Some hormone, cytokine and chemokine levels that change across lifespan vary by cognitive status in male Fischer 344 rats

Abstract

We trained and tested young (6–8 mo; n = 13), middle-aged (12–14 mo; n = 41), and aged (22–24 mo; n = 24) male Fischer 344 rats in a rapid acquisition water maze task and then quantified 27 stress hormones, cytokines and chemokines in their serum, hippocampi and frontal cortices using bead assay kits and xMAP technology. Middle-aged and aged rats learned the location of the hidden platform over training trials more slowly than their young counterparts. After training, young rats outperformed middle-aged and aged rats on both immediate and 24h retention probe trials and about half of the middle-aged and aged (aging) rats exhibited impaired performances when tested on the retention probe trial 24h later. The concentrations of many serum, hippocampal and cortical analytes changed with age often in networks that may represent age-sensitive signaling pathways and the concentrations of some of these analytes correlated with water maze learning and/or memory scores. Serum GRO/KC and RANTES levels, hippocampal GM-CSF levels and cortical IL-9 and RANTES levels were significantly higher in rats categorized as memory-impaired versus elite agers based upon their 24h probe trial performances. Our data add to the emerging picture of how age-related changes in immune and neuroimmune system signaling impacts cognition.

1. Introduction

In the United States, ten thousand baby boomers will retire each day over the next two decades (Passel and Cohn, 2008). The development of strategies that predict and treat cognitive decline with no underlying pathology in our rapidly expanding aging population is becoming increasingly critical for optimizing the quality of life for this age group and for containing the cost and burden of its care. Age-related cognitive decline has been linked to altered levels of stress hormones and circulating and central cytokines and chemokines (Kohman et al., 2012; Kohman et al., 2011; Speisman et al., 2013a; Villeda et al., 2011) that likely indicate age-sensitive changes in their signaling pathways. If changes in the levels of these or other factors in aging individuals predicted cognitive decline later in life, then early interventions such as daily exercise or environmental enrichment (Foster, 2007; Jurgens and Johnson, 2012; Kumar and Foster, 2007; Kumar et al., 2012; Speisman et al., 2013a; Speisman et al., 2013b; Williamson et al., 2012) could be employed and targets for novel therapeutic strategies revealed.

The variable performances of aging rats on a behavioral task can be capitalized upon to categorize them as memory-unimpaired or memory-impaired so that potential mechanisms mediating the impairment can be identified (Bañuelos et al., 2013; Drapeau et al., 2003; Foster, 2007, 2012; Foster et al., 2012; Gage et al., 1984; Kumar and Foster, 2013; Markowska, 1999). We have previously employed a rapid acquisition water maze task to detect memory impairments that emerge in rats as young as 13 months-old and that are exhibited reliably in rats older than ~ 18 months (Blalock et al., 2003; Foster, 2007; Kumar and Foster, 2013; Speisman et al., 2013a; Speisman et al., 2013b). Aged rats that exhibit impaired performances in this task have upregulated expression of neuroinflammation- and stress-related genes in the hippocampus, altered circulating and hippocampal cytokine concentrations and compromised measures of hippocampal synaptic plasticity and neurogenesis (Blalock et al., 2003; Foster, 2007; Kumar and Foster, 2013; Speisman et al., 2013b). Aged rats exposed to daily exercise or environmental enrichment for several months exhibit improved performances in this task, rejuvenated measures of hippocampal integrity and modulated circulating and central cytokine and leptin levels (Foster, 2007; Jurgens and Johnson, 2012; Kumar and Foster, 2007; Kumar et al., 2012; Speisman et al., 2013a; Speisman et al., 2013b; Williamson et al., 2012). These data suggest that lifestyle changes could minimize or reverse aspects of age-related cognitive decline that are related to compromised hippocampal integrity.

Several studies have linked age-related changes in immune system, neuroimmune system and hypothalamic-pituitary-adrenal (HPA) axis signaling to memory decline in rodents and humans (Chung et al., 2009; De Martinis et al., 2005; Foster, 2006; Magaki et al., 2007; Ogle et al., 2013; Rafnsson et al., 2007; Solfrizzi et al., 2006; Villeda et al., 2011). For example, cognitive scores correlate negatively with concentrations of inflammation markers in serum (Gimeno et al., 2008) and accelerated increases in stress hormone levels are associated with lower memory scores and smaller hippocampi in aging rodents and humans (He et al., 2008; Issa et al., 1990; Lupien et al., 1998; McEwen, 1998). In rodents, age-related changes in serum cytokine concentrations are associated with decreased levels of hippocampal neurogenesis and impaired cognition (Banks, 2006; Banks et al., 2002; Erickson et al., 2012; Speisman et al., 2013a; Villeda et al., 2011) and aged rodents that exercise daily exhibit rejuvenated measures of hippocampal neurogenesis and spatial ability perhaps because neuroimmune signaling is rejuvenated (Kohman et al., 2012; Kohman et al., 2011; Speisman et al., 2013a). The results suggest that chronic low-level inflammation or stress may contribute to the onset or progression of cognitive decline with advancing age. The ability to identify circulating or central stress or immune/neuroimmune system biomarkers in middle-aged subjects may warrant the adoption of lifestyle changes that minimize or prevent cognitive decline later in life.

Here, we quantify the concentrations of 27 stress hormones, cytokines and chemokines in the blood and brains of young, middle-aged and aged rats characterized behaviorally as memory-impaired or memory-unimpaired in a rapid acquisition water maze task sensitive to the onset of age-related cognitive decline. We hypothesized that 1) the concentrations of some analytes would change with age, 2) the concentrations of some of the analytes that changed with age may exhibit face validity as biomarkers of age-related cognitive decline by correlating with water maze learning and/or memory scores and 3) that biomarkers of age-related memory loss would emerge as analytes with concentrations that differed in rats categorized as memory-unimpaired versus –impaired on the water maze delayed probe trial. A few candidate biomarkers of age-related cognitive decline were identified.

2. Materials and Methods

2.1. Subjects

Young (6–8 mo; n=13), middle-aged (12–14 mo; n=41) and aged (22–24 mo; n=24) male Fischer 344 rats were purchased from the National Institute of Aging colony at Harlan Sprague Dawley (Indianapolis, IA). Based upon our previous work, this approximate number of young and aged rats yields sufficient power (π > 0.80) for detecting statistically reliable individual differences in behavioral scores. Because we expected increased behavioral performance and analyte concentration variability among middle-aged rats that we predicted should either remained cognitively intact or exhibit impaired cognition, we included approximately twice as many middle-aged versus aged rats. Upon arrival, the rats were pair-housed in shoebox cages in a standard (non-specific pathogen free barrier) University of Florida colony room that was maintained at 24±1°C on a 12:12 h light:dark cycle. The rats were fed irradiated Harlan Teklad® Rodent food diet #8604 and reverse osmosis-filtered water ad libitum for the duration of the experiment. Rats exhibiting signs of aggression (bites and scratches) or age-related health problems (i.e. poor grooming, weight loss or tumor growth) were euthanized humanely. All rats were treated in accordance with Federal and University of Florida policies regarding the ethical use of animals for experimentation.

2.2. Water maze training and testing

Water maze training and testing sessions were conducted on all rats in the behavioral room using the methodology that we have described previously (Foster and Kumar, 2007; Kumar and Foster, 2013; Speisman et al., 2013a; Speisman et al., 2013b). During training and testing sessions, that were conducted at about the same time each day during the light cycle, the rats were transported to the behavioral room that was illuminated with standard fluorescent lighting and that housed the water maze. Rats were habituated to the water maze by being released from three random locations and assisted, if needed, to a visible platform just before a being given a session of visible platform trial blocks. A session of hidden platform trial blocks was given 3 days later. Probe trials were conducted immediately and 24h after the hidden platform training session. The black cylindrical water maze tank (1.7 m diameter) was filled with 27±2°C water to a depth of either 1.5 cm below (on visible platform trials) or above (on hidden platform trials) a black platform (29 cm diameter). Water depth and extra-maze cue placement was identical on hidden platform and probe trials. Pathlengths (cm), escape latencies (s) and quadrant search times (s) were recorded by a Columbus Instruments tracking system (Columbus, OH). Rats were towel-dried and warmed under a heat lamp between trial blocks.

2.2.1. Visible platform trials

Rats were given five blocks (15 min inter-block-interval [IBI]) of 3 60-s trials (20 s inter-trial-interval [ITI]) to locate the platform and were guided to the platform if they failed to locate it within 60s. The N, S, E and W release points and flagged platform location were changed randomly on each trial. Latencies (s), pathlengths (cm) and swim speeds (cm/s) were used as measures of sensorimotor ability, visual acuity, and the ability to associate locating the visible flagged platform with escape from the water maze. The water maze was surrounded by a black curtain to obscure extra-maze cues.

2.2.2. Hidden-platform trials

Rats were given five blocks (15 min IBI) of 3 60s trials (20 s ITI)in which they were released randomly from the edge of one of the three pool quadrants not housing the platform and guided gently to the platform if they did not locate it within 60s. Latency (s) and pathlength (cm) across trials served as measures of spatial learning while swim speeds (cm/s) served as measures of sensorimotor ability. Large and visible extramaze cues surrounded the maze on hidden platform trials.

2.2.3. Probe trials

The platform was removed from the water maze on the probe trials administered immediately and 24h after hidden-platform training. Rats were released from the quadrant housing the platform on hidden platform trials and free swam for 60s. Discrimination index (DI) scores (t(G)−t(O)/t(G)+t(O), where t(G) is time spent in the goal quadrant and t(O) is time spent in the opposite quadrant) served as measures of strength of learning and memory. Note that DI scores > 0 confirm better than chance (> 25% of time spent in the goal quadrant) performance, a DI score of 0.33 confirms 2x as much time in the goal versus opposite quadrant and a DI score of 1 confirms 100% of time spent in the goal versus opposite quadrant. DI scores produce a higher fidelity memory index for aged rats that often make wider sweeping turns while navigating by swimming and that exhibit slower swim speeds than young rats. A hidden platform trial block was administered after the immediate probe trial to reinforce the association between platform localization and escape from the maze.

2.3. Sample collection

In order to minimize fluctuations on analytes concentrations that behavioral testing may produce rats were deeply anesthetized with isoflurane (Halocarbon Laboratories, River Edge, NJ) and decapitated two weeks after the final probe trial so that basal cytokine, chemokine and hormone levels could be quantified. Trunk blood was collected and stored at 4°C for 30 min and then centrifuged at 1,000 x g for 10 min so that serum could be collected and stored at −86°C until analyzed with the Bio-Plex platform. Whole hippocampi and frontal cortices (tissue anterior to the fimbria/fornix and posterior to the olfactory nerves) were dissected rapidly from both hemispheres of the extracted brain, flash frozen in a metal beaker in isopropanol over dry ice and stored at −86°C until protein harvest.

