Multiparametric profiling of plasma cell-free nucleic acids in nonagenarians and centenarians highlights age-related differences associated with longevity

Abstract

Plasma circulating cell-free nucleic acids (ccfNAs) have emerged as promising non-invasive biomarkers of aging. While age-associated changes have been reported, data in relation to extreme aging and longevity remain scarce. Here, we assessed ccfNA levels and integrity, and ccfDNA methylation in a cohort of 86 individuals, analyzed both overall and stratified by sex, including nonagenarians (NON: 90–98 years, n = 29), centenarians (CEN: 100–109 years, n = 28), and a middle-aged control group (CG: 38–67 years, n = 29) of nonagenarians’ and centenarians’ offspring, using our previously optimized multiparametric analysis workflow targeting nuclear (ccfnDNA) and mitochondrial (ccfmtDNA) DNA, ribosomal RNA (ccfrRNA), messenger RNA (ccfmRNA), and microRNAs (ccfmiRNAs). ccfnDNA levels followed non-linear trajectories, decreasing from CG to nonagenarians (28%–64.5%, significant in nonagenarian men compared to CG), then slightly increasing in centenarians. ccfmRNA and ccfmtDNA followed the opposite pattern, with significantly lower ccfmRNA levels in centenarians (44.5%), and both were strongly correlated (r = 0.59–0.83, p < 0.0001), suggesting a shared regulatory mechanism. Additionally, ccfnDNA integrity significantly decreased from CG to CEN (20.7%), while ccfmtDNA and ccfrRNA integrities, and ccfDNA methylation, did not vary. Among the seven ccfmiRNAs analyzed, miR-93-5p, miR-126-3p, and liver-specific miR-122-5p, exhibited significant age-related decreases (40.5%–70.3%), reaching their lowest levels in centenarians. Our study thereby provides novel findings regarding age- and sex- related changes in ccfNAs, highlighting both dynamic and stable characteristics linked to longevity. It identified potential ccfNA-based longevity biomarkers, and supports ccfmiRNAs as the most promising and sensitive biomarkers of longevity.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s11357-025-01981-3.

Introduction

Plasma circulating cell-free nucleic acids (ccfNAs) have garnered broad biomedical interest in recent decades, particularly for their great potential as non-invasive biomarkers for the detection of various diseases, notably cancer. The increased levels of ccfDNA containing genetic information (i.e. mutations), and their fragmentation and methylation profiles, as well as variations in ccfmiRNAs levels are particularly useful for cancer diagnostic and patient management [1–5]. ccfDNA is also widely used for prenatal testing, by analyzing fetal ccfDNA in maternal plasma [5, 6]. ccfNAs comprise diverse types and forms of DNA and RNA. ccfDNA derives from both nuclear and mitochondrial DNA, can be either single or double-stranded, and can be highly fragmented or of high molecular weight. ccfRNA includes coding circulating cell-free messenger RNA (ccfmRNA) and non-coding RNA, such as long non-coding RNA, ribosomal RNA (ccfrRNA), circular RNA, and ccfmiRNA [7]. These molecules circulate in the bloodstream completely free, associated with proteins (histones, AGO1/AGO2, …) or macromolecular lipoprotein complexes (i.e. virtosomes), or packaged within extracellular vesicles (EVs) such as exosomes, microvesicles, or apoptotic bodies [7–9]. Their release is mediated by a range of mechanisms, including mainly regulated (apoptosis, pyroptosis and NETosis), and accidental (necrosis) cell death, but also active secretion [8, 10, 11].

