Using genomics to understand severe COVID-19 Free

The coronavirus disease 2019 (COVID-19) pandemic has had devastating effects, including the deaths of over 7 million people worldwide, leading to widespread morbidity and exposing stark inequalities in health and social care provision. However, the speed with which the scientific and medical community came together to characterize the epidemiology and pathogenesis of this emerging threat, develop successful testing and treatment strategies, and produce an effective vaccine has shown just how powerful international scientific collaborations can be and leaves a legacy on which to base future translational research.

Approximately 3% of those infected with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) develop critical illness (meaning they require continuous cardiorespiratory monitoring and/or organ support) with sex, comorbidity, age, ethnicity and social factors all influencing outcome [1]. Incredibly, the very first critical COVID-19 genome-wide association study (GWAS) was published online in June 2020 [2], mere months after the start of the pandemic. Using this hypothesis-free genome-wide approach, the authors were able to detect previously unsuspected loci—an extremely valuable tool when facing a novel disease. Furthermore, by focusing on individuals with a severe phenotype (COVID-19 respiratory failure) the authors increased their power to detect genomic signals in an otherwise heterogenous cohort. This first GWAS, conducted in 1610 cases and 2205 controls from Spain and Italy, identified two significant loci: 3p21.31 containing LZTFL1, a regulator of airway ciliary function; and the ABO locus at 9q34.2 [2].

Early on in the pandemic the world's largest study of the molecular mechanisms of critical illness, the Genetics of Mortality in Critical Care (GenOMICC) consortium (https://genomicc.org), used their existing infrastructure to quickly pivot to researching critical COVID-19. In December 2020, this group, led by Professor Kenneth Baillie at the University of Edinburgh, reported another GWAS identifying four additional genome-wide significant loci associated with critical COVID-19 illness in 2244 patients recruited from 208 UK intensive care units [3]. These four loci implicated genes involved in antiviral immunity (IFNAR2, TYK2, OAS1) and lung inflammation (DPP9). TYK2 belongs to the janus kinase (JAK) family which mediates intracellular signalling of extracellular cytokines, including interleukin-6, and this association provided proof-of-concept genomic support for the JAK inhibitor baricitinib. Baricitinib, already licensed for the treatment of rheumatoid arthritis, has since been shown to reduce 28-day mortality in patients hospitalized with COVID-19 in the RECOVERY trial [4], providing a powerful demonstration of how GWAS can be used to identify potential ‘druggable’ targets.

This initial GenOMICC study used a genotyping array to sample a subset (∼700 000) of common variants [minor allele frequency (MAF) >1%] across the genome before imputing (where unobserved genotypes are statistically inferred) up to a total of ∼6 million variants. This genotyping approach is much more affordable than sequencing an entire genome, costing hundreds rather than thousands of pounds, but has limited resolution and excludes rare and structural variants, as well as those commonly seen in non-European populations. In view of this, the GenOMICC team, in partnership with Genomics England, went on to perform whole-genome sequencing (WGS) in 7491 individuals with critical COVID-19 and 48 400 controls [5]. This time, 23 genomic loci were replicated and 16 new associations identified, confirming a role for aberrant interferon (IFN) signalling (IFNA10, IFNAR2, IL10RB, PLSCR1, TYK2) and implicating coagulation factors (F8) and platelet activation (PDGFRL) in the pathogenesis of severe COVID-19 illness. Analysis of rare (MAF <0.5%), deleterious variants failed to identify any candidate genes or in fact replicate previously reported associations with IFN-pathway genes [6] and TLR7 [7], demonstrating that rare variation is not commonly linked to critical COVID-19.

Now, in their most recent update published in Nature, Pairo-Castineira et al. [8] combine WGS with genotyping data for a total of 11 440 critical COVID-19 cases and perform a trans-ancestry meta-analysis with data from other GWAS (COVID-19 Host Genetics Initiative [9], n = 8779; SCOURGE [10], n = 3502; 23andMe [11], n = 495). Analysis of these 24 202 individuals and over 1.7 million controls provided sufficient power to detect a total of 49 significant common genomic associations, 16 of which were novel. Most of these signals had relatively modest effect sizes ranging from 0.92 to 1.2, which is not uncommon in GWAS and does not preclude them from being biologically relevant. This has been shown in the case of low-density lipoprotein cholesterol levels where PCSK9 inhibitors and statins are effective treatments for hyperlipidaemia despite variant associations with small effect sizes [12]. No sex-specific genomic associations were identified despite the observation that male sex is linked with worse outcomes, and it would be interesting to explore whether sex-biased gene expression (driven either by sex hormone–specific transcription factor binding or cell-type specificity) may contribute to these differences [5, 8].

In large studies such as this, obtaining granular phenotypic data for thousands of individuals can be challenging, and although age, sex, ancestry and deprivation score were mostly accounted for in the analysis, limited information regarding comorbidities was available. In a subset of the cohort (∼7500 patients), 43% had hypertension, 14% diabetes and 8% chronic kidney disease (CKD) but failure to account for differences between cases and controls may introduce confounding. Furthermore, vaccination status, administration of glucocorticoids and treatment regimens would have varied between studies, depending on time of recruitment and country of origin, which may also have attenuated illness severity in some cohorts, perhaps masking additional genomic associations.

Fine-mapping of these 16 new loci implicated both inflammatory signalling pathways (JAK1), monocyte-macrophage activation and endothelial permeability (PDE4A), immunometabolism (SLC2A5 and AK5), and viral entry and replication (TMPRSS2 and RAB2A) as having a potential role in the pathogenesis of critical COVID-19 illness (Fig. 1). To further investigate the molecular mechanisms underlying these loci, the authors used a transcriptome-wide association study to infer how genomic variation might influence gene expression in a tissue- and cell-specific manner. Various chemokine receptors were found to be significantly associated with critical COVID-19, including CXCR6 expression in lung tissue and primary monocytes, encoding a protein known to regulate T lymphocyte migration to the lung and maintain airway-resident memory T lymphocytes.

