Circulating immune cell phenotypes are associated with age, sex, CMV, and smoking status in the Framingham Heart Study offspring participants

Summary of immune cell phenotypes profiled

Immune cell phenotypes profiled from peripheral blood mononuclear cells (PBMCs) samples are shown in Figure 1 and listed in Supplementary Table 1. We selected 24 immune cell phenotypes including the CD4+ and CD8+ naïve cells, effector cells, memory cells, senescent cells, and Granzyme B producing CD4 and CD8 cells, B cells, NK cells, classical monocytes, and the 3 ARIP measures to illustrate our main findings, with complete results for all 116 immunophenotypes for all analyses found in the online supplement. Cellular markers used to identify the 24 circulating immune cell phenotypes of primary focus are listed in Table 1. The number of participants with data and summary statistics (mean, standard deviation, min, max, and coefficients of variation (CV)) for 116 immune cell phenotypes and 3 ARIP measures are presented in Supplementary Table 2. Pairwise correlation plots are presented in Supplementary Figure 1. We note that we use “CD4” to indicate CD4+, and “CD8” to indicate CD8+ throughout.

Figure 1. Peripheral blood mononuclear cells (PBMCs) profiled in this study. (A) CD4/CD8 naïve/memory panel; (B) CD4/CD8 regulatory panel; (C) T cell, B cell and NK cell panel; (D) CD4/CD8 stimulation panel; (E) monocyte subset panel.

Participant demographics

Sample inclusion and exclusion criteria are shown in Figure 2. Participant characteristics are summarized in Table 2. Among the 996 participants included in the study, 48% were female. The average age of the participants was 62 years with a range from 40 to 88 years. CMV values were categorized into four groups by level of CMV IgG measured in U/ml. A total of 465 participants (46.7%) were CMV seropositive, 101 of whom had values above the limit of detection (CMV > 800 U/ml). More participants had CMV levels in the equivocal range (11< CMV ≤ 15 U/ml, n=399) than in the negative range (CM ≤ 11 U/ml, n=75). CMV positive (15 < CMV ≤ 800 U/ml) participants were older on average than participants who were CMV negative or equivocal (Table 3).

Figure 2. Study sample of the immune cell phenotypes in the FHS.

Immune cell phenotypes across age groups by sex and CMV status

Figure 3 (CD4), 4 (CD8) and 5 (ARIP) show the unadjusted percentages of immune cell types CD4 and CD8 and subtypes naïve (Tn), effector (Teff), effector memory (Tem) and central memory (Tcm) across age groups, stratified by sex and CMV status (as two categories by grouping the negative and equivocal, and the positive and above LOD). The percentage of CD4 T helper cells was higher in CMV negative and equivocal participants compared to CMV positive participants across all age groups and for both males and females (Figure 3), while the percentage of CD8 cytotoxic T cells in participants in the negative and equivocal CMV group was lower than in CMV positive group (Figure 4). Females had higher percentages of CD4 T helper cells and lower percentages of CD8 cytotoxic cells than males for most age and CMV serostatus groups.

Figure 3. Unadjusted mean percentages of five CD4+ T cell types by age, sex, and two-category CMV status. Note: Teff = T effector cells; Tem = T effector-memory cells; Tcm = T central memory cells.

Figure 3 shows that CD4 Tn cell percentages were lower in older age groups compared to younger age groups, while CD4 Tcm percentages were higher in older age groups compared to younger age groups. Both age trends were similar for CMV positive and negative male and female subgroups. Female participants in the CMV positive groups tended to have higher CD4 Tn percentages than CMV positive male participants in all age groups. Percentages of CD4 Teff cells and CD4 Tem cells were higher in CMV positive participants than negative/equivocal participants across all age groups, and males had higher proportions of these cells than females within CMV category for most age groups.

CD8 Tn cell percentages were lower in older age groups compared to younger, while CD8 Teff and Tem cell percentages were higher in older age groups compared to younger, and CD8 Tcm cells did not show a consistent age trend within sex and CMV strata (Figure 4). Percentages of CD8 Teff were higher in CMV positive participants compared to CMV negative/equivocal and were higher in males than females within each CMV group across all but the oldest age group. CD8 Tcm percentages were lower in CMV positive compared to negative/equivocal; males in the CMV positive group had higher percentages than females in all age groups.

