Healthy participants were selected from the EpiPath cohort (Elwenspoek et al., manuscript under review, Journal of Immunology), based on absence of chronic or acute diseases and medication use, and adequate number of bio-banked peripheral blood mononuclear cells (PBMCs) for investigation. The EpiPath cohort was recruited between 2014 and 2016 from Luxembourg and The Greater Region Saar-Lor-Lux and consisted of young adults, aged 18–35 years, that were either parent-reared (Ctrl) or experienced separation from their parents in early childhood followed by adoption (ELA). 59 Ctrl and 18 ELA participants were included in this study (Figure S1 in Supplementary Material). One adoptee was directly adopted from the birth family, all others experienced the additional stress of institutionalization, which is considered to be a form of social deprivation and structural neglect (32). All participants gave their written informed consent. The study design was approved by the Ethics Review Panel of University of Luxembourg (ERP, No 13-002) and the National Research Ethics Committee (CNER, No201303/10) in compliance with the Declaration of Helsinki.

Blood samples were collected in sodium heparin-coated tubes for PBMC isolation and in EDTA-coated tubes for DNA and plasma isolation. To minimize inter-individual variation, all samples were collected at the end of the morning (ca. 11:30 a.m. ±30 min); participants were asked to refrain from smoking, strenuous physical exercise, and drinking caffeinated or alcoholic beverages on the day of the clinical visit; women were either using hormonal contraceptives or were in the luteal phase of their menstrual cycle. Furthermore, participants’ age and sex were recorded. At a second visit, information about the age at adoption was obtained and the childhood trauma questionnaire (CTQ) was administered.

EDTA blood samples were centrifuged at 4°C within 15 min of blood collection and the plasma phase was collected. Plasma samples were transported on ice and stored at −80°C within 6 h until further analysis. Plasma was used in a 1:21 dilution to determine CMV IgG antibody indexes by ELISA (Calbiotech, El Cajon, CA, USA). As a control, HSV-1 and EBV IgG antibody indexes were also determined with ELISAs from the same company. The ELISAs were performed following manufacturer’s instructions, in duplicate, and read on a SpectraMax Plus 384 Microplate Reader (Molecular Devices, Berkshire, UK).

DNA isolation and telomere length measurements in the EpiPath were reported previously in Elwenspoek et al. (manuscript under review, Journal of Immunology). Methylation levels were measured at age-related CpGs in aspartoacylase (ASPA) (cg02228185), integrin alpha 2b (ITGA2B) (cg25809905), and PDE4CA (cg17861230) according to (22). In brief, unmethylated cytosine residues in each DNA sample were converted to uracil with a bisulfite treatment (EpiTect Bisulfite Kit, Qiagen, Venlo, Netherlands) and regions of interest were amplified with PCR (PyroMark PCR Kit, Qiagen) in the bisulfite-modified DNA according to the manufacturer’s protocols. PCR products were pyrosequenced on a Pyromark ID with Pyrogold reagents (Biotage, Uppsala, Sweden) and methylation levels were analyzed with Pyro Q-CpG SW (Biotage). A sample of pooled DNA was run in each batch as internal control, which was used to calculate relative methylation levels. These relative methylation levels were used for all further analyses.

All cell culture products were from Lonza BioWhittaker (Versviers, Belgium), unless otherwise stated. PBMCs were isolated within 3 h of sample collection using Ficoll-Paque density gradient centrifugation. Briefly, EDTA blood was diluted in sterile 1× PBS, layered over Ficoll-Paque™ PLUS (Fisher Scientific, Erembodegem-Aalst, Belgium) in Leucosep tubes (Greiner Bio-One, Vilvoorde, Belgium), and centrifuged for 5 min at 300 g. PBMCs were washed twice with PBS and stored at 4.10⁶ cells/1 mL/aliquot in 80% Heat-Inactivated Fetal Bovine Serum (Gibco, Paisley, UK) and 20% DMSO (Sigma-Aldrich, Saint-Louis, CA, USA) in liquid nitrogen until analyzed.

Peripheral blood mononuclear cells were thawed quickly and rested overnight at 37°C, 5% CO₂ in RPMI 1640 medium, with 10% Heat-Inactivated Fetal Bovine Serum, 1% penicillin/streptomycin, 1% sodium pyruvate (Gibco), 1% non-essential amino acids (Gibco), and 1% ultra-glutamine. PBMCs (1.10⁷ cells/mL) were incubated with GolgiPlug and GolgiStop (final concentrations of 2 and 1 µL/mL, respectively, BD BioSciences) for 5 h at 37°C, 5% CO₂. All subsequent steps were performed at 4°C and protected from ambient light. PBMCs were washed twice with 1× FACS Buffer [1× PBS, 1% bovine serum albumin Cohn fraction V (Sigma-Aldrich, Saint-Louis, CA, USA), 0.1% NaN3, 2 mM EDTA (Sigma-Aldrich), pH 8.0] and stained with a LIVE/DEAD dye and antibodies against CD4, CD3, CD8, CD45RA, human leukocyte antigen–antigen D related (HLA-DR), CCR7, and CD57 (Table S1 in Supplementary Material) for 30 min. Then, PBMCs were permeabilized and fixed with BD Cytofix/Cytoperm™ (BD BioSciences, San Diego, CA, USA) for 20 min, followed by a 30 min intracellular staining of granzyme B (GraB) and perforin (Table S1 in Supplementary Material).

