Convergence of cytokine dysregulation and antibody deficiency in common variable immunodeficiency with inflammatory complications

Abstract

Background:

Noninfectious complications are the greatest cause of morbidity and mortality in common variable immunodeficiency (CVID), but their pathogenesis remains poorly defined.

Objective:

Using high-throughput approaches, we aimed to identify, correlate, and determine the significance of immunologic features of CVID with noninfectious complications (CVIDc).

Methods:

We simultaneously applied proteomics, RNA sequencing, and mass cytometry to a large cohort with primary antibody deficiency.

Results:

CVIDc is differentiated from uncomplicated CVID, other forms of primary antibody deficiency, and healthy controls by a distinct plasma proteomic profile. In addition to confirming previously reported elevations of 4–1BB, IL-6, IL-18, and IFN-γ, we found elevations of colony-stimulating factor 1, IL-12p40, IL-18R, oncostatin M, TNF, and vascular endothelial growth factor A to differentiate CVIDc. This cytokine dysregulation correlated with deficiency of LPS-specific antibodies and increased soluble CD14, suggesting microbial translocation. Indicating potential significance of reduced LPS-specific antibodies and resultant microbial-induced inflammation, CVIDc had altered LPS-induced gene expression matching plasma proteomics and corresponding with increased CD14⁺CD16⁻ monocytes, memory T cells, and tissue inflammation ameliorated by T-cell–targeted therapy. Unsupervised machine learning accurately differentiated subjects with CVIDc and supported cytokine dysregulation, antibody deficit, and T-cell activation as defining and convergent features.

Conclusions:

Our data expand understanding of CVIDc proteomics, establish its link with deficiency of IgA and LPS-specific antibodies, and implicate altered LPS-induced gene expression and elevated monocytes and T cells in this cytokine dysregulation. This work indicates that CVIDc results when insufficient antibody neutralization of pathogen-associated molecular patterns, like LPS, occurs in those with a heightened response to these inflammatory mediators, suggesting a 2-hit model of pathogenesis requiring further exploration.

Graphical abstract

Common variable immunodeficiency (CVID) is the most prevalent symptomatic primary immunodeficiency and defined by profoundly impaired antibody production.^1,2 About half of those with CVID develop chronic complications such as autoimmunity and inflammation of the gastrointestinal tract, lungs, or other tissues.³ Genetic etiologies have been identified in about 25% cases of CVID, but the mechanisms underlying the inflammatory disease manifestations remain unclear.^4,5 Because they are the major source of morbidity and mortality in CVID, understanding the pathogenic basis of noninfectious complications is imperative.^6,7

Elevated type 1, or T_H1, cytokines have been associated with CVID with noninfectious complications (CVIDc). This includes increased IL-12 and IFN-γ–producing cells, increased IL-12 and IFN-γ in blood, and greater expression and phosphorylation of signal transducer and activator of transcription 1 by monocytes, the key transcription factor activated by IFN-γ.^8–15 Yet, the reason for elevated type 1 cytokines in CVIDc is unknown. Endotoxemia due to deficiency of mucosal antibodies may drive this cytokine response.^16,17 Nevertheless, noninfectious complications occur less frequently in those with selective IgA deficiency and X-linked agammaglobulinemia despite having similar (or more severe) mucosal antibody loss, indicating that undefined factors beyond immunoglobulin deficiency contribute.^18,19

Nuclear factor kappa B signaling induced by pathogen-associated molecular patterns (PAMPs), most prominently LPS, is fundamental for the induction of IL-12, a major driver of the type 1 cytokine response.^20,21 Altered function of Toll-like receptors, which recognize PAMPs, has been reported in CVID, but the significance is unclear.^22,23 Furthermore, an association with reduced isotype-switched memory B cells and the presence of noninfectious complications of CVID has long been established, but the reason for this link is unknown.²⁴ Leveraging a large primary antibody deficiency (PAD) cohort and using high-throughput approaches as well as machine learning, we aimed to reconcile the numerous immunologic observations of CVID into a unified model of pathogenesis.

METHODS

Subjects

All subjects were patients at Boston Medical Center and/or Mount Sinai. Diagnosis of CVID was defined as markedly low serum IgG and IgA and/or IgM (IgG < 400 mg/dL, IgA < 45 mg/dL, or IgM < 35 mg/dL), poor response to at least 1 vaccine, and exclusion of other causes of hypogammaglobulinemia.² CVIDc required a history of autoimmunity, inflammatory bowel disease, interstitial lung disease, and/or chronic liver disease. Antibodies used for immunohistochemistry are listed in Table E1 in this article’s Online Repository at www.jacionline.org. This study was approved by the institutional review boards of the Boston University School of Medicine and the Icahn School of Medicine at Mount Sinai. Research was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki). Written informed consent was received from participants before inclusion.

TABLE E1.

