CoChat

DLX5/6 GABAergic Expression Affects Social Vocalization: Implications for Human Evolution

Open access

Published version of a preprint

The DOI matches its Crossref record; 110 of the 114 references it cites were found in a registry or an index; 27 of the 42 sources its claims cite were checked for support.

Abstract

DLX5 and DLX6 are two closely related transcription factors involved in brain development and in GABAergic differentiation. The DLX5/6 locus is regulated by FoxP2, a gene involved in language evolution and has been associated to neurodevelopmental disorders and mental retardation. Targeted inactivation of Dlx5/6 in mouse GABAergic neurons (Dlx5/6VgatCre mice) results in behavioural and metabolic phenotypes notably increasing lifespan by 33%. Here, we show that Dlx5/6VgatCre mice present an hyper-vocalization and hyper-socialization phenotype. We identify an introgressed Neanderthal haplotype (DLX5/6-N-Haplotype) present in 12.6% of European individuals that covers DLX5/6 coding and regulatory sequences. The DLX5/6-N-Haplotype includes the binding site for GTF2I, a gene associated to Williams-Beuren syndrome.

Overview

CoChat Summary

## TL;DR

The paper reports that reducing Dlx5/6 expression in mouse GABAergic neurons increases social ultrasonic vocalizations and social interaction, and links this phenotype to a Neanderthal-derived DLX5/6 haplotype in humans. It also reports that this haplotype is underrepresented in Italian semi-supercentenarians, suggesting a possible connection between DLX5/6 regulation, sociability, and healthy aging.

## Key findings

  • In resident/intruder tests, 30% (3/10) of Dlx5/6VgatCre/+ mice and 56% (5/9) of Dlx5/6VgatCre mice emitted more than 700 calls/10 min, versus 14/15 (93%) of controls emitting less than 700 calls/10 min; some mutants emitted up to 1,400 calls/10 min.
  • High-vocalizing mice spent more than twice the time following or approaching the intruder than low-vocalizing mice, indicating a hyper-sociability phenotype.
  • The identified DLX5/6-N-Haplotype is a 37,160 bp haplotype defined by five SNPs; the paper reports it is present in 12.6% of European individuals / 6.3% of EUR alleles and is absent from continental Africa.
  • In 81 Italian semi-supercentenarians (105+), the haplotype was underrepresented relative to controls: P = 0.026 versus the ITA control group, and the paper also reports a 50% under-representation compared with other European populations; a GeneATLAS association scan found the most significant phenotype was standing height (P = 8.92E−08 for rs1207733-G).

## Methods

  • Mouse experiment: adult female control, Dlx5/6VgatCre/+, and Dlx5/6VgatCre mice were tested in a resident/intruder paradigm after 3 days of isolation; all analyzed animals were 6 months old.
  • Vocalization analysis: ultrasonic calls were recorded for 10 min and categorized into short, simple, complex, frequency jumps, and unstructured; genotype comparisons used Wilcoxon tests and χ2 tests with Bonferroni correction.
  • Human genetic analyses: the authors used 1000 Genomes data, MSSNG, UK Biobank/GeneATLAS, PheWeb, and 81 Italian semi-supercentenarians; they assessed haplotype structure, linkage disequilibrium, introgression likelihood, Fst, and case/control frequency differences.

## Limitations

  • The authors state that the phenotype/genotype association analyses in UK Biobank and other platforms had “fairly high P values” and therefore do not permit a solid correlation.
  • The paper reports a nonsignificant association with ASD in the MSSNG database (P = 0.16), so the human trait links are not uniformly supported.
  • Evident from the text: the mouse social-vocalization sample sizes are modest (10, 9, and 15 females), and the human longevity analysis is based on a relatively small 81-person semi-supercentenarian cohort.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1093/molbev/msab181

