The reciprocal associations between social deficits, social engagement, and inflammation: Longitudinal evidence comparing venous blood samples and dried blood spots and mapping the modifying role of phenotypic and genotypic depression

Background: Social psychoneuroimmunology suggests an interplay between social deficits (loneliness and isolation) and chronic inflammation, but the direction of these relationships remains unclear. We estimated the reciprocal associations of social deficits and social engagement with levels of C-reactive protein (CRP), compared the consistency of the findings depending on the biological sampling method used, and examined the modifying role of phenotypic and genotypic depression. Methods: We used longitudinal nationally representative data from the US (Health and Retirement Study, 3 waves, 2006 – 16) and England (English Longitudinal Study of Ageing, 4 waves, 2004 – 18). Loneliness, social isolation, and social engagement were self-reported. CRP was measured using dried blood spots (US) and venous blood samples (England). Cross-lagged panel models were fitted and tested interactions with phenotypic depression (above-threshold depressive symptom scores) and genotypic depression (polygenic score for major depressive disorder). Results: We included 15,066 participants (mean age = 66.1 years, SD = 9.8) in the US and 10,290 (66.9 years, SD = 10.5) in England. We found reciprocal associations between loneliness and CRP using dried blood spots and venous blood samples. Higher CRP predicted higher subsequent loneliness and higher loneliness predicted elevated CRP. Both phenotypic and genotypic depression modified this reciprocal association. There were also reciprocal associations for social engagement in venous blood samples: higher CRP predicted lower social engagement and greater social engagement predicted lower subsequent CRP. Associations between social isolation and CRP were inconsistent and unidirectional. Conclusions: Loneliness may increase chronic inflammation, whereas social engagement may reduce inflammation. As these relationships were reciprocal, there may be a loop between inflammation, loneliness, and social engagement. This loop was stronger in those with depression or at high genetic risk for major depressive disorder. This relationship for loneliness was present in both blood sampling methods despite contrasting methods of CRP measurement, indicating that the finding is not attributable to measurement bias in biomarkers.


