Primary immunodeficiency diseases – genomic approaches delineate heterogeneous Mendelian disorders

Abstract

Background

Primary immunodeficiency diseases (PIDDs) are clinically and genetically heterogeneous disorders thus far associated with mutations in more than 300 genes. The clinical phenotypes derived from distinct genotypes may overlap. Genetic etiology can be a prognostic indicator of disease severity and can influence treatment decisions.

Objective

To investigate the ability of whole-exome screening methods to detect disease-causing variants in individuals with PIDDs.

Methods

Individuals with PIDDs from 278 families from 22 countries were investigated using whole-exome sequencing (WES). Computational CNV prediction pipelines and an exome-tiling chromosomal microarray were also applied to identify intragenic copy number variants (CNVs). Analytic approaches initially focused on 475 known or candidate PIDD genes, but were non-exclusive and were further tailored based upon clinical data, family history and immunophenotyping.

Results

A likely molecular diagnosis was achieved in 110 (40%) unrelated probands. Clinical diagnosis was revised in about half (60/110) and management was directly altered in nearly a quarter (26/110) of families based on the molecular findings. Twelve PIDD-causing CNVs were detected, including seven smaller than 30 Kb that would not have been detected with conventional diagnostic CNV arrays.

Conclusion

This high-throughput genomic approach enabled detection of disease-related variants in unexpected genes, permitted detection of low-grade constitutional, somatic and revertant mosaicism, and provided evidence of a mutational burden in mixed PIDD immunophenotypes.

Introduction

Primary immunodeficiency diseases (PIDDs) are clinically and genetically heterogeneous. The severe end of the clinical spectrum includes acutely life-threatening conditions. Milder disorders may present with frequent, severe, and/or unusual infections, autoimmune, autoinflammatory or lymphoproliferative phenomena, and with or without dysmorphic features.¹ Underdiagnosis and diagnostic delay can contribute to morbidity and mortality.^2-9 Identification of the molecular cause and disease mechanism(s) may enable early protective interventions and potentially targeted or curative therapy.^10-17 Additionally, in individuals with radiosensitive PIDDs, interventions such as radiographic imaging and the use of DNA-damaging radiomimetic drugs in pre-treatment conditioning must be addressed.

Mutations in more than 300 genes are thus far known to cause PIDDs. Diagnostic molecular testing is further complicated by overlapping clinical phenotypes.¹ Genetic testing for PIDD-associated gene variants using iterative Sanger sequencing of single genes can be time-consuming and costly. Furthermore, analysis of very rare disease-related genes is not always available in diagnostic laboratories. While the utility of targeted capture of a panel of known PIDD genes has been demonstrated, this method has limitations, for example the inability to detect new disease genes.^18-20 Whole-exome sequencing (WES) has the potential to provide rapid molecular diagnoses and improve diagnostic yield,^{21, 22} which is particularly useful in clinically and genetically heterogeneous disorders such as PIDD.

In an unbiased approach to disease gene discovery and molecular diagnosis in a large cohort (n=278) with a broad range of PIDD phenotypes, we assayed all known relevant genes using WES. Study participants had been evaluated previously with conventional tools, including clinically available immunological assays and genetic tests, but lacked a molecular diagnosis at the time of enrollment. WES and Sanger sequencing detected single nucleotide variants (SNVs) and small indels. Multiplex ligation-dependent probe amplification (MLPA),²³ chromosomal microarray and tailored bioinformatic analysis of WES data enabled detection of both intragenic and larger copy number variants (CNVs). Gene variant evaluation was performed according to recommended genetic guidelines, which include information concerning variant type, its relevance to phenotype, segregation, predicted impact, and the biological evidence for a definitive or potential functional effect.

Methods

Clinical samples

Study centers included the Baylor-Hopkins Center for Mendelian Genomics (BHCMG) at Baylor College of Medicine (BCM), and the Center for Human Immunobiology, Division of Immunology Allergy and Rheumatology at Texas Children's Hospital (TCH) in Houston USA, and Oslo University Hospital (OUS), Oslo, Norway. A total of 278 PIDD families from 22 countries consecutively recruited from 2010 to 2015 participated. The 278 probands ranged in age from seven weeks to 71 years with a mean of 14.7 years. The 78 probands (47 males, 31 females) tested in Oslo ranged in age from four months to 71 years with a mean of 18.4 years. At BHCMG, the 200 probands (114 males, 86 females) ranged in age from seven weeks to 60 years with a mean of 13.2 years.

Only clinically well-characterized affected individuals for whom appropriate conventional diagnostic investigations, including immunological and clinically available genetic laboratory testing, had been previously performed were included. The diagnostic assessment was performed according to the ESID Registry diagnosis criteria (http://esid.org/Working-Parties/Registry/Diagnosis-criteria),²⁴ and/or the joint AAAAI and ACAAI Practice parameter for the diagnosis and management of primary immunodeficiency.^{25, 26} Disease manifestations ranged from mild to severe. Based on the clinical and immunologic presentation at time of inclusion, probands were assigned to one of ten PIDD subgroups (Table 1): i) antibody deficiency (humoral deficiency that does not fulfill the diagnostic criteria for CVID), ii) autoimmune disease, iii) autoinflammatory disorder, iv) severe combined immunodeficiency (SCID), v) combined immunodeficiency (not SCID) and selective T cell deficiency, vi) common variable immunodeficiency (CVID), vii) defect in innate immunity including mucocutaneous candidiasis, hyper IgE syndrome, Mendelian susceptibility to mycobacterial disease, and complement deficiency, viii) lymphoproliferative disease, hemophagocytic lymphohistiocytosis (HLH) and NK cell deficiency, ix) neutrophil defect or congenital condition with bone marrow failure such as dyskeratosis congenita and Fanconi-like phenotype, anemia and thrombocytopenia x) immuno-osseous dysplasia, chromosomal disorder or other syndromal PIDD. Due to lack of a molecular diagnosis at enrollment, rather than using the molecular based IUIS classification system ²⁷, these subgroups above were employed, as they were constructed based upon the presenting phenotype and not ultimate genetic diagnosis.