Cites

110 of 110 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1. Number of calls/10 min emitted by Dlx5/6VgatCre/+ and Dlx5/6VgatCre mice. Number of calls emitted by of Dlx5/6VgatCre/+ (n = 10), Dlx5/6VgatCre (n = 9), and control (n = 15) 6 months old female mice confronted to a genetically normal female mouse intruder for 10 min. The number and types of vocalization were quantified. (A) 93% of control mice emitted less than 700 vocalizations during the 10 min tests, only one outlier emitted 849 calls. In contrast, 30% of VgatΔDlx5-6/+ (3/10) and 56% of VgatΔDlx5-6 (5/9) mice emitted between 850 and 1400 calls/10 min. (B) Percentage of mice emitting less (pink) or more (red) than 700 calls/10 min. The difference between each genotype is highly significant (***P < 0.001, Chi-square test).
Fig. 1. Number of calls/10 min emitted by Dlx5/6VgatCre/+ and Dlx5/6VgatCre mice. Number of calls emitted by of Dlx5/6VgatCre/+ (n = 10), Dlx5/6VgatCre (n = 9), and control (n = 15) 6 months old female mice confronted to a genetically normal female mouse intruder for 10 min. The number and types of vocalization were quantified. (A) 93% of control mice emitted less than 700 vocalizations during the 10 min tests, only one outlier emitted 849 calls. In contrast, 30% of VgatΔDlx5-6/+ (3/10) and 56% of VgatΔDlx5-6 (5/9) mice emitted between 850 and 1400 calls/10 min. (B) Percentage of mice emitting less (pink) or more (red) than 700 calls/10 min. The difference between each genotype is highly significant (***P < 0.001, Chi-square test).CC BY-NC · PubMed Central
Fig. 2. Spectrograms of control and Dlx5/6VgatCre mice. Representative spectrograms from four controls (A–D) and four Dlx5/6VgatCre (E–H) mice. In each group of animals, some individuals (e.g., D, H) emitted very few vocalizations while others emitted sequences of calls separated by silent intervals (A–C, G). Hyper-vocalizing Dlx5/6VgatCre mice emitted long, uninterrupted, sequences of calls (E, F) which were never observed in controls or low-vocalizing animals.
Fig. 2. Spectrograms of control and Dlx5/6VgatCre mice. Representative spectrograms from four controls (A–D) and four Dlx5/6VgatCre (E–H) mice. In each group of animals, some individuals (e.g., D, H) emitted very few vocalizations while others emitted sequences of calls separated by silent intervals (A–C, G). Hyper-vocalizing Dlx5/6VgatCre mice emitted long, uninterrupted, sequences of calls (E, F) which were never observed in controls or low-vocalizing animals.CC BY-NC · PubMed Central
Fig. 3. Total vocalization time and call types emitted by hyper-vocalizing and low-vocalizing mice. (A) Total time spent vocalizing by mice emitting less or more than 700 calls/10 min. (B) Percentage of call types produced by mice emitting less or more than 700 calls/10 min. Five types of emissions have been considered as in Ey, et al. (2013): Short, Simple (Flat + Upward + Downward), Complex (Modulated + Complex), Unstructured, and Frequency jumps. Hyper-vocalizing animals vocalize, in average, four times more than low vocalizing animals and emit significantly more complex and modulated calls. (***P < 0.001; *P < 0.05, Mann–Whitney U test).
Fig. 3. Total vocalization time and call types emitted by hyper-vocalizing and low-vocalizing mice. (A) Total time spent vocalizing by mice emitting less or more than 700 calls/10 min. (B) Percentage of call types produced by mice emitting less or more than 700 calls/10 min. Five types of emissions have been considered as in Ey, et al. (2013): Short, Simple (Flat + Upward + Downward), Complex (Modulated + Complex), Unstructured, and Frequency jumps. Hyper-vocalizing animals vocalize, in average, four times more than low vocalizing animals and emit significantly more complex and modulated calls. (***P < 0.001; *P < 0.05, Mann–Whitney U test).CC BY-NC · PubMed Central
Fig. 4. Expression of Dlx5 in different adult brain regions involved in vocalization. X-gal staining on serial sections of Dlx5lacZ/+ brains. ß-Galactosidase activity was not detected in the brainstem (A), in the cerebellum and in the midbrain including the ventro-lateral periacqueductal gray (vlPAG) a region which integrates sensory and cognitive inputs and controls vocalization motor centers. Dlx5 was expressed in the prefrontal cortex (B) and in other cortical regions (cx), in the striatum (C) and in hypothalamic nuclei (D). AHP, anterior hypothalamic area; cc, corpus callosum; cx, cortex; DM, Dorsomedial hypothalamus; PVN, para-ventricular nucleus; SCh, suprachiasmatic nucleus; VMH, ventro medial hypothalamus.