2.4. Protein harvest

Protein was harvested rapidly from hippocampal and cortical tissue at 4°C. The tissue was suspended in 0.1 M tris-buffered saline (TBS) containing 0.1% Igepal and 1 μL/mL of two protease inhibitor cocktails combined immediately before use. Cocktail consisted of 0.5 M phenylmethylsulfonyl fluoride, 5 mg pepstatin A and 1 mg chymostatin/1 mL DMSO and Cocktail 2 consisted of 1 M G-aminocapfroic acid, 1 MP-aminobenzidine, 1 mg leupeptin and 1 mg aprotinin/mL sterile water. Tissue was mashed manually and then sonicated with a dismembrator (Thermo Fisher Scientific; Pittsburgh, PA) before centrifugation at 12,000 rpm for 10 min at 4°C to isolate protein. Protein concentrations were measured with a Bradford protein assay and Bio-Rad SmartSpec Plus Spectrophotometer (Hercules, CA). Tissue supernatant was stored at −86°C until analysis. Similar protein concentrations were detected in hippocampal samples across age (young: 5.98±0.23 mg/ml; middle-aged: 5.88±0.22 mg/ml; aged 5.66±0.25 mg/ml; F_(2,76) = 0.37; p = 0.69) but significantly smaller protein concentrations were detected in the cortices of middle-aged (19.08±0.65 mg/ml) and aged (18.80±0.85 mg/ml) versus young (25.17±1.12 mg/ml) rats (Newman-Keuls p values < 0.001; F_(2,76) = 12.65; p < 0.001).

2.5. Bio-Plex quantification of analytes

A Bio-Rad Bio-Plex 2000 suspension array system, EMD Millipore Rat Cytokine/Chemokine kit (Cat No. RCYTO-80K-PMX; Billerica, MA) and EMD Millipore Rat Stress Hormone kit (Cat No. RSH69K) was used to quantify cytokines/chemokines and stress hormones in the blood serum and hippocampal and cortical protein samples obtained from each rat according to kits instructions and (Speisman et al., 2013a). The cytokine/chemokine kit measures concentrations of CCL11 (eotaxin; 3.27–20,000 pg/mL), granulocyte-colony stimulating factor (G-CSF; 1.31–20,000 pg/mL), granulocyte macrophage colony-stimulating factor (GM-CSF; 13.11–20,000 pg/mL), CXCL1 or growth-related oncogene/keratinocyte chemoattractant (GRO/KC; 2.06–20,000 pg/mL), interferon-γ (IFN-γ; 4.88–20,000 pg/mL), IL-1α (6.23–20,000 pg/mL), IL-1β (2.32–20,000 pg/mL), IL-2 (3.67–20,000 pg/mL), IL-4 (2.30–20,000 pg/mL), IL-5 (2.89–20,000 pg/mL), IL-6 (9.80–20,000 pg/mL), IL-9 (12.85–20,000 pg/mL), IL-10 (5.41–20,000 pg/mL), IL-12 (4.13–20,000 pg/mL), IL-13 (23.2–20,000 pg/mL), IL-17 (1.61–20,000 pg/mL), IL-18 (4.78–20,000 pg/mL), CXCL10 or interferon-γ-induced protein 10 (IP-10; 3.78–20,000 pg/mL), leptin (21.50–100,000 pg/mL), CCL2 or monocyte chemotactic protein-1 (MCP-1; 3.81–20,000 pg/mL), CCL3 or macrophage inhibitory protein-1α (MIP-1α; 1.94–20,000 pg/mL), CCL5 or RANTES (54.42–20,000 pg/mL), TNF-α (4.44–20,000 pg/mL) and vascular endothelial growth factor (VEGF; 4.93–20,000 pg/mL) in a single sample. The stress hormone kit measures concentrations of adrenocorticotrophic hormone (ACTH; 3.8–4,000 pg/mL), corticosterone (CORT; 10,834–400,000 pg/mL) and melatonin (MLT; 897–400,000 pg/mL) in a single sample.

All samples were prepared according to kit instructions on ice and serum and tissue samples were run on separate plates. Standards were diluted serially with kit assay buffer for the cytokine/chemokine assay with expected concentrations of 20,000, 5,000, 1,250, 312.5, 78.13, 19.53 and 4.88 pg/mL of each analyte except leptin which had expected concentrations of 100,000, 25,000, 6250, 1,562.5, 390.63, 97.66 and 24.41 pg/mL and the hormone assay with expected concentrations of 400,000, 133,333, 44,444, 14,814, 4,938 and 1,646 pg/mL of each analyte except ACTH which had expected concentrations of 4,000, 1,333, 444.4, 148.1, 49.4 and 16.5 pg/mL. Tissue supernatant samples were used neat while serum samples were diluted with kit assay buffer before use (1:5 for cytokine/chemokine assays or 1:3 for hormone assays).

A volume of 25 μL of each standard, vendor-supplied control and sample were loaded in duplicate into a 96-well filter plate (EMD Millipore; Billerica, MA). All wells were filled to a final volume of 50μL with the addition of either 1) 25 μL assay buffer to all sample wells and 2) 25 μL kit serum matrix to standard and control wells in serum quantification plates or 3) extraction buffer to standard and control wells in tissue quantification plates. A mixture of 24 different polystyrene beads each with unique color addresses and capture antibodies for the cytokine/chemokines to be detected or 3 different color addresses and capture antibodies for the stress hormones to be detected was added to the wells and incubated overnight on a shaker at 4°C. After several washes under vacuum filtration, the beads were incubated in biotinylated detection antibodies against each analyte for 2h at RT and after several more washes under vacuum filtration, a streptavidin-phycoerythrin reporter for 30 min at RT. The bead complexes were washed, resuspended in sheath fluid (Bio-Rad; Hercules, CA) and run through a dual laser Bio-Rad Bio-Plex 2000 system with Luminex xMAP technology (Bio-Rad; Hercules, CA). Unique bead wavelength emission identified each analyte and phycoerythrin emission intensity quantified concentrations. Data were compiled using Bio-Plex Manager Software version 4.1.

All standard, control and sample protein concentrations were obtained from at least 35 beads passing through the double discriminator gated region set to exclude broken or aggregated beads. A five-parameter logistic non-linear regression model was used to generate a standard curve for each analyte based on the average of duplicate observed concentrations. A single standard was used if its duplicate was > 10% CV (coefficient of variation) or if the percent recovery (observed/expected concentration) fell outside of the accepted 70–130% range. The concentrations of positive controls supplied with the kit were confirmed to fall within the expected range before sample concentrations were calculated according to their median fluorescent intensity using the appropriate standard curve. Cytokine concentrations falling below the threshold of detection were set at 0 and those exceeding the maximum expected concentration were set to the maximum expected concentration for that analyte. If the % CV for a set of duplicate sample concentrations was > 10% and a concentration fell > ±2 standard deviations from the group mean the outlying duplicate was discarded. Duplicate sample concentrations were averaged and reported as pg/mL serum or normalized against total protein and reported as pg/mg total hippocampal or cortical protein. If the averaged concentration of an analyte fell > ±2 standard deviations from the group mean, then the concentration for that analyte (but not the other analytes in the sample) was discarded as a machine read error.

2.6 Pathway analysis to confirm and reveal cytokine signaling networks

The peripheral function and signaling pathways of the cytokines and chemokines quantified in the current study have been relatively well-described but their CNS pathways and functions are less well known. In addition, how age-dysregulated immune signaling translates to changes in neuroimmune signaling is unclear. Therefore, a pathway analysis was conducted to identify analytes with correlated concentration increases or decreases within 1) blood, 2) hippocampi and 3) frontal cortical compartments and between 4) blood and hippocampal, 5) blood and cortical and 6) hippocampal and cortical compartments with age(Baron and Kenny, 1986; Erickson and Banks, 2011; Speisman et al., 2013a). Pathway analysis was employed as described by Baron and Kenny (1986) and Erickson and Banks (2011) because it is a form of exploratory factor analysis that can identify direct as well as indirect (mediator) relationships. Briefly, pairs of analytes that changed in concentration with age were listed in descending order based upon their Spearman rank correlation coefficients (r_s) and those with coefficients deemed statistically significant after a Bonferroni-Holm corrections for multiple dependent comparisons (p = 0.0001 for 590 comparisons that included 16 serum analytes, 5 hippocampal analytes and 15 frontal cortical analytes that changed in concentration with age) were plotted in networks of analytes with correlated concentrations. Consider that analytes A and B are already plotted as network (A–B) and that the concentration of analyte C correlates significantly with the concentration of analyte A. The concentration of analyte C must also correlate with the concentration of analyte B before being added to network A–B to form network C–A–B. Otherwise, new network A–C would be drawn. If analyte pair D–E is about to be plotted but is already contained in a network (D-A-B-E, where A and B are potential mediators) then the residual correlation (the r_s of the analyte pair about to be plotted minus the product of the r_s values of all mediators) was confirmed statistically significant before analytes D and E in network D-A-B-E were connected with a loop.

2.7. Statistical analyses

Statistical analyses were conducted using STATISTICA software (Version 10; StatSoft; Tulsa, OK) and data are shown as either group averages (± S.E.M.) or individual probe trial discrimination index [DI] scores. One way analysis of variance (ANOVA) tests were employed to identify significant effects of the independent variable (age) on measures of general health (body mass and swim speeds), water maze scores obtained once (total distance and swim speed on each probe trial) and analyte concentrations. Two-way ANOVA tests were employed to identify significant effects of the independent variable (age) on water maze scores obtained across trial blocks (visible and hidden platform trial latencies and pathlengths). Significant effects identified by omnibus ANOVA tests were revealed using Newman Keuls post-hoc tests. When Levene’s tests confirmed heterogeneity of variance in analyte concentrations between age groups or for categorical data, Kruskal-Wallace ANOVA and post hoc Mann-Whitney U tests were used to reveal statistically significant effects of the independent variable (age or ‘elite’, ‘unimpaired’ and ‘impaired’ cognitive status) on the dependent variable (analyte concentration and DI scores). Spearman rank correlations were used to test relationships between analytes concentrations and other variables because analytes with concentrations that fell below the threshold of detection were set at 0. The α-level was set at p = 0.05 with the exception of pathway analyses in which the α-level was set at p = 0.0001 after a Bonferroni-Holm correction for multiple comparisons on dependent variables.

