Transcriptional signatures associated with rubella virus-specific humoral immune responses after a third dose of MMR vaccine in women of childbearing age

Abstract

In a cohort of 109 women of childbearing age, we conducted a study of rubella-specific humoral immunity before (Baseline) and after (Day 28) a third dose of MMR-II vaccine. We performed mRNA-Seq profiling of PBMCs after rubella virus in vitro stimulation to delineate genes associated with post-vaccination rubella humoral immunity and to define genes mediating the association between prior immune response status (high or low antibody) and subsequent immune response outcome.

Our study identified novel genes that mediated the association between prior immune response and neutralizing antibody titer after a third MMR vaccine dose. These genes included: CDC34/cell division cycle 34; CSNK1D/casein kinase 1 delta; APOBEC3F /apolipoprotein B mRNA editing enzyme catalytic subunit 3F; RAD18 E3 ubiquitin protein ligase; AAAS/ aladin WD repeat nucleoporin; SLC37A1/solute carrier family 37 member 1; FAS/FAS cell surface death receptor and JAK2/Janus kinase 2. The encoded proteins are involved in innate antiviral response, interferon/cytokine signaling, B cell repertoire generation, the clonal selection of B lymphocytes in germinal centers, and somatic hypermutation/antibody affinity maturation to promote optimal antigen-specific B cell immune function.

These data advance our understanding of prior immune status and genetic propensity to respond to rubella/MMR vaccination in regard to innate immunity and humoral immune outcomes after vaccination.

Graphical Abstract

Introduction

Rubella is a mild febrile exanthematous disease in children, that can lead to serious outcomes for the developing fetus during pregnancy as well as an array of birth defects designated as congenital rubella syndrome/CRS. Since 2004, rubella is no longer considered endemic in the U.S. Nevertheless, a small number of sporadic cases (on average, 10 rubella cases between 2004 and 2011) are reported to the Centers for Disease Control and Prevention each year, including cases of CRS [1]. This is mainly due to disease importation and the presence of unvaccinated individuals or vaccinated individuals with suboptimal rubella-specific immunity (usually years after last vaccination).

Protection against rubella is conferred predominantly via neutralizing antibodies directed against the two surface rubella virus glycoproteins E1 and E2 (most notably, against E1, which is suggested to be the immunodominant antigen) [2]. The current vaccine is considered highly effective in most individuals (~ 95% effectiveness) and confers sufficient immunity (including neutralizing antibodies) to protect against disease [3]. The vaccine is administered as two vaccine doses (as part of the measles-mumps-rubella [MMR] vaccine) in most developed countries.

Although rubella vaccine is effective for the majority of the population, data from the literature reveal sub-optimal long-term immunity and antibody titers below the level of protection (antibody titer <10 IU/mL) in a number of vaccinated subjects (ranging from 2 to 17%) who have received one or more doses of rubella-containing vaccine [4–16]. These findings suggest antibody waning and may indicate an increased subsequent risk for rubella among adolescents and adults (including women of childbearing age) years after vaccination.

Multiple factors—such as host genetic factors and environmental factors = play a role in the development and maintenance of immune responses to rubella vaccine and contribute to inter-individual variability in immune outcomes after vaccination [17–21]. The propensity of certain individuals to respond to repeated vaccination with a low antibody titer and/or the inability to maintain protective long-term immunity, as well as data demonstrating familial aggregation of low responders, implies the involvement of genetic factors in the immune function pathways leading to protective immunity [13, 16, 22–24]. Thus, it is likely that both low and high antibody responders to vaccination (due to inherent biology) will respond in a pre-determined manner after subsequent doses of the same vaccine.

To identify the possible factors specifically mediating the link between rubella-specific pre-immunization antibody titer (low or high) and subsequent immune response to a third dose of MMR vaccine, we designed a longitudinal vaccine study in 109 women of childbearing age from the high and the low ends of the rubella antibody response distribution (after two documented prior vaccine doses). We used whole transcriptome mRNA-Seq profiling to identify the gene expression (transcriptional) signatures mediating the relationship of prior immune response status to rubella humoral immune response outcomes (antibody titer and memory B cell ELISPOT) after a third dose of MMR vaccine.

Results

Demographic and immune response variables for the study cohort

All demographic and clinical variables of study participants have been described previously [16]. Briefly, the study sample consisted of 56 females enrolled in the low-antibody group and 53 females enrolled in the high-antibody group based on their screening rubella-specific IgG antibody titer. The majority of the study participants were White/Non-Hispanic or Latino with a median age at enrollment (third MMR dose) of 34.5 years (interquartile range/IQR 30.4, 40.3). The median age at first rubella vaccination was 15.7 months, and the median age at second rubella vaccination was 12.2 years. The median time from second rubella vaccination to enrollment in this study was 22.9 years (IQR, 18.4, 25.3). We did not observe statistically significant differences in vaccine history or demographic/clinical variables between the two groups [16]. The baseline rubella-specific neutralizing antibody titer (geometric mean with 95% confidence interval [GM, 95% CI] in NT₅₀) was 41.8 (17.4, 100.3) for the low antibody responder group and 162.2 (41.6, 632.6) for the high antibody responder group. Of note, four subjects in the low-antibody responder group had antibody titer below the level of protection (< 10 IU/mL). The peak (Day 28) neutralizing antibody titer after a third dose of MMR vaccine was 195.3 (53.3, 715.8) for the low responders and 309.9 (95.0, 1011.0) for the high responders, with no seronegative individuals The observed differences between the two groups were statistically significant (p-value=6.1E-16 for the Baseline comparison and p-value=5.6E-04 for the comparison at Day 28) [16]. The frequencies of rubella-specific memory B cells at Baseline (quantified in an ELISPOT assay in SFUs/2 × 10⁵ cells/peripheral blood mononuclear cells [PBMCs] and presented as GM, 95% CI) were 6.17 (1.31, 28.99) for the low responder group and 10.54 (2.43, 45.79) for the high responder group (p-value=0.002). The Day 28 memory B cell frequencies were 24.56 (4.34, 138.87) for the low responders and 34.29 (8.19, 143.53) for the high responders (p-value=0.12) [16]. The immune response variables in the study subjects are illustrated in Supplementary Fig. 1.

