Molecular Biology
Long Non-Coding RNAs
RNAs over 200 nucleotides that never make protein — scaffolds, guides, decoys, and enhancers that regulate the genome
A long non-coding RNA (lncRNA) is an RNA transcript longer than 200 nucleotides that is never translated into protein, yet regulates gene expression by folding into structures and binding DNA, RNA, and protein. lncRNAs act as molecular scaffolds that assemble complexes, guides that steer chromatin modifiers to precise genomic addresses, decoys that sequester regulators, and enhancers whose transcription marks active regulatory DNA. The archetype, Xist, is a ~17-kilobase lncRNA that coats an entire X chromosome and silences roughly a thousand genes to balance gene dosage between XX and XY cells. Although the ENCODE project reported in 2012 that ~80% of the human genome is transcribed, under 2% of it encodes protein — leaving on the order of 18,000 to 20,000 annotated lncRNA genes filling the so-called dark matter of the genome.
- Length cutoff>200 nucleotides
- Xist~17 kb, coats one X
- Genome transcribed~80% (ENCODE 2012)
- Protein-coding<2% of the genome
- lncRNA genes~18,000–20,000 (GENCODE)
- Four archetypesscaffold · guide · decoy · enhancer
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
Why long non-coding RNAs matter
- They rewrote the meaning of a gene. For decades RNA was seen as a passive intermediate — a working copy of a gene on its way to becoming protein. lncRNAs shattered that. Here the RNA is the functional molecule. That forced a reconceptualization of the ~98% of the human genome that does not encode protein, once dismissed as junk, and opened an entire regulatory layer.
- Dosage compensation depends on one. Every somatic cell in a female mammal silences one X chromosome using the lncRNA Xist. Without it, XX cells would express X-linked genes at twice the XY dose, which is lethal. Xist is the reason the calico cat's coat is patchwork — different X chromosomes are inactivated in different clonal patches of skin.
- They wire development. lncRNAs like HOTAIR and the transcripts flanking the HOX clusters help set the spatial patterns of Hox gene expression that define the body axis. Others, such as NEAT1, nucleate entire nuclear bodies (paraspeckles) by phase separation.
- They sit at the heart of imprinted loci. Parent-of-origin gene silencing at loci such as Igf2/H19 and Kcnq1 is executed by lncRNAs (H19, KCNQ1OT1, Airn) that spread silencing along the chromosome in cis.
- They are disease drivers and biomarkers. ANRIL occupies the 9p21 locus, the most robust common genetic risk region for coronary artery disease. HOTAIR and MALAT1 predict metastasis and survival in multiple cancers. Because many lncRNAs are exquisitely tissue-specific, they are appealing targets for antisense oligonucleotide drugs.
- They map the regulatory genome. Enhancer RNAs (eRNAs) are transcribed from active enhancers, so detecting them — via GRO-seq or CAGE — is one of the cleanest ways to find which of the millions of candidate enhancers are actually firing in a given cell.
How long non-coding RNAs work: four archetypes
Unlike microRNAs, which funnel through one canonical Argonaute-based pathway, lncRNAs have no single mechanism. Most molecular functions, however, fall into four widely used archetypes — scaffold, guide, decoy, and enhancer — and many lncRNAs combine several.
Scaffold. The lncRNA acts as a flexible structural platform, holding two or more proteins in the right relative orientation to form a functional complex. The classic case is HOTAIR (HOX transcript antisense intergenic RNA), a ~2.2-kb lncRNA transcribed from the HOXC locus. Its 5' domain binds the Polycomb repressive complex 2 (PRC2), which deposits the repressive H3K27me3 mark, while its 3' domain binds the LSD1/CoREST/REST complex, which removes the activating H3K4me2 mark. HOTAIR bridges a writer of repression and an eraser of activation into one machine. NEAT1 is a scaffold of a different kind — it seeds the assembly of paraspeckles, membraneless nuclear bodies, by liquid-liquid phase separation.
Guide. The lncRNA carries a chromatin-modifying protein complex to a specific place in the genome, providing the address that the enzyme alone lacks. It can act in cis, near its own transcription site (like Xist coating its own chromosome, or Airn and KCNQ1OT1 silencing imprinted neighbors), or in trans, at distant unlinked loci (like HOTAIR directing PRC2 to hundreds of sites across the genome). Targeting can rely on RNA-DNA base pairing, RNA-RNA interactions, or exploitation of three-dimensional genome folding.
