Our study revealed that genetic variation at several loci throughout the human genome is potentially related to the severity of the long-term sequelae of COVID-19, which was quantified here by one of three PCS scores previously developed by our group for the work-up of the SARS-CoV-2 pandemic. As can be inferred from the Manhattan plots illustrating the GWAS results (Figs. 1, 2 and 3), however, none of the observed associations reached the stringent level of so-called ‘genome-wide significance’ (5 × 10− 8). Similarly, the corresponding QQ plots do not suggest strong genetic effects upon the three PCS scores studied (Supplementary Fig. 4). Although adjusting the analyses for additional covariates would reduce the p values of individual genotype-phenotype associations further, particularly when considering BMI (Supplementary Table 5), the main results would however remain unchanged in that the allelic odds ratios estimated for the minor alleles of all lead SNPs virtually stayed the same (Supplementary Table 6).
In our view, this outcome does not detract from the scientific value of the GWAS. On the contrary: Not only would non-reporting of a lack of genome-wide significant results contribute to publication bias, the information that PCS, when regarded as a composite phenotype, does not appear to have a strong genetic basis is very important to guide future research in this field.
While we could not reproduce, even at a less stringent ‘suggestive’ significance level of 10− 5, the single genome-wide genotype-phenotype association reported so far for PCS, namely with the FOXP4 gene on chromosome 68, we nevertheless identified SNPs in other genomic regions that are at least suggestive of a functional involvement into PCS etiology and that point towards biologically meaningful links to the phenotype in question. This particularly includes olfactory impairment in the case of rs10893121, the SNP most significantly associated (p = 2.5 × 10− 6) with the original PCS score.
Previous findings that acute COVID-19 severity and individual resilience are the sole main predictors of PCS severity previously encouraged us to develop two additional PCS scores, PCS-S and PCS-R, to address different sub-domains of PCS that are related to either predictor. While the PCS-S score comprises a number of rather different symptom complexes (impairment of smell and taste, fatigue, joint or muscle pain, general signs of infection, and exercise intolerance), the PCS-R score only included two complexes, in addition to fatigue, namely neurological ailments and sleep disturbances. Interestingly, it turned out that the potential genotypic associations with the two predictor-specific scores did not overlap with one another, providing evidence that they may indeed reflect different clinical entities. Since their constituent symptom complexes were also non-overlapping, it was not surprising that the two sub-scores were associated with different SNPs, and even although the observed associations were comparatively weak, they may be worth follow-up research to shed more light on the etiology of PCS.
PCS undoubtedly is a heterogeneous clinical condition, and the major hypotheses about PCS pathophysiology currently include (I) immune dysregulation, (II) microbiota dysbiosis, (III) autoimmunity and immune imprinting, (IV) blood clotting and endothelial abnormalities and (V) dysfunctional neurological signalling3. In the following, we will assess whether the potential genetic associations identified for the three PCS scores in the present GWAS would be consistent with the above hypotheses, considering possible functional links to genes located within or near the associated regions.
PCS score
rs10893121: olfactory receptors OR4D5, OR6T1, OR10S1, OR10G4, OR10G7, OR10G8 and OR10G9
The perception of smell is triggered by odorant molecules in the nasal tissue that initiate a neural response. Olfactory receptors are members of a large family of G protein-coupled receptors that are responsible for the G protein-mediated transduction of olfactory sensory signals. OR4D5, OR6T1, OR10S1, OR10G4, OR10G7, OR10G8 and OR10G9 all belong to this protein family17,18. While their role in olfactory function is clear-defined, no particular disease has so far been associated with SNPs in the corresponding gene regions. Important in the present context, impairment of smell and taste contributes only 3.5 points to the PCS score (range: 0 to 59) and hence adds only little to the overall severity of PCS. Moreover, while impaired smell is a pathognomonic feature of acute COVID-19 that may persist for a long time, its impact on health-related quality of life is not very strong. The fact that SNP rs10893121 showed the most significant association of all with the original PCS score indicates that variation in olfactory receptor function may nevertheless be a genetic cause of longstanding olfactory impairment after (even mild) SARS-CoV-2 infection.