Results from statistical modeling of WGCNA gene clusters

Gene expression was measured in baseline PBMC samples after in vitro rubella virus stimulation. WGCNA identified 14 clusters of co-expressed genes upon in vitro rubella virus stimulation. Further, we used predictive modeling to identify clusters associated with immune response after a third dose of MMR vaccine in our cohort (i.e., clusters associated with the peak Day 28 neutralizing antibody [NA] titer or memory B cell ELISPOT, or with the change in neutralizing antibody response [Day 28 – Baseline]). When the WGCNA clusters were evaluated for their association with Day 28 neutralizing antibody titer, three clusters were selected, as they had non-zero β coefficients, (cluster of genes #1 [n=119 genes, GLMNET coefficient =−0.124]; cluster of genes #2 [n=185 genes; GLMNET coefficient = 0.097] and cluster of genes #3 [n=57 genes; GLMNET coefficient = 0.042]). The eigengene from cluster of 119 co-expressed genes (#1) was also associated with the change in neutralizing antibody response (Day 28 – Baseline) after a third dose of MMR vaccine (GLMNET coefficient =−0.282). Supplemental Table 1 presents the correlation of the eigengene with each of the genes in the identified clusters. The genes demonstrating the highest correlation with the eigengene may indicate drivers of the observed association. Gene enrichment analysis performed using the Reactome database [25, 26] identified highly enriched innate immune response pathways in the cluster of genes #1, including interferon α/β signaling (FDR=1.41E-14), interferon γ signaling (FDR=1.41E-14), cytokine signaling (FDR=1.41E-14), antiviral mechanisms of IFN-stimulated genes (FDR=1.8E-09), and mRNA editing (FDR=0.049), that demonstrate inter-individual differences in gene expression in high and low responders (Table 1, Fig. 1). Gene enrichment analysis also identified enriched immune response pathways in the #2 and #3 cluster of genes (Table 1), although the enrichment was not as pronounced as in the #1 cluster of genes. To determine if specific genes within these three gene clusters were associated with the Day 28 neutralizing antibody titer or the change in neutralizing antibody response (Day 28 – Baseline), with genes adjusted for the effects of each other, we used glmnet to select those genes most strongly associated with the immune outcome, conditional on the effects of other genes. Using this approach, thirteen genes were identified (Table 2). For these 13 genes, we illustrate their associations with the neutralizing antibody response in Table 2 by showing their associations (linear regression with one gene at a time) and their joint associations (linear regression on all genes). Highly significant univariate p-values were observed for most of the genes (Table 2), but the multivariate p-values were not significant with the exception of the TOR1B gene (torsin family 1 member B / gene cluster #1; p-value = 0.005; the gene encodes an ATPase and is involved in maintaining the integrity of the nuclear membrane and the endoplasmic reticulum). This is predominantly due to these genes being highly correlated, resulting from their selection to be in the co-expression clusters.

Table 1.

Significantly enriched pathways in the clusters of co-expressed genes associated with neutralizing antibody response after a third dose of MMR vaccine

Pathway name	# Entities found	# Entities total	p-value	FDR
Cluster #1 pathways enrichment
Interferon alpha/beta signaling	43	184	1.11E-16	1.41E-14
Interferon Signaling	61	392	1.11E-16	1.41E-14
Interferon gamma signaling	27	250	1.11E-16	1.41E-14
Cytokine Signaling in Immune system	63	1261	1.11E-16	1.41E-14
Antiviral mechanism by IFN-stimulated genes	14	94	2.14E-11	1.80E-09
ISG15 antiviral mechanism	10	83	1.18E-07	8.49E-06
OAS antiviral response	5	15	1.51E-06	9.51E-05
Formation of editosomes by ADAR proteins	2	2	3.01E-04	0.0169
Negative regulators of DDX58/IFIH1 signaling	4	36	0.0011	0.0492
mRNA Editing: A to I Conversion	2	4	0.0011	0.0492
C6 deamination of adenosine	2	4	0.0012	0.0492
TRAF3-dependent IRF activation pathway	3	17	0.0013	0.0492
DDX58/IFIH1-mediated induction of interferon-alpha/beta	6	96	0.0014	0.0492
Cluster #2 pathways enrichment
Neddylation	16	241	7.67E-06	0.0071
Antigen processing: Ubiquitination & Proteasome degradation	16	315	1.77E-04	0.0433
Cristae formation	5	31	2.49E-04	0.0433
Regulation of RAS by GAPs	7	71	3.02E-04	0.0433
Metabolism of RNA	28	782	3.22E-04	0.0433
Oxygen-dependent proline hydroxylation of Hypoxia-inducible	7	72	3.28E-04	0.0433
Factor Alpha
Cluster #3 pathways enrichment
Chemokine receptors bind chemokines	5	57	3.94E-05	0.0189
BMAL1:CLOCK,NPAS2 activates circadian gene expression	4	42	1.77E-04	0.0214
Transport of glycerol from adipocytes to the liver by Aquaporins	2	3	1.86E-04	0.0214
Reversible hydration of carbon dioxide	3	17	2.02E-04	0.0214
Iron uptake and transport	5	83	2.26E-04	0.0214
Transport of small molecules	16	963	4.76E-04	0.0376

Fig. 1.