Recently, there has been a growing interest in exploring ccfNAs in healthy individuals, particularly in relation to aging, healthy aging, and/or longevity [7]. As the worldwide population ages at an unprecedented rate, with more than 2.1 billion people who will be older than 60 years in 2050 (WHO) [12], the study of molecular biomarkers of aging and longevity is becoming increasingly important. Such biomarkers hold promise for predicting and/or monitoring the aging trajectory of an individual by distinguishing healthy from unhealthy aging or pathological processes, assessing biological age, and identifying subjects at risk of age-related diseases. In this context, ccfNAs, particularly ccfDNA and ccfmiRNAs, have emerged as potential biomarkers of aging [13, 14], healthy aging [15, 16] and/or longevity [17, 18], and have been investigated in cross-sectional studies across age groups. These studies sometimes specifically included long-lived individuals (LLI) and exceptionally LLI (ELLI), i.e. nonagenarians (NON) and centenarians (CEN), respectively, who are considered as models for healthy aging and longevity studies [7, 19]. It has been suggested that ccfDNA could reflect the chronic and low-grade inflammation observed in aged individuals, which is a phenomenon known as inflammaging and considered as a hallmark of aging [20]. Higher plasma ccfDNA levels (total, nuclear and mitochondrial ccfDNA) were observed in nonagenarians, and have been linked at older ages to this increased inflammation, ccfDNA levels being positively correlated to pro-inflammatory cytokines IL-1, IL-6, TNF-α, and to CRP and RANTES [21–23]. More than simply acting as a source of inflammatory stimuli, they may also contribute to the maintenance of inflammaging in the elderly [21]. Conversely, physical fitness and good health, especially in LLIs, were correlated to lower ccfDNA levels, with high levels linked to a health decline, such as increased frailty or cognitive decline [15]. However, due to the discordant results from different studies, the link between ccfmtDNA levels in aging or healthy aging should be interpreted with caution [13, 24]. ccfDNA methylation has also been investigated during aging, revealing a global hypomethylation and numerous age-related CpGs [25, 26]. Finally, several ccfmiRNAs were also associated to aging and longevity, showing opposite trajectories depending on the specific miRNA studied: some increased with age while others decreased. For example, several miRNAs (e.g. miR-126-3p [18], miR-21 [16, 18], miR-19a/b-3p [27]) showed an increase with age, up to 90 years old, followed by a decrease in centenarians, displaying their potential as longevity biomarkers [7]. Other miRNAs were also identified as modulators promoting inflammation, such as miR-21 [16]. Other ccfRNA species, such as messenger and ribosomal RNA, were mostly ignored in these studies.

To address this limitation, we recently developed an optimized workflow that enables the simultaneous multiparametric analysis of both ccfDNA and ccfRNA, extracted from a single plasma sample [28]. Using this workflow, we characterized age-associated molecular changes in a cohort of 139 healthy adults aged 19 to 66 years, identifying variations in the quantity and integrity of ccfNAs, and their potential as biomarkers of aging. Our results confirmed and extended previous observations by demonstrating a progressive age-related increase in total and nuclear ccfDNA, but not mtDNA, that could be linked to the increased inflammation occurring during aging. Additionally, the integrity of nuclear ccfDNA declined with age, suggesting increased fragmentation that may reflect a greater contribution of apoptosis to cell death in the elderly, while ccfmtDNA integrity did not vary. Our study was also the first to show an increase in ccfrRNA and ccfmRNA levels with age, and to highlight strong correlations and shared patterns between ccfmtDNA, ccfmRNA and ccfrRNA, suggesting common release mechanisms for these ccfNA types. Forty-seven miRNAs have also been identified to be differentially expressed during aging from additional miRNA-sequencing experiments, some of which have previously been reported as age-associated (e.g. miR-23b, miR-30c-5p, miR-483-5p, miR-30b-5p) [14, 29–31].

In the present study, we have extended this work by applying the same optimized ccfNA analysis workflow (Fig. 1) to a unique cohort of LLI and ELLI, aiming to explore ccfNA variations occurring at extreme ages and in relation to human longevity. Characterizing their patterns beyond 90 years is essential to determine whether previously reported age-related molecular features persist, reverse, or stabilize in long-lived individuals. We analyzed plasma samples from 104 French individuals from the CEPH aging cohort, comprising 70 LLI aged 90–109 years, and 34 unrelated nonagenarians’ and centenarians’ offspring, aged 38–67 years, who served as a middle-aged control group (CG). After ccfDNA quality control, 18 samples were excluded (see Supplementary Methods), leaving 57 LLI and 29 middle-aged controls (Table 1). Studying ccfNAs in this context provides a valuable opportunity to deepen our understanding of molecular changes associated with longevity, yielding new hypotheses. They also enable the identification of potential ccfNA-based biomarkers that may predict healthy aging and longevity trajectories.

Fig. 1.

