BACH2 immunodeficiency illustrates an association between super-enhancers and haploinsufficiency.

The transcriptional programs that guide lymphocyte differentiation depend on the precise expression and timing of transcription factors (TFs). The TF BACH2 is essential for T and B lymphocytes and is associated with an archetypal super-enhancer (SE). Single-nucleotide variants in the BACH2 locus are associated with several autoimmune diseases, but BACH2 mutations that cause Mendelian monogenic primary immunodeficiency have not previously been identified. Here we describe a syndrome of BACH2-related immunodeficiency and autoimmunity (BRIDA) that results from BACH2 haploinsufficiency. Affected subjects had lymphocyte-maturation defects that caused immunoglobulin deficiency and intestinal inflammation. The mutations disrupted protein stability by interfering with homodimerization or by causing aggregation. We observed analogous lymphocyte defects in Bach2-heterozygous mice. More generally, we observed that genes that cause monogenic haploinsufficient diseases were substantially enriched for TFs and SE architecture. These findings reveal a previously unrecognized feature of SE architecture in Mendelian diseases of immunity: heterozygous mutations in SE-regulated genes identified by whole-exome/genome sequencing may have greater significance than previously recognized.

Introduction

The inheritance pattern of genetic diseases consists of a spectrum, ranging from the vast majority representing polygenic susceptibility variants (usually identified on GWAS studies) to the minority, which are monogenic and manifest in either a recessive or dominant manner. It is now appreciated that mutations in over 300 different genes can cause primary immunodeficiency (PID), many of which affect T and B lymphocyte function[1-4]. PIDs are often paradoxically associated with autoimmunity[3-7]. Common variable immunodeficiency (CVID), a major form of PID with antibody deficiency, is typically associated with recurrent infections and autoimmunity[8]. Recently developed gene-sequencing technologies now allow for rapid identification of PIDs but have also raised the important question of how to interpret the many heterozygous mutations seen in both patients and healthy controls. Relatively few PID syndromes are caused by haploinsufficiency, an autosomal dominant pattern of disease inheritance, where one allele is damaged and only a single functional allele remains[9]. Genes, such as CTLA4, are particularly susceptible to haploinsufficiency and the reasons are unknown[10]. In the light of many healthy people that harbor heterozygous loss of function or hypomorphic variants, why should partial changes in gene expression have significant consequences to health?

Promoters and enhancer elements govern gene expression. Most, such as housekeeping genes like actin, are regulated by a limited number of associated enhancers, known as “typical enhancers”[11]. By contrast, 5–10% of genes have a complex enhancer structure consisting of multiple enhancers that collectively are described as SEs[12,13]. Genes with associated SEs have a highly regulated pattern of gene expression; single nucleotide polymorphisms associating in GWAS studies with autoimmune diseases are preferentially enriched within SE regions[14]. These findings suggest that minor changes in regulatory function at SE regions could have significant consequences to the immune system for genes regulated by SEs.

BACH2 is a typical example of an SE-regulated gene associated with autoimmune disease. It is a highly conserved member of the basic and leucine zipper domain (bZIP) superfamily of TFs and a critical regulator of both T and B lymphocyte differentiation and maturation[15,16]. Polymorphisms in the human gene locus associate with multiple autoimmune diseases, including asthma[17], insulin dependent diabetes mellitus[18], Crohn’s and celiac diseases[19,20], vitiligo[21] and multiple sclerosis[16,22]. The Bach2 gene locus has the largest SE structure seen in mouse lymphocytes[14]. Homozygous deletion of Bach2 in mice results in spontaneous fatal autoimmunity between 3 and 9 months of age[15]. Functionally, BACH2 acts as a repressive “guardian” TF that regulates the balance between a network of other TFs critical to T and B cell specification and maturation. In B cells, BACH2 controls the balance between Pax5 and Blimp1 by repressing the latter[23,24], to decelerate plasma cell differentiation and permit antibody class switch recombination (CSR) (allowing expression of IgA, G and E isotypes)[25]. Consequently, mice lacking BACH2 have B cells with impaired CSR that rapidly differentiate into IgM-restricted plasma cells. In T cells, BACH2 regulates networks of genes that control T cell effector lineages[14] and cellular senescence[26], thus limiting differentiation into effector cells[15] and promoting development of FoxP3+ regulatory T cells (Treg). Treg cells are a non-redundant suppressive lineage of T cells that prevent development of autoimmune diseases by controlling over-activation of the immune system[27]. Thus, mice deficient in BACH2 demonstrate both a paucity of Treg cells and an excess of memory/effector T cells that age and die prematurely, resulting in autoimmunity.

Structurally, BACH2 contains a BTB/POZ domain that mediates homo-and hetero-dimerization at its N-terminus and a bZIP domain at the C-terminus required for DNA binding. The dimerization domain is an alpha-helical structure containing a cysteine residue that is capable of forming a disulphide bond with its opposite partner[28]. Thus homo-dimerization is likely to be stabilized by a covalent modification that occurs soon after protein folding. BACH2 dimers translocate to the nucleus where they interact with target DNA loci at palindromic Maf recognition elements (MARE), either alone or in collaboration with other members of the bZIP family, such as the small Maf proteins (MafF, MafG and MafK)[16]. This interaction, for example at the Prdm1 locus that encodes Blimp1, represses gene expression.

Here we describe a novel PID caused by haploinsufficiency of BACH2 and propose a shared genetic mechanism to explain why some genes are particularly susceptible to causing disease by haploinsufficiency. We conclude that the interpretation of heterozygote variants in these genes should be regarded as significant and be prioritized in any investigation of novel genetic disease by whole exome sequencing.

Results

BACH2 mutations associate with CVID and colitis

We investigated a female (Figs. 1a and 1b – Family A) with infancy-onset colitis, who became ill at 19 years old with non-infectious fever, splenomegaly (21.7 cm, compared to 10–12 cm in normal adults) (Fig. 1c) and pancytopenia. Fever and cytopenia improved with corticosteroids, but lymphopenia, deficiency in immunoglobulin (Ig)M, IgG, IgA and IgE, ongoing colitis, lung infiltrates and recurrent upper respiratory tract infections persisted (Fig. 1c, Table 1 and Supplementary Table 1). A colonic biopsy demonstrated inflammatory changes with crypt branching and prominent lymphocytic infiltrates around the crypts (Fig. 1d), with significantly reduced FoxP3+ regulatory T (Treg) cells compared with healthy controls or patients with classical IBD (Fig. 1e). The early disease onset and unusual symptoms in the absence of family history prompted us to perform whole exome sequencing on the patient and healthy parents as a trio. After excluding all variants with minor allele frequency (MAF) >0.01, no candidate variants remained to support a hypothesis of recessive inheritance. We found a novel heterozygous de novo, non-synonymous mutation in BACH2, c.T71C, predicted to be deleterious (Supplementary Table 2), substituting a highly conserved leucine with proline (L24P), and not present in healthy family members (Fig. 1b and Supplementary Figure 1). A second family (Family B) that had been previously investigated by exome sequencing (Fig. 1a) was found to have a heterozygous point mutation in BACH2, c.G2362A (Fig. 1b), substituting glutamic acid with lysine (E788K) in a father and daughter, both of whom presented with inflammation of both small and large bowel, together with pulmonary disease, including recurrent sino-pulmonary infections, bronchiectasis and fibrosis (Fig. 1c and Supplementary Table 1). The BACH2 mutation was not seen in healthy family members (Supplementary Figure 1). The father (proband) was deficient in all Ig sub-types; his daughter had undetectable IgA (Supplementary Table 1). Detailed clinical features are described in the Supplementary notes, Tables 1 and Supplementary Fig 1. We found no low MAF variants nor causative mutations in genes causing monogenic IBD or other recognized primary immunodeficiencies[29-31].

