Whole exome sequencing in unclassified autoinflammatory diseases: more monogenic diseases in the pipeline?

Abstract

Objective

Autoinflammatory diseases (AIDs) are characterized by recurrent sterile systemic inflammation attacks. More than half of the patients remain genetically undiagnosed with next-generation sequencing panels for common AIDs. In this study, we aimed to define phenotype-genotype correlations in a cohort of unclassified AID patients via whole exome sequencing (WES).

Methods

Patients with features of AIDs were included in this study followed in the Department of Pediatric Rheumatology at Hacettepe University. They were first screened for MEFV with Sanger sequencing and then WES performed for the patients with clinically insignificant results. Pre-analysis of WES data was done by considering the 13 most common AID-related genes. Further bioinformatic analysis was performed if the patient remained genetically undiagnosed.

Results

The median age at disease onset was 1.2 years (range 0.2–16) and at the time of study recruitment was 14 years (range 3.5–17). In our cohort, WES provided a definite or probable disease-causing variant in 4 of 11 patients (36%). Heterozygous mutations for two of these genes were previously associated with neurological defects (ADAM17, TBK1), also homozygous ADAM17 mutations were observed in one family with neonatal inflammatory skin and bowel disease. Besides, two genes (LIG4, RAG1) were associated with immunodeficiency although the patients had presented with inflammatory features. Finally, for one patient, we associated a strong candidate gene (NLRC3) with autoinflammatory features.

Conclusion

WES strategy is cost-effective and provides substantial results for a selected group of undefined AID patients. Our results will contribute to the spectrum of unclassified AIDs.

Introduction

Autoinflammatory diseases (AIDs) are disorders of the innate immune system characterized by sterile systemic inflammation [1]. FMF is the most common monogenic AID with prevalences as high as 1/1073 in the eastern Mediterranean [2]. In 1997, the gene responsible for FMF,designated MEFV, was the first causative gene to be associated with an AID [3, 4]. Since then, >30 genes have been identified for the diagnosis of AIDs [5]. Although the AIDs usually have specific features, there is considerable overlap between the AIDs and there are almost no pathognomonic laboratory studies. Thus, the diagnosis is usually challenging for the caring physician and it often relies on genetic analysis where a gene panel for the common AID are sought. However, ‘gene panel analysis’ of the previously identified genes for AIDs fails to determine a causative variant in the majority of these patients. Those patients who have a clear autoinflammatory phenotype but are not associated with a known disease gene are often termed unclassified AIDs. Most of these patients may simply have a combination of variants in several genes (complex genetic trait) and thus, the genetic analysis may not end up with a certain result. However, some may have novel gene associations as well.

While next-generation sequencing (NGS) technologies developed in time, the whole exome sequencing (WES) method became a game-changer technology, allowing simultaneous analysis of multiple genes for patients with specific disease patterns. NGS panels are often used to screen a list of targeted genes for reaching the most probable genetic diagnosis, whereas WES strategy is rather used when NGS panels are not confirmative and broader screening is needed [6]. Our department is one of the main referral centres for AIDs in Turkey. In this study, we aimed to reach a causative variant with WES in AID patients who had no definitive mutations in targeted panels for the well-known AID genes.

Methods

Patient selection

Patients with recurrent attacks of fever and high CRP along with clinical features of inflammation, with a possible AID were included. Infections were excluded for all. They all had been initially tested for MEFV mutations with Sanger sequencing. The patients who did not have a causative mutation in MEFV were further selected for WES. Following WES, pre-analysis was carried out for the 13 most common AID-related genes. Further bioinformatic analysis was performed if the patient remained genetically undiagnosed with the custom-designed gene panel. We have screened 11 such AID patients with WES in this study. This study was conducted in compliance with the Declaration of Helsinki and was approved by the Ethics Board of Hacettepe University (GO 15/744–19). All patients gave written informed consent prior to their inclusion.

