Functional pathways regulated by microRNA networks in CD8 T‐cell aging

Abstract One of the most prominent immunological changes during human aging is the alteration in CD8 T‐cell subset distribution, predominated by a loss of naïve CD8 T cells. The molecular mechanisms that contribute to the loss of naïve CD8 T‐cells during aging remain unclear. Considering that many CD8 T‐cell functions are influenced by microRNAs (miRNAs), we explored miRNA expression profiling to identify novel dysfunctions that contribute to naïve CD8 T‐cell loss during aging. Here, we describe age‐dependent miRNA expression changes in naïve, central memory, and effector memory CD8 T‐cell subsets. Changes in old naïve CD8 T‐cells partially resembled those driven by an underlying shift in cellular differentiation toward a young central memory phenotype. Pathways enriched for targets of age‐dependent miRNAs included FOXO1, NF‐κB, and PI3K‐AKT signaling. Transcriptome analysis of old naïve CD8 T‐cells yielded corresponding patterns that correlated to those seen with reduced FOXO1 or altered NF‐κB activities. Of particular interest, IL‐7R expression, controlled by FOXO1 signaling, declines on naïve CD8 T cells with age and directly correlates with the frequencies of naïve CD8 T cells. Thus, age‐associated changes in miRNA networks may ultimately contribute to the failure in CD8 T‐cell homeostasis exemplified by the loss in naïve cells.

Previous studies have identified miRNAs as important regulators of effector CD8 T-cell responses and of memory differentiation (Wu et al., 2012;Zhang & Bevan, 2010). Indeed, global knockouts of miR-NAs caused by deletions in the miRNA-processing pathway affect both effector function and proliferative capacity of CD8 T-cells after activation (Trifari et al., 2013;Zhang & Bevan, 2010). Targeted approaches have also linked specific miRNAs with CD8 T-cell transition from effector to memory phenotype (Khan, Penny, Yuzefpolskiy, Sarkar, & Kalia, 2013;, with the ability of CD8 T cells to clear viral infections (Wang et al., 2013), with the accumulation of terminally differentiated effector memory CD8 T cells (Brunner et al., 2012) and with dysfunctional T-cell receptor signaling in the elderly (Li et al., 2012). During early development, miRNA expression changes in neonatal CD8 T-cells associate with inability of neonatal mice to develop memory T cells (Wissink, Smith, Spektor, Rudd, & Grimson, 2015). Moreover, multiple age-associated miRNA differences have been found in terminally differentiated effector CD8 T-cells from humans (Hackl et al., 2010;Teteloshvili et al., 2015). However, whether naïve CD8 T-cell subsets display altered miRNA expression during human aging and whether age-associated miRNAs influence the functionality of these CD8 T cells has not been investigated.
In this study, we examined the age-associated heterogeneity of miRNA expression in highly purified naïve, central memory, and effector memory CD8 T-cell subsets and the functional outcomes of these changes in the naïve CD8 T-cell compartment. We identified multiple subset-specific miRNA expression changes with age. Notably, age-dependent miRNA alterations in the naïve CD8 T-cell compartment partially reflected that of cellular differentiation toward a central memory phenotype. Moreover, global network analysis of these age-dependent miRNA changes in naïve CD8 T-cells implicated a potential role for the FOXO1 signaling pathway in defective naïve CD8 T-cell homeostasis during human aging.

| Age-dependent alterations in miRNA expression across CD8 T-cell subsets
We initially addressed whether aging caused miRNA changes in naïve and memory CD8 T-cell subsets in a cohort of young (<30 years, n = 15) and older (≥65 years, n = 9) adults. The median age of the young cohort was 21 years (range: 16-28). The median age of the older cohort was 70 years (range: 65-82). There was no statistical difference in the distribution of gender (p = 0.403) between the two cohorts (gender: 33% female in young and 56% female in older). As expected, there was a trend for higher CMV positivity in the older cohort (p = 0.09; 27% CMV positive in young vs. 67% in old).
As miRNA studies on highly purified aged CD8 T-cell subsets are limited, we sorted naïve, central memory, and effector memory subsets from peripheral blood of these healthy adults (Figure 1a). Consistent with previous studies, we found reduced frequencies of naïve CD8 T cells and increased frequencies of the memory CD8 populations in older individuals (Figure 1b). Equal numbers of naïve and memory cells were simultaneously screened for expression of 34 preselected, immune-related miRNAs by multiplex microfluidic qPCR.
We then compared miRNA expression levels between the three cell subsets in young and older individuals. miRNA expression levels relative to RNU48 are listed in Supporting Information Table S1. We found age-dependent miRNA differences in all three subsets (Figure 1c). Notably, most of the miRNA differences were unique to the individual subset, with a majority of these age-dependent miRNA changes found within the naïve compartment ( Figure 1d). One out of the 12 age-dependent changes was found in all three subsets; this change being a reduction of miR-181a in older individuals. miRNAs specific to naïve CD8 T-cells included miR-146a, miR-155, miR-142, and miR-7. miR-146a (increased) and let-7f (decreased) were the most significantly altered miRNAs in the naïve population ( Figure 1e). miR-155 and miR-7 were both expressed close to the level of detection and were not detected in some individuals giving the appearance of a bimodal distribution. Naïve and CM CD8 compartments shared a significant decrease in let-7f expression with age. CM and EM compartments shared increased miR-125a expression. Naïve and EM compartments shared no unique age-related miRNA changes.
Thus, we identified multiple miRNA expression changes in CD8 T cells that are subset-and age-dependent.

