes that encode for proteins involved in cell type-specific functions. Our data suggest that TIRs are composed—to some extent at least—of spliced-out introns and do not merely reflect the accumulation of nascent RNAs. However, further studies are needed to investigate the actual composition of TIRs and the mechanisms by which these accumulate in the nucleus. Remarkably, over a decade ago, it was shown that fetal human brain cells harbor remarkably high levels of intronic RNAs derived from long genes associated with neuronal functions 26 . While this observation was linked to co-transcriptional and alternative splicing— both of which are common for long, synapse-associated genes 47 —the exact function of these intronic RNAs has remained elusive. Leveraging high-resolution DNA and RNA FISH, we found that TIRs accumulate in the nucleus of neurons forming cloud-like agglomerates around the source loci, as well as spreading across the entire nucleus. These patterns are profoundly different from the typical localization pattern of introns of classical genes, for which only individual intronic FISH dots can be visualized at active loci, representing introns in nascent RNAs. It is not clear how much TIR patterns contribute to the signal measured by RADICL-seq, but we speculate that both contribute to forming trans contacts with various parts of the genome. Notably, the TIR clouds surrounding different TIR-producing gene loci do not colocalize with each other or with nuclear speckles, and only minimally overlap with the CT of the corresponding source locus. We speculate two distinct roles for TIRs: (i) On the one hand, TIRs might act as constituents of the nuclear matrix, playing a role in organizing chromatin in space, for example by segregating different chromatin sub-compartments or sub-nuclear organelles from each other and/or bridging different genomic regions together. In support of this scenario, a previous study reported the accumulation of introns in the insoluble chromatin fraction remaining after DNA digestion and harsh protein extraction, suggesting that intronic RNAs act as constituents of the nuclear matrix 18 . In a more recent work it was proposed that mature transcripts and spliced-out introns are transported in opposite directions, with the mRNA transiting towards the speckles, and introns moving towards the nuclear matrix 48 . We note, however, that the staining pattern of the nuclear matrix shown in that study does not resemble the staining pattern of TIRs that we have observed, although the difference might be related to the fact that different cell types were analyzed. (ii) Alternatively, TIRs might form biomolecular condensates concentrating transcription and/or splicing factors around long neuronal genes to support their highest expression in neurons, despite the exceptional length of these genes. In support of this scenario, numerous studies have shown the propensity of certain RNAs to form condensates 49 and a recent pre-print reported the identification of numerous RNAs that can presumably undergo phase separation 50 . According to the latter study, longer RNAs seem more prone to engage in condensate formation and one of the two major classes of condensate-forming RNAs identified are rich in A/U nucleotides 50 . This observation aligns with our finding that TIRs originate from ultra-long introns and are particularly rich in A/U, strengthening our hypothesis that TIRs might promote phase separation. The nuclear matrix hypothesis might explain our finding that TIRs contact a well-defined set of genomic regions (which we named TIR contact regions or TIRCs) as well as with our observation that the global 3D genome architecture becomes progressively more refined during neurodifferentiation. Indeed, in neurons we identified ∼550 TIRCs (defined as genomic regions contacted by TIRs produced by at least 16 different source loci), which are enriched in exceptionally short, GC-rich genes that become activated in neurons. Unlike TIR-producing genes, TIRC genes have high connectivity and are engaged in strong inter-chromosomal interactions, as measured by Hi-C. This strong connectivity at well-defined genomic regions indicates that, in neurons, contacts between individual chromosomes vary less between individual cells, likely reflecting the adoption of a more stable 3D genome architecture in these post-mitotic cells. We hypothesize that as cells differentiate to neurons and exit the cell cycle, TIRs form a scaffold that brings together multiple TIRCs, contributing to shaping the final neuron-specific 3D genome architecture. This 3D genome structure consolidation might additionally be favored by the radial re-localization of GC-rich TIRCs towards the nuclear center in neurons, as detected by our GPSeq method 23 . Indeed, the radial GC-gradient that GPSeq detects during neurodifferentiation becomes steeper, further suggesting that these cells likely adopt a more stereotypical genome architecture. Leveraging GPSeq, we also found that TIR-producing loci tend to be located at the very periphery of the nucleus in neural stem cells and become less peripheral in neurons, despite still being relatively close to the nuclear membrane. We suspect that this re-localization reflects the decompaction of ultra-long, neuron-specific genes as they are activated in neurons, with at least some portions of these genes occupying a larger volume, in turn ‘pushing’ them towards the nuclear center. We also harnessed Hi-C maps to investigate the neighborhoods of TIR-producing genes and found that, in neurons, these loci are depleted of long-range contacts with other loci on the same chromosome. This might be explained by our observations that TIR source loci are densely surrounded by the TIRs transcribed from them, which likely hinders the source loci from directly contacting other regions on the same chromosome. Our observations that, in neurons, TIRs originate from ultra-long, neuronally expressed genes located at the nuclear periphery are possibly related to the finding of mega-enhancer bodies in cerebellum neurons during mouse development reported in a recent pre-print 51 . In this study, long neuronal genes were found to localize to a unique nuclear sub-compartment located at the nuclear periphery and relatively far from speckles, similar to the TIR source genes. The authors speculated that the peripheral localization of mega-enhancer bodies might be driven by physical constrains linked to the large size of the genes contacted by these enhancers, for which the nuclear center might not provide sufficient space, especially considering than many long genes are expressed simultaneously during neurodifferentiation, in line with previously reported transcriptional loops formed by highly-expressing long genes 52 . Notably, the same study describing mega-enhancer bodies also reported the identification of two different A sub-compartments: one carrying long neuronal genes forming mega-enhancer bodies, and the other comprising gene-dense genomic regions localized close to speckles in the nucleus center. We speculate that these two sub-compartments correspond to the TIR source genes and TIRCs that we have identified in this study and propose that TIRs somehow mediate the communication between these two sub-compartments since they both harbor genes that become specifically activated during neurodevelopment. Future studies will need to assess whether TIRs contribute to the formation of mega-enhancer bodies, given that the mechanism remains unknown. Lastly, it was very recently reported that exceptionally long genes localize close to Polycomb bodies, away from nuclear speckles 53 . The authors of this study proposed that the separation of long genes from speckles might protect them from premature internal splicing before the long introns of these genes are fully transcribed, which further supports our observations. Functionally, our study provides several lines of evidence in support of an important role of TIRs in neurodifferentiation and suggests a possible link with the pathogenesis of NDDs. First, the expression of TIR source and TIRC genes significantly co-varies, especially between TIRC genes contacted by common