Age-related changes in microRNA levels in serum

miRNA expression in young and old individuals using deep gene sequencing

In order to study age-related changes in miRNA expression in human serum, we used Illumina small-RNA next generation sequencing (NGS) technology. We obtained serum from 11 young and 11 old individuals selected from a sub-cohort of the Healthy Aging in Neighborhoods of Diversity across the Lifespan (HANDLS) study. These participants were used in our previous examination of miRNA expression changes with age in PBMCs [14]. Demographic information for these participants is presented in Table 1. Interestingly, miRDeep2 software, which aligns sequences to both mature and precursor miRNAs, revealed that the majority (87%) of miRNA sequences in human serum were precursors. miRDeep2 detected 23 miRNAs in our samples, five miRNAs of which were found to be present in the serum from both young and old individuals (Figure 1A). Importantly, miR-181a-1, miR-1248 and miR-3607 were significantly lower in old individuals (Figure 1A). Several miRNAs had only one read in one individual and thus were excluded from further analysis (Figure 1B).

Figure 1. Identification of miRNAs in serum from young and old individuals.

(A) Frequency of miRNA reads per individual using small RNA NGS from 11 young (~30 yr) and 11 old (~64 yr) individuals for miR-181a-1, miR-3607, miR-1248, miR-151a and miR-151b. Horizontal lines represent average number of reads. miR-181a, miR-1248 and miR-3607 were significantly downregulated in serum from old individuals using Poisson regression (miR-181a-1 P=0.03, miR-3607 P=0.01, miR-1248 P=4.9×10⁻⁹). (B) miRNAs identified that had only one read per individual.

Validation of miRNA expression changes in young versus old individuals

To validate our sequencing results, we employed real-time RT-PCR using sequence-specific miRNA primers to analyze the expression levels of the five miRNAs with the highest number of sequencing reads in our original cohort (n=11 for each group). In addition, we expanded this initial cohort to include more young (n=20; mean age 30.6) and old (n=20; mean age 64.4) individuals. Demographic information for this validation cohort is in Table 1. Three of the miRNAs we analyzed, miR-151a, miR-181a-1 and miR-3607, are precursors that can be processed into two mature miRNAs according to miRBase (miR-151a-3p and −5p, miR-181a-3p and −5p and miR-3607-3p and −5p). Therefore, in our validation experiments, we tested both 3p and 5p mature sequences and also have referred to these miRNAs using the nomenclature in miRBase. In addition, miR-151b has the same sequence as miR-151a-5p and thus, we considered these miRNAs to be the same.

In order to evaluate the variation of miRNA expression among our initial cohort of 11 young and 11 old participants, we analyzed the expression of eight miRNAs in serum using reverse transcription (RT) followed by real-time, quantitative (q)PCR. Expression levels among young and old individuals were consistent with a Pearson's correlation coefficient (r) close to 1 for the all samples (Figure 2). Similar data were also observed with the entire cohort of 40 (data not shown). Comparison of the C_t values for these miRNAs showed low variability among the young and old groups (Figure 2). These data suggest that miRNA expression is reproducibly consistent among young individuals and among old individuals.

Figure 2. Comparison of expression variance of 8 miRNAs in 11 young and 11 old individuals by real-time RT-PCR.

RNA isolated from serum was reverse transcribed and real-time RT-PCR was performed with miRNA specific primers. Data was normalized to miR-191. C_t values of each miRNA for each young (A) and old (C) participant. C_t values were used for analysis of the Pearson correlation coefficient of miRNA expression between young participant samples (B) and between old participant samples (D). R values are indicated and a value of 1 indicates perfect correlation.

Through this analysis, we observed differences in miRNA abundance between young and old individuals for several miRNAs. miR-181a-5p and miR-1248 were significantly decreased in old individuals (Figure 3). In addition, although our sequencing results showed increased levels of miR-151a/b in old individuals, our RT-qPCR analysis in the larger cohort of individuals showed that miR-151a-3p is significantly lower in old individuals. Lower levels of miR-151a-5p, miR-181a-1-3p, miR-3607-3p and miR-3607-5p were also observed in old individuals, though not significantly. Consistent with our sequencing data, we found that miR-181a-5p was by far the most highly expressed serum miRNA among the miRNAs that we analyzed (Figure 3), confirming that our sequencing data gave a quantitative assessment of miRNA abundance. In addition, the expression of many of these age-associated miRNAs was highly correlated (Figure 2 and Table 2). Given the recent data that serum miR-21 increases in abundance with age [25], we also examined miR-21 levels in young and old individuals. miR-21-3p expression was not different between our age groups and miR-21-5p levels decreased, albeit not significantly, in our older cohort. These differences are likely due to the significant differences in the age span of the young and old cohorts for the Olivieri et al. study. The young cohort in that study ranged in age from 24-64 which encompasses both our young and old cohort while their old cohort is enriched with significantly aged individuals, including 30 centenarians.

