Global gene transcription profiles were assessed in circulating leukocytes from 14 individuals participating in the population-based Chicago Health Aging and Social Relations Study (CHASRS) 43. Individuals experiencing high levels of subjective social isolation were identified by scores in the top 15% of the UCLA Loneliness Scale, and an age-, gender-, and race-matched comparison group of individuals experiencing subjective social integration was defined by Loneliness scores in the bottom 15% 28. Table 1 provides demographic characteristics of each group, along with medical, behavioral, and psychosocial characteristics. This sample was composed of older American adults (median age 55 years), with diverse ethnic backgrounds (50% Anglo-American, 43% African-American, and 7% Hispanic/Latino), and varying socioeconomic status (median household income of $62,500 per year, range $25,000 to $150,000). A majority of study participants were female (78% female, 22% male), and measured levels of loneliness were stable over the 3 years prior to this gene expression analysis (1-year test-retest correlations averaged r = 0.90; intraclass correlation over all 3 years = 0.937, p < 0.0001). Sample selection procedures generated the expected difference in average levels of experienced social isolation (difference in UCLA Loneliness scores, p = 0.0002; Table 1). High-lonely participants did not differ from low-lonely controls on any of the medical or behavioral dimensions analyzed, but they were slightly younger (4 year difference, p = 0.011), had a lower household income (median $45,000 per year versus $87,500, p = 0.010), and reported higher levels of perceived stress (p = 0.023) and depressive symptoms (50% ≥ CEpidemiologic Studies Depression scale (CESD) cut-point of 10 versus 0% of controls, p = 0.038). These demographic and psychological differences were treated as potential confounders in subsequent analyses.

To assess leukocyte transcriptional alterations associated with chronic loneliness, we carried out global gene expression profiling using Affymetrix U133A high-density oligonucleotide arrays to survey 22,283 mRNA transcripts. Peripheral blood mononuclear cells were collected by antecubital venipuncture during the fourth or fifth yearly CHASRS visit. Figure 1 presents the 'transcriptional fingerprint' of subjective social isolation in human luekocytes, with green intensity indicating the magnitude of a gene's relative over-expression in high-lonely indivdiuals, and red intensity denoting the magnitude of relative under-expression. A total of 209 transcripts were differentially expressed, representing 144 distinct named human genes (listed in Additional data file 1). Of these, 78 (37%) were over-expressed in high-lonely individuals, and 131 (63%) were under-expressed, suggesting a net repressive effect on the leukocyte transcriptome (difference p < 0.0001 by binomial test).

Genes showing over-expression in leukocytes from high-lonely individuals included multiple transcription factors controlling cell growth and differentiation (EGR1, EGR3, FOSB, BHLHB2,EP400), as well as regulators of chromatin structure (the DNA topoisomerase TOP2B, the ARID1A and SMARCC1 components of the SWI/SNF chromatin remodeling complex, histone H2AFV, and the histone acetyltransferase MYST3). These findings suggest that social isolation might potentially influence basic gene-regulatory processes involved in cell growth and differentiation. Consistent with that hypothesis, over-expressed genes also included molecules promoting cell cycle progression (CDC25B, G0S2, MYBL1, DVL3, BTG2, ARFGEF2), enzymes involved in nucleotide and protein biosynthesis (PPAT, MTRR), cytoskeletal remodeling factors (DCTN1, KIF21B, RPH3A), and factors involved in processing and nuclear export of RNA (NXF3, HNRPL, DDX17, SFPQ). In the context of leukocytes, cell cycling plays a key role in immune activation and proliferation. Several other indications of immune activation emerged in the set of over-expressed genes, including pro-inflammatory cytokines, chemokines, granzymes, proteinases, and receptors for inflammatory mediators (IL1B, IL8, IL8RB, IL10RA, PTGDR,KLRC3,NKTR,GZMK, and multiple HLADR genes), as well as the master regulator of prostagladin synthesis, cyclo-oxygenase 2 (COX2/PTGS2). Additional indicators of immune cell activation that were over-expressed included the immediate-early response gene IER2, components of the insulin-like growth factor signaling pathway (IGF2R, IGFBP3), and activation-induced counter-regulators involved in pathway desensitization (TNFAIP, DUSP1, RGS1) and receptor shedding (MAN2C1, ADAM8, ARTS-1). GOstat bioinformatic analysis 44 confirmed the functional theme of a highly activated, proliferative phenotype in circulating leukocytes from socially isolated individuals by identifying over-representation of Gene Ontology (GO) annotations involving immune response and inflammation (GO:0009607, GO:0006952, GO:0006955; GO:0009613), stress response (GO: 0006950), antigen processing (GO:0019886, GO:0019884, GO:0045012), chemokine and cytokine activity (GO:0042379, GO:0008009, GO:0005125; GO:0001965), deoxyribonucleotide metabolism (GO:0009262), and nucleosome activity (GO:0000786) (all p < 0.05).

