Gene hunting in the genomic era: Approaches to diagnostic dilemmas in patients with primary immunodeficiencies

Abstract

There are more than 180 different genetic causes of primary immunodeficiencies identified to date. Approaches for identifying causative mutations can be broadly classified into 3 strategies: (1) educated guesses based on known signaling pathways essential for immune cell development and function, (2) similarity of clinical phenotypes to mouse models, and (3) unbiased genetic approaches. Next-generation DNA sequencing permits efficient sequencing of whole genomes or exomes but also requires strategies for filtering vast amounts of data. Recent studies have identified ways to solve difficult cases, such as diseases with autosomal dominant inheritance, incomplete penetrance, or mutations in noncoding regions. This review focuses on recently identified primary immunodeficiencies to illustrate the strategies, technologies, and potential pitfalls in finding novel causes of these diseases.

Over the past 4 decades, more than 180 molecular defects causing primary immunodeficiencies (PIDs) have been discovered through advances in immunology and genetics. Because the majority of PIDs are monogenic, whole-exome sequencing (WES)/whole-genome sequencing (WGS) has expedited the discovery of pathogenic mutations, particularly when combined with classical methods of identifying genetic defects (Table I). Although there are many published examples, this review will focus on selected cases to illustrate the spectrum of approaches, which includes (1) educated guesses based on known signaling pathways essential for immune cell development and function, (2) similarity of clinical phenotypes to mouse models, and (3) unbiased genetic approaches (Table II). The second half of the review will address methods of overcoming challenges in identifying molecular causes of PIDs.

TABLE I.

Definitions

Next-generation sequencing (NGS): Sequencing techniques that simultaneously amplify hundreds of millions of template DNA fragments in parallel.

Exome: The protein-coding regions of the genome.

Whole-exome sequencing (WES): A technique that uses probes to select the exonic regions from the sample before sequencing.

Single nucleotide polymorphism (SNP): Single base variants that constitute the most common type of variation between individual genomes. SNPs can be used as powerful genetic mapping tools through linkage analysis and homozygosity mapping.

Linkage analysis: A method used to define the genetic locus associated with a disease based on the principle that the causative mutation and nearby genetic markers will be inherited together. This technique identifies a region, but not the causative mutation, associated with a disease.

Homozygosity mapping: A strategy used to define loci with clusters of homozygous SNPs, which thus have a high probability of containing the causative homozygous mutation for an autosomal recessive disease. Like linkage analysis, this technique identifies candidate regions, but not specific mutations, associated with a disease.

TABLE II.

Summary of pros and cons of strategies for gene hunting

Strategy	Pros	Cons
Educated guesses based on known signaling pathways essential for cellular development and function	Efficient Does not require significant computational resources	Unlikely to identify novel or unexpected genes as causes of PIDs Might not identify the causative gene because of phenotypic variability
Similarity of clinical phenotypes to mouse models	Might suggest a relevant gene to human disease Permits in vivo study of a mutation on immune function	Might be confounded by differences between human and mouse immunity Does not account for environmental factors that affect human immunity
Unbiased genetic approaches	Unbiased comprehensive approach for identifying novel genetic causes of PIDs	Requires significant technologic resources to generate and analyze large amounts of data Might be difficult to identify the pathogenic mutation from a large candidate list of benign variations

EDUCATED GUESS BASED ON KNOWN MOLECULAR PATHWAYS

Knowledge of signaling pathways establishes conceptual frameworks for the identification of molecular defects and their clinical consequences. When a clinical phenotype and disease inheritance pattern suggest candidate genes for a PID, targeted sequencing of these genes is the most efficient approach. This has been instrumental for identifying autosomal causes of diseases that were originally discovered as X-linked disorders. A classic example involves the discovery of the defects underlying hyper-IgM syndrome. The identification of CD40 ligand as a critical signal for class-switching set the stage for the discovery of CD40 ligand deficiency as the cause of X-linked hyper-IgM syndrome.^1–6 The subsequent discovery of CD40 deficiency as a cause of autosomal recessive hyper-IgM syndrome emerged from the understanding of the CD40–CD40 ligand interaction in B-cell differentiation and class-switching.^7,8 Similarly, the signaling pathways important for T-cell development facilitated the discovery of IL2RG, which encodes the IL-2 receptor γ chain (IL-2Rγ), as the cause of X-linked severe combined immunodeficiency (SCID).⁹ This enabled the subsequent discovery of defects underlying 2 forms of autosomal recessive SCID. Janus kinase 3 was identified as a kinase downstream of IL-2Rγ,^10–12 and the IL-7 receptor was found to be a partner of IL-2Rγ in binding the cytokine IL-7, which is essential for thymocyte development.^13,14 More recently, ectodermal dysplasia and immunodeficiency (ED-ID) was identified as an X-linked disorder in male patients with mutations in the IκB kinase γ nuclear factor κB (NF-κB) essential modifier (NEMO).¹⁵ Knowledge of the NF-κB signaling pathway was instrumental for identifying mutations in IκBα, a member of the NF-κB inhibitor family, as a novel cause of autosomal dominant ED-ID in female patients.^16–18

