Author response: The lingering effects of Neanderthal introgression on human complex traits

Abstract

Article Figures and data Abstract Editor's evaluation Introduction Results Discussion Methods Appendix 1 Appendix 2 Appendix 3 Appendix 4 Appendix 5 Data availability References Decision letter Author response Article and author information Metrics Abstract The genetic variants introduced into the ancestors of modern humans from interbreeding with Neanderthals have been suggested to contribute an unexpected extent to complex human traits. However, testing this hypothesis has been challenging due to the idiosyncratic population genetic properties of introgressed variants. We developed rigorous methods to assess the contribution of introgressed Neanderthal variants to heritable trait variation and applied these methods to analyze 235,592 introgressed Neanderthal variants and 96 distinct phenotypes measured in about 300,000 unrelated white British individuals in the UK Biobank. Introgressed Neanderthal variants make a significant contribution to trait variation (explaining 0.12% of trait variation on average). However, the contribution of introgressed variants tends to be significantly depleted relative to modern human variants matched for allele frequency and linkage disequilibrium (about 59% depletion on average), consistent with purifying selection on introgressed variants. Different from previous studies (McArthur et al., 2021), we find no evidence for elevated heritability across the phenotypes examined. We identified 348 independent significant associations of introgressed Neanderthal variants with 64 phenotypes. Previous work (Skov et al., 2020) has suggested that a majority of such associations are likely driven by statistical association with nearby modern human variants that are the true causal variants. Applying a customized fine-mapping led us to identify 112 regions across 47 phenotypes containing 4303 unique genetic variants where introgressed variants are highly likely to have a phenotypic effect. Examination of these variants reveals their substantial impact on genes that are important for the immune system, development, and metabolism. Editor's evaluation Humans whose genetic ancestors lived outside Africa have a small proportion of the genome that traces back to interbreeding events with Neanderthals. To quantify the contribution of this ancestry to present-day phenotypic variation, the authors develop a convincing set of approaches that takes into account various complicating factors and apply it to a subset of the UK Biobank individuals. The work is an important contribution to human evolution and evolutionary biology more generally. https://doi.org/10.7554/eLife.80757.sa0 Decision letter Reviews on Sciety eLife's review process Introduction Genomic analyses have revealed that present-day non-African human populations inherit 1–4% of their genetic ancestry from introgression with Neanderthals (Green et al., 2010; Prüfer et al., 2014). This introgression event introduced uniquely Neanderthal variants into the ancestral out-of-Africa human gene pool, which may have helped this bottleneck population survive the new environments they encountered (Mendez et al., 2012; Abi-Rached et al., 2011; Sankararaman et al., 2014; Vernot and Akey, 2014; Racimo et al., 2015; Gittelman et al., 2016). On the other hand, many Neanderthal variants appear to have been deleterious in the modern human genetic background leading to a reduction in Neanderthal ancestry in conserved genomic regions (Sankararaman et al., 2014; Vernot and Akey, 2014; Harris and Nielsen, 2016; Juric et al., 2016; Petr et al., 2019). Systematically studying these variants can provide insights into the biological differences between Neanderthals and modern humans and the evolution of human phenotypes in the 50,000 y since introgression. In principle, studying Neanderthal-derived variants in large cohorts of individuals measured for diverse phenotypes can help understand the biological impact of Neanderthal introgression. Previously, (Dannemann and Kelso, 2017) showed that some Neanderthal introgressed variants are significantly associated with traits such as skin tone, hair color, and height based on genome-wide