Analysis of under-expressed genes complemented the portrait of generalized immune activation in showing reduced expression of cell cycle inhibitors (CDKN1C, CSPG6, p15/PAF/KIAA0101, SPARC, FKBP5), apoptosis-related genes (MCL1, BIRC1, HSXIAPAF1, and TNFSF10/TRAIL), the antiproliferative cytokine TGF-β (TGFB1), and the NF-κB inhibitor Apo J (CLU). However, several specific dimensions of immune response showed selective repression against this generalized backdrop of immune activation. Genes involved in type I interferon response were down-regulated, including the signal transducer STAT1 and the interferon response genes 2',5'-oligoadenylate synthetase 1 (OAS1), interferon-stimulated gene 12 (IFI27), interferon-induced protein 6-16 (G1P3), and interferon-inducible proteins 15 and 44 (G1P2, IFI44). Also down-regulated were genes encoding immunoglobulin light, joining, and heavy region chains (IGLC2, IGLL1, IGK/IGKC/IGKGV, IGJ, IGLJ3, IGHG1, IGHM), B lymphocyte maturation markers (CD79B, CD269/TNFRSF17), and transcription fators involved in B cell differentiation (Ikaros/ZNFN1A1, POU2AF1). Thus, the circulating leukocyte complement in chronically lonely individuals shows broad genomic indications of immune activation and pro-inflammatory signaling accompanied by selective reductions in mature B lymphocyte function and innate antiviral responses.

Eight transcripts characteristic of the two key functional GO categories showing differential activity (immune activation and transcriptional control) were selected for independent verification of differential expression by real-time RT-PCR. Results were concordant with microarray indications in seven instances, with RT-PCR confirming over-expression of IL1B,IL8,PTGS2/COX2, FOSB, EGR1, CDC25B, and under-expression of G1P3 (MANOVA difference in all transcripts simultaneously, multivariate p < 0.0001, and p < 0.05 for each transcript analyzed individually; Additional data file 2). Both micorarray data and quantitative RT-PCR found the GR NR3C1 gene to be transcribed at comparable levels in leukocytes from high- versus low-lonely individuals (difference <4% by RT-PCR, p = 0.815, and <15% by microarray, p = 0.112). Expression of CD3 (CD3G, CD3D, CD3E, CD3Z), CD8 (CD8A, CD8B1), CD4, CD56 (NCAM1), CD19, and CD14 were also comparable (all differences <10% in magnitude, all p > 0.20). Thus, observed differences in mononuclear cell gene expression do not stem from variations in the relative prevalence of leukocyte subset mRNAs within the circulating mononuclear cell RNA pool 454647.