Figure 3. Real-time RT-PCR validation of miRNA expression in serum from young and old participants.

The expression levels of miRNAs identified by sequencing analysis were validated in 20 young and 20 old participant serum samples (see Table 1 for demographic information) using real-time RT-PCR. Sequence specific primers for the indicated miRNAs were used and the data was normalized to miR-191. Box and whisker plots for each miRNA are shown. Whiskers represent ± the standard deviation, the box extends to upper and lower quartiles, lines represent the median and closed circles represent the mean. *P<0.05 comparing young and old by Student's t-test.

Age-dependent changes in miRNA expression in non-human primates

We wanted to assess whether the miRNAs that we found in human serum were also expressed in other long-lived species that are used as models of human aging [26]. Therefore, we examined levels of these seven miRNAs in the serum of 10 young (mean age = 7.7±1.5 years) and 10 old (mean age = 21.7±2.1) rhesus monkeys (Macaca mulatta). As rhesus monkeys age at approximately three times the rate of humans, the young and old groups of monkeys correspond approximately with the average ages of our young and old human cohort, ~23 for young and ~65 for old. Importantly, we found that all the miRNAs we found expressed in human serum were also present in monkey serum (Figure 4). In addition, the expression levels of these miRNAs are fairly consistent between monkeys and humans in that miR-181a-5p is highly expressed, while other miRNAs are expressed at lower levels. We found significantly lower levels of miR-151-5p and miR-1248 and a close to significant (p=0.09) decrease in miR-181a-5p levels with monkey age (Figure 4).

Figure 4. Relative expression of 7 miRNAs in rhesus monkey serum.

RNA was isolated from serum from ten young male monkeys (mean age = 7.7 yrs) and ten old male monkeys (mean age = 21.7 yrs). Real-time RT-PCR was performed with miRNA specific primers and normalized to miR-191. The histogram represents the ±SEM. *P<0.05 and #P=0.09 using Student's t-test.

Target and Pathway Analysis for serum age-associated miRNAs

To understand more thoroughly how these miRNAs may contribute to the aging process, we examined several mRNA targets of miR-151a-3p, miR-181a-5p and miR-1248 as predicted using TargetScan 6.2. We identified 115, 626 and 265 mRNAs that were predicted to be targeted by miR-151a-3p, miR-181a and miR-1248, respectively (Figure 5A). No predicted targets overlapped between the 3 different miRNAs, however, there was some overlap between the different combinations of 2 miRNAs (Figure 5A). In addition, we used the targets from TargetScan to input into Ingenuity Pathway Analysis to reveal the top 5 different diseases and disorders, molecular and cellular functions, physiological system development and function, and canonical pathways associated with each miRNA (Figure 5B). We also performed network analysis and listed the central mediator for each of the top five networks (Figure 5B).

Figure 5. Predicted targets and pathways for the indicated age-associated miRNAs.

Predicted targets for the 3 miRNAs were obtained using TargetScan software and the overlapping targets were visualized using a Venn diagram (A). The overlapping targets are indicated. (B) The predicted targets from TargetScan for each miRNA were used for Ingenuity Pathway Analysis. The top five pathways for each parameter are listed and where applicable the number of genes (#) for the particular pathway. For each parameter listed, P<0.05. Dev.=development.

