Social relationships and immune aging in early midlife: Evidence from the National Longitudinal Study of Adolescent to Adult Health

Abstract

Aging of the immune system is characterized by changes in the T-cell compartment, including a decrease in naïve T-cells and an increase in memory T-cells. Stress exposures are known to predict accelerated immune aging in older adults. However, social relationships, which are often linked to stress mechanisms, have not been widely studied in relation to these adaptive immune biomarkers, particularly in younger populations. We examined associations between social relationships, in terms of quantity (Social Network Index, Close Contacts Index) and quality of relationships (spouse/partner, friends, and family members), and immune aging in a U.S-representative early midlife population (age 33–44) from Wave V of the National Longitudinal Study of Adolescent to Adult Health (n = 4451). DNA methylation data of venous blood samples collected during Wave V were used to compute CD4⁺ memory:naïve, CD8⁺ memory:naïve, and total CD8+:CD4+ T cell ratios; higher values indicate a more aged immune profile. Results from survey-weighted linear regression models adjusted for age, sex, race/ethnicity, and education indicated higher number of close friends and frequency of contact, alongside higher quality relationships with family members were associated with decreases in CD4⁺ memory:naive ratios. The results for CD8⁺ memory:naïve and CD8+:CD4+ ratios were mostly non-significant. Our findings suggest that higher quantity and quality of social relationships may help protect against immune aging, particularly in the CD4⁺ T cell compartment, prior to midlife. This underscores the importance of interventions that enhance social relationships throughout life to promote healthy longevity.

Highlights

• •Population-based study of immunosenescence in early midlife adults (age 33–44).
• •Significant associations were found only for CD4⁺ naive:memory T cell ratios.
• •Having more close friends and frequent contact related to decreased immunosenescence.
• •High quality relationships with family were linked to decreased immunosenescence.
• •Immune aging may be a mechanism through which social relationships impact health.

1. Introduction

Aging of the immune system, known as immunosenescence, is a process resulting in declining immune efficacy that can lead to increased susceptibility to infections and reduced response to vaccinations (Aiello et al., 2019; Pera et al., 2015). Immune aging is characterized by elevated levels of systemic low-grade inflammation and alterations in the adaptive immune system, primarily composed of B and T lymphocytes. As individuals age, there is a decrease in the production of naïve T cells due to thymic involution, an accumulation of differentiated memory T cells resulting from pathogen exposure and shifts in T cell surface markers from CD4⁺ to CD8⁺ (Arnold et al., 2011; Thyagarajan et al., 2022). Additionally, differentiated memory CD8⁺ T cells have been implicated in the production of inflammatory cytokines, whereby having a larger proportion of those markers contributes to chronic low-grade inflammation (Pangrazzi and Weinberger, 2020; Aiello et al., 2019). Furthermore, infection with the latent herpesvirus cytomegalovirus (CMV) has been established to accelerate immune aging (Aiello et al., 2017; Pawelec et al., 2005). CMV is a highly prevalent infection that has been found to be associated with alterations in T cell compartments (Pawelec and Derhovanessian, 2011).

Previous research has established that accelerated immune aging is linked to multiple age-related comorbidities such as cardiovascular diseases, cancer, and aging of multiple organs (Huff et al., 2019; Yousefzadeh et al., 2021). However, studies on trends in immune aging, especially regarding changes in the T cell repertoire, have been mostly limited to older adults and rarely conducted in a population representative sample. Consequently, little is known regarding these patterns in younger adult populations, who are typically considered to have relatively robust immune systems and better health. Nonetheless, recent data suggest that young and early midlife health is deteriorating, as evidenced by increasing chronic illnesses in early midlife adults (age 35–44) (American Psychological Association, 2023). Thus, a deeper understanding of immune aging in younger adult cohorts is crucial to identifying interventions to mitigate adverse health outcomes in later life.

While younger adult populations may not be as susceptible to age-related changes in the immune system, exposure to stress has been found to similarly impact immunity through disruptions in the hypothalamic-pituitary-adrenal (HPA) axis and the sympathetic nervous system (Reed, 2019). Understanding stress-induced immune aging in younger adult populations is increasingly vital, particularly given the elevated stress levels observed among U.S. adults in the post-pandemic era (American Psychological Association, 2023). Notably, adults aged 18 to 44 report significantly higher levels of stress now compared to pre-pandemic levels, surpassing the levels reported by older adults. Furthermore, early midlife adults (age 35–44) have exhibited the highest increases in mental health diagnoses across all adult age groups.

Although a small number of studies have explored the associations between stressors (e.g. low socioeconomic status, post-traumatic stress-disorder, and psychologically adverse work conditions) and immune aging in young and middle-aged adults, they have been mostly limited to non-representative, small-scale, or neighborhood-level samples (Aiello et al., 2016a, 2016b; Bosch et al., 2009). More recent population-representative research has primarily focused on older adults, revealing that older adults with lower socioeconomic status, belonging to minoritized groups, or exposed to discrimination and stressful life events tend to exhibit a more aged immune phenotype, indicated by higher CD4⁺ and CD8⁺ memory T cells compared to naïve T cells (Noppert et al., 2023; Klopack et al., 2022). Additionally, similar associations between higher levels of distress (perceived stress, depression, anxiety, PTSD) and elevated CMV have been reported in previous research (Phillips et al., 2008; Reed et al., 2019; Uddin et al., 2010). While these findings offer valuable insight into the role of stress on immune aging, the lack of similar studies in nationally representative samples of younger adults underscores a gap in research.

