Network topology and evolution of the gene co-expression of T-cells during immuno-senescence

To better understand the potential impact of the gene expression network structure on the dynamics of immune-senescence and defects of cell functions during aging, we investigated network structures in both young and old individuals. We analyzed the gene co-expression networks (GCNs) derived from an aging signature of 130 immune-related genes obtained from CD3+ T-cell splenocytes extracted from FVB/N, C57BL/6N, and BALB/c mice at ages 2 and 22- 24 months. The network structure for the two different mouse age-groups was derived and subsequently analyzed. Analysis of network hubs using clustering coefficients, degree, betweenness, eigenvector, and closeness centralities, as well as local, indirect, and total influence measures, demonstrated changes in gene behavior and network control between the two age groups. Our quantification shows that the young, 2-month old mouse network is more organized than the 22-24-month, old mouse network, while the network structure of the older mouse GCN appears to be far more complicated but far more dispersed. Changes in network structure between the old and young mice suggest deterioration in transcription regulation with age in peripheral T- cells, particularly within the TCR signaling pathway, and potential compensatory mechanisms in older T-cells to overcome loss to regular function resulting from transcriptional irregularity. These results demonstrate the need for more research into gene co-expression in peripheral T-cells in order to better understand both network irregularities and the phenotypic dysfunction observed in older individuals. Author Summary In order to better understand the potential mechanisms of transcriptional irregularities in the immune system with aging, we analyzed the structure of gene co-expression networks of T-cells extracted from the spleens of 2 and 22-24-month old mice. Gene co-expression describes the correlation relationship between two expressed genes; as the expression of one gene goes up, the expression of another gene might also increase (or, conversely, decrease). Strong gene co- expression relationships can signal the existence of a number of important biological phenomena, such as two genes belonging to a transcription pathway or protein structure. Network diagrams visualizing these co-expression relationships in both younger and older mice demonstrated the existence of differences in network structure and properties that may be attributed to the aging of the immune system. Network mathematical methods were used to examine the complexity of each network. We found that the younger mouse network was more organized than the older mouse network. The older mouse group exhibited a 255% increase in co-expression relationships but a decrease of 92% of the connections from the young mouse network. This suggests the older mouse T-cells suffer dysfunction at a transcriptional level. This results in the loss of regular immune and cellular functions. These results demonstrate the importance of future research into gene co-expression to decipher senescence or diseases that perturb gene expression through time.


Introduction
To decipher gene expression modifications across aging, we performed a transcriptomic 153 profiling of CD3 + purified splenocytes from 12 young mice age 2 months and from 11 old mice 154 ages 22-24 months. The ICA/GSEA method was used to identify a signature of 130 genes able to 155 distinguish young from old mice. Interestingly, the age-related gene signature allows us to segregate the 3 mice strains, considering only the young mice. However, this signature cannot related to known intracellular pathways expressed in the sampled T-cells. S1 Table lists all of the 180 gene transcripts found in both of the networks, along with known human orthologs and known 181 gene functions. Network files featuring the genes that appear in each network, as well as their 182 corresponding co-expression relationships, can be found in S2 Table ( Table 2. A t-test demonstrated a statistically significant difference between the 209 mean edge counts of the two age-groups (p < 0.001).    adjacency matrix (red squares), the old adjacency matrix (blue squares) and elements that are 241 conserved for both age-groups (green squares). Note that if you count the number of green squares 242 above the diagonal, this is exactly seven which is how many conserved gene pairs are given in 243   Table 3.

210
Consequentially, because only seven edges were conserved across the aging networks, 245 most network edges were lost or gained with age. While the older mouse GCN saw an increase 246 in total edges, 92% of the edges from the 2-month old GCN were not present in the 22/24-month 247 old GCN, indicating 92% of the co-expression relationships established in young mice are absent 248 in older mice. We identified 47 nodes present in the young network but whose co-expression 249 relationships were lost with age in the senescent T-cells (Fig 5). Nodes in the older mouse GCN     Table 4 illustrates the top ten gene-transcripts with the highest degree centrality in both  Table). In 2-month mice the degree centrality of TCR related genes is 6-10, while for the 287 22/24-month it doubles to 14 to 15, meaning that the TCR from senescent T-cells establish more 288 links with other molecules. This suggests that the molecular relation is less specific (degeneracy), 289 increasing the entropy of the system and decreasing the ordered organization established during 290 development. Because degree centralities provide an overall view of the network connectivity, it is reasonable to hypothesize the fact that the 22/24-month old network is three times larger than 292 the 2-month old network. We do note that network size differences may contribute to some of the 293 overall differences in degree centrality differences. Eigenvector centrality. Eigenvector centrality is another measure of the influence of a given 300 node in a network. A high eigenvector centrality means that the given node is connected to many 301 other nodes who themselves have high eigenvector centrality scores. You can think of this as 302 follows: if a node is a big shot, then its high eigenvalue centrality provides a measure of how 303 many other big shots is it connected to.