Heatmaps of differential gene expression upon in vitro rubella virus stimulation in low and high antibody responders to rubella vaccination

The experimental conditions are as listed in the Methods section. Heat maps of log2 fold-change of the standardized rubella virus-stimulated gene expression (mRNA-Seq on human PBMCs) relative to the unstimulated gene expression for the genes in pathways highlighted by our analysis. Color scale ranges from red to blue as shown in the key. Study subjects are grouped across columns by low (purple, n=56) and high (pink, n=53) prior rubella-specific antibody response status.

A, B and C illustrate gene expression differences of specific Reactome innate immune response pathways (A: Interferon signaling; B: Cytokine signaling in immune system and C: Antiviral mechanism by IFN-stimulated genes) in low vs. high antibody responders to rubella vaccination. D illustrates gene expression differences of selected genes (Tables 3 and 4, Fig. 3) in low vs. high antibody responders.

Table 2.

Results of linear regressions for the cluster genes selected by glmnet to be associated with neutralizing antibody response after a third dose of MMR vaccine

Immune	Gene		Univariate	Univariate	Multivariate	Multivariate
outcome	symbol	Entrezgene description / (gene cluster #)	estimate1	p-value	estimate2	p-value
	TOR1B	torsin family 1 member B (cluster #1)	−0.35	0.000184	−0.29	0.005
	RNF213	ring finger protein 213 (cluster #1)	−0.26	0.005884	−0.13	0.198
Day 28 NA	SUB1	SUB1 regulator of transcription (cluster #2)	0.30	0.001237	0.13	0.310
	ATAD2B	ATPase family AAA domain containing 2B (cluster #2)	−0.31	0.000824	−0.17	0.155
	BLCAP	BLCAP apoptosis inducing factor (cluster #2)	0.28	0.004808	0.08	0.554
	GNB1L	G protein subunit beta 1 like (cluster #2)	0.27	0.003857	0.05	0.687
	HDGF	heparin binding growth factor (cluster #3)	0.24	0.009652	0.19	0.116
	SRXN1	sulfiredoxin 1(cluster 3)	0.21	0.030266	0.08	0.482
	FAS	Fas cell surface death receptor (cluster #1)	−0.50	0.000004	−0.17	0.315
	TENT5A	terminal nucleotidyltransferase 5A (cluster #1)	−0.43	0.000182	−0.19	0.128
NA change	ZNFX1	zinc finger NFX1-type containing 1 (cluster #1)	−0.48	0.000011	−0.09	0.589
(Day 28–Baseline)	SLC37A1	solute carrier family 37 member 1 (cluster #1)	−0.50	0.000002	−0.21	0.176
	CD2AP	CD2 associated protein (cluster #1)	−0.48	0.000016	−0.06	0.727

¹Coefficients from the univariate linear regressions between neutralizing antibody response and the log₂ fold-change of the rubella virus-stimulated gene expression relative to the unstimulated gene expression (standardized).

²Coefficients from multivariate linear regression between neutralizing antibody response and the log₂ fold-change of the rubella virus-stimulated gene expression relative to the unstimulated gene expression (standardized), incorporating all genes listed in the table in the model.

Results from univariate mediation analyses

We used mediation analysis of 11,610 genes (analyzing one mediator at a time) to evaluate how transcriptomic activity mediates the association of rubella-specific prior immune response status (high antibody or low antibody) on rubella-specific humoral immune response outcomes to a third dose of MMR vaccine. This analysis identified a group of genes as potential mediators for the response of Day 28 neutralizing antibody titer (Table 3) and another group of genes for the change in neutralizing antibody titer (Day 28 – Baseline) (Table 4). The most significant genes/mediators associated with Day 28 neutralizing antibody titer were the cell division cycle 34/CDC34 (p-value = 0.028) and the casein kinase 1 delta/CSNK1D (p-value = 0.033) and other mediators with suggestive association included the solute carrier family 37 member 1/SLC37A1 (p-value = 0.049), the FAS cell surface death receptor/FAS (p-value = 0.053), and the apolipoprotein B mRNA editing enzyme catalytic subunit 3F/APOBEC3F (p-value = 0.058). The most significant genes/mediators associated with the change in neutralizing antibody response (Day 28 – Baseline) were the aladin WD repeat nucleoporin/AAAS (p-value = 0.014) and the solute carrier family 37 member 1/SLC37A1 (p-value = 0.025), and other mediators with significant or suggestive association included the FAS cell surface death receptor/FAS (p-value = 0.038) and the Janus kinase 2/JAK2 (p-value = 0.057) (Tables 3 and 4). Overall, we observed modest p-values and indirect mediation effects in these analyses, while correction for multiple testing in this sample size did not reveal statistically significant mediation effects. The univariate mediation analysis also identified a group of genes as potential mediators for the Day 28 memory B cell ELISPOT response (Supplemental Table 2).