sets of TIRs, as revealed by gene expression co-variation analysis 35 . This suggests that TIRs somehow co-regulate the expression of multiple TIR source and possibly TIRC genes. Indeed, ASO-mediated knockdown of the most abundant TIR in neurons produced from intron-3 of the NALF1 gene exerts a pleiotropic effect on all the other TIR source genes identified in neurons as well as on many other genes associated with neuronal functions but, to our big surprise, not globally on TIRC genes. This pleiotropic effect is unexpected given that TIR clouds surrounding different TIR source loci do not appear to colocalize. The absence of a significant effect of ASOs on TIRC genes is also surprising; however, previous research on the effect that perturbations to the 3D genome architecture exert on gene expression suggests that this effect is typically rather minor or might take a long time to manifest 54 . In line with one of the two hypotheses outlined above, these observations lead us to speculate that TIRs form condensates, creating a high concentration of transcription and splicing factors around TIR source genes independently of speckles, thus explaining the pleiotropic effect observed upon knockdown of a single TIR species. At the same time, TIRs might provide a scaffold that organizes TIRC genes leading to the formation of highly connected TIRC networks in neurons, which, however, are less sensitive to the disruption of a single TIR species. Remarkably, TIR source genes and many of the genes that are downregulated upon TIR knockdown are significantly enriched in SNPs associated with several NDDs, including bipolar disorder and schizophrenia. Although we currently do not know whether TIRs are expressed at normal levels and form clouds in neuronal cells derived from NDD patients, we speculate that certain disease-associated variants might reduce the levels of or change the biochemical properties of the intronic sequences harbored in TIRs, in turn altering the ability of TIRs to cross-regulate the expression of many genes involved in neurodevelopment. Future studies will need to explore the landscape of TIRs in neuronal and glial cells differentiated from iPSCs derived from patients affected by various types of NDDs and to clarify whether TIRs play a role in the pathogenesis of these disorders. In sum, our study provides—to the best of our knowledge—the most comprehensive analysis of DNA-interacting RNAs in a human neurogenesis model, and highlights TIRs as a novel class of nuclear RNAs that shape the 3D genome architecture of neuronal cells and are potentially implicated in the pathogenesis of human neurodevelopmental disorders. Author Contribution Statement Conceptualization: W.K., M.B. Data curation: W.K., R.P., W.H.Y., Q.V., B.B., L.S. Sample preparation: W.H.Y., X.L., M.S., R.B., K.Y., M.M., M.K. Sequencing data acquisition: W.H.Y., M.S., R.B, X.S., M.K., C.W.Y., M.G. FISH validation and analysis: X.L., A.A.Y., Q.V. Knockdown experiments and analysis: K.Y., C.W.Y., R.P., Formal analysis: W.K., R.P., Q.V., L.S. Funding acquisition: N.C., P.C., M.B. Investigation: W.K., W.H.Y., Q.V., B.B., N.C., P.C., M.B. Project administration: B.B., N.C., M.B. Supervision: T.K., H.T., J.W.S., P.C., N.C., M.B. Visualization: W.K., L.S., Q.V., B.B., N.C., M.B. Writing: W.K., B.B., N.C., M.B. with contributions from all the authors . Competing Interest Statement P.C. is a co-founder of and H.T. is a shareholder of Transine Therapeutics, a startup company working on synthetic antisense lncRNAs that stimulate the translation of sense mRNAs. None of the other authors have financial or other competing interests related to this work to declare. Methods Experimental Methods Cell culture For RADICL-seq, Hi-C, scATAC-seq, and RNA-seq experiments we used human induced pluripotent stem cells integrated with a doxycycline-inducible NGN2 expression system at the AAVS1 safe harbor locus (hiPSC-i3N) and differentiated them to neural stem cells (NSC) and cortical neurons (NEU) as described below. This cell line is derived from a commercially available parental line WTC-11 iPSC (Coriell, cat. no. GM25256) and was generously made available to us by Dr. Michael E. Ward at the National Institutes of Health, Bethesda MA, USA. We cultured the cells in StemFit Basic04 Complete Type medium (Ajinomoto, cat. no. SF041-001) on cell culture vessels coated with iMatrix-511 (Nippi, cat. no. AMS.892 012). We refreshed the medium every other day and passaged the cells when they reached 70–80% confluence. To passage the cells, we rinsed them with 1x DPBS and incubated them with StemPro Accutase Cell Dissociation Reagent (Thermo Fisher Scientific, cat. no. A1110501) at 37 °C for 7 minutes. After centrifugation, we resuspended the cells in medium supplemented with 10 μM Y-27632 (Stemcell Technologies, cat. no. 72304) and seeded them on iMatrix-511 coated cell culture vessels. We refreshed the medium the next day after cell attachment. For GPSeq, we harnessed a different model system, which we previously used to map DNA double strand-breaks (DSBs) and 3D genome changes during neuronal cell lineage specification 55 . Briefly, we obtained a human iPSC-derived NSC line (AF22) from the human iPS core facility at Karolinska Institutet, where it was previously derived under ethical permit #2012/208-31/3 issued by the local Ethics Review Committee. We cultured the cells following a protocol 56 previously established at the same facility with some modifications, as described in ref. 55 . We then differentiated them to neural progenitor cells (NPC) and cortical neurons (NEU) as described below. Differentiation of iPSC and NSC cells to cortical neurons To differentiate iPSC-i3N to cortical neurons, we first differentiated them to NSC using the PSC Neural Induction Medium (Thermo Fisher Scientific, cat. no. A1647801) according to the manufacturer’s protocol. Briefly, we seeded 10 6 cells on a 10-cm cell culture dish coated with an iMatrix-511 (Nippi, cat. no. AMS.892 012). We exchanged the medium to Neural Induction Medium the next day (Day 1) to initiate the differentiation to NSC. We subsequently refreshed the medium every second day, on Day 2, 4 and 6. On Day 7, we dissociated the differentiated NSC with StemPro Accutase Cell Dissociation Reagent (Thermo Fisher Scientific, cat. no. A1110501) and then either expanded them in Neural Expansion Medium or cryopreserved them in STEM-CELLBANKER (Zenogen, cat. no. 11924). To further differentiate NSC into cortical neuron, we followed a previously described protocol 57 . Briefly, we seeded the NSC on cell culture dishes coated with poly-L-ornithine (Sigma Aldrich, cat. no. P3655) and cultured them in differentiation medium I containing 0.5x DMEM/F-12 medium (Thermo Fisher Scientific, cat. no. 31331093), 0.5x Neurobasal Medium (Thermo Fisher Scientific, cat. no. 21103049), 0.5x Non-Essential Amino Acids (NEAA) (Thermo Fisher Scientific, cat. no. 11140050), 0.5x GlutaMAX (Thermo Fisher Scientific, cat. no. 35050061), 0.5x N-2 Supplement (Thermo Fisher Scientific, cat. no. 17502001), 0.5x B-27 Supplement (Thermo Fisher Scientific, cat. no. 17504044), 2.5 μg/mL Human Insulin (Sigma, cat. no. I9278), 2 μM DAPT (Sigma, cat. no. D5942), 50 μM 2-Mercaptoethanol (Thermo Fisher Scientific, cat. no. 21985023), 2 μg/mL Doxycycline hyclate (Sigma, cat. no. D9891), and 5 μg/mL Mouse Laminin (Sigma, cat. no. L2020). After three days of induction, we switched the medium to differentiation medium II containing 0.5x DMEM/F-12 medium (Thermo Fisher Scientific, cat. no. 31331093), 0.5x Neurobasal Medium (Thermo Fisher Scientific, cat. no. 21103049), 0.5x NEAA, 0.5x GlutaMAX™, 0.5x N-2 Supplement, 0.5x B-27 Supplement, 2.5 μg/mL Human Insulin, 2 μM DAPT, 10 ng/mL BDNF (Peprotech 450-02), 10 ng/mL GDNF (Peprotech 450-10), 10 ng/mL NT-3 (Peprotech 450-03), 50 μM 2-Mercaptoethanol, 2 μg/mL Doxycycline hyclate, 0.5 μg/mL Mouse Laminin, and replaced half of the medium without doxycycline every 3 days until Day 10. Lastly, we harvested fully differentiated cortical neurons for RADICL-seq, Hi-C, scATAC-seq, and RNA-seq experiments. For FISH experiments on NSC, we seeded NSC on iMatrix-coated, 18-mm glass coverslips (Marienfeld, cat. no. 