Dominant atopy risk mutations identified by mouse forward genetic analysis

Abstract

Background:

Atopy, the overall tendency to become sensitized to an allergen, is heritable but seldom ascribed to mutations within specific genes. Atopic individuals develop abnormally elevated IgE responses to immunization with potential allergens. To gain insight into the genetic causes of atopy, we carried out a forward genetic screen for atopy in mice.

Methods:

We screened mice carrying homozygous and heterozygous N-ethyl-N-nitrosourea (ENU)-induced germline mutations for aberrant antigen-specific IgE and IgG1 production in response to immunization with the model allergen papain. Candidate genes were validated by independent gene mutation.

Results:

Of 31 candidate genes selected for investigation, the effects of mutations in 23 genes on papain-specific IgE or IgG1 were verified. Among the 20 verified genes influencing the IgE response, 8 were necessary for the response, while 12 repressed IgE. Nine genes were not previously implicated in the IgE response. Fifteen genes encoded proteins contributing to IgE class switch recombination or B cell receptor signaling. The precise roles of the five remaining genes (Flcn, Map1lc3b, Me2, Prkd2, and Scarb2) remain to be determined. Loss-of-function mutations in nine of the 12 genes limiting the IgE response were dominant or semi-dominant for the IgE phenotype but did not cause immunodeficiency in the heterozygous state. Using damaging allele frequencies for the corresponding human genes and in silico simulations (Monte Carlo) of undiscovered atopy mutations, we estimated the percentage of humans with heterozygous atopy risk mutations.

Conclusions:

Up to 37% of individuals may be heterozygous carriers for at least one dominant atopy risk mutation.

Introduction

Immunoglobulin E (IgE) antibodies contribute to protective immunity, particularly against parasites, but in atopic individuals, high levels of IgE mediate allergic reactions to otherwise innocuous antigens. Concentrations of IgE antibodies in circulation are much lower than those of other immunoglobulin classes, but become elevated in individuals with allergic disease. Several mechanisms limit circulating IgE concentration and the proportion of IgE-expressing B lymphoid cells. These include negative regulation of IgE class switch recombination (CSR) (1–3), the short intrinsic half-life of IgE (4); and capture or clearance of IgE from the serum by mouse FcεRII and human FcεRI, respectively (5, 6). Class switched IgE-expressing B cells are themselves tightly controlled by a strong propensity to differentiate into short-lived plasma cells rather than germinal center cells; they may also be programmed to undergo apoptosis (7–12). These effects are mediated by membrane IgE (mIgE) in an antigen-independent manner via constitutive signaling through CD19, Syk, Btk, PI3K, Akt, IRF4, and BLNK, leading to the generation of short-lived plasma cells and cell death, BCR downregulation, and/or delayed cell cycle progression (12, 13). mIgE and CD19 signaling may suppress the generation of IgE-expressing long-lived plasma cells.

Atopy is known to be heritable (14–17) but the relative contributions of variants of mIgE, its signaling pathway, and other pathways, to the overall incidence of atopy in the human population is unknown. In addition, whether de novo mutations arising in the population might be causal factors for the increasing commonness of reported allergic diseases is an open question. We present a mouse forward genetic screen to identify genes with non-redundant function in the regulation of the IgE response to allergens. Intact pedigrees of third-generation (G3) C57BL/6J mice with mixed zygosities for N-ethyl-N-nitrosourea (ENU)-induced mutations were immunized by intraperitoneal (i.p.) injection with the model allergen papain. Like many allergens, papain is a cysteine protease that stimulates Th2 responses, IgE production, and eosinophilia (18–20). Papain immunization elicits papain-specific antibodies in the absence of alum or other adjuvants; thus, papain acts as both adjuvant and antigen. Two weeks after immunization, serum papain-specific IgE was measured, along with papain-specific IgG1 in some instances. Mutations causing either an increase or decrease in the papain-specific IgE or IgG1 responses were identified by automated genetic mapping (21). We extrapolate from the number of validated mutations and the genome saturation achieved to estimate a genomic footprint for atopy under the specific conditions of our screen. We discuss the implications of our findings in understanding the etiology of allergic disease in humans.