Cancer and neurological disease overlapped among all three miRNAs, consistent with the fact that age is a significant risk factor for these two diseases. All three of these miRNAs are predicted to function in regulating gene expression, which is intriguing given that the age-associated miRNAs in our previous study of PBMCs were also predominantly predicted to target gene expression pathways [14]. Several other molecular and cellular functions overlapped between these 3 miRNAs including cellular development, cell morphology, and cellular assembly and organization. Interestingly, organismal survival was the top physiological pathway for each of the miRNAs and several predicted target genes involved in survival pathways overlapped between at least two of the three miRNAs. There was no overlap between the canonical pathways; however, in the predicted top networks, central mediators including ERK1/2, Akt, NF-κB and actin overlapped among these miRNAs. This analysis suggests that miR-151a-3p, miR-181a-5p and miR-1248 may target similar, yet distinct signaling pathways important for the aging process.

Since the top network mediator for miR-181a is TNFα and because this miRNA has been shown to target several other inflammatory markers [27-29], we hypothesized that the decline in miR-181a with age that we found may be a cause of the increased inflammation that occurs with old age. Therefore, we examined several inflammatory markers that have been reported to be targeted by miR-181a in serum from the young and old individuals of our cohort. We found that miR-181a expression was significantly negatively correlated with the pro-inflammatory cytokines IL-6 and TNFα and positively correlated with the anti-inflammatory cytokine TGFβ (Table 2). Various other age-associated serum miRNAs were also correlated with IL-6, TNFα and TGFβ levels (Table 2). Other inflammatory miR-181a targets, IL-1α, IL-1β, IL-8 and LIF, were found to be low or undetectable in the serum from our young and old individuals. Although the anti-inflammatory cytokine IL-10 was only detected in 6 participants (3 young and 3 old), it was highly correlated with miR-181a-5p levels (R=0.84, P<0.01). Low levels of circulating IL-10 is consistent with other reports that this anti-inflammatory marker is induced and not necessarily expressed under non-stimulated conditions [30, 31]. These data suggest that the downregulation of miR-181a we observed with age in serum correlates with an upregulation of pro-inflammatory cytokines and downregulation of anti-inflammatory cytokines.

Given that there is very little known about miR-1248, we sought to determine the potential targets and pathways regulated by this miRNA. Therefore, we overexpressed a miR-1248 mimic in HeLa cells and performed microarray analysis in order to determine which genes and pathways are differentially changed by miR-1248 overexpression (Figure 6). This analysis showed that 1740 mRNAs were downregulated and 1770 mRNAs were upregulated by miR-1248 overexpression (Suppl. Table S1). Since most mRNA targets are degraded by miRNAs, we focused on the top downregulated mRNAs from our microarray (Figure 6B). Interestingly, two of these mRNAs are the age-associated cytokines IL6 and IL8. We further validated by real-time RT-PCR that indeed these cytokines are decreased in abundance in miR-1248 overexpressing cells (Figure 6C). In addition, we further validated other genes that were downregulated in our microarray analysis (Figure 6C). Differentially expressed genes and Z ratios were categorized into known canonical pathways by the Parametric Analysis of Gene-set Enrichment (PAGE). The top canonical pathways that were downregulated in miR-1248 overexpressing cells are shown (Figure 6D). All canonical pathways regulated by miR-1248 are shown in the heat map in Supplemental Figure S1. Several cytokine and inflammatory-associated pathways, including NF-κB, are downregulated in miR-1248 overexpressing cells (Figure 6D). Interestingly, DNA repair pathways are upregulated in miR-1248 overexpressing cells (Suppl. Figure S1), suggesting that decreased levels of miR-1248 with age may influence the impaired DNA repair capacity that is observed in older individuals [32]. In addition, the mRNAs that were significantly changed in miR-1248 overexpressing cells were imputed into Ingenuity to obtain a miR-1248 putative target list and compared with targets for miR-1248 in TargetScan. There were 18 genes that overlapped between the 2 algorithms and our microarray data (Figure 6E). Two of these potential targets, GAPVD1 and CDK2, we validated were downregulated in miR-1248 overexpressing cells (Figure 6C). Furthermore, the miR-1248 mimic that we have used for overexpression is biotin labeled. Therefore, we performed biotin-labeled precipitations followed by RNA isolation and RT-qPCR to identify mRNAs bound to miR-1248. We found that CDK2, IL6 and IL8 mRNAs were enriched in our biotin miR-1248 pulldowns; whereas GAVD1 mRNA was not (Figure 6F). These data indicate that miR-1248 can bind to CDK2, IL6 and IL8 mRNAs.

Figure 6. Pathway and Target analysis for miR-1248.

