A review of genetic risk in systemic lupus erythematosus

ABSTRACT

Introduction

Systemic Lupus Erythematosus (SLE) is a complex multisystem autoimmune disease with a wide range of signs and symptoms in affected individuals. The utilization of genome-wide association study (GWAS) technology has led to an explosion in the number of genetic risk factors mapped for autoimmune diseases, including SLE.

Areas covered

In this review, we summarize the more recent genetic risk loci mapped in SLE, which bring the total number of loci mapped to approximately 200. We review prioritization analyses of the associated variants and experimental validation of the putative causal variants. This includes the implementation of new bioinformatic techniques to align genomic and functional data and the use of transcriptomics with single-cell RNA-sequencing, CRISPR genome editing, and Massive Parallel Reporter Assays to analyze non-coding regulatory genetics.

Expert opinion

Despite progress in identifying more genetic risk loci and variant-gene pairs for SLE, understanding its pathogenesis and applying findings clinically remains challenging. The polygenic risk score (PRS) has been used as an application of SLE genetics, but with limited performance in non-EUR populations. In the next few years, advancements in proteomics, post-translational modification estimation, and whole-genome sequencing will enhance disease understanding.

KEYWORDS:

• Expression quantitative trait loci (eQTL)
• genome-wide association study (GWAS)
• genetic variant
• immunogenetics
• Mendelian randomization
• systemic lupus erythematosus (SLE)

1. Introduction

The immune system is an interactive network of cells and molecules with specialized roles in host defense against infection. The ability to discriminate between self and non-self antigens is crucial to the function of the immune system as a specific defense against invading microorganisms. Failure of self-tolerance can result in pathological states causing autoimmune disease. Autoimmune diseases have an overall prevalence of 3–5% in the general population [1]. Although a few autoimmune diseases, such as autoimmune polyendocrinopathy syndrome type 1, which can be explained by a mutation in the autoimmune regulator (AIRE) gene, are monogenic; most autoimmune diseases are polygenic (complex genetic traits), with each susceptibility gene conveying a modest and non-exclusive elevation of risk.

Systemic Lupus Erythematosus (SLE) has a polygenic basis. Its incidence is affected by genetic ancestry and sex. The female: male incidence ratio is 10:1 in adults. The disease burden of SLE has been shown to be highest for those with African ancestry, followed by Asian (south or east) and those with European ancestry have the lowest prevalence (and severity) of SLE. The updated incidence rates of SLE incidence range from 1.5 to 11 per 100,000 person-years globally, with a prevalence of 13 to 7,713.5 per 100,000 person-years [2]. Higher mortality rates are observed among SLE patients with African ancestry compared to those of European ancestry, potentially due to disease severity and socio-economic factors.

Disease is triggered when the underlying genetic propensity for SLE interacts with environmental triggers. Some examples have been described, such as excessive UV exposure; the role of other factors such as viral infection remain unclear, particularly regarding how specific such triggers are. Once the pathological process has been initiated, the adaptive immune system plays a major contributory factor to target organ injury through immune complex formation. Thus, B cell activation plays a prominent role in SLE. However, another immune hallmark of SLE is activation of the innate immune system, which in part explains the widespread scope of the immune assault, the multiple potential targets, and the heterogeneity of the clinical disease.

Genome-wide association studies (GWAS) have led to radical increase in the identification of genetic risk factors that underpin complex diseases. In a review article published in 2017, we summarized the 84 risk loci that were reported from GWA studies [3]. This number has increased to ~ 200 following a few new large-scale GWA studies published in the last five years. In this Review, we focus on these reports from the past 5 years to provide an update on the landscape of genetic risk of SLE while discussing promising new technologies and tools that will be able to facilitate research in SLE.

2. Findings from the recently published GWA studies

Ten large-scale (>1,000 individuals) GWAS and one exome-wide association study in SLE have been published in the past five years. Among them, four [4–7] are of Asian origin only, and one [8] is of European ancestry only, another five [9–13] include individuals of multiple ancestries. Determining new results from many of these studies is complex as they frequently include meta-analyses of other out-of-study SLE association data. Full analysis of these trans-ancestral studies, including studies focusing on Asian populations, will be able to correct the European bias in the genetic basis of SLE findings previously reported. Yin X et al. [4] performed the largest genetic association study of SLE to date, which included 208,370 East Asians (largely controls) and found 46 SLE risk loci specific to Asian populations. We summarize the novel genetic risk loci for systemic lupus erythematosus identified from recently published genome/exome-wide association studies in SLE (Supplementary Table S1). This list includes two HLA loci and 127 non-HLA loci, together with two loci on the X chromosome. Among 127 novel non-HLA risk loci, 75 are Asian-specific, and 5 are European-specific.