Lastly, despite a sizable literature on the adverse effects of stress, studies investigating protective factors for immune aging are less common. One such study found higher purpose in life to be associated with increases in CD4⁺ naive T cells in older adults (Koga et al., 2024). In this study, we focus on social relationships. Social isolation and poor relationship quality can be sources of chronic stress themselves, whereas high social support has been hypothesized to be stress-buffering on adverse health outcomes (Thoits, 2011; Cohen and Wills, 1985). Substantial research exists on the associations between social relationships, stress, and overall well-being (Cacioppo and Hawkley, 2003; Steptoe et al., 2013; Kiecolt-Glaser et al., 2010). Researchers are increasingly studying the biological mechanisms through which social relationships interact with health, with a particular focus on immune responses (Leschak and Eisenberger, 2019). For instance, higher number of social ties has been linked to lower levels of inflammation and a more robust antiviral response (Cohen et al., 1997; Loucks et al., 2006a; Yang et al., 2014; Ford et al., 2019). Conversely, low social support and perceived social isolation have been associated with shorter leukocyte telomere length in older adults, a key feature of immune aging (Carroll et al., 2013). Furthermore, attachment orientations in relationships have been shown to impact telomere length of specific T cell subsets in younger adults via stress pathways (Murdock et al., 2018). Besides its relation to stress and immunity, researchers have observed better relationship quality in older adults to be predictive of slower biological aging, as measured by epigenetic aging clocks (Rentscher et al., 2023; Hillmann et al., 2023). These findings collectively suggest that social relationships may play a protective role in immune aging. However, to the best of our knowledge, this has not yet been studied in a nationally representative sample of early midlife adults.

Thus, in this study, we aim to address the existing gaps in research by investigating whether positive social relationship characteristics serve as protective factors for immune aging in an early midlife U.S-representative sample from the National Longitudinal Study of Adolescent to Adult Health (Add Health). Social relationships encompass many domains, including ties with one's spouse/partner, friends, family members, and the community in which one resides. We focus on both the quantitative characteristics of one's social network and the quality of social relationships overall, and independently with different close contacts. We hypothesize that higher number of social ties and better relationship quality will be associated with a less aged immune system, as demonstrated by lower values of CD4⁺ memory:naïve, CD8⁺ memory:naïve ratios and total CD8+:CD4+ T cell ratios.

2. Material and methods

2.1. Study design and population

Add Health is a longitudinal study of a nationally representative sample of U.S. adolescents who were in grades 7–12 (aged 12–19) during the 1994-95 school year and followed from adolescence to adulthood with five waves of data collection (Harris et al., 2019). Add Health used a multistage, stratified, and clustered sampling design with schools as the primary sampling unit to select a probability sample of more than 20,000 adolescents for the in-home Wave I interview. Further details on study design are published elsewhere (Harris, 2014; Harris et al., 2019). In this study, we used data from Wave V of Add Health.

Wave V followed eligible Wave I respondents between the ages of 33–44 years, yielding a sample size of 12,300 individuals (Biemer et al., 2019). Wave V utilized a mixed-mode survey design of web/mail questionnaires and in-home interviews, with non-response follow up on a sample of nonrespondents conducted through both in-person and telephone interviews. Furthermore, all eligible respondents were invited to complete a separate home exam, resulting in visits arranged for 5381 participants. Venous blood samples (n = 4940) were collected during the home exam. All analyses in this study are weighted using the sampling weights published by Add Health for the sample of individuals who participated in the home exam (Chen and Harris, 2020a).

The analytic sample for this study (n = 4451) consists of participants with non-missing biomarker weights; complete data on post-stratification region variable, epigenetic immune measures, age, sex, race/ethnicity, and educational attainment; and at least one non-missing social relationship measure. Subsamples were used in regression analyses for the different social relationship measures and stepwise regression models. Full details are presented in the flow diagrams in Figure A.1 and Figure A.2.

2.2. Measures

2.2.1. Epigenetic immune measures

DNA was extracted from stored venous blood samples collected from Wave V participants, and methylation analysis was conducted using the Illumina Infinium MethylationEPIC BeadChip (Illumina, Inc., San Diego, CA) (Harris et al., 2024). DNA methylation-based cellular deconvolution method, that relies on the unique DNA methylation pattern of each cell type, was applied (Salas et al., 2022) to estimate proportions of twelve leukocytes.

Based on previously published methods of measuring T cell immunosenescence (Aiello et al., 2016a) and guidelines from Salas et al., we computed the following ratios: i) CD4⁺ memory:naïve, ii) CD8⁺ memory:naïve, and iii) total CD4+:CD8+ T cells. The total composition of CD4⁺ T cells include CD4⁺ naïve, memory and T-regulatory cells while the total composition of CD8⁺ T cells include CD8⁺ memory and CD8⁺ naïve cells. To be consistent with the interpretation of the first two immune cell ratios, we inversely coded CD4+:CD8+ as CD8+:CD4+. Therefore, higher values of each immune cell ratio indicate a more aged immune profile.

Small percentages of participants in the analytic sample (n = 4451) reported zero cell proportions for CD4⁺ naïve (1.80 %), CD4⁺ memory (0.27 %), CD8⁺ naïve (6.16 %), and CD8⁺ memory (1.53 %). We added a small constant 0.001 to zero-proportions before constructing the ratios to avoid dividing by zero values, and all three ratio measures were log-transformed to account for right-skewness.