3. Results

3.1. Concentrations of many stress hormones, cytokines and chemokines change across lifespan

Table 1A shows hormone, cytokine and chemokine concentrations in the serum of young, middle-aged and aged rats. Note that all serum GM-CSF concentrations fell below the threshold of detection and were therefore excluded from further analyses. In addition, IL-9 and IL-12 concentrations were measured in smaller subsets of samples obtained from rats across age groups because assay composition was unexpectedly altered by the vendor mid-experiment. Age significantly affected serum CORT (F_(2,72) = 9.63; p < 0.01), GRO/KC (F_(2,74) = 3.79; p < 0.05), IFN-γ (H_(2,n=77) = 12.04; p < 0.01), IL-1α (H_(2,n=72) = 13.97; p < 0.001), IL-1β (H_{(2, n=77)} = 13.24; p < 0.01), IL-4 (H_(2,n=76) = 11.05; p < 0.01), IL-5 (H_(2,n=75) = 9.21; p < 0.01), IL-6 (H_(2,n=75) = 9.94; p < 0.01), IL-13 (H_{(2, n=75)} = 12.33; p < 0.01), IL-17 (H_(2,n=76) = 10.49; p = 0.01), IL-18 (H_{(2, n=75)} = 6.77; p < 0.05), IP-10 (H_(2,n=75) = 12.13; p < 0.01), leptin (F_(2,75) = 5.07; p < 0.01), MCP-1 (H_{(2, n=75)} = 15.02; p < 0.001), MIP-1α (F_{(2, 72)} = 4.49; p < 0.05) and RANTES (F_(2,71) = 3.75; p < 0.05) concentrations. Relative to concentrations detected in young rats, serum CORT (p < 0.01) and leptin (p < 0.01) concentrations were elevated in middle-aged rats, serum IFN-γ, IL-6, IL-13, IP-10 and MIP-1α concentrations were elevated respectively in middle-aged (all p values < 0.05) and aged rats (all p values < 0.01) and serum GRO/KC (p < 0.05), IL-1α (p < 0.05), IL-1β (p < 0.01), IL-4 (p < 0.05), IL-5 (p < 0.01), IL-17 (p < 0.01), IL-18 (p < 0.05), MIP-1α (p < 0.01) and RANTES (p < 0.05) concentrations were elevated in aged rats. Relative to levels detected in middle-aged rats, serum GRO/KC (p < 0.05), IFN-γ (p < 0.05), IL-1α (p < 0.05), IL-1β (p < 0.01), IL-4 (p ≤ 0.05), IL-5 (p ≤ 0.05), IL-17 (p < 0.05), IL-18 (p < 0.05), IP-10 (p ≤ 0.05) and MCP-1 (p < 0.01) were elevated in aged rats.

Table 1.

Chemokine, cytokine and hormone concentrations in young, middle-aged and aged rats

	A. Serum (pg/ml)			B. Hippocampus (pg/mg)			C. Cortex (pg/mg)
Analyte	Young	Middle-aged	Aged	Young	Middle-aged	Aged	Young	Middle-aged	Aged
ACTH	154.4±27.4	168.9±14.5	120.3±8.8	0.8±0.4	1.0±0.2	1.3±0.3	0.2±0.1	0.3±0.0a	0.4±0.0a
CORT	203324.5±20879.4	304517.8±16709.5a	254878.8±24188.6	1978.0±348.2	2265.6±196.1	1828.1±256.3	1118.8±208.9	1317.8±113.0	1160.8±150.9
eotaxin	8.9±4.7	28.8±6.6	78.7±22.8	6.1±0.5	5.6±0.3	5.8±0.3	1.2±0.1	1.4±0.1	1.7±0.1a,b
G-CSF	0.0±0.0	6.5±1.6	6.4±2.3	0.6±0.1	0.6±0.1	0.5±0.0	0.1±1.1	0.2±0.0	0.2±0.0
GM-CSF	-	-	-	1.2±0.4	1.8±0.3	2.3±0.4	0.5±0.1	0.6±0.1	0.7±0.1
GRO/KC	937.4±73.5	978.9±42.4	1194.2±92.9a,b	48.5±3.0	46.4±2.7	52.2±2.7	5.0±0.5	6.0±0.4	8.6±0.6a,b
IFN-γ	31.2±9.5	83.3±13.4a	357.4±105.1a,b	2.6±0.7	1.7±0.4	3.3±0.6	0.4±0.2	0.3±0.1	0.4±0.1
IL-1α	0.0±0.0	3.1±1.9	140.8±67.0a,b	6.7±1.2	4.4±0.5	4.5±0.7	1.2±0.2	1.7±0.2	1.9±0.3
IL-1β	5.9±2.7	23.3±6.7	78.8±19.2a,b	23.3±1.7	24.5±1.2	22.9±1.0	5.9±0.8	11.3±0.7a	11.8±1.2a
IL-2	52.2±21.6	49.2±8.9	98.9±20.5	58.2±12.2	42.0±4.1	50.2±6.3	14.5±1.6	18.3±1.4	19.9±1.5
IL-4	0.0±0.0	17.5±5.1	65.9±18.6a,b	16.6±1.7	14.3±0.7	14.4±0.6	2.3±0.2	3.4±0.2a	3.4±0.2a
IL-5	11.4±4.9	30.9±5.4	56.4±10.5a,b	16.4±3.5	8.8±1.1a	6.5±0.7a	1.3±0.2	2.6±0.3a	1.6±0.2b
IL-6	7.0±4.0	78.5±13.8a	283.54±102.4a	63.0±7.0	52.0±2.5	59.6±3.6	6.2±0.4	11.5±0.9a	9.6±1.2a
IL-9	489.5±187.0	564.2±111.7	612.3±141.2	1340.8±105.0	1274.3±95.9	1781.7±131.0a,b	206.1±26.3	284.6±17.4a	300.1±23.1a
IL-10	0.0±0.0	136.1±36.4	161.9±48.6	15.9±3.5	11.9±1.3	18.2±2.6	2.8±0.3	4.1±0.4	4.7±0.4a
IL-12	24.5±11.8	59.3±9.1	119.4±55.8	20.1±4.1	15.1±1.8	9.7±1.9a	4.6±1.1	7.4±0.2	8.2±2.2
IL-13	23.7±5.9	82.4±10.6a	190.4±57.6a	11.5±1.1	9.7±0.6	9.1±0.6	1.1±0.2	2.4±0.3	2.2±0.4
IL-17	5.3±1.9	11.5±1.7	47.2±17.7a,b	6.5±1.2	5.2±0.4	6.4±0.9	0.5±0.1	0.6±0.1	0.7±0.1
IL-18	91.9±16.9	88.1±7.7	181.4±30.9a,b	118.3±6.9	139.5±6.5	146.6±8.1	44.8±4.2	63.9±3.3a	76.6±4.5a,b
IP-10	1.4±1.0	24.4±5.8a	61.8±23.1a,b	4.5±1.1	32.9±10.5	40.8±12.8	1.0±0.2	8.3±2.9	9.9±5.1
leptin	23681.4±2644.6	36180.5±2019.0a	29841.1±3006.8	41.7±5.1	44.0±3.4	42.2±3.4	8.1±0.6	12.5±0.9a	10.8±1.1
MCP-1	564.0±30.3	650.9±19.2a	790.2±51.3a,b	156.1±18.8	157.3±12.6	169.5±15.9	17.1±2.4	33.4±2.4a	32.9±2.3a
MIP-1α	5.7±1.0	8.2±0.9	11.5±1.5a	0.6±0.0	0.7±0.0	0.8±0.1a	0.3±0.0	0.4±0.0a	0.5±0.0a,b
MLT	20923.9±4410.1	27322.4±3206.9	26203.5±2979.3	1697.4±763.1	2286.0±374.9	3474.6±421.0	1673.4±102.1	1959.3±244.7	1310.3±142.3
RANTES	23640.4±2219.2	23045.4±1345.8	31309.5±3520.3a	16.2±2.9	18.5±2.1	35.0±4.7a,b	8.7±0.8	14.2±0.9	23.9±3.7a,b
TNF-α	0.0±0.0	2.6±1.0	1.5±0.8	5.4±0.5	4.0±0.2	5.1±0.6	1.1±0.2	1.3±0.1	1.6±0.1
VEGF	4.6±3.1	52.9±13.3	27.0±11.7	12.7±1.5	15.1±0.6	13.3±1.8	1.4±0.1	2.1±0.2a	2.3±0.2a

^ap < 0.05 versus young,

^bp < 0.05 versus middle-aged

Table 1B shows hormone, cytokine and chemokine concentrations in the hippocampi of young, middle-aged and aged rats. Age significantly affected IL-5 (F_(2,72) = 7.93; p < 0.001), IL-9 (F_(2,55) = 4.88; p < 0.05), IL-12 (F_(2,42) = 3.22; p < 0.05), MIP-1α (F_(2,74) = 4.57; p < 0.05) and RANTES (F_(2,74) = 7.69; p < 0.001) concentrations in the hippocampi of adult rats. Relative to young rats, hippocampal IL-5 levels were lower in middle-aged (p < 0.01) and aged (p < 0.001) rats and IL-12 (p < 0.05) levels were lower in aged rats. In contrast, MIP-1α (p < 0.01) levels were significantly higher aged versus young rats and IL-9 and RANTES levels were higher in aged versus young (p values < 0.05) and middle-aged (p values < 0.01) rats.

Table 1C shows hormone, cytokine and chemokine concentrations in the frontal cortices of young, middle-aged and aged rats. Age significantly affected cortical ACTH (F_(2,75) = 3.58; p < 0.03), eotaxin (F_(2,73) = 6.53; p < 0.01), GRO/KC (F_(2,73) = 10.23; p < 0.001), IL-1β (F_(2,74) = 7.42; p < 0.01), IL-4 (F_(2,74) = 6.01; p < 0.01), IL-5 (F_(2,73) = 6.88; p < 0.01), IL-6 (F_(2,71) = 5.71; p < 0.01), IL-9 (F_(2,57) = 4.44; p < 0.05), IL-10 (F_(2,71) = 3.07; p = 0.05), IL-18 (F_(2,74) = 9.70; p < 0.001), leptin (F_(2,73) = 3.65; p < 0.05), MCP-1 (F_(2,74) = 8.08; p < 0.001), MIP-1α (F_(2,73) = 8.61; p < 0.001), RANTES (F_(2,73) = 9.05; p < 0.001) and VEGF (F_(2,72) = 4.34; p < 0.05) levels. Respective cortical IL-18 and MIP-1α levels increased age-dependently, such that levels were higher in middle-aged rats versus young (p < 0.01 and p < 0.05) and in aged versus middle-aged (p = 0.05 and p < 0.05) and young rats (p < 0.001 and p < 0.001) rats. Relative to young rats, middle-aged and aged rats had higher respective cortical ACTH (p < 0.05 and p < 0.05), IL-1β (p < 0.001 and p < 0.001), IL-4 (p < 0.01 and p < 0.01), IL-6 (p < 0.01 and p < 0.05), IL-9 (p < 0.05 and p < 0.01), MCP-1 (p < 0.001 and p < 0.001) and VEGF (p < 0.05 and p < 0.01) levels, middle-aged rats had higher leptin levels (p < 0.05) and aged rats had higher IL-10 levels (p < 0.01). Cortical IL-5 levels were higher in middle-aged versus young rats (p < 0.01) and lower in aged versus middle-aged rats (p < 0.05). Aged rats had higher respective cortical eotaxin, GRO/KC and RANTES levels than middle-aged (p < 0.01, p < 0.001, p < 0.01) and young rats (p < 0.05, p < 0.001 and p < 0.001).

