Microbial resistance involves innate and adaptive immune responses that prevent, diminish or clear infection, often causing collateral damage to host tissues and increased energy demands. Studies in plants1 and animals2,3 have shown that in some circumstances it can be more efficient for a host to tolerate microbes rather than resist them. Microbial tolerance involves homeostatic mechanisms to raise thresholds for initiating immune responses, to physically separate microbes from host immune receptors and to repair damage caused directly by microbes or by collateral inflammation3,4,5,6. The genetic means by which microbial resistance and tolerance are balanced remains incompletely understood. Population genetic modeling predicts that resistance traits favor host polymorphism and microbial evasion, whereas microbial tolerance traits tend toward fixation in hosts and microbial mutualism7. High mortality in indigenous human populations of Oceania and the Americas exposed to smallpox demonstrates how a tolerated pathogen in its adapted host can cause devastating disease when introduced into non-adapted populations8,9. An example in animals can be seen for European rabbits exposed to myxoma poxvirus, previously endemic to South American rabbits8. These cases illustrate the importance of fine-tuning microbial immunity and tolerance during coevolution of hosts and microbes, although little is known about the molecular pathways that underpin this.

A binary perspective on the importance of balancing microbial immunity and tolerance comes from Mendelian gene variants in mice and humans that completely inactivate one or both alleles of genes such as CTLA4, IL10, FOXP3 and TNFAIP3 (refs. 10,11,12). These variants cause severe pediatric autoimmune or inflammatory disease, particularly at mucosal barriers where large microbial populations are normally tolerated, such as the microbial burden that drives inflammatory pathology in mouse Tnfaip3 deficiency6. Genome-wide and candidate association studies in humans have implicated single-nucleotide polymorphisms at or near the TNFAIP3 locus in susceptibility to autoimmune disease13. In contrast to these disease-associated traits, few examples exist of beneficial genetic adjustments that decrease microbial tolerance in favor of heightened immunity.

A20, encoded by the TNFAIP3 gene, promotes microbial tolerance as a negative regulator of nuclear factor (NF)-κB signaling: an evolutionarily ancient and central pathway for activating innate and adaptive immune responses13. A20 has multiple domains with inhibitory activity against NF-κB, primarily preventing activation of the central IκB kinase (IKK) by upstream proteins RIPK1, TRAF6 and NEMO. The A20 ovarian tumor (OTU) domain has deubiquitinating (DUB) protease activity that cleaves activating K63-linked ubiquitin chains from RIPKI, TRAF6 and NEMO14,15,16. The A20 zinc finger 7 domain (ZnF7) binds linear polyubiquitin to suppress IKK activation, whereas ZnF4 promotes ligation of K48-linked ubiquitin chains to RIPK1, triggering RIPK1 proteolysis14,17. A20 feedback inhibition is induced at two levels: NF-κB proteins directly induce TNFAIP3 mRNA, and the inhibitory activities of A20 protein are enhanced by IKKβ-mediated serine phosphorylation near the ZnF domains, notably at S381 (refs. 18,19).

The role of the A20 OTU domain nevertheless remains enigmatic. The ZnF domains alone are sufficient for NF-κB inhibitory function in cell-based studies20,21, and mice homozygous for Tnfaip3 missense variants creating a catalytically inactive OTU domain have little15,18 or no20 evidence of excessive NF-κB signaling. Here we demonstrate that anatomically modern human, archaic Denisovan and mouse alleles encoding missense variants in the OTU domain modulate A20 phosphorylation by IKK, serving as a genetically tunable element with profound effects on the balance between microbial tolerance and immunity, and provide evidence of introgression to high frequencies during human history.


An A20 OTU domain allele acquired from Denisovans

Three convergent sequencing studies led us to identify a unique functional class of A20 missense alleles affecting the OTU domain, distinct from the ubiquitin protease catalytic site (Fig. 1a,b and Supplementary Fig. 1a). The allele with the most subtle effect, comprising T108A (rs376205580) and I207L (rs141807543) missense substitutions in cis, was identified by whole-genome sequencing in 4 of 85 families in Sydney, Australia. The majority of individuals in our cohort carrying the T108A;I207L allele were healthy family members of Māori or Pacific Islander ancestry. This allele was rare in a public variant collection (gnomAD r2.0.2) but most frequent among individuals with unassigned ancestry (Supplementary Fig. 1b).

Fig. 1: Denisovan, modern human and mouse alleles substituting non-catalytic residues of the A20 OTU domain.
figure 1

a, Schematic of the A20 protein encoded by the TNFAIP3 gene (CCDS5187), showing exons (EX1–EX9), OTU, ZnF1–ZnF7, the OTU catalytic residue (C103), the S381 phosphorylation site and the missense substitutions studied here (red lines). b, Location of amino acid residues in sliced structure 5LRX41 of the A20 OTU domain (pink) complexed with ubiquitin (blue). Structures for mouse wild-type 5DQ6 and human I325N OTU 5V3P domains are superposed in gray and orange, respectively. c, Read data of a high-coverage Denisovan genome27 across TNFAIP3 codons 108 and 207. d, Imputed frequency of the Denisovan TNFAIP3 haplotype in genotype array data from 514 individuals from indigenous populations across Island Southeast Asia and Oceania30 and directly observed in exome data from 72 Martu indigenous Australians32. Different symbols group populations from Sunda, Wallacea, Sahul and Eastern Oceania. e, Imputed Denisovan TNFAIP3 haplotype frequency from the populations described in d (filled circles) versus a surrogate estimate of Papuan ancestry (principal component 1; Supplementary Fig. 2a,b). Gray shading indicates the 95% confidence interval of a linear regression line. f, RT–PCR analysis of CXCL2, TNF and ICAM1 expression in PBMCs from individuals heterozygous (green circles) or homozygous (purple circle) for the Denisovan TNFAIP3 haplotype (hap) and 15 healthy control donors (HCs; blue squares), comparing induction kinetics following stimulation with recombinant human TNF for 0, 2 or 4 h. Fold change was determined by comparing stimulated values against those for nonstimulated PBMC preparations within the same group. Significance between groups was determined by area under the curve (AUC) analysis. Summary data are shown as mean ± s.d. and P values were determined by a two-tailed Student’s t-test: **P < 0.01; ***P < 0.001.

We next traced the global distribution of the T108A;I207L allele using the Simons Genome Diversity Project dataset that includes genome sequence data on 300 individuals representing 142 diverse populations22. An analysis of 279 individuals revealed high frequencies of the T108A;I207L allele, ranging from 25–75%, among people of Island Southeast Asia and Oceania, but an absence of the allele elsewhere in the world (Supplementary Fig. 1c and Supplementary Table 1).

Unlike people from Africa or Eurasia, people in Island Southeast Asia and Oceania acquired up to 5% of their genome from Denisovans: archaic hominins who interbred with modern humans ~50,000 years ago, migrating through Asia to settle the continent of Sahul (now Papua New Guinea and Australia)23,24,25,26. Analysis of the high-coverage genome of a Denisovan finger phalanx from a cave in the Altai Mountains of Siberia27 revealed homozygosity for the T108A;I207L allele (Fig. 1c). Both variants were absent from the genome of a Neanderthal who had inhabited the same cave28 (Supplementary Table 2), suggesting that T108A and I207L arose after divergence of the Denisovan and Neanderthal lineages 170,000–700,000 years ago27.

Multiple Denisovan-derived genomic regions, including one encompassing TNFAIP3, bear strong signatures of introgression in people from Papua25,26,29. To investigate the geographical distribution of the T108A;I207L allele in finer detail, we made use of genome-wide array data on 514 individuals from indigenous populations across Island Southeast Asia and Oceania30. By phasing and imputing haplotypes across the TNFAIP3 locus using these genotype array data, and validating our imputation in 481 modern and archaic whole genomes (Supplementary Tables 3 and 4), we found evidence that the Denisovan TNFAIP3 haplotype was present at high frequencies in modern human populations east of the Wallace Line, a 50-million-year-old faunal boundary separating organisms of Asiatic and Australian origin via deep water channels between the two continental shelves (Fig. 1d,e and Supplementary Fig. 2a–d)31. The T108A;I207L allele was absent in almost all populations west of the Wallace Line, yet accounted for 31–48% of the TNFAIP3 alleles in Wallacean people of Sumba, Flores, Lembata, Alor and Timor, 32–100% of alleles in people from Papua and 15–45% of alleles in people from Vanuatu, Tonga and Samoa (Fig. 1d and Supplementary Table 4).