Figure 1:

Miami plot for critical COVID-19 meta-analysis from Pairo-Castineira et al. [8]. Independent lead variants are annotated with their closest genes. The red-dashed line indicates the genome wide significance threshold of 5 × 10⁻⁸. Novel associations are indicated in bold. Blue boxes contain a subset of putative causal genes identified in this and other genomic studies grouped by biological pathway. Potential repurposed drugs and their protein targets are listed in green boxes.

Open in new tabDownload slide

One key aim of GWAS is to identify novel pathophysiological mechanisms underlying disease and generate a list of potential ‘druggable’ targets. This is on the basis that genomic variation is fixed at conception and therefore any detected (and replicated) associations are likely to be causal and that drugs with genomic support have a higher probability of being successful [13]. With this in mind, Pairo-Castineira et al. used a GWAS-specific form of Mendelian randomization (MR) to test for causal associations between critical COVID-19, gene expression and circulating protein levels [8]. MR uses Mendel's independent assortment rule in which random segregation of alleles (where the allele in question is linked to a specific trait) at conception separates individuals independently into two groups with confounding factors distributed equally, rather like an intention-to-treat randomized controlled trial (RCT).

Using this approach, the authors identified five new proteins unique to critical illness, including the innate immune pattern recognition receptor mannose-binding lectin-2 (MBL2), the neutrophil effector enzyme myeloperoxidase (MPO) and the von Willebrand factor cleaving protease ADAMTS13, deficiencies of which lead to thrombus formation as seen in thrombotic thrombocytopenic purpura. RNA expression of the inflammatory cytokine tumour necrosis factor (TNF) was also associated with severe disease, suggesting that inhibition of TNF signalling (e.g. with adalimumab) may be an effective therapy, although evidence for clinical benefit is currently lacking. Additional therapies being considered for repurposing include the inhaled JAK inhibitor nezulcitinib (JAK1, TYK2), the serine protease inhibitors camostat and nafamostat (TMPRSS2), and the anti-inflammatory and immune-modulatory PDE4 inhibitor roflumilast (PDE4A). However, not all significant genomic associations can be successfully translated into effective therapy: MR had previously indicated a protective role for IFNAR2 [3], an IFN receptor subunit, in which rare loss-of-function variants are associated with severe disease [6] but an RCT did not show any mortality benefit of IFN administration in over 2000 hospitalized patients [14].

This study by Pairo-Castineira et al. [8] is the largest one to date to examine the genomic factors predisposing to critical COVID-19, and uses a robust statistical approach to highlight key molecular mechanisms and propose potential therapeutic targets. One caveat is that the cohorts had slightly differing phenotype definitions and each included patients at multiple disease stages including viral exposure, infection and replication, and the development of inflammatory lung disease. It is therefore difficult to pinpoint exactly when in the COVID-19 disease course the genomic associations might be having an effect, although the focus on critical illness increases the probability that the identified signals relate to the later immune-mediated phase. It should also be noted that most of the individuals in this study (87.2% of those with critical illness) are of European ancestry which may limit the generalizability of these findings to other population groups. This emphasizes the need for a more inclusive and diverse approach to genomic studies, not only to increase discovery power but also to ensure that the benefits of these types of studies are relevant to all.

The COVID-19 Host Genetics Initiative, a collaboration bringing together the global genomics community, have subsequently released an updated meta-analysis including 21 194 cases with critical illness from 35 countries (including 7491 individuals from GenOMICC [5]) and over 2 million controls [15], replicating many of the loci reported by Pairo-Castineira et al. [8] Interestingly in this study, MR found that genomic variation linked to CKD and estimated glomerular filtration rate (eGFR) was causally associated with critical COVID-19 illness. However, given that several eGFR loci overlap with related traits such as diabetes, hypertension and ischaemic heart disease [16] (a phenomenon called pleiotropy), attributing direct causation to CKD and critical COVID-19 is difficult.

The kidney community is well aware of the increased risks of COVID-19 faced by patients with CKD. Analysis of electronic health records in over 17 million patients in the OpenSAFELY cohort study demonstrated that reduced kidney function (<30 mL/min/1.73 m²) was associated with a higher risk of COVID-19-related death (adjusted hazard ratio 2.52) than obesity, chronic heart disease, diabetes or non-haematological malignancy [17]. This increased mortality risk observed in CKD stage 4 and 5, dialysis and transplant populations persisted across successive COVID-19 waves likely reflecting the impaired vaccine responsiveness seen in these patients [18, 19] and potentially compounded by the exclusion of kidney patients from many of the COVID-19 therapeutic trials.

In summary, these statistically robust genomic studies have now identified several pathophysiological mechanisms underlying critical COVID-19 illness: innate antiviral defences (IFNAR2 and OAS1), airway mucus/cilia (MUC5B and LZTFL1), host-driven inflammatory lung injury (JAK1, DPP9, TYK2, CXCR6, CCR2) and host–pathogen interactions (ACE2, TMPRSS2 and RAB2A). They serve as an exemplar of how a collective and unbiased genomics approach can highlight key biological pathways underlying disease severity and be used to pinpoint potential therapeutic targets. Looking forward, let us hope this era of open and collaborative science is here to stay.

CONFLICT OF INTEREST STATEMENT

None declared.

REFERENCES