The 100KGP was established to sequence 100,000 genomes from around 85,000 NHS patients in the United Kingdom affected by a rare disease, cancer, or infectious disease, with detailed phenotype and clinical data (https://www.genomicsengland.co.uk/) (100,000 Genomes Project, 2019). Combining WGS data with medical records has created a ground-breaking research resource enriched for genetic diseases. As part of its recruitment target, Genomics England has highlighted the importance of engaging with the UK’s increasingly diverse black and minority ethnic population, who are underrepresented in clinical studies and research.

This dataset consisted of 6,274 patients who were tested for SARS-CoV-2, of which 1,837 tested positive. Of the positive cases, 1,411 (77%) were classified as not hospitalised, and 426 (23%) were classified as inpatients or deceased. We also used WGS data from a large control group consisting of 37,207 unrelated individuals from 100KGP (males = 17,066; females = 20,141; total allele number = 57,348), to scan for ACE2 variants and eQTL distribution in different populations.

In accordance with the 100KGP, the 13 NHS Genomic Medicine Centres have established Patient and Public Involvement Network with leads that work together with representatives from Genomics England and NHS England to curate informed consent and the literature development surrounding the data used within this study. Patients and carers also play a role in the Ethics Advisory Committee for data used within this study. In addition, patients and carers are part of a participant panel that help to assure the data access process has the interests of participants as a central goal, through involvement in the data access review committee, Genomics England Clinical Interpretation partnership board, and ethics advisory committee. As part of a public engagement program Genomics England also can disseminate the results of this study through seminars, in an open-to-everyone and free-to-the-public setting.

Using multi-sample VCFs for participants in the 100KGP cohort, we generated Principal Components (PCs), calculated pairwise relatedness among samples, and estimated probabilities of genetic ancestry for five broad superpopulations. Broad genetic ancestry was estimated using ethnicities from the 1,000 genomes project phase 3 (1KGP3), by generating PCs for 1KGP3 samples and projecting all 100KGP participants onto these. The five broad superpopulations are European, South Asian, African, East Asian, Ad Mixed American, and multiracial individuals, labelled as Other.

Patient hospitalised status is based on whether the collected specimen was from an acute (emergency) care provider, an accident and emergency department, an inpatient location, or a result of health care associate infection. “Not hospitalised” is interpreted as there is no evidence that the patient has been an inpatient in the microbiological record, but this may still be a possibility as the microbiology data source is not linked to admissions data.

We employed a computational method to predict the effect of the coding variants identified in the 100KGP dataset, based on homology modelling and computational docking of the virus-receptor protein-protein interaction. Several crystal structures of the SARS-CoV-2 RBD and ACE2 have been reported (pdb identifiers: 6VW1 (Shang et al., 2020), 6LZG (Wang et al., 2020b), 6M0J (Lan et al., 2020) 7A94 (Benton et al., 2020)) allowing us to demonstrate the utility of this model. Amino acids of interest were mutated in Coot (Emsley et al., 2010) and visually analysed using UCSF Chimera (Pettersen et al., 2004) Mutations were ranked according to their potential to affect protein stability and activity. Structure figures were prepared in Chimera.

We obtained expression data through the genotype-tissue expression (GTEx) database (https://gtexportal.org/home/) (Lonsdale et al., 2013) This data was also used to extract expression quantitative trait loci (eQTLs) for ACE2 (all data based on RNA-seq experiments). Population allele frequencies were obtained from 100KGP.

The raw list of SNVs and indels among cases and controls were annotated using ANNOVAR (Wang et al., 2010). Variants in splicing regions, 5′-UTR, 3′-UTR, and protein-coding regions, such as missense, frameshift, stop loss, and stop gain mutations, were considered. In silico prediction of pathogenicity was assessed using PolyPhen2 (Adzhubei et al., 2013), CADD (Rentzsch et al., 2019), REVEL (Ioannidis et al., 2016) and GERP++ (Petrovski et al., 2015) scores. The highest priority variants were nonsynonymous or in the splice site, with a corresponding REVEL score above the default threshold of 0.5, a GERP++ score greater than two, and PolyPhen2 prediction outcome of “probably damaging”.

The data were analysed using chi-square tests to identify instances where there were differences between any of the groups (with Fisher’s exact where any expected value <5). Statistically significant differences were followed up with z-tests to compare proportions between pairs of groups and identify where the group differences were located (with Bonferroni correction). Statistical significance was considered at the 5% level (two-tailed).