Antibodies used for immunohistochemistry

Marker	Clone	Company
CD3	LN10	Leica
CD4	4B12	Leica
CD8	4B11	Leica
CD20	L26	Leica

Cell culture

PBMCs were purified from fresh venous blood using Ficoll density gradient centrifugation. Cells were resuspended in culture media (RPMI/10% HI FBS [Gibco, Grand Island, NY]/0.1% Antibiotic-Antimycotic [Gibco]) to a final concentration of 1 to 5 × 10⁵ cells/100 μL and then added to a sterile, 96-well, flat-bottom cell culture plate (Corning, Corning, NY). LPS or TNF was added at 20 ng/μL for 18 hours where indicated.

Cytokine, chemokine, soluble CD14, and antibody measurement

The Olink Target 96 Inflammation panel was used to provide qualitative measurement of 92 analytes in subject plasma before any immunomodulatory therapy other than immunoglobulin replacement (https://www.olink.com/products/inflammation/). Plasma CXCL10, IFN-γ, IL-12p40, and TNF were also measured by multiplex assay on the Luminex 200 system. Soluble CD14 (sCD14), IL-6, IL-10, and TNF in plasma or cell supernatants were also measured by ELISA (R&D Systems, Minneapolis, Minn). Plasma LPS-specific antibodies were measured as previously.²⁵

RNA sequencing

Total RNA was extracted from cultured PBMCs using RNeasy Mini kits (Qiagen, Germantown, Md). One microgram of RNA was used for preparing RNA sequencing library with the TruSeq stranded mRNA kit (Illumina, San Diego, Calif) following manufacturer’s protocol. The xGene dual index UMI adaptors (Integrated DNA Technology, Coralville, Iowa) were used for barcoding samples. Sequencing was done on the Illumina NextSeq 550 system. Data underwent demultiplexing, followed by transcript alignment (STAR) and sequence deduplication (UMI-tools).^26,27 Final transcript counts of each sample were obtained using HTSeq.²⁸ Statistical analysis of gene counts data was carried out using DESeq2 in R software.²⁹ A 2-factor linear model with interaction effects was constructed to test the hypothesis that cultured PBMCs from subjects with different disease phenotypes show distinct gene expression profiles in response to LPS. Generalized linear regressions with the negative binomial residual model were performed on gene count data. We also carried out the likelihood ratio test using the same model to confirm the results. Significant genes were selected for those that showed differential expressions in response to LPS among different disease groups (with false discovery rate-adjusted P value <.1 and difference larger than 2-folder). Gene set enrichment analysis was performed to identify impacted pathways.³⁰

Mass cytometry

Whole blood samples from 10 patients with CVIDc, 10 patients with CVID without complications, and 10 healthy controls (HCs) selected at random were analyzed as previously; antibodies are listed in Table E2 in this article’s Online Repository at www.jacionline.org.¹³ Differential abundance analysis and cell subset definitions were done as previously.^31–35 Cluster labeling, method implementation, and visualization were done using a computational approach through the Astrolabe Platform (https://astrolabediagnostics.com/). Marker expression assigned to each leukocyte subset is summarized in Fig E1 in this article’s Online Repository at www.jacionline.org.

TABLE E2.

Antibodies used for mass cytometry

Metal	Marker	Clone	Manufacturer
89Y	CD45	HI30	Fluidigm
113In	CD57	HCD57	Biolegend
115In	CD11c	Bu15	Biolegend
141Pr	IgD	IA6-02	Biolegend
142Nd	CD19	HIB19	Biolegend
143Nd	CD45RA	HI100	Biolegend
144Nd	CD103	Ber-Act8	Biolegend
145Nd	CD4	RPA-T4	Biolegend
146Nd	CD8	RPA-T8	Biolegend
148Nd	CD16	3G8	Biolegend
149Sm	CD127	REA614	Miltenyi
150Nd	CD1c	L161	Biolegend
151Eu	CD123	6H6	Biolegend
152Sm	CD66b	G10F5	Biolegend
154Sm	CD86	IT2.2	Biolegend
155Gd	CD27	O323	Biolegend
158Gd	CD33	WM53	Biolegend
160Gd	CD14	M5E2	Biolegend
161Dy	CD56	B159	BD Biosciences
162Dy	CD20	2H7	Biolegend
163Dy	Anti-APC / APC BAFFR	APC003 / 8A7	Biolegend / Biolegend
166Er	CD169	7–239	Biolegend
168Er	CD3	UCHT1	Biolegend
170Er	CD38	HB-7	Biolegend
171Yb	CD161	HP-3G10	Biolegend
174Yb	HLADR	L243	Biolegend
209Bi	CD11b	ICRF44	Fluidigm

FIG E2.

Plasma IL-10 is increased in CVIDc, uncomplicated CVID, and other PAD compared with HCs. HIgM, Hyper-IgM syndrome. Plasma IL-10 measured by ELISA. Purple diamonds denote subjects with XLA, and teal diamonds denote subjects with HIgM. P value calculated by Kruskal-Wallis test. XLA, X-linked agammaglobulinemia. **P < .01, ****P < .0001.