Table 1.

Total number of individuals and families WES tested at the two different study centers

	Clinical phenotypesa per family	BHCMGb Houston, USA	Oslo, Norway	Total
i	Antibody deficiency	14	2	16
ii	Autoimmune disease	27	5	33
iii	Autoinflammatory disorder	7	8	15
iv	SCID	6	4	10
v	Combined or selective T cell defect	40	16	56
vi	CVID	13	7	20
vii	Defect in Innate immunity	14	7	21
viii	Lymphoproliferative or NK cell defect	53	9	62
ix	Bone marrow failure or neutrophil defect	12	17	29
x	Syndromal PIDD	14	3	17
	Sum ready WES analyzed families	200	78	278
	Total number of individuals WES analyzed	400	83	483

^aThe PIDD subgroups are based on the proband's clinical presentation pre-WES.

^bBaylor-Hopkins Center for Mendelian Genomics (BHCMG) at Baylor College of Medicine (BCM)

Whole exome sequencing

For affected individuals, WES was performed using genomic DNA extracted from whole-blood prior to hematopoietic stem cell transplantation (HSCT). WES of 78 probands (Table 1), four unaffected family members, and one additional affected family member was performed at the Department of Medical Genetics, OUS in Oslo. WES of 196 unrelated probands, 177 unaffected family members, and 27 additional affected family members was performed in Houston at the BHCMG.²⁸ In four of the 200 families tested at the BHCMG (Table 1), obligate carriers but not probands underwent WES. Nine of the samples were tested by WES at both places but were assigned to the center where the first analysis was performed, and confirmed concordance of sequencing standards.

At BCM-HGSC, exome capture was performed using the in-house developed BCM-HGSC Core design (52 Mb, Roche NimbleGen, Inc), as has been previously thoroughly described ^{21, 29, 30} In Oslo, exome capture was performed using the 50 Mb SureSelect Human All Exon kit v.5 (Agilent technologies, CA, USA). For 13 affected individuals from whom blood samples yielded a low DNA concentration (<10 ng/uL) exome capture was performed using the 45 Mb Nextera Rapid Capture Exome (Illumina, CA, USA). The three different capture methods provide similar output.^{31, 32} Sample preparation was performed according to the manufacturer's recommendations. For both locations, the captured exome was sequenced on the Illumina HiSeq 2500 platform (Illumina, CA, USA), with 100 or 125 bp paired-end reads resulting in an average coverage of 100x, and >90% of bases with >20x coverage. Details for data processing of the exome sequence is given in the Supplementary Methods. For both centers the bioinformatics pipelines were applied based on the GATK best practices and ANNOVAR.³³

Identification of potential PIDD-causing variants

Initially, rare variants were selected based on ESP, 1000 Genomes (October 2013), Exome Aggregation Consortium (ExAC; http://exac.broadinstitute.org), and in-house databases that include WES data from phenotypically well-characterized individuals. Details for variant evaluation are given in the Supplementary Methods. All affected individuals were initially screened for variants in a list of 475 known or candidate PIDD genes, designated PIDD(475), which was collated from gene lists available at Resource of Asian Primary Immunodeficiency Disease (http://rapid.rcai.riken.jp/RAPID) and International Union of Immunological Societies (http://www.iuisonline.org/iuis/),¹ and supplemented with genes known to affect telomere length or cause Fanconi anemia. If no PIDD-causing variants were identified from the PIDD(475) list, complete exome data were investigated for potential PIDD-causing variants. In the individuals who received a primary molecular diagnosis based on the PIDD(475) list, the complete exome data were also reviewed for occurrence of additional and modifying gene variants. Referring clinician(s) and molecular geneticists collaborated to assess the relevance of detected variants to the phenotype. In the exome-wide investigations, variants of interest were selected based on rarity, previously published cases with the same gene variants, and evaluation of possible genotype-phenotype correlation based on gene function, pathway, expression pattern and results from model organisms. Additional support for variants of potential impact was sought via computational prediction tools (PhyloP,³⁴ GERP,³⁵ SIFT,³⁶ PolyPhen-2,³⁷ LRT,³⁸ and MutationTaster,³⁹).

For potential disease-causing variants, the proband and available affected and unaffected family members underwent Sanger sequencing to confirm co-segregation of detected variants with disease phenotypes (see Supplementary Methods for details). More than 1,000 individual DNA samples were collected from the 278 PIDD families during the study. Altogether 400 samples underwent WES (Table 1). The remaining were collected and stored for Sanger segregation testing, and in total 367 samples were used for Sanger segregation testing from family members of the 110 probands with a likely or potential molecular diagnosis presented in this paper.

The functional impact of the detected variants was evaluated based on criteria (Table E1) adapted from the guidelines recommended by the American College of Medical Genetics and Genomics (ACMG),⁴⁰ see Supplementary Methods. The reported potential disease-causing variants were of class 5 (definitive pathological), class 4 (likely pathological) and class 3, but class 3 variants were only reported when the affected gene was consistent with the phenotype or parts of the phenotype in the affected individual. When possible, further supporting data were generated through functional studies, such as cDNA sequencing to confirm aberrant splicing using RNA extracted from PaxGene blood samples. The reported potential disease-causing variants were not present in homozygous state, and allele frequency was less than 0.0001 in heterozygous/hemizygous state (for dominant/X-linked inheritance respectively) in the ExAC database (as of March 2016), with the exception of modifying variants with known functional effect. Some candidate genes belonged to an established potential PIDD gene list.⁴¹ Additional lines of evidence included biological validation consistent with known clinically acceptable measures of a disease gene's function or at least in part with guidance proposed for unique discoveries in single cases,⁴² presence of PIDD gene protein homologues with relevant expression patterns belonging to the PIDD gene interactome, and segregation with identical immunophenotype in several individuals within the same family and/or other families. Given the residual uncertainty, we refer to the potentially causal linkage to disease as “likely”. For otherwise previously unreported genes and pending complete biological validation,^42,41 these genes are referred to in this paper as “potentially novel” in order to allow for the consideration of overall diagnostic yield appropriate fulfillment of established criteria.