Introduction
Social deficits (loneliness and isolation) are a public health threat, with approximately 25 % of people aged 65 years and above reporting feeling lonely or socially isolated in the US and UK (National Academies of Sciences Engineering and Medicine, 2020).Loneliness is a subjective negative experience caused by a perceived disconnect in the desired and actual quantity and quality of social connections (Hawkley and Cacioppo, 2010), whereas social isolation is an objective measure of infrequent contact with others (National Academies of Sciences Engineering and Medicine, 2020).These social deficits have adverse effects on mental and physical health, including stroke, heart disease (Valtorta et al., 2016), depression (Park et al., 2020), cognitive decline, dementia, and premature mortality (Shor and Roelfs, 2015;Gao et al., 2021).In contrast, social engagement refers to active participation in community or society activities (e.g.leisure activities, volunteering, and social groups) and can benefit cognitive function (Keefer et al., 2023) and reduce risk of cardiovascular disease (Han et al., 2016), depression (Cruwys et al., 2013;Bone et al., 2022), and premature mortality (Fancourt and Steptoe, 2019;Løkken et al., 2021).
A wide variety of social, psychological and biological pathways link social deficits and engagement to health and disease, with inflammatory processes being one such pathway that has received increasing attention in the past few years (Eisenberger et al., 2017).On the one hand, inflammation can influence social deficits and engagement.As an acutephase protein, C-reactive protein (CRP) is a biomarker of systemic inflammation and acute infection.Chronically elevated CRP may increase risk of cardiovascular disease (Kaptoge et al., 2010), depression (Osimo et al., 2020) and functional decline (Reuben et al., 2002), leading to increased all-cause mortality (Reuben et al., 2002;Harris et al., 1999).These mental and physical health problems could create psychological, physical and behavioural barriers to social engagement, as well as increasing loneliness and isolation.However, in the reverse direction, social deficits and engagement could also modulate inflammatory processes (Kornienko et al., 2022).Bidirectional brain-immune system pathways provide a foundation for the interplay between social experiences and inflammation (Bower and Kuhlman, 2023).Chronic loneliness has been proposed to upregulate gene transcription of proinflammatory control pathways (Cole et al., 2007), while social engagement may downregulate biological pathways for inflammatory processes (Uchino et al., 2018).Thus social behaviours could potentially act as catalysts for increased inflammation (which in turn may then reinforce patterns of social behaviour that are adverse for health).
There is some population-based evidence for the role of social factors in inflammation.A systematic review found associations between social deficits and elevated inflammatory markers (Smith et al., 2020).However, evidence was inconsistent, with social isolation linked to higher CRP and fibrinogen, and loneliness only associated with increased interleukin-6 (Smith et al., 2020).A meta-analysis of mainly crosssectional studies found that greater social integration and engagement were associated with lower CRP (Uchino et al., 2018).There are similar patterns in longitudinal research (Walker et al., 2019;Bell et al., 2022).This relationship could be bidirectional, with inflammation also influencing subsequent social deficits and engagement (Eisenberger et al., 2010).But longitudinal evidence in this direction is inconsistent (Loeffler and Steptoe, 2021).To date, the only population-based study simultaneously assessing both pathways found no evidence for a bidirectional association between volunteering and CRP (Kim and Volunteering, 2020).
An important factor to disentangle in the social behaviourinflammation relationship is the role of mental health.Depression is closely linked to social deficits and engagement, as well as inflammatory processes.Social deficits may be a driver or a consequence of depression (Heinrich and Gullone, 2006), while social engagement can help to reduce or prevent depressive symptoms (Cruwys et al., 2013).Substantial literature demonstrates associations between inflammation and depression (Osimo et al., 2020), and pro-inflammatory cytokines have been linked to the development of depressed mood (Miller et al., 2009;Wright et al., 2005).Given that inflammation is thought to have a potential causal role in depression, the relationship between social deficits, engagement and chronic inflammation may differ according to genetic propensity to depression or to current depressive symptoms (Taylor et al., 2022).Yet, no research to date has assessed the modifying role of depression.
Additionally, in population research on social behaviours and inflammation, methodological issues of biological measurement approaches are important to unpack.To date, analysis of CRP in relation to social deficits has previously largely been conducted using venous blood sampling (VBS).Recently, dried blood sampling (DBS) has become increasingly common as an alternative to venipuncture in large-scale surveys as it is low-cost, less painful and invasive, and can be carried out at home.Studies comparing DBS to VBS have shown high correlations (Crimmins et al., 2014;Brindle et al., 2010;McDade et al., 2004).However, whether social-biological relationships can be consistently elucidated using both DBS and VBS has not been ascertained.
Thus, there are several gaps remaining in understanding the relationships between social deficits and chronic inflammation.First, most population-based research has a cross-sectional design, which cannot ascertain the direction of effects.Second, limited longitudinal investigations have generally examined a unidirectional relationship (Walker et al., 2019).Third, no research has assessed the modifying roles of phenotypic and genotypic depression.Finally, whether the method of blood sampling affects findings is unknown.Therefore, in nationally representative samples of US older adults (using DBS) and English older adults (using VBS), this study aimed to simultaneously test whether i) social deficits and engagement are associated with subsequent inflammation and ii) inflammation is associated with subsequent social deficits and engagement.We also explored whether phenotypic depression (operationalised as above-threshold depressive symptoms on a validated scale) and genetic propensity for depression modify these associations.