Known pathways can also guide a targeted sequencing approach to specific clinical phenotypes, such as chronic mucocutaneous candidiasis (CMC). CMC is a feature of patients with defects in the IL-17 pathway, such as those with signal transducer and activator of transcription 3 deficiency and a resultant lack of T_H17 cells or patients with neutralizing autoantibodies against IL-17A and IL-17F.¹⁹ This knowledge prompted the discovery of the first human mutations in the genes encoding the IL-17 receptor²⁰ and the identification of gain-of-function mutations in the signal transducer and activator of transcription 1 gene as additional causes of an impaired T_H17 response in patients with CMC.^21,22

Identifying molecular defects based on known pathways is a focused approach but has a low likelihood of identifying novel genetic causes of PIDs. Additionally, hypomorphic mutations can lead to variable and unexpected phenotypes. For example, mutations in CORO1A, which encodes the actin-binding protein coronin-1A, was initially described as a cause of SCID²³ but was subsequently associated with T-cell lymphopenia and EBV-associated lymphoproliferation.²⁴ To overcome these limitations, other methods, such as mouse models and unbiased genetic approaches, have been used to broaden the scope of gene discovery.

SIMILARITY OF CLINICAL PHENOTYPES TO MOUSE MODELS OF DISEASE

Mouse models can demonstrate the physiologic importance of a gene, enabling the identification of pathogenic mutations based on phenotypic similarities between mouse models and patients. This approach was used to identify a mutation in WIPF1,²⁵ which encodes a chaperone protein necessary for stabilizing Wiskott-Aldrich syndrome (WAS) protein,^26,27 in a female patient with immunodeficiency, eczema, and thrombocytopenia. Despite phenotypic features suggestive of WAS, her WAS gene sequence was normal, and she had additional immune defects inconsistent with WAS: absence of T-cell proliferation to anti-CD3 stimulation, defective T-cell response to IL-2, and normal platelet size. These features are found in WIP-deficient mice, and targeted Sanger sequencing of WIPF1 identified a homozygous mutation causing a premature stop codon in this patient.

Additionally, mouse models can be used to prioritize a candidate gene list, as was the case in a family with congenital asplenia. WES of 3 affected siblings yielded 32 candidate genes.²⁸ Only 1 gene, NKX2-5, encoded a transcription factor essential for mouse spleen development,²⁹ and in vitro studies demonstrated that the mutation abolished NKX2-5 function.²⁸

A major limitation of this approach is that mouse models do not always recapitulate human disease. For example, TANK-binding kinase 1 (TBK1), a serine/threonine kinase downstream of Toll-like receptor 3 (TLR3), is important for multiple antiviral and antibacterial pathways in mouse models,^30,31 suggesting that TBK1 deficiency would manifest as broad susceptibility to viral and bacterial pathogens. However, TBK1 deficiency in human subjects is a risk factor only for HSV encephalitis.³² Therefore the mouse model of TBK1 deficiency did not predict the limited scope of human disease. In another example, mice deficient in the p85α subunit of phosphoinositide 3-kinase have blocked B-cell development, abnormal platelet function, abnormal mast cell development, increased production of IL-12 by dendritic cells, and increased insulin sensitivity.^33,34 However, the patient lacking p85α had disease limited to agammaglobulinemia and absent B cells.³⁵ Although mouse models have been indispensible to our understanding of the human immune system, notable differences between the 2 species underscores the need for unbiased approaches for discovery of the causative genes.

UNBIASED GENETIC APPROACHES

When knowledge of signaling pathways and animal models do not suggest candidate genes, genetic techniques have been instrumental for identifying pathogenic mutations. Polymorphic markers are used to map a genetic defect to a specific chromosomal region. These markers include restriction fragment length polymorphisms, microsatellites, and single nucleotide polymorphisms (SNPs). SNPs are single base changes in the human genome that occur with a population frequency of at least 1%. They are widely used in genetic studies because they are ubiquitous throughout the genome,³⁶ adaptable to high-throughput platforms, and catalogued in publically available databases.^37,38 Linkage studies use SNPs flanking a pathogenic mutation to define the disease loci shared by the affected subjects (Fig 1).³⁹ Genes within the candidate loci are then sequenced to identify the causative variant.