Figure 4. Unadjusted mean percentages of five CD8+ T cell types by age, sex and two-category CMV status. Note: Teff = T effector cells; Tem = T effector-memory cells; Tcm = T central memory cells.

On average, we observed a higher CD4/CD8 ratio and higher CD8 Tn/Tm ratio in the CMV negative/equivocal group compared to the CMV positive groups within each sex stratum, and higher levels in females compared to males within each CMV group. The CD4/CD8 ratio was higher in the older age groups for the CMV negative/equivocal participant group but was similar across age groups for the CMV positive groups (Figure 5). Both CD4 and CD8 Tn/Tm ratios were lower in older age groups compared to younger age groups regardless of sex or CMV serostatus, but the CD4 Tn/Tm ratio fluctuated more by age group than the CD8 Tn/Tm ratio. The unadjusted percentages of other immune cell types listed in Table 1 across age group, sex, and CMV exposure groups are illustrated in Supplementary Figure 2.

Figure 5. Unadjusted mean values of ARIP measures stratified by age, sex and two-category CMV status. Note: Tn = naïve T cells; Tm = memory T cells.

Association of immune cell phenotypes with CMV serostatus groups adjusting for age and sex

Table 4 shows that among the 27 immune cell phenotypes listed in Table 1, 20 had significantly different means across the four CMV serostatus groups after adjusting for age and sex at false discovery rate (FDR)<0.05. For most cell types, effects were not linear with increasing CMV value group, but the non-linear patterns were not the same for all cell types. For most of the significant cell types, the Equivocal CMV and Negative CMV group adjusted means were more like each other than the Above LOD group and the Positive CMV group adjusted means. We observed that CD3+ T cell levels (“T cells” in Table 4) were similar for CMV negative and equivocal participant groups but were higher in CMV positive and higher yet in the above LOD CMV positive participant groups. CD4 T helper cell levels were slightly higher for participants in the CMV equivocal group compared to those participants who were CMV negative, while the CMV positive and above LOD groups had substantially lower levels than the negative group. CD8 cytotoxic T cell levels were lower in CMV equivocal compared to CMV negative groups, and higher in the CMV positive and LOD groups compared to the negative group.

Among the CD4 cell subtypes, the CMV positive and above LOD groups had lower age and sex adjusted CD4 Tn levels than the CMV negative or equivocal groups, while adjusted Teff, Tem, TEMRA, T helper type 1 (Th 1) and Granzyme B producing cell subtype levels were higher in the CMV positive and above LOD groups than in the negative and equivocal groups. We observed similar patterns among the corresponding CD8 cell subtypes, the one exception being that CD8 Tcm cells also differed by CMV status.

B cell levels were similar in negative and equivocal CMV groups, and lower in the CMV positive groups. Finally, all three ARIP measures tested were significantly different across the CMV groups, with slightly higher ratios in the CMV equivocal group compared to the negative group, and lower ratios in the CMV positive and above LOD participant groups compared to the negative group.

The age and sex adjusted CMV associations for all 119 immune cell phenotypes for individuals with CMV measurement are presented in Supplementary Table 4. Among the 119 total immune cell phenotypes and ARIP measures, 63 had significantly different levels across the four CMV serostatus groups (Supplementary Table 4); results were similar when including “Missing” as an additional CMV category (Supplementary Table 5). We tested for interactions between age and CMV, and between sex and CMV, but did not observe significant interactions (data not shown).

Association of immune cell phenotypes with age group adjusting for multiple covariates

As shown in Table 5, seven of the 27 primary immune cell phenotypes listed in Table 1 differ across the 5 age groups after adjusting for the “Model 2” covariates described in Methods at FDR<0.05. The age effect did not appear to be linear with the increasing age groups for most of the seven significant cell types.

CD3+ T cell levels were lower for each increasing age group. CD4 Tn cell levels were the only CD4 cell subtype with significant differences by age: levels were lower in the older groups compared to the younger. The two youngest age groups differed by only 0.01 SD unit, and the two oldest age group means differed by only 0.02 SD units. CD8 Tn, Teff, and Granyzme B producing CD8 cells all differed by age group at FDR<0.05. CD8 Tn cell levels were lower in each increasing age group, while the other two cell types had higher levels in older age groups.