DNA extraction, library preparation and ion proton sequencing

Genomic DNA was extracted by standard phenol-chloroform protocol from the peripheral blood of the patients after the receipt of informed consent. Ion AmpliSeq^TM Exome RDY Kit and Ion Proton^TM sequencing platform allowed for capturing, amplification and sequencing. The 57.7 Mb target region includes 293 903 amplicons of >19 000 genes, which is >97% of the coding RefSeq exons as well as portions of flanking regions. Emulsion PCR was performed on Ion OneTouch^TM 2 instrument using the Ion PI Hi-Q OT2 200 Kit according to the manufacturer’s instructions. Enrichment of Ion Sphere Particles (ISPs) was performed using the One Touch^TM ES module. Then, ISPs were loaded on Ion PI chips and sequenced (Thermo Fisher Scientific, Waltham, MA, USA).

Whole exome sequencing

Exome sequencing was performed on 11 undefined paediatric Systemic Autoinflammatory Disease (SAID) patients. Base coverage depth was over 50× and >93% of the mapped reads were aligned over a target region in average, which is essential for detecting heterozygous germline variants. Torrent Mapping Alignment Program (TMAP) was used for aligning reads to the hg19/GRCh37 human reference genome. Raw sequence file was processed using the Torrent Suite software v5.0.5 for generating mapped reads, analysing coverage data and variant calling.

Custom-designed NGS gene panel and bioinformatic analysis

The 13 most common AID-causing genes (CECR1, FAM105B, MEFV, MVK, NLRC4, NLRP12, NLRP3, NOD2, PSMB8, PSTPIP1, TMEM173, TNFAIP3 and TNFRSF1A) were extracted and screened from WES data at first step (Fig. 1). All coding exons and exon-intron boundaries of the related genes were analysed with the custom-designed targeted NGS panel. After this preliminary analysis, WES data is further examined and filtered with various bioinformatics methods.

Variant filtering strategy

Variants with a coverage less than 5× for single nucleotide polymorphisms (SNPs) and 10× for small insertions and deletions (INDELs) were excluded for further bioinformatic analysis. Phred Quality Scores were specified as Q15 for SNPs and Q20 for INDELs. All variants passed the quality score filters were annotated with the Ion Reporter™ 5.10 software. We applied variant filtration steps and followed the strategy to select and filter the variants detected with WES (Fig. 2). We used HomSI (Homozygous Stretch Identifier) which performs homozygosity mapping from next‐generation sequencing data to identify the Runs of Homozygosity (ROH) in megabases (Mb) for index cases [7]. Variants that passed the quality check were identified for each sample, including single‐nucleotide variants and small INDELs. Common polymorphisms were discarded (Minor Allele Frequency, MAF>0.01) and exonic and nonsynonymous substitutions were selected. We also eliminated variants present in our in‐house variant database (n = 408). All previously identified Human Gene Mutation Database (HGMD) mutations were curated with Ingenuity Variant Analysis software (IVA/QIAGEN Bioinformatics). Besides, we filtered rare variants (MAF<0.01) for presence in a comprehensive list of genes known to be involved in AIDs or relevant pathways (Supplementary Table S1). The alterations that located in the boundaries of ROH (≥1.5 Mb) within all chromosomes were extracted for the families showed consanguinity. Variants present in the ‘Genome Aggregation Database Browser’ (gnomAD) as homozygous state were ruled out.

The remaining candidate variants were classified with three different variant function prediction tools (MutationTaster, PolyPhen2 and SIFT), following a manual checking step in IGV (Integrative Genomics Viewer). The alterations predicted as ‘disease causing’ and ‘damaging’ with all three prediction tools were selected. Among them, variants with a CADD (Combined Annotation Dependent Depletion) score higher than 20 were chosen. The filtered variants were ordered by REVEL (Rare Exome Variant Ensemble Learner) and VEST (Variant Effect Scoring Tool) scores [8–14]. Final candidate ensuring all the filtering criteria were defined as probable or definite disease causing variant. If the most significant variant was not associated with a human disease phenotype, but there are several supportive functional studies and evidences in the literature, we referred to these alterations as ‘novel candidate genes’. All causative and candidate variants detected with WES filtering steps were validated and familial segregation analysis was performed through Sanger sequencing.

Results

Demographics and clinical features

A total of 11 patients were included in this study. The median age at disease onset was 1.2 years (range 0.2–16 years) and 14 years (range 3.5–17 years) at the time of the present study. All of our patients were Turkish with a median disease duration of 4 years (range 1–15.8 years). All of the patients were followed in our centre for a median duration of 3 years (range 1.5–13 years).