We observed the same T108A;I207L TNFAIP3 haplotype in 31/144 (22%) exome-sequenced alleles in Martu Indigenous Australians from the Pilbara region of Western Australia32, implying that haplotype enrichment occurred before the isolation of Indigenous Australian and Papuan populations (Fig. 1d). The high frequency in Polynesia indicates that the Denisovan TNFAIP3 haplotype was retained after admixture with Austronesian farming populations, expanding from mainland Asia starting ~4,000 years ago33 in eastern Indonesia and ~3,000 years ago in the Southwest Pacific34.

Of the two coding variants, T108A was not predicted in silico to alter A20 function (Phred-scaled combined annotation-dependent depletion (CADD) score of 0.002) and was present in other vertebrate reference genomes, including that of mouse. In contrast, I207 was invariant across most jawed vertebrates and the I207L variant was predicted to be the most deleterious substitution across the Denisovan TNFAIP3 haplotype (Phred-scaled CADD score of 23.2; Supplementary Figs. 1a and 2e). Aside from the two missense variants in the introgressed Denisovan TNFAIP3 haplotype, other noncoding variants could conceivably modulate transcription. Deep sequencing of cDNA nevertheless revealed equal amounts of TNFAIP3 mRNA from Denisovan and modern human alleles in leukocytes obtained from heterozygous individuals, with or without tumor necrosis factor (TNF) stimulation (Supplementary Fig. 2f). The Denisovan TNFAIP3 haplotype was associated with heightened expression of NF-κB-induced transcripts, including CXCL2, TNF and ICAM1, in TNF-stimulated peripheral blood mononuclear cells (PBMCs) from carriers of the T108A;I207L allele (Fig. 1f, Supplementary Fig. 2g and Supplementary Table 5). The results above identify an ancient pair of substitutions in the A20 OTU domain as apparently beneficial in human history.

A mouse A20 OTU variant confers heightened immunity

Further evidence that A20 OTU domain variants could be beneficial came from another OTU substitution, I325N, identified in a genome-wide mouse mutagenesis screen in which it segregated with increased frequencies of circulating CD44hi activated/memory T cells and regulatory T cells (Fig. 2a) in otherwise healthy adult mice. Detailed analysis revealed that the I325N allele diminished IκBα levels within most immune cell populations, including T and B cells, natural killer cells, dendritic cells and macrophages (Supplementary Fig. 3a). Macrophages from the bone marrow of mice homozygous for the I325N allele produced more inflammatory cytokines in response to lipopolysaccharide (LPS) than macrophages from wild-type mice (Supplementary Fig. 3b). In isolated thymocytes, presence of the I325N allele increased NF-κB signaling in ways consistent with diminished A20-mediated inhibition (Supplementary Fig. 3c)18. Furthermore, when wild-type mice were transplanted with mixtures of bone marrow encoding mutant and wild-type A20, the I325N allele acted in a cell-autonomous manner to increase T cell antigen receptor (TCR)- and CD28-dependent formation of FOXP3+CD4+ regulatory T cells and their Helios+FOXP3 precursors within the thymus (Fig. 2b and Supplementary Fig. 3d–f). Mice homozygous for the I325N allele also harbored increased numbers of B cells in the spleen and peritoneal cavity (Fig. 2c and Supplementary Fig. 4a–c). Isolated B cells exhibited increased NF-κB activation in response to both LPS and IgM stimulation (Supplementary Fig. 4d). As for T cells, these responses were found to be cell autonomous, because when wild-type mice were transplanted with mixtures of bone marrow encoding mutant and wild-type A20, only cells harboring the I325N allele exhibited features of increased B cell activation and proliferation by LPS or antigen receptors (Fig. 2d and Supplementary Fig. 4e–g). Surprisingly, I325N had a greater effect than the C103A OTU domain substitution in T cells and B cells analyzed in parallel bone marrow transplant recipients (Fig. 2b,d), despite C103A completely abrogating the polyubiquitin protease activity of A20 (refs. 15,20), indicating that I325N must diminish additional inhibitory mechanisms.

Fig. 2: The I325N OTU variant confers heightened immunity in mice.
figure 2

a, Representative flow cytometry analysis of splenocytes, showing the percentage of CD4+ or CD8+ T cells and the percentage of CD44hi activated/memory cells within these subsets as well as the percentage of CD25+CD44int regulatory T cells and CD25CD44hi effector/memory T cells among CD4+ cells. b, Wild-type (WT) CD45.1+ mice were transplanted with mixtures containing equal amounts of congenic bone marrow from CD45.1+ donors of Tnfaip3+/+ genotype and CD45.2+ donors of Tnfaip3I325N/I325N, Tnfaip3C103A/C103A or Tnfaip3+/+ genotype. A pairwise comparison is shown of the frequency of FOXP3+ cells among CD4+ splenocytes from the same chimera. c, Numbers of the indicated B cells in the spleen of individual mice with the indicated Tnfaip3 genotype. B, B cells; NFB, newly formed B cells; FO, follicular B cells; MZ, marginal zone B cells. d, Splenocytes from the chimeras in b were cultured with 0.1 µg ml−1 LPS for 0–4 d, and the frequency of CD45.2+ cells of the indicated Tnfaip3 genotypes was measured among B cells from individual chimeras (lines), expressed relative to the frequency on day 0. eg, Male Tnfaip3+/+ (n = 14), Tnfaip3I325N/+ (n = 20) and Tnfaip3I325N/I325N (n = 10) C57BL/6 mice were injected with coxsackievirus on day 0. e, Kaplan–Meier survival data (two-tailed log-rank test). f,g, Plasma IL-6 concentration (f) and virus titers (plaque-forming units, p.f.u.) (g) per mg of pancreas on day 3. Data are from three independent experiments and are shown as mean ± s.e.m. One-way ANOVA was used for significance analysis unless otherwise stated: *P< 0.05; **P < 0.01; ***P< 0.001; ****P< 0.0001.

Consistent with heightened levels of cellular markers of immunity, mice with at least one I325N allele of Tnfaip3 had greater resistance to coxsackievirus B4 strain E2 (hereafter referred to as coxsackievirus), a virus isolated from a human neonate with a disseminated fatal infection causing extensive pancreatic necrosis35,36. A virus dose that was lethal for 90% of wild-type C57BL/6 mice was not lethal for Tnfaip3I325N/I325N littermates, and caused less mortality in Tnfaip3I325N/+ mice (Fig. 2e and Supplementary Fig. 5a). Mutant mice had less infectious virus and viral RNA in the pancreas, lower levels of mRNA encoding the immune response cytokines interleukin (IL)-1β and interferon-β, less pancreatic necrosis, higher serum IL-6 concentration and preserved body weight and euglycemia (Fig. 2f,g and Supplementary Fig. 5b–g).

The homogeneous genetic background of the Tnfaip3-mutant mice allowed testing of whether heightened immunity imposed a subclinical cost or altered the insulin anabolic axis37. Tnfaip3I325N/+ mice were healthy, of normal weight and fertile, producing homozygous offspring at the expected ratio. Homozygotes also seemed healthy, although their body weights were 5–20% lower than those of their heterozygous and wild-type littermates at 8 and 12 weeks of age (Fig. 3a), with histological analysis revealing low-grade inflammation of the pancreatic islets, colon, kidney and liver (Fig. 3b,c and Supplementary Fig. 6a–i). Pancreatic insulitis in I325N homozygotes was associated with a 50% reduction in beta cell mass (Fig. 3d), although blood glucose concentrations and glucose tolerance tests were normal (Supplementary Fig. 7a–c). Isolated islets from Tnfaip3I325N/I325N mice exhibited normal basal insulin output but reduced insulin secretion when stimulated in vitro (Fig. 3e). Islet transplant and culture experiments showed that the Tnfaip3 I325N allele acted within islet cells in ways consistent with reduced function, including exaggerating canonical and noncanonical NF-κB signaling, which lowered insulin secretion and increased inflammatory cytokine gene expression (Fig. 3f–j, Supplementary Figs. 7d,e and 8a–f, and Supplementary Table 6) (ref. 38). Together, these data demonstrate that the A20 I325N allele tunes immunity to provide beneficial protection to coxsackievirus infection at the cost of loss of tissue homeostasis in peripheral tissues, such as the pancreatic islets.

