Associations of social network size and perceived level of social support with age acceleration estimated by a proteomic aging clock: The Atherosclerosis Risk in Communities Study

Abstract

Background:

We examined the associations of social network size and social support with biological age acceleration using a protein-based aging clock.

Methods:

A total of 11,719 participants (mean age 57.1±5.7 years; 55.6% female) of the Atherosclerosis Risk in Communities (ARIC) study Visit 2 (1990–1992) were included. Biological age was calculated using a proteomic aging clock (PAC), and age acceleration was defined as the residuals after regressing the PAC on chronological age. The associations of self-reported social network size and level of support with standardized mean age acceleration were examined using multivariable linear regression.

Results:

Size of social networks and level of support were inversely associated with age acceleration. The age accelerations of small, medium, and large social networks were 0.24, 0.08, and −0.05 (p for trend = 0.003), while for low, moderate and high levels of support the age accelerations were 0.33, 0.19, and −0.06 (p for trend < 0.0001), respectively. Associations were modestly attenuated after adjusting for social and lifestyle factors.

Discussion:

Findings of the benefit of social networks and support to biological aging processes provide evidence of potential physiological underpinnings of social interactions on health outcomes.

Background

The average life expectancy is increasing worldwide, especially in developed countries. Healthy aging, which is a continuous process of optimizing opportunities to maintain and improve physical and mental health, independence, and quality of life throughout the life course¹, is a priority for individuals and the broader society.

Chronological age corresponds to the length of time a person has lived. However, people age differently due to various factors such as lifestyle, the environment and genetics. This led to the development of the concept of biological age, which estimates the age of an individual’s body and cells. Biological age is correlated with chronological age², but accruing research suggests that biological age “better predicts functional capacity at later ages than chronological age³.” Aging clocks make it possible to provide specific information about how old an individual is biologically, independent of chronological age⁴. Age acceleration occurs when biological age, estimated from aging clocks, is greater than chronological age⁵. The standard method to calculate age acceleration is by taking the residuals after regressing the aging clock on chronological age^6–9.

To create aging clocks an individual’s biological age is estimated from a set of molecules, such as DNA methylation^2,6,7,10,11 or proteomic biomarkers^4,8,9,12,13. Epigenetic clocks, based on biomarkers derived from DNA methylation in blood and/or tissues, have been the most studied^6,7 and age acceleration estimated by these clocks has been associated with adverse health outcomes. Second generation epigenetic clocks, such as GrimAge which was trained against mortality and included pack-years of smoking, have been shown to predict death, ischemic heart disease, cancer, fatty liver/excess visceral fat, menopause, etc., even when adjusted for chronological age¹⁰. However, the underlying mechanisms of biological age-related changes in DNA methylation sites remain unclear. Thanks to technological developments that allow measurements of thousands of proteins simultaneously, it has become possible to construct Proteomics Aging Clocks (PACs)^4,8,9,12,13. Proteomics provides a more direct picture of the state and characteristics of an organism at the molecular level compared to epigenetics because it allows the comprehensive composition and function of proteins to be examined¹⁴. As a result, it more accurately reflects the state of individual organs and tissues compared to epigenetic clocks and may provide a more accurate estimate of an individual’s organ age¹⁴. The PAC used in the present analyses was developed in the Atherosclerosis Risk in Communities study and validated in the Multi-Ethnic Study of Atherosclerosis, and has been shown to predict cardiovascular mortality with better accuracy than chronological age, even after adjusting for confounding factors such as age, smoking, and BMI⁸.

Social interactions, through both social support and the breadth and depth of social networks may influence health. Social support refers to actions or systems that provide psychological, emotional, material or support to individuals¹⁵. Greater social support has been associated with positive health outcomes, while social isolation has been linked to poorer health^16–19, including an increased risk of heart disease^19,20 and stroke^21,22. Although mechanisms of these associations have not been fully elucidated, psychological stress related to social isolation may adversely affect the cardiovascular system²³. Social networks are groups that individuals have personal, social or professional connections²⁴. Individuals with greater social networks and support may be more likely to engage in behaviors that promote health²⁵, which may be associated with reduced biological age and lead to lower disease risk²⁶. In recent years it has been reported that healthy lifestyle factors such as eating a healthy diet, regular physical activity, healthy sleep, healthy weight, smoking cessation, and control of blood glucose, cholesterol, and blood pressure can slow biological aging²⁷.