Unsupervised machine learning

We applied a partition around mediods clustering algorithm and multiple factor analysis (MFA) to our data from subjects with CVID (complicated and uncomplicated) and HC subjects, consisting of plasma cytokines and chemokines, total and LPS-specific antibody responses, and peripheral blood leukocyte immunophenotyping. Partition around medoids and MFA were done in R using the cluster and FactoMineR packages. To visualize the cluster results, Rtsne was used to map high-dimension data into 2-dimension coordinates.

Statistics

Categorical values were compared using the Fisher exact test. Continuous values were compared using Mann-Whitney test for 2 groups. For multiple group comparison, Kruskal-Wallis test was used for 1-way ANOVA comparisons and Sidak or Tukey multiple comparison’s test was applied when 2-way ANOVA was used. Correlation was defined via Spearman rank coefficient.

RESULTS

Elevation of cytokines and molecules involved in T-cell function in CVIDc

We used the Olink Target 96 Inflammation panel to measure plasma proteins from 79 subjects, including 26 subjects with CVIDc with no known genetic etiology after evaluation by whole-exome sequencing (in 19 subjects) and the 207 gene primary immunodeficiency panel (Invitae) (in 4 subjects)³⁶ and 31 subjects with uncomplicated CVID (Table I), as well as 12 age- and sex-matched HCs and 10 subjects with other forms of PAD (3 with hyper-IgM syndrome, 3 with selective IgA deficiency, and 4 with X-linked agammaglobulinemia; Table II). We found elevation of IL-12p40, IL-6, IL-18, IL-18 receptor, colony-stimulating factor 1, lymphotoxin alpha, the TNF superfamily member LIGHT, oncostatin M, and vascular endothelial growth factor A in CVIDc compared with uncomplicated CVID, HC, and the other forms of PAD tested (Fig 1, A). Numerous proteins involved in T-cell function were also higher in CVIDc including 4–1BB, CD40, CD5, CD6, and FMS-like tyrosine kinase 3 ligand. We were unable to measure IFN-γ or TNF by Olink. Using multiplex cytokine measurement by Luminex (Invitrogen, Waltham, Mass) and ELISA, we found elevation of IFN-γ and TNF as well as confirmed elevations of IL-12p40 and IL-6 in CVIDc relative to uncomplicated CVID, other PAD, and HC (Fig 1, B). In contrast to the elevations of proinflammatory cytokines distinguishing CVIDc, IL-10 was similarly elevated in all forms of PAD relative to HC (see Fig E2 in this article’s Online Repository at www.jacionline.org).

TABLE I.

Clinical and laboratory characteristics of subjects with CVID

Characteristic	CVIDc	CVID	P value*
No. of subjects	26	31
Sex: female (%)	15 (58)	18 (58)	>.999
Median age (y) (95% CI)	45.5 (39–52)	48 (38–54)	.636
Subjects (%) with history of:
Sinusitis	14 (54)	24 (77)	.091
Pneumonia	14 (54)	19 (61)	.601
Bronchiectasis	6 (23)	4 (13)	.486
Median lab value (95% CI)
Diagnostic IgG (mg/dL)	226 (100–345)	233 (89–366)	.922
Diagnostic IgA (mg/dL)	7 (0–13)	25 (7–49)	.045
Diagnostic IgM (mg/dL)	13.5 (0–40)	20.5 (6–33)	.595
Neutrophils (103/μL)	3.1 (2.6–4.4)	3.3 (2.8–4.1)	.846
Lymphocytes (103/μL)	1.0 (0.8–1.2)	1.5 (1.2–1.8)	.004
CD4+ T cells (/μL)	472 (333–781)	679 (504–857)	.058
CD8+ T cells (/μL)	224 (138–291)	428 (321–627)	.0004
CD16+CD56+ (/μL)	69.5 (48–135)	109 (78–222)	.127
CD19+ B cells (/μL)	65 (25–139)	190 (138–328)	<.0001

^*Categorical values were compared using Fisher exact test. Continuous values were compared using Kruskal-Wallis test. Values below the level of detection of the laboratory assay were recorded as 0. Significant P values are in boldface.

TABLE II.