Data come from the CHASRS - a 5-year cohort analysis of psychological, social, economic, and biomedical outcomes in 230 English-speaking men and women aged 50-67 years at study entry 43. Annual study visits collected detailed biomedical, social, psychological, and economic assessments, including full medical histories, body mass index and waist circumference, diurnal measures of salivary cortisol, laboratory measures of autonomic nervous system activity, and psychometric measures of social network characteristics, stress, depression, and personality (detailed below). Health-relevant behavioral characteristics assessed include sleep quality, nutrition, exercise, memory and cognitive function. In 2005, 153 of the 166 remaining participants (92%) agreed to provide a 10 ml sample of peripheral blood for gene expression analysis. Gene expression analyses were carried out on samples from 14 individuals as outlined below. Procedures were approved by Institutional Review Boards at the University of Chicago and UCLA.

Subjectively experienced social isolation was assessed by the UCLA-R Loneliness scale 53 at each yearly study visit. Biological samples from 10 individuals who consistently scored in the top 15% of the loneliness distribution during study years 1, 2, and 3, and 10 individuals who consistently scored in the bottom 15% during years 1, 2, and 3, were selected for analysis after matching for age, gender, and ethnicity. Two samples from low-lonely individuals and four samples from high-lonely individuals yielded insufficient RNA for reliable gene expression assay, and analyses are thus based on fourteen individuals (eight low-lonely, six high-lonely). Objective social isolation was assessed by the social network index 54.

Demographic, social, psychological, and medical characteristics were assessed using established instruments 43, including measures of depression 55, perceived stress 56, hostility 57, body mass index, physician-diagnosed medical conditions, use of anti-inflammatory agents (including non-steroidal anti-inflammatories), diuretics, statins, beta-blockers, anti-diabetic agents, and anti-depressants. Participants were free of infectious disease symptoms, major medical conditions, and anti-inflammatory medications at the time of data collection. Samples were collected between 10 am and 2 pm (median = 11 am for both groups).

Total RNA was extracted from 10⁷ circulating leukocytes (mononuclear cells collected from a 10 ml antecubital venipuncture sample and isolated by ficoll density gradient centrifugation; RNAlater/RNeasy, Qiagen, Valencia, CA, USA), and 5 μg of the resulting RNA was assayed using Affymetrix U133A high-density oligonucleotide arrays 58 in the UCLA DNA Microarray Core as previously described 4159. Robust multiarray averaging 60 was applied to quantify expression of the 22,283 assayed transcripts, and differentially expressed genes were identified as those showing ≥30% difference in mean expression levels in samples from high- versus low-lonely individuals (corresponding to a FDR of 10%) 59. Functional characteristics of individual genes were identified through GO annotations, Gene References into Function annotations, and PubMed literature links retrieved through NCBI Entrez-Gene 61. Functional commonalities among multiple differentially expressed genes were identified using GOstat 62 with default stringency parameters (Benjamini FDR <0.10) 44. A full list of differentially expressed genes is provided in Additional data file 1, and raw data are deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO Series GSE7148).

Transcripts identified as differentially expressed in microarray analyses were independently assayed by quantitative real-time RT-PCR using TaqMan Gene Expression Assays (Applied Biosystems, Foster City, CA, USA): IL8 (Hs00174103_m1), CDC25B (Hs00244740_m1), IL1B (Hs00174097_m1), EGR1 (Hs00152928_m1), FOSB (Hs01547109_m1), PTGS2 (Hs00153133_m1), IFI6/G1P3 (Hs00242571_m1), GAPDH (Hs99999905_m1). We also verified microarray indications of equivalent expression of GR mRNA, NR3C1 (Hs00353740_m1). Assays for each sample (n = 14) were carried out in triplicate using an iCycler instrument (Biorad, Hercules, CA, USA), Quantitect Probe RT-PCR enzymes (Qiagen), and the manufacturer's recommended 1-step thermal cycling protocol. ISGF12 was assayed using published primer sequences and reaction conditions 33. Threshold cycle numbers for each analyte were normalized to GAPDH for analysis. A multivariate analysis of variance (MANOVA) tested group differences among all analytes simultaneously, and independent sample t-tests assessed differential expression of each individual transcript.