| CMV status does not affect miRNA expression in CD8 T-cell subsets
Many molecular and cellular changes in CD8 T cells during aging have been linked with CMV infection. In our cohort, 26.7% (4 out of 15) of young donors and 66.7% (6 out of 9) of old donors had a positive CMV serology. Therefore, we investigated whether underlying CMV infection influenced miRNA expression in CD8 T-cell subsets and contributed to age-dependent miRNA changes. However, no significant differences in miRNA expression were found between CMVnegative and CMV-positive individuals in the naïve and effector memory compartments (Figure 2a). The central memory compartment showed slightly decreased miR-16 expression in CMV-positive individuals, but this miRNA did not overlap with differences observed in aging (Figure 2b). When further separated by age and CMV status, we again find no differences in miRNA expression, including miR-146a, let-7f, and miR-181a in naïve CD8 T cells (Figure 2c). Effector memory cells also showed little miRNA differences with CMV and age separation, including miR-125a and miR-93 that significantly change with age in EM populations (Figure 2d).
Although these analyses have small sample numbers, power was sufficient to detect differences of about two standard deviations (SD).
Specifically, for comparing 4 young CMV+ with 10 young CMV−, we have 80% power for detecting a difference of 1.86 SD based on two-sided t test. Similarly, for comparing three old CMV− with six old CMV+, we have 80% power for detecting a difference of 2.38 SD. One young donor was removed from the analysis, as there was no CMV status available. Overall, these data suggest that CMV infection does not significantly influence age-dependent miRNA changes in naïve, central memory, or effector memory CD8 T-cell subsets.

| Detection of differentiation-dependent miRNA signature in CD8 T-cells
One possible driver of age-dependent changes is cellular differentiation. To determine differentiation-driven miRNA changes, we first analyzed miRNA differences between naïve and memory populations in young adults. Analysis of our select panel of miRNAs identified multiple miRNA expression changes with differentiation (13 miRNAs, 32.5%), with a majority of miRNAs having increased expression in the memory populations ( Figure 3a). Central and effector memory populations displayed no distinct miRNA expression differences.
Comparing the expression differences between naïve and memory populations in young adults, we identified a group of nine miRNAs commonly changed during differentiation to central as well as effector memory cells; we termed these miRNAs as core memory miRNAs ( Figure 3b). Core memory miRNAs included miR-120a, miR-92a, and miR-146b, which are down-regulated with differentiation, and miR-16, miR-21, miR-29b, miR-146a, miR-155, and miR-301, which are up-regulated with differentiation, albeit at different levels of expression ( Figure 3c). Previous studies on miRNA profiles during CD8 T-cell differentiation also identified that miR-21, miR-146a, and miR-155 increased in the memory population with a coinciding decrease in miR-92 and miR-146b in humans (Salaun et al., 2011), suggesting these miRNAs are highly robust markers of human naïve vs. memory CD8 T cells.
We found that PC1 (32.9% of variation in dataset) was driven by age-dependent changes, seen by differences in expression between young and old (Young Na vs. Old Na [p = 0.007]; Young CM vs. Old . PC2 (25.5% of the variation) showed differences between young and old cells as well as between cell subsets, suggesting a role for differentiation in this secondary variation. No age difference was observed between young and old effector memory populations, consistent with these cells being closer to end differentiation than naïve and central memory T cells.
To understand how the individual miRNAs are driving the distribution across PC1 and PC2, we overlaid their corresponding vectors onto the PCA plot. We found that PC1 variation was driven mostly by miR-7, miR-142, miR-181a and let-7f, whereas PC2 was driven by miR-146a and miR-155 ( Figure 4c). Individual expression plots of these miRNAs confirmed strong relationships with age for miR-181a and let-7f whereas miR-146a and miR-155 were influenced by cellular subset ( Figure 4d). Notably, expression of miR-146a showed a stepwise increase from young naïve to old naïve to young central memory.
It is possible that the increase of miR-146a in aged naïve CD8 T Graph shows mean ± SEM distinguish from naïve T cells (Akondy et al., 2017;Fuertes Marraco et al., 2015;Gattinoni et al., 2011). Thus, we next interrogated single cell expression of miR-146a in young and old naïve CD8 T-cells utilizing flow cytometry in combination with in situ hybridization for miRNA detection. We found that expression of miR-146a followed a unimodal distribution with the entire population of aged naïve CD8 cells shifted toward higher miR-146a (Figure 4e), demonstrating that increased miR-146a expression in aged naïve CD8 T cells was not driven solely by a naïve-like memory subpopulation. We also find an overall decrease in miR-181a in aged naïve CD8 cells using the same detection method (Figure 4f). Therefore, the expression pattern of these six miRNAs in naïve CD8 T cells is unique to aging and not driven by contamination of "virtual" naïve cells, such as stem cell-like memory. These data also reveal that differentiation of the naïve compartment partially, but not solely, accounts for the miRNA expression profile in old naïve CD8 T cells.