Clinical and laboratory characteristics of subjects with other PAD

Characteristic		HIgM			IgAD				XLA
Gene	CD40LG	CD40LG	AICDA		NA				BTK
Mutation (cDNA location)	580delG	632C>A	*		NA		†	†	138C>T	†
Sex	M	M	F	M	M	F	M	M	M	M
Age (y)	19	40	41	73	35	15	41	37	11	19
History of:
Sinusitis	X	X	X	X	X	X		X	X
Pneumonia			X	‡		X	X	X
Bronchiectasis								X
ILD
IBD					X					X
Liver disease		X
Autoimmunity				X
Lab values
Diagnostic IgG (mg/dL)	374	NA	<51	779	871	621	272	NA	410	NA
Diagnostic IgA (mg/dL)	23	<5	5	<5	<7	<5	<7	<5	<7	<5
Diagnostic IgM (mg/dL)	707	1543	1367	45	60	15	<5	<5	<4	<5
Neutrophils (103/μL)	2	5.2	3.5	5.2	4.6	3.9	0.8	4.5	3.0	2.9
Monocytes (103/μL)	0.4	0.8	0.3	0.7	0.5	0.6	0.2	1.3	0.7	0.7
Lymphocytes (103/μL)	1.6	1.5	1.5	1.3	1.6	2.7	1	1.3	1.9	0.9
CD4+ T cells (/μL)	850	NA	NA	726	818	NA	NA	710	1103	NA
CD8+ T cells (/μL)	332	NA	NA	1462	838	NA	NA	475	442	NA
CD16+CD56+ NK cells (/μL)	166	NA	NA	190	229	NA	NA	55	62	NA
CD19+ B cells (/μL)	351	NA	NA	64	240	NA	NA	0	3	NA

F, Female; HIgM, hyper-IgM syndrome; IgAD, selective IgA deficiency; ILD, interstitial lung disease; M, male; NA, not available; NK, natural killer; SARS-CoV-2, severe acute respiratory syndrome coronavirus 2; XLA, X-linked agammaglobulinemia.

^*Genetic analysis failed to amplify exon 1 of AICDA.

^†Diagnosis confirmed by flow cytometry.

^‡No previous history of pneumonia at the time of study participation, but patient died because of respiratory failure in the context of SARS-CoV-2 pneumonia.

FIG 1.

Elevation of cytokines and molecules involved in T-cell function in CVIDc. A, Plasma cytokines by Olink in uncomplicated CVID, CVIDc, XLA, HIgM, IgAD, and HCs. B, Plasma IL-12, IFN-γ, and TNF by Luminex, IL-6 by ELISA. XLA (purple diamonds), HIgM (teal diamonds), IgAD (orange diamonds). HIgM, Hyper-IgM syndrome; IgAD, selective IgA deficiency; XLA, X-linked agammaglobulinemia. **P < .01, ***P < .001, **** P < .0001.

Cytokine and sCD14 elevation corresponds with greater antibody defect in CVID

Elevation of sCD14 marks microbial translocation, a process facilitated by immune deficiency that may drive inflammation via endotoxemia in HIV and other diseases.³⁷ Plasma sCD14 was elevated in CVIDc relative to uncomplicated CVID, other forms of PAD, and HCs as well as higher in uncomplicated CVID relative to HCs (Fig 2, A). We were unable to demonstrate differences in plasma LPS, consistent with previous reports (data not shown).^38,39 IgA is the prominent isotype at mucosal surfaces and may prevent translocation of microbial products into circulation.⁴⁰ We found serum IgA to be significantly reduced in CVIDc (Fig 2, B) and to have modest negative correlation with sCD14 (r = −0.365; P = .0053; Fig 2, C). Correspondingly, LPS-specific IgA was significantly reduced in CVIDc (Fig 2, D) and more strongly correlated with sCD14 (r = −0.447; P = .0005; Fig 2, E). Although there were no differences in total serum IgM (see Fig E3, A, in this article’s Online Repository at www.jacionline.org) or IgG while patients were on IgG replacement therapy (Fig E3, B), we found that LPS-specific IgM was significantly reduced in CVIDc relative to uncomplicated CVID and HCs (Fig 2, F), in correlation with sCD14 (r = −0.386; P = .003; Fig 2, G). There was no difference in LPS-specific IgG detected between groups (Fig E3, C). Plasma sCD14 correlated with IL-12 and TNF levels in CVID (Fig E3, D and E), but not in other PAD (Fig E3, F and G). Thus, we saw decreased levels of total IgA and LPS-specific IgA and IgM correlating with levels of sCD14 in CVIDc, with sCD14 correlating with plasma IL-12 and TNF levels in CVID. Notably, these cytokines did not correlate with sCD14 in other forms of PAD, indicating that this relationship was a distinguishing feature of CVID.

FIG 2.

Cytokine and sCD14 elevation corresponds with greater antibody defect in CVID. A, Plasma sCD14. XLA (purple diamonds), HIgM (teal diamonds), IgAD (orange diamonds). B, Serum IgA. C, IgA and sCD14 correlation. D, Plasma LPS-specific IgA. E, LPS-specific IgA and sCD14 correlation. F, Plasma LPS-specific IgM. G, LPS-specific IgM and sCD14 correlation. H, LPS-specific IgA and IL-12 correlation. I, LPS-specific IgM and IL-12 correlation. J, Isotype-switched memory B cells and IL-12 correlation. HIgM, Hyper-IgM syndrome; IgAD, selective IgA deficiency; XLA, X-linked agammaglobulinemia. Red triangles denote CVIDc, and blue triangles denote uncomplicated CVID. *P < .05, **P < .01, ***P < .001, ****P < .0001.

FIG E3.

