CoChat
Sign Up
Cited by 713Chromosomal and Genetic VariationsCancer-related molecular mechanisms researchGenomics and Chromatin Dynamics

The DOI matches its Crossref record; 115 of the 115 references it cites were found in a registry or an index.

Transposable Elements Are Major Contributors to the Origin, Diversification, and Regulation of Vertebrate Long Noncoding RNAs

Open access
Kapusta, Aurélie, Kronenberg, Zev, Lynch, Vincent J., Zhuo, Xiaoyu, Ramsay, LeeAnn, Bourque, Guillaume, Yandell, Mark, Feschotte, Cédric·2013·PLoS Genetics
DOI 10.1371/journal.pgen.1003470PMC PMC3636048
Open this paper in CoChat — free accountOpen PDF

Abstract

Advances in vertebrate genomics have uncovered thousands of loci encoding long noncoding RNAs (lncRNAs). While progress has been made in elucidating the regulatory functions of lncRNAs, little is known about their origins and evolution. Here we explore the contribution of transposable elements (TEs) to the makeup and regulation of lncRNAs in human, mouse, and zebrafish. Surprisingly, TEs occur in more than two thirds of mature lncRNA transcripts and account for a substantial portion of total lncRNA sequence (~30% in human), whereas they seldom occur in protein-coding transcripts. While TEs contribute less to lncRNA exons than expected, several TE families are strongly enriched in lncRNAs. There is also substantial interspecific variation in the coverage and types of TEs embedded in lncRNAs, partially reflecting differences in the TE landscapes of the genomes surveyed. In human, TE sequences in lncRNAs evolve under greater evolutionary constraint than their non-TE sequences, than their intronic TEs, or than random DNA. Consistent with functional constraint, we found that TEs contribute signals essential for the biogenesis of many lncRNAs, including ~30,000 unique sites for transcription initiation, splicing, or polyadenylation in human. In addition, we identified ~35,000 TEs marked as open chromatin located within 10 kb upstream of lncRNA genes. The density of these marks in one cell type correlate with elevated expression of the downstream lncRNA in the same cell type, suggesting that these TEs contribute to cis-regulation. These global trends are recapitulated in several lncRNAs with established functions. Finally a subset of TEs embedded in lncRNAs are subject to RNA editing and predicted to form secondary structures likely important for function. In conclusion, TEs are nearly ubiquitous in lncRNAs and have played an important role in the lineage-specific diversification of vertebrate lncRNA repertoires.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1371/journal.pgen.1003470