| Age-dependent miRNAs in naïve CD8 T-cells predict FOXO1 signaling defect
miRNAs primarily function as posttranscriptional regulators, by binding the mRNA and inducing mRNA degradation or preventing protein translation. However, miRNAs have hundreds of different mRNA targets and determining the function of multiple miRNA changes at a cellular level is extremely challenging. As a result, studies are often limited to one-on-one miRNA and mRNA interactions. Here, we utilized a computational tool (i.e., DIANA mirPATH; see Section 4) that provides a global picture of cellular function based on miRNA  Interestingly, TNFα signaling (p = 0.0009 with 58 gene targets) has been previously identified as a signaling defect in aging naïve T cells (Gupta & Gollapudi, 2006).
KEGG pathways usually integrate multiple upstream and downstream signaling molecules; thus, there is often a high level of redundancy in pathway assignments. Therefore, we investigated the functional overlap between three select pathways: FOXO, pluripotency in stem cells, and TNFα signaling. These pathways demonstrated multiple mRNA targets for all six age-related miRNAs with robust targeting of multiple genes within each pathway (Supporting Information Figure S1). Moreover, we found a high overlap between individual signaling pathways contained within these three pathways signaling in old naïve CD8 T cells was not due to altered FOXO1 protein expression in old naïve CD8 T cells (Supporting Information Figure S2a).