3.2. Pathway analyses reveal networks of cytokines that are impacted by age

Figure 1 shows cytokines, chemokines and hormones with concentrations that were both significantly impacted by age and that a pathway analysis revealed were networked with concentrations of other analytes within and between serum, hippocampal and cortical compartments after correcting the alpha level to p < 0.0001 using Bonferroni-Holm’s approach for multiple comparisons. Within serum we detected 11 networks. Interestingly, IFN-γ appears in 4 independent networks, IL-13, IL-18, IP-10 and MIP-1α each appear in 3 independent networks, IL-1α, IL-5, IL-10 and MCP-1 each appear in 2 independent networks, and leptin, IL-1β, IL-4, IL-6 appear in 1 network each. Notably, no single analyte exhibited a serum concentration that correlated with either a hippocampal or cortical concentration indicating that tissue measurements were not likely contaminated by blood analytes.

Figure 1. Pathway analysis reveals cytokine networks that change with age.

Spearman rank correlations between cytokines significantly modified by age (see Table 1; p-values < 0.0001 according to a Bonferroni-Holm correction) were plotted to reveal signaling networks. Pro-inflammatory (pink), anti-inflammatory (blue), and recruitment/trafficking (gray) chemokine and cytokines, growth factors (yellow) and hormones (purple) are shown. Green lines link positively correlated and red lines link negatively correlated analyte concentrations. Note that many analyte pairs are repeated frequently and a few cross-compartment (serum-hippocampus, serum-cortex and hippocampus-cortex) networks emerge.

Several pairs and trios of analytes appeared in multiple cytokine networks, supporting the hypothesis that some networked cytokines may be part of a common signaling cascade impacted by age. In serum we found 1) MCP-1 and MIP-1α, 2) IL-13 and IFN-γ, 3) IL-13 and IP-10, 4) IFN-γ and IP-10, 5) IL-1α and IFN-γ and 6) IL-13, IFN-γ and IP-10 appeared together in 2 independent networks. Circulating MIP-1α and MCP-1 were identified as factors that may regulate neuroimmune signaling across lifespan because concentrations of serum MIP-1α correlated positively with concentrations of hippocampal RANTES and concentrations of serum MCP-1 correlated positively with concentrations of cortical MIP-1α and IL-18.

Within the hippocampus, IL-5 networked with IL-12. Several corticolimbic networks that were impacted by age emerged. Hippocampal IL-12 concentrations correlated negatively with cortical MIP-1α concentrations and positively correlated hippocampal MIP-1α and RANTES concentrations correlated negatively with cortical IL-5 concentrations. Hippocampal RANTES concentrations correlated positively with positively correlated cortical MIP-1α and RANTES concentrations. RANTES occurred in most networks bridging the hippocampus and cortex with MIP-1α appearing on the peripheral side of immune-brain networks and the cortical side of the corticolimbic network.

Sixteen analyte networks were detected within the cortex. IL-1β was detected in 8 independent networks, IL-10 and IL-18 were each detected in 7 independent networks, IL-6, IL-4 and MCP-1 were each detected in 5 independent networks, Gro/KC, eotaxin and RANTES were each detected in 4 independent networks, leptin, MIP-1α and VEGF were each detected in 3 independent networks and ACTH and IL-5 were each plotted once. In the cortex, we also detected pairs and trios of analytes in multiple networks. For example, 1) IL-10 and IL-4 appeared together in 5 networks, 1) IL10 and GRO/KC appeared together in 4 networks, 1) IL-1β and IL-6, 2) IL-1β and IL-18, 3) MIP-1α and IL-18, 4) MCP-1 and IL-4 and 5) MCP-1 and IL-4 appeared together in 3 networks, 1) eotaxin and GRO/KC, 2) eotaxin and IL-4, 3) eotaxin and IL-18, 4) GRO/KC and IL-4, 5) GRO/KC and IL-18, 6) GRO/KC and MIP-1α, 7) GRO/KC and RANTES, 8) IL-1β and MCP-1, 9) IL-1β and RANTES, 10) IL-4 and IL-6, 11) IL-4 and MCP-1, 12) IL-6 and MCP-1, 13) IL-10 and IL, 14) IL-10 and MIP-1α, 15) IL-10 and RANTES, 16) IL-18 and IL-10, 17) IL-18 and MIP-1α, 18) IL-18 and RANTES and 19) IL-18 and VEGF, and MCP-1 and RANTES appeared together in 2 independent networks. 1) IL-4, IL-10 and RANTES appeared together in 3 networks, 1) IL-4, IL-6 and IL-10, 2) IL-4, IL-10, MCP-1, 3) eotaxin, GRO/KC and IL-10, 4) GRO/KC, IL-4 and IL-10, 5) GRO/KC, IL-10 and IL-18, 6) GRO/KC, IL-10 and MIP-1α, 7) GRO/KC, IL-10 and RANTES, 8) GRO/KC, IL-18 and MIP-1α and 9) IL-10, IL-18, and MIP-1α appeared together in 2 independent networks. With the exception of MIP-1α and RANTES that were detected together in several serum, hippocampal and cortical networks the analyte pairs and trios detected in the cortex were unique from those detected in the hippocampus and few cytokine pairs detected in the cortex were also detected in serum.

3.3 Visible platform trial performances improve on late training blocks, despite age-related impairment

Although pathlengths correlated positively with escape latencies across both the visible (r values ≥ 0.68, p-values < 0.001) and hidden (r values ≥ 0.84, p-values < 0.001) platform trials, we report water maze pathlengths because swim speeds change across age (see below). Figure 2A shows that visible platform pathlengths were significantly affected by age (F_(2,75) = 9.57; p < 0.001), training block (F_(4,300) = 26.51; p < 0.001) and the interaction between age and training block (F_(8,300) = 2.71; p < 0.01). On all trials combined, young rats swam more directly to the visible platform than middle-aged (p < 0.01) and aged rats (p < 0.001). Combined, all rats swam more directly to the visible platform as training commenced (block 1 > 2 > 3 > 4 and 5; all p-values < 0.05). However, pathlengths decreased across early blocks in young rats (block 1 > 2 > 3–5; all p-values < 0.05), more uniformly across blocks in middle-aged rats (blocks 1 > 3–5; block 2 > 4, 5 and block 3 > 5; all p-values < 0.05) and across later blocks in aged rats (block 2 > 4, 5; p ≤ 0.05). While all rats exhibited similar pathlengths on the first training block (p values > 0.53), young rats outperformed middle-aged rats on block 3 (p < 0.05) and aged rats on blocks 2–5 (all p values < 0.05). Middle-aged rats outperformed aged rats on block 5 (p < 0.05).

Figure 2. All rats learned to locate a visible platform.

Data are group means (±S.E.M.) for young (white circles), middle-aged (gray triangles) and aged (black squares) rats. On the bar graph, white bars depict performances on the 1^st bin (blocks 1 and 2) and black bars depict performance on the 2^nd bin (blocks 4 and 5) of visible platform trials. A) Pathlengths decreased across early blocks in young rats (block 1, 2 > 3–5; all p-values < 0.05), uniformly across blocks in middle-aged rats (blocks 1, 2 > 3–5; block 2 > 4, 5 and block 3 > 5; all p-values < 0.05) and across later blocks in aged rats (block 2 > 4, 5; p = 0.05 and p = 0.07, respectively). Young rats outperformed middle-aged rats on block 3 (p < 0.05) and aged rats on blocks 2–5 (all p values < 0.05). Middle-aged rats tended to outperform aged rats on block 4 (p < 0.10) and outperformed them on block 5 (p < 0.05). B) Young rats outperformed aged rats on Bin 1 (p < 0.05^a) and middle-aged and aged rats on Bin 2 (all p values < 0.05^b) and middle-aged rats outperformed aged rats on Bin 2 (p = 0.01^c). All age groups exhibited better performances on the 2^nd versus 1^st trial bin (all p-values < 0.01). ** p < 0.001 and *** p < 0.0001. C) On all trials combined, young rats swam more quickly than middle-aged and aged rats (p values < 0.001) and middle-aged rats swam more quickly than aged rats (p = 0.008) and all rats combined swam more quickly as training progressed (block 1 > 2–5 and block 2 > 3; all p-values < 0.05). While young rats exhibited large increases in swim speeds across blocks (block 1,2 > 3–5; all p values < 0.05), middle-aged (block 1 > 2; p = 0.03) and aged (block 1 > 2, 5; all p values < 0.05) rats exhibited smaller more inconsistent changes. (D) Venn diagram showing serum, hippocampal and cortical analyte concentrations correlating positively (green line through the analyte) or negatively (red line through the analyte) with hidden platform pathlengths averaged across training blocks in young (green zone), middle-aged (yellow zone), aged (pink zone), all (green/yellow/pink overlap) and middle-aged and aged (yellow/pink overlap) rats. Analytes are coded for their known function (see legend).

Because aged rats typically exhibit shallow learning curves on this task, we binned block 1 with 2 and block 4 with 5 to confirm improved performances across later versus earlier trials (Figure 2B). On all bins combined, young rats exhibited shorter pathlengths than either middle-aged (p < 0.05) or aged rats (p < 0.001) and middle-aged rats exhibited shorter pathlengths than aged rats (p = 0.05; F_(2,75) = 8.53; p < 0.001) and all rats combined exhibited shorter pathlengths on the last versus first bin (p = 0.0001; F_(1,75) = 107.80; p < 0.001). A significant interaction between age and block (F_(2,75) = 4.97; p < 0.01) revealed that although young rats outperformed aged rats on Bin 1 (p < 0.05) and middle-aged and aged rats on Bin 2 (all p values < 0.05) and middle-aged rats outperformed aged rats on Bin 2 (p < 0.05), all rats exhibited improved performances on the 2^nd versus 1^st bin (all p values < 0.01).