Results

Genetic screen for mutations altering antigen-specific IgE and IgG responses to papain

Wild-type mice upregulated eosinophils and IgE in response to intraperitoneal administration of papain (Fig. S1, A and B), from which both elevated and diminished IgE responses could be distinguished (Fig. S1A). Conducted in this manner, the screen detects aberrant primary IgE responses during the sensitization phase, ranging from complete failure to elevated responses characteristic of atopy (Fig. S1C).

Initial studies to validate the screen included examination of T-cell deficient Lck^−/− mice (Fig. S2A) and Nfkb1- or Sharpin-deficient mice (Fig. S2, B and C). We found that papain-induced IgE and IgG1 depended on T cells and NF-κB signaling, as previously reported (22, 23). IgE and IgG1 responses to papain were unaffected by deficiencies of MyD88, NLRP3, or CARD9 (Fig. S2, D to F), consistent with previous reports (20, 24). However, we detected slightly elevated antigen-specific IgG1 production by Tlr4^lps3/lps3 mice (Fig. S2G), in contrast to reduced amounts produced by Tlr4^−/− mice reported previously (24). This difference may reflect the different inocula (papain vs. papain+ovalbumin) and immunization routes (i.p. vs. s.c.) in the two experiments. We show for the first time the independence of IgE production from the intracellular nucleic acid sensors cGAS or MAVS (Fig. S2H).

We screened 16,177 G3 mice from 704 pedigrees for IgE responses. These mice carried 38,710 non-synonymous heterozygous or homozygous mutations in the coding regions or splice junctions of 14,654 genes. Computational assessment of the number of genes modified by damaging and/or null alleles (25) screened once or more in the homozygous state indicated 16.5% genome saturation among all annotated protein encoding genes. For IgG1 responses, the corresponding quantities were: 3,886 G3 mice from 180 pedigrees representing 12,218 mutations in 7,956 genes (5.81% genome saturation).

Selection and validation of candidate genes

We restricted our search for mutations affecting IgE or IgG1 responses to those: 1) found in pedigrees with 10 or more G3 mice, 2) with zero or more instances of homozygosity (to allow for detection of dominant or semi-dominant homozygous lethal mutations), and 3) with at least two mice homozygous for the reference allele of the affected gene. P values for linkage of mutations to altered IgE or IgG1 phenotypes were calculated using recessive, dominant, and semi-dominant models of inheritance (21). A criterion of P < 0.05 with Bonferroni correction for the number of mutations in each pedigree was used to flag mutations as candidates, which were then evaluated with an in-house machine learning tool, Candidate Explorer (CE), for the likelihood of being causative for the phenotype, i.e. recapitulating the phenotype in mice with an independently generated mutation in the same gene (Fig. 1A). CE evaluates gene-phenotype associations based on approximately 60 variables describing properties of the mutations and phenotypic data (Fig. 1A); the program is regularly trained and updated as new data are collected. For example, mutation damage probability, gauged by PolyPhen-2 and SIFT and gene essentiality, makes a minor contribution to the CE score. More highly weighted in CE scoring are properties of the phenotypic data, such as attributing multiple phenotypes to one genotype, different alleles of the same gene producing the same phenotype, and having no phenotypic overlap between REF/HET and VAR mice. In addition, the CE score will be higher for mutations that result in large differences between REF/HET and VAR phenotypes compared to mutations that result in smaller differences.

Fig. 1.

(A) Scheme showing candidate mutation selection criteria and several of the factors positively correlated with stronger Candidate Explorer ratings. (B) Pathways implicated by the atopy screen. See Supporting Information text for details. Signals from Th2 and Tfh cells induce B cell class switch recombination to IgE, which is secreted or expressed on the membrane of B cells as the B cell receptor (BCR). Signaling pathways downstream of the BCR and CD19 are shown. The phenotype name is shown in red font with the name of the mutated gene in italics and parenthesis. An arrow indicates the direction of IgE change in the atopy screen as a result of the gene mutation (green up= increased IgE, red down= decreased IgE). A red “X” indicates a mutation that failed to be validated.

Among the group of 219 genes affected by 221 mutations with a CE rating of “potential,” “good,” or “excellent candidate” (Data file S1; predicted validation frequency of 73% for this group at the time of writing), we selected 22 genes for validation testing (Table 1 and Fig. 1B). These genes generally fell into three biological categories based on published data: class switch recombination (CSR) (Fig. 2), B cell receptor (BCR) signaling (Figs. 3 and 4), and others (neither CSR nor BCR signaling, Figs. 5 and S3). All phenotypes ascribed to CSR genes were concordant with previous reports. A substantial number of genes encoding BCR signaling components scored in the screen, suggestive of biological relevance, and we therefore selected them for validation testing.