Few studies have explored associations between social interactions and biological aging^28,29. A recent analysis indicated that individuals with less social support had accelerated aging according to DNA methylation-based aging clocks²⁹. However, the associations of social network and support with proteomics-based age acceleration has not yet been reported. Therefore, using data from the ARIC study, we estimated the associations of social network size and perceived level of social support with biological age acceleration using a protein-based aging clock.

Methods

The ARIC study is a prospective epidemiologic cohort of 15,792 predominantly Black and White men and women, aged 45–64 years at the baseline Visit 1 (1987–1989)³⁰, recruited from four US communities: Washington County, Maryland; the northwest suburbs of Minneapolis, Minnesota; Jackson, Mississippi; and Forsyth County, North Carolina. A population-based sample was recruited from three of the communities, which resulted in the study sample almost exclusively identifying as White in the suburbs of Minneapolis, Minnesota and Washington County, Maryland, while in Forsyth County, North Carolina 15% of the sample self-identified as Black and 85% as White. In the Jackson, Mississippi field site, only individuals identifying as Black were recruited. Ongoing follow-up of ARIC cohort participants is conducted through annual (semi-annual since 2012) telephone interviews, and surveillance of hospitalizations and in-person clinic visits³¹. Institutional review boards at each of the participating institutions approved the study, and all participants provided informed consent.

Study population

Size of social networks and perceived social support were measured at Visit 2 (1990–1992), attended by 14,348 participants. In this study, we excluded 2,550 participants with missing information about social interactions, and 37 participants who identified as neither White nor Black, and 42 Black participants from the Maryland and Minnesota study communities due to low numbers which inhibit meaningful inference. Our final analytic sample included 11,719 participants.

Social network and perceived social support assessment

Social networks, which are groups with personal, social, or work-related connections, were measured using the 10-item Lubben Social Network Scale.³² The scale provides a composite measure of the participant’s active social network of family, friends, and interdependent relationships. The family network questions query the participant’s number of relatives seen monthly, frequency of social contact, and number they feel “close to”; The friends network queries the same about friendships. Interdependent relationships questions assess whether the participant has a confidant, is a confidant, relies upon and/or helps others, and their living arrangements. The total score ranged from 0 to 50. As has been done in prior analyses,²² we merged the smaller groups and adopted these three categories: ≤25 = small social network; 26–30 = moderate social network; ≥31 = large social network.

Perceived social support, including psychological, emotional, material, or individual behavior or systems that provides support, was measured using a modified version of the Interpersonal Support Evaluation List-Short Form (ISEL-SF)³³. This 16-item scale assessed perceived social support with four subscales; (a) appraisal support, (b) tangible assets support, (c) belonging support, and (d) self-esteem support³⁴. Each item ranged from “definitely false” to “definitely true”, with the total score ranging from 0 to 48. Similar to a previous study, four categories were created: ≤16 = lack of social support; 17–23 = low social support; 24–31 = moderate social support; ≥32 = high social support²². For analysis, we merged the smaller groups and adopted theses three categories: ≤23 = low social support; 24–31 = moderate social support; ≥32 = high social support.

Measurement of covariate

Information on covariates, including social factors, lifestyle factors, and physiologic characteristics, was assessed at ARIC Visit 1 and Visit 2. Income, education, job status, and physical activity were assessed at Visit 1. Physical activity was evaluated using a modified version of the Baecke physical activity questionnaire, which assesses activity in sports, leisure, and work domains. For this analysis, we used data from the sports component. Participants reported engagement in up to four sports during the previous year, including the number of hours per week and months per year spent on each sport. Based on responses regarding the type, frequency, and intensity of these activities, a semi-continuous index ranging from 1 (low activity) to 5 (high activity) was calculated^35,36. Age, sex, race-center, marital status, smoking status, alcohol use, and sleep quality (trouble falling asleep and waking up repeatedly during the night), weight (kg) and height (m) were assessed at Visit 2. Weight, height, and blood biomarker levels were measured by trained interviewers. BMI was^35,36assessed as weight (kg) divided by height (m) squared.