A 2-sample variant of TELiS 63 identified upstream signal transduction pathways driving differential gene expression 41. Primary analyses compared the prevalence of vertebrate V$GR_Q6 and V$CREL_01 TFBMs from the TRANSFAC database 64 in promoters of genes over-expressed in high- versus low-lonely individuals. Primary analyses utilized default scan parameters 41 and identified differences in TFBM prevalence using p values from an independent sample t-test with Welch's correction for heteroscedasticity 65. To assess the robustness of results to technical variations, sensitivity analyses examined parametric variations of promoter length (-300 bp relative to RefSeq transcription start site, -600 bp, and -1,000 bp to +200), and motif match stringency (MatSim = 0.80, 0.90, 0.95) 41. Sensitivity analyses also tested the impact of increasing stringency in the definition of differential gene expression (capture based on 1.5-fold change, corresponding to 5% FDR) 59.

Ancillary analyses utilized ANCOVA 65 to control for potential confounding between subjective social isolation (denoted x) and other demographic, psychological, or medical covariates (denoted c) that might influence gene expression. In these analyses, the profile of expression of each gene across participants (denoted y) was first adjusted to remove any variance attributable to a covariate (based on the general linear model y = c + e), and the residual covariate-adjusted gene expression profile (e) was then tested for significant difference as a function of social isolation (that is, in the general linear model e = x + z, where z represents residual error) 65. Genes identified as differentially expressed based on analysis of covariate-adjusted expression profiles were then analyzed as described above for differential prevalence of promoter TFBMs.

Reciprocal activity of NF-κB response elements and GREs was quantified by log-transformed prevalence ratios (NF-κB TFBM prevalence/GRE TFBM prevalence) in promoters of differentially expressed genes. Welch t-tests evaluated statistical significance of reciprocal bias (log ratio ≠ 0), and a Monte Carlo random sampling analysis assessed the probability that observed biases stemmed from an inverse relationship between GRE and NF-κB response elements within the cis-structure of the human promoter population (rather than the trans-activational activity targeted by TELiS). Monte Carlo analysis involved 10,000 cycles in which: 1.) a random set of genes equivalent in number to those found to be differentially expressed in socially isolated versus integrated individuals were sampled from the population of human genes assayed by the Affymetrix U133A array; and 2.) the NF-κB/GRE prevalence ratio was calculated for each random sample. The resulting sampling distribution of 10,000 null hypothesis ratios quantified the probability that a random set of human genes would generate a NF-κB/GRE prevalence ratio greater than or equal to that shown by the observed set of differentially expressed genes.

Following hypothesis-driven analyses of NF-κB/Rel and GR activity, additional exploratory analyses analyzed the prevalence of 190 other vertebrate TFBMs in the TELiS database 41, including parametric variations of promoter length and motif match stringency to evaluate the sensitivity of results to analytical parameters.

HPA axis activity was assessed by radioimmunoassay of cortisol concentrations in saliva samples collected immediately on waking (0 h), 30 minutes later (0.5 h), and immediately prior to evening repose (16 h), on 3 consecutive days during each of 3 study years 27. Log-transformed cortisol concentrations were analyzed using mixed effect linear models (SAS PROC MIXED; SAS Institute, Cary, NC, USA) treating time of day as a 3-level repeated-measures factor or a linear effect in hours post-awakening (0, 0.5, 16). Day (1-3) and year (1-4) were modeled as repeated measures, and individual differences in cortisol level and diurnal decline were modeled as a random effects nested within low- versus high-loneliness.

Systemic inflammatory biology was assessed by circulating levels of CRP. Blood spot samples obtained during study years 2, 3, and 4 were analyzed using a high-sensitivity enzyme immunoassay 66, and longitudinal observations on each participant were analyzed using a mixed effect linear model (SAS PROC MIXED) in a Year (2, 3, and 4, treated as a repeated measure) × Loneliness (high versus low) factorial design. Aberrantly high CRP values were identified by standard outlier detection algorithms (SAS PROC UNIVARIATE) and excluded from analysis because they were likely to reflect the effect of acute infection rather than chronic inflammation.