A detailed overview of the experimental design used to perform multiparametric profiling of circulating cell-free nucleic acids in plasma. a, Experimental workflow for ccfNA analysis. b, c, Quantification methods used including (b) fluorometric dsDNA quantification by Qubit and (c) HS capillary electrophoresis with the Femto Pulse instrument. An example of a ccfDNA profile obtained with a Femto Pulse is shown. D, List of all PCR assays used for ccfDNA and ccfRNA analysis. rt: room temperature, RT: reverse transcription

Table 1.

Description of the study cohort after sample exclusion based on Femto Pulse profiling

Cohort	CG (n = 29)	Nonagenarians (n = 29)	Centenarians (n = 28)
Age, Mean ± SD (range)	56.0 ± 7.2 (38–67)	93.3 ± 2.3 (90–98)	102.9 ± 2.8 (100–109)
Females, n (%)	14 (48.3%)	17 (58.6%)	16 (57.1%)

Materials and methods

Ethics statement

Participants of the CEPH Aging cohort were recruited between 1991 and 2002, and the protocol was conducted in accordance with the ethical and legal frameworks in effect at the time of the study [32]. The protocol was approved by the Commission Nationale de l’Informatique et des Libertés (CNIL) for data protection, and by the French ethic committee (Comité Consultatif de Protection des Personnes dans la Recherche Biomédicale, CCPPRB Paris-Saint-Antoine, approval No. 00479). All participants filled a written informed consent at the moment of inclusion.

Study population

The study cohort included EDTA plasma samples from 104 French donors from the CEPH Aging cohort [32], including 70 LLI and ELLI, aged 90 to 109 years, and 34 LLIs offspring aged 38 to 67 years. Donors included 56 women (F) and 48 men (M). After ccfNA extraction, 18 samples did not meet quality control criteria based on ccfDNA profiling using the Femto Pulse (see Supplementary Methods), defined as the ratio of the 1st and 2nd peaks to total DNA higher than 0.5, and were therefore excluded from the study, while 86 samples were included in the multiparametric analysis (Table 1, Fig. S1). Study participants were divided into three different age groups for analysis: a middle-aged control group composed of offspring of NON or CEN (38–67 years old, n = 29, 48.3% F), NON (90–98 years old, n = 29, 58.6% F) and centenarians (100–109 years old, n = 28, 57.1% F).

Plasma isolation

Initial plasma isolation process was performed at time of inclusion, as described by How-Kit et al. [32]. Briefly, 10 mL of blood per donors was collected in BD Vacutainer K2 EDTA tubes and was stored at room temperature up to 24 h after sampling. Around 5.5 mL of plasma was isolated by a single centrifugation at 1600 g for 10 min. Plasma was carefully collected to avoid contamination by the buffy coat and was aliquoted into one to seven tubes per sample, with each aliquot ranging in volumes from 200 µL to 5 mL. Plasma samples were stored at − 80 °C for up to 30 years without any prior thawing before ccfNA extraction (Fig. 1a). Plasma samples of the CEPH Aging cohort were provided by the CEPH Biobank (CRB du CEPH).

ccfNA extraction

Plasma samples were thawed at room temperature, and 2 µL of plasma was pipetted on Quantofix EDTA 0-400MG/L strips (Macherey–Nagel) to confirm the anticoagulant used. EDTA plasma samples were then centrifuged at 16,000 g for 10 min at room temperature (Fig. 1A). Between 1.2 and 2.6 mL of plasma were then used to isolate ccfNAs, depending on sample availability, including ccfDNAs and ccfRNAs, using the column-based QIAamp ccfDNA/RNA kit (Qiagen) according to the manufacturer’s instructions. 50 µL of H₂O were used for elution and the eluate was aliquoted to 25 µL in twin.tec® PCR Plates LoBind® (Eppendorf) and stored at − 80 °C until use.

Quality control and quantification of ccfDNA by high-sensitivity capillary electrophoresis

As described in our previous study [28], 2 µL of total ccfDNA was analyzed by high-sensitivity pulse-field capillary electrophoresis using the Ultra Sensitivity NGS kit on a Femto Pulse system (Agilent), following the manufacturer’s instructions. ccfDNA profiles, concentrations of total DNA, and of the 1st and 2nd nucleosome peaks were obtained using the calibrated ladders for quantification. The ratio of the 1st and 2nd peaks to total DNA was calculated and used as a quality threshold. Samples with a ratio below 0.5 were considered as outliers (Fig. S1), as such low values indicate a predominance of high-molecular-weight genomic DNA, which interferes with accurate ccfDNA quantification. As a result, 18 samples that did not meet the quality threshold were excluded, leaving 86 samples in the study: 57 long-lived individuals and 29 younger controls (Table 1).

