Perceived Relationship Support Moderates the Association of Contextual Stress with Inflammation Among African Americans

Abstract

We followed 402 African American young adults from ages 24 to 29, a period of emerging committed relationships, to examine the association of contextual stress (CS), e.g., experiences of financial strain, victimization, and racial discrimination, with inflammation, and to test predictions that greater relationship warmth and support (PRWS) at age 29 would moderate the association between earlier CS and inflammation, using a multiplex assessment of cytokines to construct an index of the ratio between predominantly pro-inflammatory cytokines vs predominantly anti-inflammatory cytokines. CS experienced at age 24 was associated with greater inflammation at age 29 in the full sample (b = .112, p = .004). PRWS at age 29 moderated the association of earlier CS with inflammation (b = −.114, p = .011), but there was no significant main effect of PRWS (b = −.053, p = .265). Finally, using an internal moderator approach (Mirowsky, 2012), we compared the association of CS with inflammation among those not in a committed relationship to those in more or less supportive relationships, showing a significant and stronger association of CS with inflammation for those with low PRWS (- 1 SD; b = .182, p < .001), a weaker and non-significant association of CS with inflammation among those with higher PRWS (+ 1 SD; b = −.002, p = .975), and an intermediate, and non-significant association of CS with inflammation among those with no committed romantic relationship (b = .077, p = .227). Results were robust to number of cytokines included in the inflammation index.

Contextual stressors, such as financial strain, victimization, and experiences with racial discrimination, have been shown to contribute to biological dysregulation and increased vulnerability to infectious and chronic illness, and inflammation is thought to be a key mechanism linking these stressors to outcomes (Cohen, et al., 2012; Cutrona et al., 2015). Briefly, elevated levels of inflammatory cytokines coupled with an insufficient regulating response is thought to be prompted by contextual stressors as a conserved response to adversity (Cole, 2014). This, in turn, sets the stage for the development and/or exacerbation of chronic illness and heightened risk of future morbidity and mortality (Ridker, 2007). Inflammation is a particularly important concern for African Americans, who as a group, have elevated levels of inflammation (Chyu & Upchurch, 2011). Importantly, the role of contextual stress in promoting inflammation is thought to vary across individuals such that stronger effects are expected among those experiencing lower levels of subjective social connection.

During emerging adulthood (ages 24 to 29), romantic partner relationships come to play a more central role in the extent to which individuals feel socially connected and engaged, and may have increasing importance relative to family of origin and other sources of social connection. As a consequence, perceived romantic partner support in relationships may exert an increasingly important influence on the association between contextual stress and the inflammatory response across this time period. Three considerations combine to underscore the potential role of romantic relationships in conditioning the association of contextual stress and inflammation. First, there is a large body of research on the stress buffering effects of romantic partner relationships (Cutrona, 1996; Hostinar, 2015) indicating that stressors often have less impact when individuals are embedded in relationships perceived to be supportive. Second, there have been demonstrations of direct effects of relationship quality or conflict on inflammation (e.g., Kiecolt-Glaser, Gouin, & Hantsoo, 2010; Whisman & Sbarro, 2012; but see Uchino et al., 2018 for a cautionary note), as well as a range of health outcomes (Robles, Slatcher, Trombello, & McGinn, 2014). Third, a well-developed bio-psycho-social theoretical framework suggests that social inclusion helps to offset the pro-inflammatory response to contextual stress (e.g., Cole, 2014; Irwin & Cole, 2011), with romantic relationships emerging as a key source of social inclusion during young adulthood.

Why study committed relationships and not just marriage in emerging adulthood?

Emerging adulthood has traditionally been the key time period for the development of committed, romantic partner relationships. However, individuals going through the transition to adulthood currently appear to experience a more extended process than did prior generations (Settersten and Ray 2010), with more time spent in committed relationships prior to marriage. At the same time, the ability of African American parents to buffer the stressors associated with the transition to adulthood may be diminished relative to their majority counterparts (Barr, et al., 2018; Corcoran & Matsudaira, 2008), further increasing the potential importance of non-marital, committed partner relationships for young adult African Americans. Accordingly, as the importance of emerging adulthood for future adult success has grown (Settersten and Ray 2010), the likely impact of contextual stressors on young African Americans as well as the likely importance of young adult, committed relationships has also grown.

Contextual stressors and illness among African Americans.

African Americans face elevated levels of “contextual stressors” due to discrimination, criminal victimization, and financial stress relative to others (Geronimus, Hicken, Keene, and Bound, 2006), and also bear a disproportionate share of the burden of chronic diseases of aging (CDAs), a pattern that holds for both men and women. In particular, as a group, African Americans experience earlier onset of chronic illness than majority counterparts (e.g., Office of Minority Health, 2017; Appel, Harrell, & Deng, 2002), with early onset predicted by the experience of elevated contextual stressors (Geronimus, et al., 2010). Three types of contextual stressors deserve particular attention in the current investigation because they are commonly experienced at elevated rates by young adult African Americans (e.g., Peterson & Krivo, 2010; Williams & Jackson, 2005), and are notable for their potential biological impact: financial distress (Bird et al, 2010; Koster et al., 2006), perceived discrimination (Geronimus & Shaw, 2013; Cardarelli et al., 2010; Friedman, Williams, Singer, & Ryff, 2009; Mays, Cochran, & Barnes, 2007), and criminal victimization (Peterson & Krivo, 2010; Simons et al., 2002).

