Effects of >23,000 amino acid variants on TYK2 signaling and protein abundance.

A, Schematic showing the reporter design for TYK2 DMS assays on IFN-α signaling (left) and protein abundance (right). BC, unique DNA barcode; GFP, green fluorescent protein; ISRE, interferon-stimulated response element. This panel was created using BioRender.com. B, Heatmaps showing the functional effects of each TYK2 variant, represented as Z-scores, for IFN-α signaling (left, 100 U/mL IFN-α condition) and protein abundance (right). Rows and columns of the heatmap correspond to amino acid position and variant identity, respectively. TYK2 protein domains are shown on the far right. GOF, gain-of-function; LOF, loss-of-function; WT, wild-type. C, Structures of the TYK2 kinase and pseudokinase domains (PDB: 4OLI) with residues colored blue to indicate positions that have at least two significant LOF variants at that position in the IFN-α signaling (left) or protein abundance (right) assay (FDR < 0.01).

Experimental conditions assayed by DMS

Identification of allosteric and other functionally important sites of TYK2.

A, Schematic showing approach to identify key functional sites by pinpointing TYK2 variants that impact IFN-α signaling but not protein abundance. LOF, loss-of-function. B, Scatterplot of individual TYK2 variant effects on IFN-α signaling activity (x-axis) and TYK2 protein abundance (y-axis). Variants highlighted in blue were classified as signaling-only LOF (see Methods for details). GOF, gain-of-function; WT, wild-type. C, Count of signaling-only LOF variants (horizontal axis) at each residue of TYK2 (vertical axis), shown alongside the functional domains of the protein (left). Regions important for TYK2 function are highlighted in red, including known catalytic, known allosteric, and novel allosteric sites predicted by this analysis. D, Structure of the TYK2 kinase and pseudokinase domains (PDB: 4OLI) with amino acid positions that have at least two significant signaling-only LOF variants colored blue. Novel allosteric sites predicted by this analysis are labeled with an asterisk (*). E, Structural model of inactive full-length TYK2 (colored as in D) bound to the intracellular tail of the IFNAR1 receptor (yellow). See Methods for details on how this composite model was constructed. This panel was created using BioRender.com.

Functional characterization of protein-drug interactions.

A, Conceptual schematic of how inhibitor dose influences variant effect interpretation. B, Volcano plot of variant effects under very high dose (>IC99) of BMS-986202, with significant drug-resistant variants colored pink (FDR < 0.01 used as significance threshold throughout figure). C, Surface representation of the TYK2 pseudokinase domain (PDB: 6NZP) colored as in B. D, Allosteric binding site of the TYK2 pseudokinase domain with inhibitor BMS-986202 bound (dark gray) and positions with significant drug-resistance colored pink (as in C) for increasing concentrations of BMS-986202. E, Scatter plot of variant effects (Z-statistic) in the presence of inhibitor (IC99 concentration), with variants colored by whether they exhibit resistance to BMS-986202 (pink), NDI-034858 (cyan), or both (orange). F, Heatmap of drug-resistant positions (x-axis) and variants (y-axis) colored as in E. G, Structure (PDB: 6NZP) of TYK2 pseudokinase domain (colored as in E). H, I, J, Variants that exhibit drug-potentiation to each inhibitor (at IC75 concentrations), displayed as in E, F, and G, respectively.

Human variants that protect against autoimmune disease reduce TYK2 abundance.

A, Odds ratios from comprehensive PheWAS for the TYK2 P1104A allele (see Methods), which confirm protective effect on many autoimmune phenotypes (see Supplementary Table S2 for list of phenotype abbreviations). B, Deep mutational scanning results for the IFN-α (x-axis) and protein abundance (y-axis) assays, as in Fig. 2B. Each point represents a unique missense or nonsense variant, and P1104A is highlighted in blue. GOF, gain-of-function; LOF, loss-of-function; WT, wild-type. C, TYK2 protein abundance measured using Western blotting of HEK293T and Kit225 cells harboring the wild-type (WT, gray) and P1104A (blue) alleles of TYK2. Data are shown normalized by the abundance of the wild-type allele for each cell type. Bars represent means, with error bars reporting standard error and P values determined by unpaired t-test. D, Effect sizes (beta) of association between P1104A allele and abundance of three proteins in human plasma from UK Biobank. Error bars are 95% confidence intervals. E, Rare variant dose-response curve for TYK2 variants found in the UK Biobank shows that TYK2 protein abundance (x-axis, from DMS assay) predicts the probability of autoimmune disease diagnosis (y-axis). See Methods for details. Gray shading indicates 95% confidence intervals.

IFN-α and abundance DMS reporters are TYK2-dependent.

A, IFN-α reporter assay dose-response curves for wild-type (WT) and catalytically inactive (K930E variant) TYK2. RLU, relative luciferase units. B, Flow cytometry data showing that the abundance DMS reporter distinguishes between WT and two strongly destabilizing variants (E154X and K930E).