Plasma hemolysis measurement by UV absorbance

Plasma hemolysis was assessed using a previously described spectrophotometric-based method [33], using a NanoDrop® 2000c Spectrophotometer (Thermo Scientific). We first assessed the method on calibration samples made by spiking a pool of very clear non-hemolyzed plasma samples with red blood cells (RBC) at different proportions: 0.002%, 0.004%, 0.008%, 0.016%, 0.031%, 0.0625%, 0.125%, 0.25% and 2.5%. Spiked RBC was prepared after isolation from plasma using multiple cycles of freezing, thawing and vortex mixing. Absorbance measurements were carried out using 2 µL of plasma. Absorbances at 385 nm (A₃₈₅) and 415 nm (A₄₁₄) were measured to calculate the Hemolysis Score (H.S.) with the published formula: $\text{H}.\text{S}.=A_{414}-A_{385}+0.16\times A_{385}$ [33]. A H.S. threshold of 0.062 was used to consider a sample hemolyzed, calculated by the following equation: ${H.S.}_{non-hemolyzedsamples}+3\times {\text{SD}}_{{\text{H}.\text{S}.}_{\text{non}-\text{hemolyzed samples}}}$ as recommended [33]. Thus, very clear non-hemolyzed plasma samples were used to calculate the H.S. threshold of 0.06 for hemolysis, and the calibration plasma samples spiked with 0.002% to 0.0625% RBC were thereby found non-hemolyzed.

Quantification of total ccfDNA by high-sensitivity fluorometry

Total ccfDNA was measured by fluorometric quantification using the Qubit® dsDNA HS Assay Kit on a Qubit 3 Fluorometer (Life Technologies) according to the manufacturer’s instructions. 2 µL of ccfNA was used for measurement and each sample was analyzed in duplicate. The mean value of the two measurements was used as the total ccfDNA concentration.

Primer and probe design and efficiency determination

All primers and probes are listed in Table S1, including commercial assays, previously published and newly designed primers and probes. New qPCR primers and probes were designed using Beacon Designer 8 (PREMIER Biosoft). New bisulfite sequencing PCR and pyrosequencing primers were designed using MethPrimer (http://www.urogene.org/cgi-bin/methprimer/methprimer.cgi) and the SNP Primer design software (Qiagen), respectively. All oligonucleotides were purchased from Eurogentec using RP-Gold purification for unmodified oligonucleotides and HPLC purification for modified oligonucleotides. All qPCR assay efficiencies were determined from standard curves generated from serial dilutions of a control sample (human leukocyte gDNA (Promega) for ccfDNA qPCR assays, control cDNA prepared from 1:1 PBMC/liver RNA (Ambion) for ccfrRNA, ccfmRNA qPCR and ccfmiRNA assays), and exhibited efficiencies between 1.72 and 2.12 (Table S1).

Reverse transcription

ccfmRNA and ccfrRNA

ccfRNA was reverse transcribed using the Quantitect Reverse Transcription Kit (Qiagen). 6 µL of input ccfNA was mixed with 2 µL of gDNA wipeout buffer and 6 µL of RNase-free water. The mixture was incubated for 2 min at 42 °C to eliminate DNA. Then, a 20-µL reverse transcription (RT) reaction was prepared, containing the initial 14 µL mix, 1 µL of reverse transcriptase, 4 µL of RT buffer 5X and 1 µL of RT primer mix. The RT reaction was incubated at 42 °C for 30 min, followed by 3 min at 95 °C to inactivate the reverse transcriptase. cDNA was stored at − 20 °C before ccfrRNA and ccfmRNA analysis. A control cDNA sample was also prepared from 1 µg of pooled 1:1 PBMC/liver RNA (Ambion) using the same protocol described above.

ccfmiRNA

ccfmiRNAs were reverse transcribed using the miRCURY LNA RT Kit (Qiagen). The RT reaction contained 5 µL ccfNAPs, 2 µL 5X miRCURY SYBR Green RT reaction Buffer, 1 µL 10X miRCURY RT Enzyme, 0.5 µL UniSp6 SpikeIn (10⁸ copies/µL) and 1.5 µL RNase-free water. The mixture was incubated for 60 min at 42 °C followed by 5 min at 95 °C. Samples were stored at − 20 °C. A control cDNA sample was also prepared from 1 µg of pooled 1:1 PBMC/liver RNA (Ambion) using the same protocol described above.