294
304 Table 5 lists the top 10 network nodes with respect to their eigenvalue centrality value.

305
These genes represent the top 10 genes that are thought to exert general control over the whole 306 network. It is straightforward to observe that the two GCNs showed clear differences in hub control can be found in S1 Table. node is near to all of the other nodes in a network. In Table 6 we provide the closeness 322 centralities for the top 10 genes in both networks. First, we note that there are no common genes 323 across the two networks. However, the functionalities of certain high ranking genes are 324 conserved across the networks. For example, the first rank gene in the 2-month network is gene 325 LOC386545 which is similar to T-cell receptor beta chain VNDNJC precursor (S1 Table).

326
Meanwhile first rank gene in the 22/24-month old network is A130082M07Rik, a gene that 327 functions as the T-cell receptor alpha chain variable 9D-3 region (S1 Table). This change in the 328 top-ranked closeness centrality hubs suggests differences in T-cell clonal expansions in the two 329 age-related groups, as has been previously described [24]. The functions of other high-ranking 330 genes can be found in S1 Table. 331 Table 6. Top 10 highest closeness centralities for core the young and old mouse networks. between each other. One can say that betweenness centrality represents the control a specified 335 node has over the network in that it finds nodes that act as bridges between nodes and -in a  Table 7 lists the top ten gene betweenness centrality values for both networks. Betweenness has the highest ranking in the older mouse network. Other high-ranking gene information can be 346 located in S1 Table. 347 Histone 3 (S1 Table). The gene with the highest eccentricity in the 2-month network is Whsc1l1,  Stress of a node in a biological network is defined as the number of shortest paths passing through 367 a given node. It can be used to indicate the prominence of a gene in holding together connecting 368 regulatory genes in a network pathway. The higher the value, the more importance the node has in 369 holding together communicating nodes [59]. We find that stress in the top ten highest stress nodes 370 in the 22/24-month old group appears much higher than the highest stress nodes in the 2-month 371 old group (Table 9). lower proportion of shortest paths in the total network as compared to the 2-month old group 379 ( Table 9). As a result, the genes with high stress values in the 2-month old group can be seen to highest rank ordered influence values exclusive to each network can be found in S1 Table. 398  highest clustering coefficients in the 2-month and 24-month core networks can be found in Table   430 13. None of the genes with the highest clustering coefficients in the 2-month network were in the 431 highest ranked genes for the 22/24-month network. network has a wider range of clustering coefficients compared to the 2-month network.

437
Meanwhile, the 2-month network clustering coefficients are more compactly distributed. This 438 apparent increase in highly interconnected genes in the old mouse GCN is consistent with the 439 observed large core network that makes up the old mouse GCN (Fig 3). of genes involved in these subnetworks can be found in S1 Table. 453 454 Centrality Analysis of Conserved Gene Pairs. Finally, we summarize the rankings of the 455 conserved co-expressed gene pairs (see Table 3) in variable. The braces correspond to the nodes (genes) at the end of the edges in Table 3. We note  Tregs.

542
In the 2-month old network, TCF7 falls in a three-node island with genes Lck and CD28e. related not only to activation (TCR/CD3/CD69), but also binding and adhesion/interaction 563 (VCAM1). Pathway enrichments for the genes appearing in both networks also suggest these genes 564 play a role in changes in the immune system with age (Fig 2). like 1) appears across all centrality measures, except for clustering and eccentricity, in that it 591 always ranks 1-2 for the young mouse cells but not for the old mouse cells (Table 14). This makes providing opportunities to approach these genes as potential therapeutic targets to help aging patients. Human orthologs of the genes used in this analysis (S1 Table)  an important factor is that the AGEMAP study did not look at co-expression in peripheral T-cells, 630 which may behave differently from other cell types across the aging process. Indeed, T-cells are 631 permanently selected to survive and to divide through the TCR complex and co-stimulation 632 pathways, or they undergo apoptosis if too many default signals occur (as during the cell cycle).

633
Results by Vibert & Thomas-Vaslin [24] show that across ages and genetic backgrounds, immature