Results of IFN-α and abundance DMS assays.

Heatmap showing results of the IFN-α (1, 10, and 100 U/mL stimulation conditions) and abundance DMS datasets generated in this study. Each cell represents a single variant allele colored by the magnitude of variant effect (Z-statistic). For each heatmap, the x-axis represents amino acid position, and the y-axis represents the induced amino acid change.

DMS results consistent with other variant effect assessments.

A, Bar plots quantifying the frequency of LOF (blue), neutral (gray) and GOF (red) variant effects across the IFN-α and abundance DMS datasets. B, Histograms of stop (red) vs. nonstop (gray) variant effects, where greater separation of these two distributions indicates greater dynamic range and power to detect intermediate variant effects. The x-axis represents variant effect measured as Z-statistic of the variant normalized to WT (i.e., WT = 0), and this convention is carried through all subsequent panels. C, Violin plots of IFN-α (left) and abundance (right) DMS variant effects (Z-statistic, y-axis), grouped and colored by AlphaMissense annotation. D, DMS variant effects (log2(fold-change) for a panel of 32 previously characterized TYK2 variants, grouped by whether they were predicted to be gain-of-function (GOF, top), neutral (middle), or loss-of-function (LOF, bottom). E, Violin plots of IFN-α (left) and abundance (right) DMS variant effects (Z-statistic, x-axis), grouped and colored by ClinVar annotation.

Intersecting signaling and abundance DMS data identifies allosteric and other functionally important sites.

A, Histogram quantifying the number of amino acid positions that have X signaling-only LOF variants (“count”). B, TYK2kinase-pseudokinase structure (PDB: 4OLI) with amino acid positions that have at least one significant signaling-only LOF variants colored blue.

Distinct drug-resistance and drug-potentiating profiles of TYK2 allosteric inhibitors.

A, Chemical structures of BMS-986202 (left) and NDI-034858 (right). B, Dose-response curves of allosteric inhibitors generated with the IFN-α reporter cell line. C, Structures of bound allosteric inhibitors BMS-98202 (left, PDB: 6NZP) and NDI-034858 (right, PDB: 8S9A) with residues that confer drug resistance colored pink or cyan, respectively. D, Structures of inhibitors with residues that confer drug-potentiation colored as in C.

Common TYK2 variants associated with autoimmune phenotypes reduce TYK2 abundance.

A, Effects (as z-scores) of two common partial loss-of-function missense variants of TYK2 on relatively well-powered autoimmune traits in UKBB. (**) denotes that the variant is the most likely causal variant for the trait at the 5E-08 threshold, while (*) denotes that the variant was denoted to be the most likely causal variant for the trait at the 5E-05 significance threshold. PsO, psoriasis; RA, rheumatoid arthritis; HT, hypothyroidism; UC, ulcerative colitis; IBD, inflammatory bowel disease; Pooled, autoimmune phenotypes combined; Pooled (no IBD), same as Pooled but excluding IBD cases. B, Example flow cytometry data showing variants with intermediate abundance (P1104A, L1014P) relative to WT and an early stop variant (E154X). C, Sanger sequencing traces confirming P1104A mutation in Kit225 cells. D, Representative western blot of WT and P1104A TYK2 in HEK293T and Kit225 cells. E, Deep mutational scanning results for the IFN-α (x-axis) and protein abundance (y-axis) assays, as in Fig. 2B. Each point represents a unique missense or nonsense variant, with P1104A and I684S variants colored blue and pink, respectively. F, Effect sizes (beta) of association between TYK2 alleles (P1104A, blue; and I684S, pink) and abundance of twelve proteins in human plasma from UK Biobank. All twelve proteins show a significant association with. P1104A, while only the top three show a significant association with I684S. Error bars are 95% confidence intervals.

Rare variants that reduce TYK2 abundance protect against autoimmune phenotypes.

A, Cumulative density plots quantifying the frequency of autoimmune disease in patient variants against the respective variant effect on protein abundance (left) or IFN-α signaling (right). Variants with very low abundance scores are uniquely associated with no autoimmune disease. B, Rare variant association test for TYK2 variants found in the UK Biobank, comparing probability of autoimmune disease (y-axis) against variant effects on TYK2 abundance (left, reproduced from main Fig. 4E) or IFN-α signaling (right), showing that reduced TYK2 protein abundance (left) is correlated with a reduced probability of autoimmune disease. Effects on IFN-α signaling (right) show no correlation with autoimmune disease.

Raw western blots used to validate reduced protein abundance of P1104A.

A—G, Images of all blots stained for TYK2 (top) and GAPDH loading control (bottom). For each lane, TYK2 band density was normalized to GAPDH band density. Lanes that were excluded are marked with a red X, determined by whether either band was outside of the linear detection range or contained blotting artifacts. Lanes are labeled by allele (WT or P1104A), with the cell background (HEK or Kit225) labelled above the alleles. See Methods for more details on how bands were quantified and normalized for analysis.