Fig. 3: Subclinical inflammatory and metabolic consequences of the I325N allele.
figure 3

a, Body weights of Tnfaip3+/+, Tnfaip3I325N/+ and Tnfaip3I325N/I325N mice of the indicated age and sex. b, Hematoxylin and eosin (H&E) and insulin (INS) staining of pancreas sections (scale bar, 100 µm). c, Insulitis scores (where 4 represents >75% islet mononuclear cell infiltration and 0 represents an absence of infiltrating cells). d, Calculated beta cell mass for mice with the Tnfaip3 I325N allele. Data are representative and cumulative from three independent experiments. e, Pancreatic islets were isolated from individual mice of the indicated A20 genotype and incubated overnight. Following incubation, an in vitro glucose-stimulated insulin secretion response assay in 2 mM or 20 mM d-glucose or 25 mM KCl was conducted in separate groups of islets. Cumulative data from three independent experiments are shown. f,g, Pancreatic islets were isolated from individual mice of the indicated A20 genotype and treated with 200 U ml−1 TNF for the indicated times. f, Representative immunoblot for IκBα and β-actin (loading control). hTNF, recombinant human TNF. g, Expression of TNF-induced genes. Data represent three independent islet preparations with four (Tnfaip3+/+) and three (Tnfaip3I325N/I325N) biological replicates. h, Glucose tolerance test in diabetic wild-type mice transplanted with Tnfaip3+/+ (n = 13), Tnfaip3I325N/+ (n = 8) or Tnfaip3I325N/I325N (n = 12) islets. i, Representative H&E sections from three independent experiments of islet grafts at post-operative day 30 (scale bar, 50 µm), with the fraction of grafts exhibiting immune infiltrate shown below. j, Islet grafts isolated on post-operative day 10 were analyzed for the indicated mRNAs by RT–qPCR. Error bars represent mean ± s.e.m. Statistical analysis was performed by one-way ANOVA (a, d, e, h) or two-tailed Student’s t-test (j) : *P < 0.05; **P < 0.01; ****P < 0.0001.

An A20 allelic series shows graded A20 phosphorylation

The apparent beneficial effects of the T108A;I207L and I325N OTU domain alleles detailed above contrasted with the effect of a third OTU domain substitution, A20 C243Y, found as a family-specific allele causing a dominant Mendelian inflammatory disorder resembling Behçet’s disease, with childhood onset, oral and genital ulceration and skin inflammation39. The biochemical basis for the clinically penetrant effects of C243Y was obscure, because other similarly affected cases of A20 haploinsufficiency result from nonsense or frameshift mutations that truncate or eliminate A20 protein12.

None of the three OTU domain alleles altered the accumulation of A20 protein (Supplementary Fig. 9a), but they nevertheless caused graded reductions in IKKβ-mediated phosphorylation of a critical serine residue (S381) that promotes A20 function and NF-κB inhibition18,19 (Fig. 4a–c). The C243Y allele was associated with the most severe loss of IKKβ-mediated S381 phosphorylation, diminishing it to 5% of the wild-type level, whereas I325N had an intermediate effect (50% of the wild-type level) and the Denisovan T108A;I207L allele had a mild effect (80% of the wild-type level). Reduced S381 phosphorylation was also observed by mass spectrometry of A20 protein, purified from transfected human cells (Supplementary Fig. 9b). In contrast, the C103A catalytic site substitution did not cause a significant decrease in A20 phosphorylation as measured by immunoblotting and mass spectrometry (Fig. 4b,c and Supplementary Fig. 9b).

Fig. 4: Alleles with OTU domain variants decrease A20 phosphorylation and NF-κB inhibition.
figure 4

a,b, Representative immunoblots of lysates from βTC3 mouse insulinoma cells transfected with vectors encoding wild-type A20 or A20 with the indicated amino acid substitutions, with or without cotransfected IKKβ, and probed with antibodies to p-S381 A20, total A20, IKKβ and β-actin. Molecular weight markers (kDa) and slow- or fast-migrating A20 species are indicated. c, Densitometric analysis from multiple independent experiments (dots) showing the ratio of p-S381 A20 to total fast-migrating A20 in IKKβ-cotransfected cells, expressed relative to the ratio in cells with wild-type A20. NS, not significant. d,e, Inhibition of an NF-κB luciferase reporter in βTC3 cells by cotransfection of wild-type A20 or the indicated A20 mutants, expressed relative to wild-type A20 (Supplementary Fig. 9c,d). Columns are arithmetic means ± s.e.m. Statistical comparison was performed by one-way ANOVA: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

A20-mediated inhibition of an NF-κB luciferase reporter was also reduced in a graded fashion by each OTU domain allele in the same order that each diminished S381 phosphorylation (Fig. 4d and Supplementary Fig. 9c). The Denisovan T108A;I207L allele had the smallest effect, whereas the C243Y allele was the poorest NF-κB inhibitor of the series, with the latter almost as compromised as A20 with S381 substituted to nonphosphorylatable alanine (S381A). Combining the intermediate I325N allele with the S381A mutation in cis did not cause a further decrease in A20 activity, consistent with I325N resulting in a primary deficiency in S381 phosphorylation. This conclusion was reinforced by combining I325N in cis with a substitution of S381 to the phosphoserine mimetic glutamate (S381E), which rescued the loss of activity caused by I325N (Fig. 4e and Supplementary Fig. 9d).

Phosphorylated A20 (p-S381) predominantly migrated more slowly in SDS–PAGE than unphosphorylated A20, either when tested by cotransfection with IKKβ (Fig. 4a,b) or in TNF-stimulated peripheral blood leukocytes with endogenous A20 and IKKβ (Fig. 5a). The ratio of this slowly migrating A20 species to the more rapidly migrating form was markedly decreased in blood leukocytes from healthy donors heterozygous for the Denisovan T108A;I207L allele, and further decreased in a healthy donor homozygous for T108A;I207L, as compared to healthy control noncarriers (Fig. 5b). The ratio of p-S381 A20 to fast-migrating A20 was also decreased (Fig. 5c), as was the total level of slow-migrating A20 (Fig. 5d). These biochemical changes had a functional impact, as PBMCs from T108A;I207L carriers showed increased mRNA levels of CXCL2 under unstimulated conditions (Fig. 5e), and a trend toward increased IκBα degradation and significantly increased inflammatory gene expression following TNF stimulation (Figs. 1f and 5f).

Fig. 5: Immunoblot and RT–PCR analysis of peripheral blood mononuclear cells from healthy individuals with the Denisovan T108A;I207L TNFAIP3 haplotype.
figure 5

a, Immunoblot analysis of PBMCs from individuals with (TNFAIP3I207L/+) or without (TNFAIP3+/+) the Denisovan TNFAIP3 haplotype. PBMCs were left untreated or stimulated with hTNF for 15 or 60 min. Proteins assessed included A20, phosphoserine A20 (p-S381 A20), IκBα and β-actin (loading control). b, Densitometric ratio of slow- to fast-migrating A20 species, showing the mean and s.d. from six normal donors, five T108A;I207L heterozygotes and one T108A;I207L homozygote. c, Densitometry analysis of p-S381 A20 levels in the immunoblots in a against the total A20 present. d, Densitometry analysis of total A20 levels against β-actin loading control. Data are further divided into fast-migrating (FM) or slow-migrating (SM) A20. e, RT–PCR analysis of CXCL2, ICAM1, TNF and TNFAIP3 gene expression in PBMCs of individuals heterozygous (green circles) or homozygous (purple circle) for the Denisovan TNFAIP3 haplotype and 15 healthy donors (blue squares), comparing negative ΔCt basal expression levels. f, Densitometry analysis of IκBα levels in the immunoblots in a corrected for the β-actin loading control. For all densitometry analyses, the average value from samples without the T108A;I207L haplotype was used to compare all samples. Each symbol corresponds to an individual lane in the immunoblot. Error bars represent s.d.; two-tailed Student’s t-test was used for significance analysis: *P < 0.05.