Relationship of plasma antibodies and cytokine levels. (A) Plasma IgM and (B) IgG. C, Plasma levels of LPS-specific IgG. (D) Correlation of sCD14 with IL-12 and (E) TNF. Plasma TNF levels were higher in those with lower LPS-specific IgM. Red dots denote subjects with CVIDc, and blue dots denote subjects with uncomplicated CVID. (F) No correlation between plasma sCD14 and IL-12 and (G) TNF in other PAD (purple denotes XLA, teal denotes HIgM, orange denotes IgAD). H, Correlation of plasma LPS-specific IgA with TNF. I, Correlation of plasma LPS-specific IgM with TNF. HIgM, Hyper-IgM syndrome; IgAD, selective IgA deficiency; r, Spearman rank correlation coefficient; XLA, X-linked agammaglobulinemia.

We next examined whether there was correlation between the extent of antibody deficiency and cytokine dysregulation in CVID. LPS-specific IgA showed negative correlation with IL-12 (r = −0.366; P =.0051; Fig 2, H) and TNF (r = −0.426; P = .0009; Fig E3, H). LPS-specific IgM also negatively correlated with IL-12 (r = −0.564; P < .0001; Fig 2, I) and TNF (r = −0.491; P = .0001; Fig E3, I). Isotype-switched memory B cells (IgD⁻CD27⁺CD19⁺) are known to be reduced in CVIDc relative to uncomplicated CVID.²⁴ We found strong negative correlation between isotype-switched memory B cells and plasma IL-12 (r = −0.742; P < .0001; Fig 2, K) and TNF (r = −0.766; P < .0001; data not shown). Our results demonstrate that cytokine and sCD14 elevation corresponds with greater antibody defect in CVID, including deficiency of antibodies that may neutralize endotoxemia, such as IgA, LPS-specific IgA, and LPS-specific IgM.

Altered LPS-driven leukocyte gene expression mirrors plasma cytokine dysregulation in CVIDc

Correlation of CVIDc cytokine dysregulation with loss of LPS-specific antibodies led us to test LPS-induced gene expression. We performed RNAseq on unstimulated and LPS-stimulated PBMCs and found gene expression of CVIDc to have a pattern distinct from that of uncomplicated CVID and HCs, whereas uncomplicated CVID and HCs had similar patterns (Fig 3, A). Gene set enrichment analysis revealed that type 1 cytokine and inflammatory pathways were among the most elevated in CVIDc, including pathways related to IFN-γ, IL-6, and TNF (Fig 3, B) along with IL1A, IL1B, IL6, IL23A (IL-23 shares the IL-12p40 subunit with IL-12), oncostatin M, and vascular endothelial growth factor A (Fig 2, C). RNAseq results were confirmed using NanoString nCounter analysis.⁴¹ Because apoptosis and TNF signaling were highly elevated gene sets (Fig 3, B), we measured TNF-driven apoptosis of PBMCs and found it to be increased in CVIDc (see Fig E4 in this article’s Online Repository at www.jacionline.org), validating another RNAseq finding. In summary, altered LPS-driven gene expression in CVIDc matched plasma proteomics, illustrating the potential significance of reduced LPS-specific antibodies we found in these patients.

FIG 3.

Altered LPS-driven leukocyte gene expression mirrors plasma cytokine dysregulation in CVIDc. A, Global gene expression effect of LPS stimulation on PBMCs, showing CVIDc differs from the pattern shared between uncomplicated CVID and HCs. B, Gene set enrichment analysis comparing CVIDc against uncomplicated CVID and HCs grouped together. C, Heat maps of the hallmark inflammatory response and TNF signaling via NF-kB gene sets. *P < .05, **P < .01, ***P < .001.

FIG E4.

TNF-induced cell death of PBMCs. Cell death measured by Cell Death Detection ELISA^PLUS (Roche, Basel, Switzerland) photometric detection of mono- and oligonucleosome fragments indicative of apoptosis after 18 hours of culture with 20 ng/mL TNF. P value calculated by Tukey multiple comparisons test. **P < .01, ***P < .001.

Cytokine dysregulation of CVIDc corresponds with expansion and activation of circulating monocytes

Cell-type deconvolution of our RNAseq data suggested that the findings corresponded with an increase in monocytes in CVIDc (see Fig E5, A, in this article’s Online Repository at www.jacionline.org).⁴² Using mass cytometry to more definitively measure leukocyte populations in whole blood, we found the proportion of CD14⁺CD16⁻ (classical) monocytes to be significantly elevated in CVIDc compared with either HCs or uncomplicated CVID (Fig 4, A and B). We also found CD1c⁻ (conventional) dendritic cells to be increased in CVIDc compared with uncomplicated CVID (Fig 4, A and C). We did not find differences in other dendritic cell or monocyte subsets or neutrophils, natural killer cells, and plasmacytoid dendritic cells. Proportion of CD14⁺CD16⁻ monocytes in the blood strongly correlated with plasma TNF (Fig 4, D; r = 0.907; P < .0001) and plasma IL-12 (Fig E5, B; r = 0.803; P < .0001). Demonstrating greater monocyte activation, we found expression of the costimulatory molecule CD86 to be significantly higher on all monocyte subsets in CVIDc compared with uncomplicated CVID (Fig 4, D and E). HLA-DR was also significantly higher on CD14⁺CD16⁻ monocytes from CVIDc compared with uncomplicated CVID. Thus, the cytokine dysregulation of CVIDc corresponds with expansion and activation of monocytes.