T-cells
To establish the functional activity of FOXO1 in naïve CD8 T cells, we compared gene expression of select FOXO1 target genes (IL7R, CCR7, SELL, TCF7) in a follow-up cohort of young and older individuals by qPCR. IL7R and CCR7 mRNA expression were significantly decreased in old naïve CD8 T cells (Figure 6a). SELL trended lower in older individuals. TCF7 did not have altered expression between young and old, consistent with a recently published study showing that FOXO1 induces TCF7 expression in memory, but not in naïve, CD8 T cells (Delpoux, Lai, Hedrick, & Doedens, 2017). Consistent with significantly decreased IL7R mRNA expression, IL-7R surface protein expression on old naïve CD8 T cells was lower than that of young ( Figure 6b). To demonstrate a direct link between FOXO1 activity and IL-7R expression, we inhibited FOXO1 activity in naïve expression by naïve CD8 T cells (Sheppard et al., 2014). This expression of miR-146a is alternatively inhibited by IL-7, suggesting that age-related increases in miR-146a (Figure 1a) could be, at least partially, mediated by altered signaling via homeostatic cytokines. Highaffinity TCR engagement also induces expression of many miRNAs including miR-146a and miR-155 (Teteloshvili et al., 2015); however, the effects of low-affinity TCR stimulation and the interplay between cytokines and TCR signaling in regulating miRNA expression are currently unknown.
Our study presented here revealed that old naïve CD8 T-cells also exhibit miRNA expression patterns partially related to cell differentiation. The idea that naïve T-cells undergo partial differentiation during aging is a relatively new idea and somewhat controversial, Similar to the miRNA findings presented here, our group has recently shown that naïve CD8 T-cells epigenetically and transcriptionally shift toward a more differentiated state during aging (Moskowitz et al., 2017). These different processes are likely to be highly interdependent and do not follow hierarchical models. miRNA inhibition of transcription factor activity, such as FOXO1 activity, could lead to chromatin closing as well as reduced transcription of target genes. Indeed, Ucar et. al. found chromatin closing at the IL7R locus in memory CD8 T cells from older individuals (Ucar et al., 2017).
There also appears to be modest, although not significant, closing in old naïve CD8 T-cells. This chromatin closing coincides with reduced IL-7R transcript and protein expression in old memory CD8 T-cells compared with young (Ucar et al., 2017;our unpublished observation).
Conversely, altered chromatin accessibility at loci of miRNA precursor transcripts could affect expression of the miRNAs. Indeed, we recently found the expression of miR-181a was driven by the transcription factor YY1, which is reduced in expression during aging and YY1 motifs are highly enriched at chromatin sites less accessible in old naïve CD8 T cells (Ye et al., 2018). Thus, miRNA expression changes may be both up-and downstream of epigenetic changes.
Functionally, the altered miRNA network in aged naïve CD8 T-cells predicted changes in multiple signaling pathways during aging, including those for TNFα, FOXO, and pluripotency of stem cells.
Indeed, we found that old naïve CD8 T cells had transcriptional profiles similar to FOXO1 knockout cells and display features of reduced FOXO1 activity. In the field of aging, FOXO transcription factors are important regulators of longevity and stem cell homeostasis (Boehm et al., 2012;Qin & Hubbard, 2015), sharing multiple conserved targets in humans, mice, C. elegans, and drosophila (Webb, Kundaje, & Brunet, 2016). Likewise, FOXO1 is required for T-cell homeostasis and T-cell maintenance within tissues-partially via the induction of IL-7R and other homing markers (Kerdiles et al., 2009;Ouyang, Beckett, Flavell, & Li, 2009). As we found reduced expression of IL-7R in old naïve CD8 T-cells correlated with lower naïve CD8 frequencies (Figure 6), it is possible that these cells have reduced homeostatic potential via FOXO1 pathway defects. This would also be consistent with a recent study characterizing human T-cell distribution throughout tissue compartments that found significant reductions in naïve CD8 T cells in blood, lymph nodes, and spleen with age (Thome et al., 2016). Within other tissues, such as the intestinal tract, memory T-cells predominate with very few naïve CD8 T-cells present. As our study was limited to the investigation of peripheral CD8 T cells, it may also be of interest to investigate miR-NAs and FOXO1-related changes in these tissue-specific naïve and memory T-cell subsets.
Along with tissue homing and homeostasis, FOXO1 signaling also plays a vital role in the induction and maintenance of CD8 T-cell memory in mice, with FOXO1 knockout naïve CD8 T-cells able to generate normal antigen-specific effector responses but unable to develop into long-lived memory T cells (Hess Michelini, Doedens, Goldrath, & Hedrick, 2013). Knocking out FOXO1 also causes reduced viability in CD8 T cells (Delpoux et al., 2018). These cellular changes are, in turn, linked with reduced protection against secondary antigen exposure. Interestingly, older humans display normal effector responses against the zoster vaccine but have much more rapid contraction of vaccine-specific memory cells-likely accounting for the poor efficacy of this zoster vaccine (Qi et al., 2016). In response to yellow fever vaccination, aged naïve CD8 T-cells also demonstrate reduced effector functions (Schulz et al., 2015). Thus, FOXO1 signaling may contribute both to alterations in homeostatic proliferation and memory maintenance in human CD8 T cells during aging. Further studies on the specific role of age-dependent miRNAs in modulating FOXO1 activity in CD8 T-cells would provide insight into the direct relationship between miRNAs, FOXO1, and CD8 T-cell aging.   | 9 of 12 products were then ran on the Biomark HD (Fluidigm, San Francisco, CA, USA) for miRNA quantification. A list of miRNA Taqman primer/ probes sets (Thermo Fisher Scientific) used in these studies is provided in Supporting Information Table S2. miRNA expression is relative to RNU48 expression.

| miRNA detection by FACS
miRNA was detected by FACS using PrimeFlow RNA Assay Kit (Thermo Fisher Scientific) using PrimeFlow microRNA pretreatment buffer and type 1 microRNA probe sets for detection.

Real-time quantitative PCR was performed with ABI Prism 7900HT
Detection System (Applied Biosystems, Foster City, CA, USA) using Taqman Universal Master Mix II, no UNG (Thermo Fisher Scientific) and Taqman probe sets. Probe sets used are as follows: IL7R (Hs00902334_m1), CCR7 (Hs01013469_m1), SELL (Hs00174151_m1), TCF7 (Hs01556515_m1), and RPLP0 (Hs99999902_m1). All data are presented as relative expression normalized to RPLP0 expression, as this control gene is found to be expressed the most consistently between young and old naïve CD8 T cells.

| Simple Western
Sorted naïve CD8 T cells were lysed with RIPA buffer (Santa Cruz Biotechnology, Santa Cruz, CA, USA) and sonicated for complete cell lysis. Lysates were centrifuged at 4°C for 10 min at 21,130 rcf, and

| Statistical analysis
Data were analyzed using Mann-Whitney test, paired t test, oneway ANOVA, or Pearson correlation as appropriate and as indicated in the specific Figure Legends. PCA was performed in R using prcomp. Statistical tests were performed using GRAPHPAD PRISM version 6. p-Values <0.05 are considered statistically significant.