The adjusted mean CD4 Tn/Tm ratio was similar in the two younger age groups, and lower in the three older age groups. The CD8 Tn/Tm ratio had larger differences across age groups than the CD4 Tn/Tm ratio, and the ratio was lower in each subsequent increasing age group. Together, these data are consistent with an age-related shift from naïve cells to a more senescent/exhausted phenotype, with the shift occurring in CD8 cells more evident than CD4 cells.

The multivariable adjusted associations with age groups for all 119 immune cell phenotypes are presented in Supplementary Table 6. Significant associations were seen, particularly in the CD8 compartment that are consistent with immune exhaustion from repeated replication (ie loss of CD28, CD27 and increase in CD57 expression).

Association of immune cell phenotypes with sex adjusting for multiple covariates

Table 6 shows that 18 of the 27 primary cell types differed by sex at FDR<0.05, adjusting for Model 2 covariates. Males had lower percentages of CD4 T helper cells and higher percentages of CD8 cytotoxic cells than females. Among the CD4 and CD8 subtypes, males had lower percentages than females of both CD4 and CD8 Tn cells, as well as CD8 Tcm cells. Males had higher percentages than females of both CD4 and CD8 Teff, Tem, TEMRA, and Granzyme B producing cells. Males also had a significantly higher percentage of NK cells and significantly lower percentages of B cells compared to females. The CD4/CD8 ratio and both the CD4 and CD8 Tn/Tm ratios were lower in males compared to females.

Sex associations for all 119 total immune cell phenotypes and ARIP measures are presented in Supplementary Table 7. Significant associations were observed that are consistent with male participants having higher levels of immune activation/exhaustion than female participants.

Association of immune cell phenotypes with smoking status adjusting for multiple covariates

Table 7 shows that four of the 27 primary immune cell phenotypes and ARIP measures differed significantly (FDR<0.05) between current smokers and non- smokers (former or never smokers) after adjusting for Model 2 covariates. The percentages of CD8 TEMRA, Tc1, and Granzyme B producing CD8 cells were significantly lower in smokers compared to non-smokers, while CD8 FoxP3 T regulatory cell percentages were higher in smokers compared to non-smokers. It is worth noting that, while the FoxP3 transcription factor in CD25+ cells is well-known to be a regulatory cell for CD4+ cells, less is known about the true function of CD25+FoxP3+ CD8 cells. Smoking status was not significantly associated with the three ARIP measures.

Supplementary Table 8 summarizes the effect of smoking status for all 119 immune cell phenotypes; 20 of the 119 phenotypes were significantly different (FDR<0.05). Notably, although smoking status was not significantly associated with B cell levels, 8 of the B cell subtypes were significantly different in smokers compared to non-smokers. While the memory B cells were higher with smoker, the IgM+ and IgD+IgM+ memory cells were lower with smoking. (Supplementary Table 8).

Associations of immune cell phenotypes with other variables

Adjusting for Model 2 covariates, we did not observe significant associations of the 116 immune cell phenotypes and the 3 ARIP measures with education, APOE genotype, obesity status, type 2 diabetes, prevalent CVD or AF at exam 7, hypertension treatment, lipid treatment, SBP, TC, and HDL levels, or diagnosis of chronic blood cancer prior to exam 7 (all FDR>0.05). Estimated effect sizes for each group, p–values, and FDRs are reported in Supplementary Tables 9–18.

Study sample

The FHS is a community-based prospective cohort study. Starting in 1948, the FHS recruited 5209 participants residing in Framingham, Massachusetts who were primarily white American adults of European ancestry as the Original cohort [38], followed by the Offspring cohort (n=5129) which was recruited in 1971, consisting of the children of the Original cohort and spouses of the Offspring [39, 40]. The Offspring cohort participants have been examined every 4-8 years since enrollment. A total of 3539 Offspring participants attended the seventh examination from 1998 to 2001, which is the first examination with existing stored PBMCs.