FIG 1.

Linkage analysis in a consanguineous family. The deleterious mutation arose in the patients’ great-grandmother. It is inherited with genetic markers, such as SNPs, which define the genetic locus containing the pathogenic mutation.

Linkage analysis was used to identify homozygous mutations in LRBA, which encodes LPS-responsive beige-like anchor protein (LRBA), in patients with hypogammaglobulinemia, decreased memory B-cell numbers, and autoimmunity.⁴⁰ Linkage analysis of 5 affected subjects within 4 consanguineous families identified a candidate interval containing 81 genes. Subsequent Sanger sequencing of individual genes revealed homozygous mutations in LRBA in each patient. This approach requires large families with multiple affected members and might identify large candidate regions with many genes requiring expensive and time-consuming Sanger sequencing.

In the mid-2000s, next-generation DNA sequencing (NGS) revolutionized genetics by making it possible to sequence entire human genomes within days. NGS encompasses a variety of methods that simultaneously amplify and sequence millions of DNA fragments.⁴¹ Although this technology offers comprehensive sequencing data, it is challenging to distinguish pathogenic variants within the 3.2 billion bases present in the human genome.⁴² Sequencing can be limited to only the coding region of the genome, which is known as the exome, to further focus NGS data. Although the exome constitutes only 1% of the genome, it harbors approximately 85% of deleterious mutations.⁴³ WES identifies an average of 20,000 to 35,000 single nucleotide variants per exome, depending on the sequencing technology used and the subject’s ethnicity.^41,44 The elimination of synonymous variants, which alter DNA but not amino acid sequences, decreases the candidate gene list. For rare diseases, common variants with an allele frequency of greater than 1% in public SNP databases can be eliminated (Fig 2).^41,45–47 Bioinformatics algorithms, such as Polyphen-2, SIFT, MutationTaster, and MAPP, can be used to identify missense mutations that can result in altered protein function.^48–51 Additionally, programs, such as GERP and Phylo-P SCORE, can be used to assess sequence conservation.^52,53 Finally, candidate genes are filtered based on gene function and expression. Genes essential for immune function, such as those important for lymphocyte development, are prioritized; conversely, those unrelated to immune function, such as the genes encoding the olfactory receptors, can be excluded. This approach was used to identify compound heterozygous mutations in IL10R1 as a cause of neonatal-onset Crohn disease in a patient born to nonconsanguineous parents.⁵⁴

FIG 2.

Any genome will have approximately 4 million variants that differ from the reference genome. Of these, approximately 20,000 will be in the coding and splice site regions (CDS). Approximately 10,000 variants will be nonsynonymous. Finally, approximately 200 to 1000 of these variants will be “novel” (ie, not reported in existing SNP public databases), with the majority of variants existing in heterozygous states (author’s own unpublished data and references^45–47).

An additional filter can be applied to WES data from consanguineous families through the use of homozygous mapping, a special case of linkage analysis. Homozygosity mapping identifies the genomic regions most likely to harbor the pathogenic mutation based on the assumption that the causative mutation for an autosomal recessive disease occurs within a locus containing clusters of homozygous SNPs specific to the affected subjects and inherited from a common ancestor.^55,56 The combination of homozygosity mapping and WES is an increasingly common approach for identifying genetic defects underlying PIDs. LRBA deficiency was identified not only through linkage analysis, as described previously, but also through homozygosity mapping and WES in a consanguineous family with chronic inflammatory bowel disease and a combined immunodeficiency.⁵⁷ Homozygosity mapping of 2 patients and their healthy siblings, combined with WES of 1 patient, identified only 1 novel homozygous variant: a 2-bp deletion in LRBA. In another example 2 siblings from a consanguineous family presented with naive T-cell lymphopenia, defective T-cell activation, and epidermodysplasia verruciformis.⁵⁸ Homozygosity mapping and WES of 2 patients identified only 2 novel homozygous mutations. One of these mutations resulted in a stop codon in RHOH, which encodes an atypical Rho GTPase important for T-cell development and activation.^58,59 As a third example, homozygosity mapping and WGS of 2 consanguineous parents and 1 patient identified a missense mutation in MALT1, which encodes a cysteine protease important for NF-κB activation, as a cause of combined immunodeficiency.⁶⁰ These examples show that the combination of WES/WGS and homozygosity mapping can generate a shorter list of candidate genes than those identified by means of linkage analysis alone. Although these techniques have greatly expedited the discovery of pathogenic mutations, there are still limitations inherent in WES/WGS. The remainder of this review will highlight some of these difficulties and discuss how recent studies have addressed them.