FIG E5.

Additional data supporting relationship of monocyte expansion and activation with cytokine dysregulation in CVIDc. A, Leukocyte subsets by cell-type deconvolution of RNAseq. **P < .01. B, Correlation of plasma IL-12 with circulating CD14⁺CD16⁻ monocytes. Red triangles denote subjects with CVIDc, blue triangles denote subjects with uncomplicated CVID, and green triangles denote HCs.

FIG 4.

Cytokine dysregulation of CVIDc corresponds with expansion and activation of circulating monocytes. A, Innate immune-cell subsets in whole blood by mass cytometry. (B) CD14⁺CD16⁻ monocytes and (C) CD1c⁻ dendritic cells as percentages of whole blood. D, Plasma TNF and CD14⁺CD16⁻ monocyte correlation. E, CD86 and HLA-DR median intensity. *P < .05, **P < .01, ***P < .001, ****P < .0001.

CVIDc cytokine dysregulation cytokine dysregulation corresponds with T-cell activation and infiltration of tissues

Given the importance of inflammatory cytokines and expression of CD86 and HLA-DR by antigen-presenting cells for T-cell activation, we measured CD4⁺ and CD8⁺ T central memory (T_CM) (CD45RA^lo, CD27⁺), effector memory (CD45RA^lo, CD27⁻), effector memory that reexpress CD45RA (CD45RA^hi, CD27⁻), and naive (CD45RA^hi, CD27⁺) cells in whole blood by mass cytometry.⁴³ We found the proportion of CD4⁺ T_CM cells to be elevated in CVIDc compared with HCs and uncomplicated CVID (Fig 5, A and B). We also found circulating CD4⁺ effector memory cells increased in CVIDc compared with HCs. Proportion of CD4⁺ T_CM cells strongly correlated with plasma IL-12 (Fig 5, C; r = 0.742; P < .0001). There was no difference in the proportions of CD8⁺ T_CM, CD8⁺ T effector memory, CD8⁺ effector memory that reexpress CD45RA, or naive CD8⁺ T cells between CVIDc, uncomplicated CVID, and controls. Because total CD8⁺ T-cell counts were significantly lower (P = .0004; Table I) and approached significance for reduction of CD4⁺ T cells (P = .058), overall numbers of naive CD4⁺ T cells and CD8⁺ T-cell subsets were reduced in CVIDc. Corresponding with the increase in circulating CD4⁺ memory T-cell subsets, T-cell–predominant lymphocytic infiltration was observed in the gastrointestinal tract, salivary gland, and lung (Fig 5, D) as well as liver and kidney of subjects with CVIDc (see Fig E6, A, in this article’s Online Repository at www.jacionline.org).

FIG 5.

CVIDc cytokine dysregulation corresponds with T-cell activation and infiltration of tissues. A and B, CD4⁺ and CD8⁺ T-cell subsets as percentages of whole blood. C, Plasma IL-12 correlation with CD4⁺ T_CM cells. D, Representative CVIDc biopsies illustrating T-cell infiltration. E, Olink measurement of plasma chemokines. F, Luminex measurement of plasma CXCL10. G, Weight before and after immunomodulatory therapy (AZA 5 azathioprine, ABC 5 abatacept) in subjects with CVIDc. Data points are means with SD 3 weights measured before and after treatment. G, Ileum biopsies before and after abatacept in subject P22 with CVIDc. **P < .01, ***P < .001, ****P < .0001.

FIG E6.

Additional evidence of T-cell–mediated pathology in CVIDc. A, Hematoxylin and eosin stains of liver and kidney biopsies from patients with CVIDc. B, Percentage of subjects with CVIDc with interstitial lung disease, autoimmunity, liver disease, and inflammatory bowel disease. C, Jejunum biopsies before and after azathioprine in CVIDc. D, Platelet counts before and after immunomodulatory therapy (CSA 5 cyclosporine, MMF 5 mycophenolate, ABC 5 abatacept) in CVIDc. Data points are means with SD of 3 to 5 platelet count measurements before and after treatment. **P < .01, ***P < .001, ****P < .0001.