For the current study, we identified 1332 Offspring participants who attended exam 7 (1998 to 2001) and had 2 or more vials of PBMCs stored (to ensure the resource was not exhausted) that can be used to profile immune cells. Among this group of Offspring participants, we identified a study sample of 1000 individuals aged 40 years and older who were dementia-free at exam 7. Figure 2 is a flow chart showing participant inclusion and exclusion in the study. All participants provided written informed consent at each attended examination; FHS exams are reviewed and approved by the Institutional Review Board at Boston University Medical Center.

Immunophenotyping methods

Immune cell phenotyping protocols followed previously published protocols used in other large population studies [41, 42]. Briefly, PBMCs were cryopreserved at the Framingham Heart Study Offspring Cohort Exam 7. The PBMCs were thawed in a 37° C water bath for 15 minutes. Cells were diluted to 10x initial volume by the slow addition of supplemented media with gentle mixing, pelleted, washed, and resuspended in phosphate-buffered saline and filtered. The filtered cells were divided into 5 assay panels for immunophenotyping. Fluorescent antibodies utilized for this study, manufacturer, catalog numbers, and matching isotypes used in specific assay panels are indicated in Supplementary Table 3.

For surface labeling panels (GDNK, MEMORY, MONOCYTES), samples are incubated with a fixable live/dead stain for 15 minutes. After centrifugation to remove the live/dead stain, cells were incubated with the indicated antibodies or appropriate isotype for 15 minutes in the dark at concentrations recommended by manufacturers. Cells were pelleted, washed with media, resuspended in 1% paraformaldehyde, and stored in the dark at 4C until run on the flow cytometer.

For intracellular staining, samples are stimulated with phorbal myristic acetate/Ionomycin in the presence of Brefeldin A for 3 hours at 37C as previously described [41, 42]. Cells are collected by centrifugation, washed, and incubated with CD3/CD4/CD8 antibodies for 15 minutes in the dark. Cells were washed and fixed with paraformaldehyde for 10 minutes. Cells were collected by centrifugation and incubated for 15 minutes with appropriate antibodies or isotype controls in the presence of saponin (a pore forming reagent). Cells were washed and stored in 2% paraformaldehyde in the dark at 4C until run on the flow cytometer.

Flow cytometry was performed on a MACSQuant 16. Single color compensation beads were used to set machine compensation. Isotype controls and FMO controls were used to aid in gate setting. FCS Express 6.0 was used to analyze data. A Vericell PBMC control sample (Biolegend) was analyzed with each run to assess run variability, with coefficients of variation (CVs) ranging from 2.5% for CD3+, 12.3% for CD4+, 17.5% for CD8+ and 7.8% for CD14+ monocytes.

All data are presented as a percent of their main parent population as listed in Supplementary Table 1. Major cell types analyzed for this study are displayed in Figure 1A–1E. and gating strategies utilized are shown in Supplementary Figures 3–7.

We excluded 13 immune cell phenotypes with very low variability (whose CV < 0.074 or IQR < 0.088, the 5% percentiles), leaving 116 cell phenotypes. The number of participants with data and summary statistics for each immune cell phenotype profiled are presented in Supplementary Table 2.

We calculated three ARIP measures using immune cell phenotypes profiled in our study: the CD4/CD8 ratio, and two measures of the ratio between naïve cells and memory and effector cells for both CD4 T cells and CD8 T cells. For the ratio between naïve T cells and memory T cells, Tn/Tm, we used the formula Tn/Tm = Tn / (Teff + Tem + Tcm) separately for the CD4 and CD8 Tn cells, Teff cells, Tem cells, and Tcm cells [17]. The number of participants with data and summary statistics of the three ARIP measures is summarized in Supplementary Table 2.

Cytomegalovirus

CMV is a common pathogen that influences T cell subsets in individuals infected with CMV. Therefore, we assayed CMV IgG by ELISA using the Creative Diagnostics CMV IgG kit (Catalog # DEIA326R) using an existing plasma sample from Offspring exam 7 from 943 of the participants with immune cell phenotyping. We obtained quantitative values in U/ml; the CV for the CMV assay was 5.9%.