3. Missing heritability in the genetics of SLE

Although many genetic risk loci for SLE have been identified, the explained heritability by these risk loci is still low. Morris DL et al. [14] used all genotyped SNPs in a Chinese cohort and a European cohort to calculate heritability in both populations. The results revealed 28% explained heritability in Chinese subjects and 27% in Europeans. López-Cortegano E et al. [15] reported the 17% explained heritability in SLE calculated using a selection of the most informative SNPs (P-value ≤5 × 10⁻⁸) available from the NHGRI-EBI GWAS Catalog [16]. The number was 23% when SNPs were not filtered by P-value [17].

4. From association to causality in the genetics of SLE

Two simple factors limit the functional interpretation of GWAS results: (1) The genomic loci identified by GWA studies are often physically broad. (2) More than 90% of the risk alleles reside within the non-protein coding region and thus the a priori functional effects of the genetic variants at an individual risk locus are usually not fully understood. A summary of the steps that can be followed in moving from initial mapping of a risk locus to risk gene identification at a locus and the steps toward a mechanistic link to pathology is outlined in Figure 1 [18]. Detecting the true causal genetic variant(s) within a given risk locus is relatively challenging because of the linkage disequilibrium (LD) between associated variants. The statistically associated variants can be either genuinely causal or (more likely) detected because of their strong correlation (LD) with the causal variant(s). Physical (genomic) proximity of genes to risk association signals mean that gene names are frequently used as primary means of denoting disease risk rather than SNP numbers. As gene names are far more memorable than SNP identification codes, it is often then a process of familiarity and repetition that can then confirm the identity of risk gene at a given locus rather than a formal scientific proof. The selection of risk genes is potentially further biased by the use of bioinformatic annotation/”by eye” selection of genes within LD/closest to risk alleles which have a known function or a function which fits in with current ideas about disease pathology. This resort to the known fails to capitalize on the agnostic strategy of GWAS. Furthermore, to add to the challenge of risk allele identification, there may be more than one causal variant at any one risk locus.

Figure 1. Flow of a typical process from initial GWAS to functional dissection [18]. (A) A GWAS study with appropriate population selection detects association between variants and disease. (B) Statistical fine-mapping and genomic annotations are applied to prioritize candidate causal variants. (C) Target genes are predicted using expression QTL data and/or enhancer-target gene promoter interaction (chromatin confirmation capture). (D) Application of experimental approaches to define the functional significance of causal variants and target genes.

Several methods have been proposed to disentangle the effect of LD from the GWAS data. In recent years, Bayesian methods [19] have been developed explicitly for fine-mapping multiple causal variants at a locus. Statistical fine mapping is much simpler to model when there is only one causal variant at a locus. These multi-variant approaches developed from earlier Bayesian approach for fine mapping [20] which assumed a single causal variant per genetic region to prioritize a credible set of putative causal variants. Along with accumulating evidence indicating the presence of multiple causal variants in a region [21,22], approaches that can jointly model multiple variants were required. Tools like CAVIAR (causal variants identification in associated regions) [23] FINEMAP [24], and SuSiE [25] have been developed to address this limitation. Yin X et al. [4] used FINEMAPv1.4 to pinpoint a single most-likely causal variant with high confidence (posterior probability ≥ 0.8) for four known (ATXN2, BACH2, DRAM1/WASHC3, and NCF2) and six novel (17p13.1, ELF3, GTF2H1, LRRK1, LOC102724596/PHB and STIM1) loci after recognizing 113 genomic regions including 46 novel loci for SLE risk from genome-wide association meta-analyses in 208,370 East Asian individuals. The usage of SuSiE was prohibited by its requirement for individual-level genotype and phenotype data. As an extension, SuSiE-RSS is available now to allow the application of summary level data to the SuSiE model [26], we believe more data on applying this tool to fine map SNPs from GWAS in SLE will emerge in the future.

Fine mapping genetic variants in trans-ancestral meta-analysis studies are challenging, but can be highly informative, as described below at the Major Histocompatibility Complex (MHC). Statistical fine-mapping approaches such as FINEMAP and CAVINAR are not suitable for trans-ancestral genome-wide meta-analysis because they do not consider the heterogeneous characteristics of multi-cohorts such as sample size, phenotyping, genotyping, or imputation as well as the heterogeneity between diverse populations in terms of the spectrum and allelic effects of causal variants. Several algorithms have been designed for fine mapping in trans-ancestral studies, including PAINTOR [27], MANTRA [28] and MR-MEGA [29] and have been applied for fine mapping genetic variants for SLE. For example, after accomplishing genome-wide trans-ancestral meta-analysis in SLE patients and healthy controls from East Asia and Europe, Wang YF et al. [11] performed fine-mapping using PAINTOR and identified 9, 1, 2, and 26 putative causal variants at ST3AGL4, MFHAS1, CSNK2A2, and CD226 loci, respectively. Along with the following functional annotations and colocalization analysis, they indicate that the rs2550368-G might be a putative causal variant of the trait and may regulate the expression of CSNK2A2 by affecting enhancer activities in B lymphocytes.