To explore the structural and biochemical consequences on the posterior surface of the OTU domain, we focused on the intermediate I325N allele. Crystallographic structures of A20 OTU domains with wild-type I325 or mutant N325 revealed no differences in features with known functions, including the catalytic triad and ubiquitin-binding surface (Supplementary Fig. 10a–c and Supplementary Table 7). Subtle shifts in the stem of the β7–β8 loop that contains the conserved surface residues T321, T322 and L324 (Supplementary Fig. 10d–f) were detected, but these did not alter the conserved posterior surface of the OTU domain40, including the β3–β4 loop containing C243 (Supplementary Fig. 10g). The β7–β8 loop itself is disordered in all available OTU structures but, like the disordered loops in the unliganded S1 ubiquitin-binding site, it may undergo conformational changes upon binding a cognate partner that are hindered by the I325N substitution41.

Cycloheximide treatment confirmed no difference in I325N protein stability, and wild-type and I325N OTU domains exhibited similar thermal denaturation profiles (Supplementary Fig. 11a–d). I325N did not decrease the DUB activity of the bacterially expressed OTU domain against K48-linked polyubiquitin in vitro (Supplementary Fig. 11e,f), but did diminish K63-linked polyubiquitin DUB activity and K48-linked ubiquitin ligase activity when full-length A20 was expressed in human cells (Supplementary Fig. 11g,h), consistent with these functions requiring phosphorylation of S381 (refs. 18,19). Together, these data identify a series of A20 OTU domain alleles that cause graded reductions in A20 phosphorylation and A20 control of NF-κB.

Human A20 OTU domain alleles in genome-edited mice

The mouse and human A20 OTU domains show a high degree of secondary and tertiary structural conservation, so to study the physiological effects of human OTU domain alleles in a controlled genetic and environmental context, human I207L and C243Y alleles were separately introduced into the germline of C57BL/6 mice by CRISPR–Cas9 genome editing. T108A is present in the C57BL/6 mouse and other vertebrate reference genomes, allowing us to recreate the Denisovan T108A;I207L allele in a single editing step, and also directly test the additive effect of I207L. Mice homozygous for the T108A;I207L allele (Tnfaip3I207L/I207L) were born at the expected Mendelian ratio and exhibited no overt signs of illness, weight loss, tissue pathology or early lethality (Fig. 6a and Supplementary Fig. 12a–e). In contrast, mice homozygous for the C243Y allele (Tnfaip3C243Y/C243Y) weighed 30–50% less than their wild-type littermates at 3 weeks of age and only ~50% survived to 16 weeks of age (Fig. 6a–c and Supplementary Fig. 12a). Pancreatic and intestinal pathology was also evident in 16-week-old mice homozygous for the C243Y allele, including loss of the goblet cells responsible for maintaining the mucin barrier between the intestinal epithelium and luminal microbes (Fig. 6d and Supplementary Fig. 12c–e).

Fig. 6: C243Y causes spontaneous inflammation and early lethality in homozygous mice.
figure 6

a, Weight of 3-week-old Tnfaip3I207L, Tnfaip3I325N and Tnfaip3C243Y mice of the indicated genotype. Bars represent mean ± s.e.m and one-way ANOVA was conducted to test statistical significance: **P < 0.001; ****P < 0.0001. b, Representative photographs of 16-week-old Tnfaip3C243Y mice of the indicated genotype. c, Survival of mice homozygous for the C243Y allele (n = 52) as compared to heterozygous (n = 43) and wild-type (n = 41) littermates. Significance was determined by two-tailed log-rank test: ****P < 0.0001. Data are representative sections of three independent experiments, showing H&E-stained colon sections from female mice of the indicated genotype (scale bar, 200 µm).

Comparison of mice with the three different Tnfaip3 alleles with wild-type littermate controls revealed a graded increase in the frequency of circulating CD44hi activated/memory T cells and FOXP3+ regulatory T cells in order of increasing magnitude: I207L heterozygotes < I207L homozygotes = I325N heterozygotes < I325N homozygotes = C243Y heterozygotes < C243Y homozygotes (Fig. 7a,b and Supplementary Fig. 13a–c). Flow cytometry analysis of leukocytes from age-matched and healthy human carriers of the Denisovan T108A;I207L haplotype revealed a subtle increase in CD8+ T effector memory cells and T regulatory cells and a significant reduction in the frequency of CD8+ effector memory T cells that had re-expressed CD45RA (TEMRA) (Supplementary Fig. 14a–d). CD8+ T cells play a key role in viral immunity, and the frequency of CD8+ effector memory subsets can reflect the influence of pathogen load and experience42,43. The subtle immune phenotype seen for healthy human carriers was consistent with the graded impact of the T108A;I207L allele on A20 function.

Fig. 7: Introduction of human T108A;I207L and C243Y alleles in mice tunes immunity.
figure 7

a, Percentage of CD44hi activated/memory CD8+ T cells. b, Percentage of CD4+FOXP3+ regulatory T cells from the spleens of mice encoding the T108A;I207L (I207L), I325N or C243Y allele. Each point corresponds to an individual mouse collected from more than three independent experiments. Statistical analysis was performed by a two-tailed Student’s t-test and data are shown as the mean ± s.e.m: *P< 0.05; **P< 0.01; ***P< 0.001; ****P< 0.0001. cf, Male Tnfaip3+/+ (n = 20), Tnfaip3I207L/+ (n = 11), TnfaipI207L/I207L (n = 13) (c,d) and Tnfaip3+/+ (n = 16), Tnfaip3C243Y/+ (n = 12) and TnfaipC243Y/C243Y (n= 4) (e,f) C57BL/6 mice were injected with coxsackievirus on day 0. c,e, Kaplan–Meier survival data (using a two-tailed log-rank test). d,f, Plasma IL-6 serum levels on day 3. Data are shown as the mean ± s.e.m. and one-way ANOVA was used for significance analysis unless otherwise stated: *P< 0.05.

Similarly to Tnfaip3I325N mice, Tnfaip3I207L and Tnfaip3C243Y mice exhibited increased resistance to coxsackievirus (Fig. 7c,e). A virus dose that was lethal for wild-type C57BL/6 mice caused less mortality in Tnfaip3I207L/I207L and Tnfaip3C243Y/+ mice (Fig. 7c,e and Supplementary Information Fig. 1a,b). However, protection was not complete for Tnfaip3I207L/I207L mice when compared to Tnfaip3I325N/I325N and Tnfaip3C243Y/C243Y mice, which showed 100% survival with an otherwise lethal inoculation of coxsackievirus (Figs. 2e and 7c,e). The resistant mice exhibited higher serum IL-6 concentrations, with better preservation of body weight and blood glucose, as compared to wild-type littermates (Fig. 7d,f and Supplementary Information Fig. 1c–f). These data show that the graded reduction in A20 phosphorylation had a physiological impact by beneficially enhancing protective immunity in a correspondingly graded fashion to coxsackievirus.

Shift from beneficial effect to detrimental loss of microbial tolerance

We next explored the possibility that graded loss of A20 phosphorylation and graded loss of microbial tolerance in mice harboring the I325N and C243Y alleles might become detrimental in particular environmental settings. In an experimental model for septic shock, mice homozygous for the I325N and C243Y alleles had higher mortality and greater serum IL-6 concentration following LPS injection as compared to wild-type controls or to T108A;I207L homozygotes (Fig. 8a–d). In the mouse equivalent of smallpox, infection with the orthopoxvirus ectromelia virus was tolerated and controlled by wild-type mice, yet resulted in higher mortality and higher viral titers in I325N homozygotes (Fig. 8e,f and Supplementary Information Fig. 1g,h). Furthermore, in a mouse mixed bone marrow chimera model of autoimmune diabetes, pancreatic islet-reactive CD4+ T cells escaped deletion and precipitated diabetes in the presence of cells harboring the I325N allele (Fig. 8g and Supplementary Fig. 15a,b). The rogue islet-reactive T cells were nevertheless derived equally from precursors with wild-type A20 and harboring the I325N allele, whereas I325N acted cell autonomously to increase expression of major histocompatibility complex class II and the T cell co-stimulatory molecule CD86 on B cells and dendritic cells, and to increase the frequencies of germinal center B cells (Supplementary Fig. 15c–e). These data highlight that tuning A20 to increase immunity is balanced by a corresponding cost of increased pathogenicity under different environmental and genetic scenarios, as is most clearly evident with the increasing loss of bacterial tolerance in mice homozygous for I325N and C243Y (summarized in Table 1).