Decoy. The lncRNA titrates a regulatory molecule away from its normal target. Gas5 folds into a structure that mimics the glucocorticoid response element and binds the glucocorticoid receptor, keeping it off its DNA targets. PANDA sequesters the transcription factor NF-YA. A related idea is the competing endogenous RNA or miRNA sponge: a lncRNA (or circular RNA) riddled with miRNA-binding sites soaks up microRNAs and relieves their repression of shared target mRNAs.
Enhancer. The lncRNA itself, or the act of transcribing it, boosts a neighboring gene. Enhancer RNAs (eRNAs) are produced bidirectionally from active enhancers; some stabilize the enhancer-promoter loop, load Mediator and cohesin, or evict the pausing factor NELF to release RNA polymerase II. A special class, activating ncRNAs (ncRNA-a), function like enhancers to raise the expression of nearby genes through the Mediator complex.
The Xist mechanism, step by step
Xist is the best-understood lncRNA and the cleanest illustration of the guide archetype at chromosome scale. In an early female embryo, both X chromosomes are active. The X-inactivation center integrates a counting-and-choice decision, and one X is selected to become inactive. From that chromosome, the ~17-kb Xist transcript is upregulated (its antisense partner Tsix is switched off), and it begins to accumulate.
Xist spreads in cis, coating the chromosome from which it is made without leaving it. Because chromosomes are folded, Xist reaches genomic regions that are far apart in sequence but physically close in nuclear space; it exploits the 3D architecture rather than crawling linearly. As it coats, Xist recruits silencing machinery through distinct sequence repeats. Its A-repeat binds the protein SPEN, which brings the NuRD complex and HDAC3 to deacetylate histones. Other repeats recruit hnRNPK, which nucleates Polycomb repressive complex 1 (PRC1) to deposit H2AK119 ubiquitylation, and this in turn helps recruit PRC2 to lay down H3K27me3. The chromosome loses active marks and gains repressive ones, is depleted of RNA polymerase II, replicates late in S phase, and finally condenses into the heterochromatic, DNA-methylated Barr body. Roughly a thousand genes on that copy fall silent — though a subset (~15% in humans) escapes inactivation.
lncRNA vs miRNA vs other RNAs
| Feature | lncRNA | miRNA | mRNA | eRNA |
|---|---|---|---|---|
| Length | >200 nt (often >1 kb) | ~22 nt | Hundreds to thousands nt | ~50–2000 nt |
| Codes protein? | No | No | Yes | No |
| Made by | RNA Pol II (mostly) | RNA Pol II → Drosha/Dicer | RNA Pol II | RNA Pol II (bidirectional) |
| Processing | Capped, spliced, polyadenylated | Hairpin cropped, then diced | Capped, spliced, polyadenylated | Usually uncapped/non-poly(A), unstable |
| Core mechanism | Scaffold / guide / decoy / enhancer | RISC-guided mRNA repression | Translated on ribosomes | Loops, loads Mediator, releases Pol II |
| Main location | Often nuclear (chromatin) | Mostly cytoplasmic | Cytoplasmic (translation) | At the enhancer locus |
| Targets | Chromatin, transcription, splicing, mRNA | 3'UTRs of many mRNAs | — | Its target promoter in cis |
| Example | Xist, HOTAIR, MALAT1, NEAT1 | let-7, miR-21, miR-155 | β-globin mRNA | Estrogen-induced eRNAs |
How lncRNAs are classified by genomic position
| Class | Position relative to protein-coding genes | Representative example |
|---|---|---|
| Intergenic (lincRNA) | Between genes, own transcription unit | HOTAIR, lincRNA-p21 |
| Antisense | Overlaps a gene on the opposite strand | Airn, KCNQ1OT1, Tsix |
| Intronic | Entirely within an intron of another gene | COLDAIR (in Arabidopsis FLC) |
| Sense/overlapping | Overlaps exons on the same strand | Various |
| Bidirectional/divergent | Shares a promoter, transcribed the other way | Many developmental lncRNAs |
| Enhancer-derived (eRNA) | Transcribed from an active enhancer | Activity-induced eRNAs |
Common misconceptions
- "Non-coding" means junk. The label refers only to the absence of a translated protein product. Many lncRNAs are highly regulated, conserved in structure or synteny, and required for viability — Xist knockout is embryonic-lethal in female mice. That said, not every transcript is functional (see below), so neither extreme is correct.
- The 200-nucleotide cutoff is biologically meaningful. It is not. It is a purely operational threshold that historically separated lncRNAs from the small-RNA fraction on a size-selection gel or in a sequencing library. Nothing special happens to an RNA at exactly 200 nt.