2.2.2. Social Network Index

Using Wave V survey data, we constructed the Social Network Index, adapted from the Berkman-Syme Social Network Index (Berkman and Syme, 1979). The Social Network Index measures the number of social network ties across several domains: marital status, close friends, religious attendance, and volunteer/community service, and has been utilized as a measure of social networks in multiple studies (Loucks et al., 2006a; Elliot et al., 2018). Moreover, past research has adapted the Social Network Index for use in Waves I and IV of Add Health (Yang et al., 2016; Ford et al., 2019). Using the same criteria, our index comprises the following four dichotomized items: currently married or cohabiting, at least 6 close friends, attended religious services at least 12 times in the past year, and volunteered at least once in the past year. The Social Network Index was constructed for participants who had non-missing responses to all 4 items (n = 4424). Affirmative responses to the four items were summed (range 0–4), and initially categorized as “low – socially isolated” (sum = 0 or 1), “medium-low” (sum = 2), “medium-high” (sum = 3), and “high” (sum = 4). However, due to a small number of people with a sum of 4, we collapsed the two highest categories, those with a sum of 3 or 4 into one group indicating “high” Social Network Index and labeled those with a sum of 2 as “moderate” on the Social Network Index scale.

2.2.3. Close Contacts Index

Since the Social Network Index captures the number of ties with both close contacts and the wider community, we constructed an index specifically focusing on close contacts at Wave V, following previous research (Gafarov et al., 2013). The Close Contacts Index also incorporates individuals’ frequency of contact with family and friends, and thereby acts as a measure of not only the number of social ties but also contact frequency. The Close Contacts Index comprises the following two items: at least 6 close friends and engaged in face-to-face contact with friends/family members at least once a month. This index was constructed for participants with non-missing responses to both items (n = 4443). Responses to the two items were summed (range 0–2) and categorized as “low” (sum = 0), “moderate” (sum = 1), and “high” (sum = 2), i.e. participants with a high score responded affirmatively to both items, participants with a moderate score responded affirmatively to one of the two items, and participants with a low score did not respond affirmatively to either item.

2.2.4. Quality of relationships

Quality of relationships were assessed separately for three types of social ties at Wave V: spouse/partner, friends, and family members (parents, grandparents, siblings, extended family, etc.). For each social tie, participants were asked to indicate “yes” or “no” to: i) “can you open up to them if you need to talk about your worries?”, ii) “can you rely on them for help if you have a problem?”, and iii) “whether or not they make too many demands or criticize you”. The third item was reverse coded to align in direction with the prior two items. Responses to the three items were summed (range 0–3), and since over half of the participants reported quality scores of 3, we dichotomized quality variables as high quality (sum = 3) and low quality (sum <3). Quality of relationship with one's spouse/partner was constructed for 3575 participants with valid and non-missing responses to all three items comprising the measure. Similarly, relationship quality with friends was constructed for 4151 participants, and relationship quality with family members was constructed for 4220 participants.

Finally, a composite score for quality of relationships (range 0–9) was constructed for 3827 participants. This measure was constructed as the following: i) the sum of 9 items (range 0–9) across the three social ties, for participants with complete responses to all items (n = 3352), and ii) the sum of 6 items for quality of relationships with friends and family members (range 0–6), for participants with complete responses to these 6 items who were ineligible to respond to spouse/partner items (n = 475). Initially, the composite score for the second set of participants ranged between 0 and 6. However, to account for the possibility that single participants may compensate for the absence of a romantic relationship through high quality relationships with friends and family, we rescored those with a sum of 6 as 7.5. In addition, we used a different approach to create a second composite score (range 0–1) by standardizing the scores within the two subgroups to range between 0 and 1. This method assigns the same credit regardless of marital/relationship status.

Additional details on each item comprising the social relationship measures are presented in Table A.1.

2.2.5. CMV seropositivity

Previous research has found associations between CMV levels and various stressors in population-level research (Noppert et al., 2021; Aiello et al., 2016b). Therefore, CMV might be a potential confounder or mediator on the pathway between stress and immunosenescence. In Wave V of Add Health, CMV assays were performed using two methods: initially, the enzyme immunoassay (EIA) kit from Diamedix (ERBA Diagnostics, Miami Lakes, FL) followed by the enzyme-linked immunosorbent assay (ELISA) from Creative Diagnostics (Creative Diagnostics, Shirley, NY) after discontinuation of EIA by Diamedix. Further details and data conversion protocols can be found elsewhere (Whitsel et al., 2024). CMV concentrations were standardized to a z-score. In addition, CMV concentrations were classified as: seronegative (CMV z-score≤40.19th percentile), equivocal (CMV z-score between 40.2 and 46.69th percentile), and seropositive (CMV z-score≥46.70th percentile).

2.2.6. Covariates

Covariates collected from Wave V include self-reported age, sex assigned at birth (male, female), and race/ethnicity (White, Black, Hispanic, Asian/Pacific Islander, American Indian/Alaska Native, Other). Due to the small proportion of American Indian/Alaskan Native individuals, they were combined with the “Other” category. Participants reported the number of years of education they had completed, and responses were categorized as no college, some college, and college+. To assess whether the participant experienced a recent inflammatory condition (within the past four weeks), we constructed a binary measure combining responses to survey questions asked during the in-home exam about active infection, acute illness, injury, fever, gum disease, surgery, use of Cox-3 inhibitors, corticotropins/glucocorticoids, anti-rheumatics/anti-psoriatic, and immunosuppressive agents/monoclonal antibodies. We assessed additional measures such as smoking status (never smoker, past smoker, current smoker), BMI (categorized as: normal/underweight, overweight, obese), and depressive symptoms (range 0–15; based on the 5 item Center for Epidemiological Studies-Depression (CES-D) scale (Radloff, 1977). The items included in this measure relate to how many days in the past week participants felt: “could not shake off the blues”, “depressed”, “happy” (reverse-coded), “sad”, and “life was not worth living”.