Fig. 8: Shift to detrimental effects in response to environmental stressors and increased susceptibility to autoimmunity.
figure 8

a, Kaplan–Meier survival curves following intraperitoneal injection of 2.5 mg kg−1 LPS in mice harboring the I207L, I325N and C243Y A20 OTU alleles. Tnfaip3I325N (+/+ and I325N/+), n = 16; Tnfaip3I325N/I325N, n = 18; Tnfaip3I207L, n = 6 for each genotype; Tnfaip3C243Y, n = 5 for each genotype. bd, Serum IL-6 concentration in Tnfaip3I207L (b), Tnfaip3I325N (c) and Tnfaip3C243Y (d) mice of the indicated genotype administered 2.5 mg kg−1 LPS by intraperitoneal injection at time = 0. e, Kaplan–Meier survival curves for female littermates of the indicated Tnfaip3I325N genotype and wild-type controls (C57BL/6) infected with ectromelia virus. f, Day 8 serum viral load assessed from the mice in e. g, Diminished tolerance to self-antigen caused by I325N A20. The incidence of diabetes is shown in InsHEL transgenic mice with wild-type A20 in pancreatic islets, after transplantation with a congenic bone marrow mixture derived from CD45.1+ donors of the Tnfaip3+/+ genotype transgenic for TCR3A9 and CD45.2+ donors transgenic for TCR3A9 with either the Tnfaip3I325N/I325N (red circles) or Tnfaip3+/+ (blue circles) genotype (n = 12 chimeras for Tfnaip3+/+ donors and 14 chimeras for Tnfaip3I325N/I325N donors pooled from two separate experiments). Statistical analysis was performed by two-tailed log-rank test (a, e, f) or by one-way ANOVA (b, c, d, f). Bars represent mean ± s.e.m: *P < 0.05; ***P < 0.002; ****P < 0.0001.

Table 1 Comparison of alleles in the A20 OTU allelic series


Our findings reveal genetic and biochemical mechanisms for adaptively increasing immunity, with tradeoffs against microbial tolerance and anabolic metabolism that either remain clinically silent or become detrimental in specific contexts. Three different alleles altering the N-terminal OTU domain of A20 act, to different degrees, by diminishing A20 phosphorylation, reducing the immune inhibitory activity of A20 in vitro, and increasing the frequency of activated T cells and immune responses in vivo. A rare human allele, C243Y, almost entirely eliminates A20 phosphorylation and shifts the balance away from microbial tolerance to the extremes of immunity, resulting in severe inflammatory disease in both mice and humans. This outcome is comparable to the phenotypes observed with other rare human alleles that eliminate or truncate the A20 protein expressed from one TNFAIP3 allele and engineered mouse knockouts that eliminate A20 expression from both alleles11,12. In contrast, the Denisovan T108A;I207L allele, common in people from Oceania, and the chemically induced I325N allele decrease A20 phosphorylation more modestly, without precipitating spontaneous inflammatory disease in the humans or mice who harbor them. Like C243Y, I325N in the mouse confers strong resistance to an otherwise lethal dose of a coxsackievirus, whereas the Denisovan T108A;I207L allele (which has the smallest effect on phosphorylation) confers only partial resistance. Heightened resistance to microbial pathogens may explain the beneficial effect of the Denisovan T108A;I207L TNFAIP3 haplotype, as evidenced by its high frequency in modern human populations east of the Wallace Line, who likely encountered new pathogens as they moved into environments with unique fauna and flora.

The three TNFAIP3 alleles analyzed here have much larger effects on immunity in mice than the protease-dead C103A OTU allele, revealing a previously unanticipated and critical role for the OTU domain in promoting A20 S381 phosphorylation and immune inhibitory activity. Previous studies in mice have conditionally deleted Tnfaip3 in specific tissues, causing severe inflammatory disease, or directly disabled individual enzymatic functions of A20 in all tissues (for example, via introduction of C103A), resulting in surprisingly little inflammation5,6,11,15,18,20. In contrast, phosphorylation regulates the multiple ubiquitin-editing functions of A20 (refs. 18,19), and as shown here, three OTU domain alleles that diminish phosphorylation have greater effects on immune cells, microbial tolerance and resistance in vivo than alleles affecting individual ubiquitin-editing activities. The effects of the three OTU domain alleles on S381 phosphorylation were proportional to their effects on A20 inhibition of NF-κB signaling and on immunity, and a phosphomimetic substitution (S381E) partially restored the inhibitory function of I325N in an NF-κB luciferase assay. While these results support the conclusion that the effects are due to the critical role of S381 phosphorylation18,19, they do not exclude the role of other phosphorylation sites or yet-to-be-identified steps activating the inhibitory activities of A20. I207 and T108 are buried deep within the OTU domain, whereas C243 and I325 involve residues within two separate surface loops, β3–β4 and β7–β8, respectively (Supplementary Fig. 10g), on the highly conserved posterior surface of the OTU domain40. Structural studies of full-length A20 protein may illuminate how this surface relates to the C-terminal domain harboring S381. Because the conserved surface bounded by C243 and I325 is large, it potentially offers many opportunities for substitution of buried or surface residues to tune immunity and tolerance.

Our study of the I325N allele with intermediate loss of function provides two examples of genetic tradeoffs. The first involved glucose metabolism, where the I325N allele had the surprising effect within pancreatic islet beta cells of decreasing insulin secretion while increasing inflammation. This is reminiscent of findings in Drosophila that insulin production or action decreases during infection, diminishing body glycogen supplies through increased activity of FOXO, a transcriptional inducer of starvation responses37. Resting insulin levels are a good prognostic marker in human sepsis, and exogenous insulin treatment improves outcome44. Subtly decreased A20 activity may enhance immune demands for energy and, separately, help meet these energy demands by lowering insulin-induced anabolic growth, contributing to both cachexia associated with chronic infection and inflammation.

The second example of a tradeoff is the experimental demonstration that a beneficial trait in one context can be deleterious in another. The I325N allele conferred increased resistance to coxsackievirus, an important human enterovirus, but increased mortality to ectromelia virus, a relative of the variola and myxoma viruses. The higher ectromelia virus-related mortality in mice harboring the I325N allele is reminiscent of the high mortality in indigenous populations of the Americas and Oceania exposed to variola virus relative to Europeans8. The findings here suggest that the devastation wrought by these and other microbes reflects selection during earlier environmental conditions for lower microbial tolerance and a stronger immune response, with A20 representing one critical determinant of outcome.


Human subjects

Patients and healthy family members were recruited via the Clinical Immunogenomics Research Consortium Australia (CIRCA) and The Children’s Hospital at Westmead, providing written informed consent under protocols approved by the Sydney Children’s Hospitals Network Human Research Ethics Committee. Blood samples from healthy volunteers were obtained from the Australian Red Cross Blood Service.

Genome sequencing

Parent–proband trio genomes were sequenced on the Illumina HiSeq X platform using DNA isolated from whole blood. Libraries were generated using either Illumina TruSeq PCR-free or TrueSeq Nano. Alignment, variant calling and annotation were performed as described previously45.

Population genetic analysis

TNFAIP3 genotypes were extracted from 279 publicly available genomes from Simons Genome Diversity Project samples22 or exome data from 72 Martu Indigenous Australians32. In addition, 514 individuals from across mainland and Island Southeast Asia, Papua New Guinea and Oceania were genotyped at 567,096 variants using the Affymetrix Axiom Genome-Wide Human Array, and the 538,139 autosomal variants with <5% missing data were kept for further analyses30. Weir and Cockerham’s estimated FST values were calculated using VCFtools v.0.1.14, and principal-component analysis was performed using PLINK v.1.9. Genotypes were phased and imputed using the Michigan Imputation Server v.1.0.3 ( and compared to high-coverage Denisovan27 and Altai Neanderthal28 TNFAIP3 haplotypes. Phred-scaled CADD scores were calculated for all PASS variants across the extended Denisovan haplotype with gnomAD allele frequency <0.01.