- lncRNAs never make peptides. Ribosome profiling has revealed that some annotated lncRNAs contain small open reading frames (sORFs) that are translated into functional micropeptides — for example, the muscle regulator myoregulin. The coding/non-coding boundary is fuzzier than the categories suggest.
- All lncRNAs work in trans. Many of the best-validated lncRNAs work strictly in cis, acting only on the chromosome that made them (Xist, Airn, KCNQ1OT1). Some function only through the act of transcription — transcriptional interference — and the mature RNA is dispensable.
- Transcription proves function. ENCODE's finding that ~80% of the genome is transcribed does not mean 80% is functional. Pervasive, low-level transcription from leaky promoters can produce noise. Demonstrating that a specific lncRNA does something requires conservation, a loss-of-function phenotype tied to the RNA itself, and a defined mechanism.
- lncRNAs and miRNAs are the same kind of thing. They share the "non-coding" umbrella but differ in size, biogenesis, and mechanism. miRNAs are ~22-nt guides for RISC-mediated mRNA silencing; lncRNAs are >200-nt multifunctional regulators. Confusingly, some lncRNAs act as host transcripts that are processed into miRNAs, or as sponges that sequester them.
Famous experiments and history
- H19 and Xist, the first mammalian lncRNAs (1990–1992). H19 was described by Shirley Tilghman's group as an abundant, spliced, polyadenylated, imprinted transcript with no protein product. Soon after, several groups (Willard, Brown, Ballabio, Rastan, Brockdorff) identified Xist at the X-inactivation center and showed it was expressed only from the inactive X — the founding examples of functional long non-coding RNA.
- HOTAIR acts in trans (Rinn & Chang, 2007). John Rinn and Howard Chang showed that a lincRNA transcribed from the HOXC cluster represses the distant HOXD cluster by recruiting PRC2. It was the first clear demonstration that a lncRNA could guide a chromatin complex to unlinked target genes across the genome — the guide/scaffold paradigm.
- ENCODE and pervasive transcription (2007, 2012). The ENCODE consortium reported that the overwhelming majority of the genome is transcribed and that biochemical activity extends far beyond protein-coding exons. The provocative claim that ~80% of the genome is "functional" ignited a still-running debate about how to define function versus noise.
- Xist tethering screens (2015–2017). Using RNA antisense purification (RAP-MS) and CRISPR screens, the Guttman, Lee, Brockdorff, and Zhang labs identified the direct protein partners of Xist — SPEN, hnRNPK, LBR, and others — and mapped how the A-repeat and other domains recruit HDAC3, PRC1, and PRC2 to execute silencing.
- MALAT1 and NEAT1 as nuclear architects. MALAT1 was found highly expressed in metastatic lung cancer and shown to regulate serine/arginine-rich splicing factors; NEAT1 was shown to be the essential architectural RNA that nucleates paraspeckles by phase separation — cementing the idea that lncRNAs can organize the nucleus itself.
Frequently asked questions
What is a long non-coding RNA?
A long non-coding RNA (lncRNA) is an RNA transcript longer than 200 nucleotides that is not translated into a protein. The 200-nucleotide cutoff is an operational definition — it separates lncRNAs from the small non-coding RNAs (microRNAs, siRNAs, piRNAs, snoRNAs) that are shorter — and has nothing to do with function. Most lncRNAs are made by RNA polymerase II, receive a 5' 7-methylguanosine cap, are spliced, and carry a poly(A) tail, so they look like messenger RNAs on a sequencing read; the difference is that they lack a productive open reading frame and are not loaded onto ribosomes to make protein. Instead they work as RNA molecules — folding into structures, base-pairing with other nucleic acids, and binding proteins. lncRNAs tend to be expressed at lower levels than mRNAs, are more tissue- and cell-type-specific, and are frequently localized to the nucleus, where many of them regulate chromatin and transcription.
How does Xist silence the X chromosome?
Xist (X-inactive specific transcript) is a ~17-kilobase lncRNA transcribed from the X-inactivation center. In female mammalian cells, one of the two X chromosomes is silenced for dosage compensation, and Xist is the master switch. Xist is transcribed only from the future inactive X and spreads in cis, physically coating that chromosome without leaving it. Coating is guided partly by three-dimensional chromosome architecture, so Xist reaches distant regions that are close in nuclear space. Along the way Xist recruits silencing machinery: the protein SPEN (via its A-repeat) brings the NuRD complex and HDAC3 to deacetylate histones, and Polycomb repressive complexes PRC1 and PRC2 deposit the repressive marks H2AK119ub and H3K27me3. The chromosome loses active marks, gains repressive ones, replicates late, and condenses into the heterochromatic Barr body, silencing roughly a thousand genes on that copy.