Table 1.

Summary of atopy screen genes

	High IgE	Figure	Low IgE	Figure	High IgG1	Figure	Low IgG1	Figure
Targeted: Validated	Blk D,b,d	3D			-		Nr4a3 a,c	S3C
	Cd19 D,a,d	4A
	Itch a,d	2A
	Map1lc3b D,a,d	5E
	Pik3r1 D,a,c	4E	Flcn a,c	5M
	Plcg2 D,a,c	3L	Me2 b,c	5I
	Prkd2 D,a,c	5A
	Rasgrp3 D,a,c	3P
	Scarb2 b,d	5Q
	Syk D,a,d	3H
Not Targeted, Literature Verified	Itk a Map3k8 D,a	2E S3E	Ighe a	3A			Jak3 a,b	2H
			Rel a	2N	-		Pik3cd a	4I
			Stat6 a	2K			Rel	2N
Hypothesis: Validated			Lck c	S2A
	-		Nfkb1	S2B	-		-
			Sharpin	S2C
Hypothesis: Failed validation	Prkcb b	S7A	-		-		-
Targeted: Failed validation	Arid4a b,c	S6D	Ikbkb a,c	S5A	Tnfaip2 a,c	S5O	Ikbkb a,c	S4A
	Ppp3cb a,c	S5E					Snx14 a,c	S5I
	Zfyve26 b,d	S6E					Stx17 a,c	S5N

^DDominant or semi-dominant inheritance in the screen or validation allele.

^a“Potential,” “Good,” or “Excellent” candidate” by Candidate Explorer (22 genes)

^b“Not good candidate” by Candidate Explorer (57% -predicted validation frequency) (5 genes)

^cKnockout or replacement allele generated by CRISPR/Cas9 targeting (14 genes; see Table S1)

^dPublished knockout obtained (7 genes)

Fig. 2.

Screening results of Itch, Itk, Jak3, Stat6, and Rel alleles for effects on papain-specific IgE or IgG1 responses to papain immunization. For each gene, a scatter plot of the phenotypic data for mice carrying the original ENU-induced allele, Manhattan plot of linkage analysis data, scatter plot of phenotypic data for mice carrying an independently generated allele (for validation testing), and protein domain diagram depicting ENU-mutated residue and allele name are shown. Manhattan plots display the –log₁₀ P values (Y axis) plotted versus the chromosomal positions (X axis) of all mutations identified in the G1 founder of the pedigree. Linkage peak is labeled with gene and P value. Horizontal red and beige lines represent thresholds of P = 0.05 with or without Bonferroni correction, respectively. (A-D) Itch, (E-G) Itk, (H-J) Jak3, (K-M) Stat6, and (N-Q) Rel. REF, homozygous for the reference allele; HET, heterozygous for the reference and variant allele; VAR, homozygous for the variant allele; C2, Calcium binding domain; PRR, proline-rich region; WW, Domain with 2 conserved Trp (W) residues; HECTc, Domain homologous to E6-AP carboxyl terminus; PH, Pleckstrin homology domain; Tec, Transient erythroblastopenia or childhood; SH3, Src homology 3 domain; SH2, Src homology 2 domain; B41, bank 4.1 homologues; CC, coiled coil; DNA, DNA binding domain; LK, linker domain; RHD, Rel homology domain; TAD, transactivation domain. Data points represent individual mice (A, C, E, H, K, N, P). Data are representative of 1–2 independent experiments. Central line indicates mean, error bars indicate S.D.

Fig. 3.