Table 3.

Top 20 gene mediators associated with Day 28 rubella-specific neutralizing antibody response to a third dose of MMR vaccine (univariate mediation analysis approach)

Gene Symbol	Entrezgene description	Total effect	Direct effect	Indirect effect	p-value
CDC34	cell division cycle 34	0.662	0.829	−0.167	0.028
CSNK1D	casein kinase 1 delta	0.662	0.821	−0.16	0.033
DNAJC13	DnaJ heat shock protein family (Hsp40) member C13	0.662	0.806	−0.144	0.043
MIR4426	microRNA 4426	0.662	0.804	−0.143	0.044
TARDBP	TAR DNA binding protein	0.662	0.804	−0.142	0.044
DUS1L	dihydrouridine synthase 1 like	0.662	0.804	−0.142	0.044
RPS23P8	ribosomal protein S23 pseudogene 8	0.662	0.8	−0.138	0.047
SLC37A1	solute carrier family 37 member 1	0.662	0.798	−0.136	0.049
PDXP	pyridoxal phosphatase	0.662	0.797	−0.135	0.050
NGFRAP1	brain expressed X-linked 3	0.662	0.799	−0.137	0.052
FAS	Fas cell surface death receptor	0.662	0.796	−0.134	0.053
TECR	trans-2,3-enoyl-CoA reductase	0.662	0.808	−0.146	0.053
BRAT1	BRCA1 associated ATM activator 1	0.662	0.792	−0.13	0.054
RP11–761N21.2	Pseudogene	0.662	0.791	−0.129	0.055
UMPS	uridine monophosphate synthetase	0.662	0.79	−0.128	0.056
KLHDC3	kelch domain containing 3	0.662	0.79	−0.128	0.056
CEP162	centrosomal protein 162	0.662	0.79	−0.128	0.056
AGO1	argonaute RISC component 1	0.662	0.793	−0.132	0.056
APOBEC3F	apolipoprotein B mRNA editing enzyme catalytic subunit 3F	0.662	0.789	−0.127	0.058
VPS54	VPS54 subunit of GARP complex	0.662	0.796	−0.134	0.061

The definitions of total, direct and indirect effects are provided in detail in the statistical method section.

Table 4.

Top 20 gene mediators associated with the change (Day 28 – Baseline) in rubella-specific neutralizing antibody response following a third dose of MMR vaccine (univariate mediation analysis approach)

Gene Symbol	Entrezgene description	Total effect	Direct effect	Indirect effect	p-value
AAAS	aladin WD repeat nucleoporin	−1.357	−1.132	−0.225	0.014
SLC37A1	solute carrier family 37 member 1	−1.357	−1.170	−0.187	0.025
TARDBP	TAR DNA binding protein	−1.357	−1.184	−0.173	0.029
FAS	Fas cell surface death receptor	−1.357	−1.189	−0.168	0.038
PARK7	Parkinsonism associated deglycase	−1.357	−1.213	−0.145	0.044
RPS23P8	ribosomal protein S23 pseudogene 8	−1.357	−1.216	−0.141	0.046
RASGEF1B	RasGEF domain family member 1B	−1.357	−1.196	−0.161	0.048
PMPCA	peptidase, mitochondrial processing alpha subunit	−1.357	−1.222	−0.135	0.051
CPSF6	cleavage and polyadenylation specific factor 6	−1.357	−1.223	−0.134	0.052
APOL6	apolipoprotein L6	−1.357	−1.223	−0.134	0.053
JAK2	Janus kinase 2	−1.357	−1.214	−0.143	0.057
PPIA	peptidylprolyl isomerase A	−1.357	−1.217	−0.140	0.057
RP11-632C17__A.1	Pseudogene	−1.357	−1.229	−0.128	0.057
NFS1	NFS1 cysteine desulfurase	−1.357	−1.230	−0.127	0.058
VCPIP1	valosin containing protein interacting protein 1	−1.357	−1.213	−0.144	0.059
ZNF469	zinc finger protein 469	−1.357	−1.225	−0.132	0.060
TMEM140	transmembrane protein 140	−1.357	−1.221	−0.136	0.061
UMPS	uridine monophosphate synthetase	−1.357	−1.237	−0.120	0.066
GATC	glutamyl-tRNA amidotransferase subunit C	−1.357	−1.237	−0.120	0.067
PPP1R14B	protein phosphatase 1 regulatory inhibitor subunit 14B	−1.357	−1.228	−0.129	0.071

The definitions of total, direct and indirect effects are provided in detail in the statistical method section.

Results from joint mediation analysis

Using joint mediation analysis, one gene was selected as a possible mediator associated with rubella-specific Day 28 neutralizing antibody after a third dose of MMR vaccine. This possible mediator, RAD18 (RAD18 E3 ubiquitin protein ligase), had a marginal p-value of 0.16 when tested alone; however, RAD18 was highly correlated with the genes TECR (trans-2,3-enoyl-CoA reductase) and CSNK1D (casein kinase 1 delta) from Table 3 (Fig. 2). It is possible that the correlation patterns of these genes make it difficult to determine which one of the genes is the strongest mediator. We also used our joint mediation analysis to evaluate the possible mediators associated with the change in neutralizing antibody titer (Day 28 – Baseline) after a third dose of MMR vaccine. One gene was selected based on this approach, the aladin WD repeat nucleoporin/AAAS (p-value = 0.014), which is also the most significant gene in Table 4 (univariate mediation analyses). The differential gene expression of the group of genes of interest (from univariate and joint mediation analyses) in high vs. low antibody responders is depicted in Fig. 1 and Fig. 3. The correlation patterns are depicted in Fig. 2.