Figure 1

Pedigrees and phenotype of patients with mutations in BACH2

(a) Pedigrees of two families with heterozygous missense coding mutations in BACH2, resulting in L24P (left) and E788K (right) amino acid substitutions. Shown are affected heterozygotes (filled symbols) and unaffected family members (open symbols). Arrows indicate probands; WT = wild-type allele; Mut = mutant allele. (b) Sanger sequencing chromatograms of the affected individuals in both families. For each individual, the two alleles of the sequenced region of BACH2 and base positions are shown above the chromatograms. Subject A.II.1 had a heterozygous T to C mutation at coding position 71 whereas patients B.II.1 and B.III.2 were heterozygous for G to A base substitutions at position 2362. (c) Computerized tomography scans showing splenomegaly (arrow in upper left) and pulmonary nodules (red circle in upper right) in patient A.II.1 and bronchiectasis (dilated airways; arrow in lower left) and fibrosis (“honeycombing” circled in lower right) in subject B.II.1. (d) Photomicrograph of a hematoxylin and eosin-stained section from a colonic biopsy from patient A.II.1 showing crypt branching and lymphocytic inflammatory infiltrate around the crypts. (e) Immunofluorescent staining of colonic biopsy from patient A.II.1, control IBD patient and healthy control for nuclear DNA (DAPI, blue), CD3 (green) and FoxP3 (orange). Shown are representative sections (left) and cumulative (mean ± sem) quantification (right) from four low power fields per patient (500–3000 CD3+ cells counted per low power field); white scale bar = 100 μm in main image and 2 μm in insets. *p<0.05, **p<0.01 by t-test.

Table 1

Summary clinical characteristics of patients with missense mutations in BACH2.

	Patients
Demographic and clinical characteristics	A.II.1	B.II.1	B.III.2
Age, Sex	19, F	63, M	40,F
Lymphadenopathy	Yes	Yes	Yes
Splenomegaly	Yes	No	No
Intestinal manifestations	Yes	Yes	Yes
Chronic diarrhea	Yes	Yes	Yes
IBD	Colitis	Not biopsied	UC aged 10; Crohn’s aged 32
Pulmonary manifestations	Yes	Yes	Yes
Recurrent sino-pulmonary infections	Yes	Yes	Yes
Radiographic changes on chest CT	Yes	Yes	Not imaged
Immunoglobulins†
IgM	Low	Low	High
IgG	Low	Low	High*
IgA	Low	Low	Low
IgE	Low	Low	Normal
On IvIg treatment	Yes	Yes	No
EBV antibodies	N/A (DNA negative)	N/A	High
RhF	N/A	N/A	N/A
dsDNA antibodies	Negative	N/A	N/A
ANCA	Positive (pANCA)‡	N/A	N/A
ANA	Negative	N/A	Negative

IvIg, intravenous immunoglobulin; EBV, Epstein-Barr virus; RhF, rheumatoid factor, dsDNA, double-stranded DNA; ANCA, anti-neutrophil cytoplasmic antibody; p-ANCA, perinuclear ANCA; ANA, antinuclear antibody; UC, ulcerative colitis; N/A not assessed.

Absolute values given in Supplementary Table 1;

progressive decline in IgG;

positive by immunofluorescence but negative for myeloperoxidase and proteinase III antibodies by ELISA.

In the lymphocytes of affected individuals, we found decreased expression of FoxP3 in CD4+CD25hiCD127lo regulatory T cells (Treg) (Fig. 2a) and increased expression of the TH1 transcription factor T-bet and two gut-homing receptors, CCR9 and β7-integrin on CD4+ T cells[32,33] (Fig. 2b). In the patient B cells, we found a marked reduction in CD19+CD27+ memory and IgG class-switched CD27+IgG+ B cells (Fig. 2c). These features were not present in healthy controls or patients with inflammatory bowel disease (IBD) (Supplementary Fig. 2a). Furthermore, CD24+CD38+ transitional B cells were increased in patients (Supplementary Fig. 2b). In vitro activation of naïve B cells from patients resulted in significantly impaired plasmablast generation, class-switch recombination and class-switched antibody secretion in the presence of IL-21 (Figs. 2d and 2e), suggesting a defect in B cell maturation towards memory and plasma cells, similar to Bach2 knockout mice[34]. Polyclonal activation of T cells resulted in reduced CD4+ T cell proliferation compared with healthy controls (Supplementary Fig. 2c). In summary, the immunophenotype of patients with mutations in BACH2 consisted of compromised Treg cells, enhanced TH1 differentiation, impaired proliferation and defective B cell maturation and Ig class switching.

Figure 2

Immunophenotype of patients with mutations in BACH2

(a–c) Treg cells (a), T cell (b) and B cell (c) immunophenotype of patient and healthy control peripheral blood cells. Shown are total FoxP3 expression (mean fluorescent intensity (MFI)) within CD4+CD25hiCD127lo cells (a), expression of the transcription factor T-bet and gut-homing receptors (CCR9 and β7-integrin) in bulk CD4+ T cells (b) and total memory (c, left) and class-switched memory B cells (c, right) in bulk B cells. (d-e) Plasmablast formation (d, left panels), IgG class switch recombination (d, right panels) and Ig secretion (e) in naïve patient and healthy control B cells activated in vitro as indicated. Shown are representative flow cytometry plots and cumulative data. N.D. = not detected; very low values are shown above the bars for clarity. In (a-d) representative flow cytometry plots are shown together with cumulative data from all patients and matched controls. Note that IgG secretion in (e) does not include patient B.III.2, who has normal IgG secretion. Bars show mean ± sem throughout. *p<0.05 **p<0.01 ***p<0.001 by t-test (a-c), one-way ANOVA (d) and Kruskal-Wallis test (e).

BACH2 silencing mimics patient lymphocyte phenotypes

We next measured BACH2 protein expression by flow cytometry and found it was reduced in patient CD4+, CD8+ and B lymphocytes despite normal mRNA expression (Figs. 3a,b). We measured protein expression of Flag-tagged vectors encoding wild-type (WT) or mutant forms of BACH2 in transfected HEK293T cells and found that mutant forms of the protein accumulated less than WT (Fig. 3c), at all time points measured and concentrations of vector used (Supplementary Fig. 3a–b). PRDM1, which encodes the protein BLIMP1, is a target of BACH2-mediated transcriptional repression[24]. We found that patient naïve B cells and CD4+ T cells expressed significantly higher levels of PRDM1 mRNA compared with healthy controls suggesting a release from BACH2 repression (Figs. 3d,e). Furthermore, this difference could be reversed by forced expression of WT BACH2 in patient CD4+ lymphocytes (Fig. 3e). These observations suggested a causal relationship between reduced BACH2 expression in patients and their cellular phenotype. To confirm this relationship, we silenced BACH2 expression in healthy control T and B cells using RNAi by ~50% and carried out functional phenotyping (Supplementary Figs. 4a,b). Silencing BACH2 in control CD4+ T cells led to a significant rise in PRDM1 mRNA (Fig. 3f) and resulted in reduced proliferation of CD4+ T cells (Supplementary Fig. 4c), in a similar fashion to that seen in primary CD4+ T cells from patients (Supplementary Fig. 2c). In addition, silencing BACH2 in healthy control B cells, significantly suppressed in vitro class switch recombination towards the IgG and IgA isotypes (Fig. 3g). Thus, experimental silencing of BACH2 in healthy T and B cells recapitulated the phenotype seen in primary cells of the patients.