association studies (GWAS) in British samples. However, using data from Iceland, Skov et al., 2020 found that most of the significantly associated Neanderthal introgressed single-nucleotide polymorphisms (SNPs) are in the proximity of strongly associated non-archaic variants. They suggested that these associations at Neanderthal introgressed SNPs were driven by the associations at linked non-archaic variants, indicating a limited contribution to modern human phenotypes from Neanderthal introgression. In contrast to these attempts to associate individual introgressed variants with a trait, studies have attempted to measure the aggregate contribution of introgressed Neanderthal SNPs to trait variation (Simonti et al., 2016; McArthur et al., 2021). A recent study by McArthur et al., 2021 estimated the proportion of heritable variation that can be attributed to introgressed variants though their approach is restricted to common variants (minor allele frequency >5%) that represent a minority of introgressed variants. Despite these attempts, assessing the contribution of introgressed Neanderthal variants towards specific phenotypes remains challenging. The first challenge is that variants introgressed from Neanderthals are rare on average (due to the low proportion of Neanderthal ancestry in present-day genomes). The second challenge arises from the unique evolutionary history of introgressed Neanderthal variants, resulting in distinct population genetic properties at these variants, which can, in turn, confound attempts to characterize their effects. As a result, attempts to characterize the systematic impact of introgressed variants on complex phenotypes need to be rigorously assessed. To enable analyses of genome-wide introgressed Neanderthal variants in large sample sizes, we selected and added SNPs that tag introgressed Neanderthal variants to the UK Biobank Axiom Array that was used to genotype the great majority of the approximately 500,000 individuals in the UK Biobank (UKBB) (Bycroft et al., 2018). We used a previously compiled map of Neanderthal haplotypes in the 1000 Genomes European populations (Sankararaman et al., 2014) to identify introgressed SNPs that tag these haplotypes. After removing SNPs that are well-tagged by those previously present on the UKBB array, we used a greedy algorithm to select 6027 SNPs that tag the remaining set of introgressed SNPs at r2>0.8, which were then added to the UKBB genotyping array to better tag Neanderthal ancestry. These SNPs allow variants of Neanderthal ancestry to be confidently imputed and allow us to identify a list of 235,592 variants that are likely to be Neanderthal-derived (termed Neanderthal Informative Mutations [NIMs]) out of a total of 7,774,235 QC-ed SNPs in UKBB (see ‘Methods’; Appendix 1). The goals of our study are threefold: (1) to estimate the contribution of NIMs to phenotypic variation in modern humans, (2) to test the null hypothesis that an NIM has the same contribution to phenotypic variation as a non-introgressed modern human SNP, and (3) to pinpoint regions of the genome at which NIMs are highly likely to modulate phenotypic variation. We develop rigorous methodology for each of these goals that we validate in simulations. We then applied these methods to 96 distinct phenotypes measured in about 300,000 unrelated white British individuals in UKBB. Results The contribution of Neanderthal introgressed variants to trait heritability To understand the contribution of Neanderthal introgressed variants to trait variation, we aim to estimate the proportion of phenotypic variance attributed to NIMs (NIM heritability) and to test the null hypothesis that per-NIM heritability is the same as the heritability of a non-introgressed modern human (MH) SNP. We first annotated each of the 7,774,235 QC-ed SNPs in UKBB as either an NIM or an MH SNP (see ‘Methods’). NIMs include SNPs created by mutations that likely originated in the Neanderthal lineage after the human-Neanderthal split. SNPs that are not defined as NIMs are annotated as MH SNPs that likely originated in the modern human lineage or the human-Neanderthal common ancestor. To estimate NIM heritability, we used a recently proposed method (RHE-mc) that can partition the