Copy number variants (CNVs)

Two methods were used to identify CNVs: computational CNV prediction and microarray CNV detection. CNV prediction was initially performed for the probands. Microarray CNV detection was applied in selected cases where no disease-causing variants were detected after exome sequencing or when CNV prediction data indicated the presence of a relevant CNV (see Supplementary Methods for details). One third of probands had been evaluated via a diagnostic CNV microarray prior to inclusion via 180K aCGH (Agilent technologies, CA, USA) or SNP Array 6.0 (Affymetrix, CA, USA), which did not provide a definitive diagnosis by itself.

Results

In 278 PIDD families who underwent whole exome sequencing, a likely molecular diagnosis that explains all or part of the phenotype (PIDD and non-PIDD) was established in 110 (40%) (Figure 1;Table 2). Diagnostic yield varied by PIDD subgroup (Figure 2). The highest diagnostic rate was in severe combined immunodeficiency (100%), followed by bone marrow failure (55%) and syndromal PIDD (53%). The lowest diagnostic rate was in the autoinflammatory subgroup (13%), even when transient disease was excluded (18%).

Figure 1.

Status of molecular diagnosis after WES of 278 PIDD families. Findings are grouped in categories based on whether the detected disease causing variant(s) a) affects known disease genes, b) are located in more than one gene contributing to the assumed mixed phenotype (PIDD or non-PIDD) in the individual, c) affects a confirmed novel disease gene d) affects a potentially novel PIDD gene, e) leads to immunophenotypic expansion: a modified immune phenotype observed in the affected individual, modified from that which is historically characteristic of the particular gene, f) affects a phenotypic relevant gene, that were not presumably the major disease etiology, or only one deleterious variant in a disease relevant AR gene (in total 57 cases), or g) no identifiable molecular explanation was identified using the technologies applied here (111 cases).

Table 2.

Number of WES tested family members and detection rate

Family members WES tested	Total number of families	Molecular diagnosis established	Total detection rate
One individual onlya	192	70 (36%)	36%
Proband + 1 healthy parent or other unaffected relativesb	25	7 (28%)	43% (17/40)
Proband + affected relatives	15	10 (67%)
Trio: Proband + both parents	39	17 (44%)	42% (23/55)
Trio + affected relatives	6	2 (33%)
Trio + unaffected relatives	10	4 (40%)
Sum all family WES studies	287	110	40%

^aIncluding mother of proband 90.1 (Table E1)

^bIncluding both parents of proband 95.1 (Table E1)

Figure 2.

Diagnostic yield in the PIDD cohort (n=278 families) by subgroup distribution based on the proband's diagnosis prior to WES+CNV testing.

Genetic spectrum of disease-associated variants

In the 110 probands, 148 different disease-associated, potentially causative or contributing variants were identified in 88 genes (Figure 3 and Table E1). Forty-six variants had previously been reported as disease-causing in HGMD, and another six were located in the same nucleotide position or induced a change in the same codon as a previously reported HGMD variant (Table E1). In addition, 36 other variants met criteria for ACMG class 5 (pathogenic), and 10 for class 4 (likely pathogenic)., and the rest were classified as variants of unknown significance (VUS), class 3. Altogether 98 variants (66%) were either reported before and/or likely or certain disease-causing. The 148 detected variants included 136 single nucleotide variants (SNVs): 95 missense, 13 frameshift, 11 nonsense, 14 splice variants, two indels and one in-frame deletion (Figure 4). Many of these variants likely explain the PIDD immunophenotype and some may also explain non-PIDD features of the clinical phenotype.

Figure 3.

Distribution of main likely disease genes in 111 families with established molecular diagnosis after WES+CNV testing by PIDD subgroup prior to testing.

Figure 4.

Disease-associated variants and inheritance patterns observed among the 110 families in the PIDD cohort where a likely molecular diagnosis was established. (A) Spectrum of disease-associated variants detected in the PIDD cohort. In total 148 different variants were identified among the 110 families where a likely molecular diagnosis was established. The SNVs are classified based on assumed effect on protein. The indel led to a frameshift and stop codon and the in-frame deletion is previously reported to be disease causing by altered protein function. With exception of the DKC1 duplication, all 12 CNVs are deletions. All CNVs are assumed to result in loss of function. (B) Inheritance patterns observed among the 110 families in the PIDD cohort where a likely molecular diagnosis was established. The mosaic variant in KRAS is only compatible with life in somatic mosaic state, thus a Mendelian inheritance pattern is not applicable.

PIDD-causing CNVs were detected in 12 of 110 families (11%). The CNVs constituted 12 (8%) of the 148 likely disease-causing variants (Figure 4) and involved 11 genes: DKC1 (MIM: 305000), DOCK8 (twice) (MIM: 243700), FANCA (MIM: 227650), IKZF1 (MIM: 603023), IL7R (MIM: 608971), MAGT1 (MIM: 300853), MYB (MIM: 189990), NCF1 (MIM: 233700), PGM3 (MIM: 615816), SMARCAL1 (MIM: 242900) and TERC (MIM: 127550) (Table 3). Homozygosity and hemizygosity for CNVs as well as compound heterozygosity for a CNV and a SNV were observed in 12 cases (Table 3), three of which have been published previously.^43-45 Disease-causing CNVs ranged in size from 1.6 Kb to 3.4 Mb, averaging 0.45 Mb with a median of 19.7 Kb. Seven CNVs (58%) were less than 30 Kb in size and would not have been detected by standard diagnostic microarrays (Table 3). The WES data were screened for CNVs by computational prediction.⁴³

Table 3.