Quantitative PCR

Real-time PCR was used for absolute quantification (AQ) of ccfNAs. All PCR reactions were performed in 384-well plates (Roche) with a final reaction volume of 10 µL on a LightCycler 480 II Real-Time PCR System (Roche). During PCR preparation, samples were always processed in 96-well LoBind plates (Eppendorf). PCR mix and ccfNA samples were distributed in 384-Well PCR plates by a Bravo Automated Liquid Handling Platform (Agilent). ccfDNA, ccfmRNA and ccfrRNA, and ccfmiRNA were analyzed on separate PCR plates.

ccfDNA, ccfmRNA and ccfrRNA assays

ccfDNA (Kpn I, APP and mtDNA) and 18S ccfrRNA assays were based on Syto 9 detection, while ccfmRNA assays (GAPDH, B2M and GSTA1/GSTA2) were based on hydrolysis probe detection. PCR reactions contained 1X Buffer, 1.625 mM MgCl₂, 200 µM dNTPs, 200 nM PCR primers, 1.6 µM Syto 9 or 100 nM hydrolysis probes, 0.05 U/µL HotStarTaq DNA polymerase (Qiagen), and 2 µL of a 20-fold dilution of ccfNA for ccfDNA analysis or 2 µL of a fourfold dilution of cDNA for ccfRNA analysis. Cycling conditions included an initial denaturation for 10 min at 95 °C, followed by 50 cycles of amplification (95 °C for 30 s, 60 °C for 30 s for ccfDNA or 60 s for ccfRNA, and 72 °C for 45 s), and a final melting curve analysis from 65 °C to 95 °C (5 acquisitions per °C) to assess amplicon specificity for PCR assays based on Syto 9 detection. A control human genomic DNA (Promega) and a control cDNA sample was added in PCR plates with ccfDNA and ccfRNA assays, respectively. These samples were used as interplate calibrators for AQ.

AQ in ng/mL of plasma was performed for each ccfnDNA PCR assay using standard curves obtained from serial dilutions of human genomic DNA (Promega) ranging from 2 to 0.0002 ng/µL for Kpn I assays, and 9 to 0.0009 ng/µL for APP assays. Absolute concentrations were obtained using these standard curves and two interplate calibrators included in the PCR reactions, using the AQ module of the LightCycler480 Software (Roche).

AQ in copies/mL of plasma was also performed for each ccfDNA, ccfmRNA and ccfrRNA PCR assay using standard curves generated from serial dilutions of the respective purified PCR amplicon. Briefly, a first PCR reaction was performed for each assay in a final PCR reaction volume of 20 µL, using the same concentrations and cycling conditions as above. After amplification, PCR product profiles were verified by 2% agarose gel electrophoresis and purified using AMPure XP beads (Beckman Coulter) with different bead ratios, from 1 to 3X, to eliminate unspecific PCR products. The purified PCR products were quantified in triplicate using the Qubit® dsDNA HS Assay Kit (Life Technologies) and verified for their length and absence of unspecific amplification products on a Fragment Analyzer system (Agilent) using the DNF-477 HS Small Fragment Kit, according to the manufacturer’s instructions. Finally, for each amplicon, qPCR reactions were performed in duplicate using the purified PCR products serially diluted from 10 to 10, from 25,000,000 to 25 copies/µL. PCR reactions and cycling conditions were the same as above and included two PCR replicates of the interplate calibrators. The absolute concentrations of all amplicons of each ccfNA sample were calculated using these standard curves and the interplate calibrators using the AQ module of the LightCycler480 Software (Roche). Each calculated amplicon copy number was thereby reported per mL of plasma.

ccfmiRNA assays

10 miRCURY LNA miRNA PCR Assays (Qiagen) were used for ccfmiRNA analysis (Table S1). PCR reactions were prepared using the miRCURY LNA SYBR Green PCR Kit (Qiagen). PCR mix contained 3 µL of 30-fold diluted cDNA, 1X miRCURY SYBR Green Master Mix, 1 µL miRCURY LNA miRNA PCR assay and 1 µL RNase-free water in a final volume of 10 µL. Cycling conditions included an initial denaturation for 2 min at 95 °C, followed by 50 cycles of 10 s denaturation at 95 °C, a combined annealing/elongation step at 56 °C for 1 min and a final melting curve analysis from 65 °C to 95 °C (5 acquisitions per °C) to assess amplicon specificity for SYBR detection PCR assays.