To see if there is an observable difference in SARS-CoV-2 positivity and disease severity in different ethnic backgrounds, we used a unique dataset containing genetic information within the 100KGP. This dataset includes the SARS-CoV-2 test results, genetically predicted ancestry, and hospitalisation status of each patient. We stratified positive cases in the 100KGP by the five predominant superpopulations representing the general population in the United Kingdom (Figure 1A). Our data shows that 1,254 individuals of European (68.3%), 322 of South Asian (17.5%), 49 of African (2.7%), 15 of East Asian (0.8%), 4 of Ad Mixed American (0.2%) background, as well as 193 uncategorised individuals, labelled as Other (10.5%), tested positive for SARS-CoV-2 (Figure 1B).

The chi-square test was used to compare proportions of inpatients, demonstrating that hospitalised outcomes in SARS-CoV-2 positive patients were statistically different across ethnic groups (χ² = 129.64, df = 5, p < 0.001). Individuals of East Asian and African background were approximately 50% more likely to be hospitalised compared to Europeans. Furthermore, the largest difference in hospitalisation rate was found between South Asians (16.2% hospitalised) and East Asians (46.8% hospitalised) (Figure 1C; Supplementary Table S1), suggesting that individuals of South Asian decent are less likely to be hospitalised from SARS-CoV-2 infections. In addition, Ad Mixed Americans did not statistically differ between any ethnic group, which may be due to the small sample size (n = 4). Nonetheless, these data provide evidence for varying degrees of susceptibility to COVID-19 in different ethnicities.

We hypothesise that these findings may reflect an underlying biological driver for viral susceptibility in different genetic backgrounds. To investigate this, we interrogated the 100KGP patient database for correlations between SARS-CoV-2 positivity status and genetic variants that alter ACE2 protein structure, as well as expression. Although many mechanisms are involved in SARS-CoV-2 entry and host responses, we focused on ACE2 because it is the primary host receptor for viral entry and a limiting factor for infection.

To identify whether variants within ACE2 could have a predicted influence over S protein-host cell interactions, we first identified coding variants within 37,207 unrelated individuals in the 100KGP dataset. This dataset represents the ethnic diversity of the United Kingdom population. Coding variants were stratified by the five predominant superpopulations represented within the United Kingdom population (European, South Asian, African, Ad Mixed American, and East Asian descent) and by allelic frequency. The resulting analysis provided 114 SNVs (Figure 2A), three deletions, and one insertion leading to a frameshift identified in one individual of European descent (Supplementary Table S2).

To see if any of these variants have the potential to affect binding of S protein, we compared our findings against data from a high-throughput functional mutagenesis screen that used deep sequencing to assess S-protein binding abilities to 2340 ACE2 variants (Chan et al., 2020) In the 100KGP control dataset, we identified 21 ACE2 nonsynonymous variants that alter S protein binding, seven that increase binding and 14 decrease binding (Figures 2B,C ). We stratified these variants based on allelic frequency within different ethnic backgrounds. Populations of African and South Asian descent generally had the greatest number of alleles with increased frequency, relative to the general population (Figure 2A). This was also reflected in variants that both increase and decrease S protein binding affinity for ACE2 (Figures 2B,C). Specifically, we detected a greater frequency of variants in South Asians that decrease SARS-CoV-2 S protein binding affinity for ACE2 (Figure 2C), and one variant enriched within African descent populations that increases S protein-binding affinity for ACE2 (Figure 2B).

To see if any coding variants in ACE2 correlate with decreased or increased propensity for SARS-CoV-2 positivity, we analysed WGS data of 1,837 SARS-CoV-2 positive patients and identified 18 nonsynonymous ACE2 variants (Table 1; Supplementary Table S3). Six nonsynonymous variants were novel (p.V316L, p. H345R, p. H401R, p.M455V, p. Y515N, p. P737L), of which three are predicted to be deleterious based on in silico predictions. Three variants (p.K26R, p. H378R, p. Y515N) found across the interaction surface were shown to alter receptor affinity for the viral S protein based on deep mutagenesis experiments (Chan et al., 2020). Variants p. H378R and p. K26R increased binding, whereas the novel p. Y515N variant decreased binding.