Sample
This study used nationally representative panel data from the US Health and Retirement Study (HRS, 2006-16) and the English Longitudinal Study of Ageing (ELSA, 2004-18).In HRS, social deficits and engagement were measured in the Leave-Behind Psychosocial and Lifestyle Questionnaire (HRS-LBQ).A random 50 % subsample of HRS participants was eligible for the HRS-LBQ in 2006, with the remaining 50 % of HRS participants then eligible for the HRS-LBQ in 2008.The HRS-LBQ was completed in alternate waves (i.e.every four years), with subsample A participating in 2006, 2010, and 2014 and subsample B participating in 2008, 2012, and 2016.Biomarker data was collected at the same timepoints, meaning HRS participants had a maximum of three waves (T1-T3).In ELSA, blood samples were collected during nurse visits, with follow-up every four years in three waves (2004/5, 2008/9, 2012/13).The fourth ELSA wave included in this study was completed either four or six years later (2016/17 or 2018/19), based on random selection.Social factors were measured at the same timepoints.ELSA participants thus completed a maximum of four waves (T1-T4).
We limited both samples to participants who provided biological data in at least one wave and had complete data on social deficits and engagement at the same wave.Our analyses used polygenic risk scores (PGSs) and controlled for principal components (PCs).The construction of PGSs was largely derived from genome-wide association studies (GWAS) on European ancestry groups, which may have different predictive capacities for other ancestry groups (Martin et al., 2017;Ware et al., 2017).PCs are only available for European ancestry populations in ELSA.We could only include those who identified as being of White ethnicity.This provided a final analytical sample of 15,066 participants in HRS and 10,290 participants in ELSA (Supplementary Fig. 1).Informed consent was acquired from all participants.This analysis received ethical approval from the University of Florida Institutional Review Board (IRB201901792) and the University College London Research Ethics Committee (18839/001).

Inflammation
Chronic inflammation was measured using CRP.In HRS, CRP was measured using dried blood spots at all included waves.CRP was Q. Gao et al. quantified by particle-enhanced immunonephelometric assays with a monoclonal antibody to CRP using the BNII nephelometer.We used the National Health and Nutrition Examination Survey (NHANES) equivalent CRP assay values in HRS, which matched values to the average CRP values in NHANES and retained the variability of original HRS samples (Crimmins et al., 2013).This correction transformed HRS dried blood spot (DBS) biomarkers into venous blood-based assays, making values more comparable to whole blood-based assays collected in other cohorts, including ELSA (Kim et al., 2006).In ELSA, CRP was measured using blood samples collected during nurse visits.CRP was quantified using the N Latex CRP mono immunoassay on the Behring Nephelometer II analyser.Detailed descriptions of biomarker data in HRS (Crimmins et al., 2013) and blood sample analyses in ELSA (Graig et al., 2004) are available online.We excluded participants with a CRP level over 10 mg/ L due to the possibility of acute infection or severe disease onset.Given the skewed distribution of CRP, values were log-transformed in analyses.

Social deficits and engagement
Loneliness was measured with the validated 3-item UCLA loneliness scale, which is comparable to the original 20-item version (Hughes et al., 2004).This measured whether participants felt they lacked companionship, felt left out, or felt isolated from others.Response options (1) hardly ever or never, (2) some of the time, or (3) often were summed arithmetically.Total scores ranged from 3 to 9, with higher scores indicating higher levels of loneliness.Loneliness was analysed continuously and using a cut-off of ≥ 6 to indicate loneliness (Steptoe et al., 2013).
Social isolation was defined using the social isolation index, measuring the following five domains: i) unmarried/not cohabiting (including living alone), ii) non-participation in any groups, clubs, or other social organisations, and having less than monthly contact (including face-to-face, telephone, or written/e-mail contact) with iii) children, iv) other family members, v) and friends.Total scores ranged from 0 to 5, with higher scores indicating higher social isolation.This approach has been validated in ELSA (Steptoe et al., 2013) and used in HRS (Crowe et al., 2021).We analysed scores continuously and used a cut-off of ≥ 3 to indicate social isolation (Crowe et al., 2021).
Social engagement was defined as involvement with leisure activities.We used items that were consistent across all waves of HRS and ELSA, assessing participation in: i) volunteering, ii) educational or training courses, iii) charity work or associations, iv) social or sports activities, v) non-religious community activities, and vi) hobbies.Contact with social network was not included, as this was captured in the social isolation measure.We used binary indicators of any engagement in each activity and then summed these to produce an overall score (range: 0-6).Higher scores indicate greater social engagement.We also created a binary indicator of engagement in any social activities (yes/no).