Figure 2C shows that swim speeds on visible platform trials are impacted by age (F_(2,75) = 16.23; p < 0.001) and training block (F_(4,300) = 9.03; p < 0.001) but the effect of the interaction between age and training block only approached statistical significance (F_(8,300) = 1.82; p = 0.07). On all trials combined, young rats swam more quickly than middle-aged and aged rats (p values < 0.001) and middle-aged rats swam more quickly than aged rats (p = 0.008). All rats combined swam more quickly as training progressed (block 1 > 2–5 and block 2 > 3; all p-values < 0.05). While young rats exhibited large increases in swim speeds across blocks (block 1,2 > 3–5; all p values < 0.05), middle-aged (block 1 > 2; p < 0.03) and aged (block 1 > 2, 5; all p values < 0.05) rats exhibited smaller more inconsistent changes. Age-related increases in body masses (young: 415.18 ± 11.95 g, middle-aged: 462.62 ± 7.72 g, aged: 457.89 ± 6.30 g; F_(2,75) = 6.07; p < 0.01) likely impact swim speeds.

Figure 2D shows statistically significant relationships between serum, hippocampal and cortical analyte concentrations and averaged visible platform pathlengths in young, middle-aged and/or aged rats. Note that shorter pathlengths on visible platform trials signify better performances. In all rats combined, average visible platform pathlengths correlated positively with serum MCP-1 (r_s (75) = 0.31; p = 0.006) and IL-5 (r_s (75) = 0.23; p = 0.05) levels, hippocampal IFN-γ (r_s (77) = 0.30; p = 0.001), IL-2 (r_s (76) = 0.34; p = 0.003), IL-5 (r_s (75) = 0.29; p = 0.01), IL-10 (r_s (76) = 0.31; p = 0.007), IL-17 (r_s (77) = 0.36; p = 0.001), RANTES (r_s (77) = 0.31; p = 0.005) and TNF-α (r_s (74) = 0.28; p = 0.02) levels and cortical eotaxin (r_s (76) = 0.25; p = 0.03), GRO/KC (r_s (76) = 0.38; p = 0.0008), IL-2 (r_s (77) = 0.33; p = 0.003), IL-4 (r_s (77) = 0.23; p = 0.04), IL-10 (r_s (74) = 0.32; p = 0.006), IL-18 (r_s (77) = 0.38; p = 0.0007), M1P-1α (r_s (76) = 0.40; p = 0.0003) and RANTES (r_s (76) = 0.35; p = 0.002) levels. Average visible platform pathlengths correlated negatively with serum TNF-α (r_s (77) = −0.23; p = 0.05) levels and hippocampal ACTH (r_s (78) = −0.22; p = 0.05), CORT (r_s (78) = −0.27; p = 0.02), IL-12 (r_s (75) = −0.30; p = 0.04), and VEGF (r_s (77) = −0.40; p = 0.0004) levels.

Next we looked at the relationship between visible platform performance and analyte levels in each age group independently. In young rats alone, average visible platform pathlengths correlated positively with serum MCP-1 (r_s (12) = 0.63; p = 0.03), hippocampal IL-9 (r_s (12) = 0.68; p = 0.02) and cortical MIP-1α (r_s (12) = 0.71; p = 0.009) levels but negatively with hippocampal IL-1α (r_s (13) = −0.59; p = 0.04) and IL-12 (r_s (9) = −0.67; p = 0.05) levels. In middle-aged rats alone, average visible platform pathlengths correlated positively with hippocampal IL-2 (r_s (40) = 0.49; p = 0.001), IL-5 (r_s (40) = 0.57; p = 0.0001), IL-10 (r_s (39) = 0.39; p = 0.013), IL-17 (r_s (40) = 0.49; p = 0.001) and melatonin (r_s (43) = 0.31; p = 0.02) levels and cortical G-CSF (r_s (39) = 0.42; p = 0.007) and IL-18 (r_s (39) = 0.34; p = 0.03) levels but negatively with serum IL-4 (r_s (41) = −0.45; p = 0.003), IL-6 (r_s (40) = −0.44; p = 0.005), IL-12 (r_s (30) = −0.37; p = 0.04), IL-13 (r_s (40) = −0.36; p = 0.02) and TNF-α (r_s (41) = −0.44; p = 0.004) levels, hippocampal MCP-1 (r_s (39) = −0.44; p = 0.005) levels and cortical IL-6 (r_s (39) = − 0.38; p = 0.02) and IL-13 (r_s (39) = −0.34; p = 0.03) levels. In aged rats alone, average visible platform pathlengths correlated positively with hippocampal IFN-γ (r_s (24) = 0.50; p = 0.01), IL-2 (r_s (23) = 0.53; p = 0.009), IL-5 (r_s (22) = 0.67; p = 0.0006), IL-6 (r_s (22) = 0.44; p = 0.04), IL-10 (r_s (24) = 0.48; p = 0.02), IL-17 (r_s (24) = 0.51; p = 0.01), and TNF-α (r_s (24) = 0.42; p = 0.04) levels and cortical IL-2 (r_s (24) = 0.47; p = 0.02) and IL-12 (r_s (12) = 0.60; p = 0.04) levels but negatively with serum leptin levels (r_s (24) = −0.47; p = 0.02), hippocampal ACTH (r_s (24) = −0.57; p = 0.003) and hippocampal (r_s (24) = −0.58; p = 0.003) and cortical (r_s (23) = −0.61; p = 0.002) CORT levels.

Finally, because we confirmed impaired memory in similar numbers of middle-aged and aged rats on the 24h retention probe trial, relationships between analyte levels and averaged visible platform pathlengths were quantified in these age groups combined. Average visible platform pathlengths correlated positively with hippocampal IFN-γ (r_s (64) = 0.41; p = 0.0007), IL-2 (r_s (63) = 0.51; p = 0.00002), IL-5 (r_s (62) = 0.54; p = 0.00007), IL-6 (r_s (62) = 0.32; p = 0.01), IL-10 (r_s (63) = 0.44; p = 0.0003), IL-17 (r_s (64) = 0.49; p = 0.00003), RANTES (r_s (64) = 0.27; p = 0.03) and TNF-α (r_s (62) = 0.41; p = 0.0009), levels, and cortical GRO/KC (r_s (63) = 0.33; p = 0.009) IL-2 (r_s (64) = 0.30; p = 0.02), IL-10 (r_s (61) = 0.30; p = 0.02), IL-18 (r_s (64) = 0.27; p = 0.03) and MIP-1α (r_s (64) = 0.27; p = 0.03) levels but negatively with serum G-CSF (r_s (64) = −0.30; p = 0.02), IL-6 (r_s (63) = −0.31; p = 0.01), IL-12 (r_s (43) = −0.33; p = 0.03), IL-13 (r_s (63) = −0.31; p = 0.01) and TNF-α (r_s (65) = −0.32; p = 0.01) levels, hippocampal ACTH (r_s (65) = −0.34; p = 0.006), CORT (r_s (65) = −0.31; p = 0.01) and VEGF (r_s (64) = −0.41; p = 0.0008) levels and cortical CORT (r_s (64) = −0.26; p = 0.03), IL-6 (r_s (61) = 0.31; p = 0.006) and IL-13 (r_s (62) = −0.31; p = 0.01) levels.

3.4. Spatial ability is compromised by age

Figure 3A shows hidden platform pathlengths over 5 training blocks and on a 6^th block administered after the immediate probe trial. Hidden platform pathlengths were affected by age (F_(2,75) = 5.95; p < 0.01) and training block (F_(4,300) = 48.75; p < 0.001) but not the interaction between age and trial block (F_(8,300) = 0.98; p = 0.45). Young rats swam more directly to the hidden platform than middle-aged and aged rats (p-values < 0.01) and all rats combined exhibited shorter pathlengths as training progressed (Block 1 > 2 > 3 > 4, 5; all p-values < 0.05). Planned comparisons confirmed that young, middle-aged and aged rats exhibited shorter pathlengths on the 5^th versus 1^st training block (all p-values < 0.001). Figure 3B shows pathlengths exhibited on Bins 1 (training block 1 and 2 combined) and 2 (training block 4 and 5 combined). Although young rats tended to outperform middle-aged rats (p = 0.06) and outperformed aged rats (p = 0.01) on the bins combined (F_(2,75) = 5.72; p < 0.01), all groups exhibited significantly shorter pathlengths on the 2^nd versus 1^st bin (F_(1,75) = 52.21; p < 0.0001). A dependent t-test confirmed that rats exhibited similar performances on the 5^th training block and on the 6^th refresher block administered after the immediate probe trial (t₍₇₇₎ = 1.35; p = 0.18).

Figure 3. Middle-aged and aged rats learned a hidden platform location almost as well as young rats.

Data are shown as group means (±S.E.M.) for young (white circles), middle-aged (gray triangles) and aged rats (black squares). On the bar graph, white bars depict performances on the 1^st bin (blocks 1 and 2) and black bars depict performance on the 2^nd bin (blocks 4 and 5) of visible platform trials. (A) Young rats swam more directly than middle-aged and aged rats to the hidden platform (p-values < 0.01) and all rats exhibited shorter pathlengths as training progressed (block 1 > 2 > 3 > 4, 5; all p-values < 0.05) and on the 5^th vs 1^st training block (all p-values < 0.001). (B) Young rats tended to outperform middle-aged rats (p = 0.06) and outperformed aged rats (p = 0.01) on all trial bins but all groups exhibited significantly shorter pathlengths on the 2^nd (blocks 4 and 5 combined) vs 1^st bin (blocks 1 and 2 combined; p < 0.0001). (C) On all trials combined, young rats swam more quickly than middle-aged and aged rats (p values < 0.001) and middle-aged rats swam more quickly than aged rats (p = 0.008) and all rats combined swam more quickly as training progressed (block 1 > 2–5 and block 2 > 3; all p-values < 0.05) because young rats exhibited large increases in swim speeds across blocks (block 1,2 > 3–5; all p values < 0.05) and middle-aged (block 1 > 2; p = 0.03) and aged (block 1 > 2, 5; all p values < 0.05) rats exhibited smaller more inconsistent changes. (D) Venn diagram showing serum, hippocampal and cortical analyte concentrations correlating positively (green line through the analyte) or negatively (red line through the analyte) with hidden platform pathlengths averaged across training blocks in young (green circle), middle-aged (yellow circle), aged (pink circle), all (green/yellow/pink overlap) and middle-aged and aged (yellow/pink overlap) rats. Analytes are coded for their known function (see legend).

Figure 3C shows that swim speeds were significantly affected by age (F_(2,75) = 13.76; p < 0.001) and training block (F_(4,300) = 3.89; p < 0.01) but not the interaction between age and training block (F_(8,300) = 1.23; p = 0.27). On all trials combined, young rats swam more quickly than middle-aged (p = 0.01) and aged rats (p = 0.0001) and middle-aged rats swam more quickly than aged rats (p = 0.002) and all rats combined swam more slowly as training progressed (block 1 > 2–5; all p-values < 0.05).