AQ of each miRNA in copies/mL of plasma was performed using the Ct value obtained with the UniSp6 spike-in. Due to the known and fixed quantity of this synthetic small RNA spiked in each ccfNA sample before reverse transcription and to miRNA assay efficiencies that were close to 100% (Table S1), it was possible to calculate the quantity of each miRNA in a sample using the following formula: $\text{miRNA quantity}=\text{quantity of UniSp}6\times 2^{{\mathit{Ct}}_{\mathit{miRNA}}-{\mathit{Ct}}_{UniSp6}}$. Relative expression of miR-93-5p, age-associated miR-21-5p, miR-126-3p and miR-483-5p, and liver-specific miR-122-5p could also be calculated by normalizing their concentrations by the geometric mean of the concentrations of housekeeping miR-22-3p and miR-320d.

Plasma hemolysis was also evaluated by qPCR using miR-451a and miR-23a-3p Ct values. miR-23a is not affected by hemolysis, while miR-451a is enriched in erythrocytes [34]. For each sample, we calculated the Ct_{hsa-miR-23a-3p}—Ct_hsa-miR-451a value. A difference higher than 7 indicated hemolysis [34].

Enzymatic digestion of ccfDNA

To assess the linear and circular forms of ccfmtDNA in plasma samples, we treated ccfNAs with Exonuclease V (NEB). The digestion mix contained 1 µL of ccfNA, 5 U of Exonuclease V, 1 mM of ATP and 1X NEBuffer™, in a final volume of 10 µL. The mix was incubated at 37 °C for 30 min, followed by an enzyme inactivation at 70 °C for 30 min. For each sample, a parallel reaction without the Exonuclease V was also performed under the same conditions. 2 µL of each reaction was used for qPCR analysis, using mtDNA 100-bp assay and Kpn I 149-bp assays, following the same conditions as for ccfDNA quantification. Digestion efficiency was measured using the Kpn I assay ${(2}^{-CtExoV-}-2^{-CtExoV+})/{(2}^{-CtExoV-})$. The fraction of circular DNA in a sample was calculated using the formula: ${(2}^{-CtExoV+}/2^{-CtExoV-})$. Absolute concentrations of circular and linear ccfmtDNA were obtained by applying the fraction of each form to the total ccfmtDNA levels.

Integrity analysis

The integrity of ccfDNA and ccfRNA was determined by calculating the integrity indexes from qPCR results, and also from capillary electrophoresis data for ccfDNA. The qPCR integrity index was defined as the ratio of a large amplicon quantity to the smallest. Integrity values close to 1 indicated unfragmented ccfNAs, whereas values close to zero indicated fragmented ccfNAs. For capillary electrophoresis, the 1st and the 1st + 2nd peak concentrations were divided by the total amount of ccfDNA to calculate the integrity indexes. Integrity indexes close to 0 indicated unfragmented ccfNAs, while values close to 1 indicated fragmented ccfNAs.

DNA methylation analysis by pyrosequencing

Bisulfite conversion of ccfDNA was performed using 6 µL of plasma ccfNAs and the EZ-96 DNA Methylation-Gold™ Kit (Zymo Research), according to the manufacturer’s instructions. Converted-ccfDNA was stored in LoBind plates at − 20 °C before use. For optimization experiments, DNA methylation standards with 0%, 25%, 50%, 75% and 100% DNA methylation levels and concentrations of 0.01, 0.1, 1 and 10 ng/µL were prepared using converted methylated and unmethylated human DNA from the EpiTect PCR Control DNA Set (Qiagen). Two Line-1 amplicons were amplified in a 25 µL reaction by standard PCR on a LightCycler 480 Real-Time PCR System (Roche).