Why hypothesize a moderating effect of romantic partner support and warmth on the association of CS with inflammation?

Even if contextual stressors have the ability to promote greater inflammation, this does not automatically imply that perceived qualities of romantic partner relationships should intrude upon or condition the association of contextual stress with inflammation. A broad evolutionary explanatory framework that predicts this type of moderation is offered by Irwin and Cole (2011). They posit that in the evolutionary history of humans, it likely would have been adaptive to respond to an environment characterized by heightened threat (for example, increased predation or hostile conspecifics) by increasing the body’s pro-inflammatory response-- a response that helps heal wounds and fight resulting bacterial infection (Greenberger et al. 1995). This response would be adaptive by complimenting automatic behavioral fight-or-flight stress responses. Irwin and Cole (2011) suggest that this response may have become maladaptive in the modern world, however, due to a shift in sources of threat that has introduced a preponderance of abstract (i.e., non-physical) threats. As a consequence, threats now have greatly prolonged chronicity, potentially prolonging inflammatory reactions and giving them greater potential to cause damage to the host than to the pathogen (Barton, 2008), and greater potential to result in cardiovascular, neurodegenerative and neoplastic diseases. At the same time, Irwin and Cole (2011) suggest that the programmed inflammatory response may be more maladaptive for some than for others. In particular, they suggest that the pro-inflammatory tendencies caused by contextual stressors may be activated or enhanced by perceived social isolation, or the perception that others cannot help. Conversely, this biological programming may be deactivated or neutralized by the perception that help is available from well-meaning, close others. Although relationships are multidimensional and intercorrelated (e.g. overall quality, satisfaction, commitment, perceived support, harshness, etc.), the most important dimension based on Cole’s work should be perceived relationship warmth and support (PRWS). Unresolved by the foregoing considerations, however, is whether there should be an accumulating effect of PRWS across young adulthood. Although the Irwin and Cole (2011) model suggests primarily a moderating effect of PRWS assessed concurrently with inflammation, models focused on allostatic load (e.g., McEwen, 2007) might suggest the potential for an accumulation of cross-sectional effects over time. Accordingly, it may be useful to explore potential effects of earlier PRWS as well as PRWS concurrent with the assessment of inflammation.

Although the association of contextual stress with inflammation is particularly relevant for young adult African Americans, no prior studies have examined the role of romantic partner support in moderating such effects. Available research on the association of romantic partner relationships with inflammation has typically focused on older and white individuals (e.g., Kiecolt-Glaser, et al., 2003; Whisman & Sbarra, 2012). Accordingly, in the context of a young adult African American population at risk for experience of elevated contextual stressors, questions remain regarding the extent to which PRWS will interact with relevant contextual stress (Cox et al., 2016).

Characterizing the Pro-Inflammatory Response.

To address the potential moderating role of PRWS on the association between contextual stress and inflammation, we examined pro-inflammatory response, characterizing the response broadly by assessing a number of pro- and anti-inflammatory cytokines using a “multiplex” assessment approach (e.g., Tighe, Ryder, Todd, & Fairclough, 2015). As with prior work examining relationship effects on cytokines, an advantage of testing moderating effects using cytokines is that they are not confounded with self-report and so complement prior work with self-reported outcomes such as distress, depression, or relationship outcomes (cf., Cutrona, 1996; Uchino, 2006), and also expand upon prior work with older, married couples (e.g. Kielcot-Glaser, et al., 2003; Whisman & Sbarro, 2012).

The need to include anti-inflammatory cytokines.

Pro and anti-inflammatory cytokines play different roles in response to external threats. In particular, for response to tissue damage or infection, an initial increase in pro-inflammatory cytokines is typically followed at a lag by the release of anti-inflammatory cytokines that regulate and temper the magnitude and duration of the inflammatory response. Inadequate concentrations of anti-inflammatory cytokines therefore result in excess inflammation, potentially producing harm to the host. Unfortunately, the important role of anti-inflammatory cytokines such as IL-4, IL-10, and IL-13 in regulating inflammation has been overlooked in most prior research on the effects of contextual stress and the social regulation of its impact, decreasing the likelihood of obtaining a comprehensive picture of inflammatory processes. For example, IL-4, IL-10, and IL-13 all have strong inhibitory effects on pro-inflammatory cytokines, suppressing monocyte-derived cytokines such as IL-1, TNF, IL-6, IL-8, and MIP1 (e.g., Brown & Hural 1997; Wang, et al., 1995), effects that have been shown to be causal using both animal and human models (e.g., Greenberger, et al. 1995). Because anti-inflammatory cytokines can temper the impact of pro-inflammatory cytokines, their presence may be as important or more important than pro-inflammatory cytokines alone in capturing the potential for inflammation to influence health.

Although an impressive body of research indicates that acute and contextual stress is associated with inflammatory processes, most studies have used only one or a few markers of inflammation (for reviews see Slavich & Irwin, 2014; Morisette-Thomas et al., 2014), typically limited to pro-inflammatory markers. This approach, however, fails to capture the complexity of cytokines with antagonistic effects (Abbas, Lichtman, and Pillai, 2015). Likewise, a lack of balance between pro- and anti-inflammatory cytokines has been linked to onset of a variety of chronic illnesses (see for example, Andargie and Ejara, 2015; Shao et al., 2014; Wang et al., 2013; Kumar et al., 2015; and, Chen et al., 2010). Accordingly, in the current investigation we assess inflammation using a multiplex assay to capture a number of cytokines, and we assess both pro- and anti-inflammatory cytokines (Franceschi and Campisi, 2014; Morrisette-Thomas et al., 2014), combining them into an index of the balance of pro- to anti-inflammatory activity. It should be noted, however, that our list of pro and anti-inflammatory cytokines is only “primarily” pro or anti-inflammatory, with some cytokines showing different patterns of activity in specific contexts, such as allergic response, or when acting in concert with other cytokines or receptors (e.g., Scheller, Chalaris, Schmidt-Arras, Rose-John, 2011).