Figure 3

The cellular phenotype is attributable to reduced BACH2 protein expression

(a) BACH2 protein expression in primary immune cells of patients and controls. Shown are representative flow cytometry plots with MFIs indicated (left panels) and cumulative BACH2 protein expression (right panels) from patients relative to controls. (b) Cumulative BACH2 mRNA expression from naïve B cells of patients and controls. (c) Representative immunoblot for Flag and Hsp70 from lysates of HEK293T cells transfected with empty vector (EV), Flag-tagged WT or mutant murine Bach2 (L24P or E786K, the murine equivalent of E788K). Shown are a representative blot (left) and cumulative quantifications from n = 5 experiments (right). (d) PRDM1 mRNA expression in naïve B cells from patients and healthy controls: cumulative data. (e and f) PRDM1 mRNA expression in CD4+ T lymphocytes of healthy controls and patients transfected with either control or BACH2 (e) and healthy donor CD4+ T lymphocytes transfected with control or BACH2 RNAi (f). (g) Plasmablast formation, IgG class switch recombination and IgA secretion in naïve healthy control B cells transfected with control RNAi or RNAi specific for BACH2 and activated in vitro as shown. Shown are representative flow cytometry examples and cumulative data (n = 5, 5 and 4 experiments, respectively). Bars show mean ± sem; *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001 by t-test (a, d), Wilcoxon test (f) and ANOVA (c, e and g).

BACH2 gene mutations impair protein stability

Both mutations that we identified affect highly conserved amino acid residues in BACH2 (Fig. 4a). Murine and human BACH2 share 90% sequence identity and L24 is conserved across species and with other members of the BTB/POZ domain family (Supplementary Figs. 5a,b and Supplementary Table 2). L24 resides within α-helix-1 (residues 18–34) of the BTB/POZ domain, a key part of the BACH2 homo-dimerization interface (Figs. 4b and 4c). The mutant proline residue likely perturbs α-helix-1 of the BTB/POZ domain and places a polar residue into the hydrophobic face of that helix, which we predicted would decrease dimer stability (Supplementary Table 2). We expressed and purified the BTB domains from both WT and L24P mutant proteins. The WT protein was soluble and formed dimers (Fig. 4d), whereas the L24P mutant was insoluble in solution, likely misfolded, and formed multiple aggregated species (Fig. 4e). E788, the site of the C-terminus mutation, is again highly conserved (Supplementary Fig. 5a). Though not characterized by structural studies, it is in proximity to a nuclear export signal (Fig. 4a). We found that wild-type BACH2 protein was evenly distributed in both cytoplasm and nucleus, whereas the E788K mutant protein was aggregated in the cytoplasm with relatively little in the nucleus (Fig. 4f and Supplemental Movies 1–2). Similar protein aggregates were observed in HEK293T cells transfected with this C-terminal variant (Supplementary Fig. 5c and Supplemental Movies 3–4). By contrast, aggregates were not detected in lymphocytes expressing the L24P mutant from patient A.II.1, although, as noted, L24P mutant BACH2 protein expression levels were lower than WT control (Supplementary Fig. 5d). Thus, both gene mutations impair BACH2 protein stability.

Figure 4

BACH2 mutations produce unstable proteins

(a) Domain schematic of BACH2 protein and point substitutions in patients. BTB/POZ, BR-C, ttk and bab or Pox virus and Zinc finger domain; bZIP, basic leucine zipper; NES, nuclear export signal. (b), Ribbon representations of BACH2 POZ domain (crystal structure form II, PDB: 3OHV); wild-type protein (above) with expanded and rotated interface view (below); yellow, intermolecular disulfide at position 20; orange, leucine residues at position 24. (c), (top) WT POZ domain dimer interface (PDB: 3OHV); (bottom) homology model of BACH2L24P: WT POZ hetero-dimer, illustrating local changes. In each, one monomer is rendered as a partially transparent hydrophobicity surface (orange = hydrophobic, white = intermediate, blue = hydrophilic) and the other as a ribbon (green); selected side chains are shown as sticks. Cys20 (yellow) and Ile23, Leu24, and Leu27 (all orange) form a hydrophobic patch on α-helix-1; two of these patches are in close contact at the WT dimer interface. N.B. the lower diagram is not meant to represent the structure accurately but is shown merely to indicate regional changes. (d–e) Analytical ultracentrifugation of purified wild-type (WT) p.BACH2 (d) and mutant p.BACH2L24P (e) BTB/POZ domain; sedimentation direction is left to right; M = sample meniscus. WT protein is dimeric (35 kDa), as determined by sedimentation equilibrium measurements (shown in d, right), migrating with single boundary with sedimentation coefficient (S) of 2.6. The mutant exhibits several boundaries (S values from 4 to 18), indicating heterogeneous large protein aggregates (e). (f) Representative confocal microscopy of primary lymphocytes from healthy control and patient B.II.1 stained for BACH2 (green) and Hoechst (blue); arrows highlight cytoplasmic aggregates. Scale bars: 5μm in main, 2μm in inset. Bars show quantification (mean ± sem, n=3 experiments) of cells containing aggregates per high power field (HPF) and BACH2 nuclear localization. *p<0.05 by t-test.

BACH2 mutations are not dominant negative

In both families, the BACH2 gene mutations could potentially act in a dominant negative manner. To test this, HEK293T cells were co-transfected with Flag-tagged WT together with untagged WT or mutant BACH2 protein-expressing constructs. Neither patient mutant altered WT protein expression (Fig. 5a). The experiment was repeated with HEK293T cells co-transfected with vectors encoding two tagged WT (HA-BACH2 and Flag-BACH2) forms of the proteins together with either untagged WT or mutant BACH2 protein-coding constructs (Fig. 5b). Co-immunoprecipitation studies showed that WT untagged BACH2, but not mutant forms of the protein, interfered with dimerization between HA and Flag-tagged WT BACH2. Furthermore, when WT Flag-BACH2 was co-transfected together with HA-tagged WT, L24P or E786K BACH2, we detected reduced mutant HA-BACH2 bound to Flag-tagged WT BACH2 after immunoprecipitation, in proportion to the reduction in protein accumulation, implying limited, if any, effects on WT BACH2 (Supplementary Fig. 6a). All these results were consistent with our earlier findings of loss of stability of the mutant proteins compared with wild-type proteins (Fig. 4). Finally, we used retroviral constructs encoding murine WT or mutant BACH2 to transduce Prdm1-YFP transgenic CD4+ T cells. Forced expression of wild-type BACH2 alone led to a significant reduction in the expression of Prdm1-YFP, but co-transduction with either mutant form of BACH2 did not interfere with repression of the Prdm1 reporter in primary mouse lymphocytes (Fig. 5c). Collectively, these data indicate that neither BACH2 mutation exerted a dominant negative effect.

Figure 5

Mutant forms of Bach2 do not exert dominant negative effects

(a) Immunoblot for Flag and Hsp70 in HEK293T cells co-transfected at 1:1 ratio with Flag-tagged WT murine BACH2 and untagged WT and mutant forms of murine BACH2. Shown is a representative from n = 3 independent experiments. (b) co-immunoprecipitation of Flag- and HA-tagged WT Bach2 transfected into HEK293T cells together with untagged WT and mutant forms of murine BACH2 at 1:1:1 vector ratio. Shown is a representative example from n = 3 independent experiments (left) and quantification of the co-immunoprecipitated Flag and HA signals (right). (c) Blimp1-YFP signal in Blimp1-YFP Tg mouse CD4+ T cells co-transduced at 1:1 ratio with retrovirus supernatants encoding WT and mutant forms of murine BACH2. Shown is a representative example (left) and cumulative data (mean ± sem) from n = 4 independent experiments (right). *p<0.0001 by ANOVA.