Cites

115 of 115 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 TE occurrence in lncRNAs. See text, Methods and Table 1 for more details about lncRNA datasets. A. Percentage of transcripts with at least one exon overlapping with a TEs fragment (at least 10 bp). In red, lncRNAs (human = Gencode v13; mouse = both sets). Rest corresponds to human Refseq 57: in green, small non-coding RNAs (tRNAs and sno/miRNAs); in blue, protein-coding genes (pc genes) separated in exon types (coding and non-coding = UTRs); in black, pseudo = pseudogenes. B. Distribution of percentage of human lncRNA transcripts (Gencode v13) derived from TEs (more than 0% to more than 95%). The number of transcripts with more than 80% and more than 50% TE-derived DNA exons are indicated. Distribution is also shown for the subset of 36 studied lncRNAs presented in Table 2 and Table S2.
Figure 1 TE occurrence in lncRNAs. See text, Methods and Table 1 for more details about lncRNA datasets. A. Percentage of transcripts with at least one exon overlapping with a TEs fragment (at least 10 bp). In red, lncRNAs (human = Gencode v13; mouse = both sets). Rest corresponds to human Refseq 57: in green, small non-coding RNAs (tRNAs and sno/miRNAs); in blue, protein-coding genes (pc genes) separated in exon types (coding and non-coding = UTRs); in black, pseudo = pseudogenes. B. Distribution of percentage of human lncRNA transcripts (Gencode v13) derived from TEs (more than 0% to more than 95%). The number of transcripts with more than 80% and more than 50% TE-derived DNA exons are indicated. Distribution is also shown for the subset of 36 studied lncRNAs presented in Table 2 and Table S2.CC BY · PubMed Central
Figure 2 Coverage of different TE classes in genome, lncRNA, and protein-coding exons in human, mouse, and zebrafish. For genomes, total length (100%) corresponds to total length of assembly without gaps (human: 2,897 Mb. Mouse: 2,620 Mb. Zebrafish: 1,401 Mb). For lncRNAs, total length of genomic projection of all of exons are considered (human, Genc. = Gencode v13: 14.2 Mb. Human, Cabili set: 8.5 Mb. Mouse, Ens70 = Ensembl 70: 2.8 Mb. Mouse, Kutter: 0.15 Mb. Zebrafish: 2.3 Mb). For protein coding genes (pc genes), total length of CDS exons, 5′ and 3′UTR respectively are as follow: human, 30.9 Mb, 5.2 Mb, 24.6 Mb. Mouse: 30.5 Mb, 4.0 Mb, 21.6 Mb. Zebrafish: 19.1 Mb, 33.6 Mb, 12.5 Mb. Only pc genes from Refseq annotations with CDS and UTR features are considered (see Methods). Percentage of coverage of all TEs is indicated above bars.
Figure 2 Coverage of different TE classes in genome, lncRNA, and protein-coding exons in human, mouse, and zebrafish. For genomes, total length (100%) corresponds to total length of assembly without gaps (human: 2,897 Mb. Mouse: 2,620 Mb. Zebrafish: 1,401 Mb). For lncRNAs, total length of genomic projection of all of exons are considered (human, Genc. = Gencode v13: 14.2 Mb. Human, Cabili set: 8.5 Mb. Mouse, Ens70 = Ensembl 70: 2.8 Mb. Mouse, Kutter: 0.15 Mb. Zebrafish: 2.3 Mb). For protein coding genes (pc genes), total length of CDS exons, 5′ and 3′UTR respectively are as follow: human, 30.9 Mb, 5.2 Mb, 24.6 Mb. Mouse: 30.5 Mb, 4.0 Mb, 21.6 Mb. Zebrafish: 19.1 Mb, 33.6 Mb, 12.5 Mb. Only pc genes from Refseq annotations with CDS and UTR features are considered (see Methods). Percentage of coverage of all TEs is indicated above bars.CC BY · PubMed Central
Figure 3 Examples of lncRNAs with embedded TEs. Genomic DNA is represented as a grey line, transcripts are represented by a black line, with arrows showing sense of transcription and in grey boxes the exons of the mature transcript. TEs as colored boxes (orange-red: DNA TEs. Yellow: SINEs. Pink-purple: LTR/ERVs. Green: LINEs). Only TEs overlapping with lncRNA exons are represented. See also Table S2 for details of TEs in these lncRNAs. A. BANCR [79]. B. lnc-RoR [41]. Apes = gibbon, gorilla, orangutan, bonobo, chimpanzee, human. C. lnc-ES3 [43].
Figure 3 Examples of lncRNAs with embedded TEs. Genomic DNA is represented as a grey line, transcripts are represented by a black line, with arrows showing sense of transcription and in grey boxes the exons of the mature transcript. TEs as colored boxes (orange-red: DNA TEs. Yellow: SINEs. Pink-purple: LTR/ERVs. Green: LINEs). Only TEs overlapping with lncRNA exons are represented. See also Table S2 for details of TEs in these lncRNAs. A. BANCR [79]. B. lnc-RoR [41]. Apes = gibbon, gorilla, orangutan, bonobo, chimpanzee, human. C. lnc-ES3 [43].CC BY · PubMed Central