Effect modifiers
Measures of genotypic depression were harmonised across cohorts.In HRS, the polygenic scores for major depressive disorder (MDD-PGSs) were created using a mega-analysis of GWAS of European ancestry conducted by the Psychiatric Genomics Consortium.MDD-PGSs in ELSA were derived using the same methods as HRS to allow for harmonisation (Ajnakina and Steptoe, 2019;Ware et al., 2020).In each cohort, we zstandardised the MDD-PGSs (mean = 0, standard deviation [SD] = 1) and divided them into deciles, taking the top 10 % of the PGS as the 'high genetic propensity' group and combining the remaining deciles into 'low and average genetic propensity' (Ajnakina and Steptoe, 2019).Phenotypic depression was measured using the eight-item Center for Epidemiologic Studies Depression Scale (CES-D) in both datasets, which is comparable to the well-validated 20-item CES-D scale (Karim et al., 2015;Turvey et al., 1999).The overall scores range from 0 to 8, with a cut-off of ≥ 3 used to indicate depression (Karim et al., 2015).

Covariates
We selected covariates for the study using Directed Acyclic Graphs (DAGs) (Tennant et al., 2021).A range of covariates was measured at baseline (T1) in HRS and ELSA, including demographic and socioeconomic factors: age (years), gender, educational attainment (no or basic qualification [less than high school], lower-middle [high school, GCSE or Olevel or qualification at age 16], upper-middle [college or undergraduate degree, A-levels or higher education at age 18], higher qualification [degree or further higher qualification]), household income (quintiles), and employment status (employed/unemployed or retired).Health behaviours were: regular physical activity (at least weekly engagement in mild, moderate or vigorous sports or activities; yes/no), smoking status (smoker, non-smoker), and alcohol consumption (<1, 1-2, 3-4, ≥5 times per week).Health-related factors were: baseline body mass index (BMI), chronic illness (none, one or more), and persistent moderate or severe pain (yes/no).In HRS and ELSA, 10 PCs for European ancestry groups were available to control for population stratification, in order to reduce the risk of bias resulting from ancestral differences in genetic analyses (Price et al., 2006;Hellwege et al., 2017).

