Complicated Common Variable Immunodeficiency is Driven by Aberrant IL-10/IL-21 Signaling and Predisposed Polygenic Risk

Subjects

All subjects provided informed consent for a research protocol approved by the Institutional Review Board of the University of California Los Angeles.

Stimulation Phospho-Mass cytometry

Heparinized blood samples were taken from CVID patients and age- and biological sex-matched controls.

In our studies, we sought to include the myeloid cells that might otherwise be removed in traditional gradient separation of PBMCs. We thus started our work with whole blood(20). However, to prevent serum IgM from competing with naïve B-cell-surface IgM for binding to anti-IgM antibodies, samples were first washed in PBS+1% bovine serum albumin (BSA). To prevent nonspecific binding of antibodies to Fc receptors, samples were then supplemented with 2 mM CaCl□ and incubated with FcX (BioLegend, 422302) according to manufacturer instructions. Samples were then stained with surface antibodies for 30 minutes at room temperature.

After incubation, samples were washed and then aliquotted for stimulation with a variety of individual stimuli: IL-2 (10 ng/mL, Peprotech, 200-02), IL-6 (10 ng/mL, Peprotech, 200-06), IL-10 (10 ng/mL, Peprotech, 200-10), IL-21 (50 ng/mL, Peprotech, 200-21), IFN-α (50 ng/mL, Cell Signaling Technology, 8927SC), IFN-γ (10 ng/mL, Peprotech, 300-02), R848 (1 μg/mL, Invivogen, vac-r848), PAM3CSK4 (5 μg/mL, Invivogen, tlrl-pms), LPS (0.5 μg/mL, Sigma-Aldrich, L4391), PMA/Ionomycin (10 ng/mL/1 μg/mL,Sigma-Aldrich, P1585, I3909)) or PBS as control at 37 °C for 15 minutes.

Stimulation was arrested and red blood cells were lysed together by using Lyse/Fix Buffer (BD Biosciences, 558049). Samples were then washed and permeabilized with methanol. After permeabilization, samples were stained for intracellular targets and incubated with Fix/Perm Buffer (Fluidigm, 201067) and 125 nM Iridium DNA intercalator (Fluidigm, 201192B). Samples were finally washed and then run on a Helios mass cytometer.

Antibodies used in the assay are available in supplementary Table S2.

Gating

The CytoExploreR package in R was utilized for gating (21). The gating scheme is shown (Fig. S3). Manual gates were drawn with the same cellular populations analyzed from date to date.

Gaussian Mixture Modeling

Bimodal phosphorylation responses were modeled using a 2-component Gaussian mixture modeling (GMM) framework using the expectation-maximization algorithm(22). GMMs were initialized with 2-component k-means clustering of phosphorylation responses. Up to 1000 iterations were performed on the data or if the difference between the log-likelihoods of two iterations reached less than 10^6. Models were constrained such that if two components were within 1 unit of each other or if one component represented less than 5% of the data, it was replaced by the larger model. Variance was also restricted to greater than .01. Values above 0 were considered for model fitting. Examples are included in Fig. S4.

Batch Effect Correction

Phosphorylation values from different experimental dates were adjusted to allow for accurate comparisons despite batch effects. The unstimulated Gaussian mixture model phosphorylation values from a date’s control were used to adjust each cell type-signaling pathway combination for all stimuli as was previously described(16).

Values of zero were removed from the data and are common statistical noise in all CyTOF datasets.

Exome sequencing

DNA from whole blood was extracted and then sent to Psomagen for whole-exome sequencing using the Agilent capture kit.

Polygenic Risk Score Calculations

VCFs of patients with CVID from the European Genome-Phenome Archive were processed into CSVs, then additively joined to recently described polygenic risk scoring files(23).

Gene ontology analysis

SnpXplorer was used to analyze SNP-sets for gene ontology, annotation, and relation to GWAS traits(24).

Statistical analysis

In all applicable comparisons, bootstrapping was used to generate reference distributions of given statistics. This mitigates the contribution of outliers to the generation of a statistic and allows for the calculation of 95% confidence intervals. Significance testing was done using permutation testing.

Code Availability

Scripts are hosted at https://github.com/humzalikhan/CVID_CyTOF_PRS or otherwise are available upon request.

Patient cohort and disease scoring

We performed Cytometry by Time of Flight (CyTOF/mass cytometry) on whole blood on 12 patients with CVID (Fig. 1).

• Download figure
• Open in new tab

Fig. 1.

Schematic of study design

Whole exome sequencing was also performed. Our cohort represented a diverse group, with 5 males and 7 females with an average age of 44 years (Fig. 2a). Only one patient had a known pathogenic variant in genes associated with monogenic cause for CVID (NFKB1 haploinsufficiency); others had variants of unknown significance in other genes associated with CVID such as AIRE, MALT1, and RASGRP3 (data not shown).

Given the heterogenous presentation of immune dysregulation in CVID, we used the recently described immune deficiency and dysregulation activity (IDDA) score to assess the extent of a patient’s non-infectious complications within our cohort (25,26). Patients above an IDDA score of 12 were considered IDDAhi, and others were considered IDDAlo. Patients who were classified as IDDAhi had significantly higher incidence of autoimmune cytopenias, enteropathies, lymphoproliferation, and lung disease (Fig. S1). Of note, our cohort did contain one outlier patient, with especially high disease burden and high IDDA score. To avoid excessive value to this one individual in downstream analyses, we employed bootstrapping/permutation-based analyses. Overall, we have a varied patient population composed of individuals with and without CVIDc.