Bach2− mice have impaired lymphocyte development

In the absence of a dominant negative effect we next turned to haploinsufficiency as an explanation. Complete absence of Bach2 in mice results in B cell immunodeficiency and fatal autoimmunity later in life[15,16]. If haploinsufficiency is responsible for the defects in lymphocyte development observed in our patients, we would expect to see a similar effect in mice heterozygous for WT and null alleles (Bach2−). We found that Bach2− mice manifest reduced Bach2 mRNA (Fig. 6a) and protein expression (Fig. 6b) together with elevated Prdm1 mRNA (Supplementary Fig. 7a). There was no difference in the numbers of CD4+ and CD8+ T cells, B cells or plasma cells in unchallenged mice (Supplementary Figs. 7b and 7c) but Bach2− mice did have a small but significant reduction in FoxP3+ cells together with significant increases in CCR9+ and β7-integrin+ cells in CD4+ T cells (Figs. 6c, 6d and 6e). We next immunized WT and Bach2+/− mice with 4-hydroxy-3-nitrophenylacetyl hapten-conjugated chicken gamma globulin (NP-CGG) in alum and analyzed the splenic B cell response. Immunized Bach2+/− mice exhibited minimal induction of both IgG1 class switched-B220hiCD138− B cells and B220loCD138+ plasma cells compared to WT mice (Fig. 6f). The proportion of germinal center B220+Ki67+Bcl6+ B cells was also reduced in Bach2+/− mice (Fig. 6g), supporting a haploinsufficiency model.

Figure 6

Bach2 haploinsufficient mice have abnormal B cell differentiation and Treg cell numbers

(a) Expression of Bach2 mRNA in B cells of Bach2+/+ and Bach2+/– mice. (b) Bach2 protein expression in splenic naïve B cells of Bach2+/+ and Bach2+/– mice. Shown is a representative example (left) and cumulative quantification (mean ± sem) (right) from n=3 independent experiments. (c–e) Flow cytometry analysis of CD4+ splenocytes in Bach2+/+ and Bach2+/– mice showing percentage Foxp3+ (c), CCR9+ (d) and β7-integrin+ (e) cells. (f) IgM and IgG1 staining of B cells (upper panels) and plasma cells (lower panels) in splenocytes of Bach2+/+ and Bach2+/– mice 8 days following immunization with 4-Hydroxy-3-nitrophenylacetyl hapten-conjugated chicken gamma globulin (NP-CGG) in alum. (g) B220+Ki67+Bcl6+ germinal center B cells in splenocytes of Bach2+/+ and Bach2+/– mice 8 days after immunization with NP-CGG in alum. Shown in (c-f) are representative flow cytometry plots together with bar charts (mean ± sem). In vivo experiments were carried out twice. *p<0.05, **p<0.01, ***p<0.001 by t-test (a-b), one-way ANOVA (f) and Mann-Whitney U-test (all other panels).

Super-enhancer regulated genes associate with haploinsufficiency

Taken together, our data argue that the maintenance of a threshold concentration of BACH2 is crucial for proper immunoregulation. Mutations of other TFs have been reported to cause haploinsufficient disorders[35]. BACH2 expression is regulated in a complex manner and the BACH2 locus contains an archetypal SE (Fig. 7a)[12-14,16,36,37]. We therefore hypothesized that SE structure may be enriched among genes causing haploinsufficiency (HI) diseases. To this end, we compared genetic disorders mediated by HI (372 genes) versus autosomal recessive (AR) inheritance (259 genes) to haplosufficient (HS) genes (those where single allele deletions are inconsequential; 901 genes) (Ref. 38 and Online methods). To validate these three groups, we evaluated the probability of loss-of-function intolerance (pLI) score (as estimated by ExAc[39]), where a score of 0 predicts that loss of a single copy of the gene is well tolerated whereas a score of 1 predicts that loss of a single copy is poorly tolerated and likely to result in a disease. As expected, the median pLI score for our HI list was significantly higher than the others (median values of 0.86, 0.0005 and 0.004 for HI, HS and AR recessive genes, respectively) (Fig. 7b). Moreover, HI genes were substantially more likely to have SE architecture, as denoted by especially high acetylated histone H3 lysine 27 (H3K27Ac) signal, a hallmark of active enhancers[40] (Figs. 7c,d, Supplementary Fig. 8a and Supplementary Tables 3,4). In contrast, there was no difference in the frequency of typical enhancers between the three groups (Fig. 7d). We next compared the function of genes between the three groups and found that HI genes were more likely to encode transcription factors than genes associated with AR inheritance or HS genes (Fig. 7e, Supplementary Figs. 8b, c). To address any potential confounding abundance of transcription factor genes in SE, we also divided our list of HI genes into those that code for transcription factors and those that code for all other proteins and compared the frequency of SEs (Supplementary Fig. 8c). We found that even after discounting TF genes, haploinsufficiency disease-causing genes are heavily enriched for SE architecture compared to HS and AR genes (Supplementary Fig. 8d). We next asked whether SE-bearing genes have lesser tolerance to loss-of-function mutations and whether the ‘size’ of the SE correlates with this effect. We expanded our analysis to a collection of genes regulated by SEs from across more than 100 tissues (dbSuper database[41]) and observed both a striking increase in the probability of loss-of-function intolerance score with increasing SE signal size and a concomitant increase in the proportion of transcription factor genes (Fig. 7f). Thus, not only the presence of an SE but also its “size” correlates with likelihood of disease caused by haploinsufficiency.

Figure 7

Super-enhancer (SE)-regulated genes associate with haploinsufficiency

(a) The BACH2 locus has SE structures in multiple human immune cell types demarcated by H3K27Ac loading. Red fill denotes the presence of an SE in the BACH2 locus in a tissue. Source data are indicated. (b) Violin plots showing probability of loss of function intolerance scores in haplosufficient (HS), autosomal recessive (AR) and haploinsufficient (HI) gene sets. The white circles show median values. Source data: ExAc database[39]. (c) Number of HS, AR or HI genes with and without associated SE architecture in humans (see also supplementary Fig. 8a and supplementary Table 3). (d) Pie charts indicating the frequency of SE (upper panels) and typical enhancer (TE; lower panels) structures in HS (left), HI (middle) and AR (right) genes. (e) Gene ontology (GO) functional annotation enrichment in HI genes. Shown are enrichment scores (blue bars) and Benjamini p-values (in orange) for the top 5 most significantly enriched terms. (f) Median probability of loss of function intolerance (black line) against SE signal size; the percentage of genes that are transcription factors (TF, red line) against SE signal size is shown in the inset. For reference, the red line asymptotes to the expected level (mean percentage of genes in the human genome that are TFs is 7.5%). Source data: ExAc[39] and dbSuper[41] databases. (g) Pie charts indicating the percentage of HS or HI genes that have GWAS disease associations. p-values in d and g are Fisher exact tests; NS = non-significant; GWAS = genome-wide association study.

SE architecture was previously shown to associate with human disease loci in genome-wide association studies (GWAS)[14,36,42]. This is the case for BACH2[16,22], which was consistently in the top 1% of human SE genes by H3K27Ac SE signal intensity in naïve CD4+ T, naïve CD8+ T and B cells (Supplementary Figs. 8e–g). Based on the SE enrichment among HI genes, we next asked whether there would be general enrichment of GWAS “hits” in genes associated with haploinsufficient disease. In agreement with this hypothesis, we found that there was a highly significant enrichment of disease-associated SNVs within this gene set (Fig. 7g and Supplementary Table 5). To exclude gene size as a potential confounding factor, the analysis was repeated on subsets of genes of less than 50 kb and again we found more GWAS associations in genes associated with HI syndromes compared to HS genes (Supplementary Fig. 8h). Thus, HI genes are enriched for both SEs and GWAS “hits”.

