Database mining and study selection
We searched the Gene Expression Omnibus (GEO), BaseSpace Correlation Engine (BSCE, Illumina, Inc, USA), and ArrayExpress databases to identify microarrays from subjects with known loneliness or social isolation (See Methods). The database search retrieved the following arrays: GEO = 397, ArrayExpress = 6, BSCE = 6. One array, GSE80696, containing transcriptomic data from individuals with known loneliness or social isolation met our inclusion/exclusion criteria and was analyzed further. The overall bioinformatic workflow is presented in Fig. 1.
Identification Of Switch Genes Associated With Loneliness
The dataset GSE80696 was imported into SWIM to identify switch genes in the nucleus accumbens of individuals with known loneliness. The SWIM analysis was performed using the following comparisons: all subjects (high vs. low loneliness), males (high vs. low loneliness), and females (high vs. low loneliness).
The SWIM algorithm was performed as described previously 10, 18, 19. In the first step, genes were included (red bars) or discarded (gray bars) using a cut-off of 2.0 or higher (Fig. 2A). The gene matrix was imported into SWIM to build the loneliness gene correlation network based on the average Pearson correlation coefficient (APCC). Using the APCC, three hubs are defined; date hubs with low positive co-expression with their partners, party hubs with high positive co-expression, and the fight-club hubs with negative APCC values (Fig. 2B). Two parameters identify the plane: Zg (within-module degree) and Kπ (clusterphobic coefficient), and it is divided into seven regions, each defining a specific node role (R1-R7). High Zg values correspond to hubs nodes within their module (local hubs), whereas low Zg values correspond to nodes with few connections within their module (non-hubs within their communities, but they could be hubs in the network). Each node is colored according to its APCC value. Yellow nodes are party and date hubs, positively correlated in expression with their interaction partners. The switch genes are characterized by low Zg and high Kπ values and are connected mainly outside their module. The switch genes are the blue nodes in region R4 (Fig. 2B).
A heat map of the expression of switch genes is presented in Fig. 2C. The data is clustered according to rows and columns representing the switch genes and samples, respectively. The samples depicted in red are from subjects with high loneliness. Most switch genes identified in individuals with high loneliness were downregulated (shown in blue). In contrast, those with low loneliness were upregulated (shown in yellow) (Fig. 2C). Fight-club hubs differ from the date and party hubs, and switch genes are significantly different from random, confirming the analysis’s robustness (Fig. 2D). The x-axis represents the cumulative fraction of removed nodes. In contrast, the y-axis represents the average shortest path. Each curve corresponds to the variation of the average shortest path of the correlation network as a function of removing nodes specified by the colors of each line.
SWIM analysis identified 48 switch genes in postmortem nucleus accumbens from individuals with high vs. low loneliness. The same analysis was performed by stratifying samples by sex and level of loneliness. This analysis yielded 27 switch genes in males with increased loneliness compared to low (Supplementary Fig. S1). Switch genes from males with high loneliness depicted with red bars were downregulated (shown in blue) compared to males with low loneliness (Supplementary Fig. S1). Analysis of samples from females did not yield any switch genes.
Biological And Functional Analysis Of Switch Genes
Functional associations were explored for each switch gene using the gene ontology annotations from the HUGO database. Gene ontology analysis revealed that some switch genes from individuals with high loneliness associated with angiogenesis and hemostasis (SERPINA1, FN1, KLK3, COL4A1), innate immunity, and inflammation (CD59, GPNMB, LILRA2, NECTIN2, UBE2V1), lipid metabolism (GPR3, SRD5A1, ACSL5, ACACB), neuronal development and function (ARF1, FN1, DPP10, DMD, TH, SYT8, FOXN4). In males with high loneliness, switch genes were involved in the inflammatory response (HAMP, ZFP36, PELI1, CXCL1, HLA-DRB5, S100A8, SPN) and regulation of transcription (NPAS3, AGO2, FOS, TAF6). The list of switch genes and their gene ontology annotations is provided in Supplementary Table S2.
Network and pathway analysis revealed 12 unique dysregulated pathways associated with loneliness (Fig. 3a and b, Supplementary Table S3). The top dysregulated pathways implicated in loneliness were adherens junctions, TGF-β, FOXO, Hippo, PI3K-AKT, WNT, AGE-RAGE, acute myeloid leukemia, microRNAs in cancer, JAK-STAT, and endometrial cancer (Fig. 3b). Network and pathway analysis identified 18 unique dysregulated pathways associated with loneliness in males (Fig. 3c and d, Supplementary Table S3). The male pathways were predominantly associated with infection, innate immunity, cancer-related pathways, and autoimmune diseases (Fig. 3d). Venn diagram analysis indicated that 15 pathways were shared between both groups. The complete list of pathways is provided in Supplementary Table S3.
Gene-disease Association Analysis
A gene-disease association network analysis was performed in NetworkAnalyst. Switch genes obtained from lonely individuals were connected to 16 diseases, including cancer, liver cirrhosis, HIV infection, bipolar disorder, depression, schizophrenia, and mental retardation (Fig. 4a, Supplementary Table S4). Male-specific switch genes were connected to 6 diseases, including liver cirrhosis, cocaine-related disorders, alcoholic intoxication, mammary neoplasms, and hypersensitivity (Fig. 4b, Supplementary Table S4).