5. The role of the major histocompatibility complex (MHC) in the genetics of SLE

The MHC region on chromosome 6 contains strong association signals for SLE susceptibility and the presence of diverse autoantibodies [30–32]. There has been a long debate over the ensuing 50 years about the identity of the risk genes at the MHC in SLE. One key area of debate relates to an extended risk MHC haplotype, tagged by HLA-DRB1 × 03:01 (shortened to DR3). In the context of SLE, it is noteworthy that DR3 is in strong linkage disequilibrium with a deletion event that removes an entire C4A gene from the class III region of the MHC [33]. Gaining insight into the SLE risk alleles of the MHC requires combining the power of large-scale genetic studies across ancestry and integrating genetic data from SNPs, HLA alleles, and C4 (C4A and C4B) variants. One such comprehensive study was recently published [34]. C4 structural variants and C4A/B allotypes were imputed in European and African American case–control SNP association data. Some HLA alleles were genotyped directly, and some were imputed. Compilation of HLA and C4 alleles in the two populations revealed striking differences in haplotype frequency and composition. The two populations carried very different frequencies of the C4 allele, including the frequency of the C4B(S) haplotype, which is completely devoid of the C4A gene. Furthermore, the association of the C4 allele with the HLA allele was found to be generally much weaker in African Americans, which is not surprising given the overall lower LD in this ancestral population. These results were then used to determine whether the genetic effects of the SLE association were consistent. These analyses provide strong evidence that C4 is a shared genetic risk factor in both populations. The odds ratios for different cohort sizes were compared, but the strength of the association appeared to be well maintained, with the deletion of C4A (the C4B(S) haplotype) having the most potent effect (Figure 2a). Furthermore, it is clear that C4A deletion appears to have a greater effect on SLE sensitivity than C4B deletion. To clearly separate the effect of C4 from other possible effects of the DR3 haplotype, the odds ratios for SLE risk for C4B(S) were calculated for individuals with varying numbers of HLA-DRB1 × 03:01 alleles. This was only possible in the African American cohort, as the strong LD in the MHC hindered this analysis in Europeans. The results clearly showed that the SLE association was from C4 gene copy number and not from HLA-DRB1 × 03:01 (Figure 2b). An interesting finding for C4 is that the effect size associated with C4 loss was larger in males than in females. It has been shown that males have higher serum levels of C3 and C4 compared to females [35], the mechanism for which remains unknown but does not appear to be present at the transcript level [34].

Figure 2. Trans-ancestral analysis at the C4 locus [34]. (a) Each C4 allele associated with effect sizes of similar magnitude on SLE risk in Europeans and African Americans with loss of C4A (the C4B(S) haplotype) having the strongest effect. (b) Analysis of SLE risk across combinations of C4-B(S) and DRB1 × 03:01 genotypes indicate that on each DRB1 × 03:01 genotype background, additional C4-B(S) alleles increase risk, whereas, on each C4-B(S) background, DRB1 × 03:01 alleles have no appreciable relationship with risk.

6. From causality to function in the genetics of SLE

The integration of statistical fine-mapping with functional analysis is crucial to understand how genetic risk loci impact biological processes in complex diseases like systemic lupus erythematosus (SLE) [3,36,37]. Pathway analysis methods, taking into account non-significant signals and considering the collective effects of multiple variants, have identified various immune-related pathways associated with SLE, such as HLA regulation, immune complex clearance, and Toll-like receptor signaling [6–8].

In addition, the integration of expression quantitative loci (eQTL) studies with genome-wide association studies (GWAS) provides insights into the function of risk single-nucleotide polymorphisms (SNPs). Colocalization methods are used to determine whether the same variant is responsible for both GWAS and eQTL signals in a locus. Ota M et al. [38] built a cell-type specific eQTL landscape of 28 distinct immune cell subsets from 10 categories of immune-mediated diseases, including data from 62 SLE patients. The study is of value for lupus research because this is the first large-scale tissue-disease-specific eQTL analysis in SLE while previous studies mainly used eQTL data in healthy populations to accomplish colocalization analysis between genetic risk variants and eQTL variants. Using eQTL data in healthy controls to represent the cellular state in disease environment may cause bias because disease conditions can induce different expression patterns and different variant-gene interactions [39]. The study [38] identified 20 GWAS loci colocalize with at least one immune cell eQTL. The report also recognized the genes that had opposite regulatory functions in different immune cells when regulated by the same variant. For example, the transcriptional regulator LBH (Limb Bud-Heart) is downregulated in myeloid cells but upregulated in plasmablasts by the risk allele, and both effects showed strong colocalization with the GWAS signal [38]. PTPRC (Protein Tyrosine Phosphatase Receptor Type C) also exhibited opposing regulatory functions among plasmacytoid dendritic cells and regulatory T cells by an SLE risk allele [38]. These subset-specific and opposing eQTLs likely significantly impact immune cell orchestration and can be useful to explain the pathogenesis of SLE.