Fig. 4. Expression of Dlx5 in different adult brain regions involved in vocalization. X-gal staining on serial sections of Dlx5lacZ/+ brains. ß-Galactosidase activity was not detected in the brainstem (A), in the cerebellum and in the midbrain including the ventro-lateral periacqueductal gray (vlPAG) a region which integrates sensory and cognitive inputs and controls vocalization motor centers. Dlx5 was expressed in the prefrontal cortex (B) and in other cortical regions (cx), in the striatum (C) and in hypothalamic nuclei (D). AHP, anterior hypothalamic area; cc, corpus callosum; cx, cortex; DM, Dorsomedial hypothalamus; PVN, para-ventricular nucleus; SCh, suprachiasmatic nucleus; VMH, ventro medial hypothalamus.CC BY-NC · PubMed Central
Fig. 5. Social behavior of hyper-vocalizing and low-vocalizing mice. The social interaction behavior of Dlx5/6VgatCre/+ (n = 10), Dlx5/6VgatCre (n = 9), and control (n = 15) mice placed in the same cage with a control visitor was monitored using a protocol (de Chaumont, et al. 2019; de Lombares, et al. 2019), which associates computer vision and artificial intelligence to identify and categorize individual social interactions. The types of behavior recorded were classified as follows: 1) isolated behavior (moving alone, rearing alone, stopped alone, jump alone, huddling alone), 2) static interactions (rearing in contact, oral–oral contact, oro-genital contact, side-by-side in same or opposite orientation, 3) dynamic interactions (moving in contact, train of 2 mice, following the same path), 4) Social approach (making contact, approaching in rearing, approaching), 5) social escape (breaking contact, escaping), a detailed description of each behavior is found in (de Chaumont, et al. 2019). The time spent by each mouse (s) adopting a specific behavior was measured over a 10 min recording. (A) Normalized results: for each behavior, the average time spent by low-vocalizing animals was set as one (pink) and the ratio with the time spent by hyper-vocalizing mice shown (red). For example, “Dynamic Interaction” was adopted 2.3 times more frequently by hyper-vocalizing mice than by low-vocalizing animals. (B) Actual values in seconds recorded for each social behavioral trait. (***P < 0.001; *P < 0.05, Mann–Whitney U test).
Fig. 5. Social behavior of hyper-vocalizing and low-vocalizing mice. The social interaction behavior of Dlx5/6VgatCre/+ (n = 10), Dlx5/6VgatCre (n = 9), and control (n = 15) mice placed in the same cage with a control visitor was monitored using a protocol (de Chaumont, et al. 2019; de Lombares, et al. 2019), which associates computer vision and artificial intelligence to identify and categorize individual social interactions. The types of behavior recorded were classified as follows: 1) isolated behavior (moving alone, rearing alone, stopped alone, jump alone, huddling alone), 2) static interactions (rearing in contact, oral–oral contact, oro-genital contact, side-by-side in same or opposite orientation, 3) dynamic interactions (moving in contact, train of 2 mice, following the same path), 4) Social approach (making contact, approaching in rearing, approaching), 5) social escape (breaking contact, escaping), a detailed description of each behavior is found in (de Chaumont, et al. 2019). The time spent by each mouse (s) adopting a specific behavior was measured over a 10 min recording. (A) Normalized results: for each behavior, the average time spent by low-vocalizing animals was set as one (pink) and the ratio with the time spent by hyper-vocalizing mice shown (red). For example, “Dynamic Interaction” was adopted 2.3 times more frequently by hyper-vocalizing mice than by low-vocalizing animals. (B) Actual values in seconds recorded for each social behavioral trait. (***P < 0.001; *P < 0.05, Mann–Whitney U test).CC BY-NC · PubMed Central
Fig. 6. Linkage disequilibrium analysis of the distribution of the DLX5/6-N-Haplotype in different populations. (A) For each population, the width of the color bar represents the allele frequency for each SNP. SNPs linked by a solid bar cosegregate; if cosegregation is less than 100%, the width of the bar diminishes accordingly. The allele frequency of the most represented Neanderthal SNP of the DLX5/6-N-Haplotype is indicated for each population. The major SNPs constituting the haplotype are indicated. The scale bar represents a 5% allelic frequency within a population. EAS, East Asia populations; SAS, South Asia populations; EUR, European populations; AMR, American