Gene-transcription Factor Network Analysis
Transcription factor analysis of loneliness-related switch genes identified 65 master regulators. The most significant transcription factors according to degree and betweenness centrality were IRF1 and TGIF2 (Supplementary Table S5). Analysis of loneliness switch genes from males identified 41 transcriptional regulators. The most significant transcription factors based on network topology measurements were KLF9 and ZFX (Supplementary Table S5).
Loneliness-related Switch Genes Associated With Neuropsychiatric And Neurodegenerative Diseases
We investigated whether loneliness-related switch genes were involved in neuropsychiatric and neurodegenerative diseases. To this end, we compared the results from this study to our previous analyses of switch genes in AD, FTD, ALS, and physical activity 12, 13, 17. Several loneliness-related switch genes were identified as switch genes in different neurodegenerative diseases. For instance, ACACB and DLEC1 were identified as switch genes in the entorhinal cortex of AD patients 12. Further, ACACB and GPNMB were identified as switch genes in the frontal cortex of FTD patients 13.
We next manually curated the literature to explore the associations between loneliness-related switch genes and brain diseases. Specifically, we used the search terms “neurodegeneration”, “dementia”, “Alzheimer’s disease”, “Parkinson’s disease”, “Frontotemporal dementia”, Amyotrophic lateral sclerosis”, “Lewy body dementia”, “neuropsychiatric disorders”, “major depressive disorder”, and schizophrenia” for each switch gene individually. This search identified the association of 25 switch genes with neurodegenerative and neuropsychiatric diseases. For instance, 13 switch genes, GPNMB, TH, CD59, COL4A1, ZBTB16, TSPAN15, DMD, LEF1, GPR3, UBE2V1, DPP10, NECTIN2, LGALS3BP, CDKN1A, SERPINA1, and DMP1 were linked to AD, PD, FTD, PD dementia, Creutzfeldt Jakob disease, and LBD (Table 1). Nine switch genes from the male dataset, AGO2, HLA-DRB5, ALDOA, S100A8, CTSG, CXCL1, CYTH4, PELI1, and FPR1 associated with AD, HD, PD, LBD, and FTD. Five switch genes, HLA-DRB5, CYTH4, NPAS3, DMD, and DPP10 related to neuropsychiatric disorders (Table 1).
Comparative Gene Correlation Analysis Between Loneliness And Neuropsychiatric And Neurodegenerative Diseases
Given the associations between loneliness and brain diseases, we performed a correlation analysis between the list of switch genes identified from chronically lonely subjects (GSE80696) and the most common neuropsychiatric and neurodegenerative diseases using the gene correlation tool in BSCE. We used the search terms “Alzheimer’s disease”, “dementia”, “Parkinson’s disease”, “major depressive disorder”, “depression”, and “schizophrenia”, to identify gene expression datasets. Studies were filtered only to include human studies.
In the context of neurodegeneration, the correlation analysis in BSCE showed that loneliness-related switch genes significantly overlapped in 82% (53/65) of human studies on AD deposited in BSCE (Table S6). The most significant genetic overlap was observed in studies from early-stage AD patients' entorhinal and frontal cortex pyramidal neurons. Specifically, 23 (p = 1.60E-09) and 36 (6.10E-07) switch genes overlapped in the entorhinal and frontal cortices, respectively. Furthermore, several loneliness-related switch genes were associated with variants previously identified as risk factors for AD in 30 different GWAS studies (Supplementary Table S7). For instance, variants in BCAM and NECTIN2 have been related to the risk of AD in more than 10 GWAS from diverse populations, including European, Caucasian, and Japanese (Supplementary Table S7). NPAS3, RBM38, and PELI1 were associated with the risk of AD in APOE4 (-) individuals and AD with psychosis in a European cohort. Finally, ASGR2 is associated with response to cholinesterase inhibitors in discovery and replication cohorts of AD individuals (Supplementary Table S7).
The same correlation analysis was performed with PD studies. Loneliness-related switch genes significantly overlapped in 68% (40/59) of human studies on PD (Table S8). The most significant genetic overlap was observed in the globus pallidum internal in PD patients with 12 overlapping switch genes (p = 2.50E-06). Several switch genes overlapped with known risk loci in PD patients. For example, variants in NPAS3, HLADRB5, ALDOA, and GPNMB have been linked to PD risk in several populations (Supplementary Table S9).
In the context of neuropsychiatric diseases, loneliness switch genes overlapped in 70% (16/23) of human studies on the major depressive disorder (Supplementary Table S10). Nine switch genes, HLA-DRB5, ARHGAP15, COL4A1, RBM38, DMD, LGALS3BP, WSCD2, CYTH4, and CNTRL, overlapped with known genetic variants in depression (Supplementary Table S11). Similarly, switch genes significantly overlapped in 64% (16/25) of human studies in schizophrenia (Supplementary Table S12). Seven switch genes, NPAS3, ARHGAP15, LGALS3BP, DPP10, SMYD3, CPXCR1, and HLA-DRB5, were associated with known risk factors for schizophrenia (Supplementary Table S13).