To validate the imputation efficiency of the Denisovan T108A and I207L missense variants, we made use of a combined dataset of 481 modern and archaic genomes, including 161 high-coverage genomes from the Indonesian Genome Diversity Project, sampled from 14 islands across Island Southeast Asia46. Also included were 292 genomes from the Simons Genome Diversity Project22, 25 additional Papuan genomes47, and 2 archaic genomes (Altai Denisovan27 and Neanderthal28). The combined dataset was phased using read-aware phasing with SHAPEIT v.2.r837 (ref. 48), using the HapMap phase II b37 recombination map49 and the following arguments: --states 400 --window 0.5 --states-random 200. Informative phased genotypes across the interval (chr.6:138160825–138246614) were then segregated into 18 unique haplotypes using Haploview (v.4.2) (ref. 50) and used to impute TNFAIP3 variants in 514 individuals with single-nucleotide polymorphism array data30.


The Tnfaip3 lasvegas strain (Tnfaip3I325N) was generated by ENU mutagenesis in C57BL/6 mice, and propagated by backcrossing to C57BL/6 mice. The strain was maintained as heterozygous breeding pairs so that wild-type littermates could be used for controls. To generate Tnfaip3C243Y and Tnfaip3I207L strains, respective guide RNAs (5′-GGGATATCTGTAACACTCC-3′ and 5′-TGACAATGATGGGTCTTCTGAGG-3′) were microinjected into C57BL/6 zygotes in combination with Cas9 mRNA and respective oligonucleotide templates (5′-CACTGTCCTTCAGGGTCACCAGGGGTACAAAGTGCTGGCTGTCATAGCCTAGGACGATGGGATATCTGTAATACTCCTGAGCAGGCCAGTGAAGAGGCAGATAAATCCCACCCACTTTCAAAGGAGCAAAATTGGAACCAGATTCCAAAC-3′ and 5′-CTCCTCAGAGCTGAAACTCACCCAGGGAACCTAGAAACTCTCTGAGGCACCTCACCTGAAATGACAATGAGGGGTCTTCTGAGAATGTTGCTGAGGACAAATATGTGGATTTCTTCCAGGGAATTGTACTGAAGTCCACTTCGGGCTGCA-3′). Founder mice carrying the appropriate substitutions were then crossed to C57BL/6 mice, and heterozygous offspring were intercrossed to generate homozygous, heterozygous and wild-type littermates. Mice were housed at the Australian Phenomics Facility (Australian National University) or at the Australian BioResources Centre. Tnfaip3C103A (ref. 18), TCR3A9-transgenic51 and insHEL transgenic52 mice have been described previously. Mice were genotyped for transgenes and A20 by PCR and assessed 7–28 (typically 10–16) weeks after birth. For TCR3A9-transgenic mouse experiments, mice were hemizygous for the 3A9 or insHEL transgene on the B10.BR and B10.BR.SJL-Ptprca (CD45.1) background, and the Tnfaip3I325N mice were backcrossed to B10.BR mice. For bone marrow chimeric mice, congenic C57BL/6.SJL-Ptprca or B10.BR.SJL-Ptprca (CD45.1) mice were irradiated with two doses of 4.5 Gy 4 h apart, and injected intravenously with mixtures of 1.8 × 106 bone marrow cells from C57BL/6 or B10.BR SJL-Ptprca mice and 1.8 × 106 bone marrow cells from C57BL/6 (CD45.2) mutant or control mice, and analyzed 8–14 weeks later. Animal studies were approved by the Garvan/St Vincent’s or the Australian National University Animal Ethics Committees. All procedures performed complied with the Australian Code of Practice for Care and Use of Animals for Scientific Purposes.

Coxsackievirus infection model and virus quantification

Mice were intraperitoneally injected with normal saline (control) or 20 p.f.u. in 200 μl of saline of the coxsackievirus B4 strain E2 grown in HeLa cells. Mice were monitored daily and euthanized if they displayed gross signs of illness (for example, ruffled coat, hunching). The pancreas was dissected at the indicated times for RT–qPCR, histopathological analysis and viral titer determination. Serum was also collected by terminal cardiac puncture for measurement of IL-6 by ELISA (BD; OptEIA Set Mouse IL-6), according to the manufacturer’s instructions. Plaque assays were performed to determine viral titers within the pancreas following infection. HeLa cells (0.6 × 106 cells per well) were seeded in 2 ml of complete medium (RPMI with 10% FCS, 2 mM l-glutamine, 100 U ml−1 penicillin, 100 μg ml−1 streptomycin and 100 μg ml−1 Normocin) in six-well plates and incubated overnight at 37 °C. Samples were collected in RPMI, homogenized in a Dounce tissue grinder and passed through a 22-μm filter, before preparing tenfold serial dilutions. HeLa cells, at 90% confluency, were washed with 1× PBS and 400 μl of infectious homogenate added to each well and incubated for 60 min at 37 °C under gentle rocking. Infectious medium was removed and 3 ml of agar mix (2× 1.8% agar, 2× MEM containing 10% FCS) was added to each well before incubating at 37 °C for 3 d. Cells were then fixed with Carnoy’s reagent for 60 min and subsequently stained with 0.5% crystal violet for 60 s. Wells were extensively washed with water and plaques were counted.

Ectromelia virus model

Mice were inoculated subcutaneously with 103 p.f.u. ectromelia virus (Moscow strain; ATCC no. VR-1374) in the flank of the left hind limb between the regio tarsi and regio pedis under avertin anesthesia. Clinical scores were recorded daily, and animals were weighed on days 0 and 5, and every 2 d thereafter until day 21, and bled on days −1 and 8 to measure viral load. Animals with a significant clinical score or a ≥20% decrease in starting weight were euthanized and considered dead the following day. Viral load was measured in the blood by qPCR for ECTV-Mos-156 viral genomes and in organs by viral plaque assay as the log10 p.f.u. per gram of tissue53.

Macrophage cultures

Bone marrow cells from femurs and tibiae were cultured for 7 d in complete RPMI-1640 medium with 10% FBS, 100 U ml−1 penicillin, 100 μg ml−1 streptomycin, 2 mM l-glutamine (Life Technologies) and 50 ng ml−1 recombinant human macrophage colony-stimulating factor (Peprotech). Cells were placed into six-well plates for stimulation with 10 ng ml−1 Salmonella minnesota R595 (Re) ultra-pure LPS (List Biological) for different periods of time. RNA was extracted by TRIzol (Life Technologies) and reverse transcribed to cDNA with SuperScriptII (Life Technologies), and Cxcl1 transcript abundance was measured using the TaqMan assay (Applied Biosystems). RNA abundance relative to that of Ef1a was calculated with ΔCt values. Conditionally Hoxb8-immortalized bone marrow progenitor cells were generated as described54.

LPS sepsis model

A dose of 2.5 mg kg−1 LPS was administered by intraperitoneal injection. Mice were monitored every 1 h for 10 h and killed if the ethical end point was reached. Monitoring was conducted using body conditioning score and Grimace scale (National Centre for the Replacement, Refinement and Reduction of Animals in Research) approved by the Garvan/St Vincent’s Animal Ethics Committee. Monitoring continued twice daily for 7 d in surviving mice, where weights and blood glucose concentrations were also measured. Serum was collected by tail tipping before LPS injection and 2 and 4 h following injection for determination of IL-6 concentration by ELISA (BD, OptEIA Set Mouse IL-6), according to the manufacturer’s instructions.

Immunohistochemistry and beta cell area determination

Tissues were fixed in 10% neutral-buffered formalin (Sigma-Aldrich) and embedded in paraffin, and parallel sections (5 μm) were prepared. Sections were stained with H&E (Sigma-Aldrich) and, for pancreatic tissue, parallel sections were stained for insulin (purified rabbit polyclonal anti-mouse insulin; 4590, Cell Signaling Technology). Visualization of bound anti-insulin was achieved using horseradish peroxidase (HRP)-labeled polymer-conjugated goat anti-rabbit IgG (Dako EnVision+ System), followed by counterstaining with hematoxylin. For pancreatic beta cell mass determination, consecutive pancreatic serial sections at 200-μm intervals were stained for insulin, and beta cell area was quantified from the total area (taken by insulin-positive cells as compared to nonpositive tissue) (ImageJ, NIH). Beta cell mass (mg) was calculated by multiplying the relative insulin-positive area by the mass of the isolated pancreas before fixation. Images were captured using a Leica DM 4000 or Leica DM 6000 Power Mosaic microscope (Leica Microsystems).