Blood collection, protein measurement and quality control

Proteomics were measured in 11,719 blood samples collected at Visit 2 using standardized venipuncture and processing procedures.³⁷ Plasma samples were analyzed using a SOMAmer (Slow Off-rate Modified Aptamers) based capture array called SomaScan^® by Somalogic, Inc. (Boulder, CO, USA)^38–41. The reproducibility and variability of the analytes measured by this platform have been previously evaluated⁴². Among the 5,284 available aptamers, we excluded aptamers with a Bland-Altman coefficient of variation (CVBA) greater than 50% or a variance of less than 0.01 on the log scale, or binding to mouse Fc-fusion, contaminants, or non-proteins⁴³. About 5% of proteins had more than one aptamer binding to the same protein. Each aptamer was treated as a variable in the construction of PACs. The Bland-Altman coefficient of variation (CVBA) for split samples was 6% at Visit 2.

Proteomic aging clocks

In this study, we employed the midlife (Visit 2) ARIC PAC that was developed by Wang et al in “healthy participants”³⁷, and the weight of the identified aptamer (protein) was used to calculate the PAC for each participant in this study. Healthy participants were defined as participants without major age-associated diseases that have been linked to premature mortality³⁷. Those diseases include abnormal kidney function (i.e., estimated glomerular filtration rate (eGFR) less than 60 mL/min/1.73m²), cancer, chronic obstructive pulmonary disease (COPD), CVD (heart failure, definite or probable stroke, or coronary heart disease^44,45), diabetes, and hypertension³⁷.

Details of the validation have been detailed previously³⁷. To summarize, to construct the midlife ARIC PAC, we first randomly selected two-thirds of healthy participants at Visit 2 and used them as the training set. The remaining one-third of healthy participants were used as the test set. The training set was used to train the midlife ARIC PAC against chronological age, obtain the appropriate hyperparameter values, and weighting for each aptamer: chronological age=β₀+∑ⁿ_i=1  β_i×aptamer_i, where aptamer_i is the level of the ith aptamer. The midlife ARIC PAC has been validated by calculating its correlation with chronological age in the test set and by examining its association with mortality in total ARIC participants, including persons who develop a disease in the future after excluding the training set.³⁷ As a result, elastic net regression selected 788 aptamers out of 4955 aptamers (4712 proteins) for the midlife ARIC PAC in our study. ARIC PAC has also been validated in the Multi-Ethnic Study of Atherosclerosis cohort³⁷.

In addition to examining the midlife ARIC PAC, for comparison we also computed the previously published PAC by Sathyan’s et al¹³. Sathyan’s PAC was originally developed in the LonGenity cohort (age range: 65–95, 55.7% female). We calculated Sathyan’s PAC using the proteins identified in that study as well as the published weights¹³. The original Sathyan’s PAC was constructed from 1,025 participants (mean age 75.8 years [(range: 65–95 years)], and used an elastic net regression model that aimed at selecting proteins that predicted calendar age. Sathyan identified 162 proteins associated with calendar age out of 4,265 proteins examined, of which 61 proteins have been associated with chronological age.

We created age acceleration for ARIC PAC and Sathyan’s PAC as residuals after regressing the PAC on chronological age⁴⁶. A higher number indicates an older biological age relative to the chronological age. The cutoff point of the largest quintile of the age acceleration was 1.98. We also conducted analyses categorizing participants by whether they had >2 as older biological age, to facilitate interpretability.

