CoChat

Combining genome-wide association studies highlight novel loci involved in human facial variation

Open access

The DOI matches its Crossref record; 46 of the 46 references it cites were found in a registry or an index.

Abstract

Standard genome-wide association studies (GWASs) rely on analyzing a single trait at a time. However, many human phenotypes are complex and composed by multiple correlated traits. Here we introduce C-GWAS, a method for combining GWAS summary statistics of multiple potentially correlated traits. Extensive computer simulations demonstrated increased statistical power of C-GWAS compared to the minimal p-values of multiple single-trait GWASs (MinGWAS) and the current state-of-the-art method for combining single-trait GWASs (MTAG). Applying C-GWAS to a meta-analysis dataset of 78 single trait facial GWASs from 10,115 Europeans identified 56 study-wide suggestively significant loci with multi-trait effects on facial morphology of which 17 are novel loci. Using data from additional 13,622 European and Asian samples, 46 (82%) loci, including 9 (53%) novel loci, were replicated at nominal significance with consistent allele effects. Functional analyses further strengthen the reliability of our C-GWAS findings. Our study introduces the C-GWAS method and makes it available as computationally efficient open-source R package for widespread future use. Our work also provides insights into the genetic architecture of human facial appearance.

Overview

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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

Integrity signals

Registry

Crossref · 10.1038/s41467-022-35328-9

Cites

46 of 46 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Power comparison between EbICoW and Wald test for combining two GWASs under the same true effect via simulations. The summary statistics of two GWASs with the same true effect configuration, i.e., E(χ2) = 1.3, were simulated, each consisting of 50,000 SNPs with 10% having true effect, according to the following parameter grid: ψ and π∈{−0.98, −0.9, −0.8,…,−0.1, 0, 0.1,…, 0.8, 0.9, 0.98}. For each of the 441 (21×21) different combinations of π and ψ, 1000 replicates were carried out. The power relationship between EbICoW and Wald is illustrated using a heat map in the center panel. The maximum power of the two methods is proportional to the block size in the heat map. The power ratio of EbICoW and Wald is expressed by color from red to blue. Taking seven specific combinations of π and ψ as examples, we show in detail the performance of EbICoW and Wald (panels (a–g) surrounding the central panel). For illustration purposes, each surrounding panel is a scatter plot of the test statistics from the two GWASs consisting of 2000 SNPs with 50% increased true effect compared with the corresponding cell in the heat map, in which the null SNPs are indicated in gray dots and the SNPs with true effects are indicated in colored crosses. Different colors are used to distinguish the significance of EbICoW and Wald, i.e., orange for both significant, blue for EbICoW only significant, green for Wald only significant, and black for non-significant results. The areas outside of the blue lines are the detection range of EbICoW and the area outside of the green ellipse is the detection range of Wald.
Fig. 1 Power comparison between EbICoW and Wald test for combining two GWASs under the same true effect via simulations. The summary statistics of two GWASs with the same true effect configuration, i.e., E(χ2) = 1.3, were simulated, each consisting of 50,000 SNPs with 10% having true effect, according to the following parameter grid: ψ and π∈{−0.98, −0.9, −0.8,…,−0.1, 0, 0.1,…, 0.8, 0.9, 0.98}. For each of the 441 (21×21) different combinations of π and ψ, 1000 replicates were carried out. The power relationship between EbICoW and Wald is illustrated using a heat map in the center panel. The maximum power of the two methods is proportional to the block size in the heat map. The power ratio of EbICoW and Wald is expressed by color from red to blue. Taking seven specific combinations of π and ψ as examples, we show in detail the performance of EbICoW and Wald (panels (a–g) surrounding the central panel). For illustration purposes, each surrounding panel is a scatter plot of the test statistics from the two GWASs consisting of 2000 SNPs with 50% increased true effect compared with the corresponding cell in the heat map, in which the null SNPs are indicated in gray dots and the SNPs with true effects are indicated in colored crosses. Different colors are used to distinguish the significance of EbICoW and Wald, i.e., orange for both significant, blue for EbICoW only significant, green for Wald only significant, and black for non-significant results. The areas outside of the blue lines are the detection range of EbICoW and the area outside of the green ellipse is the detection range of Wald.CC BY · PubMed Central