There have been technical advances in characterizing the transcriptome, which can now be studied on a single-cell basis. Ghodke-Puranik Y et al. [40] conducted expression quantitative trait locus (eQTL) analysis using qPCR in single classical (CL) and non-classical (NCL) monocytes from patients with SLE and documented multiple SLE-risk allele eQTLs in single monocytes which differ greatly between CL and NCL subsets. The greatest number of eQTLs was observed with the SPP1 (Secreted Phosphoprotein 1) and TNFAIP3 (TNF Alpha-Induced Protein 2) loci. Single-cell approaches are, not surprisingly, liable to missing data given the technical challenge of the process. That being said, the depth of data available, if sufficient cells are studied with sufficient read length, is a radical advancement. Perez RK et al. [41] conducted a single-cell RNA-seq and cell type and disease-specific eQTL analysis using more than 1.2 million peripheral blood mononuclear cells from 162 SLE patients and 99 controls. Joint analysis of cis-eQTLs and genome-wide association study identified 25 cell-type specific variant-gene pairs using COLOC.

We summarize the variant-gene pairs obtained from colocalization analysis between SLE GWAS data and eQTL data (either in healthy controls or SLE patients) published so far (Supplementary Table S2). 129 variant-gene pairs are listed from six publications [10,11,38,41–43]. The majority of studies in the field have primarily focused on European populations due to the limited availability of eQTL data in non-European samples. Because of the different genetic architecture between diverse populations, a bias will occur when we use the eQTL datasets which were derived from individuals mainly of European origin to identify risk loci identified from non-European populations. Thus, caution needs to be taken when interpreting colocalization analyses of GWAS risk loci from multiple ancestries with eQTLs identified from large public available eQTL datasets such as GTEx (https://gtexportal.org/home/) and eQTLGen [44] which are built mainly from the European population. To overcome this problem, eQTL datasets in non-European cohorts should be generated. These additional datasets would also have utility across a range of immune-related diseases.

7. Experimental validation of the findings from GWA studies in SLE

Computational fine mapping can only provide probabilistic causality and its success is contingent on contrasting patterns of LD in different ancestries. In view of this, wet-lab experimentation is important to obtain conclusive evidence of the connection between genetic risk alleles, target genes, and traits. Traditional approaches like luciferase reporter assays and CRISPR-Cas9 can only assess one genomic region at a time, resulting in limited validation of genetic variants. Massively Parallel Reporter Assays (MPRAs) is a valuable tool that allows the testing of the potential regulatory role of thousands of sequences with unique variants in a single experiment [45]. Lu et al. [46] employed MPRA to screen 3073 GWAS-linked systemic lupus erythematosus (SLE) variants at 91 loci in lymphoblastoid, GM12878 and Jurkat T cells. Fifty-one variants showing allelic dependent enhancer activity in Epstein-Barr virus-transformed B cell line GM12878. Ninety-two allele-specific enhancer variants were identified in the Jurkat T cell line, 25% of which were shared with the B cell line. This study offered insight into elucidating the transcriptional regulatory mechanisms associated with SLE and provided invaluable resource for further studies. However, it is important to note that these studies are limited in their interpretation because of the use of cell lines as the functional readout. Recent advances in MPRA technology have supported the use of ex vivo immune cells as the hosts for transfection, which further augments their biological relevance [47]. Once the likelihood of causality of non-coding regulatory variants has been determined using MPRA, the function can be investigated by using CRISPR-Cas9 methods as confirmation of effect and as a means of linking the risk allele to a causal gene.

Pooled CRISPR screens, similar to Massively Parallel Reporter Assays (MPRAs), provide a scalable approach for validating findings from GWA studies. These screens have been utilized to gain insights into non-coding functional elements. Simeonov et al. [48] used tiled CRISPR activation (CRISPRa) screening to investigate the autoimmune disease susceptibility loci associated with CD69 and IL2RA. Multiple CRISPRa-responsive elements (CaRE) were found in the CD69 and IL2RA loci, including one containing the autoimmunity risk variant rs61839660. Fulco et al. [49] developed an approach called CRISPRi-FlowFISH, which combines CRISPR interference, RNA fluorescence in situ hybridization (FISH), and flow cytometry to simultaneously perturb hundreds of noncoding elements in parallel and quantify their effects on the expression of an RNA of interest in a single experiment. They applied this method to test > 3,500 potential enhancer-gene connections for 30 genes. Using the results generated by this method, the authors then propose and test a new computational method for predicting enhancer – gene interactions, the activity-by-contact (ABC) model. Leveraging the outcomes, the authors devised a computational model called the activity-by-contact (ABC) model to predict enhancer-gene interactions, which demonstrated success in identifying enhancer-gene connections in 74 different cell types. This integrated approach of CRISPRi-FlowFISH and the ABC model presents an exciting prospect for accurate experimental characterization of endogenous enhancer–gene interactions, offering valuable insights into the functional interpretation of disease risk variants in the noncoding genome.

