Mid-life adults (N = 113; 55% white, 87.5% female) were recruited from 9 counties within a 50-mile radius of Chapel Hill, NC via paper, television and online advertisement. Our recruitment protocol adopted several a priori inclusion and exclusion criteria. To test whether an established biomarker of the hypothesized “biological residue” of low childhood SES might be reversed in midlife, we sampled from the population of individuals with the highest odds of showing this biological residue. Prospective participants (N =360; see Online Supplementary Materials (OSM) for full CONSORT diagram) were assessed for eligibility through a pre-screening survey that measured multiple indices of low childhood SES, as well as variables known to amplify the relationships between low childhood SES and midlife biological health risks, including delay discounting, low maternal warmth, and mistrust of others (Miller et al., 2011).

Although we initially planned to only recruit participants who met the criteria for low childhood SES and scored below median levels of maternal warmth, and above median levels of delay discounting and mistrust, upon encountering difficulty identifying eligible participants, we relaxed screening criteria to be based solely on childhood SES (See OSM for measures of omitted screening variables and associated descriptive statistics). Specifically, the revised inclusion criteria determined low Childhood SES by scoring a 39 or below on Hollingshead Four-factor Index (see measures section 2.3.1; Hollingshead, 1975), which is the highest score before entering into mid-high range occupations based on Hollingshead scoring system using occupational data from the 1970 U.S. Census. Eligible participants also were between the ages 30-69, had little to no experience with meditation, and had home internet access to complete nightly online surveys.

Participants could earn up to $235 for full participation in the study. Participants were between the ages of 30 and 69 (M = 47.26). See Table 1 for full demographic characteristics. Rates of attrition were moderate (N = 24; 21%), and did not differ significantly across groups χ²(1) = 0.01, p = .93. Participants who dropped out of the study did not differ significantly from the analysis sample on any demographic variables or variables of interest (See Supplementary materials). One person was excluded from final analyses for missing data on control covariates. Only participants who had complete data for CTRA gene expression profiles both pre- and post-intervention were retained for final analyses (N = 88). Data from this larger NIH-supported study [5R01AG048811] have been reported elsewhere (West et al., 2022). Study data and analytic syntax are available at the Open Science Framework (https://osf.io/ve8hx/)

Descriptive statistics confirm that the analysis sample represents those raised in low-SES households. Specifically, the mean Hollingshead Four-Factor Index Score was 26, well-below the cut-off of 39 (Hollingshead, 1975). Homeownership is also a common indicator of social status and income (Cohen et al., 2013), and about a third of the analysis sample reported that their parents never owned their home while they were growing up. See Measures section and Table 1 for details on SES indicators and further sample characteristics.

The minimum sample size for detecting effects of interest was calculated using a structural equation model based on the root-mean-square error approximation (RMSEA) fit index (MacCullum et al., 1996). We defined the null hypothesis to represent lack of close model fit in the population, and as such, we set the RMSEA0 to .07 and the RMSEAa to .01, the latter suggesting excellent model fit in the population. With an alpha level of .05 and considering the degrees of freedom across models, these analyses suggested that 107 participants are needed to achieve power of .80 for our hypothesized mediational process latent growth curve models. While this number is lower than our final recruited sample (N = 113), it is larger than the number of complete cases retained in our final analysis sample (N = 88). We thus interpret our results with caution.

The Institutional Review Board at the University of North Carolina at Chapel Hill approved all procedures for the study (Study #14-1938). Data collection took place in two waves between February 2015 and May 2016. Participants were recruited to participate in a study about how to best reduce health risks. Prospective participants were first directed to a website which informed them on the purpose and procedures of the study and invited them to complete pre-screening measures. Participants indicated consent during pre-screening and were assigned a numeric identification code. Those who met the revised inclusion/exclusion criteria were invited to participate in the study. Once eligibility was confirmed and participants were scheduled for an initial lab visit, they were allocated (via a random number generator) to either the loving kindness meditation group or the mindfulness meditation group. All study personnel involved in lab visits were blind to participants’ condition. The study lasted 12 weeks, with the 6-week meditation workshops occurring from Week 3 through Week 8. Baseline biological samples, covariates, and psychological measures were collected at Week 1 and again at Week 12. Daily emotion reports were collected throughout the study, but only data from the second week (baseline measures) and the weeks during the meditation intervention (Weeks 3-8) were analyzed here.