Figure 4 Evidence of purifying selection in TE–derived DNA transcribed as lncRNAs. LncRNAs correspond to Gencode v13 (human) and protein coding genes to Refseq 57 (human, 20,848 genes). Boxplots show primate PhyloP scores computed in order to compare the conservation of different sets (see upper panel). Random set is size and number-matched for TE-derived DNA in lncRNA exons. Intronic lncRNA TEs correspond to TE-derived DNA in lncRNA introns that don't overlapp with splicing sites and all annotated chromatin marks were removed (see Methods), in order to obtain a most neutral set [inactive chromatin, see 32]. Statistical test used: permutation test with 1000 permutations were performed in R. Boxplots depicts the median upper (75%) and lower (25%) quantiles. The whiskers extend beyond the upper and lower quantile by 1.5× the inner quartile range. Outliers have been removed for visualization.
Figure 4 Evidence of purifying selection in TE–derived DNA transcribed as lncRNAs. LncRNAs correspond to Gencode v13 (human) and protein coding genes to Refseq 57 (human, 20,848 genes). Boxplots show primate PhyloP scores computed in order to compare the conservation of different sets (see upper panel). Random set is size and number-matched for TE-derived DNA in lncRNA exons. Intronic lncRNA TEs correspond to TE-derived DNA in lncRNA introns that don't overlapp with splicing sites and all annotated chromatin marks were removed (see Methods), in order to obtain a most neutral set [inactive chromatin, see 32]. Statistical test used: permutation test with 1000 permutations were performed in R. Boxplots depicts the median upper (75%) and lower (25%) quantiles. The whiskers extend beyond the upper and lower quantile by 1.5× the inner quartile range. Outliers have been removed for visualization.CC BY · PubMed Central
Figure 5 Contribution of TEs to different gene features of lncRNAs. A. Schematic of the type of overlap between TE and lncRNA sequences. Upper panel shows an idealized lncRNA transcription unit, and lower panel shows a protein-coding gene (only genes with annotated 5′ and 3′UTRs were analyzed; see Methods). Exons (grey boxes) overlapping a TE are categorized based on the type of overlap: the TE may provide functional feature(s), as a transcription start site (TSS), the first exon (including TSS and splicing site: TSS+SPL), a splicing site (SPL), a middle exon (including the 2 splicing sites (Both SPL), a polyadenylation site (polyA), the last exon (including splicing site and PolyA: PolyA+SPL). A TE not overlapping with any feature is called exonized. B. Comparison between observed (Obs) and random (Rand) distribution (see Methods). Note that a given exon can belong to several categories since a given TE can hit different exons and therefore be counted multiple times. Unhit exons correspond to exons with no TE overlap. Human: lncRNAs from Gencode v13. Mouse: lincRNAs from Ensembl release 70. With the exception of ‘exonized’, ‘TSS’ and ‘polyA’ categories in mouse (p-values = 1, 0.001 and 0.298 respectively) and ‘exonized’ category in zebrafish (p-value = 0.001), the p-values were systematically <0.0001.
Figure 5 Contribution of TEs to different gene features of lncRNAs. A. Schematic of the type of overlap between TE and lncRNA sequences. Upper panel shows an idealized lncRNA transcription unit, and lower panel shows a protein-coding gene (only genes with annotated 5′ and 3′UTRs were analyzed; see Methods). Exons (grey boxes) overlapping a TE are categorized based on the type of overlap: the TE may provide functional feature(s), as a transcription start site (TSS), the first exon (including TSS and splicing site: TSS+SPL), a splicing site (SPL), a middle exon (including the 2 splicing sites (Both SPL), a polyadenylation site (polyA), the last exon (including splicing site and PolyA: PolyA+SPL). A TE not overlapping with any feature is called exonized. B. Comparison between observed (Obs) and random (Rand) distribution (see Methods). Note that a given exon can belong to several categories since a given TE can hit different exons and therefore be counted multiple times. Unhit exons correspond to exons with no TE overlap. Human: lncRNAs from Gencode v13. Mouse: lincRNAs from Ensembl release 70. With the exception of ‘exonized’, ‘TSS’ and ‘polyA’ categories in mouse (p-values = 1, 0.001 and 0.298 respectively) and ‘exonized’ category in zebrafish (p-value = 0.001), the p-values were systematically <0.0001.CC BY · PubMed Central
Figure 6 TE amounts and types in human lncRNA and their surrounding regions. Regions are genome, intergenic regions and exons. In the case of protein coding genes, exons include UTR exons as well as coding exons. 1 or 10 kb up and dw = intergenic regions up to 1 or 10 kb upstream of the TSS and downstream of the polyA respectively. Any annotated exons (RefSeq and Gencode v13 lncRNAs) have been subtracted from intergenic and intronic regions. A. Coverage of all TEs. LncRNA set corresponds to Gencode v13, separated in lincRNA transcripts (intergenic) and genic transcripts. Coverage is calculated as described for Figure 2 and in Methods and is shown per TE class (LTR/ERV, nonLTR/LINE, nonLTR/SINE, DNA) with an additional separation between ERVs (LTR/LTR) and internal parts (LTR/int) of ERV elements. B. Same as A, except that only TEs that overlap with DNaseI hypersensitive sites (‘TE-DHS’) are considered (see Methods). C. Heatmap of distance between LTR and lincRNA (left) and protein-coding genes (right) aggregated for all chromosomes (Jaccard test see Methods). The x-axis is the alignment of all reference features (protein coding exons and lncRNAs). The line depicts the total percentage of TEs found along the reference feature. The color quantifies the departure from null distribution generated from permutation. “Hot” (red) and “cool” (blue) colors mean that there was more or less TEs observed at a given position than by chance, respectively. All p–value <0.001.