Discussion

Adaptive immunity is critically dependent on appropriate differentiation and maturation of lymphocytes. Several complex differentiation steps are required to form mature cells that occupy specific niches and carry out defined roles within the immune system. Key to the regulation of lymphocyte differentiation is precise control over expression of many transcription factors (TFs) that form complex regulatory networks. The identification of both mice and humans with dramatic early onset stereotypical autoimmune disease associated with a homozygous loss of gene expression has led to the identification of many key regulatory TFs, most notably FoxP3, the master TF of Treg cells[27].

The reduction in the cost and time it takes to perform whole exome sequencing has allowed patients with no family history to be analysed for genetic mutations. Comparing patients’ DNA sequence with healthy parents identifies the appearance of de novo mutations that would otherwise be missed if a positive family history was required prior to any investigation. Using this strategy a number of heterozygous mutations associated with autoimmune diseases have recently been discovered.

BACH2 plays a major role in the regulation of the adaptive immune system. Its own expression is tightly regulated by the presence of a large super-enhancer region within the Bach2 locus[14]. The role of BACH2 has been elucidated by the investigation of BACH2-deficient mice that have a defect in B cell class switch recombination together with a deficiency of Treg cell differentiation. In mice, this combination results in a chronic variable immunodeficiency together with a late onset, but progressively fatal, autoimmune syndrome that includes inflammatory enteropathy and respiratory infiltrates[15]. In keeping with its place as an SE-associated gene, there is a link between single nucleotide polymorphisms within the BACH2 locus and a number of autoimmune/inflammatory diseases.

Herein we describe three patients from two families that have heterozygous mutations in BACH2. Two of the three presented with a history of early onset autoimmune gastrointestinal disease and the third presented later in life. All three have developed a chronic variable immunodeficiency characterized by recurrent respiratory tract infections associated with an inability to generate appropriate antibody responses to vaccination. Our findings support a role for human BACH2 as a key regulator of the human adaptive immune system critical to maintain Treg cell function and B cell maturation. BACH2-deficient mice exhibit accelerated T cell senescence[26,43] and, in keeping with this, T cells from our patients have a defect in cell proliferation associated with a progressive T cell lymphopenia. Many of the autoimmune phenomena in our patient with the L24P mutation have been successfully treated with corticosteroids although this has not reduced her chronic variable immunodeficiency nor her pneumonitis, which is of some concern as this is a key cause of early mortality in BACH2-deficient mice. The father with the E788K mutation developed bronchiectasis later in life. It remains to be seen whether the pneumonitis will be progressive in our L24P patient and result in chronic lung damage.

In the first family, the mutant BACH2 gene resulted in a protein that is predicted to be unable to dimerize and is unstable. In the second family, the mutant BACH2 protein again showed some evidence of a defect in stability but this was less dramatic, and we saw more evidence of a defect in the localisation of the protein with reduced nuclear localisation. We found little evidence that either mutant protein acted in a dominant negative manner. Thus we attribute the clinical phenotype to BACH2 haploinsufficiency and this conclusion was consistent with our findings and previous reports[44] that Bach2+/− heterozygote have defects in CSR antibody responses.

Mammalian cells contain tens of thousands of gene enhancer sites that cluster in large numbers around a select subset of genes that make up some 5–10% of the total human genome. These clusters are collectively known as SEs. GWAS mutations tend to associate with these gene loci but the significance of this remains unclear. Previous work would suggest that SE genes code for proteins whose function is highly dependent on transcription, small changes of which would lead to significant changes in cell development. From this we hypothesise that SE genes would be susceptible to gene dosage effects in patients with heterozygous mutations. We conclude that the relationship between GWAS studies and SE regulated genes occurs not simply because these genes transcribe proteins that are important per se but because small changes in the expression of SE genes result in large functional changes in the affected cells.

In summary, we describe a new disorder, BACH2-related immunodeficiency and autoimmunity (BRIDA) due to heterozygous mutations in BACH2. We found that the mechanism of disease is BACH2 haploinsufficiency and that BACH2 is a prototype haploinsufficiency gene exhibiting SE architecture. Given the prevalence of heterozygous variants in non-consanguineous human genomes[45], it is difficult to predict which ones cause disease. We demonstrate that haploinsufficiency diseases are associated with heterozygous mutations in SE-regulated genes. As SEs allow complex regulation of gene transcription, we conclude that haploinsufficiency genes are carefully regulated due to their SE association and that small changes in their expression level can potentially lead to amplified changes in their associated network, especially for TF genes, resulting in significant pathology. Thus, SE-regulated genes should be more significantly prioritized when interpreting heterozygous variants discovered on whole exome/genome sequencing.

Online Methods

Ethics approvals

Patients and their relatives provided written informed consent and were investigated under National Institute of Allergy and Infectious Diseases (NIAID) Institutional Review Board–approved research protocols 89-I-0158 and 06-I-0015, West London Research Ethics Committee approval (Ethics Protocol Reference Number 11/LO/0883) and Oxford IBD cohort study (monogenic IBD subproject). All animal studies were performed according to National Institutes of Health guidelines for the use and care of live mice and were approved by the Institutional Animal Care and Use Committee of National Institute of Arthritis, Musculoskeletal and Skin Diseases (Protocol number A014-03-02).

Histology and Immunohistochemistry

A colonic biopsy was performed on patient A.II.1 at the time of her presentation, aged 19 years. The biopsy was stained with Hematoxylin and Eosin stain and reviewed by pathologists at the Hammersmith hospital, London, UK. Immunohistochemical staining of formalin-fixed paraffin-embedded (FFPE) sections was performed on patient and tissue-matched FFPE sections from healthy control donors as well as age-matched donors diagnosed with classical Crohn’s Disease (provided by the Oxford Centre of Histopathology Research and the Oxford Gastrointestinal Illness Biobank) using antibodies to FOXP3 (Abcam; 236A/E7) followed by TSA amplification (PerkinElmer) and CD3 (Dako; F7.2.38) followed by Alexa Flour 488-conjugated goat anti-mouse IgG (LifeTechnologies). Nuclei were stained using Vectashield antifade mounting medium with DAPI (Vector Laboratories) and slides were examined with a Zeiss LSM510 inverted confocal microscope. ImageJ (ImageJ) and Photoshop (Adobe) were used for the processing and presentation of the images.