Screening and validation testing of Ighe, Blk, Syk, Plcg2, and Rasgrp3 alleles for effects on papain-specific IgE or IgG1 responses to papain immunization. Data are displayed as in Fig. 2. (A-C) Ighe, (D-G) Blk, (H-K) Syk, (L-O) Plcg2, and (P-S) Rasgrp3. The Rasgrp3^aster and Rasgrp3^centre pedigrees were combined to a scatter plot of normalized serum papain-specific IgE after papain immunization (P) and corresponding Manhattan plot of combined linkage analysis data (Q). Manhattan plots display the -log₁₀ P values (Y axis) plotted versus the chromosomal positions (X axis) of all mutations identified in the G1 founder(s) of the pedigree(s). REF, homozygous for the reference allele; HET, heterozygous for the reference and variant allele; VAR, homozygous for the variant allele; C, CRISPR/Cas9-targeted allele; Cre, engineered Cre recombinase allele; −/−, engineered null allele; IGc1, Immunoglobulin C-type; IG-like, Immunoglobulin like; SH2, Src homology 2 domain; SH3, Src homology 3 domain; PH, Pleckstrin homology domain; EF, EF hand domain; PLCXc, Phospholipase C, catalytic domain X; PLCYc, Phospholipase C, catalytic domain Y; C2, Calcium binding domain; RasGEFN, Guanine nucleotide exchange factor for Ras-like GTPases, N-terminal motif; RasGEF, Guanine nucleotide exchange factor for Ras-like small GTPases; C1, Diacylglycerol (DAG) binding domain. Data points represent individual mice (A, D, F, H, J, L, N, P, R). Data are representative of 2–3 independent experiments. Central line indicates mean, error bars indicate S.D.

Fig. 4.

Screening data for Cd19, Pik3r1, and Pik3cd alleles on papain-specific IgE or IgG1 responses to papain immunization. These alleles were considered validated based on previously published reports. Data are displayed as in Fig. 2 except validation testing using an independently generated allele was not performed. (A-D) Cd19, (E-H) Pik3r1, (G-I) and Pik3cd, (I-K). REF, homozygous for the reference allele; HET, heterozygous for the reference and variant allele; VAR, homozygous for the variant allele; C, CRISPR/Cas9-targeted allele; Cre, engineered Cre recombinase allele; IG, Immunoglobulin; TM, transmembrane; SH3, Src homology 3 domain; RhoGAP, GTPase-activator protein for Rho-like GTPases; SH2, Src homology 2 domain; p85B, p85 binding domain; RBD, Ras binding domain; C2, Calcium binding domain; PI3Ka, Phosphoinositide 3-kinase, accessory domain; PI3Kc, Phosphoinositide 3-kinase, catalytic domain. Data points represent individual mice (A, C, E, G, I). Data are from 1–2 independent experiments. Central line indicates mean, error bars indicate S.D.

Fig. 5.

Screening and validation testing of Prkd2, Map1lc3b, Me2, Flcn, and Scarb2 alleles for effects on papain-specific IgE or IgG1 responses to papain immunization. These alleles were considered robust for validation despite being ranked “not good candidate” by Candidate Explorer. Data are displayed as in Fig. 2. (A-D) Prkd2, (E-H) Map1lc3b, (I-L) Me2, (M-P) Flcn, and Scarb2 (Q-T). (E and F) Multiple pedigrees containing distinct mutations in Me2 were combined for linkage analysis. Mutations, each from a single pedigree, are color coded in the phenotypic data scatter plot and designated by the associated missense mutation (E) (papain-specific IgE normalized to B6 controls, not shown). REF, homozygous for the reference allele; HET, heterozygous for the reference and variant allele; VAR, homozygous for the variant allele; C, CRISPR/Cas9-targeted allele; −/−, engineered null allele; C1, Diacylglycerol (DAG) binding domain; PH, Pleckstrin homology domain; S TKc, Serine/Threonine protein kinases, catalytic domain; SH3, Src homology 3 domain; SH2, Src homology 2 domain; Malic, Malic enzyme, N terminal domain; Malic M, Malic enzyme, NAD binding domain; CD36, scavenger receptor domain. Data points represent individual mice (A, C, E, G, I, K, M, O, Q, S). Data are representative of 1–3 independent experiments. Central line indicates mean, error bars indicate S.D.

Validation of the candidate genes was pursued by generating CRISPR-targeted mutations in wild-type genomes. CRISPR-induced knockout strains were created when the candidate phenotype was consistent with a loss-of-function mutation and evidence existed that homozygous null alleles were not lethal (10 genes). Replacement alleles (replicating the original ENU-induced DNA change) by CRISPR targeting are relatively more challenging to obtain; these were produced when evidence of lethality existed for homozygous null mutants or when a gain-of-function mutation was suspected (4 genes). Mutations that phenocopied the original ENU-induced mutation were considered validated. Some candidate variants were in genes where loss of function variants have a well-established impact on IgE production (Itk, Jak3, Stat6, Rel [Fig. 2]; Pik3cd [Fig. 4], and Ighe [Fig. 3]). When these candidates phenocopied the published phenotype, we considered them “literature verified” and did not pursue experimental validation (7 genes).