8. Mendelian randomization in SLE

Mendelian randomization (MR) has become more widely used in recent years to study causation in complex disease traits. This methodology can be applied to GWAS summary statistics data to estimate the causal relationship between traits and diseases: its original uses were primarily focused on risk factors in cardiovascular disease. MR requires an exposure and an outcome and uses genetic variation to provide evidence that supports or rejects the hypothesis that the exposure has a causal effect on the outcome. In MR, it is assumed that specifically selected genetic variants behave similarly to treatment assignment, that the population is divided into subgroups in a way that mimics randomization (Figure 3 [50]), and that the instrumental variables selected (usually SNP) should satisfy the relevance, exchangeability, and exclusion restrictions [51]. To date, MR analyses have been conducted between the SLE GWAS and many other traits, including other diseases and gene/protein expression etc.

Figure 3. The analogy between Mendelian randomization and randomized controlled trial [50].

8.1. Other diseases

Patients with SLE tend to have a higher risk of cardiovascular disease (CVD) [52], but the relationship between genetic susceptibility to SLE and CVD is unclear. Using traditional Mendelian randomization (MR) approaches, Kain J et al. [53] found a positive causal effect of SLE on CAD and identified a net positive causal estimate of SLE-associated non-HLA SNPs on CAD. Pathway analysis using SNP-to-gene mapping followed by unsupervised clustering based on protein–protein interactions (PPIs) identified biological networks composed of positive and negative causal sets of genes.

Epidemiological evidence suggests that patients with SLE are at increased risk of developing cancer, particularly hematologic diseases [54,55]. Zhang M et al. [56] verified a weak association (odds ratio = 1.0004, P = 0.0035) between SLE with a genetic predisposition and risk of lymphoma, and Gu D et al. [57] also found a weak negative causal association with prostate cancer risk (OR = 0.99, P = 0.039), and Peng H et al. [58] found a causal association between genetically predisposed SLE and increased risk of lung cancer (OR = 1.045, P = 0.0276).

Observational studies have found an increased incidence of depression in patients with SLE, with a recent estimate showing a pooled prevalence of depression of 35% in SLE patients [59]. Using the SLE GWAS, Chen J et al. [60] found that genetically predicted SLE mildly reduced the prevalence of depression (OR = 0.995, P = 0.025) and major depression (OR = 0.985, P = 0.009).

8.2. Gene/Protein expressions

A case–control study showed that patients with lupus had higher serum levels of growth and differentiation factor 15 (GDF15), a differentiated member of the TGF-β family associated with cancer malignancy, cardiovascular disease, and a range of other diseases with an inflammatory etiology [61]. Ye D et al. [62] found that genetically predicted high circulating GDF-15 levels were associated with a reduced risk of SLE (OR = 0.80, 95% CI: 0.68–0.92). MR methods have also been applied to data from methylation and gene expression quantitative trait loci (meQTL and eQTL) and plasma protein-level QTL (pQTL) studies to explore potential causal factors for SLE. Mo X et al. [63] found that DNA methylation of 15 loci and mRNA expression of 21 genes were causally associated with SLE, particularly methylation and mRNA expression of known SLE risk genes, UBE2L3 and BLK.

9. Translating genetic discoveries in SLE into practical clinical applications

9.1. Risk stratification using polygenic risk scores

In complex diseases, each genetic variant on its own has a relatively small effect on disease risk with limited predictive power; however, when considered cumulatively, they could provide meaningful risk predictions and improve risk stratification. A polygenic risk score (PRS) aggregates genetic risks identified from GWA studies to compose a score to predict the risk of an individual developing a disease using his/her genotype. Chen L et al. [64] conducted a genetic risk score analysis for SLE across Chinese and European populations. Utilizing three European and two Chinese GWAS datasets and training on a dataset for one population, they found that the best performing GRS results in good predictive power with an area under the Receiver Operator Curve ROC curve (AUC) for SLE equal to 0.72 and 0.67 in two different cohorts. Wang YF et al. [12] found that an ancestry-matched risk score demonstrates 73.4% sensitivity and 65.4% specificity in individuals of Guangzhou ancestry and disease risk increased with higher PRS, with individuals in the highest PRS decile having a much higher disease risk than those in the lowest decile. In addition to providing value in the diagnosis of SLE, PGR is also shown to be useful in the differential diagnosis of SLE from other autoimmune diseases: Knevel et al. [65] developed a genetic probability tool (G-PROB) to calculate the probability of different inflammatory arthritis causing conditions (rheumatoid arthritis, systemic lupus erythematosus, spondyloarthropathy, psoriatic arthritis, and gout) for a patient using genetic risk scores. Calibration of G-probabilities with disease status was high, with regression coefficients from 0.90 to 1.08 (1.00 is ideal). G-probabilities discriminated true diagnoses across the three cohorts with pooled areas under the curve (95% CI) of 0.69 (0.67 to 0.71), 0.81 (0.76 to 0.84), and 0.84 (0.81 to 0.86), respectively. attain a reasonable diagnostic accuracy with ROC-AUC of 0.84. They further observed that 35% of the patients were misdiagnosed at the initial visit. In 77% of patients, the final diagnosis was within the top two diseases with highest G-probabilities. This demonstrated that converting genotype information before a clinical visit into an interpretable probability value could significantly improve differential diagnosis.