populations; AMR/AFR, Afro-americans; AFR, Africans. (B) Geographical distribution of the DLX5/6-N-Haplotype. The allele frequency of the most represented SNP constituting the DLX5/6-N-Haplotype is represented on the earth map. The higher frequency of the DLX5/6-N-Haplotype is seen in Europe and South Asia, the haplotype is not present in African populations and in Peruvians from Lima. CHB, Han Chinese in Beijing, China; JPT, Japanese in Tokyo, Japan; CHS, Southern Han Chinese; CDX, Chinese Dai in Xishuangbanna, China; KHV, Kinh in Ho Chi Minh City, Vietnam; CEU, Utah Residents (CEPH) with Northern and Western European Ancestry; TSI, Tuscans in Italy; FIN, Finnish in Finland; GBR, British in England and Scotland; IBS, Iberian Population in Spain; YRI, Yoruba in Ibadan, Nigeria; LWK, Luhya in Webuye, Kenya; GWD, Gambian in Western Divisions in the Gambia; MSL, Mende in Sierra Leone; ESN, Esan in Nigeria; ASW, Americans of African Ancestry in South-West USA; ACB, African Caribbeans in Barbados; MXL, Mexican Ancestry from Los Angeles USA; PUR, Puerto Ricans from Puerto Rico; CLM, Colombians from Medellin, Colombia; PEL, Peruvians from Lima, Peru; GIH, Gujarati Indian from Houston, Texas; PJL: Punjabi from Lahore, Pakistan; BEB: Bengali from Bangladesh; STU: Sri Lankan Tamil from the UK; ITU, Indian Telugu from the UK. (C) Neighbor-joining tree computed on five modern human haplotypes (I–V) identified through haplotype network analysis, three archaic human haplotypes (Vindija, Denisovan, and Altai) and the chimpanzee. The values in blue boxes represent the support for each branch (1,000 bootstraps). The pie charts point out the frequency in the five continental population groups and in the semi-supercentenarians of each modern human core haplotype. The number reported near each pie-chart indicates the frequency of each core-haplotype among 1000 Genomes individuals and, in brackets, the frequency of the core haplotypes among the semi-supercentenarians.
Fig. 6. Linkage disequilibrium analysis of the distribution of the DLX5/6-N-Haplotype in different populations. (A) For each population, the width of the color bar represents the allele frequency for each SNP. SNPs linked by a solid bar cosegregate; if cosegregation is less than 100%, the width of the bar diminishes accordingly. The allele frequency of the most represented Neanderthal SNP of the DLX5/6-N-Haplotype is indicated for each population. The major SNPs constituting the haplotype are indicated. The scale bar represents a 5% allelic frequency within a population. EAS, East Asia populations; SAS, South Asia populations; EUR, European populations; AMR, American populations; AMR/AFR, Afro-americans; AFR, Africans. (B) Geographical distribution of the DLX5/6-N-Haplotype. The allele frequency of the most represented SNP constituting the DLX5/6-N-Haplotype is represented on the earth map. The higher frequency of the DLX5/6-N-Haplotype is seen in Europe and South Asia, the haplotype is not present in African populations and in Peruvians from Lima. CHB, Han Chinese in Beijing, China; JPT, Japanese in Tokyo, Japan; CHS, Southern Han Chinese; CDX, Chinese Dai in Xishuangbanna, China; KHV, Kinh in Ho Chi Minh City, Vietnam; CEU, Utah Residents (CEPH) with Northern and Western European Ancestry; TSI, Tuscans in Italy; FIN, Finnish in Finland; GBR, British in England and Scotland; IBS, Iberian Population in Spain; YRI, Yoruba in Ibadan, Nigeria; LWK, Luhya in Webuye, Kenya; GWD, Gambian in Western Divisions in the Gambia; MSL, Mende in Sierra Leone; ESN, Esan in Nigeria; ASW, Americans of African Ancestry in South-West USA; ACB, African Caribbeans in Barbados; MXL, Mexican Ancestry from Los Angeles USA; PUR, Puerto Ricans from Puerto Rico; CLM, Colombians from Medellin, Colombia; PEL, Peruvians from Lima, Peru; GIH, Gujarati Indian from Houston, Texas; PJL: Punjabi from Lahore, Pakistan; BEB: Bengali from Bangladesh; STU: Sri Lankan Tamil from the UK; ITU, Indian Telugu from the UK. (C) Neighbor-joining tree computed on five modern human haplotypes (I–V) identified through haplotype network analysis, three archaic human haplotypes (Vindija, Denisovan, and Altai) and the chimpanzee. The values in blue boxes represent the support for each branch (1,000 bootstraps). The pie charts point out the frequency in the five continental population groups and in the semi-supercentenarians of each modern human core haplotype. The number reported near each pie-chart indicates the frequency of each core-haplotype among 1000 Genomes individuals and, in brackets, the frequency of the core haplotypes among the semi-supercentenarians.CC BY-NC · PubMed Central