Statistical analysis

Our analyses excluded 4 participants with missing data for most of the immune cell subtypes, resulting from failure of these samples for multiple assays. The 116 immune cell phenotypes had sample sizes ranging from n=922 to 996. The number of participants with data and summary statistics for each immune cell phenotype profiled are presented in Supplementary Table 2. Missingness in cell phenotypes data was due to poor sample quality, technical assay errors, or not enough cells to assay a specific subset. Number of missingness for each type is listed in Supplementary Table 2. We stratified age into five groups: age ≤ 55, 55 < age ≤ 60, 60 < age ≤ 65, 65 < age ≤ 70, and age > 70. We categorized CMV into 4 groups: negative (CM ≤ 11 U/ml), equivocal (11< CMV ≤ 15 U/ml), positive, below the limit of detection (LOD) (15 < CMV ≤ 800 U/ml) and positive, above the LOD (CMV > 800 U/ml). CMV was unavailable for 57 participants due to lack of available plasma sample from exam 7. Participant characteristics of individuals missing CMV values did not differ substantively from participants with CMV measured. For complete case analyses, individuals missing CMV status were omitted.

We computed unadjusted mean percentages of immune cell phenotypes for Table 1 immune cell types stratified by age, sex, and CMV group using complete-case data.

For all association analyses, immune cell phenotypes were rank-based inverse normal transformed and treated as the outcome variable. To test for association between CMV status and immune cell phenotypes, we used individuals with CMV status available and adjusted for age and sex (Model 1) using a linear mixed effect model, accounting for correlations among related individuals using a random effect with the kinship matrix.

To test for associations of immune cells with age, sex, smoking status as current smokers vs. non-smokers (never or former), and other covariates, we used a multivariable model (“Model 2”) that included CMV, age, sex, education level, APOE genotypes, and other CVD risk factors including BMI, smoking, presence of diabetes, prevalent CVD, prevalent AF, SBP, on hypertension treatment at exam 7, TC, HDL, on lipid treatment at exam 7, and prevalent cancer prior to exam 7. All covariates except APOE ε4 carrier status and education level were directly measured at the FHS Offspring exam 7. Education level was ascertained at FHS Offspring exams 2 and 8, and during off-exam-cycle neuropsychological testing. We used highest education level reported and categorized it into four levels: attended high school, high school graduate, attended college, and college graduate. APOE genotype groups were defined as ε3ε3, ε2 carriers, and ε4 carriers with the ε3ε3 group treated as the reference level. BMI was grouped into three levels: normal (BMI < 25), overweight (25 ≤ BMI < 30), and obese (BMI ≥ 30). The presence of diabetes was determined if any of the following were satisfied: fasting blood glucose level of 126 mg/dL or higher, random blood glucose level of 198 mg/dL or higher, or use of antidiabetic medications. Prevalent CVD was defined at exam 7 based on current or previous diagnosis of coronary heart disease (myocardial infarction, angina pectoris, coronary insufficiency), transient ischemic attack, intermittent claudication, and congestive heart failure determined by adjudication of a panel of senior investigators.

We used multiple imputation by chained equations [43] to create 10 imputed data sets to impute missing CMV values, immune cell phenotypes, and other missing data among the 996 participants in variables obtained from the Offspring exam 7, including the covariate data for the multivariable adjusted models (Model 2 covariates). In the imputation models, log-CMV levels were treated as quantitative with values above LOD set to the LOD. Age, sex, education level, APOE genotypes, all immune cell phenotypes, and available exam 7 variables were used in the imputation. For each immune cell phenotype outcome and each imputed data set, we fit a linear mixed effect model adjusting for Model 2 covariates accounting for correlations among related individuals using a random effect with the kinship matrix. Effect estimates and likelihood ratio p-values were pooled across data sets.

To account for multiple testing, false discovery rates [44] (FDRs) across the 116 immune cell phenotypes plus 3 ARIP measures were calculated separately for each predictor tested for association.

We used a threshold of FDR < 0.05 to declare significance. All analyses were conducted in R software version 4.0.2 [45]. The multiple imputations including pooling of the effect estimates and p–values from the fitted models were carried out using the mice package [43]. Linear mixed effect models were fitted using lmekin function from the coxme package [46].