630-2200) placed in 12-well plates and cultured the cells for 2-3 days in Neural Induction Medium supplemented with 10 μM Y-27632 during the first 24 hours before fixation. For NEU, we instead NSC on poly-L-ornithine coated coverslip and differentiated as abovementioned. To differentiate AF22 cells to NPC and cortical neurons for GPSeq, we followed the same procedure that we previously described 55 . Briefly, we plated 78,000/cm 2 AF22 cells in two 10 cm dishes each containing six 22 x 22 mm microscope coverslips (Marienfeld, cat. no. 0107052) pre-coated with poly-L-ornithine (Sigma Aldrich, cat. no. P3655) and laminin (Sigma, cat no. L2020). Before plating the cells, we resuspended the cells in a differentiation medium composed of DMEM-F/12 medium (Thermo Fisher Scientific, cat. no. 31331093) supplemented with 1x N-2 Supplement (Thermo Fisher Scientific, cat. no. 17502001), 1x B-27 Supplement (Thermo Fisher Scientific, cat. no. 17504044), and 1% Penicillin/Streptomycin (Thermo Fisher Scientific, cat. no. 15140122). We prepared fresh differentiation medium every two weeks and replaced the growth medium every second day. After Day 10, we added 1 μg/mL Mouse Laminin (Sigma-Aldrich, cat. no. L2020) to the differentiation medium to promote cell attachment to the flask. RADICL-Seq We performed RADICL-Seq as we previously described 21 . Briefly, we fixed 1 million iPSC, NSC and NEU cells in 1% formaldehyde (FA) (Thermo Fisher Scientific, cat. no. 28906) solution in 1x PBS and stored the cell pellets after FA quenching and washing at –80 °C. For the proximity ligation of RNAs and chromatin, we resuspended the cell pellets in a lysis buffer containing 10 mM Tris-HCl pH 8.0, 10 mM NaCl, 0.2% NP-40, followed by a mild digestion with DNase I (Thermo Fisher Scientific, cat. no. EN0525) at 37 °C for 10 minutes. We then performed end-repair with T4 DNA Polymerase (Thermo Fisher Scientific, cat. no. EP0062) and DNA Polymerase I, Large (Klenow) Fragment (Thermo Fisher Scientific, cat. no. 18012021) at room temperature for 1 hour followed by A-tailing using Klenow Fragment (3’→5’ exo-) (Thermo Fisher Scientific, cat. no. EP4021) at 37 °C for 1 hour. We added RNase H (New England BioLabs, cat. no. M0297L) to the reaction and incubate at 37 °C for 40 minutes. We ligated the pre-adenylated RADICL-seq adapter to the RNA using T4 RNA Ligase 2, truncated KQ (New England BioLabs, cat. no. M0373L) at 20 °C overnight. The next day, we performed ligation of the adapter-ligated RNA to chromatin by incubating the cell pellets with T4 DNA Ligase (New England BioLabs, cat. no. M0202L) at room temperature for 4 hours. We then proceeded to extract genomic DNA and prepare sequencing libraries as described in ref. PMID: 32094342. We sequenced each RADICL-seq library on NovaSeq 6000 using S2 Reagent Kit v1.5 (200 cycles) kit (Illumina, cat. no. 20028315) with single-end 150 cycles mode, aiming at generating 600 million reads per sample. Hi-C We performed Hi-C on iPSC-i3N, NSC, and NEU cells using the Arima HiC+ kit (Arima Genomics, cat. no. A510008) and the Arima Library Prep Module (Arima Genomics, cat. no. A303011). Briefly, we fixed XXX iPSC, NSC and NEU cells in 2% paraformaldehyde (PFA) (Thermo Fisher Scientific, cat. no. 28906) solution in 1x PBS and stored the cell pellets at –80°C after PFA quenching and washes. We then performed in situ proximity ligation and library preparation following the manufacturer’s protocol. We sequenced the libraries on NovaSeq 6000 using the 2 lane of S4 Reagent Kit v1.5 (300 cycles) (Illumina, cat. no. 20028312) with paired-end 150 sequencing mode, aiming at generating 800–1,000 million reads per library. GPSeq We performed GPSeq on NSC, NPC, and NEU as we previously described 23 with minor modifications. Briefly, for each cell type, we fixed 0.3 million cells on 22 x 22 mm coverslips in 4% paraformaldehyde (PFA) (Thermo Fisher Scientific, cat. no. 28906). We performed radial chromatin digestion using DpnII for 6 different timepoints (10 sec, 30 sec, 2 min, 5 min and 30 min for NSC; 10 sec, 30 sec, 2 min, 5 min, 10 min and 30 min for NPC; 10 sec, 20 sec, 30 sec, 45 sec, 2 min and 5 min for NEU). To visually confirm that radial digestion had occurred properly, we performed YFISH and analyzed the images as described in ref. 23 . We prepared one sequencing library for each timepoint and sequenced all the libraries on NextSeq 500/550 using High Output Kit v2.5 (75 Cycles) (Illumina, cat. no. 20024906) with single-end 75 cycles mode, aiming at generating 50 million reads per library. scATAC-seq We isolated nuclei from iPSC-i3N, NSC and NEU following the 10x Genomics protocol for nuclei extraction for scATAC-seq (10x Genomics, Demonstrated Protocol no. CG000169). We performed DNA tagmentation, single-cell droplet encapsulation and cell indexing using the Chromium Next GEM Single Cell ATAC Library & Gel Bead Kit (10x Genomics, cat. no. 1000175) aiming at 5000 nuclei. We sequenced all the libraries on HiSeq X using one lane of HiSeq X Ten Reagent Kit v2.5 (Illumina, cat. no. FC-501-2501) with paired-end 150 sequencing mode, aiming at generating 150 million reads per library. Total RNA-seq We extracted total RNA from iPSC-i3N, NSC and NEU using the Qiagen RNeasy Mini kit (Qiagen, cat. no. 74104) and prepared libraries using the TruSeq Stranded Total RNA Sample Prep Kit with Ribo-Zero Human/Mouse/Rat kit (Illumina, cat. no. 20020596) and TruSeq RNA Single Indexes (12 indexes, 24 samples) Set A (Illumina, cat. no. 20020492). We sequenced all the libraries on NovaSeq 6000 using 1 lane of S4 Reagent Kit v1.5 (300 cycles) (Illumina, cat. no. 20028312) with paired-end 150 sequencing mode, aiming at generating 300–400 millions reads per library. Cap Analysis of Gene Expression (CAGE) We extracted total RNA in the same way as for total RNA-seq and prepared CAGE libraries as previously described 25 . We sequenced all the libraries on NextSeq 1000/2000 using the P2 Reagent Kit (Illumina, cat. no. 20100986) in paired-end 100 sequencing mode, aiming at generating 50 million reads per library. DNA and RNA FISH Probe design and production We designed DNA/RNA FISH probes targeting various intronic and exonic sequences of TIR source genes and matching TIR control genes; TIRCs and matching TIRC control loci; as well as chr12 and chr13 spotting probes using the iFISH pipeline that we previously developed 27 . We appended a probe-specific barcode pair to each oligo and ordered multiple probes as oligopools (Twist Bioscience). The genomic coordinates and oligo sequences of all FISH probes used in this study are available in Supplementary Table 2 . We amplified individual probes from the oligopools by PCR using the corresponding primers, complemented with a readout sequence at the 5’ end and a T7 promoter sequence at the 3’ end, using the PowerUp SYBR Green Master Mix (Thermo Fisher Scientific, cat. no. A25776). We purified the PCR product with AMPure XP magnetic beads (Beckman Coulter, cat. no. A63882) following the manufacturer’s instructions. We performed in vitro transcription (IVT) with the purified PCR product as template, using the HiScribe T7 Quick High Yield RNA Synthesis Kit (New England Biolabs, cat. no. E2050S) in presence of RNaseOUT (Thermo Fisher Scientific, cat. no. 10777019). We purified the RNA product with RNAClean XP magnetic beads (Beckman Coulter, cat. no. A63987) and converted it to single-stranded DNA (ssDNA) by reverse transcription using the Maxima H Minus Reverse Transcriptase (Thermo Fisher Scientific, cat. no. EP0752). Finally, we hydrolyzed the RNA in alkaline conditions and purified the resulting probes using the Monarch PCR & DNA Cleanup Kit (New England Biolabs, cat. no. T1030L). RNA FISH After growing cells on coverslips, we fixed them in 4% PFA (Fisher Scientific, cat. no 11481745) in PBS for 10 min at room temperature, before quenching with 125 mM glycine (Sigma Aldrich, cat. no. 4810) in PBS for 5 min. We permeabilized the cells using 0.5% Triton X-100 (Sigma Aldrich, cat. no. T8787) in PBS for 20 min at room temperature, followed by two washes with 0.05% Triton X-100 in PBS; or, alternatively, 70% ethanol for 10 min at room temperature. We incubated the cells in 0.1 N HCl (Sigma Aldrich, cat. no. 1090571000) for 5 min at room temperature, followed by two washes with 0.05% Triton X-100 in PBS. We equilibrated the cells in RNA wash buffer, composed of 25% formamide (Ambion, cat. no. AM9344) in 2x saline-sodium citrate (SSC) buffer (Ambion, cat. no. AM9765), for at least 5 min at room temperature, before proceeding with hybridization. We mixed the ssDNA FISH probes at a ratio of 1:99 with RNA hybridization buffer containing 2x SSC, 25% formamide, 10% dextran sulfate (Sigma Aldrich, cat. no. S4030), 1 mg/mL tRNA from E. coli (Sigma Aldrich, cat. no. 10109541001), 0.2 mg/mL BSA (Ambion, cat. no. AM2616), and 2 mM Ribonucleoside Vanadyl Complex (RVC; New England Biolabs, cat. no. S1402S), to achieve a final concentration of 1 nM per oligo in the final hybridization mix. We performed hybridization overnight at 37℃ in a humidity chamber. On the next day, we washed the coverslips twice with RNA wash buffer for 30 min at 37℃. Afterwards, we prepared the readout hybridization mix, by combining fluorescently-labeled readout oligos with the readout hybridization buffer containing 2x SSC, 10% dextran sulfate and 25% formamide, at a ratio of 1:99 to achieve a final concentration of 20 nM per readout oligo. We incubated the cells with the readout hybridization mix overnight at 30℃ in a humidity chamber. The next day, we washed the cells with 0.2x SSC, 0.2x Tween-20 (Sigma Aldrich, cat. no. P9416) for 2x 10 min at 45℃, then performed counterstaining with 1 ng/μL Hoechst 33342 (Thermo Fisher Scientific, cat. no. 62249) in 2x SSC for 30 min at room temperature. After washing the cells with 2x SSC, we mounted the coverslips in GLOX buffer, containing 10 mM Trolox (Sigma Aldrich, cat. no. 238813), 37 μg/mL glucose oxidase (Sigma Aldrich, cat. no. G2133), 100 μg/mL catalase (Sigma Aldrich, cat. no. C3515) in an equilibrium buffer with 10 mM Tris-HCl (Invitrogen, cat. no. 15567027), 0.4% glucose (Sigma Aldrich, cat. no. 49139) and 2x SSC, sealed the sides of the coverslips with rubber cement (Fixogum, Triolab, cat. no. LK071A) and proceeded immediately with imaging. Alternatively, we mounted the coverslips in Prolong Glass Antifade Mountant (Thermo Fisher Scientific, cat. no. P36980), let them cure for 24 hours and proceeded with imaging. DNA FISH and DNA-RNA FISH After growing cells on coverslips, we fixed them in 4% PFA in PBS for 10 min at room temperature, before quenching with 125 mM glycine in PBS for 5 min. We permeabilized the cells using 0.5% Triton X-100 in PBS for 20 min at room temperature, followed by two washes with 0.05% Triton X-100 in PBS. We incubated the cells in 0.1 N HCl for 5 min at room temperature, followed by two washes with 0.05% Triton X-100 in PBS. We equilibrated the cells in FPS buffer composed of 50% formamide, 50 mM sodium phosphate (Thermo Fisher Scientific, cat. no. J60158-AP) and 2x SSC buffer overnight at room temperature. The next day, we incubated the cells for 1 hour in the pre-hybridization buffer containing 50% formamide, 5x Denhardt’s solution, 1 mg/mL tRNA from E. coli , 100 μg/mL salmon sperm DNA (Invitrogen, cat. no. 15632011), 0.2 mg/mL BSA and 2x SSC, before proceeding with hybridization. We mixed the ssDNA FISH probes at a ratio of 1:10 with DNA hybridization buffer, composed of 2.2x SSC, 5.5x Denhardt’s solution, 11% dextran sulfate and 55% formamide, to achieve a final concentration of 0.05 nM per oligo for FISH probes against DNA, and 1 nM per oligo for FISH probes against RNA. We placed the coverslips with cells face-down on microscope slides with 10 μL of the hybridization mix and sealed the coverslips with rubber cement. We performed DNA denaturation at 75–78℃ for 2-3 min on a heating block, then transferred the slides to a humidity chamber for hybridization overnight at 37℃. On the next day, we released the coverslips in 2x SSC, 0.2% Tween-20, then washed the samples twice for 7 min at 56–65 ℃ in 0.2x SSC, 0.2% Tween-2. We rinsed the coverslips with 4x SSC, 0.2% Tween-20, followed by 2x SSC, before equilibrating the cells in RNA wash buffer, composed of 25% formamide in 2x SSC, for at least 5 min at room temperature. Afterwards, we prepared the readout hybridization mix, by combining fluorescently-labeled readout oligos with the readout hybridization buffer containing 2x SSC, 10% dextran sulfate and 25% formamide, at a ratio of 1:99 to achieve a final concentration of 20 nM per readout oligo. We incubated the cells with the readout hybridization mix overnight at 30℃ in a humidity chamber. The next day, we washed the cells twice with 0.2x SSC, 0.2x Tween-20 for 2x 10 min at 45℃, then performed counterstaining with 1 ng/μL Hoechst 33342 in 2x SSC for 30 min at room temperature. After washing the cells with 2x SSC, we either mounted the coverslips in GLOX buffer, sealed the sides of the coverslips with rubber cement and proceeded immediately with imaging; or mounted the coverslips in Prolong Glass Antifade Mountant, let cure for 24 hours and proceeded with imaging. Imaging We acquired fluorescent images on either a custom-built Nikon Eclipse Ti inverted wide-field microscope, equipped with a Plan Apochromat Lambda 100x/1.45 oil-immersion objective (Nikon) and an Andor iXon Ultra 888 EMCCD camera (Oxford Instruments), or on a custom-built Nikon Eclipse Ti2 inverted wide-field microscope, equipped with a Plan Apochromat Lambda 100x/1.45 oil-immersion objective (Nikon) and an Andor Sona sCMOS camera (Oxford Instruments). For all datasets, we collected z-stacks with 200-300 nm between imaging planes. Knockdown of TIRs Design of ASOs and siRNAs We designed locked nucleic acid (LNA) phosphorothioate gapmer antisense oligos (ASOs) against the NALF1 TIR based on the transcript models defined by CFC-seq 58 and purchased them from Qiagen. Additionally, we designed siRNAs targeting specific sequences in exon 1 or exon junction 1-2 and 2-3 of NALF1 and purchased them from Thermo Fisher Scientific. As negative ASO or siRNA control, we purchased Antisense LNA GapmeR Controls (Qiagen, cat. no. 339515) and Silencer Select Negative Control No. 1 siRNA (Thermo Fisher Scientific, cat. no. 4390843), respectively. The sequences of all the ASOs and siRNAs used in this study are listed in Supplementary Table 3 . RNA knockdown We differentiated NSC on Poly-L-Ornithine (PLO)-coated 12-well plates at a density of 3 x 10 5 cells per well as described above and transfected ASOs and siRNAs at Day 9 of differentiation following the protocol described in ref. 59 . Briefly, we prepared a final concentration of 20 nM of each ASO or siRNA by mixing 2 µL of Lipofectamine RNAiMAX (Thermo Fisher Scientific, cat. no. 13778150) in 200 µL of OptiMEM (Thermo Fisher Scientific, cat. no. 11058021). We incubated the transfection reagents for 10 minutes at room temperature before combining them with 1 mL of fresh differentiation medium II without doxycycline. We performed the transfections in duplicate alongside with negative control ASO and siRNA. After 24 hours, we refreshed the medium and after 48 hours, we harvested RNA using the RNeasy kit (Qiagen, cat. no. 74004) and on-column DNase digestion (Qiagen, cat. no. 79254). We stored the purified RNA at –80°C until proceeding to real-time quantitative RT-PCR or RNA-seq library preparation. Real-time quantitative PCR (RT-qPCR) We quantified purified RNA using a NanoDrop Onespectrophotometer (Thermo Fisher Scientific, cat. no. ND-ONE-W) and performed real-time quantitative RT-PCR using the One Step PrimeScript RT-PCR Kit (Takara Bio Inc., cat. no. RR064B). We designed primers targeting the intronic TIR regions and exon regions of NALF1 (see Supplementary Table 3 for the list of primer sequences). We normalized the expression levels to GAPDH and calculated the knockdown efficiency based on the fold-change between the knockdown samples and the negative controls. Computational Methods The following sections are ordered based on when the corresponding analyses are first described in the Results. RNA-seq data analysis We processed and mapped all the Total RNA-seq datasets, including the data from the NALF1 gapmer ASO and siRNA experiment, using the nf-core rnaseq pipeline (v 3.10.1; https://github.com/nf-core/rnaseq ). We aligned