Fig. 7. Association of different pathological conditions with the DLX5/6 genomic region. The central black line represents the 7q21.3 locus, which includes at least 5 genes DLX5, DLX6, SEM1, SLC25A13, and DYNC1I1. A long noncoding antisense RNA, DLX6-AS, is also transcribed from this region. Purple boxes in the upper part of the figure represent the phenotypic consequences observed in Split Hand Foot Malformation syndrome (SHFM) patients carrying deletion/mutations in this locus. While all SHFM patients present ectrodactyly, in certain cases this malformation is accompanied by craniofacial defects, hearing loss and/or cognitive defects. The closer the deletion are to DLX5/6 coding sequences the higher is the frequency of patients presenting also cognitive impairment. Point mutations within the DLX5/6 region and proximal or intragenic enhancers have been associated with autism spectrum disorders (ASD) without ectrodactyly or craniofacial defects and are included in the DLX5/6-N-Haplotype. Three conserved enhancers (I56i, I56ii, and Mef2-BA) that direct DLX5/6 expression in the brain (indicated in yellow) are also part of the haplotype. A point mutation within the I56i enhancer, prevents binding of GTF2I, a transcription factor deleted in Williams–Beuren syndrome, a condition characterized by hyper-vocalization and hyper-sociability (Poitras, et al. 2010).
Fig. 7. Association of different pathological conditions with the DLX5/6 genomic region. The central black line represents the 7q21.3 locus, which includes at least 5 genes DLX5, DLX6, SEM1, SLC25A13, and DYNC1I1. A long noncoding antisense RNA, DLX6-AS, is also transcribed from this region. Purple boxes in the upper part of the figure represent the phenotypic consequences observed in Split Hand Foot Malformation syndrome (SHFM) patients carrying deletion/mutations in this locus. While all SHFM patients present ectrodactyly, in certain cases this malformation is accompanied by craniofacial defects, hearing loss and/or cognitive defects. The closer the deletion are to DLX5/6 coding sequences the higher is the frequency of patients presenting also cognitive impairment. Point mutations within the DLX5/6 region and proximal or intragenic enhancers have been associated with autism spectrum disorders (ASD) without ectrodactyly or craniofacial defects and are included in the DLX5/6-N-Haplotype. Three conserved enhancers (I56i, I56ii, and Mef2-BA) that direct DLX5/6 expression in the brain (indicated in yellow) are also part of the haplotype. A point mutation within the I56i enhancer, prevents binding of GTF2I, a transcription factor deleted in Williams–Beuren syndrome, a condition characterized by hyper-vocalization and hyper-sociability (Poitras, et al. 2010).CC BY-NC · PubMed Central
Fig. 8. Allele frequency of the DLX5/6-N-Haplotype in the semi-supercentenarian population. (A) The allele frequency of DLX5/6-N-Haplotype European populations (1000 Genomes Project) is confronted to the CEN population of 81 semi-supercentenarians (105+ yo) and to the ITA population (Tuscany 1000 Genomes population [TSI] + 36 controls genomes matching the CEN). While the ITA population does not show any significant difference from the Europeans, a significant reduction of the DLX5/6-N-Haplotype allele frequency is observed in the semi-supercentenarian population (*P < 0.05, Fst distribution). (B) Density plot build considering all the Fst values calculated considering all the archaic haplotypes described in Skov, et al. (2020) for CEN versus ITA groups. The blue arrow indicates the value of Fst calculated for the DLX5/6-N-Haplotype.
Fig. 8. Allele frequency of the DLX5/6-N-Haplotype in the semi-supercentenarian population. (A) The allele frequency of DLX5/6-N-Haplotype European populations (1000 Genomes Project) is confronted to the CEN population of 81 semi-supercentenarians (105+ yo) and to the ITA population (Tuscany 1000 Genomes population [TSI] + 36 controls genomes matching the CEN). While the ITA population does not show any significant difference from the Europeans, a significant reduction of the DLX5/6-N-Haplotype allele frequency is observed in the semi-supercentenarian population (*P < 0.05, Fst distribution). (B) Density plot build considering all the Fst values calculated considering all the archaic haplotypes described in Skov, et al. (2020) for CEN versus ITA groups. The blue arrow indicates the value of Fst calculated for the DLX5/6-N-Haplotype.CC BY-NC · PubMed Central