Proposed Hypotheses.

The foregoing review suggests the hypothesis (H1) that earlier contextual stress will be positively associated with inflammation ratio even after potentially confounding concurrent variables are controlled. In addition, it suggests the hypothesis (H2) that there should be a smaller association between earlier contextual stress and inflammation ratio among those currently in a committed relationship who have greater perceived relationship warmth and support (PRWS) at age 29. Finally, (H3) we hypothesize that those with greater PRWS will show a less robust association between earlier contextual stress and inflammation ratio than will counterparts who are not in committed relationships, but those with low PRWS will show a more robust association than counterparts not in a committed relationship. The review also suggests a number of exploratory analyses including (E1) to examine whether change in contextual stress from age 24 to age 29 has an additive effect on inflammation beyond that associated with contextual stress at age 24; (E2) to test whether PRWS at age 24 moderates the association of CS at age 24 with inflammation; (E3) to test whether there are similar or different direct and moderating effects using relationship harshness, or global relationship satisfaction in place of PRWS; and (E4) to test whether findings are robust to different decision rules for inclusion of cytokines in the inflammatory index.

METHODS

The protocol and all study procedures were reviewed by the University Institutional review Board of the University of Georgia. (Title: FACHS IV; Protocol ID#: STUDY00000172).

Participants

We examined the lagged effect of chronic stress at age 24 on inflammatory response at age 29, using data from the Family and Community Health Study (FACHS), an ongoing longitudinal research project designed to increase understanding of contextual risk and protective factors associated with the health and well-being of African Americans. Youth and their families were approached for participation in FACHS when youth were, on average, 11 years old, and the first wave of the FACHS data was collected in 1997–1998 from 889 African American, fifth-grade children (467 from Iowa and 422 from Georgia). Comparison of those who did not participate in the blood draw at age 29 vs. those who did indicated no significant differences on caregivers’ education, household income, family structure, or neighborhood characteristics at age 11. However, there was a difference in the percent male, indicating increased attrition across waves for male participants, with 53.2% of those lost to attrition vs. 37.8% of those providing blood at age 29 being male, t(886) = −4.64, p < .001. Additional details of the original recruitment, attrition across waves, and the comparison of individuals who participated in the analyses described herein with those not included can be found in the supplemental description of the sample, and Supplemental Table S1.

Directly relevant to the current investigation, for the blood draw we decided to restrict assessment to those in Georgia, Iowa, or a contiguous state, due to the logistics of scheduling home visits by phlebotomists. After locating all available individuals who had participated in either of the prior two waves (and excluding persons who were deceased, incarcerated, or otherwise unreachable), we were left with a total potential sample at age 29 (2015–2016) of 545 individuals, of these 470 (86%) agreed to be interviewed and to provide blood. Assays were completed for 411 (87.4%) and these comprise the sample for the current analyses, with the other 59 lost secondary to technical problems. Rate of missing data was 1.45% for romantic partnership warmth and for partner harshness. Of the 411, 6 were missing data for romantic partner warmth and/or harshness and 3 were missing data for BMI, leaving 402 cases that could be used in the current analyses. In all analyses missing values for control variables were handled by multiple imputation using the “MI” function of the STATA 15 software. All 402 participants self-identified as African American. The final sample can be characterized as low to moderate in annual income (M = $21,111.98), with moderate average educational attainment, percent graduating HS (89.6%) and percent graduating college (13.9%). In addition, most were employed (80.1%), had health insurance (82.1%), and had 0.61 children on average, with 62.4% of participants reporting a committed romantic partner relationship at age 29. The 251 participants designated as having a committed romantic partner indicated that they were in a committed relationship characterized by either exclusively dating one person (51.4 %), being engaged and cohabiting (11.9%), engaged or cohabiting but not both (7.2%), or married (29.5%).

Measures

Contextual stress.

We assessed three sources of contextual stress (racial discrimination, victimization, and financial stress) at ages 24 and 29. Racial discrimination was assessed using a 13-item scale (Landrine & Klonoff, 1996) focused on respondents’ experience of discriminatory events, with responses ranging from 1 (never) to 4 (several times) during the preceding year (e.g., “How often has someone yelled a racial slur or racial insult at you just because you are African American?”). The unstandardized scale had a mean of 19.51 (7.06) at age 24 and 17.97 (7.49) at age 29, suggesting that exposure to some form of discrimination in the preceding year was common. Cronbach’s alpha at age 24 was .908 and at age 29 was .851. Two items with three response options each assessed financial stress (Conger & Elder, et al., 1990): “During the past 12 months, have you had (a) serious money problems or (b) not enough money?” The scale had a potential range of 2 to 6, and an unstandardized mean of 3.52 (sd = 1.15) at age 24 and 2.98 (sd = 1.14) at age 29, suggesting that some participants experienced substantial financial strain. Two items assessed victimization, (e.g., “Were you a victim of a violent crime in the past 12 months; yes = 1, no = 0”). The scale had a potential range from 0 to 2, and an unstandardized mean of .15 (sd = .39) at age 24 and .12 (sd = .35) at age 29. Scale scores were normalized before the mean of the scales was calculated to create an overall composite stress measure at each age. The overall mean was used to index cumulative stress. The correlation between age 24 and 29 CS was (.373). Using Nunnally’s (1978) reliability formula for composite variables, the reliability for the index was .92. at age 24 and .95 at age 29. Additional background detail about the contextual stress measures can be found in the appendix.