Statistical analysis

Descriptive statistics of participant characteristics, according to categories of social network score and perceived social support score, were calculated. ARIC PAC was identified in the training set, in ‘all participants minus the training set’, and in ‘all participants’. The main results included “all participants”. Linear regression was conducted to obtain marginal mean age acceleration calculated by the midlife ARIC PAC and Sathyan’s PAC according to the size of the social network and perceived level of social support. Logistic models were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) for the associations of social network and social support with having a 2-year older biological age. Nested models were used to adjust for covariates. Model 1, our primary model, adjusted for age (continuous), sex (men and women), and a 5-level variable cross-classifying race and center (White and Washington County, White and Forsyth County, Black and Forsyth County, Black and Jackson, and White and Minneapolis). Model 2 adjusted for model 1 covariates, and additionally for education (completed less than high school, high school or vocational school, and college or graduate school) and marital status (married and not married). Model 3 adjusted for model 2 covariates as well as behavioral and physiologic factors which may mediate the associations of social network and social support with the PAC (i.e., smoking status (smoker and non-smoker), alcohol use (current drinker and non-current drinker), exercise (leisure-time sports index that ranged from 1 to 5), sleep quality (trouble falling asleep or not), and body mass index (<23, 23–25, 25–27.5, 27.5–30, 30–32.5, and ≥32.5 kg/m²). Additionally, we adjusted for model 3 covariates for eGFR (quintiles), since protein concentrations from the SomaLogic platform have been shown to be negatively correlated with kidney function⁴³. P-values for trend were obtained from testing beta coefficients for ordinal forms of social network and social support variables. Effect modification by sex, race (White or Black), age category (≥60 or <60) and marital status were tested using cross-product terms of those variables by social network or social support score (continuous). Stratified results were provided regardless of whether an interaction was present. As sensitivity analyses, the same linear regression analyses were repeated 1) after excluding the ARIC participants who were in the healthy aging PAC training set and 2) using the Sathyan’s PACs. SAS version 9.4 was used for all analyses.

Results

The 11,719 participants in our final analytic sample were on average 57.1 (SD±5.73) years old, 56% were female, and 23% self-identified as Black. Average age acceleration was −0.002 (±2.73) (Table 1).

Table 1:

Means and proportions of participant characteristics according to the size of social network score: The ARIC Study (1990–1992)

Size of social network	Small	Moderate	Large
Social network score	≤25	26–30	≥31
N total	1216	1607	8896
Age and ARIC PAC
Age, mean years ± SD	57.3±5.81	57.2±5.89	57.0±5.69
ARICPAC, years ± SD	58.0±4.94	57.8±4.82	57.5±4.63
Other demographics
Male, %	52.4	49.8	42.3
Black, %	31.3	25.9	21.8
Income, %
<$25,000	50.2	39.2	30.5
$25,000–$49,999	29.2	35.8	37.4
≥$50,000	14.8	20.2	26.6
Education, %
Less than high school	30.6	23.0	20.2
High school graduate	38.5	42.1	42.5
Beyond high school	30.6	34.6	37.2
Employed, %	66.0	71.6	69.7
Married, %	56.3	64.0	80.5
Behavioral characteristics
Physical activity*	2.28±0.79	2.40±0.79	2.48±0.80
Current drinker, %	54.6	57.4	56.9
Current smoker, %	30.1	27.3	20.4
Trouble falling asleep, %	25.6	24.4	21.1
Wake up repeatedly during the night, %	43.3	39.8	37.8
Physiologic characteristics
BMI, kg/m2 ± SD	27.9±5.44	27.7±5.30	28.0±5.33
eGFR, ml/min/1.73m2 ± SD	95.6±15.8	96.4±14.3	96.8±13.7
ARIC PAC of development subset
ARICPAC, years ± SD (training set)	274	407	2305
	56.1±4.6	56.3±4.5	56.0±4.3
ARICPAC, years ± SD (without training set)	942	1200	6591
	58.5±4.9	58.3±4.8	58.1±4.6

PAC: proteomic aging clock, BMI: body mass index, eGFR: estimated glomerular filtration rate

^*Physical activity was assessed using a leisure-time sports index that ranged from 1 to 5