Disease causing copy number variants detected in the PIDD cohort

Gender	Age	Diagnosis pre-WES (PIDD subgroup)	PIDD gene involved	Inheritance pattern for disease	Mutation type	Size of CNV	State	Affected family members
Male	13y	Hoyeraal Hreidarssons syndrome, X linked (ix)	DCK1	XL	Duplication	14 kb	Hemi	3
Male	NA	Combined immunodeficiency (v)	DOCK8	AR	Deletion	355 kb	Hom	2
Female	5y	HLH and NK cell defect (viii)	DOCK8	AR	Deletion	84 kb	Hom	1
Female	34y	Fanconi anemia, mild (ix)	FANCA	AR	Deletion	22-24.6 kb	Het	2
Male	30y	Agammaglobulinemia (i)	IKZF1	AD	Deletion	15 kb	Het	1
Female	2y	SCID, later debut (iv)	IL7R	AR	Deletion	224 kb	Het	1
Male	16y	Immunodeficiency, X linked, extensive warts (v)	MAGT1	XL	Deletion	16 kb	Hemi	2
Female	8y	Immunodeficiency, progressive bone marrow failure. Short stature, dysmorphic facial features. (x)	MYB	AD	Deletion	3.4 Mb	Het	1
Female	71y	Chronic Granulomatous Disease (ix)	NCF1	AR	Deletion	15 kb	Hom	1
Male	6y	T and B cell deficiency and neutropenia (iv)	PGM3	AR	Deletion	1.24 Mb	Het	3
Female	4y	Immunosseous dysplasia (x)	SMARCAL1	AR	Deletion	1.6 kb	Hom	1
Male	49y	Dyskeratosis Congenita, progressive bone marrow failure, Short telomere lengths (ix)	TERC	AD	Deletion	3 kb	Het	1

CNVs predicted as potentially disease-associated were confirmed either by an exon-tiling CNV microarray, MLPA or PCR-based breakpoint analysis. Disease-causing CNVs in PGM3,⁴⁵ FANCA, SMARCAL1, NCF1, TERC, MAGT1, and IKZF1 ⁴⁶ were detected by computational prediction (Figures E1-E3) and in IL7R, MYB and DKC1 by exon-wise CMA (Figure E1). The two homozygous DOCK8 deletions were first noted on a SNP array used for WES quality control and subsequently visualized by HMZDelFinder (Figure E4). The CNVs in NCF1 ⁴⁷ and DKC1 were confirmed by MLPA.

The DKC1 duplication was detected by CMA in samples from the family reported in 1970 by Hoyerall et al,⁴⁸ the first report of dyskeratosis congenita, subsequently designated Hoyeraal-Hreidarsson syndrome (MIM: 305000). Our bioinformatic CNV tools were unable to detect this duplication of this X-linked gene, even in retrospect. PCR located the duplication within the gene. Although it is unclear how the duplication affects gene function, the detected CNV is the likely cause of the disease in the family because DKC1 is the only X-linked gene known to cause the specific phenotype. All three female carriers had a 100% skewed X-chromosome inactivation pattern. cDNA sequencing of one carrier failed to identify an aberrant DKC1 product, in keeping with inactivation of the mutant allele.

Mendelian patterns in PIDD

Among the hereditability of variants observed (n = 126), autosomal recessive (AR) inheritance was most frequent (n = 57, 45%), followed by autosomal dominant (AD)(n = 49, 38%, excluding one case of somatic mosaicism,) and X-linked inheritance (n = 20, 16%; Figure 4, Table E1).¹ Where parental samples were available, the likely disease-causing variants with dominant inheritance were de novo in fourteen probands (Table E1). This corresponds to a de novo mutation rate for autosomal dominant mutations of at least 29%. Reduced penetrance or parental mosaicism was present in nine families. Two variants in X-linked genes occurred de novo in affected males (Table E1). In four families obligate carriers were sequenced because affected probands were deceased or had undergone HSCT (families 90, 93, 95, and 111, Table E1). A molecular diagnosis was established in three of these families, with variants in DKC1, FANCC (MIM: 227645) or IL2RG (MIM: 300400). In a single family, two individuals suspected to have identical disease in fact had different PIDDs: one caused by a homozygous SNV in RAG1 (individual 81.5)(MIM: 601457), and the other by a homozygous CNV in DOCK8 (proband 81.1). In another family two affected cousins with neutropenia and a congenital heart defect were suspected to have identical mutations but in fact did not (family 91, Table E1 and Figure E5).

A potential mutational burden

More than one gene may contribute to the PIDD phenotype in a given family. The presence of a dual molecular diagnosis is considered likely when diagnostic criteria for both genes can be fulfilled by the single patient. In our cohort, this was observed in 12 (11%) of families with a likely molecular diagnosis. For example, in one family three deceased males (proband 93.1, Table E1, Figure E5) had immunodeficiency, leukemia, lymphoma, anemia, and solid tumors. The affected individuals shared a SH2D1A variant c.80G>A, p.Gly27Asp (NM_002351) causing X-linked lymphoproliferative syndrome (MIM: 308240). An additional variant (c.989T>C, p.Ile330Thr (NM_001018113) in FANCB (MIM: 300515), the gene for X-linked Fanconi anemia that maps 2.5 Mb from SH2D1A, co-segregated and may have contributed to the complex phenotype (Figure E5). As another example, in another family, an adult female (proband 27.1, Table E1) presented with combined immunodeficiency, low absolute lymphocyte counts, very low levels of CD8⁺ T cells, poor phytohemagglutinin (PHA) mitogen proliferative response, disseminated HPV infection, recurrent respiratory infections, vaccine-induced varicella, and cryptococcal meningitis. WES identified compound heterozygous disease-causing variants in ZAP70 (p.Gly245Arg in the SH2 domain and p.Pro502Leu in the kinase domain). Hypomorphic variants in ZAP70 have been reported in individuals with a later onset immunodeficiency,⁴⁹ and may explain this proband's clinical presentation, low CD8⁺ T cell counts, and combined immunodeficiency. In addition, she harbored a homozygous missense variant in RNF168, the gene associated with another PIDD, RIDDLE syndrome (MIM: 611943). Radiation sensitivity analyses performed by colony survival assay (CSA)⁵⁰ of the proband's EBV transformed lymphoblasts demonstrated a reduced survival fraction (8%) and exhibited increased radiosensitivity, a hallmark of RIDDLE syndrome and to our knowledge not observed in ZAP70-deficient individuals. Of interest, the patient also developed HPV-associated extensive scalp squamous cell cancer. All variants co-segregated with the disease in the family (Figure E5). Impaired function of both proteins may underlie the blended phenotype.^{21, 22} In 12 of the 110 families in which the proband's disorder was attributed to deleterious variants in a disease gene, variants in additional genes were considered having a potentially modifying effect (Figure 1, Figure E5, and Table E1 – marked with footnote ^fp).