There is also interest in exploring the usage of PRS in predicting disease severity. A study by Reid et al. demonstrated that high GRS is associated with increased risk of organ damage, renal dysfunction and all-cause mortality [66]. Chen et al. [64] also observed that a high PRS for SLE correlates with poorer prognostic factors, such as young age-of-onset and lupus nephritis. These findings support the applications of polygenic risk scores into risk stratification and guiding decision-making in the management of SLE patients. However, caution needs to be taken with interpretation of the performance of the PRS models. Firstly, most of the PRS models were developed using data from European ancestry populations and predictive performance may decrease when applying PRS models from European ancestry to other ancestries. Privé et al. [67] examined the portability of PRS models for 245 traits trained using individuals from Northwestern European ancestry in nine different ancestry groups. The authors observed a significant reduction in the accuracy of PRS models when applied to other ancestries and the performance systematically decreased as the ancestries became genetically distant. Secondly, currently developed polygenic risk score models mostly only aggregate common risk alleles without including rare variants and copy number variants, which may be important for the development of the disease. Thus, future studies should make more efforts on developing polygenic risk score models using individuals from non-European ancestries and taking rare variants into account to overcome these challenges.

9.2. Predicting outcome of treatment

Genetic polymorphism has been shown to be able to predict the outcome of treatment in SLE patients. The study by Sun XX et al. [68] showed that rs7160651, rs10873531, and rs2298877 polymorphisms of HSP90B1 (Heat Shock Protein 90 Beta Family Member 1) gene were related to glucocorticoid response and CCCGAACATCCC haplotypes of HSP90B1 (Heat Shock Protein 90 Beta Family Member 1) gene were associated with glucocorticoid efficacy. Li S et al. [69] found that rs6500552 on TRAP1 (TNF Receptor Associated Protein 1) gene may be related to efficacy of glucocorticoid in SLE patients and the rs3794701 polymorphisms on TRAP1 gene were associated with the improvement in the role-emotional (RE) of SLE patients. In another study [70], a statistically significant difference was observed in V allele frequency of FCGR3A (Fc gamma-receptor III a) gene, 158F/V, between responder (38%) and nonresponder (16%) autoimmune disease patients (p = 0.01; odds ratio [OR] = 3.24, 95% confidence interval [CI] 1.17–11.1) to Rituximab. The drug was also shown to be more effective in V allele carriers (94%) than in homozygous FF patients (81%): p = 0.02; OR = 3.96, 95% CI 1.10–17.68. These results suggest that FCGR3A-158F/V (rs396991) gene polymorphism plays a role in the response to rituximab in autoimmune diseases. This is supported by a recent study that is conducted in 262 SLE patients by Robinson JI et al. [71]. They found that FCGR3A was associated with increased odds of British Isles Lupus Assessment Group (BILAG)-2004 major clinical response when analyzed at the genotypic level, with FCGR3A-158 V homozygotes demonstrating a 2.5-fold improved responses (p = 0.03). Carriage of FCGR3A-158 V was associated with a 1.9-fold (p = 0.02) and a 1.8-fold (p = 0.04) improvement in odds of BILAG response and BILAG MCR, respectively.

The association between genetic polymorphism and lupus nephritis (LN) treatment outcome was also studied. One study found that LN patients with GSTA1 (Glutathione S-Transferase Alpha 1 (−69C > T, rs3957356)) TT genotype have the highest risk of cyclophosphamide unresponsiveness and toxicity. LN patients with the wild genotype of GSTA1 have the greatest probability of achieving a complete renal response to IV cyclophosphamide [72]. Another study showed that the carriers of rs6697139, rs10917686, and rs10917688 alleles were located between Fc Gamma Receptor IIb (FCGR2B) and Fc Receptor Like A (FCRLA) genes and had a low response to Cyclophosphamide treatment for Lupus Nephritis [73].

9.3. Hints for novel therapeutic targets

Extensive follow-up of common and rare variants has been fruitful for discovering new therapeutic agents. A recent study showed that 66% of FDA-approved drugs in 2021 have genetic evidence [74]. Furthermore, GWAS have identified drug-repurposing opportunities, i.e. finding and developing new uses for preexisting drugs. One study demonstrated that SLE risk genes are more likely to be interacting proteins with targets of the approved SLE drugs. The authors also found 19 SLE risk genes as targets of drugs not intended for SLE, providing a clue to repurposing existing drugs for the SLE treatment such as JAK2 inhibitors ruxolitinib and erlotinib (Tarceva) and an experiment drug momelotinib (CYT387, NCT01969838) [10].