2.3. Statistical analyses

We calculated weighted descriptive statistics for the analytic sample and subsamples for each non-missing social relationship measure. In addition, we examined the distribution of the social relationship measures by age, sex, race/ethnicity, and educational attainment. Furthermore, we calculated the median CD4⁺ memory:naïve, CD8⁺ memory:naïve, and CD8+:CD4+ T cell ratios and 95 % confidence intervals, by age group, sex, race/ethnicity, educational attainment, inflammatory condition, smoking status, BMI, depression, and social relationship measures.

We conducted survey-weighted least squares regression to quantify the association between each social relationship measure and log-transformed immune cell ratios, with Taylor series linearization method to estimate sampling errors of estimators. We fit separate models for each of the five social relationship measures (Social Network Index, Close Contacts Index, quality of relationships with spouse/partner, friends, and family members) and each of the three T cell ratio outcomes. In our main analyses, we controlled for covariates that were hypothesized beforehand to be potential confounders of the relationship between social relationships and immune aging. Considering each social relationship measure separately, we fit two sets of models for each outcome: Model 1 adjusted for age and sex, and Model 2 additionally adjusted for race/ethnicity and educational attainment.

We performed five secondary analyses. First, we conducted an additional set of models (Model 3) where we controlled for measures that were hypothesized to affect immunosenescence, such as smoking status, BMI, depressive symptoms, and presence of a recent inflammatory condition. In a following set of models (Model 4), we also included adjustments for z-scored CMV levels, which may be a strong predictor of immunosenescence. Third, we investigated whether the associations between social relationships and cellular immunosenescence may vary by CMV infection status by rerunning our main analyses stratified by CMV seropositivity. Given infection status and the immune response to CMV may play a role in the relationship between stress and immune aging, we included an additional model with adjustments for CMV z-score for CMV seropositive participants. Fourth, to assess whether associations between social relationships and immunosenescence differ between males and females, we conducted sex-stratified analyses of our main models.

In sensitivity analyses, we fit models with the continuous measure of the Social Network Index (standardized, range 0–1), quality of relationships with spouse/partner, friends, and family members (range 0–3) instead of the categorical measures we used in our main analyses, due to variations in how researchers have scored these measures in the past. Lastly, we conducted our main analyses with the composite quality score (range 0–9).

All analyses were weighted using Add Health Wave V biomarker sampling weights, applying appropriate subpopulation procedures (Chen and Harris, 2020b). Statistical analyses were conducted in SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina) and figures were created in R (version 4.3.1; R Core Team, 2023).

3. Results

3.1. Sample characteristics

Descriptive characteristics of the analytic sample are provided in Table 1. The sample was split evenly by sex, with a weighted mean age of 37.9 (range 33–44) years. More than two-thirds of the sample identified as White (70.2 %) and 41.7 % attained a college education or higher. Regarding health status, almost half (46.4 %) of the sample was obese, 25.8 % reported smoking at the time of the survey, 14.9 % reported a recent inflammatory condition, and 52.3 % of the participants were categorized as seropositive for CMV. The mean CES-D score was 2.46 (range 0–15). Table 2 displays the medians and 95 % confidence intervals of the outcome measures, CD4⁺ memory:naive, CD8⁺ memory:naive, and CD8+:CD4+ ratios by sociodemographic and health behavior variables. Correlations between raw immune cell proportions, ratios, and CMV are presented in Figure A.4.

Table 1.

Descriptive characteristics of analytic sample.

Characteristics	Frequency (%)	N
Age (years), mean (range)	37.9 (33–44)	4451
Sex		4451
Male	1784 (49.6)
Female	2667 (50.4)
Race/ethnicity		4451
White	2872 (70.2)
Black	850 (17.2)
Hispanic	457 (8.61)
Asian/Pacific Islander	227 (2.67)
Other	45 (1.26)
Educational attainment		4451
No College	667 (17.7)
Some College	1712 (40.6)
College+	2072 (41.7)
Smoking Status		4446
Never Smoker	2473 (50.5)
Past Smoker	965 (23.6)
Current Smoker	1008 (25.8)
BMI		4386
Normal/underweight	1114 (24.2)
Overweight	1261 (29.2)
Obese	2011 (46.4)
Depression (CES-D),mean (range)	2.46 (0–15)	4396
Recent inflammatory condition	665 (14.9)	4451
CMV		4423
Seronegative/equivocal	2067 (47.6)
Seropositive	2356 (52.3)
Social Network Index, mean (range)	0.40 (0–1)	4424
Social Network Index		4424
Low – socially isolated	2079 (50.7)
Moderate	1365 (29.2)
High	980 (20.1)
Close Contacts Index		4443
Low	1214 (27.9)
Moderate	2635 (59.3)
High	594 (12.8)
Relationship quality: spouse/partner		3575
Low quality	1149 (33.3)
High quality	2426 (66.7)
Relationship quality: friends		4151
Low quality	1053 (27.3)
High quality	3098 (72.7)
Relationship quality: family members		4220
Low quality	1724 (40.8)
High quality	2496 (59.2)
Relationship quality (composite),mean (range)	7.36 (0–9)	3827

Survey-weighted percentages are shown, with unweighted frequencies. N depicts the number of non-missing observations.