Genome-wide transcriptional profile data were quantile normalized, log2-transformed, and z-score standardized within-gene for statistical analysis. Following previous analyses (Fredrickson et al., 2015), primary analyses focused on a 53-gene CTRA indicator contrast involving a pre-specified set of 19 pro-inflammatory indicator genes (e.g., IL1B, IL6, CXCL8/IL8, PTGS2/COX2, TNF) and 34 genes involved in Type I interferon signaling and antibody response (e.g., IFI-, MX-, OAS-, and IRF-family genes) with z-score signs inverted for the latter group to reflect its inverse contribution to the CTRA profile. Fredrickson et al., 2015 provides more detail on the genes involved and contrast specification. Data were analyzed in a 2 (Group) x 2 (Time point: baseline vs. post-intervention) x 53 (Gene) mixed effect linear model including random effect subject-specific intercepts to account for dependency from repeated measurement across genes and time points. The analysis treats each of the 53 CTRA indicator genes as a distinct indicator (i.e., a repeated measures vector of responses) and tests whether there is an association between experimental condition and change over time in the average value of that response vector. Data were analyzed using SAS (v 9.3; See OSM for syntax). Inclusion of partial cases led to non-convergence in maximum likelihood estimations of variance-covariance matrices for mixed-model residuals. Thus, cases that were missing data at either time point (i.e., baseline or follow-up) were excluded from analysis (n = 25). Primary analysis focused on the Group x Time interaction term to quantify group differences in change over time in average expression of the 53 CTRA indicator genes. Ancillary analyses focused on the same interaction term after adding to the model above several covariates typically controlled for in social genomics studies (Fredrickson et al., 2013, 2015; Cole et al., 2015) : age, sex, race/ethnicity (coded White vs People of Color), body mass index, history of regular smoking, history of regular heavy alcohol consumption, current acute illness symptoms at the time of blood sampling, and 7 gene expression values indicating the relative prevalence of major leukocyte subsets within the PBMC sample (CD3D, CD4, CD8A, CD19, FCGR3A/CD16, NCAM1/CD56, CD14).

Independent t-tests and Chi-square tests indicated no significant differences between experimental conditions in age, sex, race, baseline BMI, BMI change, childhood and present-day SES indicators, smoking and alcohol use, illness symptoms, baseline emotions or other psychological variables of interest. Furthermore, there were no differences in workshop attendance (M = 4.92, SD = 1.11), meditation self-practice (days a week: M = 3.23; hours a week: M = 1.20), or response rate to daily reports (M = .75; SD = .19). However, those in the LKM condition reported significantly more informal meditation practice, more than once a day (M = 1.35, SD = .65) compared to the mindfulness group (M = .93, SD = .62; t (86) = 3.122, p = .002). Additional exclusion analyses on T1 primary variables and demographics indicated no significant differences in the above variables between the analyzed and attritted samples (See Table S1 in OSM).

To test for group differences in mean levels and changes in emotions over time, we fit a two-level mixed model with random intercepts using the nlme package (Pinherio et al., 2022) in R 4.0.0. For each model, the Groups, Time and Group X Time interactions are included as predictors and positive and negative emotions were used as the outcome variables. Results indicated significant main effects of Time on both positive emotions (b = .007, p<.001, SE = .001) and negative emotions (b = −.005, p <.001, SE = .001), reflecting increases in positive emotions and decreases in negative emotions. No significant Group X Time interactions emerged. Results from this data set relating to positive and negative emotions over time have been reported in (West et al., 2022). We also explored two daily items separate from the composites, namely, stress and compassion. No significant Group by Time interactions emerged for stress (b = −.004, p = .164, SE = .003) or compassion (b = .003, p = .888, SE = .002). Because no significant main effects of Group or Group by Time interactions emerged for emotions, we did not proceed with planned tests for mediation by emotion.