Ten of the 22 selected genes (Table 1) were validated by testing mice for their IgE or IgG1 responses (Cd19, Flcn, Itch, Map1lc3b, Nr4a3, Pik3r1, Plcg2, Prkd2, Rasgrp3, and Syk). Five genes were similarly tested but failed to be validated (Ikbkb, Ppp3cb, Snx14, Stx17, and Tnfaip2). Lastly, for seven genes (Ighe, Itk, Jak3, Map3k8, Pik3cd, Rel, and Stat6) mutant mice were not subjected to phenotypic testing but the genes were regarded as validated based on published reports (“Not Targeted, Literature Verified” in Table 1).

Additionally, five genes (Blk, Me2, Scarb2, and the linked pair Zfyve26/Arid4a) containing mutations ranked “not good candidate” (lower CE-predicted validation frequency of 39%), but which had P < 0.05 for linkage to their respective IgE or IgG1 phenotypes, appeared otherwise robust and were selected for validation testing (Table S1). Three of the five genes (Blk, Me2, Scarb2,) were validated; two genes (Zfyve26, Arid4a) representing one phenotype failed to be validated.

Summary of implicated genes and their effects

Mutations in 17 genes—either validated by robust LOF alleles or strongly supported by previous experimental observations—influenced IgE responses and were detected in screening (Table 1). For 12 of these genes, the effect of mutations was to augment the IgE response (Blk, Cd19, Itch, Itk, Map1lc3b, Map3k8, Pik3r1, Plcg2 Prkd2, Rasgrp3, Scarb2 and Syk). We noted that a substantial fraction of these mutations (9 out of 12) displayed dominant or semi-dominant phenotypic effects (Table 1, marked with superscripted “D”). For the remaining 5 validated genes, mutations diminished or abolished IgE responses (Flcn, Ighe, Me2, Rel, and Stat6). Null alleles of Ppp3cb, Zfyve26, and Arid4a, and a replacement allele of Ikbkb failed to validate as affecting IgE (Table 1). Finally, one gene, Prkcb, was not implicated in screening, but its well-described function in BCR signaling led us to test an existing Prkcb allele, Untied (26), for an effect on the papain induced IgE response. No effect was observed (Fig. S7A, B).

Diminished IgG1 responses were attributed to mutations in four genes (Jak3, Nr4a3, Pik3cd, and Rel; Table 1); mutations in Ikbkb, Snx14, and Stx17 failed to be validated (Fig. S5). No validated ENU-induced mutations were found to increase the IgG1 response (Tnfaip2 failed to validate; Fig. S5Q).

Discussion

The forward genetic analysis of mice with a defined genetic background permitted unbiased and unambiguous assignment of causality to single gene variants responsible for aberrant IgE or IgG1 phenotypes. Because mice and humans generate similar adaptive immune responses, our findings likely have implications for atopy in humans. To our knowledge, no other group has approached the genetic basis of atopy in this way.

Including Lck, Prkcb, Sharpin, and Nfkb1, we investigated 32 mutations in 31 genes for their effects on the quantity of serum papain-specific IgE or IgG1 produced in response to immunization. We suggest that measurement of allergen-specific IgE more accurately reflects atopy than total IgE or basal IgE, which have not consistently been associated with anaphylaxis or food allergy (27). Using CRISPR-targeted knockout/replacement alleles or published, available knockout alleles, we validated the effects of mutations in 23 (74%) of the 31 genes; the effects of eight mutations (26%) were not validated. Many of the validated genes aligned to signaling pathways important for the production of IgE. Among 20 validated gene candidates identified in screening or hypothesized to affect the IgE response, 15 were in either the CSR or BCR signaling pathways (Fig. 1B). We discuss putative mechanisms by which specific mutations alter IgE responses to papain in the Supporting Information; we describe relevant pathways beginning with Th2 stimulation leading to IgE CSR, then BCR signaling, and ending with a summary of genes with unknown function in the IgE response.