rs61739314: zinc fingers and homeoboxes 3 (ZHX3), lipin 3 (LPIN3), Elastin microfibril interfacer 3 (EMILIN3), chromodomain helicase DNA-binding protein 6 (CHD6)
The products of the ZHX gene family, including ZHX3, comprise two C2H2-typic zinc fingers and other proteins that may function as transcriptional repressors19,20. While there is evidence that these gene products represent (unfavourable) prognostic markers of renal, urothelial, endometrial and thyroid cancer, no associations have been reported so far to cardiovascular or inflammatory diseases21,22,23. Furthermore, no ZHX gene-encoded mRNA is detectable in immune cells, and the corresponding proteins are not found in blood. Therefore, a causal role of these genes in PCS pathophysiology is not very likely.
The protein encoded by the CHD6 gene can function as a transcriptional repressor and is involved in the cellular repression of influenza virus replication24. This role in host-virus interaction, and the capability of the protein to activate gene transcription in response to oxidative stress through an interaction with NFE2L2, may be a plausible explanation for the association between local SNP rs61739314 and PCS25.
The LPIN3 gene is ubiquitously expressed in duodenum, skin and more than 20 other tissues. One of the functions of lipin complexes is to contribute to gene regulation by acting as transcriptional co-activators in the nucleus26,27. In addition, lipin complexes process precursors of triglycerides and phospholipids in the cytoplasm28. However, protein functions that might be relevant in PCS pathophysiology are not immediately apparent.
Finally, the EMILIN3 gene product is part of the collagen-containing extra-cellular matrix. While the gene is highly expressed in connective tissue in both female and male reproductive organs, a connection to pathophysiological processes of PCS is also not obvious29.
PCS-S score
rs9792535 – NIMA related kinase 6 (NEK6), proteasome 20 S subunit beta 7 (PSMB7) and adhesion G protein-coupled receptor D2 (ADGRD2)
The ‘never in mitosis A’ (NIMA) gene of Aspergillus nidulans encodes a serine/threonine kinase that controls initiation of mitosis. Human NIMA-related kinases (NEKs), like NEK6, are homologues of fungal NIMA and perform similar functions. Inhibition of the proteins can lead to apoptosis30,31. The NEK6 gene is most abundantly expressed in gallbladder but is neither specific to immune cells nor expressed in blood. While there are some connections between the gene and tumorigenesis, e.g. by suppressing p53-induced cancer cell senescence32, no link to PCS pathophysiology is obvious.
Proteasome 20 S subunit beta 7 (PSMB7) is a multi-catalytic proteinase complex that is distributed throughout eukaryotic cells and cleaves peptides in a non-lysosomal pathway. Gamma interferon may downregulate this proteosomal catalytic subunit. The PSMB7 complex plays many essential roles in the cell by associating with different regulatory particles. Removing misfolded or damaged proteins that could impair cellular function is one of the major capacities of the gene product33,34. Adhesion G protein-coupled receptor D2 (ADGRD2) is predicted to be involved in adenylate cyclase-activating G protein-coupled receptor signaling pathway and is an integral component of the membrane35,36. Therefore, an involvement of PSMB7 or ADRG2 also in processes hypothesized to cause PCS appears plausible.
rs58659000, rs10465914 – MAST4, FBXO42 and SZRD1
The MAST4 gene encodes a protein belonging to the microtubule-associated serine/threonine protein kinases, which are mainly expressed in cytoplasm, predominantly in esophagus and urinary bladder37. FBXO42 is a member of the F-box protein family and is characterized by a 40 amino acid F-box motif. Full-length cloning of FBXO42 in a mammary library identified the so-called ‘Just one F-box and Kelch domain-containing protein’ (JFK), a critical negative regulator of p5338. Both FBXO42 and SZRD1 are negative markers of liver cancer. Finally, SZRD1 suppresses cell proliferation by inducing cell cycle arrest and apoptosis39. All three genes are thus not obviously connected to pathophysiological processes relevant for PCS onset or severity.