10. Useful web portals and tools for conducting genetic research in SLE

Interactive web portals of public datasets and online analytical tools can significantly facilitate research. Here, we summarize web tools that could be useful for future genetic research in SLE.

ImmuNexUT (https://www.immunexut.org/) [38] is an immune cell Gene-regulation atlas built from the results of a large-scale gene-expression analysis with whole-genome sequence analysis consists of 28 distinct immune cell subsets from 79 healthy volunteers and 337 patients with 10 categories of immune-mediated diseases including systemic sclerosis (SSc), systemic lupus erythematosus (SLE), idiopathic inflammatory myopathy (IIM), Behçet’s disease (BD), Sjögren’s syndrome (SjS), Rheumatoid arthritis (RA), Adult-onset Still’s disease (AOSD), ANCA-associated vasculitis (AAV), Takayasu arteritis (TAK) and mixed connective tissue disease (MCTD). It is a public database that allows users to search for the expression level of a specific gene in the 28 immune cell subsets in healthy individuals and 10 immune-mediated diseases and SNPs with an eQTL effect on the gene. This is an invaluable tool for studying cell-type-context specific gene expression and eQTL and linking genetic risk loci and gene expression.

Open Targets Genetics (https://genetics.opentargets.org/) [75] is open access resource that integrates several sources of GWAS data and functional genomics data including gene expression, quantitative trait loci data, protein abundance, chromatin interaction, and conformation data from a vast variety of cell types and tissues. This enables systematic identification and prioritization of putative causal variants and genes. Searching can be done by variant, gene, or study/phenotype. It also includes the results of statistical fine-mapping and disease-molecular trait colocalization analysis across 92 tissues and cell types.

ezQTL (https://analysistools.cancer.gov/ezqtl) [76] is a user-friendly interactive web-based tool for integrative QTL (Quantitative Trait Loci) visualization and colocalization with GWAS data for individual loci. It contains several publicly available GWAS and QTL resources and supports users using their data for colocalization analysis.

LocusFocus (https://locusfocus.research.sickkids.ca/) [77] is another web-based tool for testing colocalization between GWAS data and eQTL. Users must supply GWAS summary statistics and eQTL data to apply the test.

OneK1K cohort database (https://onek1k.org/) [43] consists of results of single-cell RNA sequencing (scRNA-seq) data from 1.27 million peripheral blood mononuclear cells (PMBCs) collected from 982 healthy donors. It also includes information of 26,597 cell-type-specific independent cis – expression quantitative trait loci (eQTLs) and 990 trans-eQTLs in 14 immune cell types obtained by combining the scRNA-seq data with genotype data and colocalization results of single-cell eQTL and genome-wide association study (GWAS) loci in seven common autoimmune diseases including Systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), Crohn’s disease, Inflammatory bowel disease (IBD), multiple sclerosis (MS), ankylosing spondylitis (AS).

11. Conclusion

More risk loci for SLE have been identified by genome-wide association studies in the last 5 years. Bayesian methods and statistical fine-mapping approaches have been applied to fine map genetic variants in SLE, uncovering putative causal variants and providing insights into regulatory mechanisms. Advances in characterizing the transcriptome, particularly through single-cell RNA sequencing (scRNA-Seq), have furthered our understanding of cell types and activities within the immune system. Traditional methods like luciferase reporter assays and CRISPR-Cas9 have the limitation of examining one genomic region at a time, which restricts the validation of genetic variants. Recent advancements in Massively Parallel Reporter Assays (MPRAs) and pooled CRISPR screens have emerged as promising approaches for validating potential regulatory elements on a larger scale. By utilizing these innovative techniques in diverse cell types relevant to SLE, future research can be accelerated, providing a valuable resource for interpreting the genetic associations observed in SLE.

12. Expert opinion

Although many novel genetic risk loci have been identified in past five years and the studies on prioritizing causal variants based on eQTL recognized numerous causal variant-gene pairs. However, we are still far away from fully elucidating the pathogenesis of SLE and applying these findings clinically.

One route through which to apply SLE genetics is the polygenic risk score (PRS). This is a summation of risk alleles, weighted on the effect size, that any individual carries. We have previously shown that PRS in SLE is an informative tool and that there is an inverse correlation with age-of-inset of disease and a direct correlation with disease severity [64,74]. The PRS in SLE has been shown to have good reproducibility between cohorts with an AUC in ROC analysis [76]. Larger genetic studies and improved fine-mapping will lead to better performance of PRS as disease predictors. More precise implementation of PRS in SLE has the potential to inform disease taxonomy, which given the clinical heterogeneity in SLE is an important consideration and activity. One limitation of current PRS models is reduced performance in non-EUR populations [78]. However, the recent rise in SE Asian genetic studies has begun to address this issue, it is a matter of urgency to increase representation of those with African ancestry and also complex mixed ancestry, as in Hispanic ethnicities from the Americas.