the reads to the human GRCh38/hg38 reference genome and used the GENCODE gene annotation v38 as annotation reference. We used the tximport (v 1.18.0) R package to generate a count table containing transcript abundances in transcripts per million (TPM) derived from salmon quant.sf output files, which were previously obtained in the nf-core rnaseq pipeline. For differential gene expression analysis, we used the DESeq2 (v 1.30.1) R package. We defined expressed genes as those having expression higher than the mean expression (TPM ≥ 9.6, corresponding to ∼0.4 quantile of all the protein coding genes in the NEU replicate 1 sample (see Supplementary Table 1 ). The DESeq raw count tables can be found in the can be found in the Source Data available at https://github.com/wenjingk/TIR_code/ . Gene ontology analysis We performed gene ontology (GO) analysis using DOSE (v3.28.1), clusterProfiler (v4.10.0) and org.Hs.eg.db (v3.18.0) R packages. We obtained differentially expressed genes with a P value below 0.01 without filtering of fold-change. We fed the genes to the enrichDO function from DOSE and the enrichGO function from clusterProfiler to search for enriched ontologies, testing the provided genes against a background universe consisting of expressed genes only. We retained only findings related to molecular functions or biological processes. We then summarized and represented the results using ggplot2 (v 3.5.1). RADICL-seq data analysis We processed and mapped the RADICL-seq data as previously described 21 . In brief, we mapped the data to the human GRCh38/hg38 reference genome. We assigned RNA tags to annotated genes while preserving strand specificity, and grouped DNA tags into 25 kb genomic bins, with each fragment represented by its central nucleotide to minimize overlap with multiple bins. We excluded all blacklisted genomic regions in the ENCODE hg38 genome blacklist and filtered out duplicated RNA-DNA contacts using the BEDtoolsr R package. To determine statistically significant RNA-DNA contacts, we used a modified negative binomial distribution model implemented in CHiCANE 60 , 61 . We retained only significant contacts that met a q-value threshold of 0.05 (q ≤ 0.05) for further analysis. We summarized contacts of a gene in the same 100 kb genomic bin using intersect and groupby in bedtools (v2.30.0). The binned RADICL-Seq data tables for iPSC, NSC and NEU can be found in the Source Data available at https://github.com/wenjingk/TIR_code/ . Since correlation analysis showed a high degree of similarity between biological replicates, we merged the two replicates from each cell line. We classified the significant RNA-DNA contacts as cis if the RNA contacts a DNA locus less than 5 Mb apart on the same chromosome or as trans if the contact occurs beyond 5 Mb or on a different chromosome. Integration of RADICL-seq and RNA-seq data To integrate RADICL-seq and RNA-seq data, we first used significant RNA-DNA contacts retained after CHiCANE filtering and used them to generate the RADICL-Seq RNA coverage profile (in bedGraph format) over the genome using bedtools genomecov (v 2.29.2). We then converted the RNA-seq RNA coverage (in bedGraph format) from the bigWig files generated by nf-core rnaseq pipeline, using bigWigToBedGraph . For each gene of interest, we first generated sliding windows with window size 100 bp and step size 10 bp using the makewindows tool in bedtools and annotated them based on whether the windows are located in exons (including 3′ and 5′ UTRs) or introns. We then intersected the windows with the RNA coverage of RADICL-Seq and RNA-seq and summed up the coverage counts falling in each window. For each gene, we used the maximum coverage count of each exon and intron to calculate the intron vs. exon coverage ratio. This is equal to the mean of the maximum coverage counts for introns divided by the mean of the maximum coverage counts for exons. FISH image analysis Deconvolution We processed all FISH datasets using our deconvolution package Deconwolf 62 (v0.4.5, available at https://deconwolf.fht.org/ ). Briefly, we generated in silico point-spread functions (PSF) for each of the fluorescent dyes used for FISH as well as for Hoechst 33342 used for DNA staining, with the imaging parameters matching the dataset to process. As an example, we generated a PSF for images of ATTO 647, with an emission maximum at 662 nm, and imaged with an oil-immersion objective with NA 1.45, pixel size of 65 nm, and 200 nm between z-planes at acquisition, using the following parameters: dw_bw --lambda 662.0 --NA 1.450 --ni 1.515 --threads 4 --resxy 65.0 --resz 200.0 ’./ATTO647_PSF.tif’ We applied the obtained PSF to deconvolve the corresponding datasets, by processing single-color channels individually, then tiling the image stack in four along the x-y axis, and using GPU acceleration to process the 4 stacks separately before recombining them at output. As an example, we ran Deconwolf on an ATTO 647 image stack using the following settings, for 50 iterations with GPU acceleration: dw --bq 2 --method shbcl2 --tilesize 1024 --iter 50 --threads 4 ’image_path_ATTO647.tif’ ’./ATTO647_PSF.tif’ For visualization, we used 50 iterations of Deconwolf to deconvolve all the images shown in the figures. For nucleus segmentation and identification of dots, we used images deconvolved for 50–100 iterations, to reduce background signals without reaching oversharpening. Feature extraction We segmented cell nuclei using either thresholding of the Hoechst 33342 channel, or manual annotation. We detected FISH dots using the “ df_getDots ” function in our DOTTER 63 analysis suite, using default values. Briefly: (i) We applied a Difference-of-Gaussians filter to the image; (ii) We extracted local maxima to identify possible dots of interest; (iii) We set a manual threshold with visual feedback, to separate false from true localization events; and (iv) We fitted a Gaussian model on the remaining dots using Maximum Likelihood, to retrieve sub-pixel localization. For intron and exon quantifications, we obtained dot counts per field of view (130 x 130 μm 2 ) and normalized them to the number of nuclei. To estimate the relative proximity of loci to nuclear speckles, we identified DNA FISH dots within manually corrected nuclei masks, then measured the average MALAT1 fluorescence intensity within 0.5 μm around each dot compared to 2.5 μm away from it and reported this ratio as the local enrichment of MALAT1 transcripts. Hi-C data analysis We first processed the Hi-C data following Arima Genomics’ recommendations to trim 6 nt from the left end, and 70 nt from the right end of each read using trimfq -b 6 -e 70 in the seqtk tool ( https://github.com/lh3/seqtk ). We then fed the remaining 75 nt from each read to the nf-core hic pipeline (v 2.1.0, https://nf-co.re/hic/2.1.0/ ) with the following parameters: --genome ’hg38’ --restriction_site ’[^GATC,G^ANTC]’ --ligation_site ’[GATCGATC,GANTGATC,GANTANTC,GATCANTC]’ --digestion ’arima’ --res_compartments ’500000,250000,100000’ --tads_caller ’insulation,hicexplorer’ --bin_size ’1000000,500000,100000,50000,25000,10000,5000,2500,1000’ Since correlation analysis showed a high degree of similarity between biological replicates, we merged the two replicates from each cell line. To call A/B compartments, we processed the Hi-C contact maps at 100 kb resolution generated by the nf-core hic pipeline using a customized R script with the following steps: Load the ICE normalized contact map. Convert the contact map to the observed versus expected (O/E) matrix. We first calculated the expected Hi-C signal for intra-chromosomal contact map using cooltools expected-cis (v 0.6.1) with the balanced cool file at 100 kb resolution as input and with the options --smooth --aggregate-smoothed . For each DNA-DNA contact, we then calculated the O/E value by dividing the ICE normalized value by the expected value (the ‘balanced.avg.smoothed’ reported from the cooltools expected-cis ) of the given linear genomic distance between the two DNA loci. Convert the O/E matrix to correlation matrix using R cor with the use=“pairwise.complete.obs” option. Perform Principal Component Analysis (PCA) on the correlation matrix