heritability of a phenotype measured in large samples across various genomic annotations (Pazokitoroudi et al., 2020). We applied RHE-mc with genomic annotations that correspond to the ancestry of each SNP (NIM vs. MH) to estimate NIM heritability (h2NIM). We also attempted to estimate whether per-NIM heritability is the same as the per-SNP heritability of MH SNPs (Δh2). A positive (negative) value of Δh2 indicates that, on average, an NIM makes a larger (smaller) contribution to phenotypic variation relative to a MH SNP. To assess the accuracy of this approach, we performed simulations where NIMs are neither enriched nor depleted in heritability (true Δh2=0). Following previous studies of the genetic architecture of complex traits (Evans et al., 2018; Evans et al., 2018), we simulated phenotypes (across 291,273 unrelated white British individuals and 7,774,235 SNPs) with different architectures where we varied heritability, polygenicity, and how the effect size at a SNP is coupled to its population genetic properties (the minor allele frequency [MAF] at the SNP and the linkage disequilibrium or LD around an SNP). We explored different forms of MAF-LD coupling where BASELINE assumes that SNPs with phenotypic effects are chosen randomly, RARE (COMMON) assumes that rare (common) variants are enriched for phenotypic effects, and HIGH (LOW) assumes that SNPs with high (low) levels of LD (as measured by the LD score; Finucane et al., 2015) are enriched for phenotypic effects (see ‘Methods’). Estimates of hNIM2 and Δh2 tend to be miscalibrated (Figure 1ab). The miscalibration is particularly severe when testing Δh2 so that a test of the null hypothesis has a false-positive rate of 0.55 across all simulations (at a p-value threshold of 0.05). Figure 1 with 2 supplements see all Download asset Open asset Benchmarking approaches for estimating the heritability components of Neanderthal introgression. We group simulations by relationships between minor allele frequency (MAF) and local linkage disequilibrium at an SNP on effect size (MAF-LD coupling): BASELINE, COMMON, RARE, HIGH, LOW. In each group, we perform 12 simulations with varying polygenicity and heritability (see ‘Methods’). Additionally, we combine results from all simulations together as ALL. We plot the distributions of two Z-scores (y-axis), one on each row: (a) Z-score (h2^NIM = hNIM2) tests whether the estimated and true Neanderthal Informative Mutations (NIM) heritability are equal, and (b) Z-score (Δ^h2=0) tests whether the estimated per-NIM heritability is the same as the per-SNP heritability of modern human (MH) SNPs (see ‘Methods’). In each panel, we present results from a variance components analysis method (RHE-mc) using four different input annotations: ancestry only where ancestry is either NIM or MH, ancestry + MAF, ancestry + LD, ancestry + MAF + LD. A calibrated method is expected to have Z-scores distributed around zero and within ±2 (shaded region). Among all tested approaches, only RHE-mc with ancestry + MAF + LD annotations is calibrated across simulations. Figure 1—source data 1 RHE-mc results in simulated data. https://cdn.elifesciences.org/articles/80757/elife-80757-fig1-data1-v2.zip Download elife-80757-fig1-data1-v2.zip To understand these observations, we compared the MAFs and LD scores at NIMs to MH SNPs. We observe that NIMs tend to have lower MAF (Figure 2a) and higher LD scores compared to MH SNPs (Figure 2b) (the average MAF of NIMs and MH SNPs are 3.9% and 9.9%, respectively, while their average LD scores are 170.6 and 64.9). Among the QC-ed SNPs, 76.9% of NIMs have MAF >1%, and 27.7% have MAF >5%, in contrast to 61.6% and 41.6% of MH SNPs. Distinct from MH SNPs, the MAF and LD score of NIMs tend not to increase with each other (Figure 2cd). We replicated this observation using NIMs that had been identified by an alternate approach (McArthur et al., 2021; Appendix 5). Figure 2 Download asset Open asset Distributions of minor allele frequency (MAF) and LD-score in Neanderthal Informative Mutations (NIMs) and modern human (MH) SNPs. Empirical cumulative distribution functions of (a) MAF and (b) LD scores of NIMs (in solid green line) and MH SNPs (in pink dashed line) estimated in the UK Biobank (UKBB). (c) Boxplots