Clan genomics and founder mutations

In families with known consanguinity or from areas where consanguineous marriages are common, disease-causing homozygous variants were observed as expected (Table E1; 5.1, 30.1, 31.1, 40.1, 64.1, 81.1 and 81.5, 85.1, 91.4, 98.1, 101.1, 108.1, 110.1, 114.1, 122.1, 123.1 and 124.1). As an unexpected finding, we were also able to use WES to detect genomic regions with region of absence of heterozygosity (AOH) in individuals from presumed outbred populations, potentially reflecting distant parental kinship (Table E1; 33.1, 33.4, 39.1, 56.1, 87.1, 100.1 and 105.1), consistent with the Clan Genomics hypothesis.⁵¹

Variable expressivity

Reduced penetrance or variable expressivity was observed in families for which probands (Table E1; 10.1, 11.1, 14.1, 15.1, 20.1, 47.1, 79.1) had a likely autosomal dominant disease-causing variant in CTLA4 (MIM: 123890), PIK3CD (MIM: 602839), NFKB2 (MIM: 164012), FASLG (MIM: 134638), and COPA (MIM: 601924). Some heterozygous mutation carriers in these families appeared clinically unaffected.⁵² For example, proband 20.1 presented with neutropenia, enteropathy, and joint disease, and harbored a missense mutation in FASLG, causing autoimmune lymphoproliferative syndrome (ALPS). Sanger sequencing confirmed that the variant was inherited from the father, who had only minor symptoms and laboratory signs of disease, suggesting reduced expressivity or lack of environmental triggering. Splicing variants were also identified that may be incomplete and cause milder phenotypes (Figure E6). Milder disease could as well be due to low-grade mosaicism, such as in the PIK3CD family (proband 67.1) with the ratios of variant to total reads 37%, 54%, and 42% in the three affected siblings but only 15% in their mildly affected father. (Table E1; 67.1).

Mendelian exceptions - somatic and revertant mosaicism

In three families, affected individuals displayed a milder phenotype and had findings suggesting somatic revertant mosaicism. Their less severe phenotypes could be related to a reduced number of cells harboring mutant alleles (Figure E7). In the first family, a six month old boy (proband 59.1) presented with eczema, diarrhea, BCG vaccination associated infection, recurrent severe staphylococcal infections and eosinophilia but normal T, B and NK cell numbers. A previously reported X-linked disease-causing IKBKG mutation (c.1167dupC, p.Glu390Argfs*5(NM_001099857)) was identified; his unaffected mother was a heterozygous carrier (69/284 reads by WES). Interestingly, the IKBKG mutant allele was observed in only 35% (55/159) of reads in the boy's DNA extracted from blood (Table E1). This finding argues for revertant mosaicism, which could explain his relatively mild phenotype compared to classic IKBKG associated immunodeficiency (MIM: 300291) in males. Revertant mosaicism and milder phenotypes were also detected in two other families with disease-causing variants in FANCA (Figure E7) and IL7R⁴⁴.

Female subject 65.1 (Table E1) serves as an example of somatic mosaicism with severe clinical consequences. She presented with recurrent infections in early childhood and subsequently developed lymphadenopathy, hepatosplenomegaly, and ALPS. She died at 4 years of age with a clinical phenotype reminiscent of HLH. WES identified a de novo variant in exon 2 of KRAS in 15/110 reads (13.6%) (c.37G>T, p.Gly13Cys (NM_004985)). Mosaicism in blood was confirmed by Sanger sequencing (Figure E7). The child had no dysmorphic features and a normal head circumference/stature. The variant was likely confined to her hematopoietic cells. Mosaicism for the identical variant was previously reported in a child with ALPS-like disease.⁵³ The phenotype is now designated RAS-associated autoimmune leukoproliferative disorder (RALD) (MIM: 614470).

Unexpected genomic and correlated clinical findings

We included affected individuals from all subgroups of PIDD through an unbiased genomic approach. In addition to using the PIDD(475) gene list we further examined other disease-related OMIM genes and all detected variants in other genes in an effort to identify new PIDD genes. For 90 out of 278 probands, variants in disease-causing genes with known, new or expanded phenotypes were observed (Table E1).^{30, 54, 55} Expansion of the immunophenotypic spectrum was associated with nine OMIM genes (SP110 (MIM: 235550), CBL (MIM: 613563), MBTPS2 (MIM: 308205), CORO1A (MIM: 615401), STAT3 (MIM: 147060, 615952), IRF8 (MIM: 614894), KRAS, MYB, and TGFB1 (MIM: 131300)) (Figure 3). Unexpected rare congenital syndromes were also detected in our cohort. In six families, variants were detected in OMIM genes which have not previously been reported in association with immunodeficiency or bone marrow failure. These variants may or may not explain the immunological defect, or perhaps only parts of the affected individual's phenotype. This could be the case for example in proband (113.1) with pathological CFTR variants and recurrent pulmonary infections. Some cases may represent multi-disease burden, others true phenotypic expansions. Biological studies of relevant gene products in the immune system are in progress in such families. The insights derived from immunophenotypic expansion can be substantial and may relate to the specifics of individual mutations or broadened clinical presentations.