Fig. 2 Power comparison between EbICoW, MTAG and TWT via simulations. The summary statistics of ten GWASs were simulated, each consisting of 50,000 SNPs with 10% having true effect. We evaluated the power of EbICoW, MTAG, TWT and MinGWAS at α = 0.05/50,000 under two different ψ ∈ {0, 0.5} and four different patterns of true effect. For pattern 1, one GWAS with true effect E(χ2) = 2 and nine GWASs under the null E(χ2) = 1 were simulated; for pattern 2, three GWASs with true effect E(χ2) ∈ {1.6, 1.3, 1.1} and seven GWASs under the null were simulated; for pattern 3, six GWASs with true effect E(χ2) ∈ {1.3, 1.2, 1.2, 1.1, 1.1, 1.1} and four GWASs under the null were simulated; for pattern 4, all ten GWASs were simulated with true effect E(χ2) = 1.1. For each of the 8 different combinations of ψ and pattern of true effect, 1,000 replicates were carried out. The x-axis is π and the y-axis is the power. Each column has the same configuration of the true effect and each row has the same ψ. The configuration of the patterns of true effect is displayed in the top row. Note that the E(χ2) is the sum of mean χ2 of true effect and the mean χ2 under the null, the latter is always 1.0. The power estimates of EbICoW, TWT, MTAG, and the MinGWAS of ten GWASs are indicated in orange, green, blue, and black dots, respectively. Dots were fitted using local polynomial regression curves. Note that the p-values from MTAG, TWT, and the MinGWAS of ten GWASs were additionally adjusted to have the same α with EbICoW.
Fig. 2 Power comparison between EbICoW, MTAG and TWT via simulations. The summary statistics of ten GWASs were simulated, each consisting of 50,000 SNPs with 10% having true effect. We evaluated the power of EbICoW, MTAG, TWT and MinGWAS at α = 0.05/50,000 under two different ψ ∈ {0, 0.5} and four different patterns of true effect. For pattern 1, one GWAS with true effect E(χ2) = 2 and nine GWASs under the null E(χ2) = 1 were simulated; for pattern 2, three GWASs with true effect E(χ2) ∈ {1.6, 1.3, 1.1} and seven GWASs under the null were simulated; for pattern 3, six GWASs with true effect E(χ2) ∈ {1.3, 1.2, 1.2, 1.1, 1.1, 1.1} and four GWASs under the null were simulated; for pattern 4, all ten GWASs were simulated with true effect E(χ2) = 1.1. For each of the 8 different combinations of ψ and pattern of true effect, 1,000 replicates were carried out. The x-axis is π and the y-axis is the power. Each column has the same configuration of the true effect and each row has the same ψ. The configuration of the patterns of true effect is displayed in the top row. Note that the E(χ2) is the sum of mean χ2 of true effect and the mean χ2 under the null, the latter is always 1.0. The power estimates of EbICoW, TWT, MTAG, and the MinGWAS of ten GWASs are indicated in orange, green, blue, and black dots, respectively. Dots were fitted using local polynomial regression curves. Note that the p-values from MTAG, TWT, and the MinGWAS of ten GWASs were additionally adjusted to have the same α with EbICoW.CC BY · PubMed Central
Fig. 3 The null distributions before and after calibration from C-GWAS and MinGWAS via simulations. Vectors of T statistics and corresponding two-side p-values of 30 GWASs were simulated under the null, each consisting of 10,000 SNPs without any true effect. All pairs of T statistics vectors were simulated to have the same ψ. C-GWAS and MinGWAS were used to combine the 30 simulated GWASs. Null distributions of C-GWAS (the right column) and the MinGWAS (the left column) were obtained at ψ = 0 (the upper row) and ψ = 0.5 (the lower row). The crude p-values (blue), the p-values after the Tippett’s method based correction of the number of independent tests obtained from the getNtest function (green), the p-values calibrated using the getCoef function (orange), and the uniform distribution (black) are distinguished in different colors. The median and 95% intervals at all quantiles are indicated using dots and dashed lines respectively.
Fig. 3 The null distributions before and after calibration from C-GWAS and MinGWAS via simulations. Vectors of T statistics and corresponding two-side p-values of 30 GWASs were simulated under the null, each consisting of 10,000 SNPs without any true effect. All pairs of T statistics vectors were simulated to have the same ψ. C-GWAS and MinGWAS were used to combine the 30 simulated GWASs. Null distributions of C-GWAS (the right column) and the MinGWAS (the left column) were obtained at ψ = 0 (the upper row) and ψ = 0.5 (the lower row). The crude p-values (blue), the p-values after the Tippett’s method based correction of the number of independent tests obtained from the getNtest function (green), the p-values calibrated using the getCoef function (orange), and the uniform distribution (black) are distinguished in different colors. The median and 95% intervals at all quantiles are indicated using dots and dashed lines respectively.CC BY · PubMed Central