Table 2.

Median immune cell ratios and 95 % confidence intervals by descriptive characteristics.

Overall	CD4+ memory:naïve			CD8+ memory:naïve			CD8+:CD4+
	Median	95 % CI		Median	95 % CI		Median	95 % CI
	1.72	(1.64,	1.79)	1.96	(1.84,	2.07)	0.52	(0.50,	0.53)
Age
≤median (37.5)	1.61	(1.49,	1.71)	1.64	(1.52,	1.75)	0.52	(0.51,	0.54)
> median	1.81	(1.67,	1.96)	2.25	(2.09,	2.41)	0.51	(0.49,	0.53)
Sex
Male	1.75	(1.61,	1.89)	2.22	(1.99,	2.46)	0.54	(0.51,	0.57)
Female	1.69	(1.60,	1.77)	1.76	(1.61,	1.91)	0.50	(0.48,	0.51)
Race/ethnicity
Asian/Pacific Islander	1.21	(1.04,	1.37)	2.45	(1.25,	3.65)	0.72	(0.63,	0.81)
Black	2.35	(2.11,	2.58)	1.80	(1.49,	2.11)	0.50	(0.47,	0.53)
Hispanic	1.88	(1.47,	2.29)	2.12	(1.67,	2.56)	0.57	(0.52,	0.62)
Other	2.20	(1.15,	3.26)	6.06	(0.79,	11.3)	0.53	(0.42,	0.64)
White	1.63	(1.53,	1.72)	1.93	(1.81,	2.06)	0.50	(0.49,	0.52)
Education
No College	2.22	(1.95,	2.48)	2.35	(1.78,	2.92)	0.50	(0.48,	0.52)
Some College	1.77	(1.67,	1.87)	1.90	(1.71,	2.09)	0.51	(0.48,	0.53)
College+	1.50	(1.42,	1.57)	1.90	(1.72,	2.08)	0.53	(0.51,	0.56)
BMI Categories
Normal/Underweight	1.52	(1.37,	1.67)	1.89	(1.62,	2.16)	0.53	(0.51,	0.56)
Overweight	1.63	(1.47,	1.78)	1.76	(1.54,	1.98)	0.53	(0.50,	0.56)
Obese	1.88	(1.76,	2.01)	2.09	(1.92,	2.27)	0.49	(0.47,	0.52)
Smoking Status
Never Smoker	1.67	(1.58,	1.75)	2.06	(1.90,	2.21)	0.54	(0.52,	0.56)
Past Smoker	1.60	(1.47,	1.73)	1.94	(1.68,	2.20)	0.50	(0.48,	0.53)
Current Smoker	2.00	(1.83,	2.16)	1.83	(1.62,	2.04)	0.47	(0.44,	0.50)
Inflammatory Condition
No	1.69	(1.62,	1.75)	2.05	(1.93,	2.16)	0.52	(0.51,	0.54)
Yes	1.96	(1.70,	2.23)	1.63	(1.43,	1.84)	0.47	(0.45,	0.50)
Depression
≤median (1.17)	1.72	(1.60,	1.84)	2.08	(1.89,	2.26)	0.53	(0.51,	0.55)
>median	1.71	(1.62,	1.80)	1.87	(1.70,	2.04)	0.50	(0.49,	0.52)
CMV
Seropositive	2.08	(1.96,	2.19)	3.00	(2.71,	3.28)	0.60	(0.57,	0.63)
Seronegative/equivocal	1.41	(1.33,	1.50)	1.25	(1.16,	1.34)	0.44	(0.42,	0.46)
Social Network Index
Low – socially isolated	1.77	(1.65,	1.89)	1.95	(1.77,	2.13)	0.50	(0.48,	0.52)
Moderate	1.71	(1.58,	1.83)	1.92	(1.71,	2.12)	0.53	(0.50,	0.56)
High	1.62	(1.48,	1.75)	2.05	(1.77,	2.32)	0.54	(0.52,	0.56)
Close Contacts Index
Low	1.92	(1.74,	2.09)	1.93	(1.68,	2.19)	0.50	(0.48,	0.52)
Moderate	1.69	(1.60,	1.78)	1.93	(1.79,	2.08)	0.52	(0.50,	0.53)
High	1.46	(1.27,	1.64)	2.10	(1.80,	2.40)	0.56	(0.52,	0.61)
Relationship quality: spouse/partner
Low quality	1.80	(1.66,	1.94)	1.89	(1.67,	2.11)	0.52	(0.49,	0.55)
High quality	1.63	(1.52,	1.73)	1.98	(1.83,	2.13)	0.52	(0.50,	0.54)
Relationship quality: friends
Low quality	1.91	(1.73,	2.09)	1.83	(1.60,	2.07)	0.49	(0.47,	0.51)
High quality	1.67	(1.58,	1.75)	2.04	(1.89,	2.19)	0.53	(0.51,	0.54)
Relationship quality: family members
Low quality	1.83	(1.68,	1.97)	2.02	(1.85,	2.18)	0.51	(0.49,	0.52)
High quality	1.63	(1.53,	1.73)	1.96	(1.79,	2.13)	0.53	(0.51,	0.54)

Just over 20 % of individuals with non-missing Social Network Index and 12.8 % of individuals with non-missing Close Contacts Index were observed to score high on these scales. Amongst those with non-missing responses for each relationship type, 66.7 % reported high-quality spouse/partner relationships, 72.7 % reported high-quality friendships, and 59.2 % reported high-quality relationships with family members. Table A.2. provides details of the distributions of social relationship measures by median age, sex, race/ethnicity, educational attainment and CMV seropositivity. All social relationship measures were positively correlated with one another and negatively correlated with depressive symptoms (Figure A.3).