For example, proband 115.1 (Table E1) presented with a SCID-like immunophenotype shortly after birth, as well as microcephaly, severe scaling erythrodermatitis, conjunctivitis, keratitis, and oligoclonal T cells – all features of Omenn syndrome – but with a normal IgE level and eosinophil count. He was ultimately found to have ichthyosis follicularis, alopecia and photophobia (IFAP) syndrome, due to alteration of MBTPS2, which has been previously associated with immunodeficiency, albeit not of this severity (Figure E8).^{56, 57} Other unexpected variants, which explained all or parts of the probands’ phenotypes, were identified in RET (MIM: 142623), PLXNA1 (MIM: 601055), ATP6V0A2 (MIM: 611716), FBN1 (MIM: 154700), MX2 (MIM: 147890), CFTR (MIM: 219700), CCDC40 (MIM: 613808), and CLPB (MIM: 16271). None of these variants would have been identified if the samples were analyzed with a gene panel restricted to targeted capture of known PIDD genes only.

Change in diagnosis and management after WES

In 55% (60/110) of the families in whom a likely molecular diagnosis was identified by WES, the prior PIDD diagnosis based on clinical features and immunophenotyping was altered (Figure 5). Post-WES revision of diagnosis was observed across all PIDD subgroups. A likely etiological diagnosis was established in approximately half (range 18% to 100%) of the affected individuals in each PIDD subgroup prior to WES. In 25% of cases (26 families, 32 affected individuals), WES findings directed a significant change in management (Figure 5). Ten individuals underwent hematopoietic stem cell transplantation (HSCT) primarily guided by WES findings. In four additional individuals, WES results supported a previous decision to treat with HSCT. Six probands had already undergone HSCT. For one of them, proband 115.1 with a molecular diagnosis of MBTPS2-IFAP, HSCT might not have been performed if the molecular diagnosis had been known.

Figure 5.

Alteration of specific diagnosis and change in management due to molecular findings by WES+CNV testing. The distribution is shown by PIDD subgroup which was based on clinical presentation at time of inclusion.

Recent and new disease genes

Twenty-five percent of molecular findings were in genes reported as disease-causing during the last five years; five in 2011, four in 2012, seven in 2013, three in 2014 and three in 2015 (Table E1). Eight were new disease genes (CCDC40,⁵⁸ CLPB,^{59, 60} COPA,⁵² CTLA4,^{61, 62} IKZF1,⁴⁶ MX2,^{63, 64} NFKB2,^65-67 PGM3,^{45, 68, 69} at the time of detection and were either confirmed biologically by us or others. Six likely candidate PIDD genes were also identified in this study (listed as potentially novel; Figure 3). These are rare, damaging, conserved variants that fit biologically with the immunophenotype, but full biological confirmation to comply with proposed criteria ⁴² remains presently incomplete. This group of patients is included so as not to underestimate the potential impact of WES as applied to PIDD. One example of a biologically confirmed new disease gene identified by WES comes from two families in our cohort with autosomal recessive variants in PGM3. Functional studies detected reduced specific phosphoglucomutase enzymatic activity.⁴⁵ Additional PIDD affected individuals with PGM3 mutations were reported simultaneously,^{68, 69} and more than 10 families have been published since.^{19, 45, 68-70} New and potentially novel PIDD genes were found in 21 families in our cohort (19%), and biological investigations to confirm these associations according to published standards ⁴² have been performed or are in progress.

In general, and as would be expected, every exome tested contained several rare, exonic variants (MAF < 0.001 in 1000 Genomes data, in-house databases, ESP and ExAC) in PIDD genes and HGMD-reported variants with MAF <0.05. With the 148 likely causative variants excluded, the majority of the other rare variants were definitively not disease-causing, but some represented single variants in AR trait genes where disease genes were compatible with the affected individual's phenotype. These are included in Figure 1 among the 20% families with findings in a phenotypically relevant gene, but findings are considered inconclusive and molecular diagnoses classified as unsolved.

Discussion

Using genomic approaches, we identified the potential molecular basis for disease in 40% of 278 probands with various PIDD phenotypes of unknown etiology who had previously been investigated with conventional methods (albeit with some variation across the 24 different hospital centers contributing patients to this study). We found evidence for a mutational burden effect,⁷¹ which might contribute to disease variability and underlie blended immunophenotypes, as well as somatic and revertant mosaicism. With no preset bioinformatic cut-off level for variant reads, we were able to detect mosaicism below 15% (KRAS, Figure E7). Several disorders with significant nonimmunological features, such as primarily skin or intestinal diseases, were diagnosed. Some cases might represent the phenotypic expansion of a known disorder to include immunodeficiency. For autosomal dominant (AD) traits, we noted reduced penetrance associated with variant alleles. For autosomal recessive (AR) traits, we identified founder variants in outbred populations. Our approach resulted in the expansion of the phenotypic spectrum associated with nine previously known disease genes (Figure 3), and the identification of eight new PIDD genes previously reported by us (PGM3,⁴⁵ COPA,⁵² IKZF1,⁴⁶) or others (CTLA4,^{61, 62} NFKB2,^65-67 CLPB,^{59, 60}), new non-PIDD genes (CCDC40,⁵⁸ MX2,^{63, 64})reported by others and detected in our cohort as well, and six potentially novel candidate genes. All of these new disease genes and potential novel candidate genes affected 21 probands (Figure 2, Table E1). We found PIDD relevant variants in 88 different genes, 25% of which were first reported during the last five years (Table E1). Twelve probands had presumably phenotypically relevant sequence alterations in more than one gene, compatible with a blended phenotype in 11%. As this has not previously been reported for major PIDD cohort studies, variations in PIDD phenotypes may need to be revisited. Importantly, molecular findings resulted in a revised diagnosis in 60 families and altered management in 32 cases (Figure 5).