What is the difference between lncRNA and miRNA?
Both are non-coding RNAs, but they differ in size and mechanism. MicroRNAs (miRNAs) are ~22 nucleotides long, processed from hairpin precursors by Drosha and Dicer, and loaded into Argonaute to form the RISC complex; a miRNA guides RISC to partially complementary sites in the 3' untranslated regions of target mRNAs, repressing translation and destabilizing the message. Long non-coding RNAs are over 200 nucleotides — often thousands — and act through diverse mechanisms rather than one canonical pathway: they scaffold multi-protein complexes, guide chromatin modifiers to specific genomic sites, decoy transcription factors or even sponge miRNAs away from their targets, and mark active enhancers. A miRNA typically works in the cytoplasm on many mRNAs at once; many lncRNAs work in the nucleus on chromatin and transcription. Some lncRNAs even give rise to miRNAs, and both classes fall under the broad umbrella of the non-coding transcriptome.
What are enhancer RNAs?
Enhancer RNAs (eRNAs) are non-coding transcripts produced from active enhancers — the distal regulatory elements that loop to and boost the transcription of their target promoters. When an enhancer is engaged, RNA polymerase II transcribes it bidirectionally, generating short, often unstable, largely non-polyadenylated eRNAs. eRNA production is one of the most reliable molecular signatures of an active enhancer: genome-wide run-on sequencing (GRO-seq) detects the paired divergent transcription that marks these elements. eRNAs are not merely transcriptional noise — several stabilize enhancer-promoter loops, help load the Mediator and cohesin complexes, and titrate the transcriptional repressor NELF to release paused polymerase. Because they are cell-type specific and appear rapidly in response to signals such as estrogen or neuronal activity, eRNAs are used to map the active regulatory landscape of a given cell state.
How much of the human genome is non-coding?
About 98 to 99% of the human genome does not code for protein. Protein-coding exons occupy under 2% of the roughly 3.2 billion base pairs, and there are only about 19,000 to 20,000 protein-coding genes. Yet the ENCODE project reported in 2012 that around 80% of the genome shows some reproducible biochemical activity — it is transcribed into RNA, bound by proteins, or marked by chromatin signatures. That figure is debated: 'biochemically active' is not the same as 'functional,' and much pervasive transcription may be low-level noise or the byproduct of an accessible chromatin state. Even under stricter, evolution-based definitions of function, however, the non-coding fraction that is conserved and regulatory is far larger than the coding fraction. GENCODE currently annotates on the order of 18,000 to 20,000 lncRNA genes, a number comparable to the count of protein-coding genes, though how many of these transcripts are truly functional remains an open and active question.
Do long non-coding RNAs cause disease?
Yes. Many lncRNAs are dysregulated in disease and some are causal. HOTAIR overexpression, which retargets the PRC2 chromatin-silencing complex, correlates with metastasis and poor prognosis in breast, colorectal, and other cancers. MALAT1 (a nuclear lncRNA that regulates splicing-factor phosphorylation) is a prognostic marker in lung adenocarcinoma. ANRIL, transcribed antisense to the INK4 tumor-suppressor locus, sits in the 9p21 region — the single most reproducible common risk locus for coronary artery disease and type 2 diabetes in genome-wide association studies. At imprinted loci, misregulation of lncRNAs such as H19, KCNQ1OT1, and Airn contributes to overgrowth syndromes like Beckwith-Wiedemann and to Silver-Russell syndrome. Because lncRNAs are often tissue-specific, they are attractive therapeutic targets: antisense oligonucleotides against MALAT1 and other lncRNAs have shown efficacy in preclinical cancer and neurological models.
Are all long non-coding RNAs functional?
No, and this is one of the most contested questions in genome biology. Pervasive transcription means most of the genome is copied into RNA at least occasionally, but transcription alone does not prove that the resulting RNA does anything. Some lncRNAs matter only because the act of transcribing them — not the transcript itself — remodels the local chromatin or interferes with a neighboring promoter, a mechanism called transcriptional interference. Others are almost certainly non-functional transcriptional noise from leaky, accessible promoters. Rigorous criteria are needed to call a lncRNA functional: sequence or structural conservation across species, a reproducible loss-of-function phenotype when the RNA (not just its DNA) is depleted, a defined molecular mechanism, and specific expression. Well-validated lncRNAs like Xist, HOTAIR, and NEAT1 pass these bars; the large majority of annotated lncRNAs have not yet been tested, so the true count of functional lncRNAs is unknown.