Statistical analyses
First, we used cross-lagged panel models to test the reciprocal relationships between social deficits and engagement and inflammation.The cross-lagged pathways tested the longitudinal relationships by exploring whether social factors (i.e.loneliness, social isolation and engagement at T1-T2/3) were associated with CRP at subsequent waves (T2-T3/4); and simultaneously tested whether CRP (T1-T2/3) was associated with subsequent social factors (T2-T3/4).Models accounted for the shared variance at each measurement occasion (occasion effects), the autoregression of individual differences over time and the correlations of social factors and inflammation for all waves.Standardised regression coefficients (β) were estimated.We adjusted for all covariates measured at T1.Each social factor (loneliness, isolation, engagement) was included in a separate model, and models were run separately across HRS and ELSA, giving a total of six models.These models were estimated using maximum likelihood robust (MLR) estimation.
Second, we tested whether there is evidence that the cross-lagged associations between social factors and CRP differ according to depression, measured both phenotypically (meeting criteria for depression on the CES-D) and genotypically (high polygenic propensity for MDD).First, we included two-way interaction terms between phenotypic depression and social factors, as well as phenotypic depression and CRP in each wave.Second, we repeated this for genotypic depression.The genotypic model was additionally adjusted for the 10 PCs for European ancestry individuals.Each social factor (loneliness, isolation, engagement) was modelled separately with the Bayes estimator.
We used model fit indices for all analyses, including the Chi-square goodness-of-fit, the Tucker-Lewis Index (TLI), the Comparative Fit Index (CFI), the root mean square error of approximation (RMSEA), and the standardised root mean square residual (SRMR) (Hooper et al., 2008).Full information maximum likelihood was applied to address missingness in modelling.All analyses were based on weighted data, using the sample weights for the biomarker sub-sample in HRS and the blood weights for ELSA.In sensitivity analyses, we repeated all analyses using binary measures of each outcome.All statistical analyses were performed using STATA 16.0 (StatCorp LP, Texas, USA) and Mplus 8.0 (Los Angeles, CA: Muthén & Muthén).
There was evidence for an interaction between phenotypic depression and CRP on subsequent loneliness in both the DBS sample (β CRP*dep T1 = 0.05 [95 % CI 0.02-0.09],β CRP*dep T2 = 0.08 [0.03-0.12])and VBS sample (β CRP*dep T1 = 0.07 [0.02-0.11],β CRP*dep T2 = 0.08 [0.05-0.12];Supplementary Table 4).Interaction terms indicate the association between CRP and subsequent loneliness was stronger in those with depression.However, there was no evidence that the reciprocal association between loneliness and subsequent CRP differed according to phenotypic depression.There was also evidence for an interaction between genotypic depression and CRP on subsequent loneliness in the DBS sample (β CRP T1*PGS = 0.04 [0.01-0.07])(Supplementary Table 5).In the VBS sample, this interaction was in the opposite direction, with evidence for an interaction between genotypic depression and loneliness on subsequent CRP (β lon T1*PGS = 0.05 [0.01-0.08]).There was a stronger association between CRP and subsequent loneliness in those at higher risk of MDD.

Social isolation
There was very little evidence for concurrent associations between social isolation and CRP (Fig. 2, Supplementary Table 6).There was also limited evidence for longitudinal associations.In the VBS sample, there was no evidence that CRP predicted subsequent isolation.Although isolation at T2 predicted CRP at T3 (β iso T2 = 0.05, p = 0.01), this lagged association was not present at other timepoints.In the DBS sample, CRP at T2 predicted isolation at T3 (β CRP T2 = 0.04, p = 0.04), but this was not present at earlier timepoints, and there was no evidence that isolation predicted subsequent CRP.All models had acceptable fit (CFI 0.89 to 0.92; TLI 0.83 to 0.86; RMSEA 0.04 to 0.05; SRMR 0.05 to 0.06).Sensitivity analyses did not produce any significant findings (Supplementary Table 6).There was no consistent evidence for interactions between social isolation and depression (Supplementary Tables 7-8).
There was evidence for an interaction between social engagement and phenotypic depression on later CRP in the DBS sample (β SE*dep T2 = 0.05 [0.01-0.09]),but not in the VBS sample (Supplementary Table 10).The negative association between social engagement and CRP was weaker in those with depression.For genotypic depression, there was only evidence in the same direction in the VBS sample, with an interaction between social engagement and genotypic depression on subsequent CRP (β SE T1*PGS = 0.05 [0.01-0.09])(Supplementary Table 11).

Discussion
This was the first study to estimate the longitudinal reciprocal associations between social deficits, social engagement, and inflammatory processes and to simultaneously unpack the moderating role of depression and the stability of findings across different biological measurement approaches.Longitudinally, there was a reciprocal association of loneliness and social engagement with CRP.Net of sociodemographic and health-related confounders, higher CRP predicted greater loneliness and lower social engagement.In turn, lower social engagement and greater loneliness predicted higher CRP.But there was no clear relationship between social isolation and CRP.Findings for loneliness and social isolation were fairly consistent across two different biological measurement approaches (VBS vs DBS), although there were discrepancies in social engagement findings.Some of the associations between loneliness, social engagement, and CRP also differed according to phenotypic and genotypic depression.