3.2. Linear regression

Regression coefficients and 95 % confidence intervals from Models 1, 2, 3, and 4 are presented in Fig. 1, Fig. 2, Fig. 3. Full regression results can be found in Tables A.3 - A.5.

Fig. 1.

Social networks and immunity. Regression coefficients and 95 % confidence intervals from models estimating the associations between Social Network Index and each log-transformed immune cell ratio. Covariate adjustments: Model 1 = age and sex; Model 2 = Model 1 + race/ethnicity and education; Model 3 = Model 2 + BMI, smoking status, depressive symptoms, presence of an inflammatory condition; Model 4 = Model 3 + CMV Z-score.

Fig. 2.

Close contacts and immunity. Regression coefficients and 95 % confidence intervals from models estimating the associations between Close Contacts Index and each immune cell ratio. Covariate adjustments: Model 1 = age and sex; Model 2: Model 1 + race/ethnicity and education; Model 3: Model 2 + BMI, smoking status, depressive symptoms, presence of an inflammatory condition; Model 4: Model 3 + CMV Z-score.

Fig. 3.

Quality of relationships and immunity. Regression coefficients and 95 % confidence intervals from models estimating the associations between quality of relationships and each log-transformed immune cell ratio. Covariate adjustments: Model 1: age and sex; Model 2: Model 1 + race/ethnicity and education; Model 3: Model 2 + BMI, smoking status, depressive symptoms, presence of an inflammatory condition; Model 4: Model 3 + CMV Z-score.

3.2.1. Immune cell ratios by quantitative measures of social relationships

The results of the regression models for Social Network Index and Close Contacts Index are presented in Fig. 1, Fig. 2. Higher number of social ties as measured by the Social Network Index was associated with statistically significant decreases in log-CD4⁺ memory:naïve ratios, adjusted for age and sex. For example, high Social Network Index scores (vs. low– socially isolated) was associated with a 0.10-unit decrease in log-CD4⁺ memory:naive ratios (95 % CI: 0.20, −0.00). However, this association was not statistically significant after adjusting for race/ethnicity and education (Model 2). We did not find any statistically significant associations between log-CD8⁺ memory:naïve ratios and the Social Network Index. For overall CD8+:CD4+ ratios, high Social Network Index scores (vs. low - socially isolated) was associated with a 0.05 (95 % CI: 0.00, 0.10) increase in log-CD8+:CD4+ in Model 1. This association was no longer significant after adjustments for race/ethnicity and educational attainment.

Adjusted for age and sex, moderate and high Close Contacts Index scores (vs. low) were associated with a 0.17 (95 % CI: 0.27, −0.07) and 0.30 (95 % CI: 0.44, −0.15) decrease in log-CD4⁺ memory:naïve ratio, respectively. After controlling for race/ethnicity and education, only the association between high Close Contacts Index scores (vs. low) and CD4⁺ memory:naive ratios remained statistically significant (β = −0.16 95 % CI: 0.31, −0.01). For the CD8⁺ T cell compartment, we found moderate Close Contacts Index scores (vs. low) to be associated with lower log-CD8⁺ memory:naive ratios (β = −0.16, 95 % CI: 0.31, −0.01). Lastly, similarly to the Social Network Index, we found high Cloes Contacts Index scores (vs. low) to be significantly associated with increases in log-CD8+:CD4+ in Model 1 (β = 0.08, 95 % CI: 0.01, 0.16), but this lost statistical significance after further covariate adjustments.

In secondary analyses, after additional adjustments for smoking status, BMI, presence of an inflammatory condition, depressive symptoms and CMV (Model 3 and 4), the association between high Close Contacts Index scores (vs. low) and CD4⁺ memory:naive ratios was slightly attenuated but remained statistically significant in Model 4 (β = −0.14, 95 % CI = −0.28, 0.00).

3.2.2. Immune cell ratios by quality of relationships

Regression coefficients for quality of relationships and immune cell ratios are presented in Fig. 3. High quality relationships with all three social ties were significantly associated with decreases in log-CD4⁺ memory:naïve ratios, adjusted for age and sex (Model 1). For example, in comparison to low quality, high quality relationships with spouse/partner were associated with a decrease of 0.13 (95 % CI = −0.23, −0.02), higher quality friendships were associated with a decrease of 0.15 (95 % CI = −0.27, −0.02), and higher quality relationships with family members were associated with a decrease of 0.14 (95 % CI = −0.23, −0.05) in log-CD4⁺ memory:naive.

After adjusting for race/ethnicity and educational attainment (Model 2), only high-quality relationships with family members (vs. low quality) remained statistically significant (β = −0.10, 95 % CI = −0.18, −0.02). No statistically significant associations were found between log-CD8⁺ memory:naïve ratios and relationship quality. Finally for log-CD8+:CD4+ ratios, high quality relationships with friends (vs. low quality) was associated with a 0.08 (95 % CI = 0.01, 0.15) increase in log-CD8+:CD4+ ratio, in Model 1. However, this lost statistical significance after adjusting for race/ethnicity and educational attainment.