PCR reactions contained 2 µL converted ccfDNA (or 1 µL of DNA methylation standard), 1X Buffer, 1.625 mM MgCl₂, 200 µM dNTPs, 200 nM PCR primers, 1.6 µM Syto 9 and 0.08 U/µL Hot Start Taq DNA polymerase (Qiagen). Cycling conditions included an initial 10-min denaturation at 95 °C, followed by 45 cycles of amplification (30 s denaturation at 95 °C, 30 s primer annealing at 60 °C, and 30 s elongation at 72 °C), and a final elongation step at 72 °C for 5 min. PCR products were stored at − 20 °C before sequencing. 10 µL of PCR products was purified using a previously described protocol [35]. Pyrosequencing was performed using the PyroMark Gold SQA Q96 Kit (Qiagen) on a PyroMark Q96 MD (Qiagen) using pyrosequencing sequences (Table S1), and the DNA methylation data were analyzed with PyroMark CpG software (Qiagen). We excluded one, seven, and thirteen DNA methylation outliers for CpGs 1, 2 and 4, respectively, as their DNA methylation percentages were close to or equal to 100%, which is much higher than the 100% DNA methylation standard.

Statistical analysis

All analyses were performed using GraphPad Prism version 10.2.3 for Windows (GraphPad Software) and R (GPL) for PCA analysis. Correlations analyses were performed using Pearson test with the associated r correlation coefficient and p-values. We performed both global correlation analyses on the entire cohort and stratified correlation analyses grouping either the control group (CG) and nonagenarians (NON), or NON and centenarians (CEN). Significant differences between groups were assessed using the Mann–Whitney U test, either by comparing age groups of women, men, and all individuals pairwise, or by considering both sexes within the same age range. p-values are two-sided, and p-value < 0.05 is considered significant; * < 0.05, ** < 0.01, *** < 0.001, **** < 0.0001. PCA analysis was performed using the FactoMineR_2.3 package, using reduced centered data from our analysis. Missing values were replaced by the median of the variable in the data matrix.

Results

Levels of ccfnDNA and ccfmtDNA do not significantly vary in nonagenarians and centenarians, but present sex-related differences

We first compared the absolute quantification of ccfDNA using three methods: fluorometric quantification using Qubit, HS capillary electrophoresis using Femto Pulse, and qPCR Kpn I 149-bp assay. ccfDNA concentrations measured using Qubit, Femto Pulse, and qPCR progressively decreased across methods, with mean values of 11.81 ± 10.94, 7.94 ± 13.58 and 3.01 ± 3.25 ng/mL of plasma, respectively, with similar values across age group for each method (Fig. S2). This was expected as the Femto Pulse instrument and qPCR assay do not quantify DNA higher than the upper marker (6000 bp), and DNA lower than 149 bp, respectively. Fluorometric quantification and HS capillary electrophoresis did not reveal significant variations in total ccfDNA levels, either in correlation analyses or in age group comparisons (Fig. S3).

Absolute quantification of ccfnDNA was also performed more finely using qPCR assays targeting Kpn I repeats and APP single copy gene for ccfnDNA, amplifying DNA sequences ranging from 67 to 421 bp, and assays targeting ccfmtDNA, amplifying 100 and 417-bp fragments. Quantifications showed concordant results, with moderate to strong correlations between the different APP and Kpn I PCR assays used (0.57 < r < 0.95, 2.37 × 10⁸ < p < 3.7 × 10³⁹) (Fig. S4), with an over-quantification bias with smaller qPCR amplicons, as expected. Correlation analysis of the entire cohort showed no variation in ccfnDNA copies per mL of plasma levels with age, regardless of the qPCR assay used (Figs. 2a, S5a and b). Using the 149-bp Kpn I assay, values were highly variable, ranging from 9.54 × 10⁵ to 1.15 × 10⁸ (RSD = 111.2%), with mean values of 1.28 × 10⁷, 1.08 × 10⁷ and 1.65 × 10⁷ copies per mL of plasma, in CG, NON and CEN, respectively (Fig. 2a). When assessing the variations of ccfnDNA levels in the CG–NON and NON–CEN groups, we observed opposite trends between both stratified groups, as well as between men and women. Although not significant, all qPCR assays (Kpn I and APP) showed a slight decrease in ccfnDNA levels between CG and NON men (− 0.30 < r < − 0.12), followed by an increase from NON to CEN men (0.061 < r < 0.28), while the opposite variations were observed in women (Figs. 2a, S5a and b). This pattern was also observed with Qubit and Femto Pulse quantifications (Fig. S3). Age group comparisons showed significantly lower ccfnDNA levels in NON men compared to CG men when using the 149, 168 and 291-bp Kpn I assays, with p-values of 0.028, 0.028 and 0.025, respectively, while no significant differences were observed between CG and CEN (Figs. 2a and S5a). Additionally, the 291 and 421 bp assays revealed significant sex differences in CG (p = 0.030 and p = 0.027, respectively), with men exhibiting higher copy numbers than women. Using both APP qPCR assays, no significant age or sex-related changes were observed (Figs. 2a and S5b).