Fig. 4 C-GWAS analysis of 78 correlated facial traits and replication. C-GWAS is applied to combine summary statistics of 78 GWASs based on 10,115 samples from four European cohorts4. a The 78 correlated facial traits were derived as the pair-wise Euclidean distances between 13 well-defined facial landmarks. b, c A simulation analysis derived the null distributions of C-GWAS and MinGWAS in terms of crude p-values (gray) and p-values calibrated using getCoef (orange), which are compared with the uniform distribution (black). Median values and 95% intervals at all quantiles are indicated using dots and dashed lines. After distribution calibration, the C-GWAS results (upper part of e) and the MinGWAS p-values (lower part of e) are plotted using an overlayed Q–Q plot (d) and Miami plot (e). The study-wide significance threshold (p = 5 × 10−8) is indicated using dashed lines and the study-wide suggestive significance threshold (p = 3 × 10−6) is indicated using solid lines. The overlapping between previously established and novel loci is illustrated in a Venn diagram. The closest genes to the regional lead SNPs are listed in (f) (orange for novel). A list of 327 previously established face-associated SNPs was looked up in the C-GWAS and MinGWAS results (N = 10,115, g). The regional lead SNPs at 56 loci from C-GWAS were looked up in the combined replication dataset (N = 13,622, h). One-side Kolmogorov-Smirnov test is used to test if the observed p-value distribution is deviated from the uniform distribution under the null. In the replication sample of 1174 individuals of RS, the facial variance explained by C-GWAS polygenic risk score (PRS) (i) is compared with that explained by MinGWAS PRS (j).
Fig. 4 C-GWAS analysis of 78 correlated facial traits and replication. C-GWAS is applied to combine summary statistics of 78 GWASs based on 10,115 samples from four European cohorts4. a The 78 correlated facial traits were derived as the pair-wise Euclidean distances between 13 well-defined facial landmarks. b, c A simulation analysis derived the null distributions of C-GWAS and MinGWAS in terms of crude p-values (gray) and p-values calibrated using getCoef (orange), which are compared with the uniform distribution (black). Median values and 95% intervals at all quantiles are indicated using dots and dashed lines. After distribution calibration, the C-GWAS results (upper part of e) and the MinGWAS p-values (lower part of e) are plotted using an overlayed Q–Q plot (d) and Miami plot (e). The study-wide significance threshold (p = 5 × 10−8) is indicated using dashed lines and the study-wide suggestive significance threshold (p = 3 × 10−6) is indicated using solid lines. The overlapping between previously established and novel loci is illustrated in a Venn diagram. The closest genes to the regional lead SNPs are listed in (f) (orange for novel). A list of 327 previously established face-associated SNPs was looked up in the C-GWAS and MinGWAS results (N = 10,115, g). The regional lead SNPs at 56 loci from C-GWAS were looked up in the combined replication dataset (N = 13,622, h). One-side Kolmogorov-Smirnov test is used to test if the observed p-value distribution is deviated from the uniform distribution under the null. In the replication sample of 1174 individuals of RS, the facial variance explained by C-GWAS polygenic risk score (PRS) (i) is compared with that explained by MinGWAS PRS (j).CC BY · PubMed Central
Fig. 5 Multi-trait effects of the face-associated SNPs with examples. The estimated multi-trait effect (MTE, y-axis) of the 57 C-GWAS-identified study-wide suggestively significant regional lead SNPs were plotted against their −log10(p) from C-GWAS (−log10(pC), x-axis, a). SNPs are indexed according to their closest genes for illustrative purpose. SNPs are distinguished with two colors of orange for previous known face-associated loci and blue for novel loci. The boost of significance of C-GWAS in relative to MinGWAS (−log10(pC/pG)) is indicated using the dot size. The SNPs that are suggestively significant in both C-GWAS and GWAS are indicated using circles with solid lines. The squared projected effects (χ2) are plotted against the quantile of the expected effects on the 26 projected vectors (see “Methods” for details, b). The expected effects under the null are indicated using black dots connected by black lines. The observed effects are detailed for eight examples with large or small values of MTE, including EIF4G3, CDK2AP1, PAX3, SFRP2, TBX3, SIX2, INTU, and RPE65, which are distinguished using different colors and dot shapes. The effect of the lead SNPs in terms of the T-statistic from GWASs on 78 facial traits are superimposed on face maps (c). Face maps of full 57 C-GWAS lead SNPs are available in Supplementary Fig. 15.