To test this study’s primary hypothesis, we tested for group differences in the magnitude of change over time in average expression of 53 CTRA indicator gene transcripts (assessing up-regulation of proinflammatory genes and down-regulation of Type I interferon and antibody-related genes). Primary analyses indicated a significant Group x Time interaction, b = .137 ± SE= .039, t (9186) = 3.49, p < .001 (See Figure 1), and follow-up simple effects tests (Time effect nested within Group) showed CTRA gene expression to decrease significantly from baseline to follow-up in those randomized to MM (b = −.058 ± .028, t(9186) = −2.10, p = .036) and to increase significantly from baseline to follow-up in those randomized to LKM (b = .078 ± .027, t(9186) = 2.86, p = .004). In analyses that additionally controlled for age, sex, race, body mass index, smoking, drinking, illness symptoms at the time of blood sampling, and the relative prevalence of major leukocyte subsets within the PBMC sample, results continued to indicate a significant Group x Time interaction (b = .119 ± SE .040, t(9177) = 2.98, p = .003). In covariate-adjusted analyses, those randomized to MM continued to show a significant reduction in CTRA gene expression over time (b = −.066 ± .028, t(9177) = −2.31, p = .021) whereas the LKM group showed no significant change over time (b = .054 ± .029, t(9177) = 1.86, p = .063). Finally, because we observed significant differences in informal meditation practice by condition, we assessed whether those differences predicted significant gene expression change. Informal meditation did not significantly predict gene expression change when included as an additional covariate (b = −.003 ± .022, t(9176) = −0.12, p = .904), whereas condition remained a significant predictor (b = .121 ± .042, t(9176) = 2.89, p = .004). Thus, results were not substantively influenced by group differences in informal meditation practice.

We explored whether any main effects of randomized Group emerged on changes in depression, loneliness, forgiveness, meaning in life and mental health between T1 and T2. None were significant. (See OSM for details).

The present research had several limitations. First, without a waitlist control group, this work is limited in its ability to assess what changes in CTRA are due to time alone, without either meditation intervention. There was also moderate attrition, as this particular sample was difficult to recruit and maintain. Future work might overcome this obstacle by limiting potential barriers to participation such as paying for or providing childcare and transportation options. While no differences emerged between groups in the demographics of participants who left the study, it could also be that difficulty recruiting and maintaining our sample was partially due to the absence of efforts to tailor the intervention for a primarily low-income and racially diverse demographic. For example, a qualitative study exploring the cultural relevance of mindfulness meditation interventions for African American adults found participants felt mindfulness training helped with stress-management, yet thought the presentation should be adapted to incorporate more familiar spiritual or cultural practices, include more reading material from African American writers, and have overall more African American representation across instructors and participants (Woods-Giscombé & Gaylord, 2015). Future work should consider how adaptations to presentation and implementation to better fit socio-economic or racial demographics may enhance therapeutic value and participation in at-risk samples.

Finally, the present study limited the sample to low childhood SES, making it impossible to assess whether the effects of the intervention are specific to this group, or if the same findings would emerge in a sample with higher childhood SES. The initial study hypotheses were based on evidence suggesting relatively healthy mid-life adults showed improved biological outcomes in response to LKM compared to MM (Le Nguyen et al., 2019; Kok et al., 2013;). This suggests that the present findings may indeed be specific to this group. However, further research is needed to understand the potential differences in responses to meditation practices depending on one’s socio-economic background. Additionally, future research is needed to address whether the effect of MM on CTRA gene expression is comparable to that of a non-low-SES sample.

In the coming decades, rising income inequality coupled with health inequities may further widen the gap in mortality and longevity between low and high socio-economic groups. Altering the structural features of the environmental context, such as access to basic health care and an efficient social safety net, may help mitigate the impact of socio-economic-related stress on the development of detrimental biological residues. Yet at present, it remains imperative to understand whether and to what degree the health-relevant biological effects of having been raised in low SES households may be mutable with behavioral changes in midlife. Although further replication and understanding mechanisms of action are needed, this work presents preliminary evidence that mindfulness meditation may be a viable option for improving health in at-risk populations.