Figure 3D shows statistically significant relationships between hidden platform pathlengths averaged over the first 5 training blocks and analyte concentrations. Note that longer pathlengths indicate poorer performances. In all rats combined, average hidden platform pathlengths correlated positively with serum IL-9 (r_s (50) = 0.29; p = 0.04), IL-10 (r_s (73) = 0.32; p = 0.006), IL-18 (r_s (75) = 0.29; p = 0.01), IP-10 (r_s (75) = 0.33; p = 0.004), MCP-1 (r_s (75) = 0.25; p = 0.03) and MIP-1α (r_s (75) = 0.24; p = 0.04) levels, hippocampal GM-CSF (r_s (78) = 0.27; p = 0.02), IL-10 (r_s (76) = 0.24; p = 0.03), leptin (r_s (78) = 0.23; p = 0.04), melatonin (r_s (59) = 0.31; p = 0.02) and RANTES (r_s (77) = 0.29; p = 0.01) levels and cortical IL-1β (r_s (77) = 0.29; p = 0.009), IL-18 (r_s (77) = 0.31; p = 0.006), M1P-1α (r_s (76) = 0.29; p = 0.01), RANTES (r_s (76) = 0.33; p = 0.003) and VEGF (r_s (75) = 0.23; p = 0.05) levels.

We next tested the relationships between hidden platform performances and analyte concentrations in each age group independently. In young rats alone, statistically significant relationships did not emerge between analyte concentrations and averaged hidden water maze performances. In middle-aged rats, average hidden platform pathlengths correlated positively with serum IL-9 (r_s (25) = 0.46; p = 0.02), IL-18 (r_s (40) = 0.38; p = 0.02) and leptin (r_s (41) = 0.37; p = 0.017) levels, hippocampal CORT (r_s (41) = 0.32; p = 0.04), IL-2 (r_s (12) = 0.68; p = 0.02), IL-10 (r_s (39) = 0.33; p = 0.04), leptin (r_s (41) = 0.40; p = 0.01) and melatonin (r_s (31) = 0.43; p = 0.02) levels and cortical leptin (r_s (40) = 0.32; p = 0.04) levels but negatively with serum IL-4 (r_s (41) = −0.35; p = 0.03) levels and hippocampal IL-12 (r_s (25) = −0.43; p = 0.03) levels. In aged rats, average hidden platform pathlengths correlated negatively with serum IL-2 (r_s (23) = −0.52; p = 0.01) and leptin (r_s (24) = −0.46; p = 0.03) levels. In middle-aged and aged rats combined, average hidden platform pathlength correlated positively with serum IL-18 (r_s (62) = 0.52; p = 0.0004) levels, hippocampal GRO/KC (r_s (62) = 0.52; p = 0.01), IL-2 (r_s (63) = 0.23; p = 0.04), IL-10 (r_s (63) = 0.31; p = 0.01), leptin (r_s (65) = 0.27; p = 0.03) and RANTES (r_s (64) = 0.26; p = 0.04) levels and negatively with serum IL-13 (r_s (63) = −0.25; p = 0.05) levels and hippocampal IL-12 (r_s (36) = −0.34; p = 0.04) levels.

3.5. Aging rats exhibit impaired memory

Recall that DI scores > 0 on probe trials confirm better than chance performances with > 25% of time spent in the goal quadrant. On the 60s immediate strength of learning probe trial, age rats swam significantly slower (22.22 ± 0.60 cm/s) than either young (26.0 ± 0.81 cm/s; p < 0.001) or middle-aged (24.66 ± 0.46 cm/s; p < 0.001) rats (F_(2,75) = 8.29; p < 0.001) and therefore, aged rats also swam significantly shorter distances (1333.10 ± 35.93 cm) than either young (1557.78 ± 48.83; p < 0.001) or middle-aged (1479.48 ±27.49; p < 0.01) rats (F_(2,75) = 8.29; p < 0.001). We therefore, compared discrimination index scores that are insensitive to age-related differences in distances swum over a fixed time period on both the immediate and 24h probe trials. Figure 4A shows that young (DI scores = 0.65 ± 0.07), middle-aged (DI scores = 0.39 ± 0.04) and aged (DI scores = 0.40 ± 0.05) rats readily discriminated the goal versus opposite water maze quadrant on the probe trial administered immediately after training, but that young rats exhibited significantly higher discrimination index scores than either middle-aged (p = 0.01) or aged (p = 0.01) rats (H_(2,n=78) = 8.18; p < 0.05). Young rats also crossed the platform location (6.00 ± 0.56) significantly more times than middle-aged (3.49 ± 0.32; p = 0.001) or aged rats (3.17 ± 0.36; p = 0.0006; F_(2,75) = 9.52; p < 0.001), suggesting that young rats are capable of more precise navigation than older rats. Figure 4B confirms that young, middle-aged and aged rats spent a greater % of time in the goal versus opposite quadrant (all p values ≤ 0.05; F_(1,75) = 30512.1; p < 0.0001) but also shows that young rats spent significantly greater % of time in the goal quadrant than either middle-aged (p < 0.001) or aged (p < 0.001) rats (interaction effect: F_(2,75) = 1102.0; p < 0.01). Overall, these data show that while all rats learned the hidden platform location, young rats may have learned the location more precisely than middle-aged and aged rats.

Figure 4. Some aging rats exhibit impaired memory.

Data points show discrimination index scores on the immediate and 24h 60-s probe trials for young (white circles), middle-aged (gray triangles) and aged (black squares) rats and % Time spent in the water maze goal (black bars) or opposite (white bars) quadrant by young, middle-aged and aged rats. (A) Young rats exhibited significantly higher discrimination index scores on the probe trial administered immediately after training than either middle-aged (p < 0.05) or aged (p < 0.01) rats. (B) Young, middle-aged and aged rats spent more time in the goal versus opposite quadrant (all p values ≤ 0.05) but young rats spent significantly more time in the goal quadrant than middle-aged (p < 0.001) or aged (p < 0.001) rats. (C) On the memory probe trial, young rats also exhibited higher discrimination index scores than either middle-aged (p < 0.001) or aged (p < 0.001) rats on the probe trial administered 24h after the last training trial (D) and young rats spent significantly more time in the goal versus opposite quadrant (p ≤ 0.0001) and more time in the goal quadrant than either middle-aged (p < 0.01) or aged (p < 0.01) rats. (E) Venn diagram showing serum, hippocampal and cortical analyte concentrations correlating positively (green line through the analyte) or negatively (red line through the analyte) with memory probe trial DI scores in young (green circle), middle-aged (yellow circle), aged (pink circle), all (green/yellow/pink overlap) and middle-aged and aged (yellow/pink overlap) rats. Analytes are coded for their known function (see legend).

On the 24h memory probe trial, age rats swam significantly slower (22.81 ± 0.71cm/s) than either young (27.58 ± 0.96 cm/s; p < 0.001) or middle-aged (25.94 ± 0.54 cm/s; p < 0.001) rats (F_(2,75) = 9.71; p < 0.001) and therefore, aged rats also swam shorter pathlengths (1368.84 ± 42.40 cm) than either young (1655.06 ± 57.61; p < 0.001) or middle-aged (1556.49 ±32.44; p < 0.01) rats (F_(2,75) = 9.71; p < 0.001). Figure 4C shows that on average, young rats (DI scores = 0.37 ± 0.09) performed above chance levels but middle-aged rats (DI scores = 0.03 ± 0.05) and aged rats (DI scores = −0.05 ± 0.06) performed just above and just below chance levels, respectively. Young rats exhibited higher discrimination index scores than either middle-aged (p = 0.01) or aged (p < 0.0001) rats on the probe trial administered 24h after the last training trial (H_(2,n=78) = 13.65; p < 0.01). On this probe trial, young rats crossed the platform location (3.85 ± 0.44) significantly more times than either middle-aged (2.20 ± 0.23; p = 0.0006) or aged (1.71 ± 0.28; p = 0.0001; F_(2,75) = 8.08; p < 0.001) rats. Figure 4D confirms that young (but not middle-aged or aged) rats spent significantly more time in the goal versus opposite quadrant (p ≤ 0.0001; F_(1,75) = 1336.05; p < 0.01) and young rats spent significantly more time in the goal quadrant than either middle-aged (p < 0.01) or aged (p < 0.01) rats (F_(2,75) = 1399.33; p < 0.001). Overall, these data suggest that while all young rats recall a hidden platform location after a 24h retention period, many middle-aged and aged rats do not recall the platform location.

Figure 4E shows statistically significant relationships between DI scores on the 24h probe trial and serum, hippocampal and cortical analyte concentrations. Note that higher DI scores reflect better probe trial performances. In all rats combined, DI score correlated negatively with serum GRO/KC (r_s (77) = −0.33; p = 0.004), IL-1β (r_s (77) = −0.24; p = 0.04), IL-4 (r_s (76) = −0.25; p = 0.03) and MCP-1 (r_s (75) = −0.25; p = 0.03) levels, hippocampal GM-CSF (r_s (78) = −0.37; p = 0.0007), IL-9 (r_s (58) = −0.33; p = 0.01), melatonin (r_s (59) = −0.27; p = 0.04) and RANTES (r_s (77) = −0.30; p = 0.007) levels and cortical ACTH (r_s (78) = −0.25; p = 0.03), eotaxin (r_s (76) = −0.26; p = 0.02), IL-1β (r_s (77) = −0.34; p = 0.003), IL-4 (r_s (77) = −0.24; p = 0.04), IL-6 (r_s (74) = −0.25; p = 0.03), IL-9 (r_s (60) = −0.44; p = 0.0004), IL-18 (r_s (77) = −0.24; p = 0.03), melatonin (r_s (78) = −0.26; p = 0.02), MCP-1 (r_s (77) = −0.31; p = 0.005), RANTES (r_s (76) = −0.42; p = 0.0001) and VEGF (r_s (75) = −0.35; p = 0.002).

In young rats alone, DI score correlated negatively with serum IL-1β (r_s (13) = −0.63; p = 0.02) levels and cortical eotaxin (r_s (13) = −0.60; p = 0.03), GM-CSF (r_s (13) = −0.65; p = 0.02), IL-1α (r_s (13) = −0.56; p = 0.05), IL-4 (r_s (13) = −0.58; p = 0.04), IL-9 (r_s (13) = −0.71; p = 0.006), melatonin (r_s (13) = −0.83; p = 0.0004), TNF-α (r_s (13) = −0.62; p = 0.03) and VEGF (r_s (13) = −0.76; p = 0.002) levels but positively with hippocampal leptin (r_s (13) = 0.63; p = 0.02) levels. In middle-aged rats alone, statistically significant correlations between DI scores and serum analyte concentrations did not emerge. In aged rats alone, hippocampal GM-CSF (r_s (24) = −0.61; p = 0.001), IL-9 (r_s (17) = −0.57; p = 0.02), IP-10 (r_s (23) = −0.43; p = 0.04) and RANTES (r_s (23) = −0.48; p = 0.02) and cortical IL-18 (r_s (24) = −0.57; p = 0.003), MIP-1α (r_s (24) = −0.44; p = 0.03), RANTES (r_s (22) = −0.74; p = 0.00009) and VEGF (r_s (23) = −0.45; p = 0.03) correlated negatively with DI score.