Antibodies, cell lines and media

The following antibodies and reagents were used in the study: anti-human BACH2 (ab83364) was purchased from Abcam, anti-human CD19 (HIB19), anti-human CD24 (ML5), anti-mouse CD3 (145-2C11), anti-mouse CD8 (53-6.7), anti-mouse CCR9 (9B1), anti-human-CCR9 (LO53E8), anti-human/mouse β7-integrin (FIB504) (all BioLegend), anti-human CD4 (OKT4), anti-human CD25 (2A3), anti-human CD27 (M-T271), anti-human CD38 (HB-7), anti-human IgG (GI8-145), human Fc Block, anti-mouse CD4 (RM4-5), anti-mouse CD25 (7D4), anti-mouse CD44 (IM7), anti-mouse CD62L (MEL-14), anti-mouse CD138 (281-2), anti-mouse B220 (RA3-6B2), anti-mouse CXCR5 (2G8), anti-mouse IgG1 (A85-1), anti-mouse IgM (R6-60.2), streptavidin-APC, streptavidin-FITC (all BD), anti-human CD3 (OKT3), anti-human CD8 (RPA-T8), anti-human CD38 (HB7), anti-human-CD127 (eBioRDR5), anti-human T-bet (eBio4B10), anti-human FoxP3 (PCH101), anti-mouse CD25 (BC61.5), anti-mouse CD127 (A7R34), anti-mouse GL7 (GL-7), anti-mouse Fas (15A7), anti-mouse NKp46 (29AI.4), anti-mouse IgD (11–26), anti-mouse IgM (11/41), anti-mouse PD1 (J43), anti-mouse GITR (DTA-1), anti-mouse Foxp3 (FJK-16s), anti-Thy1.1 (HIS51) (all eBioscience), mouse anti-FLAG M2 (Sigma) and goat anti-rabbit-IgG-AlexaFluor488 (A-11034) (LifeTechnologies). Live-Dead Flixable Aqua Dead Cell stain was purchased from Thermofisher (Boston, USA). Raji, Ramos and HEK293T cell lines were purchased from ATCC. Unless specified, human cells and cell lines were maintained in RPMI 1640 supplemented with 2mM L-glutamine, penicillin/streptomycin (100 IU/mL and 100 ug/mL respectively; all from LifeTechnologies) and 10% FBS (Atlanta Biologicals). Mouse cells were cultured in identical medium supplemented in addition with 2 mM β-mercaptoethanol (Sigma Aldrich). HEK293T cells were maintained in DMEM (LifeTechnologies) supplemented as with human cell culture medium.

Mice

C57BL/6J mice were purchased from The Jackson Laboratory. Bach2−/− and Bach2+/− mice were generated and housed as previously described[15]. Blimp1-YFP BAC transgenic mice have been previously described[49]. No statistical methods were used to predetermine sample size.

Cell isolation and culture

Human PBMC were isolated from patient and healthy donor blood by density gradient centrifugation using Ficoll (GE Healthcare) followed by lysis of red blood cells with RBC lysis buffer (eBioscience). CD4+ T cells, naïve CD4+ T cells and naïve B cells were purified from PBMC by negative selection using human CD4 T cell isolation kit, human naïve CD4 T cell isolation kit II and human naïve B cell isolation kit II, respectively (all MiltenyiBiotec) according to manufacturer’s instructions. B-cell subsets were sort purified by FACSAria (BD Immunocytometry Systems, San Jose, CA, USA.) using APC conjugated anti-CD19 (BioLegend, San Diego, CA, USA), PE conjugated anti-CD27 (BD Biosciences, San Jose, CA, USA.), PerCP-Cy5.5 conjugated anti-IgM (BD Biosciences). Naïve B cells were defined as CD19+CD27-IgM+ B cells with a purity typically more than >98%[50].

CD4+ T cells from spleens and lymph nodes of 6- to 8-week-old mice were purified by negative selection and magnetic separation (Miltenyi), followed by sorting of naive CD4+CD25−CD62L+CD44− population with a FACSAria II. Naïve Blimp1-YFP CD4+ T cells were activated for 3d by plate-bound anti-CD3 (2C11; BioXCell) plus CD28 (37.51; BioXCell), each at a concentration of 10 μg/ml in medium. Cells were stimulated in the presence of mouse IL-12 (20ng/ml) and anti-mouse IL-4 (10 μg/ml) (Th1 conditions) (both from R&D systems) for 3 days, then split into fresh uncoated plates and supplemented with fresh medium and 100 IU/mL human IL-2 (NIH/NCI BRB Preclinical Repository).

B cell cultures and induction of class-switch recombination

Purified naïve B cells were cultured in RPMI 1640 containing L-glutamine (Sigma Aldrich, St. Louis, MO, USA), 10% fetal bovine serum (Sigma Aldrich), 10 mM HEPES (pH 7.4; Sigma-Aldrich), 0.1 mM nonessential amino-acid solution (Sigma- Aldrich), 1 mM sodium pyruvate and 40 μg/ml apo-transferrin (Sigma-Aldrich) and supplemented with 60 μg/ml penicillin and 100 μg/ml streptomycin. To induce class switch recombination, recombinant human CD40L (1μg/ml; R&D Systems, Minneapolis, MN, USA), Fab fragment anti-human IgM (Jackson Immunoresearch, West Grove, PA, USA), IL-2 (100 IU/ml; PeproTech) and IL-21 (50 ng/ml; PeproTech, Rocky Hill, NJ, USA) were added at the beginning of the culture. Cells were cultured in 96-well round bottom well plates (NuncTM, Roskilde, Denmark) for 5 days. Culture supernatants were collected for ELISA at the end of the culture.

IgG and IGA ELISA

IgG and IgA secretion was determined with the Ready-set-go total IgG and IgA kits (Thermofisher) according to manufacturer protocols. Absorbance was read at 450 nm within 30 minutes of stopping of the reaction. The sensitivities and linear ranges were obtained using the provided standard immunoglobulin.

Whole exome sequencing

DNA was extracted from EDTA blood using Maxwell 16 Blood DNA Purification Kit (Promega) or PBMC using DNeasy Blood & Tissue Kit (Qiagen). Total of 3 ug of DNA were sheared using E220 focused sonicator (Covaris) and exome libraries were generated using the SureSelect Human All Exon Kits (Agilent) according to manufactures’ protocol. The quality of generated libraries was inspected using Agilent High Sensitivity DNA Kit (Agilent) and quantified using qPCR kit (Agilent). Samples were sequenced on Illumina HiSeq2000 (Illumina) generating 100 bp paired end reads. Sequences were aligned to a human reference genome GRCh37 using bwa v 0.6.1 with default parameters[51]. Variant calling (Single nucleotide variants and indels) was performed using GATK v.2[52] and variants were annotated using Annovar[53]. An in-house custom analysis pipeline was used to filter and prioritize variants based on the likely genetic models and clinical pedigree for patients.

Sanger sequencing

DNA samples were extracted from blood or saliva using Maxwell 16 Blood DNA Purification Kit (Promega) and Oragene DNA (OG500) (Oragene), respectively. The candidate mutations in affected and unaffected individuals of both families were validated using BigDye Terminator Sequencing kit (Life technologies) and sequenced on ABI3730xl genetic analyser (Applied Biosystems). PCR primer sequences are available on request.

Flow cytometry

All flow cytometry was carried out in a final staining volume of 100–200 μL, with data acquisition on an LSR II, LSRFortessa or FACSVerse (all BD Biosciences) within 24 h. Appropriate internal controls, isotype controls and Fluorescence Minus One (FMO) controls were used to assign gates. Rat anti-mouse CD16/CD32 (clone 2.4G2; BD) was used for Fc blockade in mouse flow cytometry experiments. FACS data were analysed using FlowJo (Tree Star Inc., Oregon). For Intracellular staining, BD Cytofix/Cytoperm™ plus Fixation/Permeabilization Solution Kit was used according to manufacturer’s instructions. For cytokine staining, 4h re-stimulation with PMA (50ng/mL) and ionomycin (1mM) (both Sigma) in the presence of Brefeldin A (GolgiPlug™ (BD) was carried out prior to fixation and permeabilization. Foxp3 staining was carried out using the kit from eBiosciences as per manufacturer’s instructions. Relative FoxP3 and BACH2 levels were calculated by dividing the geometric mean fluorescence intensity (MFI) of patient cells by that of matched healthy control in each run. For assessment of cell proliferation by flow cytometry, T cells were stained with CellTrace™ Violet as per manufacturer’s instructions followed by culture in the presence of anti-CD3 and anti-CD28 (1ug/mL of each) (clones HIT3α and CD28.2, respectively, both from Biolegend) for five days before live/dead staining and data acquisition.