Fig. 2.

Correlation plots for genes mediating the association of prior immune response status and rubella-specific neutralizing antibody titer following a third dose of MMR vaccine

The experimental conditions are as listed in the Methods section. Graphical display of the correlation patterns (positive and negative correlations) for log2 fold-change of the rubella virus-stimulated gene expression relative to the unstimulated gene expression (mRNA-Seq) for the top gene mediators associated with rubella-specific neutralizing antibody response following a third dose of MMR vaccine A. top genes (Table 3) mediating the association of prior immune response status and Day 28 rubella-specific neutralizing antibody titer, and B. top genes (Table 4) mediating the association of prior immune response status and the change (Day 28 – Baseline) in rubella-specific neutralizing antibody titer. The colors and size of the circles illustrate the pair-wise correlations, with blue indicating positive correlations and orange/red negative correlations. There are two clusters of positively correlated genes (the upper left and bottom right blue clusters) for the genes in plot A and the genes in plot B.

Fig. 3.

Gene expression differences for selected genes in low vs. high antibody responders to rubella vaccination

The experimental conditions are as listed in the Methods section. Box and whisker plots of standardized log2 gene expression values for rubella virus-stimulated samples minus the standardized log2 gene expression values for unstimulated samples (shown on the y axis); separated by study subjects with low (n=56) and high (n=53) prior antibody response status. The p-value is calculated using the Wilcoxon rank-sum test. The gene expression is illustrated for selected genes with plausible link to immunity (see Discussion and Tables 3 and 4).

Discussion

We and others have previously shown that inherent genetic and immune factors play an important role in the generation/maintenance of rubella-vaccine-induced humoral immunity [13, 16–24], and in explaining inter-individual differences in immune response. We have also demonstrated that high responders to rubella vaccination respond with a statistically significant higher peak neutralizing antibody titer ~28 days following a third dose of MMR vaccine, compared to low responders, who respond with a lower antibody titer, but have a higher relative boost/change (Day 28 – Baseline) in antibody titer [16]. Understanding the yet unknown factors controlling rubella immunity is essential to predict the rubella-specific response to MMR vaccination, the durability of protection absent wild virus boosting, and may provide valuable insights into immunity to rubella virus and other viral pathogens.

Using a comprehensive assessment of transcriptomic information and a mediation analysis approach [27], we were able to identify genes and plausible biologic processes/mechanisms underlying the differential response to rubella vaccination.

Our WGCNA analysis identified three gene clusters of interest, in particular, a cluster of 119 genes (cluster #1)associated with both the Day 28 neutralizing antibody titer and the change/boost in rubella-specific neutralizing antibody titer after a third dose of MMR vaccine. This cluster is highly enriched in genes of the interferon and cytokine signaling pathways, antiviral mechanism/activity by IFN-stimulated genes, TRAF3-dependent IRF activation pathway, mRNA editing pathway and other relevant innate immune pathways (Table 1, Fig. 1). Thus, it is likely that innate immune function (antiviral response) and the innate immune pathways are key factors in the development of rubella-specific neutralizing antibody response following vaccination.

The top plausible mediator linking prior response status (high or low) to the Day 28 neutralizing antibody titer following a third dose of MMR vaccine was CDC34 (see Table 3, Fig. 3), which encodes a protein that is intimately interconnected with genes and functions regulating the B cell receptor repertoire, antibody diversity, and B lymphocyte development and differentiation. The ubiquitin conjugating (E2) activity of CDC34 is essential (via functional interaction) for the recombinase activating gene 1/RAG1 auto-ubiquitylation, which in turn is necessary for an optimal V(D)J recombination process [28–32]. In addition, CDC34 is involved in the ubiquitin/proteasome-mediated degradation of the inhibitor NFKB inhibitor alpha (encoded by NFKBIA), which is needed to activate the transcription factor NF-kappaB, a process central in many immune response pathways, including B cell maturation and antibody production [33, 34].

Interestingly, a different plausible gene mediator marginally associated with the Day 28 neutralizing antibody response in our study (i.e., APOBEC3F; Table 3, Fig. 3) is also involved in mechanisms influencing the magnitude and quality of the humoral immune response to viruses. Studies in mouse models of Friend retrovirus (FV) infection have repeatedly demonstrated that APOBEC3 might directly influence neutralizing antibody response by mechanisms including antibody class switching and affinity maturation, as well as by fine-tuning the balance between marginal zone B-cell response (fast, but characterized by less antibody affinity maturation) and germinal center B-cell response (supporting antibody affinity maturation and the generation of highly specific and functional antibodies) [35–37]. Using immunoglobulin heavy chain V gene sequencing of antigen-specific hybridomas and germinal center B cells from FV-infected APOBEC3-competent mice compared to APOBEC3-malfunctioning mice, Halemano et al. revealed the role of APOBEC3 in somatic hypermutation (similar to AID) for the development of high-affinity antibody response against retroviruses [38]. Thus, it is possible that APOBEC3 factors are directly involved in the biological processes that fine-tune the antibody response after infection and vaccination. APOBEC3 factors have been extensively studies as intrinsic/innate antiviral factors that restrict the replication of retroviruses (e.g., HIV-1, HTLV-1, human foamy viruses), DNA viruses (e.g., adeno-associated AAV, hepatitis B virus/HBV, human papillomavirus/HPV, herpes simplex virus 1/HSV-1 and Epstein-Barr virus/EBV) and typical RNA viruses (e.g., measles, mumps, respiratory syncytial viruses and coronaviruses ) through various enzyme-dependent/editing and enzyme independent mechanisms [39–42]. By studying the sequence evolution of rubella vaccine-descending viruses in the cutaneous granulomas of children with immunodeficiency, Perelygina et al. recently reported the key role of APOBEC3 editing activity for the genetic diversity as well as replicative and persistence characteristics of rubella viruses in vivo [43]. Thus, it is likely that APOBEC3 members, and APOBEC3F identified in our study, contribute to innate immune restriction and/or modulation of the RA27/3 rubella virus replication and characteristics in vivo, and, in this way, modulate the cumulative antigen load, virus evolution in the host, and the elicited adaptive immunity.