PCS-R score
SLC7A2, PDGFRL, MYOCD, ARHGAP44, CHST11, TMEM200A and CADM2
The protein encoded by the SLC7A2 (solute carrier family 7 member 2) gene is a cationic amino acid transporter belonging to the APC (amino acid-polyamine-organocation) family of transporters40. Located in the cell membrane, it is responsible for the cellular uptake of arginine, lysine and ornithine. Three transcript variants encoding different isoforms have been detected for this gene40. SLC7A2 may play a role in macrophage activation through its role, as a member of the cationic amino acid transporter protein family, in L-arginine transport.
The PDGFRL gene encodes a protein with significant amino acid sequence similarity to the ligand binding domain of platelet-derived growth factor receptor beta41. Mutations in this gene, and the deletion of a chromosomal segment containing this gene, are both associated with sporadic hepatocellular carcinomas, colorectal cancers and non-small cell lung cancers, suggesting that the gene product may function as a tumor suppressor42,43.
The protein encoded by the MYOCD gene is found in smooth muscle cells and cardiac muscle cells. Through forming a complex with serum response factor, it functions as a transcriptional activator of CArG box-dependent cardiac promotors44. The two proteins play a crucial role in cardiogenesis, urinary bladder development and the differentiation of smooth muscle cells (myogenesis)44. Potential clinical phenotypes associated with variation in the MYOCD gene include congenital megabladder, but no association to inflammatory, auto-immune or other processes relevant to PCS are known.
ARHGAP44 enables phospholipid binding activity and is involved in actin cytoskeleton dynamics for filopodia protrusion and cell migration45. It is predicted to play a role in several processes, including modification of dendritic spine, negative regulation of Rac protein signal transduction and regulation of plasma membrane-bounded cell projection organization. The protein is not detected in immune cells or blood. Previous studies suggested that ARHGAP44 has an indirect influence on viral-induced tetherin signaling46. There is also growing evidence that ORF7a and Spike act as tetherin antagonists in SARS-CoV-2 infections47,48, and another study suggested that SARS-CoV-2 inhibits ORF3a tetherin by trapping it in late endocytic organelles49.
The protein encoded by the CHST11 gene belongs to the sulfotransferase 2 family. A chromosomal translocation, t(12;14)(q23;q32), involving this gene and IgH has been reported in a patient with B cell chronic lymphocytic leukemia34. Furthermore the protein is a negative marker of renal and urothelial cancer50. TMEM200A is predicted to be an integral component of the membrane, but its detailed function is still unknown.
The protein encoded by the CADM2 gene belongs to the synaptic cell adhesion molecule 1 (SynCAM) family, which is part of the immunoglobulin superfamily51. It has a cytosolic binding site for members of the protein 4.1 family, known to interact with cytoskeletal proteins. Variants in the CADM2 gene are associated with post-bronchodilator FEV1 and FEV1/FVC ratio, alcohol consumption, general risk-taking tendency and adventurousness, and BMI52,53,54.
Strengths and limitations
The major strength of our GWAS is that it was based upon a prospective population-based cohort study of PCS with deep phenotyping of participants. Although the PCS scores used in our study drew upon patient self-reports, clinical characteristics of study participants, including laboratory measurements, vital signs, lung function, neurological testing, and echocardiography are available for possible follow-up analyses.
Compared to GWAS of other complex human phenotypes, the sample size of the present study (maximum n = 2,216) was modest, which may have resulted in limited power to detect weak to moderate genetic effects. However, by lowering the significance level accordingly, we aimed to reduce the risk of false negative results to a level that, we believe, ensured that clinically and biologically truly significant associations were unlikely to have been missed. Indeed, inspection of the PCS score-specific QQ plots obtained in the three GWAS reveals that an enrichment of such associations is conceivable at least for the acute severity-sensitive PCS-S score (Supplementary Fig. 4).