Loneliness
The bidirectional relationship between loneliness and CRP builds on previous observational findings that loneliness is associated with higher CRP (Uchino et al., 2018;Smith et al., 2020;Walker et al., 2019).Lonely individuals may be at higher risk of chronic disease due to systemic inflammation (as indicated by consistently raised CRP).In turn, heightened CRP may increase loneliness by increasing risk for chronic disease, in turn limiting mobility, social contact, and social network size (Gao et al., 2021).Replicating this association across both DBS and VBS indicates that this finding is not attributable to specific attributes of the biological measurement.It also lends weight to work incorporating DBS into cohort studies as pragmatic approaches to measuring biological data.While previous studies have demonstrated the high correlations between DBS and VBS (Crimmins et al., 2014;Brindle et al., 2010;McDade et al., 2004), our research extends this by highlighting the consistency in findings when utilising the two different approaches for complex social-biological analyses.

Social engagement
In the English VBS sample, we also found evidence for a reciprocal association between social engagement and inflammation, independent of confounders.A previous systematic review found that social engagement and integration were associated with reduced inflammatory responses (Uchino et al., 2018).Longitudinal studies have found that social engagement (e.g.volunteering) was associated with lower subsequent CRP (Walker et al., 2019;Bell et al., 2022).The positive loop of high social engagement and low inflammation is consistent with evidence showing the health benefits of a range of social activities, especially in reducing the risks of chronic illness (Keefer et al., 2023;Han et al., 2016;Cruwys et al., 2013;Bone et al., 2022).However, this reciprocal association was not present in the US DBS sample.Considering the results in the context of the biological sampling, the DBS may have been slightly less sensitive.Indeed, there was some tentative evidence of DBS results moving towards significance in some of the directions (e.g.T2 social engagement to T3 CRP).We could also consider the findings in terms of the two different cultural settings (England vs the US).Variations in cultural norms of social embeddedness may exist, which could partially explain the differences in the findings.Levels of social engagement were similar across the two countries, but the samples are from different countries with demographic and behavioural variations.Further, the effects of social engagement in modulating immune responses could depend on the magnitude and types of social activity engaged in, as well as country-level social, economic, and cultural factors, including the Gini index (measuring income inequality in a population) and life expectancy (Mak et al., 2023).

Social isolation
Surprisingly, there was no evidence for bidirectional associations between social isolation and CRP, or even longitudinal associations in either direction.Neither biological measurement approach provided substantial evidence of a relationship.Social isolation is a more objective measure of social deficits, particularly compared to subjective loneliness.A recent meta-analysis found only concurrent associations between social integration and lower CRP (Uchino et al., 2018).It is possible that subjective perceptions of social connection are more strongly associated with inflammation than objective measures.This hypothesis requires further investigation, particularly as it contradicts previous evidence for a relationship between social isolation and adverse health outcomes, including declines in functional ability and long-term conditions (1).Further research should compare the reciprocal associations between subjective and objective measures of health, social factors, and inflammation to confirm our findings, especially simultaneously including both loneliness and isolation measures.

Moderation by phenotypic and genotypic depression
We found evidence that the association between loneliness, social engagement, and inflammation might differ according to depression.For loneliness, the association between CRP and subsequent loneliness was stronger in those with phenotypic depression and those at high genetic risk for MDD.These novel findings extend previous evidence that both social deficits and inflammation can increase the risk of depression (Heinrich and Gullone, 2006;Miller et al., 2009;Wright et al., 2005), indicating that high levels of depressive symptoms potentially indicative of depression and genetic risk for depression may both interact with inflammation, leading to stronger reciprocal associations between loneliness and inflammation.Individuals with higher phenotypic or genotypic propensity for depression may be at greater risk of the self-perpetuating negative loop between loneliness and inflammation, leading to more adverse outcomes in the long term (Kaptoge et al., 2010;Osimo et al., 2020;Reuben et al., 2002).There was also some evidence of the modifying role of phenotypic depression, but not genotypic propensity, in the healthy loop of social engagement and inflammation.The beneficial cycle of social engagement and subsequent CRP may be weaker in those with depression.It is possible that individuals with depression have higher chronic inflammation (Miller et al., 2009;Wright et al., 2005), which is more difficult to alter via