Another possible mediator of the association between prior immune response status and Day 28 rubella neutralizing antibody response is RAD18 (joint mediation analysis, Fig. 3), which is a ubiquitin E3 protein ligase that is involved in ubiquitin conjugation as well as DNA replication and repair mechanisms. Its connection to immune function is largely unknown, although Bachl et al., [44] postulated its likely contribution to the complex processes of somatic hypermutation and Ig diversification via involvement in the ubiquitination of proliferating cell nuclear antigen, which in turn signals switch to error-prone DNA lesion bypass/mutagenesis [44]. RAD18 is also highly correlated with the genes CSNK1D, and TECR (Fig. 2A) and the correlation patterns make it difficult to untangle the exact mediator. Of the latter three genes, CSNK1D, a member of the casein kinase I gene family of serine/threonine-specific kinases (comprised of 6 isoforms) has plausible immune-related functions (and is one of our top plausible mediators, Table 3). Casein kinase I isoforms have been implicated in the regulation of antiviral response by controlling the activation/phosphorylation of TRAF3, a major signaling adaptor involved in the induction of type I interferons by viruses [45]. In addition to its involvement in a wide range of cellular functions (microtubule formation, apoptosis, DNA replication and repair, p53 pathway effects, circadian rhythm), CSNK1D was demonstrated to regulate cell-cell contacts, remodeling of cytoskeleton and centrosome positioning during T cell activation/immune synapse formation, and the process of connexin gap junction assembly, which is essential for communication between immune cells [46–49]. It has been demonstrated that inhibition of gap-junction communication between immune cells influences key B cell functions such as antibody and cytokine production [50].

The top gene linking the prior immune status to neutralizing antibody response change (Day 28 – Baseline) in antibody titer following a third dose of MMR vaccine is the AAAS gene, encoding the aladin WD repeat nucleoporin (top gene in Table 4; significant gene in joint mediation analysis). Its relation to immune response is largely unknown. Interestingly, some viruses (poliovirus, rhinovirus, vesicular stomatitis virus, influenza virus) have developed strategies to block or modulate nuclear transport (host mRNA and other) through interaction with nucleoporins, degradation of specific nucleoproteins or interaction with mRNA export factors [51]. This may lead to evasion of host antiviral response and enhanced viral replication, a mechanism by which nucleoporins may potentially influence immunity to viruses [51, 52].

Another factor associated with the neutralizing antibody response is the SLC37A1 gene, which encodes the solute carrier family 37 member 1 protein (Tables 2, 3, 4, Fig. 3).Although its biological link to viral infection/vaccination and immunity remains to be illuminated, it has been suggested that the encoded protein may have a key role in the metabolism of some immune (e.g., neutrophils) and other cells [53]. The link of this gene to adaptive humoral immunity is elusive and warrants further investigation.

Other important factors from our study, which were associated with rubella-specific antibody response following vaccination, are also likely to impact B cell immune function. For example, FAS (Tables 2, 3, 4, Fig. 3) encodes the Fas cell surface death receptor that is involved in the selection of high-affinity B cells in the germinal centers, the establishment of memory B cells, and the homeostasis of both B and T cells [54, 55].

JAK2 (Table 4, Fig. 3) encodes a Janus tyrosine kinase that is critical for the signaling through receptors that lack intrinsic tyrosine kinase activity, including receptors for erythropoietin, thrombopoietin, prolactin, GM-CSF and a variety of cytokine receptors, including IFNγ, IL-12, IL-23, IL-3, IL-5 and IL-6 [56]. Due to the pleiotropic effects of JAK2-signaling cascades, it is difficult (without in-depth functional studies) to indicate the exact mechanism by which JAK2 impacts B cell function and antibody titers. It is possible that specific cytokine signaling (IL-5, IL-6, IFNγ) promotes B cell proliferation and differentiation, germinal center responses, antibody class switching, and/or the generation of Tfh cells [57–59]. The JAK2/STAT3 signaling pathway has also been demonstrated to regulate the activation of dendritic cells, including the upregulation of MHC class II molecules, costimulatory molecules, and the ability to stimulate T cells [60], all of which can impact humoral immunity.