Corresponding with increased lymphocytic tissue infiltration, we found elevations of numerous T-cell chemoattractants in CVIDc plasma (Fig 5, E), including CCL19, CCL20, CCL23, CCL25, CXCL9, CXCL10, and CXCL11 via Olink. Profound elevation of plasma CXCL10, a potent T-cell chemoattractant induced by IFN-γ, was confirmed by Luminex (Fig 5, F). Most subjects with CVIDc had interstitial lung disease (65%) or autoimmunity (61%), whereas 23% had chronic liver disease and 12% had inflammatory bowel disease (Fig E6, B). Two subjects with CVIDc received T-cell–targeted therapy for T-cell–predominant inflammatory bowel disease: One subject with CVIDc received the purine synthesis inhibitor azathioprine, and another received abatacept. Abatacept restrains T cells by provision of the checkpoint inhibitor cytotoxic T-lymphocyte–associated protein 4. In both cases, weight (Fig 5, G) and inflammation (Fig 5, H, and Fig E6, C) improved after 4 months of therapy. Three subjects with CVIDc received treatment that modulates T cells for recurrent immune thrombocytopenia: 1 received cyclosporine, 1 received mycophenolate mofetil, and 1 received abatacept. Cyclosporine inhibits the calcineurin pathway, vital to lymphocyte function, and mycophenolate mofetil selectively targets lymphocytes by inhibiting the de novo purine synthesis these leukocytes uniquely rely on. These therapies significantly increased platelet counts in the 3 subjects (Fig E6, D). Together, these results demonstrated that CVIDc cytokine and chemokine dysregulation corresponds with marked activation and tissue infiltration of T cells that may be pathogenic, because complications can be ameliorated by T-cell–targeted therapy.

FIG 6.

Unsupervised machine learning reinforces link between cytokines, antibodies, and T cells in CVIDc. A, Partition around medoids (PAM) clustering of subjects with CVID and HC subjects. Subjects with CVID denoted by circles, HCs by triangles. Cluster x and y coordinates were determined by the Rtsne algorithm. Each dot is for 1 human subject, and its size indicates the number of subject’s noninfectious complications. B, Percentage of subjects with CVIDc assigned to cluster 1 and clusters 2 and 3. C, Percent contribution of parameters to the largest principle components (PC1) in which there is significant separation between subjects with CVID and HCs. Red dashed line indicates threshold of significant contribution.

Unsupervised machine learning reinforces link between cytokines, antibodies, and T cells in CVIDc

We next applied a partition around medoids algorithm incorporating all the data we collected from our subjects, including plasma cytokines and chemokines, total and LPS-specific antibodies, peripheral leukocyte immunophenotyping, and laboratory results from the electronic medical record. This unbiased analysis identified 3 clusters of subjects. All HCs segregated to cluster 1, whereas subjects with CVID were assigned to clusters 1 and 2 (1 subject with very high plasma cytokines formed its own third cluster) (Fig 6, A). About 16.7% of patients with CVIDc were assigned to cluster 1, whereas 83.3% were assigned to cluster 2 or 3 (Fig 6, B). Conversely, 88.9% of patients with uncomplicated CVID were assigned to cluster 1 and 11.1% to cluster 2. Applying principal-component MFA to our clustering algorithm showed that it was most significantly driven by measurements of CXCL10, antibodies (total immunoglobulin isotype levels and LPS-specific antibodies), isotype-switched memory B cells, and CD4⁺ and CD8⁺ T-cell subset measurement (Fig 6, C). All variables, including CVID complications, were included in this MFA to confirm correlation and demonstrate that inclusion of more variables does not bias analysis (see Fig E7 in this article’s Online Repository at www.jacionline.org). Thus, unsupervised learning validated the association of elevated plasma cytokines with greater antibody defect and skewed T-cell immunophenotype, demonstrating that convergence of these traits is a distinguishing feature of CVIDc.

FIG E7.

MFA of data from subjects with CVID and HC subjects. The data consist of plasma cytokines and chemokines, total and LPS-specific antibody responses, peripheral blood leukocyte immunophenotyping, and medical complications. A, Projection of individual subjects on the first 2 principal components (Dim1 and Dim2) showing clear separation of subject groups. B, Projection of variable groups on the first 2 principal components, with antibodies, lymphocyte subset percentages, and cytokines in blood correlating closely with the disease phenotype. C, Projection of individual quantitative variables on the first 2 principal components. Plasma sCD14, antibodies, and CXCL10 contribute largely to PC1 (Dim1), and CD4⁺ and CD8⁺ T-cell subsets providing the major contribution to PC2 (Dim2). All other quantitative variables are also shown on the figure; however, they contribute very little to either PC1 or PC2 and are therefore located very closely around the origin.

DISCUSSION

We found plasma proteomics distinguishing CVIDc with prominent elevation of IL-12p40 and TNF, cytokines that correlated with high sCD14 and deficiency of IgA and anti-LPS IgM and IgA. LPS-induced leukocyte gene expression mirrored the proteomics and also corresponded with monocyte and T-cell expansion and activation in CVIDc, with T-cell–targeted therapy having demonstrable efficacy in these patients. Unbiased computational analysis further emphasized an interrelationship of cytokine dysregulation, antibody dysfunction, and T-cell activation in CVIDc.