In addition, we analysed the effect of these variants on protein structure and assessed if they might alter the interaction between ACE2 and the SARS-CoV-2 S protein (Figure 3). While predicting the effect of single mutations in the core of the protein on the interaction of ACE2 with the S protein is challenging, we can nevertheless make predictions on the impact on protein stability. We selected five mutations for structural analysis (H345R, H378R, H401R, M455V and Y515N). H378R is the mutation with the most severe destabilisation effect on ACE2 as it directly coordinates the Zn²⁺ ion bound by the HEXXH + E motif of the ACE2 metallopeptidase. Similarly, H401 disrupts an H-bond between H378 and H401, destabilising the motif, likely to a lesser extent. H345R does not interact directly with either of the amino acids that coordinate the Zn²⁺ ion, but its substitution with arginine could have a negatively affect. Y515 can be found lining the pocket providing access to the Zn²⁺ motif and mutations could interfere with potential interaction partners. S-aromatic motifs have been reported to have stabilising effects on proteins (Aledo, 2019); thus, substituting methionine for valine at position 455 can negatively affect protein stability as M455 interacts with W477 through an S-aromatic interaction (Figure 3).

We also identified an ACE2 splice region variant (rs2285666 C/T) that was significantly more prevalent among SARS-CoV-2 patients than the control population (p = 0.015). Further, intra-ethnic comparisons showed that this variant was more prevalent in inpatients (minor allele frequency (MAF) = 23.9%) compared to patients not hospitalised (MAF = 18.4%) within the European population (p = 0.029), suggesting it plays a role in susceptibility and severity (Supplementary Table S4).

The rs2285666 C/T SNP is located on the fourth base position of intron three, which may affect gene expression via mRNA alternate splicing mechanisms (Yang et al., 2015). According to the HaploReg tool (v4.1), rs2285666 alters transcription factor binding sites; namely, HNF1, and Ncx motifs. Using an ELISA method, Wu et al. (2017) showed that the rs2285666 SNP elevated serum ACE2 levels, with the T/T genotype increasing expression by almost 50%. Similarly, the GTEx database classifies this variant as a significant eQTL, which is associated with an increased expression of ACE2, CA5BP1, PIR and TMEM27, and decreased expression of CA5B.

Of note, previous studies have identified rs2285666 as a risk factor for hypertension, type 2 diabetes, and coronary heart disease (Roach and Bugan, 2014; Asselta et al., 2020). Taken together, this polymorphism may increase SARS-CoV-2 infection rates by altering ACE2 expression and predisposing to comorbidities observed in COVID-19 patients.

We also analysed ACE2 variants in the 3′UTR, 5′UTR and promoter regions, where the promoter is defined as 2 kb upstream of the start codon. We found no significant differences in the allele frequencies between SARS-CoV-2 positive patients and controls, or patients stratified by hospitalised status. However, when examining intra-ethnic differences in allele and haplotype frequencies, the rs199651576 T/C 5′UTR polymorphism was found to be higher in SARS-CoV-2 patients (MAF = 0.12%) compared to controls (MAF = 0.016%) within the European population (p = 0.039) (Supplementary Table S4). The function of this variant remains unclear, although it may be involved with the translational regulation of the ACE2 mRNA transcript.

To see if any of the variants found within SARS-CoV-2 positive patients differ between ethnicities, we compared the MAFs from 37,207 unrelated individuals, representing the general population of the United Kingdom Z-test comparisons were made between ethnic groups (with Bonferroni correction where there were more than two groups). The chi-square test was used to see whether distributions of variables differ from each other. Variants p. K26R (χ² = 23.12, df = 3, p < 0.001) and p. N720D (χ² = 137.55, df = 4, p < 0.001) were statistically different between groups. These variants were also the most frequent in the 100KGP cohort and Genome Aggregation Database (gnomAD) (Karczewski et al., 2020).

The p. K26R variant was more common in Europeans (MAF = 0.710%) compared to South Asians (MAF = 0.275%; p < 0.001) and Africans (MAF = 0.155%; p = 0.001). The K26 residue lies on the proximal end of the SARS-CoV-2 RBD–ACE2 interface (Lan et al., 2020). Structural predictions have shown that the K26R variant enhances the affinity of ACE2 for SARS-CoV-2 by strengthening the hydrogen bond between H34 and Y453 of the S-protein (Stawiski et al., 2020; Al-Mulla et al., 2021). This is consistent with deep mutagenesis experimental data using a synthetic human ACE2 mutant library (Chan et al., 2020).