Thirty thousand lymphocyte events were acquired on the BD LSRFortessa (BD BioSciences) using FACSDiva (BD BioSciences, version 8.0). Data analysis was performed with FlowJo (version 10.2, Tree Star, Ashland, OR, USA) using the gating strategy presented in Figure S2 in Supplementary Material. T cell differentiation was determined by CCR7 and CD45RA expression: naïve (CCR7⁺CD45RA⁺), CM (CCR7⁺CD45RA⁻), EM (CCR7⁻CD45RA⁻), and terminally differentiated cells (terminally differentiated effector memory; TEMRA, CCR7⁻CD45RA⁺). Relative numbers of cells (e.g., CD57⁺ cells) and median fluorescent intensity (MFI) were analyzed.

Group differences in telomere length, relative methylation levels of age-related CpGs, CMV titers, CTQ sum scores, and age at adoption were investigated with a Wilcoxon rank sum test with continuity correction. In the initial analysis, we constructed linear regression models to investigate group differences in cell types (flow cytometry data), in which “cell type” was included as outcome variable and both groups and experimental day were included as fixed effects; the latter to account for variation between experiments. Cell percentages were transformed with the arcsine transformation (asin(sign(x) × sqrt(abs(x)))) to stabilize variance and MFI values with a log-transformation to approximate normality. Analyses of variance (ANOVAs) were performed on each model to test the significance of the group effect. The Benjamini–Hochberg procedure was performed on all p-values generated by the ANOVAs to correct for multiplicity (33). When plotting CMV titers, a “high” and “low” antibody group emerged. Log(CMV titers) = −1.2 was chosen as cutoff. To investigate the relationship between CMV levels and CD57 expression, Spearman’s rank correlation rho were determined. To investigate the mediating effect of CMV on the association between ELA and senescent cells, a mediation analysis was performed in R [version 3.3.3 (34) using mediation version 4-4.5 (35)]. Because age at adoption can be considered to be proportional to the duration of adversity and can thus be used as proxy for ELA severity (36), in the mediation analysis, age at adoption (months) was used as continuous variable for ELA, in which controls were set to 0. p-Values below 5% were considered significant.

The Ctrl and ELA groups did not differ in age or sex. The median age at adoption was 3.4 months. Adoptions took place at an early age, so participants had no memory of the time before adoption. Consequently, the CTQ scores, based on the participant’s memory of trauma experiences before age 16, reflects experiences after adoption, which was similar between groups. Thus, apart from the adoption, the experimental groups were comparable in age, sex, subsequent childhood trauma exposure, and all participants were in good health (Table 1; Figure S1 in Supplementary Material).

Immunosenescence in leukocytes was measured with two distinct techniques. First, as previously reported, we did not observe a difference in telomere length between the two groups in the complete EpiPath cohort (Elwenspoek et al., manuscript under review, Journal of Immunology). Also in the subset of participants used in this investigation, which only included healthy participants, there was no effect of ELA on telomere length (median [interquartile range; IQR]; Ctrl: 1.2 [0.8–1.8], ELA: 1.1 [0.8–1.7], p = 0.714; Figure 1A). Second, methylation levels at three age-related CpGs that have been linked to chronological and biological age were measured (22). We found similar methylation levels in ELA and Ctrl (ASPA Ctrl: 1.01 [0.97–1.03], ELA: 1.01 [0.98–1.04], p = 0.452; ITGA2B Ctrl: 1.05 [1.00–1.12], ELA: 1.08 [1.02–1.12], p = 0.438; phosphodiesterase (PDE4C) Ctrl: 0.99 [0.93–1.10], ELA: 1.05 [0.97–1.13], p = 0.258; Figures 1B–D).

A 10-color flow cytometry panel was used to investigate cell-specific immunosenescence. Although immunosenescence is related to changes in the ratio of naïve and memory T cells, the ratios and numbers of naïve, CM, EM, and terminally differentiated T cells (TEMRA) were similar between Ctrl and ELA (data not shown). However, we found a higher number of T cells expressing the senescence marker CD57. ELA was associated with a significant increase in both the total number of T cells (linear regression, adjusted p = 0.017) and T helper (Th) cell subset (adjusted p = 0.038), expressing CD57 (Figure 2A). The cytotoxic T lymphocytes (CTLs) showed a similar trend toward higher CD57, albeit this did not reach statistical significance (adjusted p = 0.061). The increase in CD57⁺ cells between Ctrl and ELA appeared to be highest in Th cells, showing almost a 1.5-fold increase, and lowest in overall lymphocytes, consisting of a mixture of B cells, T cells, and NK cells.