using R prcomp , where the correlation matrix is filtered by removing DNA bins with missing values. Correct the sign of the PC1 value if needed to obtain a positive correlation with GC-content, so that DNA bins with positive sign represent the A compartment, while DNA bins with negative sign represent the B compartment. The GC-content at 100 kb resolution is obtained using genome gc in cooltools with the coordinates of 100 kb genomic bins and the human reference genome hg38 as input. The customized R script is available at https://github.com/wenjingk/TIR_code/tree/main/Hi-C_analysis . GPSeq data analysis GPSeq data preprocessing We processed GPSeq data using a customized pipeline available at https://github.com/wenjingk/GPSeqNP . We installed all necessary software and packages in a docker and singularity container. Briefly, we indexed the reference genome (GRCh38) using bowtie2 (v 2.4.1) and identified the coordinates of the recognition site (GATC) of the restriction enzyme DpnII in the reference genome using fastx-barber (v 0.1.5). We then identified the 8-nt unique molecular identifier (UMI), 8-nt sample barcode, 4-nt cut site in subsequent order from the 5′ end of the reads. We trimmed these sequences from the read but stored them and their corresponding sequencing quality scores in the header using fastx-barber . We discarded reads if the barcode or the cut site sequence did not match the expected sequence with up to 1 mismatch or if more than 20% of the nucleotides in the UMI had a quality score less than 30. We aligned the remaining reads to the reference genome using bowtie2 with options --very-sensitive -L 20 --score-min L,-0.6,-0.2 --end-to-end . After the mapping, we retained only primary alignments with a mapping quality equal to or higher than 30, while discarding unmapped, multimapping, chimeric, chrM, low mapping quality reads. Next, we grouped reads derived from the same DNA locus and calculated the distance between the DNA locus and the nearest cut site using the customized script group_umis.py . We discarded groups that could not be confidently assigned to a nearest cut site located within 20 nucleotides using the customized script umis2cutsite.py . Finally, we removed duplicated reads at each cut site if they had the same UMI using the customized script umi_dedupl.R . These preprocessing steps resulted in a bed file containing the number of unique reads of each cut site for each digestion time point. GPSeq score and percentile calculation To calculate the GPSeq score 23 , we input the bed files generated from the preprocessing above and a config file containing sample information to the gpseq-radical.R script (v 0.0.9) with the following parameters: --normalize-by lib --chromosome-wide --binsize 1e6:1e5,5e5:5e4,1e5:1e4,5e4:5e4,25e3:25e3 --ref-genome hg38 --chrom-tag 22:X,Y --site-domain universe --mask-bed “blacklist.bed” The blacklist file contains regions with low mappability (< 80% coverage of k50 UMAP data, https://bismap.hoffmanlab.org/ ), including highly repeated regions (e.g., centromeres). We excluded these regions for the GPSeq score calculation. To enable comparison across cell types and account for variable ranges of the GPSeq score observed in different experiments, we calculated the GPSeq percentile as: (# of genomic bins with score less than or equal to the selected score)/(Total # of bins) × 100. The GPSeq score and GPSeq percentile tables for iPSC, NSC and NEU can be found in the Source Data available at https://github.com/wenjingk/TIR_code/ . scATAC-seq data analysis We processed the scATAC-seq data using cellranger-atac-2.0.0 , which yields three files per cell differentiation stage: 1) ‘peaks.bed’ contains the coordinates of scATAC-seq peaks; 2) ‘matrix.mtx’ contains the peak index, cell index and peak abundance; and 3) ‘barcodes.tsv’ contains the barcode of single cell. We obtained the coordinates of scATAC-seq peaks from ‘peaks.bed’ and calculated the height of the peaks based on the number of cells sharing the peak by parsing ‘matrix.mtx’. We then assigned the scATAC-seq peaks to the 100 kb DNA bins over the genome using bedtool intersect and used the average height of the peaks mapped to the same DNA bin to represent the abundance of scATAC-seq signals per bin. The scATAC-seq count table per bin can be found in the Source Data available at https://github.com/wenjingk/TIR_code/ . Single-cell gene expression covariation analysis We downloaded gene count tables of two independent batches (replicates) of Smart-seq2 on iPSC-derived motor neurons (94 and 87 in batch-1 and -2, respectively) from GSE138120. We first removed low-quality cells having less than 6,000 genes expressed and removed not expressed genes in 56 of the cells from batch-1 and 48 of the cells from batch-2. We calculated correlations between gene pairs using Spearman’s rank correlation coefficient (SCC) on the filtered count tables of batch-1 (73 cells × 5,291 genes) and batch-2 (69 cells × 5,379 genes). We calculated P values using a z-score for the Fisher transformation of the SCC. We applied further filtering steps as previously described 35 : 1) we only considered gene pairs that were significantly correlated (P < 0.01) in both replicates; 2) we discarded gene pairs if the sign of the SCC differed between replicates or if the sign was negative; 3) we discarded riboprotein gene pairs, as these tend to dominate the positive covariations as previously described 35 . We then used the mean SCC values of gene pairs to calculate the Covariation Enrichment Score (CES). We separately calculated the CES for different gene sets: Expressed protein-coding (pc) genes within genomic bins (100 kb resolution) that are co-contacted by TIRs produced from 1, 2, 3, 4, 5, 6, or 7 out of the top-10 TIR source genes identified in NEU. We randomly selected each combination of TIR-producing genes for 100 times and each time calculated the CES of the gene set selected. Expressed pc genes within genomic bins (100 kb resolution) belonging to the same radial layer defined based on GPSeq: peripheral layer (GPSeq percentile < 30%), middle layer (55% < GPSeq percentile 0.95%). For each radial layer, we randomly selected 300 genes for 20 times and each time calculated the CES of the gene set selected. TIR-producing genes and expressed pc genes within genomic bins (100 kb resolution) belonging to TIRCs, TIRC control set 1 and TIRC control set 2. For each gene group, we randomly selected 300 genes for 20 times and each time calculated the CES of the gene set selected. GWAS trait-associated SNP enrichment analysis To check whether disease risk loci are enriched inside TIR source and TIRC genes, we applied Stratified Linkage Disequilibrium Score Regression (S-LDSC) 40 implemented in ldsc (v1.0.1) available at https://github.com/bulik/ldsc . Briefly, we calculated stratified linkage disequilibrium (LD) scores from the European ancestry samples in the 1000 Genomes Project Phase 3 data with a minor allele frequency of > 0.05 and only included HapMap3 SNPs. We then partitioned the SNP heritability of neurodevelopmental disorder traits ( n = 18), neurodegenerative disorder traits ( n = 7), immune related traits ( n = 6) and other traits ( n = 6) based on the annotations used in the Baseline Model v2.2 41 , which include DNaseI hypersensitivity sites, promoter and enhancer sites, protein-coding regions, untranslated regions, evolutionarily conserved regions, and more ( Supplementary Table 4 ). In addition to these publicly available annotations, we also added TIR-related annotations, including TIR-producing gene regions, intronic regions of TIR-producing genes, TIR control gene regions, TIRCs, and TIRC control regions. We downloaded the 1000 Genomes Project Phase 3 data, the HapMap3 SNPs, the baseline model, and the GWAS summary statistics of 37 traits from https://alkesgroup.broadinstitute.org/LDSCORE/ . We plotted the heritability enrichment (HE) of the selected annotations using R (v 4.0.5) and ggplot2 (v 3.3.5) if the enrichment of the trait-annotation pairs was significant (P < 0.01). The output of SNP heritability partitioning can be found in the Source Data available at https://github.com/wenjingk/TIR_code/ . Data