Perceived Relationship Warmth and Support (PRWS).

When participants had committed romantic partner relationships at age 24 or 29, they were asked three questions concerning the degree to which warmth and support was displayed in the relationship by the partner and three parallel questions about their own behavior (e.g., helped do something important or showed affection during the past month) (Surjadi, Lorenz, Wickrama, & Conger, 2011). Responses ranged from 1 (never) to 7 (always). Larger values indicated greater PRWS. Factor analysis of the six items comprising PRWS indicated a single factor, with all items loading in the expected direction (λ > .70). (See supplemental Table S2 for results of the factor analysis). Accordingly, PRWS was scored as a single scale. The unstandardized scale had a mean of 18.60 (sd = 5.05) and a range of 9 to 54 at age 24, and a mean of 17.84 (sd = 3.45) and a range of 6 to 24 at age 29. Alpha for the scale was .866 at age 24 and .884 at age 29.

Other Perceived Romantic Relationship Characteristics.

The Relationship Hostility Scale (Cui et al., 2005) assessed at age 29 was used as a control in all analyses, to control effects of harsh, aggressive actions by the partner. In addition, we examined harshness as a potential alternative to PRWS in supplemental analyses. This scale consists of five items about partner behavior such as insult or swear, and shout or yell, and five parallel questions about own behavior toward one’s partner. The scale had a mean of 13.12 (sd = 3.00) and a range of 10 to 29 at age 24, and a mean of 12.49 (sd = 3.03) and a range of 10 to 35 at age 29, suggesting low levels of harshness on average, but with some participants indicating elevated harshness. Alpha for the scale was .80 at age 24 and .72 at age 29. The Global satisfaction was also used as a potential alternative to PRWS in supplemental analyses. It is a single-item measure with potential responses ranging from 1 (extremely unhappy) to 6 (extremely happy) in response to the question “How happy or satisfied are you, all things considered, with your relationship?”

Control variables.

Because they have been shown to affect cytokine levels (see O’Connor et al., 2009), statistical covariates were included to control the life style factors of drinking, diet, exercise, sleep quality, and BMI. In addition, we controlled age and sex in all analyses. To control potential effects on relationship processes, we also controlled relationship instability (1 = a change in partner or relationship status from waves 6 to 7; 0 = the same relationship partner status at waves 6 and 7). Exercise was measured with two items: On how many of the past 7 days did you exercise or participate in physical activity for at least 30 min that made you breathe hard such as running or riding a bicycle fast? And, on how many of the past 7 days did you exercise or participate in physical activity for at least 30 min that did not make you breathe hard, but was still exercise such as fast walking, slow bicycling, skating, pushing a lawn mower, or doing active household chores? The response categories ranged from 1 (0 days) to 5 (all 7 days). Scores on the two items were averaged to form the exercise measure. Healthful diet was assessed using two items that asked about frequency of fruit and vegetable consumption during the previous 7 days. Responses ranged from 1 (none) to 6 (more than once every day) and were averaged to form the healthful diet variable. Episodic alcohol consumption (binge) was defined as the consumption of 3 or more drinks of alcohol (1 = never, 6 = several times per week). Sleep quality was measured using the subjective item (1 = very bad, 4 = very good): “During the past month, how would you rate your sleep quality overall?” Depression was assessed using a nine-item measure of depressive symptoms (the University of Michigan Composite International Diagnostic Interview; Kessler & Mroczek, 1994). Respondents were asked to report (0 = no; 1 = yes) whether they experienced several symptoms of depression (e.g., “felt sad, empty, or depressed most of the day” and “lost interest in things”) for at least a 2-week period in the past year. The Cronbach Alpha for the scale was .830. Finally, BMI was assessed based on measured height and weight at the time of the blood draw (range: 16.61 – 61.62).

Assessment of Pro-inflammatory and Anti-inflammatory Cytokines

A certified phlebotomist drew blood at each participant’s home. Two tubes were spun immediately to separate serum into cryo-vials that were then frozen and stored in a −80° freezer until used for the analysis of cytokines described below.

Levels of cytokines in plasma were determined using a Bio-Plex 200 system (Bio-Rad, USA) and a standard 17-plex cytokine detection kit according to the manufacturer’s protocol. The Bio-Plex Human Cytokine 17-Plex panel includes human interleukin IL-1β, IL-2, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-12, IL-13, IL-17, granulocyte colony-stimulating factor (G-CSF), granulocyte-macrophage colony-stimulating factor (GM-CSF), interferon-gamma (IFN-g), monocyte chemotactic protein-1 (MCP-1), macrophage inflammatory protein-1b (MIP-1b), and tumor necrosis factor alpha (TNF-a). The Bio-Plex assay combines fluorescent flow cytometry and ELISA technology. Because IL-1β, IL-2, IL-4, IL-5, IL-6, IL-12, IL-17, G-CSF, IFN-g, GM-CSFR, and MCP-1 were present at undetectable levels in the majority of samples (i.e., ≥ 50%) these cytokines were excluded from the index, leaving 6 cytokines to be included in the index. Of the cytokines included, 2 are anti-inflammatory and 4 are pro-inflammatory (see formula below). Cytokine measurement reproducibility was good (overall average intra-assay coefficients of variation for IL-7, 3.4%; IL-8, 6.2%; IL-10, 2.4%; IL-13, 3%; MIP-1β = 3.6%; TNF-α = 2.8%). Supplemental table S3 provides raw descriptive statistics for all cytokines, including the coefficient of variability for each standard as well as the median and the interquartile range for detectable samples. In addition, in the last column of Table S3 we provide the percent undetectable for each cytokine.