Many validated genes were kinases (Itk, Jak3, Map3k8, Blk, Syk, Pik3cd, and Prkd2) or GTPases (Rasgrp3 and Pik3r1) affected by ENU-induced mutations in the catalytic domain. It may be that catalytic domain mutations generally have a more significant impact on protein function than mutations in non-catalytic domains. For these kinase domain mutations, we recapitulated the original phenotypes with null alleles, confirming that the ENU-induced mutations conferred a loss of function. However, in several instances the complete knockout revealed a heterozygous increase in IgE not observed for the ENU-induced mutation, which was recessive (Blk, Pik3r1, and Rasgrp3; also Cd19, not a kinase). These findings suggest that the ENU-induced mutations in Blk, Cd19, Pik3r1, and Rasgrp3 were relatively mild and heterozygosity resulted in retention of >50% of wild-type protein function, which was sufficient to confer the wild-type phenotype. In contrast, 50% of wild-type protein function caused by heterozygosity for a null allele was insufficient for normal IgE production. Our genetic findings support previous reports that constitutive signaling from the IgE BCR initiates gene expression changes that affect cell fate and lifespan to limit IgE production. These specific events are still unclear but could be elucidated by transcriptomic analysis using bulk RNA-Seq and single cell RNA-Seq of IgE B cells (28, 29). Recurrent haploinsufficient phenotypes caused by mutations of BCR signaling components suggest moderate decreases in signaling curtail the events that limit IgE production. Other mutations causing a dominant decrease in IgE were in nuclear transcription factors that dimerize to exert their activity (Rel and Stat6).

Considering the number of validated genes in the CSR and BCR immune pathways (10 increasing IgE) and the genome saturation (16.5%) for the atopy screen to date, we estimated that a total of 60 CSR/BCR-related genes might be expected to increase papain-specific IgE should saturation reach 100%. However, this conservative calculation likely underestimates of the number of genes that affect antigen-specific IgE responses since only a single allergen, papain, was used and genes not known to participate in these pathways were found to increase antigen-specific IgE (Map1lc3b, Prkd2, and Scarb2). Alternative atopic sensitization methods or allergens may find additional genes or pathways. Further, the immunization method of intraperitoneal injection was artificial but more practical for large scale screening compared to a labor intensive but more physiologic method such as exposure by skin, inhalation or ingestion. This could prevent detection of genes such as Flg (encoding filaggrin) that lead to atopy by disturbing skin integrity (30–32).

Determining a model system that would resemble the human situation should be disease-specific. For example, several of the BCR signaling mutants may appropriately model human food allergy and allergic rhinitis, which involve increased allergen-specific IgE in the absence of immunodeficiency. Some mutants may be unsuitable for modeling non-IgE mediated disease such as eosinophilic esophagitis, but the Stat6 mutant may be an interesting model to study the effect of blocking IL-4 signaling to treat allergies (33). The Plcg2 mutant displayed increased allergen-specific IgE and may be a good model to study the pathology and mechanisms underlying the increased IgE and atopy associated with mutations in human PLCG2 (34).

A significant finding from our screen was the high incidence of dominant or semi-dominant inheritance patterns among elevated IgE phenotypes. Of the 12 validated gene candidates for increased IgE, nine were dominant or semi-dominant allelic variants, a frequency significantly higher than the frequency of dominant or semi-dominant alleles causative of all other CRISPR-validated phenotypes screened in our group (P = 3.33 × 10⁻⁵, Fisher exact test, n= 9 Atopy^Dom, 3 Atopy^Reccessive, 190 Other^Dom, 854 Other^Reccessive, Table S2). These dominant or semi-dominant mutations were apparently innocuous in the heterozygous state with respect to immune competence (as indicated by phenotypic testing in other immunologic screens; Table S2), suggesting that they are not gain-of-function variants, and importantly, reflecting the clinical presentation of most allergic patients. We speculate that heterozygous mutations with dominant effects on atopy, together with the large number of genes estimated to comprise the genomic footprint of the high IgE-response phenotype (~60 genes from CSR/BCR signaling alone), may explain the high incidence of heritable atopic disease in humans (as in normally non-atopic mice modified by a random mutagen).