In contrast to a previous report,²² we found an abundance of disease-causing CNVs. Disease-causing CNVs were detected in 11% of the families (Table 3), half of which would not have been identified by conventional chromosomal arrays. A broad spectrum of likely disease-related SNV types was observed, but the paucity of indels in our cohort deviates from the data reported for selected PIDD genes in HGMD. This finding may reflect characteristics of some immunodeficiency related genes or perhaps challenges in algorithm detection of indels from WES data. However, the distribution of SNV types we report is similar to the results demonstrated using WES in other heterogeneous Mendelian cohorts.²²

X-linked disorders contribute to the male excess in population-based PIDD studies.² In our cohort with 58% molecularly undiagnosed males, many individuals had already undergone testing for X-linked PIDD genes such as BTK, WAS, IL2RG, IPEX, DKC1, TAZ and CD40L. However, 16% of the potential causal variants in our cohort were located in these and other X-linked genes, including variants in BTK (c.141+11C>T) (MIM: 300755) and IL2RG (c.924+5G>A) not detected by diagnostic Sanger sequencing. The same was true for the variant in JAK3 (c.1695C>A) (MIM: 600802) undetected by Sanger sequencing prior to WES. The indels in CTLA4 and CD40LG (MIM: 308230) were inadequately annotated on the initial WES data but visualized in IGV and correctly classified using Sanger sequencing. This emphasizes the ongoing value and importance of validation. The mosaic variants in FANCA, IL7R and KRAS had not been detected without WES, but were visualized by Sanger sequencing as small peaks on electropherograms, which could easily have been misinterpreted as artifact.

Of the 364 genes in our PIDD(475) list with known inheritance patterns, 23 (6%) are associated with X-linked, 76 (21%) with AD, 247 (68%) with AR, and with 18 (5%) AD or AR inheritance. The distribution of inheritance patterns we observed mirrors the inheritance patterns documented for known PIDD genes with slight skewing toward variants in AR genes and an increased trend towards variant discovery in AD genes. In our families with documented AD or AR inheritance, we identified disease-related variants in 38% and 45% of families, respectively. In cohorts with molecularly confirmed Mendelian disorders, including neurological and neurodevelopmental disorders associated with reduced reproductive fitness, the trend is reversed, with a 3:2 ratio of AD and AR disorders and more than 70% de novo variants in the AD genes.^{22, 72, 73} In contrast, fewer than 30% of variants were proven to be de novo amongst the AD PIDD genes identified in this study. Even after including potential de novo variants in AD PIDD genes for whom parental samples were missing (GATA2 in proband 88.1, IKZF1 in 1.1, RPL5 (MIM: 612561) in 99.1, TERC in 63.1, STAT3 in 66.1, STAT1 (MIM: 614162) in 50.1, 60.1 and 37.1, Table E1) the percentage (45%) was substantially lower than seen in neurodevelopmental disorders.

Immunologically important genes are numerous and spread throughout the genome. Caution is required in attributing causality to variants identified by WES because even rare inherited or de novo variants may not cause disease. Widespread use of WES in diagnostics and research reveals that even variants listed as deleterious in HGMD can be misclassified.⁷⁴ A case in point is the intronic BTK variant c.141+11C>T (NM_000061.2) (Table E1) detected in individuals without antibody deficiencies and reported frequently (160 hemizygotes) in healthy males in the ExAC database. Another example is the affected male in our cohort with an X-linked, maternally inherited missense variant in WAS (exon 10 c.995T>C, NM_000377) (MIM: 301000). This p.Val334Ala variant has previously been reported to be disease-related (CM072118; HGMD), but has a minor allele frequency (MAF) of 0.005. With 115 hemizygotes reported in ExAC, it seems unlikely that such a large number of affected males with classic Wiskott Aldrich syndrome or X-linked thrombocytopenia are included among the individuals in the ExAC database. Nonetheless, while the data argue that the variant may not be a primary cause of disease, we cannot exclude the possibility that the variant may have a milder or disease-modifying pathogenic effect. We did not identify any other disease-causing variants in this individual, and he is not included in the 110 families with a likely established molecular diagnosis. Every exome contained several rare, exonic variants in PIDD genes and HGMD-reported variants that were definitively not disease-causing or only one variant in an AR trait gene which could theoretically contribute to the observed clinical phenotype. This general observation underscores the importance of evaluating a genetic variant in relationship to the clinical phenotype, especially when limited biological or functional evidence for causality exists.⁷⁵ Optimal utilization of WES data in PIDD therefore requires a dynamic iterative collaboration between clinicians, laboratory immunologists, molecular geneticists and bioinformaticians. The standards for biological consideration of novel single gene discoveries, as exemplified by PIDD, have been proposed and represent an important threshold to clear when evaluating candidate genes.⁴² We have elected not to include the names of ‘potentially novel’ otherwise unreported PIDD-candidate genes in the present work in deference to these published criteria and their current status as not yet biologically confirmed. Investigations and/or collaborative evaluations of these candidates are currently in progress in order to provide the necessary mutational impact testing and biological validation.

A few factors were considered that could potentially affect discovery rates. The ages of the probands among the families who received a likely or potential molecular diagnosis from our study were similar to those who did not. Thus, clear Mendelian traits were observed across all age groups and not confined to the youngest children with PIDD (Table E1). In Houston, the overall discovery rate was 37.5% (75/200). Among the 200 families, 204 additional family members underwent WES, which may have improved the discovery rate (Table 2). Mainly singleton cases were WES tested in Oslo; however the overall discovery rate was still 44.9% (35/78). In general, across all subgroups WES testing of more than one affected individual in the same family increased the diagnostic rate, as demonstrated in Table 2 and Table E1, and a larger proportion of potentially novel genes were discovered among those families. In 21 families, more than one affected family member underwent WES as part of our study. Among all 110 families that received a likely or potential molecular diagnosis, 12 had more than one affected family member WES tested, and 5 of these had disease related variants in novel PIDD genes. Formal WES trios (proband plus both parents) may have increased the discovery rate, but not to the extent as observed in other cohorts.^76-78 Since the numbers are small (Table 2), however, definitive conclusions cannot be made. Finally, the fact that all patients were evaluated according to diagnostic criteria and conventional methods prior to inclusion, many of the more straight forward molecular diagnoses had been already made, and those patients did not contribute to this study. Thus, if all presenting PIDD patients were included, discovery rates in a PIDD cohort would have likely been higher.