of MAFs of NIMs (on the left filled in green) and MH SNPs (on the right side filled in pink) while controlling for LD score (UKBB). (d) Boxplots of LD score (UKBB) of NIMs and MH SNPs while controlling for MAF. NIMs and MH SNPs are divided by the 20, 40, 60, 80, 100 (c) LD score (UKBB) percentile or MAF percentile (d) based on all QC-ed SNPs (7,774,235 imputed SNPs with MAF >0.001). The lower and upper edges of a box represent the first and third quartile (qu1 and qu3), respectively; the horizontal red line inside the box indicates median (md); the whiskers extend to the most extreme values inside inner fences, md ± 1.5 (qu3–qu1). To account for the differences in the MAF and LD scores across NIMs and MH SNPs, we applied RHE-mc with annotations corresponding to the MAF and the LD score at each SNP (in addition to the ancestry annotation that classifies SNPs as NIM vs. MH) to estimate NIM heritability (h2NIM) and to test whether per-NIM heritability is the same as the per-SNP heritability of MH SNPs, that is, Δh2=0 (see ‘Methods,’ ‘Appendix 4’). Our simulations show that RHE-mc with SNPs assigned to annotations that account for both MAF and LD (in addition to the ancestry annotation that classifies SNPs as NIM vs. MH) is accurate both in the estimates of hNIM2 (Figure 1a) and in testing the null hypothesis that Δh2=0 (the false positive rate of a test of Δh2=0 is 0.017 at a p-value threshold of 0.05; Figure 1b). On the other hand, not accounting for either MAF or LD leads to poor calibration (Figure 1; we observe qualitatively similar results when estimating genome-wide SNP heritability; Figure 1—figure supplement 1). To further assess the robustness of our results to the genetic architecture, we also performed simulations under a model that assumes an even greater enrichment of SNP effects among rare variants wherein SNPs with MAF <1% constitute 90% of the causal variants (ULTRA RARE). RHE-mc with MAF and LD annotations remains accurate in its estimates of hNIM2 and in testing the null hypothesis that Δh2=0 (Figure 1—figure supplement 2). We then applied RHE-mc with ancestry + MAF + LD annotations to analyze a total of 96 UKBB phenotypes that span 14 broad categories. In all our analyses, we include the top 5 PCs estimated from NIMs (NIM PCs) as covariates in addition to the top 20 genetic PCs estimated from common SNPs, sex, and age (see ‘Methods’). The inclusion of NIM PCs is intended to account for stratification at NIMs that may not be adequately corrected by including genotypic PCs estimated from common SNPs (we also report concordant results from our analyses when excluding NIM PCs; ‘Appendix 3’ and Figure 4—figure supplement 1). We first examined NIM heritability to find six phenotypes with significant NIM heritability (Z-score (h2^NIM=0)>3): body fat percentage, trunk fat percentage, whole body fat mass, overall health rating, gamma glutamyltransferase (a measure of liver function), and forced vital capacity (FVC) (Figure 3a and c). Meta-analyzing within nine categories that contain at least four phenotypes, we find that meta−h2^NIM is significantly larger than zero for anthropometry, blood biochemistry, bone densitometry, kidney, liver, and lung but not for blood pressure, eye, lipid metabolism (p<0.05 accounting for the number of hypotheses tested). Meta-analyzing across all phenotypes with low correlation, we obtain overall NIM heritability estimates (meta−h2^NIM) = 0.12% (one-sided p=6.6×10–31). The estimates of NIM heritability are modest as would be expected from traits that are highly polygenic and given that NIMs account for a small percentage of all SNPs in the genome (see ‘Methods). Figure 3 with 1 supplement see all Download asset Open asset Neanderthal Informative Mutation (NIM) heritability in UK Biobank (UKBB) phenotypes. (a) Estimates of NIM heritability (h2^NIM) and (c) the Z-score of h2^NIM (testing the hypothesis that NIM heritability is positive) for each UKBB phenotype. Analogously, (b) estimates of Δ^h2 and Z-score (d) of Δ^h2 (testing the hypothesis that per-NIM heritability is equal to per-SNP heritability at modern human [MH] SNPs after controlling for MAF and LD). Phenotypic categories are shown in alphabetical order and listed on the top of