Indeed, among the seven validated, dominant CSR/BCR mutations described here, 1.3% of human subjects in the ExAC database are heterozygous for null or probably damaging alleles (n= 780 alleles; Data file S2) of the genes; on a per gene basis, this amounts to an allele frequency that is on average less than the 1% allele frequency needed for identification by genome-wide association studies. Overall, we estimate that 37.5% of the population carries a dominant CSR/BCR signaling atopy risk mutation in the heterozygous state (Materials and Methods). An atopic mutation will predispose an individual to being sensitized to an allergen and they may not manifest allergic symptoms. An additive effect of heterozygous atopic mutations at multiple loci may increase the risk of allergic disease. Thus, rare, damaging mutations in the heterozygous state may underlie allergy and atopy in otherwise healthy children and adults.

Conclusions

We present the first forward genetic screen for mutations that influence atopy, which identified 31 candidates of which 74% (23/31) were validated. Of 20 genes that affected IgE in our screen, 55% (11/20) were not previously known to influence atopy. Although further work will be necessary to confirm similar associations in humans, a forward genetic approach in mice will yield important contributions to resolving genetic determinants of allergy. These discoveries can guide genome-wide association studies, which so far have revealed only weak associations with allergy pathophysiology. Given the current genomic saturation rate, we expect that many more novel gene-atopy associations remain to be discovered. We estimate that a heterozygous mutation in at least one of these genes may be present in 37% of the human population.

Materials and Methods

Detailed descriptions of CRISPR/Cas9 gene targeting in mice, measurement of eosinophils, ELISA analysis, flow cytometry, T cell-dependent and –independent antigen immunization and measurement of antibody responses, and measurement of TNF production by macrophages are provided in the online supplement.

Mice

Male C57BL/6J mice (designated G0) were mutagenized with ENU as previously described (35). Three generations (G1, G2, G3) of mice were bred from the G0 mice; G3 mice were subjected to phenotypic screening (21). The sources of previously reported mouse strains are provided in the online supplement. Male and female mice were used in all experiments and data for males and females were combined for analysis. Mice were maintained at the University of Texas Southwestern Medical Center. All studies were performed in accordance with the guidelines established by the Institutional Animal Care and Use Committee of the University of Texas Southwestern Medical Center and with the National Institutes of Health Guide for the Care and Use of Laboratory Animals.

Immunizations

For dose-response studies, C57BL/6J mice were injected i.p. on day 0 with the indicated dose of papain (EMD Millipore, 5125 Carica papaya) in 0.5 ml PBS. For screening, G3 mice were injected i.p. on day 0 with 0.5 mg papain in 0.5 ml PBS; subcutaneous immunization (0.5 mg papain in 0.2 mL PBS injected over the shoulders) was performed on a small subset of G3 mice (n=1,709) with equivalent effects and screening data were combined with those from i.p. immunizations. On day 15 after immunization, blood was collected in microcentrifuge tubes (Fisher Scientific) for ELISA analysis. G3 mice were injected with ovalbumin (OVA, Sigma-Aldrich, A5503) adsorbed to Alum (InvivoGen, vac-alu-250) in a 1:1 ratio and administered by intramuscular injection; blood was collected 14 days after immunization for ELISA analysis. Intraperitoneal NP-Ficoll immunizations were performed as previously described (36).

ELISA

ELISA analysis to detect serum IgE or IgG1 was performed using standard techniques detailed in the online supporting information.

Automated genetic mapping

Automated genetic mapping was performed as previously described (21). Briefly, genotypes at all mutation sites present in the exomes of G3 mice were determined prior to phenotypic screening: tail DNA from G1 males was subjected to whole exome sequencing using an Illumina HiSeq 2500 instrument; G2 and G3 mice were then genotyped at the identified mutation sites using an Ion PGM (Life Technologies). Following phenotypic screening, linkage analysis using recessive, additive, and dominant models of inheritance was performed for every mutation in the pedigree using the program Linkage Analyzer; phenotypic data scatter plots and Manhattan plots were displayed using the program Linkage Explorer.

