We screened the literature through the Gene Expression Omnibus for relevant peer-reviewed datasets based on organism type (Rattus norvegicus), expression profiling (microarray), sample type (brain hippocampal tissue), and condition (aging), using the search terms “aged” or “aging” or “old” or “young” and “brain” or “hippocampus.” No restrictions based on language and geographic origin were used in our search, while no exclusion criteria in the baseline characteristics of animals from which tissue sections were obtained, were applied. Duplicate gene expression sample and series or those lacking expression data for controls or with incompatible platforms, were excluded. Author (PG) formulated the search strategy and in conjunction with a second author (KP), the yielded datasets were screened and any discrepancies in the literature search were resolved.

Brain hippocampal samples from healthy young and aged rat models were compared and differentially expressed genes (DEGs) were retrieved using ImaGEO (Toro-Domínguez et al., 2019). Significance analysis and integration were conducted using the fixed effect model to identify those DEGs with the strongest average effect across the collected datasets. This approach was followed as both the treatment (i.e., aging) and expression array platforms were orthogonal between datasets. Genes with a P < 0.05 corrected by the Benjamini-Hochberg (BH) false discovery rate (FDR) were regarded as significant. DEGs with Z score > 1.96 were classified as upregulated, while those with Z score < 1.96 as downregulated (corresponding to a significance level of 5%).

Encoded hippocampal proteins of the aging DEGs and their interactome were predicted into a protein-protein interaction (PPI) network using The Search Tool for the Retrieval of Interacting Genes via a medium probabilistic confidence score of >0.4 and mapped with Cytoscape (Szklarczyk et al., 2019). A reasonably moderate cut-off score was ensued to amplify the coverage of all potential protein interactions without inflating their precision. Proteins lacking any interactions were excluded from the network.

Central gene elements of the aging rat hippocampal network were inferred by measuring network features from their complex interactome and by identifying sub-networks and hub objects (Giannos et al., 2021a,b, 2022). Highly clustered DEGs or densely connected modules in the PPI network were identified using the molecular complex detection (Bader and Hogue, 2003). Cut-off selection was ensued based on the default network scoring parameters: degree cut-off = 2, haircut cluster finding, node score cut-off = 0.2, K-core = 2, and max depth = 100. The interactome interference of module DEGs in the PPI network was quantified using CytoHubba (Chin et al., 2014). Module DEGs were ranked based on the intersection of 11 established topological algorithms, as previously described: Degree, Closeness, Betweenness, Radiality, Stress, EcCentricity, BottleNeck, Edge Percolated Component, Maximum Neighborhood Component, Density of Maximum Neighborhood Component, and Maximal Clique Centrality.

Enrichment of aging DEGs in the rat hippocampus was predicted using Gene Ontology (GO) annotations from the Molecular Signatures Database (Liberzon et al., 2015). Functional classification was ensued using a probability density P < 0.05 following BH FDR correction and categorized into three groups of GO terms: biological process (BP), molecular function (MF), and cellular component (CC).

The search of the GEO database resulted in 4811 datasets, of which 3981 contained gene expression data from organisms other than rats. From these, 111 microarray datasets were retrieved after exclusion of 717 datasets which included non-profiling by array expression data from samples other than tissues. Further exclusion of 106 studies with either duplicate gene expression samples or series and incompatible gene expression platform resulted in 5 eligible datasets: GSE14505 (Nagahara et al., 2009), GSE20219 (Blalock et al., 2010), GSE14723 (Haberman et al., 2011), GSE14724 (Haberman et al., 2011), and GSE14725 (Haberman et al., 2011; Figure 1). The retrieved datasets consisted of 38 young (5–8 months) and 29 aged (21–26 months) male rat hippocampal tissue samples (Supplementary Table 1).

Integration analysis revealed a total of 527 aging DEGs, of which 278 showed significantly increased hippocampal expression while 249 DEGs decreased expression in the rat brain (Supplementary Table 2).

A PPI network of DEGs in the aging rat hippocampus was constructed and consisted of a total of 474 encoded proteins with 1,862 interactions. A single highest-ranked clustering module was retrieved and consisted of 19 upregulated aging DEGs with 362 interactions among 92 genes: allograft inflammatory factor 1 (AIF1), complement C1q A chain (C1QA), complement C1q B chain (C1QB), complement C1q C chain (C1QC), CD53 molecule (CD53), CD74 molecule (CD74), colony stimulating factor 1 receptor (CSF1R), cathepsin S (CTSS), Fc epsilon receptor Ig (FCER1G), Fc gamma receptor IIa (FCGR2A), Fc gamma receptor IIb (FCGR2B), Fc gamma receptor IIIa (FCGR3A), fibrinogen like 2 (FGL2), integrin subunit beta 2 (ITGB2), lysosomal protein transmembrane 5 (LAPTM5), macrophage expressed 1 (MPEG1), pleckstrin (PLEK), triggering receptor expressed on myeloid cells 2 (TREM2), and transmembrane immune signaling adaptor TYROBP (TYROBP; Figure 2, Table 1, and Supplementary Table 3). Multi-algorithmic topological analysis revealed MPEG1 as the highest ranked hub gene (Supplementary Table 4). DEGs in the highest-ranked clustering module were most enriched with immune response (BP, P = 1.71E-17), immunoglobulin G binding (MF, P = 1.92E-08), and intrinsic component of plasma membrane (CC, P = 1.25E-06; Supplementary Table 5).

This is the first study that comprehensively examined the potential role of DEGs and their interactome as gene biomarkers in hippocampal aging using 5 publicly available datasets with almost 70 included rat samples and using a multi-algorithmic protein-interaction based functional approach that was dependent on different levels of annotation. Nevertheless, certain conceptual and methodological limitations exist. Hippocampal tissue distribution from which gene expression was derived in the included studies was sparsely heterogeneous, ranging from whole hippocampus samples to those from dorsal and ventral areas, the CA1/CA3 regions or the dentate gyrus. Although histologically distinguishable subfields, these likely share common genomic organizations and thus, aging related accumulation of DNA damage from these subregions may be overlapping (Van der Meer et al., 2020). Moreover, despite that the expression profiling platforms employed by the included datasets were indifferent, lab effects due to experimental variation are known to hinder statistical power in DEGs detection, a challenge which remains even after normalization (Johnson et al., 2007; Roberts et al., 2011; Hansen et al., 2012; Dillies et al., 2013; Lyu and Li, 2016). Yet, an attempt to address unknown cross-study heterogeneity was made by integrating gene expression rather than considering their overlap between datasets.