panel (a) in the same color and alphabetical order (from top to bottom, and left to right) as they are in the figure. The estimate for each phenotype is shown as one colored dot, on the x-axis based on its phenotypic category, and on the y-axes based on its Z-score (h2^NIM=0) and Z-score (Δ^h2=0), for panels (c) and (d) respectively. For each phenotypic category with at least four phenotypes, their Z-scores from random effect meta-analysis are plotted with the flat colored lines (see ‘Methods’). The color shades cover Z-scores around zero and within ±2. g. Figure 3—source data 1 UKBB phenotype annotation. https://cdn.elifesciences.org/articles/80757/elife-80757-fig3-data1-v2.zip Download elife-80757-fig3-data1-v2.zip Figure 3—source data 2 RHE-mc results with Ancestry+MAF+LD annotations and NIM PCs included in covariates applied to 96 UKBB phenotypes. https://cdn.elifesciences.org/articles/80757/elife-80757-fig3-data2-v2.zip Download elife-80757-fig3-data2-v2.zip We next tested whether the average heritability at an NIM is larger or smaller compared to a MH SNP (Δ^h2=0). We find 17 phenotypes with significant evidence of depleted NIM heritability that include standing height, body mass index, and HDL cholesterol (Z-score < –3; Figure 3b and d). Five phenotypic categories show significant NIM heritability depletion (anthropometry, blood biochemistry, blood pressure, lipid metabolism, lung) in meta-analysis. Meta-analyzing across phenotypes, we find a significant depletion in NIM heritability (meta−Δ^h2 = –1.7 × 10–3, p=2.1×10–36). On average, we find that heritability at NIMs is reduced by about 57% relative to an MH variant with matched MAF and LD characteristics. In contrast to the evidence for depletion in NIM heritability, we find no evidence for traits with elevated NIM heritability across the phenotypes analyzed. We repeated these analyses using NIMs that had been identified using a different approach (Browning et al., 2018) and obtained concordant results (‘Appendix 5’). Our observations are consistent with NIMs having been primarily under purifying selection for thousands of generations (Harris and Nielsen, 2016; Petr et al., 2019). Nevertheless, as evidenced by their overall heritability, NIMs still make a significant contribution to phenotypic variation in present-day humans. We also investigated the impact of controlling for MAF and LD on our findings in UKBB. Analyses that do not control for MAF and LD tend to broadly correlate with our results that control for both (Pearson’s r = 0.96, 0.68, and 0.65 and p<10–12 among h2^ , h2^NIM , and Δ^h2). However, these analyses underestimate both heritability (Figure 4a) and NIM heritability (Figure 4b), resulting in apparent NIM heritability depletion (Z-score < –3) in 83 of the 96 phenotypes (Figure 4c). While yielding qualitatively similar conclusions about the depletion in heritability at NIMs relative to MH SNPs, prior knowledge that per SNP heritability of complex traits can be MAF and LD dependent (Evans et al., 2018) coupled with our extensive simulations lead us to conclude that controlling for MAF and LD leads to more accurate results. Figure 4 with 1 supplement see all Download asset Open asset Comparing heritability analyses with and without controlling for minor allele frequency (MAF) and LD in UK Biobank (UKBB) phenotypes. Each phenotype is shown with one dot colored by the phenotypic category it belongs to, on the y-axis based on its point estimate and standard error (estimated by RHE-mc with Ancestry annotation) and on the x-axis based on its point estimate and standard error (estimated by RHE-mc with ancestry + MAF + LD annotation). Estimates shown are (a) total heritability h2^ , (b) Neanderthal Informative Mutation (NIM) heritability h2^NIM , and (c) the difference between per-NIM heritability and matched modern human (MH) SNPs heritability Δ^h2 . Not controlling for MAF and LD leads to underestimation of NIM heritability, which leads to false positives when testing whether heritability at an NIM is elevated or depleted relative to an MH SNP. Figure 4—source data 1 RHE-mc results with Ancestry only annotation and NIM PCs included in covariates applied to 96 UKBB phenotypes. https://cdn.elifesciences.org/articles/80757/elife-80757-fig4-data1-v2.zip