Metabolic studies

Blood glucose concentrations were determined using a FreeStyle Lite glucometer and blood glucose test strips (Abbott Diabetes Care) via tail tipping. Measurements were taken from 8- and 12-week-old male or female non-fasted mice. Glucose tolerance tests and measurements of glucose-stimulated insulin secretion were conducted as previously described38.

Islet isolation, transplantation and in vitro studies

Islets were isolated as previously described55, and counted for islet transplantation or in vitro experiments using a Leica MZ9.5 stereomicroscope. Islets were transplanted under the kidney capsule of diabetic C57BL/6 littermates as described56. For in vitro studies, islets were placed into 12-well non-tissue-culture-treated plates (150–200 islets per well; Fisher Scientific) and were treated with 200 U ml−1 recombinant human TNF (R&D Systems) for 1, 4 or 24 h. Total RNA was extracted using the RNeasy Plus Mini kit (Qiagen) and reverse transcribed using the Quantitect Reverse Transcription kit (Qiagen). Primers were designed using Primer3 (ref. 57) with sequences obtained from GenBank and synthesized by Sigma-Aldrich (Supplementary Tables 5 and 6). PCR reactions were performed on the LightCycler480 Real-Time PCR System (Roche) using FastStart SYBR Green Master Mix (Roche). Cyclophilin (CPH2), RPL13A and ACTB were used as housekeeping genes, and data were analyzed using the 2ΔΔCt method.

Flow cytometry

Flow cytometry staining was performed as described58,59. Human PBMCs were labeled with the following monoclonal antibodies: anti-CD3 BV421 (UCHT1), anti-CD4 BUV737 (SK3), anti-CD8 BUV379 (RPA-T8) (BD Biosciences); anti-CD127 BV650 (A019D5), anti-CCR7 (CD197) PE-Cy7 (BioLegend); anti-CD25 FITC (2A3) (Becton Dickinson, BD Biosciences) and anti-CD45RA PerCP-Cy5.5 (HI100) (Invitrogen, eBioscience). For mouse lymphocytes, fluorochrome-conjugated antibody clones to the following surface antigens were used: CD4 (RM4-5), CD8 (53-6.7), CD21 (7G6), CD23 (B3B4), CD25 (PC61.5), CD44 (IM7), CD69 (H1.2F3), CD93 (AA4.1), CD45.2 (104), B220 (RA3-6B2), FOXP3 (FJK-165), IgM (II/41), IgD (11-26). B cells were stimulated with: F(ab′)2 goat anti-mouse IgM (10 μg ml−1; Jackson ImmunoResearch Laboratories), LPS from Escherichia coli 055:B5 (10, 1.0 or 0.1 μg ml−1; Sigma), anti-CD40 (10 μg ml−1 FGK4.5; BioXCell). TCR3A9-transgenic cells were detected with the 1G12 hybridoma (specific for TCR3A9). Data were acquired with LSRII and FORTESSA flow cytometers (BD) and analyzed using FlowJo software (Tree Star).

Mass cytometry

Unstimulated spleen cells from four wild-type mice and four mice homozygous for the I325N allele were individually labeled with mass barcodes, mixed, permeabilized and stained with mass-labeled antibodies to a panel of cell-surface markers and intracellular proteins including IκBα, and analyzed by CyTOF60. Spanning-tree progression analysis of density-normalized events (SPADE)61 analysis was used to resolve leukocyte lineages and subsets, and the relative intensity of IκBα in each subset was compared between mutant and wild-type cell counterparts.

Mass spectrometry analysis of A20 phosphorylation at S381

Human and mouse A20 proteins were reduced in sample buffer containing DTT (Sigma) and alkylated in 0.176 M N-isopropyliodoacetamide (synthesized in house, mass addition of 99.0684 Da to all cysteine residues). Samples were separated by SDS–PAGE and stained. Bands around 90 kDa in each lane were excised and digested with trypsin as previously described62. After overnight digestion, the peptides were extracted from gel slices in acetonitrile and the solution was evaporated to near dryness. Samples were then reconstituted in 10 μl of 0.1% formic acid containing 2% acetonitrile and analyzed by liquid chromatography and tandem mass spectrometry. Reconstituted peptides were injected via an autosampler onto a Dionex UltiMate 3000 RSLCnano Proflow system (Thermo Fisher Scientific) over a gradient of 2% buffer A (98% water and 2% acetonitrile with 0.1% formic acid) to 35% buffer B (98% acetonitrile, 2% water, 0.1% formic acid) and a flow rate of 450 nl min−1. Samples were separated over a 25-cm capillary column (100-μm internal diameter) packed with Waters nanoAcquity M-Class BEH (1.7 μm) material (New Objective). Samples were analyzed online via nanospray ionization into an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific). Precursor ion scans were performed in the Orbitrap (resolution of 60,000), and selected peptides were isolated using the quadrupole before fragmentation by high-energy collisional dissociation at a normalized collision energy of 30. The resulting fragment ions were analyzed in the ion trap (resolution of 15,000; AGC target = 2 × 10–4; max IT = 11 ms). Tandem mass spectrometry data were searched against candidate sequences using the Byonic and Byologic software packages (Protein Metrics) with variable modifications of oxidation (methionine) and phosphorylation (serine, threonine, tyrosine). AUCs were manually extracted from Xcalibur (Thermo Fisher Scientific).

Immunoblot analysis and immunoprecipitation

Protein concentration was measured using the Bradford assay (Bio-Rad), and total protein (20–25μg) was resolved on a 7–10% SDS–PAGE gel and then transferred to a nitrocellulose membrane, Immobilon-P (Merck Millipore). Membranes were incubated with anti-IκBα (9242), anti-phospho-IκBα (9256), anti-JNK (9252), anti-phospho-JNK (9255), anti-IKKα (2682), anti-phospho-IKKα/IKKβ (16A6; 2697), anti-NIK (4994), anti-NF-κB2 p100/p52 (4882), anti-RelB (C1E4, 4922, Cell Signaling Technology), anti-β-actin (AC15, Sigma-Aldrich) or anti-S381 A20 (ref. 19) followed by HRP-labeled secondary goat anti-mouse IgG Fc (Pierce Antibodies) or donkey anti-rabbit IgG (GE Life Sciences). HRP conjugates bound to antigen were detected and visualized by using an ECL detection kit (GE Life Sciences).

For immunoprecipitation anti-TNFR1 (ab7365) or anti-A20 (59A426) antibody (Abcam) were used. Membranes were incubated with anti-A20 (56305/D13H3), anti-IκBα (9242), anti-IKKβ (2684), anti-JNK (9252), anti-TAK1 (52065), anti-phospho-IκBα (9256), anti-phospho-JNK (9255), anti-phospho-TAK1 (4536/90C7), anti-TNFR1 (3736C25C1, Cell Signaling Technology), anti-RIP1 (H-207, Santa Cruz), anti-RIP1 (610458, BD Bioscience), anti-TAK1 (491840, R&D Systems) or anti-β-actin (AC15, Sigma-Aldrich). For B cell stimulation, splenic B cells purified by magnetic-activated cell sorting with CD43 depletion were stimulated with anti-IgM F(ab′)2 or LPS for the indicated times. Cells were lysed with TNE buffer (1% Nonidet P-40, 20 mM Tris-HCl (pH 8.0), 150 mM NaCl, 0.1 mM sodium orthovanadate and complete protease inhibitor; Roche). The following antibodies were used: anti-IκBα (9242), anti-TNFAIP3 (5630/D13H3, Cell Signaling Technology), anti-A20 (A-12), anti-ubiquitin (P4D1, Santa Cruz Biotechnology) and anti-β-actin (AC15, Sigma-Aldrich).