Figure 6 TE amounts and types in human lncRNA and their surrounding regions. Regions are genome, intergenic regions and exons. In the case of protein coding genes, exons include UTR exons as well as coding exons. 1 or 10 kb up and dw = intergenic regions up to 1 or 10 kb upstream of the TSS and downstream of the polyA respectively. Any annotated exons (RefSeq and Gencode v13 lncRNAs) have been subtracted from intergenic and intronic regions. A. Coverage of all TEs. LncRNA set corresponds to Gencode v13, separated in lincRNA transcripts (intergenic) and genic transcripts. Coverage is calculated as described for Figure 2 and in Methods and is shown per TE class (LTR/ERV, nonLTR/LINE, nonLTR/SINE, DNA) with an additional separation between ERVs (LTR/LTR) and internal parts (LTR/int) of ERV elements. B. Same as A, except that only TEs that overlap with DNaseI hypersensitive sites (‘TE-DHS’) are considered (see Methods). C. Heatmap of distance between LTR and lincRNA (left) and protein-coding genes (right) aggregated for all chromosomes (Jaccard test see Methods). The x-axis is the alignment of all reference features (protein coding exons and lncRNAs). The line depicts the total percentage of TEs found along the reference feature. The color quantifies the departure from null distribution generated from permutation. “Hot” (red) and “cool” (blue) colors mean that there was more or less TEs observed at a given position than by chance, respectively. All p–value <0.001.CC BY · PubMed Central
Figure 7 Wordle representation of the most enriched TE families in lncRNAs. Colors refer to different TE classes: purple = LTR, green = LINE, yellow = SINE, red = DNA. A. See also Figure S4. Human lncRNA set is from Gencode v13, mouse is from Esembl. The expected and observed counts of fragments corresponding to each TE are calculated using RepeatMasker output (see Methods). Observed values are obtained by considering overlapping TEs lncRNA exons. Expected values are calculated based on the overall density of each TE family in the genome according to the RepeatMasker output assuming a random distribution of TE family members throughout the genome.). Only families statistically enriched in term of counts (fragment numbers) are kept (at least p-value<0.05, binomial distribution test) and only ratios above 2 are represented on wordle. For human sets, TEs with less than 5 fragments in lncRNAs are removed, 4 fragments for mouse and zebrafish. Size of the TE family name is proportional to its over-representation (scales of 5× or 10× are represented). B. Visual representation of the 25 most abundant TE families in the 3 species. Size of the TE family name is proportional to its percentage of TE derived DNA in the genome (scale of 2% is represented).
Figure 7 Wordle representation of the most enriched TE families in lncRNAs. Colors refer to different TE classes: purple = LTR, green = LINE, yellow = SINE, red = DNA. A. See also Figure S4. Human lncRNA set is from Gencode v13, mouse is from Esembl. The expected and observed counts of fragments corresponding to each TE are calculated using RepeatMasker output (see Methods). Observed values are obtained by considering overlapping TEs lncRNA exons. Expected values are calculated based on the overall density of each TE family in the genome according to the RepeatMasker output assuming a random distribution of TE family members throughout the genome.). Only families statistically enriched in term of counts (fragment numbers) are kept (at least p-value<0.05, binomial distribution test) and only ratios above 2 are represented on wordle. For human sets, TEs with less than 5 fragments in lncRNAs are removed, 4 fragments for mouse and zebrafish. Size of the TE family name is proportional to its over-representation (scales of 5× or 10× are represented). B. Visual representation of the 25 most abundant TE families in the 3 species. Size of the TE family name is proportional to its percentage of TE derived DNA in the genome (scale of 2% is represented).CC BY · PubMed Central