In secondary analyses, the association between quality of relationships with family members and log-CD4⁺ memory:naive ratios remained statistically significant after additional covariate adjustments in Model 3 (β = −0.09, 95 % CI = −0.18, −0.01). Adjusting for CMV slightly reduced the effect size and reduced the association to non-significance (β = −0.08, 95 % CI = −0.17, 0.00).

3.2.3. Secondary analyses

The results of the models stratified by CMV seropositivity are shown in Table A.6 - A.8.

Overall, we observed stronger results in CMV-seropositive individuals compared to CMV-seronegative individuals. Associations remained statistically significant in CMV-seropositive individuals after adjusting for standardized CMV antibody levels. The results of our sex-stratified analyses are displayed in Tables A.9 - A.11. We found the associations between social relationships and immune cell ratios to be stronger in females than males. In sensitivity analyses (Table A.12), we found the association between spouse/partner quality and CD4⁺ memory:naive ratios to be statistically significant in fully adjusted models, and as strong as the association between quality of relationships with family members and CD4⁺ memory:naive ratios. We did not observe statistically significant associations between composite relationship quality scores and outcomes. Further details on secondary and sensitivity analyses are presented in Supplementary Information.

4. Discussion

In this study, we investigated whether the quantity and quality of positive social relationships were linked to reduced cellular immunosenescence in a U.S. early midlife population. Our findings showed that higher number of close friends (alongside frequency of interactions) and higher-quality relationships with family members (excluding spouse/partner and children) were consistently associated with improved functional capacity of the adaptive immune system, particularly in the CD4⁺ T cell compartment. Similar results were observed in the CD8⁺ T cell compartment for individuals moderately integrated with close contacts (e.g., having at least six close friends or visiting friends/family at least once a month). Although higher-quality relationships with friends and spouse/partner were not statistically significant after adjusting for race/ethnicity and education, the trends suggested that these relationships might still protect against immune aging, even at this younger stage of life. Sensitivity analyses also supported the importance of quality relationships with spouse/partner and family members, though no associations were found between relationship quality and changes in the CD8⁺ T cell compartment. Unexpectedly, higher number of social ties and better-quality friendships were linked to increased CD8+:CD4+ ratios, which indicate a more aged immune system.

Our study contributes to the growing literature on the importance of social ties and social support for immune health, investigating the potential role of positive social relationship factors as protective for immune aging. To the best of our knowledge, no previous study has assessed the associations of social relationships with immune aging, as measured by the distribution of the T cell subtypes. Therefore, we compare our results with studies that have assessed the role of social relationships in immune health. Our results suggest quality of relationships, specifically with one's family members, is a stronger predictor of immune aging than the Social Network Index. These findings are consistent with what other researchers have found when studying the Social Network Index and relationship quality concurrently (Ford et al., 2019). Moreover, additional findings (Yang et al., 2016) suggest that the number of social ties may be more important for physical health in early and late adulthood, whereas quality of social relationships may be better at capturing health in midlife. With respect to the three social ties we assessed, our findings suggest quality of relationships with family members are more strongly associated with cellular immune age than spouse/partner relationship and friendship quality. This aligns with a previous study where social strain with family members was more strongly associated with inflammatory burdens than strain in other relationship types among middle-aged adults (Yang et al., 2014).

In stratified analyses, we found quantitative measures of social relationships and quality of relationships to be more strongly associated with markers of immunosenescence in females compared to males, as depicted by larger decreases in CD4⁺ memory:naive ratios. Our results align with a study that reports older females may benefit from higher social support with regards to their adaptive immunity, but not males (van der Velpen et al., 2024). Older males in their sample were reported to gain benefits of higher social support with respect to innate immunity. In our study however, we focused on measures of adaptive immunity, therefore testing associations with innate immunity was beyond the scope of this paper. Prior research suggests sex differences in the association between social relationships and immune health vary across age groups with more robust findings reported in older adults compared to middle-aged adults, and results are not consistent across different inflammatory markers such as interleukin-6 and C-reactive protein (Yang et al., 2013; Elliot et al., 2018). Structural and quantitative measures of social networks such as the number of social ties or marital status may be more strongly associated with inflammation in older men (Yang et al., 2013; Loucks et al., 2006b), whereas marital quality may be more predictive of inflammatory burdens in women (Donoho et al., 2013). However, it is important to note that the population we are studying is much younger than samples typically used in previous studies, and therefore, previous studies may not accurately reflect expected associations in an early midlife cohort.

Our findings also revealed that the associations between social relationships and T cell ratios were mostly significant in CMV-seropositive individuals. This is consistent with another study where they found, among CMV-seropositive individuals, high levels of perceived stress were associated with higher proportions of late-differentiated T cells, and that there was a significant interaction between CMV IgG levels and perceived stress in relation to the proportion of late-differentiated T cells (Reed et al., 2019). High stress was associated with higher proportions of late-differentiated T cells in CMV-seropositive individuals, regardless of CMV IgG levels. CMV is a burden on the immune system (Aiello et al., 2017). Thus, CMV seropositive individuals may be more susceptible to either the beneficial effects of social support or the detrimental effects of social isolation than those who are not subject to the burden of CMV infection.

Focusing on the different T cell compartments, we found social relationship measures to be significantly associated mainly with changes in the CD4⁺ T cell compartment. Other studies have found similar results when investigating the associations of stressors and immunosenescence. For instance, income levels were found to be significantly associated with CD4⁺ memory:naïve ratios but not CD8⁺ memory:naïve and total CD8+:CD4+ ratios in a community-based sample of middle-aged adults (Aiello et al., 2016a). Furthermore, increases in CD4⁺ memory cells in older adults were strongly associated with lifetime discrimination and chronic stress, more so than CD8⁺ T cells (Klopack et al., 2022). Similarly, previous research studying the associations between psychological well-being such as purpose in life and immune cells also reported more robust results in the CD4⁺ T cell compartment (Koga et al., 2024).