Social network

Table 1 shows the age-adjusted means and proportions of characteristics of participants according to social network size categories. The number (and percentage) of participants who had a small, moderate, and large social network was 1,216 (10.4%), 1,607 (12.7%), and 8,896 (75.9%), respectively. Those with a large social network mere more likely were more likely to be female, White, nonsmokers, have higher income, have more education, be married, and experienced less trouble falling asleep and waking up repeatedly during the night. Mean age acceleration was approximately the same in men (−0.02±0.04) and women (0.02±0.03) (p for difference=0.41), and in younger age groups (0.03±0.03) and older age groups (−0.05±0.04) (p for difference=0.13). Mean age acceleration was higher in Black participants (0.29±0.05) than White participants (−0.09±0.03) (p for difference <0.0001), and higher in non-married (0.33±0.04) than married (−0.23±0.03) (p for difference <0.0001). (These results are only in the text). Among the participants, 19.9% had ARIC PAC ≥2 years older than their chronological age.

Figure 1 and Table S1 show the average age acceleration calculated by ARIC PAC according to the size of the social network, overall, without the training set, and stratified by sex, race, age category and marital status. Overall, having a larger social network was associated with smaller age acceleration (i.e., the age, sex, and race-center adjusted average age accelerations for those with small, moderate, and large social network were 0.24, 0.08, and −0.05, respectively; p for trend 0.003). This association was not significant after adjusting for social factors, lifestyle factors, BMI, and eGFR suggesting that at part of the association can be explained by these factors (p for trend was 0.07 for Model 2, 0.49 for Model 3, and 0.45 in a model additionally adjusting for eGFR ). The trend of the average age acceleration, calculated by ARIC PAC according to the size of the social network, was similar in ‘all participants minus the training set’ to those with ‘all participants’. In stratified analyses, there were no statistically significant interactions, but the association appeared somewhat more pronounced for women, participants self-identifying as White, participants <60 years, and participants who were not married.

Figure 1. Marginal means age accelerations calculated by ARIC PAC according to the size of social network: The ARIC Study (1990–1992).

The y-axis shows the number of participants by social network size within each subgroup. The results present age acceleration values from Model 1, which was adjusted for age, sex, and race-center. In the figure, solid lines indicate p-values for trend, while dotted lines indicate p-values for interaction.

Table 2 shows the odds ratios and 95% confidence intervals of having greater biological age acceleration (age acceleration of >2 years) across categories of social network size. The age, sex and race-center adjusted odds ratio (OR) and 95% confidence interval (95% CI) was 0.75 (0.65–0.87) for those with large social network versus small social network, and p for trend was <0.0001. With additional adjustment for behavioral and physiologic factors which may be mediators of the association, the OR CIs for large versus small social networks crossed 1, though the p-trend remained significant, at 0.006 for Model 2, 0.04 for Model 3, and 0.03 in a model adjusting for Model 3 covariates and eGFR.

Table 2:

Odds ratios and 95% confidence intervals of having older aging clock (age acceleration of >2) for those with larger social network compared to smaller: The ARIC Study (1990–1992)

Size of Social network	Small	Moderate	Large	P for trend
Social network score	≤25	26–30	≥31
N total	1216	1607	8896
Total
N (age acceleration >2)	288	347	1692
Model 1	Reference	0.88 (0.74–1.06)	0.75 (0.65–0.87)	<0.0001
Model 2	Reference	0.93 (0.78–1.12)	0.83 (0.72–0.96)	0.006
Model 3	Reference	0.99 (0.82–1.19)	0.88 (0.76–1.03)	0.04

Model 1: age, sex, and race-center

Model 2: Model 1 covariates and additionally adjusted for education and marital status

Model 3: Model 2 covariates and additionally adjusted for smoking status, alcohol use, exercise, sleep quality, and body mass index

Social support

Table 3 shows the age-adjusted characteristics of participants according to the levels of social support. The numbers of participants who had low, moderate, and high perceived social support were 923 (7.9%), 1,203 (10.3%), and 9,593 (81.9%), respectively. Those who perceived higher levels of social support tended more likely to be self-identify as White, have higher income, more education, were more likely to be employed and married, to be a current drinker, a nonsmoker, and to have less trouble sleeping.