In vitro reporter transfection studies

Reporter assays were carried out as described previously63. NF-κB activity experiments were conducted using βTC3 cells transfected with 0.3 μg of the NF-κB.Luc reporter (Promega) and 0.2 μg of CMV.β-galactosidase. pcDNA vectors encoding human wild-type or variant A20 constructs or the empty pcDNA3.1 reporter was then added (0.5 μg) to make 1 μg total DNA. Transfection was conducted using Lipofectamine 2000 (Invitrogen). Following transfection, cells were stimulated with 200 U ml−1 of hTNF (R&D Systems). Luciferase activity was assayed in cell lysates collected 8 h after stimulation, using a luciferase assay kit (Promega). Results were normalized to β-galactosidase activity (Galactostar) to give relative luciferase activity. Expression plasmids and reporters were obtained and maintained as described previously63.

Ubiquitination assays

Wild-type A20, A20 I325N or A20 C103A OTU domain protein (1 μg) purified from E. coli was added to 2 μg of K48-linked ubiquitin chains of mixed chain length (Ub2–Ub7) or to purified tetra-ubiquitin (Ub4) (UC-230, UC-210, Boston Biochem). Recombinant FLAG-tagged full-length wild-type A20 or A20 I325N was expressed in HEK-293T cells and purified as described previously14,17. Full-length A20 deubiquitination reactions were performed using 100 ng of recombinant A20, 500 ng of the indicated ubiquitin chain and DUB reaction buffer with or without phosphatase inhibitor cocktail and were incubated for the indicated times at 37 °C with agitation at 1,000 r.p.m. Following incubation, samples were placed on ice and 20 μl was collected and added to SDS sample buffer to stop the reaction. A full-length ubiquitin ligase assay was performed as previously described14. Samples were subjected to one-dimensional SDS–PAGE and immunoblotted for ubiquitin (clone P4D1, Santa Cruz Biotechnology or Cell Signaling Technology) or A20 (clone 59A426, Abcam or A-12, Santa Cruz Biotechnology), as described above.

OTU protein preparation and crystallization

The N-terminal OTU domains of human and mouse A20 (human wild-type and I325N A20, residues 1–366; mouse wild-type and C103A A20, residues 1–360) were cloned into vector pGEX-6P-1 (GE Healthcare), facilitating bacterial expression as a glutathione S-transferase (GST) fusion. Expression was performed in E. coli strain BL21 (DE3) Gold, where cells were induced at an OD600 of 0.5 with 0.2 mM IPTG, followed by incubation at 20 °C overnight in LB medium. Eluted A20 was stabilized with iodoacetamide (Sigma; 30 mM, 30 min at 20 °C, terminated by addition of an equivalent amount of β-mercaptoethanol). For purification of mouse wild-type and C103A proteins, GST-I fusion protein was first eluted (50 mM Tris-HCl (pH 8.0), 200 mM NaCl, 5 mM DTT, 10 mM glutathione) and then the A20 OTU component was released by overnight incubation (4 °C) with PreScission protease. Cleavage products were subjected to anion-exchange chromatography (HiTrap Q FF, GE) where the OTU domain eluted as a single peak between 5 and 500 mM NaCl, 25 mM Tris-HCl (pH 7.5). All proteins were further purified by gel filtration chromatography (ÄKTA, S200 26/60, buffers as above).

Crystals of both the A20 I325N variant (long triangular rods) and wild-type A20 (long rods) OTU domains grew under the same conditions; equal volumes of protein (2.7 mg ml−1) and well solution (50 mM CaCl2, 100 mM MES (pH 6.0), 5% PEG1500) were combined in a hanging-drop setup. Crystals grew over several weeks at 20 °C. Cryoprotection was achieved by briefly (for 1–5 s) swimming crystals in mother liquor and doping with glycerol (25% (vol/vol) final) before being plunge vitrified in liquid nitrogen. For mouse wild-type A20, crystallization was achieved by hanging-drop vapor diffusion at a concentration of 8 mg ml−1 in 1.8–2.4 M NaCl, 0.1 M MES (pH 6–6.7). Crystals were soaked in mother liquor containing 30% ethylene glycol for 1 min and immediately vitrified in a nitrogen cryostream.

Crystallographic data reduction and model refinement

Diffraction data were collected at 100 K at beamline MX2 at the Australian Synchrotron and then processed using software as described64. Although grown under the same conditions, the crystals for the human wild-type and I325N A20 proteins were in different space groups. The data were highly anisotropic, resulting in poor completeness, low multiplicity and noisy electron density maps. In the case of the mouse OTU crystal, there was marked thermal diffuse scattering. See Supplementary Table 7 for data reduction and refinement statistics.

Structures were solved by molecular replacement using PHASER65. The search model was the A-chain of PDB entry 3DKB stripped of surface loops that displayed conformational variability in other PDB entries (2VFJ and 3ZJD). In the case of the human I325N data, the structure was originally solved in space group P312 with four molecules in the asymmetric unit. However, unaccounted-for electron density suggested a solution in the lower-symmetry P31 space group, with six molecules in the asymmetric unit, with sensible packing and no unaccounted-for density. The human wild-type A20 structure, solved as a control for iodoactamide alkylation, also has six molecules (three dimers) in a different asymmetric unit and space group. The mouse OTU structure was solved in the P32 space group, with a dimer in the asymmetric unit. Restrained B-factor refinement, using local noncrystallographic symmetry restraints, was performed with REFMAC5 (ref. 66). For the mouse A20 OTU structure, torsion liberation screw parameterization and restrained refinement were carried out using phenix.refine67. Between rounds of refinement, electron density maps and composite OMIT maps68 and their fit to the model were examined. Amino acid side chains were added if suggested by difference map electron density. Difference maps identified the active site cysteine (C103) as the only cysteine residue alkylated by the iodoacetamide treatment in both human structures 5V3B and 5V3P (PDB residue descriptor YCM; for full list see The six molecules refined in each structure are highly similar in fold, both to themselves and as compared to each other. All molecules form dimers with neighboring molecules, as observed in other crystal structures. Structure validation was performed using the MOLPROBITY web server69. The final human I325N, human wild-type and mouse wild-type structures contain Ramachandran-favored/outlier components of 88.64/0.00%, 88.13/2.87% and 82.13/7.87%, respectively.

Measurement of OTU domain thermal stability

Purified A20 OTU domains (0.4 mg ml−1) were exchanged into 100 mM NaCl, 5 mM DL-Dithiothreitol and 20 mM Na2HPO4 (pH 7.5). Circular dichroism data were collected with a Chirascan circular dichroism spectrometer (Applied Photophysics) in a 1-mm cuvette. The protein was heated from 20–90 °C at a rate of 1 °C min−1 while ellipticity was monitored at 220 nm. The thermally induced unfolding of A20 OTU domains was not reversible. The unfolding of the A20 OTU domains was described by a two-state model70 (equation (1))

$$y_{\mathrm{obs}} = \frac{{y_{\mathrm{n}} + m_{\mathrm{n}}T + \left( {y_{\mathrm{u}} + m_{\mathrm{u}}T} \right)\exp \left( {\frac{{\Delta H_{{\mathrm{vH}}}}}{{\mathrm{R}}}\left( {\frac{1}{{\mathrm{T}}} - \frac{1}{{T_{\mathrm{trs}}}}} \right)} \right)}}{{1 + {\mathrm{exp}}\left( {\frac{{\Delta H_{{\mathrm{vH}}}}}{{\mathrm{R}}}\left( {\frac{1}{{\mathrm{T}}} - \frac{1}{{T_{\mathrm{trs}}}}} \right)} \right)}}$$

where yobs is the observed elipticity and yn and yu are the elipticity values observed for the native and unfolded states, respectively. The mn and mu values are the linear temperature dependencies of yn and yu. ΔHvH is the apparent van’t Hoff enthalpy, R is the universal gas constant and Ttrs is the temperature at which the population of unfolded protein is 50%. Curves were fitted by nonlinear regression using GraphPad Prism v.6 (GraphPad Software).

Statistical methods

Results are expressed as mean ± s.d. or mean ± s.e.m. Statistical analysis was performed using Student’s t-test, ANOVA or log-rank test, where indicated.

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.