Fig. 5 Multi-trait effects of the face-associated SNPs with examples. The estimated multi-trait effect (MTE, y-axis) of the 57 C-GWAS-identified study-wide suggestively significant regional lead SNPs were plotted against their −log10(p) from C-GWAS (−log10(pC), x-axis, a). SNPs are indexed according to their closest genes for illustrative purpose. SNPs are distinguished with two colors of orange for previous known face-associated loci and blue for novel loci. The boost of significance of C-GWAS in relative to MinGWAS (−log10(pC/pG)) is indicated using the dot size. The SNPs that are suggestively significant in both C-GWAS and GWAS are indicated using circles with solid lines. The squared projected effects (χ2) are plotted against the quantile of the expected effects on the 26 projected vectors (see “Methods” for details, b). The expected effects under the null are indicated using black dots connected by black lines. The observed effects are detailed for eight examples with large or small values of MTE, including EIF4G3, CDK2AP1, PAX3, SFRP2, TBX3, SIX2, INTU, and RPE65, which are distinguished using different colors and dot shapes. The effect of the lead SNPs in terms of the T-statistic from GWASs on 78 facial traits are superimposed on face maps (c). Face maps of full 57 C-GWAS lead SNPs are available in Supplementary Fig. 15.CC BY · PubMed Central
Fig. 6 Biological annotation of significant findings. a Gene ontology (GO), human phenotype ontology (HP), mouse phenotype ontology (MP) terms enrichment analyses were carried out based on C-GWAS (83 genes in 56 loci) and MinGWAS (24 gene in 17 loci) findings. Biological terms with their Bonferroni corrected enriched p-values from C-GWAS and MinGWAS were plotted above and below the black solid line, respectively. Shared and significant terms are indicated in green (more significant) or blue (less significant). Unique terms are indicated in orange with those non-significant in gray. The dashed lines indicate Bonferroni corrected p = 0.05. A regulation network in CNCC is derived based on C-GWAS finding (b, 56 loci) and MinGWAS finding (c, 17 loci). Transcription factor (TF)—regulation element (RE)—target gene (TG) regulation triplets are superimposed on chord diagrams and are distinguished using colors. The TF-RE edge width represents the number of REs associated with TF. TGs linked to their RE are highlighted and the number of RE-TG is proportional to arrows size. Closest genes to lead SNPs are highlighted in orange text. The genes that are both TF and TG are marked using “*”.
Fig. 6 Biological annotation of significant findings. a Gene ontology (GO), human phenotype ontology (HP), mouse phenotype ontology (MP) terms enrichment analyses were carried out based on C-GWAS (83 genes in 56 loci) and MinGWAS (24 gene in 17 loci) findings. Biological terms with their Bonferroni corrected enriched p-values from C-GWAS and MinGWAS were plotted above and below the black solid line, respectively. Shared and significant terms are indicated in green (more significant) or blue (less significant). Unique terms are indicated in orange with those non-significant in gray. The dashed lines indicate Bonferroni corrected p = 0.05. A regulation network in CNCC is derived based on C-GWAS finding (b, 56 loci) and MinGWAS finding (c, 17 loci). Transcription factor (TF)—regulation element (RE)—target gene (TG) regulation triplets are superimposed on chord diagrams and are distinguished using colors. The TF-RE edge width represents the number of REs associated with TF. TGs linked to their RE are highlighted and the number of RE-TG is proportional to arrows size. Closest genes to lead SNPs are highlighted in orange text. The genes that are both TF and TG are marked using “*”.CC BY · PubMed Central
Fig. 7 Colocalization analysis of C-GWAS significant findings. Colocalization analysis was carried out for the 56 loci from C-GWAS in 22 tissues from GTEx. a Posterior probabilities in the heatmap are indicated using the box size and the gradience in color (posterior probabilities > 0.7 in solid boxes). Closest genes to the regional lead SNPs are indicated using orange text. Four examples of eQTL colocalizations at distinct loci in different tissues (b–e), where the regional lead SNPs were cis-associated with gene expressions (lower right) and were simultaneously associated with facial variation in C-GWAS (upper right), suggesting that these SNPs are responsible for both the cis-eQTL signal and the C-GWAS signal (left).
Fig. 7 Colocalization analysis of C-GWAS significant findings. Colocalization analysis was carried out for the 56 loci from C-GWAS in 22 tissues from GTEx. a Posterior probabilities in the heatmap are indicated using the box size and the gradience in color (posterior probabilities > 0.7 in solid boxes). Closest genes to the regional lead SNPs are indicated using orange text. Four examples of eQTL colocalizations at distinct loci in different tissues (b–e), where the regional lead SNPs were cis-associated with gene expressions (lower right) and were simultaneously associated with facial variation in C-GWAS (upper right), suggesting that these SNPs are responsible for both the cis-eQTL signal and the C-GWAS signal (left).CC BY · PubMed Central