Figure 8 lncRNAs with cell-type specific expression are also associated with cell-type specific TE–DHS. Cell-type specific lncRNA based on RNA-Seq expression (cutoff of 10-fold higher) were identified in GM = GM12878, H1 or K562. Numbers of cell-type specific lncRNAs are written above graphs. For each lncRNA, only the most active proximal TE-DHS (<10 Kb) was retained and the distribution of normalized tag counts over these elements are shown in each cell type (*** for P<0.0001, ns for P>0.5.).
Figure 8 lncRNAs with cell-type specific expression are also associated with cell-type specific TE–DHS. Cell-type specific lncRNA based on RNA-Seq expression (cutoff of 10-fold higher) were identified in GM = GM12878, H1 or K562. Numbers of cell-type specific lncRNAs are written above graphs. For each lncRNA, only the most active proximal TE-DHS (<10 Kb) was retained and the distribution of normalized tag counts over these elements are shown in each cell type (*** for P<0.0001, ns for P>0.5.).CC BY · PubMed Central
Figure 9 Lineage-specific TE insertions in cyrano. Symbols and graphics are as in Figure 3. The structure of cyrano (lnc-oip5) [14] is based on coordinates of Gencode v13 transcript OIP5-AS1-001. Vertebrate PhastCons: peaks of sequence conservation across 46 vertebrate genomes displayed in the UCSC genome browser.
Figure 9 Lineage-specific TE insertions in cyrano. Symbols and graphics are as in Figure 3. The structure of cyrano (lnc-oip5) [14] is based on coordinates of Gencode v13 transcript OIP5-AS1-001. Vertebrate PhastCons: peaks of sequence conservation across 46 vertebrate genomes displayed in the UCSC genome browser.CC BY · PubMed Central
Figure 10 TE contribution to predicted lncRNA secondary structures. A. High and low TE content groups of 100 lncRNA were extracted from Gencode v13 set (TE content from 96.74% to 100% and from 0.49% to 2.27% respectively; see Table S7). P-values were calculated by Randfold and provide an indication of predicted secondary structure stability. The boxplot depicts the maximum, upper quantile, median, lower quantile and minimum value in a standard way. The mean of these 2 groups are significantly different by Wilcox rank sum test (p = 0.0022). B. Predicted secondary structures (RNAfold [115]) and compensatory mutations for two zebrafish lncRNAs containing ANGEL (DNA TEs) elements. In structures, TE derived regions are marked by solid line and base pairing probability by color spectrum (from 0 in violet to 1 in red). Zoom-in windows show part of stem with compensatory mutations: nucleotide substitution are boxed and the corresponding nucleotide found in ANGEL consensus are shown under/above actual RNA sequence. Sites of compensatory mutations are marked by asterisks and written p-values are adjusted by Bonferroni methods.
Figure 10 TE contribution to predicted lncRNA secondary structures. A. High and low TE content groups of 100 lncRNA were extracted from Gencode v13 set (TE content from 96.74% to 100% and from 0.49% to 2.27% respectively; see Table S7). P-values were calculated by Randfold and provide an indication of predicted secondary structure stability. The boxplot depicts the maximum, upper quantile, median, lower quantile and minimum value in a standard way. The mean of these 2 groups are significantly different by Wilcox rank sum test (p = 0.0022). B. Predicted secondary structures (RNAfold [115]) and compensatory mutations for two zebrafish lncRNAs containing ANGEL (DNA TEs) elements. In structures, TE derived regions are marked by solid line and base pairing probability by color spectrum (from 0 in violet to 1 in red). Zoom-in windows show part of stem with compensatory mutations: nucleotide substitution are boxed and the corresponding nucleotide found in ANGEL consensus are shown under/above actual RNA sequence. Sites of compensatory mutations are marked by asterisks and written p-values are adjusted by Bonferroni methods.CC BY · PubMed Central
Figure 11 Long stem in two human lincRNAs (Cabili set) formed by inverted TEs. Two examples of heavily edited human lincRNA transcripts with editing sites located in TEs. RNA structures are predicted by RNAfold [115]. Nucleotide color in structures indicates base pairing probability (from 0 to 1). Inverted TE pairs are marked by solid lines, the arrow illustrating TE strand. The stem pair of TCONS_00017795 is composed by inverted Alu elements, while the structure of TCONS_0001109 is formed by 2 LTRs (MLT2B3).
Figure 11 Long stem in two human lincRNAs (Cabili set) formed by inverted TEs. Two examples of heavily edited human lincRNA transcripts with editing sites located in TEs. RNA structures are predicted by RNAfold [115]. Nucleotide color in structures indicates base pairing probability (from 0 to 1). Inverted TE pairs are marked by solid lines, the arrow illustrating TE strand. The stem pair of TCONS_00017795 is composed by inverted Alu elements, while the structure of TCONS_0001109 is formed by 2 LTRs (MLT2B3).CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 115 references