AUTOSOMAL DOMINANT DISEASES

Autosomal dominant diseases represent a challenge because there are approximately 69% more heterozygous than homozygous variants in any given genome.⁶¹ Consequently, effective mapping of an autosomal dominant gene requires large families to narrow the candidate gene list: a pedigree consisting of multiple generations with 6 to 12 affected subjects is often necessary but not always available.³⁹ To circumvent this difficulty, investigators have sequenced affected subjects from unrelated families, as was done in the study identifying heterozygous mutations in GATA2 as a cause of monocytopenia, NK- and B-lymphocytopenia, severe infections with M avium complex (MonoMAC syndrome).⁶² Assuming that the disease is caused by mutations in a single gene, the authors hypothesized that WES of 4 unrelated patients would identify the gene containing the causative variants while eliminating the majority of nonpathogenic mutations. Only 1 gene, GATA2, which encodes a transcription factor important for stem cell maintenance, contained heterozygous, novel, and deleterious mutations shared by all 4 patients. Another group also identified GATA2 as a cause of MonoMAC syndrome after noting that a small proportion of these patients also have myelodysplasia, acute myeloid leukemia, or chronic myelomonocytic leukemia,⁶³ which have been previously associated with mutations in GATA2.⁶² Together, these cases illustrate how multiple approaches can be used to successfully identify the genetic cause of autosomal dominant PIDs.

DISEASES CAUSED BY MUTATIONS IN NONCODING REGIONS

Mutations in noncoding regions typically will not be detected by using WES, which primarily captures only exonic regions. Noncoding regions comprise approximately 99% of the genome, and the effect of a variant in a noncoding region is often indeterminate. Therefore identification of deleterious noncoding mutations typically occurs only after an excellent candidate gene is identified. Exonic mutations in UNC13D, which encodes Munc13-4, were known to cause familial hemophagocytic lymphohistiocytosis type 3, prompting the discovery of mutations in conserved intronic regions that abolish protein expression and result in familial hemophagocytic lymphohistiocytosis type 3.^64,65 Additional examples include intronic GATA2 mutations in patients with MonoMAC syndrome⁶⁶; an intronic mutation in SH2D1A, which encodes SAP, resulting in X-linked lymphoproliferative disorder⁶⁷; and a mutation in the 5′ untranslated region of NEMO as a cause of X-linked ED-ID.⁶⁸ In all these examples, identification of a causative intronic mutation was possible because a candidate gene was identified based on the clinical phenotype and the mutation severely impaired protein expression. In the future, the decreasing cost of WGS will facilitate efficient identification of intronic mutations. Recent research has been directed at improving both bioinformatic and laboratory tools for predicting the effect of intronic mutations,⁶⁹ but anticipating the effect of intronic mutations remains a nascent and challenging field.

DISEASES CAUSED BY STRUCTURAL VARIATIONS OF THE GENOME

Identifying large structural variations, such as deletions, inversions, and translocations, by using WES/WGS can be problematic for multiple reasons. It can be difficult to differentiate a bona fide deletion from a genomic interval with poor exome capture or sequencing. Because of its large size, the human genome is fragmented before NGS. Biotinylated oligonucleotide probes complementary to the exome bind to DNA fragments encoding the exome and are collected with magnetic streptavidin beads. The fragments are amplified and sequenced in parallel. Although sequencing is performed to achieve an overall average sequencing depth, or coverage of typically 30× to 150×, a fragment might be sequenced from 1 to more than 200 times because of genomic variations, such as repetitive regions or differences in guanosine-cytosine content. As a result, WES might not adequately sequence as much as 10% of genes.⁴⁴ The sequences are then aligned to the reference human genome. Fragments containing a large deletion or inversion will not align properly. Instead, the fragment must be analyzed as series of smaller subunits to identify the best-fit alignment for each subunit until the structural variations are pinpointed, a process that requires significant bioinformatics resources.⁷⁰

Yet even WGS, which avoids the step of exome capture, does not guarantee easy detection of structural variations and copy number variations, which include deletions and duplications. This was illustrated by a study of patients from 3 families with dominantly inherited cold-induced urticaria, antibody deficiency, and autoimmunity.⁷¹ Linkage analysis of 2 families identified 24 candidate genes, but WGS did not reveal novel variants in this region. PLCG2, which encodes phospholipase Cγ2, was the primary candidate because of its known role in B-cell function and signaling. Sanger sequencing of PLCG2 identified a heterozygous deletion in all the patients, which was confirmed by using retrospective reanalysis of the WGS data. It is difficult to distinguish a heterozygous deletion from an area of low coverage because of the significant variability in coverage inherent in WES.