Verification

Claims the paper makes

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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 46 references

Research works printed but not linked: 1

  1. Joanes, D. N. & Gill, C. A. Comparing measures of sample skewness and kurtosis. J. R. Statistical Soc. Ser. D (The Statistician) 47, 183–189 (1998).

Cited by

See all 37

Metadata

Authors
Xiong, Z., Gao, X., Chen, Y., Feng, Z., Pan, S., Lu, H., Uitterlinden, A. G., Nijsten, T., Ikram, A., Rivadeneira, F., Ghanbari, M., Wang, Y., Kayser, M., & Liu, F.
Affiliations
  • Erasmus MC (NL)
  • Chinese Academy of Sciences (CN)
Corresponding author
  • Kayser, Manfred
  • Liu, Fan
Venue
Nature Communications
Published
2022-12-20
Volume
13
Issue
1
Pages
7832-7832
ISSN
2041-1723 (ISSN-L)
OpenAlex
W4313339753
Cited by
37
References
46
Citations by year
2026: 11 · 2025: 15 · 2024: 5 · 2023: 6
FWCI
3.91
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • EUR Research Repository (Erasmus University Rotterdam) (publishedVersion, cc-by)
  • EUR Research Repository (Erasmus University Rotterdam) (submittedVersion, cc-by)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Purdue University
  • Indiana University-Purdue University Indianapolis
  • European Commission
  • National Natural Science Foundation of China (KFJ-STS-QYZD-2021-08-001, XDC01000000, XDB38010400, 2017SHZDZX01, 81930056)
  • ZonMw
  • Erasmus Universiteit Rotterdam
  • Chinese Academy of Sciences (2017SHZDZX01, KFJ-STS-QYZD-2021-08-001, XDC01000000, KFJ-STS-ZDTP-079, XDB38010400)
  • Nederlandse Organisatie voor Wetenschappelijk Onderzoek
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Genetic Associations and Epidemiology
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Genetic Mapping and Diversity in Plants and Animals
  • Life Sciences › Agricultural and Biological Sciences › Ecology, Evolution, Behavior and Systematics › Morphological variations and asymmetry
Keywords
facial morphology, facial variation, novel loci, MTAG, statistical power, facial anthropometry, R package
MeSH terms
Computer Simulation, Humans, Phenotype, Reproducibility of Results, Polymorphism, Single Nucleotide, Genome-Wide Association Study

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Combining genome-wide association studies highlight novel loci involved in human facial variation. https://app.cochat.ai/papers/10.1038/s41467-022-35328-9
The paper
Xiong, Z., Gao, X., Chen, Y., Feng, Z., Pan, S., Lu, H., Uitterlinden, A. G., Nijsten, T., Ikram, A., Rivadeneira, F., Ghanbari, M., Wang, Y., Kayser, M., & Liu, F. (2022). Combining genome-wide association studies highlight novel loci involved in human facial variation. Nature Communications. https://doi.org/10.1038/s41467-022-35328-9

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.