Cited by

Sorted by citations

See all 713

Related

Metadata

Authors
Kapusta, A., Kronenberg, Z., Lynch, V. J., Zhuo, X., Ramsay, L., Bourque, G., Yandell, M., & Feschotte, C.
Affiliations
  • University of Utah (US)
  • University of Chicago (US)
  • McGill University and Génome Québec Innovation Centre (CA)
Corresponding author
Feschotte, Cédric
ORCID
Venue
PLoS Genetics
Publisher
Public Library of Science (PLoS)
Published
2013-04-25
Volume
9
Issue
4
Pages
e1003470
ISSN
  • 1553-7390 (ISSN-L)
  • 1553-7404
DOI
10.1371/journal.pgen.1003470
PMC
PMC3636048
OpenAlex
W2029868997
Cited by
687
References
115
Citations by year
2026: 21 · 2025: 27 · 2024: 60 · 2023: 52 · 2022: 57 · 2021: 70 · 2020: 53 · 2019: 52 · 2018: 68 · 2017: 58 · 2016: 50 · 2015: 62 · 2014: 42 · 2013: 13
FWCI
67.23
Type
journal article
Language
en
Open access
Licence
Locations
  • PLoS Genetics (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Knowledge@UChicago (University of Chicago) (submittedVersion, cc-by-nc-nd)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Figshare (submittedVersion, cc-by)
  • University of Chicago (cc-by)
Funders
  • National Science Foundation (1126998)
  • National Institutes of Health
  • Canadian Institutes of Health Research (115090_1)
  • National Human Genome Research Institute (R01HG004694)
  • National Institute of General Medical Sciences (T32GM007464, R01GM077582)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Chromosomal and Genetic Variations
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Cancer Research › Cancer-related molecular mechanisms research
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Chromatin Dynamics
Keywords
transposable elements, long non-coding RNA, lncRNA evolution, lncRNA biogenesis, cis-regulation, open chromatin, RNA editing, secondary structure, transcription initiation, splicing, polyadenylation, human, mice, zebrafish
MeSH terms
Animals, DNA Transposable Elements, Exons, Humans, Introns, Vertebrates, RNA, Long Noncoding

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review