Table 3:

Means and proportions of participant characteristics according to the levels of social support score: The ARIC Study (1990–1992)

Levels of social support	Low	Moderate	High
Social support score	≤23	24–31	≥32
N total	923	1203	9593
Age and ARIC PAC
Age, mean years ± SD	57.1±5.90	57.1±5.69	57.1±5.72
ARICPAC, years ± SD	57.9±4.86	57.8±4.77	57.6±4.67
Other demographics
Male, %	43.3	47.1	44.1
Black, %	34.7	25.4	22.0
Income, %
<$25,000	51.8	39.5	31.3
$25,000–$49,999	28.8	34.6	37.3
≥$50,000	13.7	20.1	26.1
Education, %
Less than high school	39.0	27.3	19.3
High school graduate	39.1	45.2	41.9
Beyond high school	21.6	27.4	38.7
Employed, %	61.6	68.4	70.5
Married, %	66.1	72.4	77.3
Behavioral characteristics
Physical activity*	2.22±0.73	2.33±0.78	2.49±0.80
Current drinker, %	48.8	54.9	57.7
Current smoker, %	31.3	28.2	20.7
Trouble falling asleep, %	37.5	29.7	19.8
Wake up repeatedly during the night, %	52.2	48.0	36.5
Physiologic characteristics
BMI, kg/m2 ± SD	28.4±0.18	28.3±5.72	27.8±5.23
eGFR, ml/min/1.73m2 ± SD	95.5±0.46	96.1±15.2	96.8±13.7
ARIC PAC of development subset
ARICPAC, years ± SD (training set)	182	284	2520
	56.3±4.6	56.0±4.4	56.0±4.4
ARICPAC, years ± SD (without training set)	741	919	7073
	58.3±4.8	58.3±4.7	58.1±4.6

Abbreviations: PAC - proteomic aging clock; BMI - body mass index; eGFR - estimated glomerular filtration rate

^*Physical activity was assessed using a leisure-time sports index that ranged from 1 to 5

Figure 2 and Table S2 show the average age accelerations calculated by ARIC PAC according to the levels of social support, overall, without the training set, and stratified by sex, race, age category and marital status. Age acceleration inversely associated across categories of social support, with an age, sex, and race-center adjusted p for trend of <0.0001. The average age accelerations for persons with low, moderate, and high social support were 0.33, 0.19, and −0.06, respectively. This association remained significant after adjusting for social factors (p for trend was 0.01 for Model 2), but was not significant after additionally adjusting for lifestyle factors and BMI (p for trend was 0.28 for Model 3). Results were also null in a model that additionally adjusted for eGFR (p for trend 0.39). The trend of the average age acceleration, calculated by ARIC PAC according to the levels of the social support, was similar in ‘all participants minus the training set’ to those with ‘all participants’. A significant interaction was observed whereby the association was stronger in women than in men (p-interaction=0.03). Though the interactions were not statistically significant, the associations appeared somewhat more evident for participants self-identifying as White, and those aged <60 years.

Figure 2. Marginal means age accelerations calculated by ARIC PAC according to the levels of social support: The ARIC Study (1990–1992).

The y-axis shows the number of participants by level of social support within each subgroup. The results present age acceleration values from Model 1, which was adjusted for age, sex, and race-center. In the figure, solid lines indicate p-values for trend, while dotted lines indicate p-values for interaction.

Table 4 shows the odds ratios and 95% confidence intervals of having an older aging clock (age acceleration of >2 years) for those with higher social support compared to low. Those with high social support had a lower odds ratio of having a 2-year biologically older age compared to those with low support. The age, sex, and race-center adjusted OR and 95% CI was 0.77 (0.66–0.91) for those with high social support versus low social support, and p for trend across the social support categories was 0.0005. This association was not significant after further adjustment.