In vivo class switch assay

8–10 week old Bach2+/− heterozygous and Bach2+/+ WT mice were i.p. injected with 50 ug of NP-conjugated chicken gamma globulin (NP-CGG)(Biosearch technologies) in 1:1 Alum (Thermo Scientific) (vol:vol). Spleens were harvested after 8 days and single cell suspensions were made by passing the cells through 40 μm strainer followed by surface staining and flow cytometry as described above.

Quantitative RT-PCR

Total RNA was extracted using TRIzol reagent (Invitrogen) and treated with DNAseI (Qiagen). RNA was reverse transcribed to cDNA using iScriptcDNA synthesis kit (Bio-Rad) following the manufacturer’s instructions. Quantitative real-time PCR (qRT-PCR) was performed in triplicate using Taqman® Universal PCR Master Mix (Applied Biosystems) in total reaction volumes of 20 μL and thermocycled in a CFX284 TouchTM Real-Time PCR Detection System (Bio-Rad). The following Taqman gene-specific primer probes were purchased from Applied Biosystems: human BACH2 (Hs00222364_m1), PRDM1 (Hs00153357_m1), ACTB (Hs99999903_m1) and 18S (Hs99999901_s1), mouse Bach2 (Mm00464379_m1), Prdm1 (Mm00476128_m1), Bcl6 (Mm00477633_m1) and Actb (Mm00607939_s1). Cycle threshold (Ct) values were exported and normalized against the control probe using the 2−ΔCt method and reported as expression relative to a control condition.

Silencing of BACH2 and BACH2 over-expression

5 × 106 PBMCs per sample were nucleofected with 300 nM DsiRNA negative control or predesigned BACH2 DsiRNA (both TriFECTa®, Integrated DNA technologies) using Amaxa human T cell nucleofector kit (Program-U014, Lonza), according to manufacturer’s instructions. 24 hours after nucleofection cells were labeled with CellTrace violet cell proliferation kit (Thermo) and rested for 6 hours in culture before activation of 1 × 105 cells per 96-well plate with plate bound anti-CD3 (1ug/ml, clone HIT3α) and anti-CD28 (1ug/ml, clone CD28.2 both BioLegend). Cells were surface stained and proliferation was analyzed by flow cytometry after 5 days.

Naïve B cells or CD4+ T cells were nucleofected with 2 uM MISSION universal negative control siRNA (Sigma) or BACH2 siRNA (Hs01_00214431, Sigma) using P3 primary cell 96-well Nucleofector™ kit (Lonza) according to manufacturer’s instructions. Cells were cultured for 24h at 37°C in the presence of 100 ng/ml human IL-7 before activation for class-switch recombination as described earlier.

5×106 blasting human CD4+ T cells or were mixed with 2–5μg of either BACH2 or eGFP mRNA (TriLink) in 50 μl of HyClone™ MaxCyte® buffer and electroporated in OC-100 PA electroporation chamber using MaxCyte® GT Instrument (Program T-02). After electroporation cells were incubated 20 min at 37°C in electroporation buffer in 96-well plates and after that transferred to 12-well plates in complete RPMI containing 100 IU/ml human IL-2. PRDM1 expression was analyzed after 24 – 48h by qPCR.

Plasmid DNA and point mutagenesis

Wild-type Bach2 cDNA expression vectors pMSCV-IRES-GFP (pMIGR1-Bach2) and pMSCV-IRES-Thy1.1 DEST (pMIT-Bach2) have been described previously[15]. Gene synthesis was performed to achieve an N-terminal fusion of Flag and HA sequences preceded by a methionine translation initiation codon (MDYKDDDDK and MYPYDVPDYA, respectively) to the wild-type BACH2 open reading frame. Synthesized DNA was subcloned into pMIT to generate pMIT-Flag-BACH2 and pMIT-HA-BACH2. Point mutagenesis to introduce the Bach2 (Bach2L24P) and Bach2 (Bach2E786K) mutations were carried out using Agilent QuickChange II XL Site-directed mutagenesis kit (Agilent Technologies) according to the manufacturer’s instructions, with the following primer pairs: Bach2: forward, 5′-CATTGAGGCCCAGGGGGATGTTGGCACAG-3′ and reverse, 5′-CTGTGCCAACATCCCCCTGGGCCTCAATG-3′; Bach2: forward, 5′-AGAGGTACAATTCTTAGAGGTGTTGCTGGGCACC-3′ and reverse, 5′-GGTGCCCAGCAACACCTCTAAGAATTGTACCTCT-3′.

Transfection and production of retrovirus

Transfection was carried out in antibiotic-free medium using lipofectamine LTX and Plus reagent (Invitrogen). Medium was replaced 7 h later. For production of retrovirus, payload retroviral plasmid was co-transfected with pCL-Eco helper virus plasmid as previously described[54]. Transfected cells were harvested and viral supernatant collected 48 h after transfection.

Retrovirus transduction

Prdm1-YFP BAC Tg CD4+ T cells were activated for 24 h with plate-bound anti-CD3 + anti-CD28. Activated cells were transduced with supernatants containing retrovirus encoding Thy1.1 alone (EV) or together with mouse Bach2 or mutant mouse Bach2 conforming to the L24P or E786K mutation, in the presence of polybrene (4 μg/ml) by centrifugation at 2200 rpm for 50 min at 22°C. Medium was replaced afterwards with fresh culture medium and cells harvested 48 h after transduction.

Western blotting and FLAG immunoprecipitation (IP)

Clarified protein extracts were prepared by lysis of cell pellets in Pierce™ IP lysis buffer (ThermoScientific) containing 1x cOmplete Protease Inhibitor cocktail (Roche). Protein concentrations were quantified (Micro BCA protein assay kit (ThermoScientific) to ensure equal loading. Proteins were resolved by SDS-PAGE on Any kD™Criterion™ TGX™ gels (Bio-Rad) and electrotransferred onto nitrocellulose membranes (Bio-Rad). Immunoblotting was performed using rabbit anti-BACH2 (Abcam), mouse anti-FLAG® M2 (Sigma), mouse anti-Hsp70 (SantaCruz Biotechnology) and goat anti-mouse IRDye® 800CW (Li-Cor) following by scanning on an Odyssey imaging system (Li-Cor Biotechnology) or anti-HA-HRP for development using SuperSignal® West Pico Chemiluminescent Substrate (ThermoScientific) and imaging on a ChemiDocTM MP Imaging system (Bio-Rad). FLAG IP was carried out using EZview™ Red Anti-FLAG® M2 Affinity gel (Sigma) according to manufacturer’s instructions followed by elution using 3X FLAG® Peptide (Sigma).

Confocal microscopy

HEK293T cells (ATCC) were cultured and transfected on poly-L-lysine (Sigma) coated round cover slips. Primary PBMC were spun onto poly-L-lysine coated cover slides using a Cytospin3 centrifuge (Shandon). Cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% TritonX-100 in TBS, blocked with TBS containing 5% horse serum and 0.01% NaN3 and stained with primary antibodies for 1–2 h at room temperature. Staining with secondary antibodies was performed for 40 min at room temperature in the dark together with 1:10000 of Hoechst. Cells were mounted with ProLong Diamond antifade mountant (LifeTechnologies). The following antibodies and dilutions were used for confocal microscopy: 1:100 mouse anti-FLAG M2 (Sigma), 1:25 rabbit anti-human BACH2 (Abcam), 1:500 goat anti-mouse IgG-AlexaFluor 488 (LifeTechnologies), 1:500 goat anti-rabbit IgG-AlexaFluor 568 (LifeTechnologies). Confocal microscopy of immunostained cells was performed using Leica SP8 inverted 5 channel confocal microscope equipped with a motorized stage and ultra-sensitive hybrid detectors (Leica Microsystems). The following laser lines were used: diode for 405 nm, Argon for 488 nm, and DPSS for 561 nm excitation wavelengths. Microscope configuration was set up for 3D (x, y, z) sequential scanning using 63x objective, and z stacks of 0.3 μm optical slices (total of 10–15 μm) were collected. For statistical analysis of BACH2 localization, tiled images of transfected cell layer at total cell number of 200 cells per field were collected. Images were processed using Imaris (Bitplane, Switzerland) and Huygens (Scientific Volume Imaging, Netherlands) software. The number of cells containing protein aggregates was determined from at least 3 tiled images. Pearson’s Correlation Coefficients was calculated using Imaris.