Primary immunodeficiency diseases with severe, opportunistic infections represent distinct phenotypes with strong genetic components and simple Mendelian inheritance patterns, even in PIDD subgroups historically considered as complex polygenic disorders, such as CVID. ^{79, 80} Conversely, due to heterogeneity, known occurrence of genetic modifiers and susceptibility factors (e.g. MBL deficiency, (MBL2, MIM:614372)), reduced penetrance and variable expressivity between mutation carriers even within families, and strong influence by environmental factors (bacteria, viruses, fungi, nutrition status, age), the delineation of genetic causality in each individual can be challenging. Our 40% gene variant identification rate is high compared to other WES studies of various non-PIDD disease cohorts.^{21, 22} However, the rate is lower than for WES trios in cohorts with severe intellectual disability ^{76, 81} and brain abnormalities.⁸² Those conditions are characterized by highly penetrant gene variants, with skewing towards de novo deleterious variants in AD genes, and perhaps less influenced by environmental factors such as infection-triggered immunodysregulation. Further biological verification of causality will be needed for the variants classified as VUS, and despite our effort taking the genome-wide approach, additional disease-causing variants may be discovered even among the 40% with an assumed molecular diagnosis.

For the 60% of families in this cohort who still lack a molecular diagnosis (Figure 1), further genomic studies are indicated in an effort to identify variants in non-captured regions, including regulatory regions, low-grade mosaicism, small CNVs and indels. Some of these families may also have polygenic PIDD, which remains an active area of investigation. A stepwise approach for PIDD diagnostics with targeted next generation sequencing (NGS),²⁰ followed as needed by WES and/or whole-genome sequencing (WGS) has been previously proposed.¹⁹ Our findings support such an approach, although use of WES and/or WGS may ultimately become preferred to targeted multi-gene panel testing as technologies advance to provide results in a comparable timeframe. The need for rapid analysis is emphasized by the fact that acquisition of the proper diagnosis directly altered management of 25% of probands in our cohort. Targeted gene panel testing currently has advantages in terms of high coverage of genes of interest (including deep intronic regions), ease of data interpretation, and rapid turnaround time. Targeted capture of a limited number of PIDD genes can often deliver molecular diagnostic results within weeks or days, at a lower cost, while gene hunting with inadequate coverage of regions of interest may highlight pitfalls associated with currently available WES and WGS.^{18-20, 83-86} WES, however, allows for detection of variants in potentially novel gene candidates. It can uncover mutations in other genes, such as CFTR, that mimic PIDD. Our PIDD(475) gene list was an efficient means for detecting causative variants in known genes in 77 probands. We have also demonstrated the utility of WES combined with tools for intragenic CNV detection to identify small CNVs. It remains unknown whether targeted gene panels will prove more cost-effective than WES, as various laboratory-based, insurance-related, and nationally-implemented policies factor greatly influence cost analyses. WGS is not capture based, unlike targeted gene panel testing and WES, allowing for direct discovery of structural abnormalities, such as CNVs and inversions, and SNVs within nonexonic genomic regions, albeit currently at a substantially higher cost.⁸⁷

Based on our data and a review of the literature, we suggest a two-tiered approach: 1) Initial testing with a limited capture kit targeting a) common PIDD-causing genes, b) actionable gene variants, and c) genes causing severe phenotypes and requiring a rapid molecular diagnosis, such as disorders characterized by low TRECs on newborn screening for SCID ⁸⁸ and lymphoproliferative disorders.⁸⁹ Optimal probe design of a targeted capture kit to ensure sufficient coverage of the targeted genes may enable replacement of Sanger sequencing by rapid testing on a bench-top NGS platform.²⁰ Given substantial read depth-of-coverage, targeted gene test data may allow for detection of intragenic CNVs and low-grade mosaicism. 2) If no PIDD-causing variants are identified by the above testing, we suggest proceeding to a genome-wide approach using either WES or WGS with a large, frequently updated PIDD gene candidate list such as PIDD(475) that includes other extended immunologically PIDD-relevant genes as a bioinformatic filter in the initial variant screening.⁴¹ Genome-wide data can be used to identify AOH regions that can help guide the search for disease-causing variants in consanguineous families and in individuals from outbred, but homogeneous populations.

Clinical implications.

With a rate of likely molecular diagnosis in 40%, WES in tandem with CNV screening tools is effective for detecting a broad spectrum of disease-causing variants in individuals with PIDD.

Capsule summary.

Whole-exome sequencing and CNV screening underlines the genetic component of PIDDs. Revised clinical diagnosis in half and altered management in a quarter of the families emphasizes the relevance of a molecular diagnosis.

Abbrevations

aCGH

array comparative genomic hybridization

AD

Autosomal dominant

ALPS

Autoimmune lymphoproliferative syndrome

AOH

Absence of heterozygosity

AR

Autosomal recessive

CMA

Chromosomal microarray analysis

CNV

Copy number variant

CVID

Common variable immunodeficiency

HLH

Hemophagocytic lymphohistiocytosis

HPV

Human papillomavirus

HSCT

Hematopoietic stem cell transplantation

MAF

Minor allele frequency

MPLA

Multiplex ligation-dependent probe amplification

NGS

Next generation sequencing

PCR

Polymerase chain reaction

PHA

Phytohemagglutinin

PIDD

Primary immunodeficiency disease

SCID

Severe combined immunodeficiency

SCN

Severe congenital neutropenia

SNP

Single nucleotide polymorphism

SNV

Single nucleotide variant

WES

Whole-exome sequencing

WGS

Whole-genome sequencing