The limitations of our study include the possibility of false-positive associations due to testing many genes with a relatively small sample size, as well as the measurement of gene expression in PBMCs. The observed median gene expression changes, and the observed difference between the low and the high antibody responder groups looks minimal. However, this is expected when measuring gene expression changes in mixed cell culture (i.e., PBMCs) rather than in specific relevant cell subsets. This may impact the effect of a specific gene if it is only expressed in one cell subset. In mixed cell culture, the actual gene expression and fold change measured for a specific gene are often diluted by gene expression in other cell types in the sample. In addition, the subtraction of the baseline gene expression (unstimulated condition) may also impact the median gene expression change. The statistical tests used in our study are derived using both the change and the variance of the gene around that change to ascertain significance. Although we used stringent QA/QC procedures to minimize variability in our assays, and we used different analytical approaches, including joint mediation analysis and Weighted Gene Co-expression analysis (WGCNA), our exploratory strategy to discover new genes related to immune response has the risk of multiple testing and may lead to false discoveries. Hence, the functional validation of our findings in follow-up studies is warranted to validate our findings. The clinical significance of the identified genes/factors, as well as their functional/mechanistic effects on protective immunity, also merits further investigation. Further validation and functional studies may also provide insights into how to manipulate specific immune functions (e.g., adjuvants, cytokines, kinase inhibitors) to achieve optimal neutralizing antibody response and immune protection, and to inform new vaccine candidates and strategies.

Our study identified previously unknown mediators of rubella virus-specific humoral immune responses in women of childbearing age. This enhances our understanding of the genetic drivers of immune response to rubella vaccination, and provides potential insights into the specific mechanisms underlying inter-individual variations in vaccine response and failure.

Methods

The methods described below are similar or identical to those published for our previous studies [13, 15, 16, 18, 19, 61–64].

Study cohort

The study sample has been described previously [16]. The sample consisted of 109 healthy female subjects (20–45 years old) from Olmsted County, MN, and surrounding areas enrolled at Mayo Clinic, Rochester, MN, with two documented doses of MMR vaccine. Subjects were selected for this study if they were in the highest 30% (high antibody responder group) or lowest 30% (low antibody responder group) of the rubella-specific IgG antibody-titer distribution based on enzyme-linked immunosorbent assay (ELISA) screening of 1,117 available serum samples from the local community (obtained from the Mayo Clinic Biobank). Samples of study participants (serum and PBMCs) were collected prior to vaccination (Baseline) and at two time points after the receipt of a third MMR vaccine dose (Day 8 and Day 28). Vaccination history was recorded. All study participants provided written informed consent, all research was performed in accordance with relevant guidelines, and all study procedures, including the Mayo Clinic Biobank protocol, were approved by the Mayo Clinic Institutional Review Board.

Rubella neutralizing antibody assay

A soluble immuno-colorimetric-based neutralization method (sICNA), optimized to a high-throughput micro-format, was used to assess rubella neutralizing antibody titers [13, 61–63, 65]. The sICNA development step was performed after 72 hours (37^oC, 5% CO₂) incubation as previously described [13, 61–63], except that the procedure was automated. The NT₅₀ was determined using Karber’s method. The neutralization titer represented the highest dilution at which the input virus signal was reduced by at least 50% within the dilution series (NT₅₀) [62]. For quality control purposes and to confirm assay reproducibility, 129 serum samples were retested. The reported intra-class correlation coefficient (ICC) based on log-transformed estimates from repeated NT₅₀ measurements for this assay was 0.89, which demonstrates a high degree of reproducibility [61].

Memory B cell ELISPOT assay [16, 64]

Rubella virus-specific memory-like IgG B cells were quantified pre- (Baseline) and post-vaccination (Day 28) in subjects’ PBMCs using the Mabtech ELIspot^PLUS kit for human IgG (Mabtech Inc.; Cincinnati, OH) according to the manufacturer’s specifications and as previously described [64, 66, 67]. In vitro non-specific pre-stimulation of thawed PBMCs/B cells was performed for three days in the presence of human recombinant IL-2 and the Toll-like receptor (TLR) agonist R848 [16, 66]. ELISPOT plates were coated with rubella virus antigen (HPV77 RV strain) from Meridian Life Science Inc. (Memphis, TN). Antigen-specific memory B cell frequencies were measured and presented as spot-forming units (SFUs) per 2×10⁵ cells as subjects’ medians (median of rubella virus-specific SFUs response, measured in quadruplicate with subtracted subject-specific no-antigen background measure). The reproducibility of this assay (assessed as intra-class correlation coefficients between replicate measurements) was high (average 0.88) [16, 64].

Next Generation Sequencing (NGS):

For assessment of gene expression in response to rubella virus stimulation, PBMC samples from the baseline blood draw were thawed and stimulated with live rubella virus (the Therien-W strain) at multiplicity of infection/MOI of 5, or were left unstimulated. Cells were incubated for 48 hours at 37^oC, 5% CO₂ and were harvested in RNAProtect (Qiagen, Valencia, CA), and frozen at −20°C until RNA extraction. mRNA was extracted using the Qiagen RNeasy Plus Mini Kit (Qiagen, Valencia, CA). All samples passed RNA and cDNA QC check on an Agilent 2100 Bioanalyzer (Agilent; Palo Alto, CA) before and after library preparation. Library generation and sequencing was performed at the Mayo Clinic Advanced Genomics Technology Center using 101-base paired-end mode on Ilumina’s HiSeq 4000 at 8 samples per lane as described previously [64, 68–70]. The MAP-RSEQ version 3.0 [71] pipeline was used for processing the raw RNA sequence paired-end reads. The pipeline aligns reads using STAR [72] to the hg38 human reference genome and gene expression counts are obtained using featureCounts [73] utilizing the gene definition files from Ensembl v78.