We identified a definite or probable disease-associated variant in 4 of 11 patients (36%) with WES. One patient with neurocognitive impairment and inflammatory features was homozygous for a TBK1 nonsense mutation, previously reported in a heterozygous state, who had a partial response with anakinra treatment (Table 1). The heterozygous loss-of-function (LoF) form of p.Arg440Ter (c.1318C>T) mutation in TBK1 was shown on both familial Amyotrophic Lateral Sclerosis (fALS) and ALS-dementia patients [15, 16]. This is the first case with homozygosity of this nonsense mutation. The p.Arg440Ter mutation was not observed in both gnomAD and our in-house database, neither in homozygous nor heterozygous state. Sanger sequencing showed that the grandmother of the index case, diagnosed as having early dementia, was heterozygous for the p.Arg440Ter mutation as well as the father and mother (Supplementary Fig. S1). The parents were asymptomatic probably due to young age as both of them were younger than 40.

One other patient had a rare homozygous p.Ile284Thr (c.851T>C) missense variant in ADAM17 (NM_003183.4), which was not observed in gnomAD and our in-house database. Following WES, variant filtering ended up with six homozygous missense alterations. Between these, p.Ile284Thr variant on ADAM17 gave the highest pathogenity scores and estimated as ‘disease causing’ within six different variant prediction tools (Supplementary Table S2). This variant was further validated by Sanger sequencing. Parents and two unaffected siblings were each heterozygous for the p.Ile284Thr variant and were again younger than 40.

Two of the patients had mutations previously associated with immunodeficiency and published by our group [17, 18]. One of them had a homozygous hypomorphic p.Arg699Trp (c.2095C>T) RAG1 (NM_000448.2) mutation which causes T-B-NK+ SCID. He presented with vasculitic features such as pulmonary haemorrhage, digital necrosis and finger amputation. His skin biopsy showed necrotizing arteritis. His initial diagnosis was polyarteritis nodosa and he was treated with cyclophosphamide and high dose steroid followed with azathioprine and monthly IVIG treatment. He is now in remission, waiting for hematopoietic stem cell transplantation [17]. The other patient had a homozygous missense LIG4 (NM_206937.1) mutation, p.Arg871His (c.2612G>A), presented with aseptic meningitis and Behçet-like phenotype and was in remission with colchicine and azathioprine [18] (Table 1). Both mutations were validated by Sanger sequencing and completely segregated within family members.

One patient presented with recurrent fever attacks together with myalgia, abdominal pain, chest pain, arthralgia, and fatigue occurring every 6 months, lasting for 15–30 days. He was treated with on-demand corticosteroids and had a partial response (Table 1). Following WES, data analysis and variant filtering steps revealed a single homozygous p.Ala509Val (c.1526C>T) missense SLC27A6 (NM_014031.3) variant. This variant was absent in our in-house database as well as gnomAD. This gene has been previously defined to be associated with leukodystrophy [19].

Another patient was a heterozygous carrier for a previously reported MEFV variant (p.Ala289Val; c.866C>T) which is classified as ‘Variant of Uncertain significance/VOUS’ in the Infevers database [20]. However, he also had oral ulcers and diarrhoea during the attacks and was unresponsive to colchicine. Thus, he was not accepted as having FMF. We were not able to define a clear causative variant for this patient (Table 1).

A final patient with parental consanguinity had a rare homozygous Pp.Leu90Pro (c.269T>C) missense variant in NLRC3 (NM_178844.2) which belongs to the NOD-like receptor family (Fig. 3A). Following WES, variant filtering ended up with four homozygous missense alterations located in three different homozygous chromosome regions (Fig. 3B and C). Between these, the p.Leu90Pro (c.269T>C) variant on NLRC3 gave the highest pathogenity scores and estimated as ‘disease causing’ within six different variant prediction tools (Supplementary Table S3). The p.Leu90Pro variant was observed in gnomAD database with a low minor allele frequency (MAF: 0.00004512) – no homozygous individuals were present. Besides, this alteration was absent in our in-house WES database. Sanger sequencing showed the homozygosity for index case and heterozygosity for parents and unaffected female sibling (Fig. 3D). The homozygous rare missense variant is located in Leucine at amino acid position 90, which was found to be completely evolutionary conserved among different species (Fig. 3E). This patient presented with recurrent attacks of fever, rash, abdominal and back pain, arthralgia, oral ulcers and diarrhea,and have responded to on-demand anakinra treatment (Table 1).