In middle-aged and aged rats combined, DI score on the 24h retention probe trial correlated negatively with serum GRO/KC (r_s (64) = −0.27; p = 0.03) levels, hippocampal GM-CSF (r_s (65) = −0.40; p = 0.001), IL-9 (r_s (46) = −0.38; p = 0.009) and RANTES (r_s (64) = −0.33; p = 0.007) levels and cortical RANTES (r_s (63) = −0.32; p = 0.009) levels.

3.6. Some stress hormone, cytokine and chemokine concentrations vary by cognitive status in aging rats

Next we identified cytokine, chemokine and hormone concentrations in aging (middle-aged and aged) rats combined that varied by performance on the 24h memory probe trial. We categorized aging rats as ‘elite’ agers if their DI scores were at least as good the average young rat (DI scores ≥ 0.37), as ‘memory-unimpaired’ agers if their DI scores were better than chance but worse than the average young rat (DI scores > 0 but < 0.37) and as ‘memory-impaired’ agers if their DI scores were at or below chance levels (DI scores ≤ 0; Figure 4D). Figure 5 shows significant differences in blood serum (first panel), hippocampal protein (second panel) and cortical protein (third panel) analyte concentrations in ‘elite agers (n=10)’ versus ‘memory-unimpaired (n=21)’ agers and ‘memory-impaired (n=34)’ agers. Based upon this categorization, higher mean ± S.E.M. DI scores were exhibited by elite agers (0.49 ± 0.06) than by memory-unimpaired agers (0.17 ± 0.04; p < 0.001) and memory-impaired agers (−0.25 ± 0.03; p < 0.001) and higher DI scores were exhibited by memory-unimpaired agers than memory-impaired agers (p < 0.001; F_(2,62) = 84.55.33; p < 0.0001).

Figure 5. Some analytes vary in concentration according to cognitive status.

Middle-aged and aged rats were classified as either elite agers (performing as well as the average young rat; DI score ≥ 0.37), memory-unimpaired (performing better than chance; 0.37 > DI scores > 0) or memory-impaired (performing worse than chance; DI scores ≤ 0) agers based on their 24 h retention probe trial performance (see Figure 4). Significant differences in (A) serum, (B) hippocampal and (C) cortical analyte concentrations between elite agers and their memory-unimpaired and memory-impaired cohorts were tested using Mann-Whitney U tests. Black boxes show similar analyte concentrations between elite agers and their memory-impaired or –unimpaired cohorts, while green boxes show higher analyte concentrations in memory-impaired or -unimpaired rats versus elite agers.

Figure 5 shows that, relative to elite agers, memory-unimpaired agers had significantly elevated serum GRO/KC (median_elite = 797.00 and median_unimpaired = 984.05, U = 54.00, p < 0.05) levels, hippocampal IL-12 (median_elite = 9.25 and median_unimpaired = 17.16, U = 11.00, p < 0.05) levels and cortical IL-1α (median_elite = 0.92 and median_unimpaired = 2.13, U = 32.00, p < 0.01) and RANTES (median_elite = 10.25 and median_unimpaired = 15.11, U = 49.00, p < 0.05) levels.

Relative to elite agers, memory-impaired agers had significantly elevated serum GRO/KC (median_elite = 797.00 and median_impaired = 1114.11, U = 69.00, p < 0.01) and RANTES (median_elite = 17559.69 and median_impaired = 24914.35, U = 67.00, p < 0.01) levels, hippocampal GM-CSF (median_elite = 0.00 and median_impaired = 2.92, U = 84.00, p < 0.05) levels and cortical IL-9 (median_elite = 193.22 and median_impaired = 319.10, U = 37.00, p < 0.05) and RANTES (median_elite = 10.35 and median_impaired = 16.73, U = 61.00, p < 0.01) levels.

4. Discussion

Transient age-related changes in the concentrations of some analytes emerged earlier in middle-aged rats while some changes emerged later in aged rats (Table 1) and in some cases changes were coordinated, revealing potential signaling networks that may be sensitive to age (Fig. 1). Furthermore, some circulating and central analytes appear to be valid biomarkers of age-related cognitive decline because their concentrations changed across age and correlated with water maze learning (Figs. 2 and 3) and memory scores (Fig. 4). Indeed, we identified analytes that differed significantly in aging rats categorized as memory-unimpaired or –impaired relative to elite agers that performed as well as the average young rat (Fig. 5) based upon their 24h retention probe DI scores (Fig. 4). Each of these points is discussed below.

Cytokines and Chemokines

Generally, serum biomarker concentrations changed robustly with age. Consistent with human and rodent studies, the largest ~10–40-fold age-related increases were observed in serum IFN-γ, IL-1α, IL-1γ, IL-4, IL-6, IL-13, 1L-17 and IP-10 levels (Barrientos et al., 2015; Barrientos et al., 2003; Braida et al., 2004; Gerli et al., 2000; Mariani et al., 2006; Seidler et al., 2010; Villeda et al., 2011; Villeda et al., 2014; Zhao et al., 2010)^{for review} with more subtle but significant age-related changes were observed in serum CORT, GRO/KC, IL-5, IL-18, leptin, MCP-1, MIP-1α and RANTES levels (Galimberti et al., 2006; Lupien et al., 2009; Mawhinney et al., 2011; Petitto et al., 2002; Speisman et al., 2013a; Vallejo et al., 2011; Villeda et al., 2011; Villeda et al., 2014). To our knowledge, the age-related changes detected in serum IP-10 concentrations are novel.

Identifying serum makers that can predict behavioral or cognitive variability in aging individuals may facilitate early intervention strategies and reveal novel therapeutic targets. The concentrations of some serum analytes correlated with behavioral scores (IL-9, IL-12 and TNF-α) but their utility as biomarkers of age-related cognitive decline was limited by their stable concentrations across lifespan. Of analytes that changed in concentration across age groups (Table 1) and were associated with behavioral scores in middle-aged and aged rats, several inflammatory regulators may predict impaired cognition. For example, higher pro-inflammatory IL-18 levels related to poorer spatial learning scores (Fig. 3) and higher GRO/KC levels related to poorer memory scores (Fig. 4). Furthermore, higher GRO/KC levels were observed in memory-unimpaired and -impaired older animals versus elite agers that performed as well as the average young rat (Fig 5). These findings are consistent with previous work showing that elevated IL-18 levels are associated with age-related cognitive decline (Blalock et al., 2003; Mawhinney et al., 2011)that serum GRO/KC levels decline in aged rats that exercise daily and that exhibit better spatial ability (Speisman et al., 2013a). IL-13 levels changed across lifespan (Table 1) and higher IL-13 levels related to better spatial learning (Figure 3), suggesting that IL-13 may not only be a good biomarker of healthy cognition across lifespan but also a good pharmacotherapeutic replacement strategy target. Interestingly, IL-13 and IL-4 exhibit structural and functional homology and signal through a common pathway [Fig. 1 and (Ansel et al., 2006)] and factors exert anti-inflammatory and neuroprotective effects in the brain (Opal and DePalo, 2000; Ponomarev et al., 2007; Shin et al., 2004).

Interestingly, serum MIP-1α levels were unrelated to the behavioral scores of aging rats per se, but serum MIP-1α networked with hippocampal RANTES (Fig. 1) and higher hippocampal RANTES levels differentiated memory-impaired aging rats from their unimpaired and elite cohorts (Fig. 5) and were associated with poorer spatial learning and memory scores (Fig. 3) suggesting that serum MIP-1α levels could indirectly predict age-related cognitive decline. Elevated serum MIP-1α levels has been implicated in a handful of studies to be associated with impaired cognition (Stuart and Baune, 2014). Some serum factor concentrations also related to non-spatial behavioral performances. For example, higher pro-inflammatory IL-6 and IL-13 levels related to better visual discrimination performance scores. Our finding that higher IL-6 levels were associated with better non-spatial learning in aging rats adds complexity to the emerging picture of detrimental and beneficial effects of IL-6 cognition across lifespan (Braida et al., 2004). How circulating inflammatory cytokines might promote visual discrimination is unclear but one possibility is that impaired hippocampal function may enhance reliance on other cognitive systems including habit or cue learning (McDonald and White, 1994; Packard and McGaugh, 1992; Kumar et al., 2011).

Because hippocampus-dependent spatial memory declines with age (Eichenbaum, 1998; Ergorul and Eichenbaum, 2004; Foster et al., 2012; Morris et al., 1982), we predicted that candidate hippocampal biomarker concentrations would better correlate with spatial learning and memory performance scores than frontal cortical biomarker concentrations. Across lifespan, the largest increases were observed in pro-inflammatory IL-9 and RANTES concentrations and relatively modest decreases were observed in pro-inflammatory IL-12 and anti-inflammatory IL-5 concentrations and a small but significant increase in MIP-1α was detected (Table 1). Of these analytes, higher hippocampal RANTES concentrations related to poorer spatial learning and along with IL-9, poorer memory scores. RANTES receptors are expressed by hippocampal neurons and RANTES both modulates calcium signaling and apoptosis in cultured hippocampal neurons (Meucci et al., 1998), which likely impacts cognition. To our knowledge, the effects of hippocampal IL-9 on cognition are unknown. In contrast, lower IL-12 levels related to poorer spatial learning. Although lower serum IL-12 levels have been associated with better cognitive outcomes in aging individuals (Baune et al., 2008), serum IL-12 levels were neither related to central IL-12 levels nor other central factors associated with poor performance scores. Finally, some hippocampal analyte concentrations correlated with visual discrimination (ACTH, CORT, IFN-γ, IL-6, IL-10, IL-17, TNF-α, VEGF; Fig. 2D), spatial learning (GRO/KC, IL-2, IL-10; Fig. 3D) and memory (GM-CSF; 4D) scores, but their stable concentrations across lifespan (Table 1) limit their utility as biomarkers of age-related cognitive decline. Overall, the results show, that like serum inflammatory biomarkers, changes in the concentrations of some hippocampal neuroinflammatory markers may influence or predict spatial ability and revealed that hippocampal IL-12 may be a good pharmacotherpeutic replacement target.