Table 4:

Odds ratios and 95% confidence intervals of having older aging clock (age acceleration >2) for those with larger perceived social support compared to smaller: The ARIC Study (1990–1992)

Social support levels	Low	Moderate	High	P for trend
Social support score	≤23	24–31	≥32
N total	923	1203	9593
Total
N (age acceleration >2)	218	258	1851
Model 1	Reference	0.88 (0.72–1.09)	0.77 (0.66–0.91)	0.0005
Model 2	Reference	0.96 (0.78–1.18)	0.90 (0.77–1.06)	0.18
Model 3	Reference	1.03 (0.83–1.28)	0.99 (0.83–1.18)	0.78

Model 1: age, sex, and race-center

Model 2: Model 1 covariates and additionally adjusted for education and marital status

Model 3: Model 2 covariates and additionally adjusted for smoking status, alcohol use, exercise, sleep quality, and body mass index

Secondary analyses using Sathyan’s PAC

Social network

Table S3 shows the average age accelerations calculated by Sathyan’s PAC according to the size of social network. The larger the size of the social network, the smaller the age acceleration. The age, sex, and race-center adjusted p for trend was 0.0001, and the adjusted average age accelerations for those with a small, moderate, and large social network were 0.24, 0.10, and −0.05, respectively. This association remained significant after adjusting for social factors in model 2 (p for trend = 0.002), but was not significant after adjusting for lifestyle factors and BMI in model 3 (p for trend = 0.12).

Social support

Table S4 shows the average age accelerations calculated by Sathyan’s PAC according to the levels of social support. The higher the levels of social support, the smaller the age acceleration. The age, sex, and race-center adjusted p for trend was <0.0001, and the average age accelerations for persons with low, moderate, and high social support were 0.29, 0.13 and −0.03, respectively. This association remained significant after adjusting for social factors in model 2 (p for trend = 0.002), but was not significant after adjusting for lifestyle factors and BMI (p for trend = 0.10).

Discussion

Larger social networks and higher levels of perceived social support were associated with less biological age acceleration and lower odds of having an older aging clock (defined as 2 years older than chronological age) in this population-based cross-sectional study of 11,719 men and women aged 46–70 years. These associations were weakened after adjustment for other social and lifestyle factors which may mediate the associations of social network size and perceived social support with biological aging. That lifestyle may mediate the associations is supported by prior studies, which have reported that having a large social network or high support is associated with having a healthier lifestyle^25,27,47,48.

The cross-sectional design of this study limits causal interpretations. However, these findings suggest that biological aging is inversely associated with social network size and perceived level of social support, and provide evidence of potential physiological underpinnings for associations between social interactions and healthy aging. Replication is needed in independent populations and with prospective designs to better understand the role of social networks and social support in biologic aging.

Findings in Context

Few reports have examined the relationship between social networks or support and biological age. A recent analysis of 3,647 participants of the Health and Retirement Study, aged 50 to 100 years, investigated the associations of social support and strain with biological age ascertained by epigenetics.²⁹ Participants with greater support had slower Hannum age acceleration (b = −0.380, SE = 0.184, p = .04)²⁹. Additionally, it has been reported that loneliness and isolation in the community due to contact with the police in childhood was associated with slightly accelerated epigenetic age in adulthood²⁸. The present study adds to these findings, and provides new insights as we assessed biological age using the a PAC. Although we did not directly compare findings to those from epigenetic clocks, more pronounced associations were observed with the PAC as compared to prior studies in different populations. One potential reason why PACs may show more pronounced associations is that proteins can more directly capture molecular phenotypes compared to epigenetic aging clocks^14,37 and thus may allow for more accurate estimation of an individual’s biological age. Overall, our findings were similar by sex, race, age category and marital status. The exception was that we observed a sex interaction, whereby the inverse association of social support with biological age acceleration was stronger in women than in men. Prior work has suggested that the effects of social networks and support on health tend to be stronger among women compared to men. For example, in a longitudinal study of 6,928 adults in California, the age-adjusted relative risk of mortality for the most isolated compared to those with the most social contacts was 2.8 for women compared to 2.3 for men⁴⁹. Also, in a small study of 292 participants in the Connecticut Epidemiology of the Elderly Study, lack of emotional support was significantly associated with cardiovascular outcomes in patients hospitalized for heart failure, with an odds ratio of 8.2 (95% CI, 2.5–27.2) for women compared to 1.0 (95% CI, 0.3–3.3) for men⁵⁰. Further investigation is needed to understand how demographic and social factors influence biological age acceleration.