To correct for potential method variance reflecting variance associated with plate rather than the variables of interest, we corrected the 6 cytokines used in the index for plate-to-plate variation using linear regression with the eight plates entered as categorical covariates. For all cytokines, we then used the residuals after the removal of plate effects in subsequent analyses. In line with recommendations for conservative treatment of biological assays for which non-detectable values are moderately frequent (KHRC, 2015), we trichotomized cytokine values into 1 = undetectable, 2 = detectable but not elevated, 3 = detectable and in upper quartile of detectable scores (i.e., scores that were at or above the 75^th percentile for detectable scores were considered elevated. Numerical cutoffs varied across cytokines and are provided in table S3).

To capture the relative balance of pro-inflammatory to anti-inflammatory activity, cytokines primarily involved in pro-inflammatory responses were summed separately from cytokines involved primarily in anti-inflammatory responses. The following equation was used to calculate the relative balance of pro-inflammatory cytokines (in the numerator) to anti-inflammatory cytokines (in the denominator):

$$\frac{(\mathrm{I}\mathrm{L}7+\mathrm{I}\mathrm{L}8+\mathrm{M}\mathrm{I}\mathrm{P}1+\mathrm{T}\mathrm{N}\mathrm{F})}{(\mathrm{I}\mathrm{L}10+\mathrm{I}\mathrm{L}13)}$$

Using this ratio, higher scores indicate a greater pro-inflammatory response without a corresponding, mitigating anti-inflammatory response. As expected, levels covaried across all pro- and anti-inflammatory cytokines.

Supplemental analyses were conducted to test generalization using both more and less conservative cutoffs to characterize the set of cytokines. (See replication of primary analysis in supplemental table S7 using these alternative indices).

Plan of Analysis

We first examined simple correlations and characterized the data in terms of mean and SD for all variables (Table 1). We then examined each of the three primary hypotheses in turn, followed by examination of the exploratory analyses. First, we used the full sample to examine the association of contextual stress (CS) with the inflammation ratio (Table 2, model 1). Second, for the subsample that was in a committed relationship, we examined the main effects of CS, PRWS, and perceived relationship harshness on inflammation (Table 2, model 2) and then examined moderation of the association between CS and inflammation ratio by PRWS (Table 2, model 3). Finally, we used the full sample to conduct an internal moderator analysis (see Mirowsky, 2012), allowing us to directly compare the effect of CS for those in a romantic relationship (at either higher or lower levels of relationship support) with the main effect of contextual stress among those with no committed partner relationship at age 29 (Table 2, column 4). The internal moderator approach explicitly models the fact that the hypothesized moderator applies only to some members of the data set (see Mirowsky, 2012), providing a “contingent” interaction effect that is similar to the observed effect of the moderator in analyses restricted to those with partners, but allowing comparisons between those with and without romantic partners.

Table 1.

Correlations, Means, and Standard Deviations for the Study Variables at age 29 (N = 402)

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
1. Inflammation ratio	―
2. Contextual stress (W6)	.185**	―
3. Change in stress (W6–7)	.064	.000	―
4. Romantic partners	.025	−.001	.040	―
5. PRWS (warmth support)	−.092†	−.140**	−.074	.000	―
6. Relationship harshness	.048	.150**	.070	.000	−.295**	―
7. Sex (Male =1)	−.086†	.007	−.051	.001	−.027	.059	―
8. Relationship instability	.013	.094†	.015	−.399**	.113*	−.093†	.010	―
9. Healthy diet	.083	.042	.038	−.012	.090†	−.038	−.194**	−.014	―
10. Exercise	.125*	.158**	.016	.048	−.030	.041	.173**	−.057	.178**	―
11. Episodic alcohol	.031	.224**	.134**	−.008	.010	.094	.092†	.101*	.046	.129*	―
12. Sleep quality	−.166**	−.305**	−.189**	−.071	.113*	−.126*	.137**	−.047	−.084†	−.075	−.163**	―
13. Age	.017	−.011	.028	.042	−.028	−.061	−.025	−.033	−.078	−.029	−.014	−.048	―
14. BMI	−.052	.015	−.004	.019	−.026	−.044	−.216**	−.029	.094†	−.130**	−.004	−.077	−.018	―
15. Depression	.026	.206**	.341**	−.075	−.068	.077	−.158**	.107*	.013	−.047	.211**	−.305**	.012	.082	―
Mean	2.404	−.004	.000	.624	.000	.000	.378	.751	6.575	4.944	1.860	3.101	28.597	31.465	1.659
SD	.727	.696	.600	.485	.790	.790	.486	.433	2.414	2.324	1.204	.850	.762	8.523	2.149

Note: A point-biserial correlation is used to test a dichotomous variable (e.g., romantic partners and males)

^†p ≤ .10,

^*p ≤ .05,

^**p ≤ .01 (two-tailed tests).