Cellular subsets correlated with disease burden in CVID

We performed resampled correlation analysis between cell frequencies in patients and IDDA scores. We found that patients with higher frequencies of CD21lo B cells among all B cells were highly correlated to higher disease burden (Fig. 2b). CD21lo B cells are transitional B cells that are autoreactive and contribute to autoimmunity (27–30). This finding supports previous work demonstrating the link between CD21lo B cells and CVIDc (15). Furthermore, the frequencies of all subsets of monocytes (CD16lo, mid, and hi) were positively associated with higher disease activity. We view high monocyte frequencies as a proxy for monocyte activation, as monocyte activation induces effector differentiation and proliferation (31). Indeed, monocyte activation and proliferation has been related to NICs in CVID (13,14). Higher proportions of switched memory B cells (CD27+ IgD-IgM-), regulatory T cells, and naïve T cells were associated with lower IDDA scores (Fig. 2c-d). These findings support that low IDDA scores may be driven by the presence of switched memory B cells, which have undergone receptor editing and lack polyreactivity, or regulatory T cells, which also oppose autoimmunity.

• Download figure
• Open in new tab

Fig. 2.

Cohort description and cellular frequencies associated with disease course in CVID. a) distribution of age, sex and Immune deficiency and dysregulation activity of CVID cohort. b. Correlation analysis of cell type frequencies and IDDA scores. c. Cell subtypes with differential abundance between IDDAhi and IDDAlo patients. Circle sizes are relative and relate the proportions of each cell type among peripheral blood lymphocytes. d. Proportions of each identified cell type in c.

We compared frequencies of immune cells between healthy controls, IDDAhi patients, and IDDAlo patients. IDDAhi patients had much lower percentage of switched memory B cells and higher percentage of CD21lo B cells in comparison with both healthy controls and IDDAlo patients. Additionally, FOXP3+ regulatory T cells were diminished in IDDAhi patients (Fig. 2c-d). These findings are in line with the previous observations and support the hypothesis that autoimmunity in CVIDc develops via lack of proper immune regulation by regulatory T cells and an expansion of the CD21lo immature B cell subset.

Overresponsive IL-10/IL-21 Signaling found in CVIDc

Since overresponsive cellular signaling has previously been linked to autoimmunity, we investigated cellular signaling in CVIDc using a multiplexed stimulation assay using phospho-CyTOF (32).

We employed heatmap analysis of these results to visualize the separation between healthy controls and IDDA-stratified CVID patients. We find a clear distribution of these responses, with healthy controls and IDDAlo patients barely responding to IL-10 in the STAT1/STAT3 pathways, while IDDAhi patients signaled much more strongly than either of the other subgroups (Fig. 3a). Similar results were found for IL-21 and STAT1/STAT3 (Fig. 3b).

• Download figure
• Open in new tab

Fig. 3.

Profound signaling defects in the IL-10 and IL-21 axes are correlated non-infectious complications in CVID. A. Heatmap indicating aberrant IL-10 driven pSTAT1 and pSTAT3 signaling. B. Heatmap indicating aberrant IL-21 driven pSTAT1 and pSTAT3 signaling. C. T cell subsets associate with aberrant IL-10 signaling in CVIDc.

Furthermore, IL-10 stimulation induced strong decreases in phosphorylation of AKT, CREB, and S6 in healthy controls and IDDAlo patients. However, in IDDAhi patients had increased levels of phosphorylated AKT, CREB, and S6 in response to IL-10 (Fig. S5).

We then performed hypothesis testing of which cell-type subsets displayed aberrant signaling in CVIDc. After stimulating cells with IL-10, we found that patients with CVIDc (IDDAhi patients) had increased responses in the phosphorylation of STAT1 and STAT3 in many T cell subsets and natural killer cells (Fig. 3c). IL-21 also induced similar over-responsivity of STAT1 and STAT3 phosphorylation in these cell types (Fig. 3b, Fig. S5).

An opposite effect was found with the phosphorylation of S6, while also including CD16hi/lo monocytes, monocytoid dendritic cells, and neutrophils (Fig. S5).

Eosinophil and Allergy/Eczema Polygenic Risk Scores Confer Higher Likelihood of Developing Autoimmune NICs in CVID

To expand the findings in our small local cohort of patients, we utilized genome sequencing data of IEI patients from the National Institute for Health and Care Research (NIHR) Bioresource. This cohort included 451 patients with CVID and phenotyping data on their comorbidities (380 with full phenotyping on autoimmunity). This large group represents around 1/3 of all CVID patients registered in the UK(33).

We explored polygenic risk scores (PRS) to find the effect of common variants related to some traits. We performed odds-ratio analysis, focusing on PRSs where patients in the 9^th and 10^th deciles for a given PRS had high odds-ratios for autoimmune complications in CVID. We chose to examine autoimmune complications specifically given that this is a particularly discriminating feature between CVID and CVIDc. We found two PRSs that satisfied our decile criterion: one for eosinophil count and one for diagnosed allergy/eczema (Fig. 4a, Fig. 4b).

• Download figure
• Open in new tab

Fig. 4.

Eosinophil count and allergy/eczema polygenic risk scores associate with complicated CVID. A. Schematic of computational analysis of NIHR CVID cohort. B. CVID patients with autoimmunity are enriched in the top two deciles of eosinophil count/allergy and eczema polygenic risk scores.

We then performed gene ontology and genome-wide association study (GWAS) catalogue analysis on the 10,000 most contributing SNPs of the two relevant PRS variant sets. The two variant sets are not orthogonal and contain some of the same SNPs. This is seen with many genes sharing the same ontology. We found that SNPs from both PRSs related to the GWAS catalogue traits of type II diabetes, hemoglobin measurements, and lymphocyte count indicating a strong inflammatory signal in CVIDc.