We found elevation of plasma proteins to distinguish CVIDc relative to uncomplicated CVID, other PAD, and HCs. This included corroborating previous reports of elevated IL-6, IL-18, and IFN-γ in CVIDc,^9,14 as well as novel findings of increased soluble IL-18 receptor, colony-stimulating factor 1, lymphotoxin alpha, LIGHT, oncostatin M, TNF, and vascular endothelial growth factor A. We also found elevation of IL-12p40, the common subunit of IL-12 and IL-23, in CVIDc, previously reported elevated in CVID but not CVIDc specifically.⁴⁴ Further efforts are needed to elucidate why these specific cytokines are elevated in CVIDc and their significance to therapy. Elevations of 4–1BB, CD40, and T-cell chemoattractants in CVIDc are consistent with a previous report,¹⁴ and, along with concurrently observed increases of CD5, CD6, and FMS-like tyrosine kinase 3 ligand, correspond with activation, expansion, and tissue infiltration of T cells in CVIDc, because these are all proteins involved in T-cell function. CXCL10 measurement was among the strongest differentiating factors by machine learning, and, because it is strongly induced by IFN-γ, provides a noted connection between the cytokine dysregulation and T-cell inflammation.⁴⁵ Continued research is needed to understand how features of cytokine and T-cell dysregulation can be used to improve diagnosis, prognosis, and treatment of CVIDc.

We found plasma sCD14, a marker of bacterial translocation as well as systemic inflammation, elevated in CVIDc relative to uncomplicated CVID, other forms of PAD, and HCs. Although others have previously reported sCD14 elevation in CVID, ours is the first to demonstrate the highest levels in CVIDc.^38,39 Concurrently, we found reduced levels of IgA as well as LPS-specific IgA and IgM in CVIDc, and levels of these antibodies inversely correlated with sCD14 as well as IL-12 and TNF. These data suggest that antibodies reduced in CVIDc may play a role in neutralizing PAMPs, such as LPS, that drive inflammation. Moreover, it provides an explanation of why CVID complications occur despite IgG replacement therapy, because specific IgA and IgM responses may be important in limiting microbial-induced inflammation. However, there are likely nuances to this conclusion because we found no correlation between sCD14 and IL-12 or TNF in other PAD. Notably, sCD14 helps induce IL-1β, IL-6, and TNF,⁴⁶ all cytokines we found elevated in CVIDc. Additional exploration of microbial influence and sCD14 levels on CVID phenotype could help further clarify their role in promoting noninfectious complications.

LPS stimulation of PBMCs demonstrated gene expression matching plasma proteomics. RNAseq analysis also revealed that altered LPS-induced gene expression was shaped by the increased proportion and activation of monocytes in CVIDc. Elevation of peripheral monocytes was revealed to be predominantly the CD14⁺CD16⁻ classical monocyte, a subset previously found elevated in patients with CVID with progressive interstitial lung disease.¹³ An increase in CD14^bright monocytes was previously reported in CVID in association with levels of sCD14 and T-cell activation, though information regarding noninfectious complications is not included.³⁸ Increased CD14⁺CD16⁺⁺ nonclassical monocytes are also reported in CVID, but levels may be altered by IgG replacement therapy and we did not find elevation of this subset in our cohort.^17,47

We found expression of CD86 and HLA-DR, proteins involved in activation of T cells, increased on CD14⁺CD16⁻ classical monocytes in CVIDc, corresponding with elevated proportion of memory CD4⁺ T cells, particularly T_CM, higher levels of T-cell attractant chemokines in the blood, and lymphocyte infiltration of tissues. Providing evidence for significance of the T-cell activation and expansion in CVIDc, we found improvement in gastrointestinal disease and autoimmune thrombocytopenia with therapies that suppress T cells in a small group of patients.

Our study is limited by small numbers of subjects, requiring us to combine patients with CVIDc with different complications as well as multiple diagnoses into our other PAD cohort. Studies specifically assessing one type of CVID complication may reveal distinct features of that condition, such as dysregulation of B-cell–activating factor and B-cell hyperplasia in patients with CVID with interstitial lung disease.^13,48 Despite these limitations, our data support a model that should be further studied in which deficiency of antibody neutralization of PAMPs, such as LPS, couples with heightened PAMP-induced inflammation to promote CVIDc.

FIG E1.

Definition of immune-cell subsets by mass cytometry. Figure displays the markers (top axis) and expression levels (darker green indicates higher expression) used to define leukocyte subsets (left axis) in mass cytometry analysis.

Key messages.

• Patients with CVIDc have a distinct cytokine and chemokine profile coinciding with deficiency of LPS-specific antibodies, together corresponding with expansion and activation of monocytes and T cells.

• The mélange of immunologic observations of CVIDc can be simplified into a model characterized by convergence of the loss of antibodies against microbial stimuli with heightened microbial-induced inflammatory response.

Abbreviations used

CVID

Common variable immnodeficiency

CVIDc

CVID with noninfectious complications

HC

Healthy control

MFA

Multiple factor analysis

PAD

Primary antibody deficiency

PAMP

Pathogen-associated molecular pattern

sCD14

Soluble CD14

T_CM

T central memory