Table 2.

Regression models of contextual stress (CS) at age 24, PRWS (warmth support) at age 29, and their interaction on Inflammation Ratio. Model 1 tests direct effects in full sample, Model 2 tests direct effects for those with a committed partner, Model 3 tests moderated effects for those with a committed partner, Model 4 tests internal moderator effects using the full sample.

	Model 1		Model 2		Model 3		Model 4
	b	β	b	β	B	β	b	β
Contextual stress (CS) W6	.112** (.039)	.155	.117* (.053)	.161	.078 (.054)	.107	.077 (.057)	.105
Change in CS (W6-W7)	.040 (.038)	.055	.043 (.046)	.067	.039 (.045)	.059	.032 (.038)	.044
PRWS			−.047 (.048)	−.067	−.053 (.047)	−.075
PRWS × CS					−.114* (.044)	−.168
Having romantic partner							.006 (.081)	.004
× PRWS							.013 (.075)	.013
× CS							−.063 (.048)	−.068
× CS × PRWS							−.116* (.045)	−.132
Relationship harshness			−.010 (.048)	−.013	−.004 (.047)	−.006	−.001 (.048)	−.002
Males	−.153† (.079)	−.102	−.100 (.099)	−.069	−.101 (.098)	−.070	−.157* (.079)	−.105
Relationship instability	.004 (.084)	.002	−.024 (.096)	−.016	.008 (.096)	.006	.048 (.094)	.028
Healthy diet	.011 (.016)	.036	.008 (.020)	.027	.011 (.020)	.036	.014 (.016)	.048
Exercise	.029† (.017)	.093	.013 (.020)	.045	.016 (.020)	.053	.030† (.016)	.095
Episodic alcohol	−.012 (.033)	−.020	−.002 (.039)	−.004	−.001 (.039)	−.001	−.009 (.033)	−.015
Sleep quality	−.095* (.046)	−.111	−.093 (.060)	−.110	−.096 (.060)	−.114	−.091† (.047)	−.106
Age	.015 (.047)	.015	.034 (.061)	.035	.014 (.060)	.015	.001 (.047)	.001
BMI	−.006 (.004)	−.068	−.001 (.006)	−.007	−.001 (.006)	−.008	−.006 (.004)	−.069
Depression	−.021 (.019)	−.061	.007 (.026)	.021	.009 (.025)	.024	−.020 (.019)	−.059
Collection time
Morning before noon	−.062 (.111)	−.031	−.115 (.141)	−.058	−.098 (.140)	−.049	−.045 (.111)	−.023
Afternoon	.063 (.081)	.043	.026 (.102)	.019	.003 (.101)	.002	.052 (.082)	.036
Constant	2.340† (1.393)		1.689 (1.807)		2.204 (1.797)		2.665* (1.399)
R-square	.080		.094		.119		.101

Notes: Unstandardized (b) coefficients shown with robust standard errors. Reference group = Evening.

^†p ≤ .10;

^*p ≤ .05;

^**p ≤ .01 (two-tailed tests).

In all analyses, variables found to be related to cytokines in prior research (O’Connor et al., 2009) were controlled, as was relationship instability (i.e. changing relationship status across assessments), and relationship harshness. After reporting the primary analyses, we also briefly report supplemental analyses that examined the exploratory hypotheses.

Results

Table 1 presents descriptive statistics and correlations for all variables in the analysis, including control variables. As expected, there was a significant zero-order correlation between contextual stress (CS) and the inflammation ratio measured at age 29 (r = .185, p < .001). There was no significant zero-order correlation between change in CS and inflammation ratio. Also noteworthy are significant correlations of CS with PRWS (r = −.140, p = .005), perceived relationship harshness (r = .150, p = .003), drinking (r = .224, p < .001), exercise (r = .158, p = .002), and sleep quality (r = −.305, p < .001). These correlations are consistent with suggestions by Miller et al. (2011) that there may be spillover effects of contextual stress into relationship processes, lifestyle choices, and health behaviors (both health promoting and potentially health damaging). Patterns of associations by sex were similar to those for the full sample. In supplemental table S4 we also provide zero-order correlations with all major study variables for pro and anti-inflammatory cytokines components considered separately.

H1: Earlier contextual stress will be positively associated with inflammation ratio even after potentially confounding concurrent variables are controlled.

As Model 1 of Table 2 indicates, there is a significant main effect association of CS with inflammation ratio using data from all participants (N = 402), with gender, diet, exercise, drinking, BMI, and other potential confounders controlled. The main effect association of CS with later inflammation ratio was b = .112, p = .004, supporting hypothesis 1. There was no significant additional main effect of change in CS from wave 6 to wave 7.

H2: There should be a smaller association between earlier contextual stress and inflammation ratio among those currently in a committed relationship who have greater perceived relationship warmth and support (PRWS) at age 29, even after controlling for potentially confounding effects.