Strengths and limitations
A key strength of this study is the cross-lagged panel approach, allowing the examination of the reciprocal relationships between social deficits, engagement, and inflammation.We utilised data from England and the US, involving two different measurement approaches for CRP and also allowing us to compare these relationships in two differing contexts, strengthening causal inferences.Nationally representative longitudinal data provided rich information on social deficits and social engagement over time, including both subjective and objective measures of social deficits.Additionally, we adjusted for a wide range of potential sociodemographic, behavioural, and health-related confounders identified in previous research, limiting the potential for residual confounding.
However, our study had several limitations.First, although considered an inflammatory marker, CRP has diverse additional roles in somatic maintenance (Del Giudice and Gangestad, 2018).Thus our findings indicate a relationship between social factors and CRP, but further research with alternative inflammatory biomarkers is required to confirm if this does indeed support an overall heightened inflammatory profile.Second, we tested the modifying role of phenotypic and genotypic depression separately.Genotypic risk for depression (PGS-MDD) is fixed at conception, providing a genetic predisposition to depression.In contrast, the depression phenotype is shaped by both genetic and environmental factors, with depressive symptoms and PGS-MDD strongly related.We therefore cannot rule out the effect of genotype in the modifying role of phenotypic depression.Further, our phenotypic depression was measured with a validated depression scale, but we did not explore the effects of sub-clinical effects of depression or alternative psychiatric diagnoses on findings.We could only include individuals of White ancestry due to the way PGSs were calculated.Our findings may not be generalisable to other ethnic groups and should be replicated in more diverse samples with PGSs calculated in more equitable ways.The findings may also be specific to the cohorts, and future research would need to test the social-biological relationships across both DBS and VBS within the same individuals.Additionally, we limited our sample to participants with complete biomarker data in at least one wave, who may have been healthier than the overall HRS and ELSA samples.However, we used weights to minimise the impact of attrition and sampling biases.Social factors were self-reported and so may be subject to recall bias.Although we used well-established validated measures of loneliness and social isolation, their cultural specificity remains unclear.We were unable to identify the extent to which differences in findings between the US and England were due to the social measures used.We could not measure social engagement frequency, or explore different types of social engagement separately, so this remains a priority for future research.Future research is encouraged that could consider how findings are moderated depending on life transitions that disrupt social networks (such as retirement).Our work highlights the importance of analyses using multiple cohorts from different countries, confirming whether underlying social-inflammatory processes are consistent and identifying individual or societal moderators responsible for any differences.

Conclusions
In summary, we found evidence for reciprocal relationships between loneliness, social engagement, and inflammation, but not social isolation.There may be an adverse loop whereby higher loneliness leads to elevated inflammatory load, which in turn increases loneliness.However, there may also be a beneficial loop whereby more social engagement leads to less inflammation, which could further support social engagement.These associations between social deficits and engagement and inflammation remain independent of depression, but the loop between loneliness and inflammation was potentially stronger in older adults with high levels of depressive symptoms above the level that can indicate clinical depression and those with higher genetic risk of depression.Our findings illustrate one mechanism through which loneliness and social engagement may influence health (via chronic inflammation), and on the flip side mechanisms through which inflammation may influence health (via loneliness and social engagement), particularly for people at increased risk of depression.Promisingly, results were relatively consistent across two different biological measurement approaches, although given some variations in findings, further research is encouraged to confirm whether measurement or demographic and cultural differences between samples explain differences.Loneliness and social engagement could thus provide modifiable targets for mitigating chronic inflammation, reducing long-term risk of chronic health conditions.

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.

Table 1
Participant characteristics at baseline.