No definite or probable disease-causing variants were identified for the rest of the patients.

Discussion

The scope of AIDs are broadening. AIDs are characterized by the presence of inflammation of the innate immune system. Inflammation is often concentrated in the serosal membranes and skin, but may involve any organ system including the neurological system. Fever is a common feature in most of these diseases. Thus, an autoinflammatory syndrome can be defined as a disease of the innate immune system, presenting with episodes of clinical inflammation along with raised acute phase reactants. A genetically unclassified autoinflammatory syndrome is one where the disease cannot be associated with any certain mutation in a previously defined gene. Some of these patients may not have a monogenic disease, and thus this may be the result of multiple low penetrant genetic and environmental factors. The aim of this study was to examine a group of unclassified autoinflammatory syndromes to characterize further patients and associate phenotypes with genetic variants in previously undefined genes. We present detailed features of these patients in hope of inviting more patients from the community.

When the associated genes are discussed separately, a number of novel features stand out. Two patients had mutations in genes that are associated with neurological features as well. In one of these patients (case 1) who had neurological involvement and recurrent fever attacks, WES variant filtering revealed a previously described heterozygous mutation (c.1318C>T; p.Arg440Ter) in homozygous state in TBK1 (TANK-binding kinase 1). Although we were not able to carry out functional studies, we suggest that the homozygous LoF mutation in TBK1 in this patient is associated with the inflammatory features besides neurological dysfunction.

TBK1 is a 729 amino acid long, four functional domain-containing multifaceted molecule involved in various cellular pathways, including innate immune response, inflammation, autophagy, proliferation and interferon signalling [21, 22, 23, 24] (Fig. 4A). The serine/threonine kinase domain phosphorylates the stimulator of interferon genes (STING) and interferon regulatory factor-3 (IRF3) leading to type-I interferon and pro-inflammatory cytokine induction [25]. Hammaker et al. demonstrated that TBK1 contributes to the pathogenesis of rheumatoid arthritis (RA) by regulating IRF3-mediated IFN-β and IP-10 (CXCL10) expression and proposed the TBK1 molecule as a potential therapeutic target for RA patients [26]. The homozygous p.Arg440Ter nonsense mutation in our patient is likely to cause a truncated TBK1, lacking complete CCD2 in addition to a large part of CCD1, which would mainly cause dysregulation in autophagy (Fig. 4B). In a murine study, Goncalves et al. observed that both CCD1 and CCD2 deleted TBK1 mutants were completely inactive, which confirms the crucial role of these domains for protein functioning [27]. On the other hand, heterozygous LoF mutations resulting with haploinsufficiency cause amyotrophic lateral sclerosis (ALS) and/or frontotemporal dementia (FTD) (MIM (Mendelian Inheritance in Man): #616439) [15, 16]. In fact, TBK1 has been clearly associated with neuroinflammation, data linking this protein to neurological dysfunction supports the causality for this patient having neurocognitive retardation and seizures. The grandmother did have dementia as well and the parents were young to observe FTD phenotype.

A homozygous missense variant (c.851T>C; p.Ile284Thr) was identified in ADAM17 for the other patient (case 2). Previously, a homozygous mutation in ADAM17 has been reported to cause neonatal inflammatory skin and bowel disease (NISBD1; MIM: #614328) in one family; however, this pleiotropic gene is also associated with late-onset familial Alzheimer disease in heterozygous form [28]. Blaydon et al. have described siblings with homozygous LoF ADAM17 mutations presenting with inflammatory skin, nail and bowel lesions [29]. Our patient also had severe pustular skin lesions and nail abnormalities. Biopsy of the skin revealed neutrophilic dermatosis (Supplementary Fig. S2). Thus, the skin and nail features of this patient were similar to that reported by Blaydon et al. serving as a possible second case. It is known that broad-spectrum metalloprotease inhibitors suppress soluble TNF-α levels by inhibiting ADAM17 activity. In addition, Saad et al. have shown that ADAM17 activates the IL6 trans-signalling pathway, which is another critical cytokine in inflammation [30]. We thus suggest that the homozygous p.Ile284Thr variant in ADAM17is associated with the inflammatory phenotype in our patient.