As is shown in Table 2, Model 2, there is a main effect of CS (b = .117, p = .027), and a non-significant effect of PRWS (b = −.047, NS), perceived relationship harshness (b = −.009, NS), and change in CS, among those in a committed romantic relationship at age 29 (N = 251). Model 3 shows that there is a significant, negative, interaction of CS and PRWS predicting inflammation ratio (b = −.114, p = .011). The interaction is explicated in Figure 1. Those with greater perceived romantic relationship support (+1 SD) showed a weak (and non-significant) association of CS with inflammation ratio (b = −.036, NS), whereas those with low romantic partner support (- 1 SD) showed a significant, positive association of composite stress with inflammation ratio (b = .192, p < .001). There was no significant effect of harshness. As can be seen in supplemental Table S5, however, when analyzed separately, global satisfaction, and harshness show similarly shaped, but non-significant, moderating effects (albeit with the expected reversed effect for harshness) (see Supplemental figures 2 and 3). As can be seen in supplemental table S6, PRWS measured at age 24 also failed to produce a significant interaction with CS in the prediction of the inflammation ratio. Finally, as can be seen in Supplemental Table S7, we examined the impact of changing the cutoff for inclusion of cytokines in the inflammation index either up or down resulting in sets of pro and anti-inflammatory cytokines based on a cut off of less than 20% undetectable (4 cytokines), less than 60% (7 cytokines); and less than 95% undetectable (14 cytokines). In all cases a significant interaction of PRWS and CS on inflammation ratio was observed.

Figure 1.

Explication of Significant Interaction between contextual stress (CS) and Perceive relationship warmth and support (PRWS) in Model 3 of Table 2 among those in a committed relationship at age 29, showing a significant association of CS on later inflammation ratio for those reporting low support but a non-significant effect among those reporting high support. Contextual stress is graphed ranging from −1 sd to +2 sd because there were very few observations lower than one sd below the mean (the range of observed contextual stress values was −1.258 sd to 3.147 sd).

H3: Using an internal moderator approach to compare those with and without a committed romantic partner relationship, individuals with greater PRWS will show a less robust association between earlier contextual stress and inflammation ratio than will counterparts who are not in committed relationships, but those with low PRWS will show a more robust association than counterparts not in a committed relationship.

In Model 4 of Table 2, we used an internal moderator approach to compare the moderated effects attributable to romantic partner support with the effect of CS among those with no romantic partner. Corroborating the results restricted to those with a romantic partner, the interaction of PRWS and CS was significant, b = −.116, p = .002. Focusing on the three simple slopes representing the association of composite stress with inflammation ratio among (a) those in committed relationships reporting a warm, supportive romantic partnership (1 SD above the mean); (b) those not in committed, exclusive romantic partnerships, and c) those reporting committed romantic partnerships that were not warm (1 SD below the mean), we found that for those with perceived supportive couple relationships, the slope for the prospective effect of contextual stress on inflammation is essentially 0, b = −.002, ns. Whereas for those reporting romantic partner relationships that were not perceived to be supportive the slope was significant, b = .182, p < .001. For those not reporting a committed, exclusive romantic partnership, the slope was intermediate but still not significant, b = .077, ns. This effect is shown in Supplemental Figure 1. This indicated that the prospective effect of stress on inflammation was significantly lower among those with a warm, supportive romantic partner relationship relative to those with a non-supportive partner, and explains why mere presence of a partner did not have a significant main or interactive effect on inflammation ratio. That is, the divergence between those with more vs. less supportive relationships is such that, relative to those without a committed partner, they may experience either a significant or a non-significant association that is numerically larger or smaller than those not in a committed relationship (cf. Holt-Lunstad, Smith, & Layton, 2010).

DISCUSSION

Characterizing the role of perceived relationship warmth and support (PRWS) in emerging adulthood in moderating the association of contextual stress with inflammation is an important avenue of inquiry given implications for long-term health effects. The current results support the hypothesis that elevated contextual threat is associated with pro-inflammatory tendencies for young adult African Americans, and that the association may be moderated by perceived level of relationship warmth and support in a romantic partnership (e.g., Cole, 2014; Irwin & Cole, 2011). In the current investigation we focused on contextual stressors that are typically seen as affecting young adult African Americans to a greater extent than their majority counterparts, including discrimination, financial stress, and victimization (Geronimus, et al., 2006; 2010; Geronimus & Snow, 2013; Peterson & Krivo, 2010).

To provide a comprehensive measure of inflammatory tendencies, we used an index capturing the balance of pro- vs. anti-inflammatory cytokines. Our primary analyses utilized cytokines with detection rates > 50%, and, for our primary analyses, we characterized each person as undetectable, detectable, or elevated on each of the retained cytokines. We also controlled variables previously shown to be associated with inflammation, indicating that the pattern we observed was robust to controls for potentially confounding variables. Further showing the robustness of the association, in supplemental analyses, we replicated all primary findings using a range of cut-offs for cytokine inclusion, resulting in a series of cytokine indices utilizing differing numbers of cytokines. Each of these alternative indices also showed moderation of the association of CS with inflammation ratio by PRWS.

An important focus of the current investigation was to test the hypothesis that the association of contextual stress with inflammation would be moderated by current PRWS. Several clear conclusions emerged. First, as would be expected based on Cole’s model, current PRWS moderated the association of earlier contextual stress with inflammation ratio. Past experience of PRWS was not a significant moderator of the association, nor was perceived harshness in the relationship. That is, the perception of having a current committed relationship characterized by mutual warmth and support moderated the association of contextual stress with inflammation in a manner consistent with it playing a role in reducing the conserved response to elevated adversity, as predicted by Irwin and Cole (2011). At the same time, supplemental analyses indicated that harshness and global relationship satisfaction had similarly shaped, albeit reduced and non-significant moderating patterns. It is possible that the lack of significant moderating effects for perceived harshness, and perhaps global satisfaction was due primarily to lower reliability of measurement for harshness and satisfaction in the current data set, a factor known to be associated with considerably reduced power to detect moderating effects (Aiken & West, 1991). However, the observed non-significant effects for harshness are also consistent with prior examination of main effects on inflammation showing an effect for perceived support but not for harshness (e.g., Whisman & Sbarra, 2012). Likewise, current results provide a conceptual replication and extension of prior work on the impact of romantic relationship quality on self-reported health (Barr, Culatta, & Simons, 2013; see also Robles, et al., 2014).