Recombinant protein expression and purification of BACH2 and variants

Synthetic genes with codons optimized for E. coli expression were from Genscript. BL21(DE3) cells with pET 28 vectors were grown in a fermenter and cells were broken and initially processed as previously described[55]. The proteins: full-length human p.BACH21–841 and p.L24P variant; murine p.Bach21–133 and murine p.Bach21–133 L24P all contained an N-terminal his-tag to facilitate purification (NB The sequence difference between human p.BACH21–133 and murine p.Bach21–133 is at one position, amino acid 8, which is Asp in human and Ala in murine). Human WT p.BACH21–841 was extracted from cell lysate with 100 mM sodium bicarbonate, pH 9.5 containing 2 M urea and the L24P variant with 8 M guanidine-HCl. WT proteins were expressed as a soluble protein but L24P variants were insoluble and extracted with 8M guanidine-HCl. Proteins were purified using a combination of Ni-chelate and size exclusion chromatographies using Ni-chelate Sepharose and Sephadex S200 (both from GE Healthcare). The L24P variants were folded by dialysis against 4 M urea and then stepped through lower concentrations until the urea was removed. DTT was present in all buffers to keep proteins reduced.

Analytical ultracentrifugation

A Beckman Optima XL-I analytical ultracentrifuge, absorption optics, an An-60 Ti rotor and standard double-sector centerpiece cells were used. Equilibrium measurements were at 20°C and concentration profiles recorded after 16 h at 20,000 rpm (BACH2133) or 10,000 rpm (BACH2841). Baselines were established by over-speeding at 45,000 rpm for 3 h. Data (the average of eight scans collected using a radial step size of 0.001 cm) were analyzed using the standard Optima XL-I data analysis software. Sedimentation velocity experiments were performed at 40,000 rpm with scans recorded every 6 minutes for 3 h. Protein partial specific volumes, calculated from the amino acid compositions, and solvent densities were estimated using the program SEDNTERP (http://www.rasmb.bbri.org/).

Protein concentrations

Estimated from amino compositions: absorbencies at 280 nm of 1 mg/ml of mBach2133 and hBACH2841 of 0.69 and 0.41 respectively, were used.

Analysis of mutations

Conservation scores for mutated sites (PhyloP, PhastCons and GERP) were obtained from the UCSC genome browser (GRCh37/hg19). Polyphen2, SIFT, LRT, MutationAssessor Functional Impact, MutationTaster and CADD scaled scores were derived using dbNSFP, as described[56,57]. The CADD-based mutation significance cutoff (MSC) at 99% confidence interval (CI) was calculated as described[58].

Curation of haploinsufficient and autosomal recessive disease genes and haplosufficient genes

Haploinsufficient genes were retrieved from PubMed and Online Mendelian Inheritance in Man (OMIM), using the semi-automated method of Dang et al.[59]. Searches were restricted to the period from 12th November 2007 to the 25th of October 2015 and merged with the existing dataset prior to 12th November 2007[59]. All retrieved items were manually curated by two independent physicians, to ensure that only true positives (genes causing haploinsufficient disease) were kept for further analysis. Autosomal recessive genes were identified by downloading the OMIM database and extracting all entries inherited in an autosomal recessive fashion. Haplosufficient genes were obtained from a list of high-confident predictions (Pr(HI)<0.05) in Huang et al.[38]. The predictions were further screened by removing those that match HI genes (3 genes in total). Functional annotation analysis for genes was carried out using Gene Ontology enrichment analysis via DAVID[60,61] and Ingenuity Pathway Analysis (Qiagen).

Super-enhancer (SE) structures

Sequencing data were downloaded from GEO. URLs for data used in this manuscript are listed in table S4. Reads were mapped to hg19 with bowtie0.12.8[62]. The HOMER suite of programs[63] was used to call super enhancers and typical enhancers following the guidelines presented by Whyte et al[64]. Enhancers were assigned to the closest genes with PAPST[65]. K27Ac signal graphs were created using data generated with HOMER.

We obtained estimated probabilities of human gene intolerance to loss of function mutations from the EXAC database[39] (n=18,225 genes, release 0.3.1: ftp://ftp.broadinstitute.org/pub/ExAC_release/release0.3.1/functional_gene_constraint/fordist_cleaned_exac_r03_march16_z_pli_rec_null_data.txt; accessed 2016 Aug 18). We obtained super-enhancers calls (n=65,950 super-enhancers from 99 tissues/cells) from dbSuper[41] (http://bioinfo.au.tsinghua.edu.cn/dbsuper/data/bed/hg19/all_hg19_bed.bed; accessed 2016 Aug 17). These super-enhancers were ranked according to signal intensity within each cell/tissue. We assigned each super-enhancer to the closest protein-coding gene promoter within 50kb[66] (ENSEMBL GRCh37.75; http://ftp.ensembl.org/pub/release-75/gtf/homo_sapiens/Homo_sapiens.GRCh37.75.gtf.gz; accessed 2016 Aug 18) using BEDTOOLS[67]. If a gene was near multiple super-enhancers, we assigned it the highest observed super-enhancer rank. Finally, to explore the relationship between pLI score and enhancer architecture, we combined this gene-centric table of super-enhancer ranks with the EXAC pLI table. Specifically, we determined the median pLI score observed with varying thresholds of super-enhancer rank. To explore the specific role of transcription factors, we obtained a comprehensive list of human transcription factors from AnimalTFDB[68] (http://www.bioguo.org/AnimalTFDB/download/Homo_sapiens_TF_EnsemblID.txt; accessed 2016 Sep 14). We then determined the fraction of transcription factors with varying thresholds of super-enhancer rank. We created the plots using the R project.

GWAS data (gwas_catalog_v1.0) were downloaded from http://www.ebi.ac.uk/gwas/docs/downloads. The hg38 SNP coordinates were converted to hg19 coordinates with liftOver from the UCSC Genome Browser (http://hgdownload.cse.ucsc.edu/downloads.html#source_downloads). Genomic region overlapping analyses were conducted with BEDTools[67]. A SNP was assigned to a gene if its co-ordinate was within the gene body (transcription start to transcription end, as defined by RefSeq hg19). HS and HI genes with GWAS associations are listed in table S5. Fisher exact tests were carried out using R3.2.0. Data extraction, data reformatting, and data preparation for analysis were all facilitated with customized scripts of Bash, Python, and R.

Data analysis and visualization

Data were analyzed using Microsoft Excel and GraphPad Prism (Graph Pad Software) and visualized using CLC Main Workbench 7 (CLCbio, Qiagen) and DataGraph 3.2 (Visual Data Tools, Inc). Molecular graphics and analyses were performed with the UCSF Chimera package. Chimera is developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco (supported by NIGMS P41-GM103311). Statistical analyses were performed using appropriate parametric and non-parametric tests as appropriate. Multiple datasets were compared by repeated measures ANOVA. Statistical analysis of data in contingency tables was carried out using the Fisher exact test. Two-tailed p-values of <0.05 were considered statistically significant throughout.

Data availability

The data that support the findings of this study are available from the corresponding author upon request.