Statistical Analysis

Three response variables were analyzed: 1) change in neutralizing antibody titer/response after a third dose of MMR vaccine, whereby change in response was log base-2 neutralizing antibody measured at Day 28 (after a third dose of MMR vaccine) minus the log base-2 neutralizing antibody at Baseline; 2) Day 28 neutralizing antibody response, log base-2 neutralizing antibody; 3) Day 28 memory B cell ELISPOT response, determined by fitting a Poisson regression model to adjust for batch effects and using the residuals from that model as the adjusted response.

Gene expression data were normalized using conditional quantile normalization [74]. Only protein coding genes and processed pseudogenes were included in analyses. Genes with low counts were excluded from analyses if their coefficient of variation was less than the 25th percentile or if the median of the gene expression response (gene expression in antigen-stimulated cells minus gene expression in unstimulated cells) was less than 16. The gene expression responses for the remaining 11,610 genes were centered on their means and scaled by their standard deviations.

To select genes that had the most influence on immune response and account for the large number of genes, we used the R package glmnet to fit penalized linear models, with 90% of the penalty placed on the lasso penalty and 10% of the penalty placed on the ridge penalty. The penalty parameter was selected as the value that minimized the mean squared prediction error from 10-fold cross-validation [75]. These models were fit using all 11,610 genes as input to select genes, as well as to select eigengenes from weighted gene co-expression network analysis (WGCNA). Genes or clusters with coefficients (β-values) set to zero were considered to be non-informative and were not included in further analyses. The WGCNA was used to cluster genes into modules using unsigned networks and bi-weight correlations; the remaining parameters were set to package default values. The module eigengene, the first principal component of each cluster, was used as a summary measure of each cluster [76].

To identify genes that mediated the association between prior antibody response status (high or low) and each of the response variables, we used two types of mediation analyses. The first analysis was the traditional Sobel marginal test for mediation that analyzes one potential mediator at a time (i.e., for each of the 11,610 genes) [77]. This entailed using three linear regression models: 1) regress the response variable on prior antibody response status (the regression coefficient for prior antibody response status is a measure of its total effect); 2) regress the response variable on both prior antibody response status and gene expression (the regression coefficient for prior antibody response status is a measure of its direct effect); 3) regress gene expression on prior antibody response status. The mediated indirect effect was calculated by multiplying the regression coefficient for prior gene expression in model #2 by the regression coefficient for prior antibody response status in model #3. Although analyzing one mediator at a time simplifies the analyses, the correlation among the mediators can cause the marginal analysis to miss a mediator that has effects in a joint analysis [78]. For this reason, we used a novel second approach based on penalized structural equation models (SEMs), which jointly estimate the mediators with direct and indirect effects, while accounting for correlation [79, 80]. SEMs become unstable when there are a large number of parameters. To overcome these limitations, we used a newly developed penalized SEM that is suited for multivariate mediation analyses; note that we have previously described this SEM [27]. With utilizing our methods, it is not possible to fit 11,610 genes as potential mediators for a sample size of 109 subjects.To address this, we used sure-independence screening to reduce the number of potential mediators [81]. This method is based on ranking marginal correlations and selecting the highest ranked ones such that their number is less than the sample size. Mediation depends on the two correlations: and , where x is prior antibody response, y is the response variable, and is a potential mediator (e.g., change in gene expression). Because of this, we ranked the absolute values of their products, . From these, we selected the highest 40 ranked ones to establish which potential mediators to add in our penalized mediation models. This resulted in 80 parameters for the potential mediators. The penalized SEM uses a penalty parameter to penalize the mediation effects, shrinking small parameter estimates to 0. The penalty parameter that minimized the Bayesian Information Criterion (BIC) was used to choose the best fitting model. To avoid excess shrinking of the parameters, we refit the selected model using the lavaan R package that fits standard SEM without penalties [82].

Supplementary Material

Supplementary Material

Supplementary Fig. 1 Rubella-specific humoral immune response in women of childbearing age

Immune responses after a third dose of MMR vaccine. (A) rubella-specific neutralization antibody titers and (B) rubella-specific memory B-cell ELISPOT over time are shown using boxplots for all subjects (“Overall”, grey box) , the high responders (“High”; black box) and the low responders (“Low”, white box).The neutralizing antibody titers and the number of rubella-specific memory B cells are plotted in log2 scale. Each box was plotted using the 25% to 75% interquartile range and the median was represented by the bold line in the box. The “whiskers” extend up to 1.5 times the interquartile range above or below the 75^th or 25^th percentiles respectively.(From Haralambieva IH, Ovsyannikova IG, Kennedy RB, Goergen KM, Grill DE, Chen MH, Hao L, Icenogle J, Poland GA. Rubella virus-specific humoral immune responses and their interrelationships before and after a third dose of measles-mumps-rubella vaccine in women of childbearing age. Vaccine. 2020 Jan 29;38(5):1249–1257. doi: 10.1016/j.vaccine.2019.11.004.)

Supplemental Table 1. Correlations of co-expressed genes with the eigengene of the WGCNA clusters associated with neutralizing antibody response after a third dose of MMR vaccine

Supplemental Table 2. Top 20 gene mediators associated with the Day 28 rubella-specific memory B cell ELISPOT response after a third dose of MMR vaccine (univariate mediation analysis approach)