Second, as would also be expected from Irwin and Coles’s (2011) model, current perceived relationship warmth and support at age 29 was a significant moderator of the association between CS and inflammation ratio whereas perceived support at earlier ages (i.e., age 24) was not. That is, at least in young adulthood, when relationships are more fluid, pro-inflammatory tendencies were moderated by current perceived relationship support but not by past perceived relationship support. This result appears consistent with the expectation that a conserved pro-inflammatory response to heighted threat should reflect the dual perception of heightened threat level and the perception of being on one’s own as one faces that threat.

Third, consistent with expectations that committed relationships can be both advantageous (when perceived as supportive) or disadvantageous (when perceived as non-supportive), we found a numerically intermediate and non-significant association between contextual stress and inflammation for those not in a current committed relationship. That is, the magnitude of the association for those not in a current committed relationship fell between the significant association of CS and inflammation ratio observed for those in a committed relationship with low PRWS, and the non-significant association observed for those high in PRWS. Accordingly, the current results also reinforce theorizing that relationship status alone may not be informative about the stress moderating capacity of romantic partner relationships (e.g., Holt-Lunstad, Smith, & Layton, 2010).

Overall the observed pattern of associations supports expectations that contextual stressors of the sort occurring frequently for African American young adults are associated with an elevated ratio of pro-inflammatory to anti-inflammatory cytokines. This may reflect the modern structure of contextual stress in which increasingly symbolic stressors exert longer-term effects on inflammatory response, perhaps to a greater degree than was the case earlier in human evolutionary history. Or, conversely, it may reflect a tendency for those with elevated inflammation to experience greater stress. Mitigating the latter interpretation, CS was measured prior to the assessment of inflammatory cytokines. At the same time, there was broad support for the expectation that aspects of current romantic partner relationships, such as greater warmth and support in the relationship, would moderate these associations. As suggested by Irwin and Cole (2011), perhaps this reflects the role of current relationship support in turning off or downregulating inflammatory reactions prompted by the experience of contextual stress. Conversely, perhaps this reflects an important role for the perception of lack of support in amplifying stress effects, a pattern also consistent with Irwin and Cole’s (2011) theorizing. In either case, this pattern suggests an important role for relationships during emerging adulthood, and it is a role that may begin prior to marriage.

From the perspective of health promotion, these findings suggest that relationships may play an important role in moderating the impact of contextual stressors that are common for African American youth during emerging adulthood. Accordingly, emerging adulthood may be relevant as a developmental period during which preventive intervention efforts designed to counter the corrosive effects of contextual stress might be effective. For those suffering the effects of discrimination, financial strain, and victimization, reminders of threat are ubiquitous. As a consequence, the importance of protective relationships, and the negative consequences of unsupportive relationships, may have increased in proportion to the increase in contextual stressors, and this dynamic may be particularly apparent for African Americans during emerging adulthood. Accordingly, as contextual threats proliferate and become more chronic (Col, 2009), the potential role of romantic partner relationships deserves additional attention.

The current investigation does not provide a direct of test of a causal role of current PRWS. However, a preventive intervention study could test the possible causal relationship between enhanced perceived support from a romantic partner during emerging adulthood and the prevention of pro-inflammatory tendencies, as well as the attendant accumulation of negative health effects attributable to chronic inflammation. Accordingly, direct tests of causal effects via preventive intervention would be an important step in confirming or disconfirming the hypothesized relationships identified in the current investigation. As was highlighted in our internal moderator analyses (Mirowsky, 2012), having a committed partner relationship is not, in itself, a panacea. Accordingly, preventive intervention programs designed to protect against adverse effects of contextual stress will need to focus on quality of perceived relationships rather than relationship status alone.

While this study has a number of strengths, limitations must also be addressed. One general limitation of the study is its inability to account for baseline differences in inflammation prior to the occurrence of the contextual stressors. We cannot rule out the possibility that earlier levels of inflammation influenced the experience of stress or the experience of romantic relationships. Accordingly, future work examining change in inflammation over time will be useful in refining the proposed theoretical model and developing novel potential points of preventive intervention to support enhanced health outcomes for African Americans. Second, although the sample size is substantial for longitudinal investigations of this sort, we were underpowered to test hypotheses involving differential patterns by gender or relationship history (i.e., three-way interactions). Future replications with larger samples could clarify the presence or absence of such subgroup effects and sex differences. Third, the multiplex assessment we used was not a high sensitivity assay and so many individuals had undetectable levels of cytokines, including for IL6, which was therefore not used in our index. It is possible that a high sensitivity multiplex assay would have produced a different pattern of results, or included other cytokines, suggesting the need for replication and comparison of results across platforms. In addition, characterizing cytokines as pro vs. anti-inflammatory is a simplification of complex cytokine interactions. Finally, although we found no evidence of a significant moderating effect of either perceived relationship harshness or global satisfaction on inflammation ratio, these variables may assume greater importance in other samples, as couples marry, or develop longer histories together, and interaction effects may be detected using other measurement approaches.