Two patients had presented with attacks of inflammation although they had mutations in genes associated with immunodeficiency. The patient with RAG1 deficiency (case 3) had presented with vasculitis and features suggesting inflammation. Although he had pulmonary haemorrhage, digital necrosis, finger amputation and a skin biopsy showing necrotizing vasculitis compatible with polyarteritis nodosa, we showed that he had RAG1 deficiency, which is classically regarded as an immunodeficiency [17].

The patient associated with LIG4 deficiency (case 4) presented with Behçet disease-like symptoms such as oral and genital ulcers, anterior uveitis, aseptic meningitis and non-erosive arthritis. He also had developmental delay, atypical facial features and dysembryoplastic neuroepithelial tumuor (DNET). He was treated with corticosteroids during the acute phase and colchicine and azathioprine as a maintenance treatment. His inflammatory symptoms subsided with these treatments [18].

We also defined a possible novel gene association with a less clear implication. We identified a homozygous missense variant (c.269T>C; p.Leu90Pro) in NLRC3 (NM_178844.2) in one of our patients (case 5) with recurrent attacks of fever, abdominal and back pain, rash, arthralgia, oral aphthosis, nausea and diarrhoea who responded to anti-IL1 treatment. NLRC3 encodes a NOD-like receptor family protein that directly interacts and inhibits STING protein function. Actually, both TBK1 and NLRC3 bind to STING and form a complex that controls secretion of type-I interferon and proinflammatory cytokines [31]. It has been shown that HSV-1-infected Nlrc3^-/- mice had enhanced interferon and cytokine production and NLRC3 deprivation results with elevated NF-κB activation and increased pro-inflammatory cytokine production via lipopolysaccharide (LPS) stimulation [32]. It is tempting to speculate that NLRC3 variants have pleiotropic effects resulting in distinct immunologic and inflammatory phenotypes depending on variant type, localization and zygosity.

Another patient (case 6) had recurrent fever attacks accompanied by myalgia, abdominal pain, chest pain, arthralgia and fatigue every six months lasting 2–4 weeks and responding to on-demand corticosteroid treatment. He had the homozygous SLC27A6 variant, encoding a very long chain acyl-coA synthetase. Heterozygous mutation of this gene has been shown in a patient with leukodystrophy; however, we could not define any neurological feature in our patient [19].

Similar attempts to define novel genes in AIDs lacking mutations in the defined genes often have lower yields (4%) to identify a definite or probable disease-causing variant [33]. We mainly selected patients who were offspring of consanguineous marriages (except two), which probably explains our high yield of 36%. An important handicap of our study was the lack of functional studiesfor the defined genes. We suggest that these genes are associated with the phenotype of these patients; thus, we hope that further studies will enlighten the underlying mechanisms.

In conclusion, we present the genetic analysis results for 11 patients with unclassified autoinflammatory features. We were able to define causative or probable variants in4 of the 11 patients. While this is a high success rate compared with previous studies on genetic aetiologies of SAID patients, we reached this rate in our clinic department by choosing a selected population of undiagnosed SAID patients, high parental consanguinity rate and pre-screening of MEFV. We believe sharing the results may enable the identification of novel patients and possibly lead to further studies for personalized treatment.

Funding: This work was supported by the INSAID Project ‘A comprehensive clinical and experimental approach to personalized molecular medicine in patients with defined and undefined autoinflammatory disorders’ (E-rare 2015, 15-pp-156), TUBITAK Project (315S096). Also, we acknowledge Hacettepe Exome Project by using the in-house database (Grant Number: TAY-2015–7335).

Disclosure statement: The authors have declared no conflicts of interest.

Supplementary data

Supplementary data are available at Rheumatology online.

References

Author notes