Array comparative genomic hybridization is an alternative method of identifying copy number variations in which DNA is fragmented, fluorescently labeled, and hybridized to microarrays containing oligonucleotide probes complementary to the reference genome. Increased signal intensity corresponds to a duplication, whereas decreased or absent signal intensity would correspond to a heterozygous or homozygous deletion.⁷² This technique identified novel deletions in dedicator of cytokinesis 8 (DOCK8) in patients with autosomal recessive hyper-IgE syndrome.⁷³ Array comparative genomic hybridization has also identified large deletions in CYBB, which encodes the gp91^phox subunit of the NADPH oxidase, resulting in chronic granulomatous disease.⁷⁴ These arrays cannot identify structural variations that do not lead to changes in copy number, such as inversions and translocations. Detection of such variants remains an area of active research in the development of NGS technology.

DISEASES WITH INCOMPLETE PENETRANCE

All of the cases discussed thus far are Mendelian diseases with full penetrance. This assumption governs the filtering strategies used to eliminate candidate mutations. For example, in an autosomal recessive disease, all candidate mutations must be homozygous in the patient and heterozygous in at least 1 parent because the patient might have a de novo mutation in the second affected allele. However, the possibility of incomplete penetrance should be considered when developing a filtering strategy so that causative variants are not inadvertently eliminated. This was demonstrated in the study identifying an autosomal dominant mutation in TRIF, which encodes an adaptor protein important for TLR3 signaling, as a risk factor for herpes virus encephalitis (HSE).⁷⁵ A heterozygous missense mutation in TRIF was identified in a patient with HSE, but the patient’s mother and maternal grandfather had the same mutation and HSV-1 antibodies without a history of HSE. The authors demonstrated the deleterious effect of this mutation by showing that fibroblasts from the patient and her mother had impaired cytokine production to TLR3 stimulation, which was restored by transfection with wild-type TRIF. This study indicates the importance of functional assays in proving the effect of a variant on host defense because a strictly Mendelian filtering strategy would have excluded the causative variant. Furthermore, because only one of the 3 subjects with the mutation had HSE, this case also suggests how other modifier genes, epigenetic factors, or environmental exposures can affect the clinical phenotype. This phenomenon occurs with autosomal recessive PIDs as well. Even patients with the same homozygous mutation in RAG2 can present with different phenotypes, such as Omenn syndrome and hyper-IgM syndrome.⁷⁶ Identification of these modifier genes represents a challenging frontier for research.

CONCLUSIONS AND FUTURE DIRECTIONS

Advances in immunology and genetics have facilitated the discovery of novel defects underlying PIDs. However, there is still much progress to be made. Epigenetic modifications regulating gene expression, such as DNA methylation, histone modification, and noncoding RNAs, modulate the immune system,⁷⁷ and defects in these mechanisms might contribute to PIDs. NGS can be used to investigate the transcriptome to detect disease-causing splice variants, leading to exon skipping, alternative splicing, and alternative start and polyadenylation sites.⁷⁸ For our patients, the identification of the defects underlying PIDs enables genetic counseling and preimplantation diagnosis. Lastly, pinpointing these genetic defects is the foundation for the development of gene therapy as a cure.

Abbreviations used

CMC

Chronic mucocutaneous candidiasis

ED-ID

Ectodermal dysplasia and immunodeficiency

HSE

Herpes virus encephalitis

IL-2Rγ

IL-2 receptor γ chain

LRBA

LPS-responsive beige-like anchor protein

MonoMAC

Monocytopenia NK- and B-lymphocytopenia, severe infections with M avium complex

NF-κB

Nuclear factor κB

NGS

Next-generation DNA sequencing

PID

Primary immunodeficiency

SCID

Severe combined immunodeficiency

SNP

Single nucleotide polymorphism

TBK1

TANK-binding kinase 1

TLR3

Toll-like receptor 3

WAS

Wiskott-Aldrich syndrome

WES

Whole-exome sequencing

WGS

Whole-genome sequencing