Mechanisms

These findings provide additional context for how social network and support may be linked to better health outcomes. The mechanisms underlying the association between social networks and support and slower biological age acceleration are not fully elucidated, but likely include both behavioral and physiological factors. It has been reported that individuals with larger social networks are more likely to participate in more health-promoting behaviors (e.g., eating a healthier diet, exercising, quitting smoking)^{25,27,47,48,51} and are less susceptible to psychological stress.¹⁵ Both participation in health-promoting and psychological stress or resilience may influence biological age acceleration through a variety of mental and physical changes. Activation of the hypothalamus-pituitary-adrenal (HPA) axis is an adaptive response to stress, but persistent elevations in glucocorticoids may impair the neuroimmune system and neuronal survival^52,53, and prolonged stress and HPA activation may affect age acceleration⁵⁴. In the present analysis, after adjusting for lifestyle factors and BMI the associations of social networks and social support with biological aging weakened and became nonsignificant. Previous studies have also suggested that older adults with less social support have shorter telomeres,⁵⁵ which is considered a hallmark of aging. Additionally, individuals with smaller social network scores are more likely to have elevated circulating levels of hsCRP and interleukin-6, even after adjustment for confounders including other social and economic factors⁵⁴. These findings suggest that social interactions may also influence biological age acceleration via the inflammatory system.

Strengths and Limitations

A strength of this study is that it is based on a large community-based sample of Black and White individuals. Previous studies of social interactions and biological age acceleration have typically had small sample sizes or focused primarily on White individuals^9,12,13. The PAC we employed is also novel in that it was developed and validated using approximately 5,000 proteins measured through the SomaLogic platform. In addition, the PAC has been validated in MESA, a large independent cohort, and the reliability of the PAC is high. We were also able to adjust for a broader range of confounding factors, such as lifestyle and social factors, whereas previous studies on social interactions and PACs have typically adjusted only for demographic factors^8,13. In addition to the ARIC PAC, we conducted a secondary analysis using Sathyan’s PAC¹³ and obtained similar results. Furthermore, similar results were obtained in supplemental analyses excluding participants from the training set used to create the ARIC PAC. These sensitivity analyses support the robustness of our findings.

Our study has several limitations. First, the design is cross-sectional, and therefore temporality cannot be established and causality is unclear. It is possible that individuals with lower biological age acceleration (e.g., likely healthier on average) are better able to maintain their social networks. Second, although we adjusted for potential confounders, residual or unmeasured confounders may have influenced the relationship between social networks and support and biological age acceleration through other pathways. Third, the social network and social support scales were self-reported and these questionnaires were administered at a single point in time, which could have resulted in measurement error and misclassification. Fourth, the possibility of protein degradation during long-term storage cannot be excluded. However, since the blood samples were frozen immediately after collection and never thawed, the possibility of degradation is low. Furthermore, repeatability of measurements in blind duplicates was high (Bland-Altman coefficient of variation of split samples = 6% at Visit 2)³⁷. Lastly, the aging clock was created in a population aged 46–70 years, therefore results may not be generalizable outside this age-range.

Conclusion

In conclusion, we found that having a larger social network and a higher perceived level of social support were associated with less biological age acceleration in a community-based sample of Black and White individuals at midlife. Social engagement, in addition to risk factor control, may be beneficial to overall health.

Data Availability Statement:

The ARIC and MESA datasets are available through BioLINCC, with appropriate study approvals from the ARIC and MESA studies. Data request through BioLINCC can be accessed at https://biolincc.nhlbi.nih.gov/studies/aric/ and https://biolincc.nhlbi.nih.gov/studies/mesa/.