(A) HeLa cells transfected with Con-miR or miR-1248 were analyzed 48 hrs after transfection for miR-1248 expression using real-time RT-PCR. Data was normalized to U6 expression and the histogram represents the mean + SEM from 3 different experiments. (B) Total RNA from (A) was analyzed using genome-wide Illumina microarrays. The top downregulated genes are listed and the corresponding numerical complete dataset is presented in Suppl. Table S1. (C) RT-qPCR validation of microarray results using mRNA-specific primer pairs. (D) Top canonical pathways for genes downregulated by miR-1248 overexpression are shown on the heatmap. Red arrows indicate inflammatory and cytokine pathways. (E) Venn diagram of putative targets from TargetScan, Ingenuity and miR-1248 microarray analysis. Overlapping mRNAs are listed. (F Biotinylated miR-1248 or Con-miR were transfected into HeLa cells and were precipitated using streptavidin beads. Associated mRNAs were isolated and gene-specific primers were used for RT-qPCR.

Study Participants

Fasting blood samples were obtained from participants in the Healthy Aging in Neighborhoods of Diversity across the Life Span (HANDLS) study of the National Institute on Aging Intramural Research Program (NIA IRP), National Institutes of Health (NIH). Samples were collected in vials with no additives, centrifuged and serum was collected, aliquoted and immediately frozen at −80°C until use. HANDLS is a longitudinal, epidemiologic study based in Baltimore, MD that seeks to determine the source of age-associated health disparities by examining the relationship between race, socioeconomic status and health [39]. Participants are Whites and African Americans between the ages of 30-64 at baseline residing in Baltimore, MD. This study has been approved by the Institutional Review Board of the National Institutes of Environmental Health Sciences, NIH and all participants provided written informed consent. The demographics of the cohort used in this study are presented in Table 1. All data presented were collected from participant interviews and self-reported medical histories. For this cohort, we excluded participants with documented Hepatitis B, Hepatitis C or human immunodeficiency virus (HIV) infection. Inflammatory cytokines were measured in serum using SearchLight immunoassays by Aushon Biosystems (Billerica, MA).

Non-human Primates

Blood samples were collected from anesthetized (Ketamine 7-10 mg/kg, IM or Telazol 3-5 mg/kg, IM) rhesus monkeys (Macaca mulatta) after an overnight fast. Serum was aliquoted and immediately frozen at −80°C until use. The monkeys were control animals on NIA IRP studies under protocols approved by the NIA Animal Care and Use Committee. Monkeys were considered healthy at the time of sample collection.

Next Generation Sequencing

Total RNA was isolated from 400 μL of serum from 11 young (30 yrs) and 11 old (64 yrs) HANDLS participants using Trizol LS (Life Technologies) according to manufacturer's instructions with the addition of a second phenol/chloroform extraction prior to RNA precipitation in order to reduce protein and DNA contamination. Samples were resuspended in 15 μL RNase-free water and frozen at −80°C until further use. In order to identify miRNAs present in serum, we used Illumina's NGS technology. Total RNA samples, as isolated above, were prepared for sequencing using the TruSeq small RNA sample prep kit (Illumina) with multiplexing according to manufacturer's instructions. Briefly, 3' and 5' RNA adapters were ligated to small RNA molecules. RNA was then reverse transcribed and PCR amplified. Prior to amplification, indices to allow for sample multiplexing were incorporated. cDNA was then validated using an Agilent Technologies Bioanalyzer with a high sensitivity DNA chip. Samples in the same multiplex group were pooled and run on a 6% SDS-PAGE gel to purify. After gel purification, small RNA libraries were again validated using an Agilent Technologies Bioanalyzer high sensitivity DNA chip and loaded onto a flow cell for sequencing. Data was analyzed using miRDeep software (v2)[40]. Illumina adapter sequences were clipped and any short sequence tags of length less than 17 bases were removed. miRNA sequences were aligned to human genome (hg19) with no more than 1 base mismatch in the seed region using mirDeep mapper program which uses the popular bowtie algorithm [41]. miRNAs were identified using the MirDeep2 program using human miRBase-19 database [42] of both mature and miRNA precursors. All matches and predicted novel miRNAs were reported and analyzed further by randfold program [43] to identify the folding free energies. Differences in expression of miRNAs identified by sequencing were analyzed in R using Poisson regression.