There was also a significant group effect on the intensity of CD57 fluorescence (MFI, T cells: adjusted p < 0.001, Th cells: p = 0.024, CTLs: p = 0.001; Figures 2B,C), suggesting that CD57 expression was higher in ELA. The median fluorescent intensity (MFI) of CD57 was 5–10% higher in ELA than Ctrl on lymphocytes, total T cells, Th, and CTLs. CTLs showed the lowest increase in CD57 expression. As expected (37), CD57⁺ cells were not equally distributed among the different stages of T cell differentiation (Figure 3). In both Th and CTL subsets, EM and TEMRA cells had the highest number of CD57⁺ cells. The increase in CD57⁺ cells in ELA was mainly happening in EM Th cells (Figure 3A) and in both TEMRA and EM CTLs (Figure 3B).

Cytolytic potential was measured by GraB and perforin staining. As expected (38), CD57⁺ cells had higher GraB, perforin, and HLA-DR expression (Figures 4A,B), suggesting higher cytolytic potential and a higher activation status in senescent Th and CTLs. When comparing the senescent cells (CD57⁺) between groups, GraB and perforin expression was elevated in ELA, although the expression of the activation marker HLA-DR was similar (Figures 4C,D). The level of fluorescent intensity of neither GraB nor perforin differed between the groups (data not shown).

In the final step of the analysis, we investigated whether CMV infection had a mediating effect on the relationship between ELA and immunosenescence. Initially, we tested the difference in CMV titers between the two groups. Indeed, titers were higher in the ELA group (medium [IQR], Ctrl: 0.13 [0.09–0.64], ELA: 0.82 [0.33–0.94]; p = 0.023; Figure 5A), caused by an increased number of seropositive participants. To further investigate the relationship between CMV titers and CD57 expression, we divided the control participants into two groups with either “high” or “low” CMV titers. Participants with high CMV titers had significantly higher levels of CD57⁺CD4 EM (p < 0.001) and TEMRA (p < 0.001) cells, as well as higher CD57⁺ in all subsets of CD8 (Naïve, p = 0.007; CM, p < 0.001; EM, p = 0.001; TEMRA, p = 0.020; Figure 5B). Subsequently, we tested the correlation between the number of senescent cells (CD57⁺) or the level of CD57 expression on these cells and CMV titers. Indeed, we found highly significant and strong correlations in T cells, and its subsets Th and CTLs (Figure 5C). In contrast, titers of EBV and HSV, herpes viruses that cause similar lifelong latent infections, were not elevated in ELA (EBV: Ctrl: 0.91 [0.53–1.78], ELA: 1.43 [0.72–2.10], p = 0.213; HSV: Ctrl: 0.22 [0.12–1.78], ELA: 0.29 [0.13–1.82], p = 0.906) nor where they correlated to the number of senescent T cells (EBV: rho = 0.10, p = 0.370; HSV: rho = −0.01, p = 0.900). A causal mediation analysis demonstrated a large mediating effect of CMV titers (Figure 6A), which could explain 27.4% of the total effect of ELA (age at adoption) on T cell senescence (Table 2). Interestingly, when investigating ELA as mediator of the effect of CMV on senescent T cells, we only found a trend (Table 3; Figure 6B). Furthermore, there was a clear correlation between the age of adoption and CMV titer (Figure S3 in Supplementary Material, Spearman’s rho = 0.71, p = 0.0071), while there was no clear correlation between CMV titers and age at time of study participation among control participants (Figure S4 in Supplementary Material, Spearman’s rho = 0.17, p = 0.15). This suggests that ELA is not just a marker for CMV exposure, but that there are other ELA-related factors involved (Table 3; Figure 6B).

Differences in ethnic background between the Ctrl and ELA group and small sample size were the main limitations in this study. Nevertheless, we have taken several measures to reduce variability. ELA participants came from various regions in the world, and there was not one particular region overrepresented, so we believe that a possible effect of genetic background would have been limited (Table 4). Furthermore, all blood samples were collected at the same time of day, which minimized the impact of circadian rhythms of hormones and immune cells on the results. It is important to note that in the original sample of the EpiPath cohort, ELA participants were almost four times more likely to have a chronic disease than controls. Participants with any chronic disease had higher CMV titers than healthy participants, which may have been secondary to medication use or the disease itself. To avoid effects of present disease and medication on the immune variables, we specifically selected a healthy subset of the cohort. Finally, we accounted for possible variation between experimental days by including this as a factor in the final statistical model.

It is important to note that freezing and thawing as well as resting overnight of PBMCs may have introduced artifacts in the analysis. For instance, resting overnight before staining has been shown to decrease viability and can change T cell phenotype (59). Nevertheless, we could repeat our previous findings on fresh samples of the same cohort concerning HLA-DR expression on T cells (Elwenspoek et al., manuscript under review, Journal of Immunology).