Download elife-80757-fig4-data1-v2.zip An interesting hypothesis is whether the depletion in heritability that we observe here reflects selection specifically against Neanderthal alleles or whether these could represent selection against functional changes in general since prior work has shown that Neanderthal alleles tend to be distributed further away from regions of the genome under selective constraint (Sankararaman et al., 2014; Juric et al., 2016). To answer this question, we can compare the average heritability at NIMs to modern human SNPs matched for B-value, a measure of background selection (McVicker et al., 2009). We attempted to estimate the difference in average heritability between NIMs and MH SNPs (Δh2) while matching on quartiles of B-value bins, in addition to MAF and LD bins. A challenge with this approach is the large number of annotations leads to annotations with few SNPs so that h2NIM estimates are substantially less precise (estimated with standard errors that are about 10 times larger on average than in the setting where we do not match on B-values). Consequently, we do not find a significant difference in the per-SNP heritability at NIMs compared to MH SNPs. Instead, we estimated (Δh2) matching on B-values and MAF having confirmed that the h2NIM estimates are estimated with precision comparable to the setting where we do not match on B-values. In this setting, we continue to observe a significant depletion in NIM heritability across phenotypes (55 phenotypes with Z-score < –3) with no evidence for traits with elevated (Figure 3—figure supplement 1). Taken together, our analyses suggest that depletion in heritability likely reflects selection against Neanderthal alleles rather than selection against variation in functionally constrained regions of the genome in general. Identifying genomic regions at which introgressed variants influence phenotypes Having documented an overall contribution of NIMs to phenotypic variation, we focus on identifying individual introgressed variants that modulate variation in complex traits. We first tested individual NIMs for association with each of 96 phenotypes (controlling for age, sex), 20 genetic PCs (estimated from common SNPs), and 5 NIM PCs (that account for potential stratification that is unique to NIMs). We obtained a total of 13,075 significant NIM-phenotype associations in 64 phenotypes with 8018 unique NIMs (p<10–10 that accounts for the number of SNPs and phenotypes tested) from which we obtain 348 significant NIM-phenotype associations with 294 unique NIMs after clumping associated NIMs by LD (see ‘Methods). A limitation of the association testing approach is the possibility that an NIM might appear to be associated with a phenotype simply due to being in LD with a non-introgressed variant (Skov et al., 2020). We formally assessed this approach in simulations of phenotypes with diverse genetic architectures described previously where the identities of causal SNPs are known. An NIM that was found to be associated with a phenotype (p<10–10) was declared a true positive if the 200 kb region surrounding the associated NIM contains any NIM with a non-zero effect on the phenotype and a false positive otherwise. Averaging across all genetic architectures, the false discovery proportion (FDP; the fraction of false positives among the significant NIMs) of the association testing approach is around 30% (Figure 5b). Hence, finding NIMs that are significantly associated with a phenotype does not confidently localize regions at which introgressed variants affect phenotypes. Figure 5 Download asset Open asset Fine mapping of Neanderthal Informative Mutations (NIMs) in simulations and the UK Biobank (UKBB). (a) Fine mapping pipeline to identify NIMs that aims to identify genomic regions at which NIMs are likely to modulate phenotypic variation (credible NIM regions). (b) Comparison of approaches for identifying credible NIM regions. For each simulation, false discovery proportion (FDP) is computed for association testing compared to our pipeline (combining association testing and fine-mapping). The distributions of the FDP are shown across genetic architectures (summarized across groupings of