Verification

Claims the paper makes

20 Valid · 3 Not valid · 10 Not checked · 71 Own results and interpretations

55 more claims · 5 with their cited sources checked

Create a free account to see all 104 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 34
  • Probably not a substantial claim. 12
  • The quote only partly states this claim. 4

Create a free account to see these claims.

Self-check

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

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 110 references

Research works printed but not linked: 1

  1. Vacante, 2012, BMC Surg, S36

Cited by

Metadata

Authors
Levi, G., de Lombares, C., Giuliani, C., Iannuzzi, V., Aouci, R., Garagnani, P., Franceschi, C., Grimaud-Hervé, D., & Narboux-Nême, N.
Affiliations
  • Centre National de la Recherche Scientifique (FR)
  • University of Bologna (IT)
  • Karolinska Institutet (SE)
  • N. I. Lobachevsky State University of Nizhny Novgorod (RU)
  • Physiologie Moléculaire et Adaptation, CNRS UMR7221, Département AVIV, Muséum National d'Histoire Naturelle, Paris, France
Venue
Molecular Biology and Evolution
Published
2021-06-16
Volume
38
Issue
11
Pages
4748-4764
ISSN
  • 0737-4038 (ISSN-L)
  • 1537-1719
OpenAlex
W3171099679
Cited by
23
References
114
Citations by year
2026: 3 · 2025: 4 · 2024: 10 · 2023: 2 · 2022: 3 · 2021: 1
FWCI
1.93
Type
journal article
Language
en
Open access
Locations
  • Molecular Biology and Evolution (publishedVersion, cc-by-nc)
  • bioRxiv (Cold Spring Harbor Laboratory) (submittedVersion)
  • Europe PMC (PubMed Central) (publishedVersion, other-oa)
  • PubMed Central (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
  • Archivio istituzionale della ricerca (Alma Mater Studiorum Università di Bologna) (submittedVersion, cc-by-nc)
  • HAL (submittedVersion)
Funders
  • ANR (ANR-17-CE14-0024, ANR-17-CE37-0007)
  • Fondation-NRJ (216612)
  • French Ministry of Research
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Genetics and Neurodevelopmental Disorders
  • Life Sciences › Agricultural and Biological Sciences › Ecology, Evolution, Behavior and Systematics › Animal Vocal Communication and Behavior
  • Social Sciences › Cultural Studies › Language and cultural evolution
Keywords
GABAergic neurons, social vocalizations, sociability, domestication syndrome, self-domestication, Neanderthal introgression, GTF2I, Williams syndrome, longevity, human evolution, speech evolution, FOXP2, lifespan
MeSH terms
Animals, Genes, Homeobox, Humans, Transcription Factors, Vocalization, Animal, Homeodomain Proteins, Transcription Factors, TFII, Mice, Neanderthals

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-11, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-11 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.