Variant p. N720D was also more common in Europeans (MAF = 2.64%) compared to South Asians (MAF = 0.609%; p < 0.001), Africans (MAF = 0.285%; p < 0.001) and East Asians (MAF = 0.198%; p < 0.001). The p. N720D variant lies in the C-terminal collectrin domain of ACE2. Although it is not involved in the SARS-CoV-2 S-protein interaction, studies have shown that it enhances transmembrane protease, serine 2 (TMPRSS2) binding and cleavage of ACE2. As TMPRSS2 cleavage of the ACE2 receptor increases viral entry, the p. N720D variant may lead to increased COVID-19 susceptibility (Mohammad et al., 2020; Al-Mulla et al., 2021).

Finally, the p. Y497H variant was more prevalent in East Asians (MAF = 0.198%) compared to Europeans (MAF = 0.00203%) (p < 0.001), but absent in other populations (Table 1; Supplementary Table S5).

While protein-protein interactions can explain some components of viral entry and potential susceptibility in variant carrying at-risk groups, differential expression of ACE2 may also be a contributor to viral-host cell interactions.

Among the 31 GTEx human tissues, the highest ACE2 expression levels are found in the testis, small intestine, kidneys, heart, thyroid, and adipose tissue (Lonsdale et al., 2013). The lowest ACE2 expression levels are found in whole blood, spleen, brain, and skeletal muscle. The lungs, pancreas, oesophagus (muscularis) and liver had intermediate ACE2 expression levels.

The Human Protein Atlas (HPA) (Thul and Lindskog, 2018) database shows that ACE2 had relatively high expression levels in the duodenum, small intestine, gallbladder, kidneys, testis, seminal vesicle, colon, rectum, and adrenal gland. The HPA database also showed that the gastrointestinal tract (duodenum, small intestine, colon, and rectum), kidney, gallbladder, and male tissues (testis and seminal vesicle) had high expression levels of both the ACE2 protein and gene. Interestingly, the results obtained from the HPA (fresh frozen tissue) and GTEx (postmortem tissue) are similar, suggesting negligible effects of the sampling procedures used by the GTEx consortium on RNA degradation.

We also interrogated the GTEx database for the distribution of ACE2 eQTLs (GTEx analysis Release V8). For ACE2, we identified 288 significant (FDR <0.05) eQTLs in 20 different tissues, including, adipose (n = 31; 10.8%), brain (n = 246; 85.4%), testis (n = 2; 0.7%), and prostate (n = 60; 20.8%). It would be valuable to compare the frequency of eQTL variants with ACE2 expression specifically in the lung with susceptibility to viral infection and severity of COVID-19. However, to date, no eQTL for ACE2 in the lung has been reported in the GTEx database, thus warranting further investigations in this regard.

The allele frequencies of the 288 eQTL variants were compared among different ethnicities in the 100KGP dataset (Figure 4; Supplementary Table S6). Across this cohort, we performed a case-control genetic association analysis on ACE2 eQTLs. Of the 288 eQTLs associated with tissue expression of ACE2 in the GTEx database, 132 were significant between patients and controls. Twenty-three were significantly different after Bonferroni correction, seven of which had a Cohen’s effect size, d, of >0.2. The largest MAF difference was observed in rs12006793 (d = 0.91). This eQTL decreases ACE2 expression and was more common in the control group (Supplementary Table S7,S9), suggesting it reduces COVID-19 risk. This eQTL was statistically different between ethnic groups (χ² = 801.06, df = 4, p < 0.001). The eQTL prevalence significantly differed between all ethnic comparisons, except between Ad Mixed Americans and East Asians, and Ad Mixed Americans and Europeans. It was more common in Africans (MAF = 66.6%) and South Asians (MAF = 60.1%). However, according to GTEx, it was only expressed in adipose tissue (omentum) with a normalised enrichment score of -0.1.

When comparing hospitalised SARS-CoV-2 positive patients with patients not hospitalised, only two eQTLs were statistically significant after multiple testing correction, but none had a Cohen’s effect size greater than 0.2. Interestingly, however, rs12006793 was one of the two significantly enriched among the patients stratified by hospitalised status (Supplementary Table S8). Based on this data, rs12006793 may be an important marker for COVID-19 susceptibility and severity.