Frontal cortical analytes were quantified because of possible corticolimbic contributions to spatial behavior (Banuelos et al., 2013; Seamans et al., 1998), the influence of frontal cortical circuitry in procedural learning that is required to learn the rules of solving the water maze task (Della-Maggiore et al., 2000; Vorhees and Williams, 2006) and to examine whether age-related changes in neuroinflammatory signaling may be regional. Age-related increases in the concentrations of anti-inflammatory (IL-4 and IL-10) markers, pro-inflammatory (IL-1β, IL-5, IL-6, IL-18) markers, recruitment/trafficking (eotaxin, Gro/KC, MCP-1, MIP-1α, RANTES) markers and VEGF were detected in the frontal cortices of rats. Interestingly, higher frontal cortical RANTES concentrations were related to poorer non-spatial learning scores and poorer spatial learning and memory scores, which suggests that either potentially networked regional changes or diffuse changes in RANTES levels modulate corticolimbic circuitry that mediates non-spatial and spatial behaviors. Poorer visible platform performances were exhibited by rats with higher frontal cortical IL-10, IL-18, GRO/KC and MIP-1α concentrations (Fig. 2D) and better visual discrimination performances were exhibited by rats with higher cortical IL-6 concentrations. Cortical concentrations of pro-inflammatory IL-2 and anti-inflammatory IL-13 correlated with the visible platform performance scores of aging rats but their predictive utility was limited by their stable concentrations across lifespan. Interestingly, serum MCP-1 levels were unrelated to behavioral scores per se, but serum MCP-1 networked with cortical IL-18 (and MIP-1α) and higher cortical IL-18 levels were associated with poorer visual discrimination scores, suggesting that serum MCP-1 levels could indirectly predict age-related cognitive decline. The reliability and robustness of some of the small but significant age-related changes in concentrations of cytokines that also correlated with behavioral scores should be confirmed in future work that also tests their biological relevance (e.g. cortical ACTH, IL-4, IL-5, MIP-1α and hippocampal GM-CSF and MIP-1α). Overall, the results suggest that serum inflammatory biomarkers are predictive of cognitive decline and that hippocampal biomarkers (e.g. RANTES and IL-9) better describe impaired spatial memory while several cortical biomarkers better describe altered procedural learning.

Hormones

Changes in HPA axis activity across lifespan are well-documented and glucocorticoids are hypothesized to act upon the hippocampus to impair hippocampus-dependent behavior (Issa et al., 1990; Lupien et al., 1998; Lupien et al., 2009; McEwen, 1998; Ogle et al., 2013; Sandi and Touyarot, 2006). Consistent with this body of work, serum CORT levels changed across lifespan. Hippocampal and cortical CORT and ACTH levels were stable across age, but CORT levels exhibited high variability within each age group and tissue examined. Higher hippocampal CORT levels related to poorer spatial learning scores (Fig. 3). On the other hand, higher hippocampal ACTH and hippocampal and cortical CORT levels related to better cue discrimination performances (Fig. 2), which is consistent with the notion that while age-dysregulated HPA activity impairs spatial behavior, CORT promotes the use of multiple memory systems and that stress and glucocorticoids can facilitate non-spatial behavior (Issa et al., 1990; Lupien et al., 1998; Lupien et al., 2009; McEwen, 1998; Sandi and Touyarot, 2006; Schwabe et al., 2012).

Previous work indicates that serum leptin responds to inflammatory stimuli (Mastronardi et al., 2005; Sarraf et al., 1997), can stimulate neuroinflammation (Hosoi et al., 2002a, b), correlates negatively with memory scores and rates of hippocampal neurogenesis in aged rats (Speisman et al., 2013a) and compromises forms of plasticity (Fadel et al., 2013). Here, serum and cortical leptin levels were higher in middle-aged rats and correlated with poorer spatial learning in this group. In contrast, hippocampal leptin levels did not vary by age (Table 1); however, higher hippocampal leptin levels were associated with poorer spatial learning across the two older groups (Fig. 3). Thus, although leptin has been implicated as biomarker of age-related cognitive decline, our current data suggest that leptin’s utility as a biomarker requires further validation.

Biomarkers of age-related memory loss

Potential biomarkers of age-related memory loss were confirmed by comparing analyte concentrations in the serum and brains of rats categorized as elite agers (performing as well as the average young rat) versus aging rats categorized as memory-unimpaired (performing better than chance but worse than the average young rat) or memory–impaired (performing worse than chance) based upon their 24h probe trial performances (Fig. 5). Relative to elite agers, unimpaired agers had higher serum GRO/KC levels, hippocampal IL-12 levels and cortical IL-1α and RANTES levels. Relative to elite agers, impaired agers had higher GRO/KC and RANTES levels, hippocampal GM-CSF levels and cortical IL-9 and RANTES levels. Disentangling whether age-related increases in single factors (i.e. serum GRO/KC and cortical RANTES) or whether concomitant age-related increases in several cytokines (i.e. serum GRO/KC and RANTES and cortical IL-9 and RANTES) leads to cognitive impairment is an important goal of future work. The finding that serum GRO/KC levels are lower in elite agers versus their cohorts is consistent with our correlational data and published data (Speisman et al., 2013a) supporting that GRO/KC could be a valid biomarker of age related cognitive decline and memory loss. Higher hippocampal GM-CSF levels were detected in memory-impaired versus –unimpaired aging rats, which is consistent with published work showing that hippocampal GM-CSF levels can impact spatial memory (Krieger et al., 2012). However, the utility of GM-CSF as a biomarker of age-related memory loss should be validated in future experiments because we did not find statistically significant age-related changes in the low and variable GM-CSF levels detected in the hippocampus (Table 1). Hippocampal IL-12 levels decreased across lifespan and were lower in elite agers versus their unimpaired cohorts and lower in aging rats that exhibited better spatial and non-spatial learning scores. Interestingly, hippocampal IL-12 was networked independently with hippocampal IL-5 and cortical MIP-1α (Figure 1). Elevated concentrations of both of these factors were associated with poorer spatial learning (Figure 2).

Conclusions

Chronic inflammatory responses stimulated by pathogen exposure, excess fat tissue, poor hygiene, smoking, reduced sex steroid production and chronic disorders including atherosclerosis have been hypothesized to produce cognitive decline in elderly individuals (Chassagne et al., 1996; Green et al., 2014; Krabbe et al., 2004; Rooney, 2014; Vasto et al., 2007). Indeed, lifelong or repeated pathogenic insults can dysregulate peripheral pro-inflammatory cytokine milieus (Banks et al., 2014; Batty et al., 2013; Cambier, 2005; Green et al., 2014; Krabbe et al., 2004; Lockhart et al., 2009; Rooney, 2014; Vasto et al., 2007; Wårdh and Wikström, 2014). Blood-borne and neural routes of communication between the peripheral and central nervous systems have been well-defined (Dinarello et al., 1988; Ericsson et al., 1994; Maier, 2003) and systemic immune challenge dramatically alters neural activity (Barrientos et al., 2015; Chapman et al., 2010; Maier and Watkins, 1998) and ablates the production of new highly excitable hippocampal granule neurons (Chen et al., 2011; Ekdahl et al., 2003; Ekdahl et al., 2009; Monje et al., 2002; Monje et al., 2003; Ormerod et al., 2013). Systemic inflammation can stimulate the de novo synthesis of brain parenchymal cytokines primarily by microglia (Layé et al., 1996; Nguyen et al., 1998; Van Dam et al., 1995) but also by other CNS cells (Liu et al., 2014; Vincent et al., 1998). In turn, neuroinflammation can impact the production of systemic cytokines (Campbell et al., 2007). A plausible hypothesis is that chronic or repeated immunostimulation across lifespan leads to dysregulated immune and/or neuroimmune signaling and subsequent age-related cognitive decline. Note that the rats used in the current study were housed in standard colony rooms that may permit pathogen exposures at a higher probability than barrier facilities would.

Identifying cytokines and chemokines linked to the onset and progression of disease may provide insight about the nature of chronic inflammation and their cellular players that negatively impacts aging and cognition. In middle-aged and aged rats combined, several serum (GRO/KC, IL-6, IL-13 and IL-18), hippocampal (IL-5, IL-9 and IL-12) and cortical (GRO/KC, IL-6, IL-10, IL-18, MIP-1α and RANTES) analytes emerged as potential candidates for testing in future experiments. Note that higher levels of serum IL-13 and in some cases serum IL-6, hippocampal IL-12 and cortical IL-6 biomarkers were associated with better behavioral performances. In particular, serum GRO/KC and RANTES and hippocampal GM-CSF and cortical IL-9 and RANTES may be validated as good candidate biomarkers of age–related memory loss because they differed in concentration between elite agers and aging rats categorized as memory-impaired based upon their 24h retention probe trial performances. For example, some but not all of these cytokines have been associated with dysregulated Type 2 immune responses that drive disease (Wynn, 2015). The extent or speed of progression of age-related cognitive decline may depend upon the unique interactions between several cytokines/chemokines that are impacted by age.

Discovering whether inflammatory and neuroinflammatory biomarkers that we and others have identified interact to compromise cells or tissues is important for developing strategies to combat their effects on cognition across lifespan. Chronic systemic inflammation and chronic neuroinflammation can compromise glia and synaptic mechanisms thought to underlie memory (Barrientos et al., 2015; Chapman et al., 2010; de Haas et al., 2007; de Rivero Vaccari et al., 2014; Foster, 1999; Godbout and Johnson, 2009; Hauss-Wegrzyniak et al., 2002)^{for review}. Hippocampal neurogenesis has been linked to cognition, can decline with age and is ablated by neuroinflammation (Belarbi et al., 2012; Ekdahl et al., 2003; Kuhn et al., 1996; Monje et al., 2003; Monje et al., 2007; Ormerod et al., 2013; Speisman et al., 2013a; Valero et al., 2014). Finally chronic age-associated neuroinflammation may render microglia dystrophic and primed to respond to repeated insult leading to disrupted feedback that compromises their typically protective and regenerative effects on neurons (Barrientos et al., 2015; Chapman et al., 2010; Streit, 2006; Streit et al., 2008; Streit et al., 2004) and other CNS cell types (de Haas et al., 2007; de Rivero Vaccari et al., 2014; Mawhinney et al., 2011). For example, microglia that shift from a resting to activated state shift from producing growth factors like BDNF that promote learning, memory and synaptic plasticity to cytokine production (Barrientos et al., 2015; Barrientos et al., 2003; Chapman et al., 2010; Cotman et al., 2007). Whether biological outcomes are produced by single initiating signals or interactions between cytokines and chemokines that impair cellular mechanisms is currently unclear.

• Similar numbers of middle-aged and aged rats exhibited memory probe impairments

• Some serum, hippocampal and cortical analyte levels quantified changed across lifespan

• Candidate serum biomarkers predicted spatial and non-spatial scores in aging rats

• Cortical biomarkers generally predicted non-spatial behavioral scores in aging rats

• Hippocampal biomarkers predicted spatial scores in aging rats