Comparison of frequency of dominant and semi-dominant phenotypes in other screens

We compared the frequency of dominant and semi-dominant mutations encountered in the papain screen to that identified in all other screens performed in our laboratory. First, we calculated the ratio of validated dominant or semidominant phenotypes among all papain-specific IgE phenotypes. Then we counted all phenotypes from all screens that had a dominant, semi-dominant, or recessive mode of inheritance and met the following criteria: 1) probably damaging or probably null causative mutation, 2) total mouse number in pedigree ≥ 20, 3) at least 3 mice of every genotype (wild type, heterozygous and homozygous for the variant allele), and 4) P < 0.001 with Bonferroni correction. The proportion of dominant and semi-dominant mutations was then calculated. The Fisher exact test was used to compare the ratios for the papain screen vs. all other screens.

Calculating prevalence of atopy-risk genes with heterozygous mutations

Monitoring genomic saturation for genes with mutations screened once in the homozygous state in the papain screen was performed as previously described (21) with adjustments using correction factors for PolyPhen-2 classifications [CorrectionFactorPPN2] to account for the overestimation of damage by PolyPhen-2 as previously described (25). Dividing the number of validated dominant and semi-dominant CSR/BCR gene mutations (seven) by the percent genome saturation (16.5%) equals the estimated total number of dominant/semi-dominant CSR/BCR genes expected to increase allergen-specific IgE (42 genes).

To estimate the frequency of persons carrying a mutation in the heterozygous state that would increase atopy, we started by calculating the average probability of genetic damage for each gene (P_GeneDamaged) based on the variant alleles and their frequencies in the ExAC (F_ExAC) database (alleles from TCGA database excluded) (37). The damaging effect of an allele was predicted using PolyPhen-2 and then F_ExAC for that allele was corrected by multiplying with CorrectionFactorPPN2 to reflect the “true” probability (25) of damage. For each gene, corrected frequencies for all alleles were summed to give P_GeneDamaged:

P_GeneDamaged = [(F_ExAC-Null1 + F_ExAC-Null2 + … + F_ExAC-NullX) x CorrectionFactorPPN2_Null (0.61)] + [F_{ExAC-ProbDamg1} + F_{ExAC-ProbDamg2} + … + F_{ExAC-ProbDamgX}) X CorrectionFactorPPN2_ProbDamg (0.17)]

Next, we performed a Monte Carlo simulation to randomly select a set of genes equal to the total number of dominant atopy genes (n=42) with the seven validated genes set as constants in the simulation. We used P_GeneDamaged to determine the probability that a person has no damaging mutations in any of the 42 randomly selected genes. Then we subtracted that value from 1 to calculate the probability that an individual is heterozygous for at least 1 damaging mutation in a dominant atopy risk gene (P_{AtopicMutation}):

$$P_{AtopicMutation}=1-\prod _1^{42}\left(1-P_{GeneDamaged}\right)$$

We performed 1000 unique Monte Carlo simulations and averaged the resulting P_{AtopicMutation} across all 1000 simulations to obtain 37.5%.

Candidate Explorer

CE is a software tool to assess the likelihood that a putative mutation is causative for an observed phenotype. CE utilizes a supervised machine learning algorithm to classify putative mutations into four categories: excellent candidate, good candidate, potential candidate, and not good candidate. The training set consisted of 1,378 verified and 2,120 excluded mutation-phenotype associations. We used the CRISPR/Cas9 system to generate germline mutant alleles and assayed them for phenotypic effect. The input vector is comprised of over 60 features describing the properties of the mutations and the characteristics of the phenotypic data, including mutation damage probability by PolyPhen-2 and SIFT, gene lethality, clustering of observations within one genotype group, overlaps between different genotype groups, and consistency among multiple pedigrees. The prediction model is built using a random forest algorithm implemented in the R caret package. Using potential candidate as a cutoff, the current version of CE achieves 73% accuracy (90% sensitivity and 62% specificity). The program had worse performance for the atopy screen (n=19) with an overall accuracy of 42% (recall/sensitivity 67% and specificity 20%). The atopy screen made up a minority of cases in the training set for CE (n=25), so the weighting of certain characteristics is not optimized for this screen.

Statistical analysis

Data represent mean ± SD in all graphs depicting error bars. The statistical significance of differences between experimental groups in Fig. S2 was determined by Student’s t test using GraphPad Prism 6. *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001; ****P ≤ 0.0001; ns, not significant with P > 0.05. The P values of association between genotype and phenotype were calculated using a likelihood ratio test from a generalized linear model or generalized linear mixed effect model and Bonferroni correction applied (21).