Lastly, the associations between total CD8+:CD4+ ratios and social relationship measures were mostly not in the direction we expected. Previous studies in older adults have found this measure to be significantly associated with socioeconomic factors and chronic stress. Our contrasting results could be partly due to the younger age range of our cohort (age 33–44). Past research shows that the prevalence of individuals with more CD8⁺ T cells than CD4⁺ T cells is much greater among individuals aged 60+ than younger adults (Strindhall et al., 2013; Li et al., 2019).

Our study has limitations that need to be addressed. Firstly, past research suggests that quality of relationships should be assessed independently as social support and social strain, as these may underlie distinct constructs (Yang et al., 2014, 2016; Rentscher et al., 2023). Few studies have comprehensive measures of social strain and social support alongside these epigenetic immune measures. Since we chose to create comparable measures of quality across all three social ties, we were only able to assess three items in Add Health for each relationship. Due to the lack of measures to capture social strain across all three ties, we chose to reverse-code the item that pertains to strain and create an overall measure for quality of relationships. Future research could aim to incorporate more comprehensive measures beyond the 3-item scale we created for each relationship quality. It should be noted that the Social Network Index for individuals in Wave V indicated a high percentage of people with “low – socially isolated” scores, which contrasts with the Social Network Index for Wave IV constructed by researchers previously (Yang et al., 2016). This was driven by far fewer people reporting at least 6 close friends in Wave V (14.5 % in Wave V vs. 29.3 % in Wave IV). Therefore, to understand how changes in this index are associated with immune aging, different operationalizations depending on the life stage may be worth considering in future research. Furthermore, researchers have scored the Social Network Index in various ways for other population-based studies and a different approach to scoring may be worth assessing in Add Health (Elliot et al., 2018; Kornej et al., 2022).

Moreover, no previous study has studied these DNA methylation immune cell proportions in population-based studies. Most research in this area has been conducted using immune cell estimates obtained through flow cytometry methods. The algorithms used herein are relatively new, developed and published by Salas et al., in 2022. As part of their work, Salas et al. validated their DNAm measures of immune markers with reference to flow cytometry (i.e. gold standard) measures. While the sample size was small, they found good validity across most markers. However, the algorithm for the specific naive and memory subtypes of CD4⁺ cells was less reliable than other measures.

Finally, we only investigated the cross-sectional association between social relationships and immune aging, and thereby cannot make causal inferences. DNA methylation measures of immune cell proportions were first obtained in Wave V of Add Health, and therefore the changes in immunosenescence cannot be studied with the data currently available.

5. Conclusions

Our study identified significant associations between social relationships and greater cellular aging in a population-based cohort of early midlife adults. Our results suggest higher scores on the Close Contacts Index (number of close friends and frequency of contact) and high quality relationships with family members (parents, grandparents, siblings, etc.) were protective for the functional capacity of the immune system. Therefore, immune aging may be one mechanism through which social relationships impact health.

CRediT authorship contribution statement

Farizah I. Rob: Writing – original draft, Formal analysis, Conceptualization. Rebecca C. Stebbins: Writing – review & editing, Supervision, Conceptualization. Jennifer Momkus: Writing – review & editing, Conceptualization. Chantel L. Martin: Writing – review & editing, Funding acquisition. Kathleen Mullan Harris: Writing – review & editing, Supervision, Funding acquisition, Conceptualization. Allison E. Aiello: Writing – review & editing, Supervision, Funding acquisition, Conceptualization.

Ethics statement

This study uses data from Wave V of the Add Health study and the Add Health Epigenome Resource, which were conducted in compliance with relevant laws and institutional guidelines approved by the University of North Carolina at Chapel Hill Institutional Review Board (Add Health Wave V: IRB #13–1946, initial approval August 05, 2013; The Add Health Epigenome Resource: IRB #18–2176, initial approval: 08/30/2018). In addition, we received Columbia University IRB approval (Add Health Wave V: IRB AAAU4491, initial approval: 12/20/2022; Add Health Epigenome Resource: IRB AAAU4174, initial approval: 01/17/2023).

Add Heath participants provided written informed consent for participation in all aspects of Add Health in accordance with the University of North Carolina Institutional Review Board guidelines that are based on the Code of Federal Regulations on the Protection of Human Subjects 45CFR46. Privacy rights of human subjects were respected, and further details on privacy can be found on the Add Health website.

Funding

This research uses data from Add Health, funded by NIH grant P01HD31921 (Harris) from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), with cooperative funding from 23 other federal agencies and foundations. Add Health is currently directed by Robert A. Hummer and funded by the National Institute on Aging cooperative agreements U01 AG071448 (Hummer) and U01AG071450 (Aiello and Hummer) at the University of North Carolina at Chapel Hill. Add Health was designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill. This project was funded by grant R01MD013349 from the National Institute on Minority and Health Disparities to Kathleen Mullan Harris and Allison Aiello. Additional funding was received from R01AG057800 & P30AG066615 from NIA and T32HD091058 & P2CHD050924 from NICHD. The funders had no role in study design, data collection, data analysis, interpretation, or writing of the report.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article:

Data availability

The authors do not have permission to share data at this time. The data will be publicly released by 2026 in the Add Health website.