Reverse Transcription (RT) followed by real-time, quantitative PCR analysis

Total RNA was isolated from 25 μL of serum from 20 young and 20 old HANDLS participants as well as 10 young and 10 old rhesus monkeys using Trizol LS (Invitrogen) according to the manufacturer's instructions with the following modifications. After precipitation, RNA was re-suspended in 85 μL RNase-free water, 10 μL 10x DNase buffer and 5 μL DNase I (RNase free, Ambion) and incubated at 37 °C for 30 min. 200 μL NT2 buffer [44] and 300 μL Acid:Phenol CHCl₃ (Ambion) were then added and samples were vortexed for 1 min and centrifuged at room temperature for 5 min. 250 μL of the top layer was extracted and precipitated overnight with ethanol, sodium acetate and GlycoBlue (Ambion) at final concentrations of 70% ethanol, 80 mM sodium acetate and 3×10⁻⁵ mg/mL GlycoBlue. Samples were then incubated 30 min at 4°C, washed with 75% ethanol, air-dried and resuspended in 15 μL RNase-free water. Samples were frozen at −80°C until further use.

For each participant, 10 μL of RNA was reverse transcribed using the QuantiMir™ cDNA Kit (System Biosciences, Mountain View, CA) according to manufacturer's directions. Real-time RT-PCR reactions were performed using 2x SYBR Green Master Mix (Applied Biosystems) according to the manufacturer's directions on an Applied Biosystems 7500 Real-Time PCR System. In order to amplify cDNA generated from serum samples, we used the Universal Reverse Primer provided by the QuantiMir kit and miRNA specific forward primers designed to match the exact sequences of each miRNA analyzed. We normalized to miR-191 in serum as it was expressed in all participant serum, did not change with age and was the least variable normalization miRNA in all individuals. Data was analyzed using Minitab 16 software.

Microarray and validation of microarray results

HeLa cells were transfected with 20 nM biotinylated miRIDIAN miR-1248 mimic (Thermo Scientific) or a negative control, biotinylated miRIDIAN C. elegans miR-67 (here named Con-miR) using Lipofectamine 2000. Forty eight hours post transfection, total RNA was isolated using the mirVana kit (Invitrogen) according to manufacturer's procedure. Three different experiments were performed and a technical repeat was included for each sample. Samples were analyzed using Illumina's Sentrix Human HT-12 ver4 Expression BeadChips (Illumina, San Diego, CA). Details about the microarray procedure and data analysis are in the Supplemental Methods.

To validate the microarray results, total RNA from the same samples from above (Con-miR and miR-1248 overexpressing) were reverse transcribed using random hexamers and SSII reverse transcriptase (Invitrogen) followed by RT-qPCR using SYBR green PCR master mix (Life Technologies). Sequences used for gene specific primers are in the Supplemental Methods. RT-qPCR was performed on an Applied Biosystems 7500 Real-Time PCR System.

Immunoprecipitation of mRNAs associated with biotinylated-miR-1248

HeLa cells were transfected with 20 nM biotinylated miR-1248 (miR-1248) or a negative control, biotinylated C. elegans miR-67 (Con-miR) using Lipofectamine 2000. Twenty-four hours post transfection, cells were lysed in buffer containing 20 mM Tris-HCl (pH 7.5), 100 mM KCl, 5 mM MnCl₂, 0.3% NP-40, 50 U RNase Out (Invitrogen) and a protease inhibitor cocktail (Roche Applied Science). Samples were then centrifuged for 10 minutes at 4°C at 10,000 × g. After a 2 hour pre-incubation of Streptavidin Dynabeads in lysis buffer containing BSA (1 mg/mL) and yeast tRNA (1 mg/mL), cytoplasmic lystates were applied to beads and incubated at 4°C for 4 hours with rotation. Samples were washed five times with lysis buffer after which, RNA bound to beads was isolated using Trizol LS according to manufacturer's instructions with the following modifications: after RNA precipitation, RNA was DNase treated for 30 min at 37°C and then extracted with Acid:Phenol CHCl₃ (Ambion) before a second precipitation with 100% Ethanol. RNA pellets were re-suspended in 15 μL water. mRNAs associating with biot-miR-1248 were analyzed using gene specific primers (see Supplemental Methods) followed by RT-qPCR.