coupling of effect size, minor allele frequency [MAF] and LD) and summarized across architectures (ALL). Our approach to identifying credible NIMs decreases FDP in all studied architectures (the LOW LD setting has a median and quartiles of zero across replicates). (c) The distribution of the length of credible NIM regions across 96 UKBB phenotypes. (d) Distribution of the ratio between the number of credible NIMs and number of tested NIMs (in the example of panel (a), the number of tested NIMs is the union of NIMs in input to the fine-mapping software (SuSiE) 1 and 2) showing that our approach is effective in prioritizing NIMs that affect phenotype. (e) The distribution of the number of credible NIM regions among phenotypes. The number of credible NIM regions is positively correlated with (f) heritability (g) NIM heritability. Figure 5—source data 1 Fine mapping FDP in simulated data. https://cdn.elifesciences.org/articles/80757/elife-80757-fig5-data1-v2.zip Download elife-80757-fig5-data1-v2.zip To improve our ability to identify NIMs that truly modulate phenotype, we designed a customized pipeline that combines association testing with a fine-mapping approach that integrates over the uncertainty in the identities of causal SNPs to identify sets of NIMs that plausibly explain the association signals at a region (Figure 5a). Our pipeline starts with a subset of significantly associated NIMs that are relatively independent (p<10–10) followed by the application of a statistical fine-mapping method (SuSiE) within the 200 kb window around each NIM signal (Wang et al., 2020) and additional post-processing to obtain a set of NIMs that have an increased probability of being causal for a trait. We term the NIMs within this set credible NIMs while the shortest region that contains all credible NIMs in a credible set is termed the credible NIM region (see ‘Methods; Figure 5a). We employed the same simulations as previously described to evaluate our fine-mapping approach. The fine mapping approach yields a reduction in the FDP relative to association mapping (FDP of 15.6% on average; Figure 5b) while attributing the causal effect to a few dozen NIMs within the credible NIM set (mean: 79, median: 54 NIMs across all simulations). Applying our pipeline to the set of 96 UKBB phenotypes, we identified a total of 112 credible NIM regions containing 4303 unique credible NIMs across 47 phenotypes (Figure 6a). The median length of credible NIM regions, 65.7 kb (95% CI: [4.41 kb, 469.3 kb]) is close to the expected length of Neanderthal introgressed segments (Skov et al., 2020) suggesting that the resolution of our approach is that of an introgressed LD block (Figure 5c). While fine mapping generally attributes the causal signal to a subset of the tested NIMs (mean: 55.8, median: 37 NIMs across phenotypes), the degree of this reduction varies across regions likely reflecting differences in the LD among NIMs (Figure 5d). We do not detect any credible NIM in 49 out of 96 phenotypes potentially due to the limited power of our procedure that aims to control the FDR (Figure 5e). The sensitivity of our method is affected by both total heritability (Figure 5f, Pearson’s r = 0.49, p=3.3×10–7) and NIM heritability (Figure 5g, Pearson’s r = 0.36, p=3.3×10–4). A linear model that uses both total heritability and NIM heritability to predict the number of credible sets yields r2 = 0.29, p=1.3×10–5 and 0.015, respectively, while linear models with only total heritability or only NIM heritability result in statistically lower r2 (0.24 and 0.13, respectively). Figure 6 with 3 supplements see all Download asset Open asset Analysis of credible Neanderthal Informative Mutations (NIMs). (a) Distribution of credible NIMs across the genome. (b) High and moderate impact credible NIMs annotated by SnpEff software (Cingolani et al., 2012). A total of 26 credible NIMs have high (marked in bold) or moderate impact effects on nearby genes (chromosome number and hg19 coordinates). The effects of the SNP and the gene name are displayed. This plot shows significant associations of these NIMs with specific phenotypes (color denotes the phenotype category). (c) Plot of 300 kb region surrounding rs60542959 (marked in black diamond; hg19 coor