Availability All the raw sequencing data are available under different accession numbers, as described in Supplementary Table 1 . The source data needed reproduce all the plots in the main and Supplementary Figures is available on GitHub at: https://github.com/wenjingk/TIR_code/ . Code Availability All the custom code used for processing and analyzing the data described in this study as well as to reproduce all the plots in the main and Supplementary Figures is available on GitHub at: https://github.com/wenjingk/TIR_code/ . Acknowledgements We acknowledge the RIKEN IMS Sequencing Platform and the Human Technopole National Genomics Facility for support with sequencing. We are grateful to Chiara Medaglia and Simona Pedrotti (Human Technopole), Simonetta Guarrera (Italian Institute of Genomic Medicine, Turin), Sara Bellomo and Enrico Berrino (Candiolo Cancer Institute, Turin) for helping with sample shipment and preparation and sequencing of the RNA-seq libraries from the NALF1 siRNA experiment, with very short notice before the 2024 Winter Holiday break. We acknowledge Nicola Pirastu (Human Technopole) for providing valuable suggestions on accessing and analyzing publically available GWAS datasets. We thank Andreas Lennartsson (Karolinska Institute) for facilitating the transfer of cells and reagents relevant to this project from RIKEN, Japan. The Hi-C, RNA-seq, and GPSeq data pre-processing was enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS) at UPPMAX (project no. 2024/22-849) funded by the Swedish Research Council (grant no. 2022-06725). W.H.Y. was supported by a PhD fellowship from the European Union’s Horizon 2020 Research and Innovation programme under the Marie Skłodowska-Curie Actions Innovative Training Network (MSCA ITN ‘Cell2Cell’ – grant no. 860675). B.A.M.B. was supported by a postdoctoral scholarship from the Karolinska Institutet Strategic Programme in Neurosciences (StratNeuro). This work was funded through research grants from the Swedish Brain Foundation (Hjärnfonden, grant no. PS2023-0023) to W.K.; the Swedish Research Council (VR, grant no. 2022-00721) and Karolinska Institutet (KI Consolidator Grants 2020) to N.C.; the RIKEN Center for Integrative Medical Sciences; the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) to P.C.; the Swedish Research Council (grant. no. 2020-02657), Karolinska Institutet (KI Consolidator Grants 2020), and the European Union (ERC, RADIALIS, GA n. 101088408) to M.B.; and intramural funding from Human Technopole to N.C., P.C., and M.B. Views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency. Neither the European Union nor the granting authority can be held responsible for them. References 1. ↵ Bonev , B. & Cavalli , G . Organization and function of the 3D genome . Nat. Rev. Genet . 17 , 661 – 678 ( 2016 ). OpenUrl CrossRef PubMed 2. ↵ Bouwman , B. A. M. & de Laat , W . Architectural hallmarks of the pluripotent genome . FEBS Lett . 589 , 2905 – 2913 ( 2015 ). OpenUrl CrossRef PubMed 3. ↵ Rahman , S. & Roussos , P . The 3D Genome in Brain Development: An Exploration of Molecular Mechanisms and Experimental Methods . Neurosci. Insights 19 , 26331055241293455 ( 2024 ). 4. ↵ L, L. , et al. Robust Hi-C Maps of Enhancer-Promoter Interactions Reveal the Function of Non-coding Genome in Neural Development and Diseases . Mol. Cell 79 , ( 2020 ). 5. ↵ Bonev , B. et al. Multiscale 3D Genome Rewiring during Mouse Neural Development . Cell 171 , 557 – 572 .e24 ( 2017 ). OpenUrl CrossRef PubMed 6. ↵ Rahman , S. et al. Lineage specific 3D genome structure in the adult human brain and neurodevelopmental changes in the chromatin interactome . Nucleic Acids Res . 51 , 11142 – 11161 ( 2023 ). OpenUrl CrossRef PubMed 7. ↵ H, L. , et al. Single-cell DNA methylome and 3D multi-omic atlas of the adult mouse brain . Nature 624 , ( 2023 ). 8. ↵ Won , H. et al. Chromosome conformation elucidates regulatory relationships in developing human brain . Nature 538 , 523 – 527 ( 2016 ). OpenUrl CrossRef PubMed 9. Carballo-Pacoret , P. , Carracedo , A. & Rodriguez-Fontenla , C . Unraveling the three-dimensional (3D) genome architecture in Neurodevelopmental Disorders (NDDs) . Neurogenetics 25 , 293 – 305 ( 2024 ). OpenUrl CrossRef PubMed 10. Spiess , K. & Won , H . Regulatory landscape in brain development and disease . Curr. Opin. Genet. Dev . 65 , 53 – 60 ( 2020 ). OpenUrl CrossRef PubMed 11. ↵ Hu , B. et al. Neuronal and glial 3D chromatin architecture informs the cellular etiology of brain disorders . Nat. Commun . 12 , 3968 ( 2021 ). OpenUrl CrossRef PubMed 12. ↵ Fujita , Y. & Yamashita , T . Spatial organization of genome architecture in neuronal development and disease . Neurochem. Int . 119 , 49 – 56 ( 2018 ). OpenUrl CrossRef PubMed 13. ↵ Rajarajan , P. et al. Neuron-specific signatures in the chromosomal connectome associated with schizophrenia risk . Science 362 , eaat4311 ( 2018 ). OpenUrl Abstract / FREE Full Text 14. ↵ Girdhar , K. et al. Chromatin domain alterations linked to 3D genome organization in a large cohort of schizophrenia and bipolar disorder brains . Nat. Neurosci . 25 , 474 – 483 ( 2022 ). OpenUrl CrossRef PubMed 15. ↵ Dueva , R. et al. Neutralization of the Positive Charges on Histone Tails by RNA Promotes an Open Chromatin Structure . Cell Chem. Biol . 26 , 1436 – 1449 .e5 ( 2019 ). OpenUrl CrossRef PubMed 16. ↵ Cremer , T. et al. Chromosome territories--a functional nuclear landscape . Curr. Opin. Cell Biol . 18 , 307 – 316 ( 2006 ). OpenUrl CrossRef PubMed Web of Science 17. ↵ Hall , L. L. et al. Stable C0T-1 repeat RNA is abundant and is associated with euchromatic interphase chromosomes . Cell 156 , 907 – 919 ( 2014 ). OpenUrl CrossRef PubMed Web of Science 18. ↵ Creamer , K. M. , Kolpa , H. J. & Lawrence , J. B . Nascent RNA scaffolds contribute to chromosome territory architecture and counter chromatin compaction . Mol. Cell 81 , 3509 – 3525 .e5 ( 2021 ). OpenUrl CrossRef PubMed 19. ↵ Lu , J. Y. et al. Homotypic clustering of L1 and B1/Alu repeats compartmentalizes the 3D genome . Cell Res . 31 , 613 – 630 ( 2021 ). OpenUrl CrossRef PubMed 20. ↵ Quinodoz , S. A. et al. RNA promotes the formation of spatial compartments in the nucleus . Cell 184 , 5775 – 5790 .e30 ( 2021 ). OpenUrl CrossRef PubMed 21. ↵ Bonetti , A. et al. RADICL-seq identifies general and cell type–specific principles of genome-wide RNA-chromatin interactions . Nat. Commun . 11 , ( 2020 ). 22. ↵ Lieberman-Aiden , E. et al. Comprehensive mapping of long range interactions reveals folding principles of the human genome . Science 326 , 289 – 293 ( 2009 ). OpenUrl Abstract / FREE Full Text 23. ↵ Girelli , G. et al. GPSeq reveals the radial organization of chromatin in the cell nucleus . Nat. Biotechnol . 38 , 1184 – 1193 ( 2020 ). OpenUrl CrossRef PubMed 24. ↵ Yekelchyk , M. , Li , X. , Guenther , S. & Braun , T . Single-Nucleus ATAC-seq for Mapping Chromatin Accessibility in Individual Cells of Murine Hearts . Methods Mol. Biol. Clifton NJ 2752 , 245 – 257 ( 2024 ). OpenUrl CrossRef 25. ↵ Takahashi , H. , Nishiyori-Sueki , H. , Ramilowski , J. A. , Itoh , M. & Carninci , P . Low Quantity Single Strand CAGE (LQ-ssCAGE) Maps Regulatory Enhancers and Promoters . Methods Mol. Biol. Clifton NJ 2351 , 67 – 90 ( 2021 ). OpenUrl CrossRef 26. ↵ Ameur , A. et al. Total RNA sequencing reveals nascent transcription and widespread co-transcriptional splicing in the human brain . Nat. Struct. Mol. Biol . 18 , 1435 – 1440 ( 2011 ). OpenUrl CrossRef PubMed 27. ↵ Gelali , E. et al. iFISH is a publically available resource enabling versatile DNA FISH to study genome architecture . Nat. Commun . 10 , 1636 ( 2019 ). OpenUrl CrossRef PubMed 28. ↵ Arun , G. , Aggarwal , D. & Spector , D. L . MALAT1 Long Non-Coding RNA: Functional Implications . Non-Coding RNA 6 , 22 ( 2020 ). 29. ↵ Bhat , P. et al. Genome organization around nuclear speckles drives mRNA splicing efficiency . Nature 629 , 1165 – 1173 ( 2024 ). OpenUrl CrossRef PubMed 30. ↵ Miyagawa , R. et al. Identification of cis- and trans-acting factors involved in the localization of MALAT-1 noncoding RNA to nuclear speckles . RNA N. Y. N 18 , 738 – 751 ( 2012 ). OpenUrl CrossRef 31. ↵ Hutchinson , J. N. et al. A screen for nuclear transcripts identifies two linked noncoding RNAs associated with SC35 splicing domains . BMC Genomics 8 , 39 ( 2007 ). 