Short-term end points are risk allele and risk gene identification. However, these stages must be seen simply as stepping stones in the full interpretation of disease genetics. Defining causal genes in SLE is relevant to another clinical application for SLE genetics relates directly to drug target get development. This suggestion is based on the observation that success through phase III clinical trials is influenced by the relationship between genetics and the drug target [79]. However, to capitalize fully on this observation, the precise detail of the causal genes and its role in disease should be known. Furthermore, our current knowledge of SLE genetics does not support a deviation away from simple models of linear risk. With larger studies and more precise risk allele identification, we will have more power to establish whether risk does deviate from linearity. Such results would have implications for disease taxonomy as well as combination strategies using targeted therapeutics.

Missing heritability remains an unsolved problem. One speculation about missing heritability from GWA studies is that rare variants that are not captured by SNP-chips could be possible contributors to disease susceptibility. Such variants can be captured by exome sequencing or even more reliably by whole-genome sequencing (WGS). WGS has the additional advantage that with appropriate depth and computational alignment it can be used to detect structural variants. Such changes have the potential to have a significant impact on function and can be challenging to detect in areas of coding homologous genes. In SLE, there are common examples of causal structural variants (complement C4 and IgG Fc gamma receptor genes) [80]. Genetic changes with larger functional impact will most likely be relevant to those with greater genetic effects on disease – namely childhood-onset lupus. This is a key area for progress to define mechanisms and gene identification even if mutations are rare and not directly relevant to many with SLE.

Genetics of complex traits is dominated by the task of defining risk genes. We know that these are relevant to outcome in SLE because of the correlation of polygenic risk scores with clinical metrics of disease severity. However, the topic of genetic factors affecting outcome, once the disease process has commenced and been neglected. This has primarily been because of the limited availability of high-quality longitudinal outcome data over years. Such data are expensive and challenging to collect due to the heterogeneity of the disease.

Spatial transcriptomics has become increasingly feasible in recent years and has the potential to further the understanding of SLE pathology in organs that are commonly affected and can be readily biopsied, such as the skin and kidney. With improving resolution, the proximity and nature of the physical proximity of cell types infiltrating affected tissues can be investigated. That being said, it is important to be reminded that effector functions are largely protein driven. Proteomic data are becoming available on larger scale. This has hitherto been a limitation. As costs fall, larger transcriptomic data (for eQTL), and proteomic data (for pQTL) in health and disease will support the identification of causal genes and the functional interpretation of genetics data.

Genetic characteristics remain constant and easily and cheaply measured in a person. They also drive a large proportion of expression and epigenetic changes. Despite our incomplete knowledge of disease genetics, we have shown that current polygenic risk scores show clinical and disease correlation. With refinement in genetics, the predictive power of these scores will improve. Given the functionality of proteins and the observation that protein expression and is more functionally removed from genetics than transcription, it is likely that greatest information likely obtained from a combination of genetic and protein estimation. The use of high throughput proteomics should yield optimal biomarkers, to combine with genetics to yield factors with utility in diagnosis, prognosis, and disease classification. Thus, more pathological relevant sets of criteria will be used.

The technology and analysis methods are changing so rapidly in this field that predicting 10 years into the future is a highly imprecise. In the next few years, high throughput proteomics will expand and become more reliable and cheaper and likely methods to estimate post-translational modifications (such as phosphorylation) will become much higher throughput and will include the capability to study both intracellular and membrane-bound proteins. This, combined with cheaper and more readily annotated whole-genome sequencing, will generate a much greater understanding of disease pathology. The challenge remains to correlate such with the marked clinical heterogeneity of SLE.

Article highlights

• Recent genome-wide association studies have found numerous novel genetic risk loci of SLE, which bring the total number of loci mapped to approximately 200. More trans-ancestry studies and the studies focusing on Asian population analysis were published which will be able to correct the European bias in findings hitherto.

• The studies on colocalizing GWAS signals with eQTLs prioritized numerous causal variant-gene pairs.

• Studies also managed to refine the eQTL mapping detection to cell type and disease (SLE) specific levels. Efficient leverage of such resources by future studies will certainly aid the exploration of the biology underlying the genetics.

• Mendelian randomization analyses provided evidence of correlation between SLE and several other traits including celiac disease, cardiovascular diseases, type 2 diabetes, and solid or hematological tumors.

• Single-cell RNA-sequencing provided valuable insight into the disease’s molecular mechanisms and raised the research standard to cell subtype level.

Declaration of interest

The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer disclosures

Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/1744666X.2023.2240959

Additional information

Funding

The study was supported by China Scholarship Council (CSC), no. 201908330377 and no.202008060031.