32. ↵ Chen , Y. et al. Mapping 3D genome organization relative to nuclear compartments using TSA-Seq as a cytological ruler . J. Cell Biol . 217 , 4025 – 4048 ( 2018 ). OpenUrl Abstract / FREE Full Text 33. ↵ Chen , Y. & Belmont , A. S . Genome organization around nuclear speckles . Curr. Opin. Genet. Dev . 55 , 91 – 99 ( 2019 ). OpenUrl CrossRef PubMed 34. ↵ Winick-Ng , W. et al. Cell-type specialization is encoded by specific chromatin topologies . Nature 599 , 684 – 691 ( 2021 ). OpenUrl CrossRef PubMed 35. ↵ Tarbier , M. et al. Nuclear gene proximity and protein interactions shape transcript covariations in mammalian single cells . Nat. Commun . 11 , 5445 ( 2020 ). OpenUrl CrossRef PubMed 36. ↵ Huang , X. et al. Downregulation of the silent potassium channel Kv8.1 increases motor neuron vulnerability in amyotrophic lateral sclerosis . Brain Commun . 6 , fcae202 ( 2024 ). OpenUrl CrossRef 37. ↵ Lennox , K. A. & Behlke , M. A . Cellular localization of long non-coding RNAs affects silencing by RNAi more than by antisense oligonucleotides . Nucleic Acids Res . 44 , 863 – 877 ( 2016 ). OpenUrl CrossRef PubMed 38. ↵ Gabel , H. W. et al. Disruption of DNA-methylation-dependent long gene repression in Rett syndrome . Nature 522 , 89 – 93 ( 2015 ). OpenUrl CrossRef PubMed 39. ↵ Gold , W. A. et al. Rett syndrome . Nat. Rev. Dis. Primer 10 , 84 ( 2024 ). OpenUrl CrossRef 40. ↵ Finucane , H. K. et al. Partitioning heritability by functional annotation using genome-wide association summary statistics . Nat. Genet . 47 , 1228 – 1235 ( 2015 ). OpenUrl CrossRef PubMed 41. ↵ Gazal , S. et al. Linkage disequilibrium-dependent architecture of human complex traits shows action of negative selection . Nat. Genet . 49 , 1421 – 1427 ( 2017 ). OpenUrl CrossRef PubMed 42. ↵ Davydov , E. V. et al. Identifying a high fraction of the human genome to be under selective constraint using GERP++ . PLoS Comput. Biol . 6 , e1001025 ( 2010 ). OpenUrl CrossRef PubMed 43. ↵ Zylka , M. J. , Simon , J. M. & Philpot , B. D . Gene length matters in neurons . Neuron 86 , 353 – 355 ( 2015 ). OpenUrl CrossRef PubMed 44. ↵ Wilusz , J. E. et al. A triple helix stabilizes the 3’ ends of long noncoding RNAs that lack poly(A) tails . Genes Dev . 26 , 2392 – 2407 ( 2012 ). OpenUrl Abstract / FREE Full Text 45. Carter , K. C. , Taneja , K. L. & Lawrence , J. B . Discrete nuclear domains of poly(A) RNA and their relationship to the functional organization of the nucleus . J. Cell Biol . 115 , 1191 – 1202 ( 1991 ). OpenUrl Abstract / FREE Full Text 46. ↵ Shopland , L. S. , Johnson , C. V. & Lawrence , J. B . Evidence that all SC-35 domains contain mRNAs and that transcripts can be structurally constrained within these domains . J. Struct. Biol . 140 , 131 – 139 ( 2002 ). OpenUrl CrossRef PubMed 47. ↵ Ray , T. A. et al. Comprehensive identification of mRNA isoforms reveals the diversity of neural cell-surface molecules with roles in retinal development and disease . Nat. Commun . 11 , 3328 ( 2020 ). OpenUrl CrossRef PubMed 48. ↵ Park , G. et al. Topology of gene regulatory compartments relative to the nuclear matrix . 2022.05.10.491284 Preprint at doi: 10.1101/2022.05.10.491284 ( 2022 ). OpenUrl Abstract / FREE Full Text 49. ↵ Roden , C. & Gladfelter , A. S . RNA contributions to the form and function of biomolecular condensates . Nat. Rev. Mol. Cell Biol . 22 , 183 – 195 ( 2021 ). OpenUrl CrossRef PubMed 50. ↵ Klobučar , T. et al. Integrative profiling of condensation-prone RNAs during early development . 2024.10.14.618182 Preprint at doi: 10.1101/2024.10.14.618182 ( 2025 ). OpenUrl Abstract / FREE Full Text 51. ↵ Zhao , Z. et al. Mega-Enhancer Bodies Organize Neuronal Long Genes in the Cerebellum . 2023.07.19.549737 Preprint at doi: 10.1101/2023.07.19.549737 ( 2023 ). OpenUrl Abstract / FREE Full Text 52. ↵ Leidescher , S. et al. Spatial organization of transcribed eukaryotic genes . Nat. Cell Biol . 24 , 327 – 339 ( 2022 ). OpenUrl CrossRef PubMed 53. ↵ Khyzha , N. , Ahmad , K. & Henikoff , S . Profiling transcriptome composition and dynamics within nuclear compartments using SLAM-RT&Tag . Mol. Cell 0 , ( 2025 ). 54. ↵ Rao , S. S. P. et al. Cohesin Loss Eliminates All Loop Domains . Cell 171 , 305 – 320 .e24 ( 2017 ). OpenUrl CrossRef PubMed 55. ↵ R, B. , et al. An atlas of endogenous DNA double-strand breaks arising during human neural cell fate determination . Sci. Data 9 , ( 2022 ). 56. ↵ Falk , A. et al. Capture of neuroepithelial-like stem cells from pluripotent stem cells provides a versatile system for in vitro production of human neurons . PloS One 7 , e29597 ( 2012 ). OpenUrl CrossRef PubMed 57. ↵ Fernandopulle , M. S. et al. Transcription Factor-Mediated Differentiation of Human iPSCs into Neurons . Curr. Protoc. Cell Biol . 79 , e51 ( 2018 ). OpenUrl CrossRef PubMed 58. ↵ Yip , C. W. et al. CFC-seq: identification of full-length capped RNAs unveil enhancer-derived transcription . 2024.10.31.620483 Preprint at doi: 10.1101/2024.10.31.620483 ( 2024 ). OpenUrl Abstract / FREE Full Text 59. ↵ Yip , C. W. et al. Antisense-oligonucleotide-mediated perturbation of long non-coding RNA reveals functional features in stem cells and across cell types . Cell Rep . 41 , 111893 ( 2022 ). 60. ↵ Holgersen , E. M. et al. Identifying high-confidence capture Hi-C interactions using CHiCANE . Nat. Protoc . 16 , 2257 – 2285 ( 2021 ). OpenUrl CrossRef PubMed 61. ↵ Shu , X. , Kato , M. , Takizawa , S. , Suzuki , Y. & Carninci , P . RADIP technology comprehensively identifies H3K27me3-associated RNA-chromatin interactions . Nucleic Acids Res . 52 , e104 ( 2024 ). OpenUrl CrossRef PubMed 62. ↵ Wernersson , E. et al. Deconwolf enables high-performance deconvolution of widefield fluorescence microscopy images . Nat. Methods 21 , 1245 – 1256 ( 2024 ). OpenUrl CrossRef PubMed 63. ↵ Wernersson , E. DOTTER (v0.0.1) . Zenodo doi: 10.5281/ZENODO.7112586 ( 2022 ). View the discussion thread. Back to top Previous Next Posted March 14, 2025. Download PDF Supplementary Material Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Cell type-specific intronic RNAs shape genome architecture during neuronal lineage specification Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Cell type-specific intronic RNAs shape genome architecture during neuronal lineage specification Wenjing Kang , Wing Hin Yip , Quentin Verron , Britta A.M. Bouwman , Xiaoze Li-Wang , Andrea Abou Yaghi , Mitsuyoshi Murata , Merula Stout , Lorenzo Salviati , Xufeng Shu , Kayoko Yasuzawa , Marco Gaviraghi , Rodrigo Pracana , Johan Lord , Roberto Ballarino , Anna Falk , Jay W. Shin , Takeya Kasukawa , Chi Wai Yip , Masaki Kato , Hazuki Takahashi , Nicola Crosetto , Piero Carninci , Magda Bienko bioRxiv 2025.03.13.641826; doi: https://doi.org/10.1101/2025.03.13.641826 Share This Article: Copy Citation Tools Cell type-specific intronic RNAs shape genome architecture during neuronal lineage specification Wenjing Kang , Wing Hin Yip , Quentin Verron , Britta A.M. Bouwman , Xiaoze Li-Wang , Andrea Abou Yaghi , Mitsuyoshi Murata , Merula Stout , Lorenzo Salviati , Xufeng Shu , Kayoko Yasuzawa , Marco Gaviraghi , Rodrigo Pracana , Johan Lord , Roberto Ballarino , Anna Falk , Jay W. Shin , Takeya Kasukawa , Chi Wai Yip , Masaki Kato , Hazuki Takahashi , Nicola Crosetto , Piero Carninci , Magda Bienko bioRxiv 2025.03.13.641826; doi: https://doi.org/10.1101/2025.03.13.641826 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Genomics Subject Areas All Articles Animal Behavior and Cognition (7622) Biochemistry (17650) Bioengineering (13871) Bioinformatics (41880) Biophysics (21424) Cancer Biology (18566) Cell Biology (25461) Clinical Trials (138) Developmental Biology (13365) Ecology (19866) Epidemiology (2067) Evolutionary Biology (24290) Genetics (15590) Genomics (22475) Immunology (17713) Microbiology (40328) Molecular Biology (17148) Neuroscience (88473) Paleontology (666) Pathology (2827) Pharmacology and Toxicology (4816) Physiology (7635) Plant Biology (15114) Scientific Communication and Education (2044) Synthetic Biology (4286) Systems Biology (9815) Zoology (2268)