Fig. 2.

Plasma ccfDNA quantity and integrity variations in CG, NON and CEN (n = 86). a, ccfnDNA and ccfmtDNA quantification. b, ccfnDNA and ccfmtDNA integrity assessment using the integrity index. c, Global plasma ccfDNA methylation variation using the mean DNA methylation value of Line1 (CpG_{1,3–7,9–10}). Mann–Whitney U tests were performed in box-plots between each age group and between men and women of the same age group. The ‘ + ’ symbol indicates the mean value obtained for each group. Pearson r coefficients and associated p-values as well as linear regression lines are shown in scatterplots. p-values < 0.05 are considered significant; * < 0.05

Regarding ccfmtDNA, quantification of copy number with the two qPCR assays (100 bp and 417 bp) did not show any significant correlation with age or any differences between age groups, either using the absolute quantification, or ccfmtDNA copy number per genome equivalent (Figs. 2a and S5c). Using the 100-bp assay, ccfmtDNA copy number ranged from 1.37 × 10⁴ to 4.54 × 10⁶ (RSD = 134.9%), with mean values of 5.19 × 10⁵, 8.37 × 10⁵ and 6.70 × 10⁵ in CG, NON and CEN, respectively. Stratified correlation analyses showed in men a slight but not significant increase with age in CG and NON (r = 0.28 and r = 0.30, respectively), and a non-significant decrease with age in NON and CEN (r = − 0.22 and r = − 0.22, respectively) (Figs. 2a and S5c). We also assessed levels of the linear and circular forms of ccfmtDNA after Exonuclease V treatment. The proportion of circular ccfmtDNA did not significantly change with age, but greatly varied between individuals, ranging from 18.5% to 100%; with a median circular/linear ratio of 0.92 (Figs. S6a, S6b and S7a). Global correlation analyses displayed no significant correlation of linear and circular ccfmtDNA with age, both considering ccfmtDNA levels or copy number per genome equivalent (Fig. S6c and d). Moreover, ccmtDNA levels exhibited a slight (r ≥ 0.11) but non-significant positive correlation with the circular ccfmtDNA percentages (Fig. S6e). In men, both linear and circular mtDNA expressed in copies per mL of plasma or per haploid genome followed the same trends observed for total ccfmtDNA, with an increase between CG and NON (0.24 < r < 0.36), and a decrease between NON and CEN (− 0.38 < r < − 0.12) (Fig. S7b).

ccfDNA integrity declines in both nonagenarians and centenarians

We assessed the ccfDNA integrity by calculating an integrity index based on the ratio of a large amplicon to the smallest from each set of qPCR assays. Kpn I 291/149 and 421/149 integrity indexes showed a decreasing trend with age (r = − 0.20 and r = − 0.23, respectively), the latter being statistically significant (p = 0.034) (Fig. 2b). In age-group analysis (Figs. 2b and S8a), only 291/149 integrity index displayed significant differences, with lower integrity index in NON and CEN men compared to CG men (p = 0.032 and p = 0.037, respectively). Kpn I 291/149 and 421/149 integrity indexes were also significantly lower in CG women compared to men of the same group (p = 0.029 and p = 0.020, respectively). On the contrary, APP 180/67 integrity index displayed no significant changes, neither in correlation analyses nor in age group comparisons (Fig. S8b). Using Femto Pulse quantification, we did not observe any significant age-related differences in ccfDNA integrity. Consistent with the qPCR results, women in CG showed a significantly lower integrity compared to men in the same age-group, as measured by the (1st + 2nd peaks)/Total integrity index (p = 0.0027) (Fig. S8c). Moreover, we did not observe any significant variations of ccfmtDNA integrity using the 417/100 ratio (Fig. 2b). Stratified correlation analyses presented trends similar to those observed in global correlations using the Kpn I qPCR assays, decreasing with age in almost every subgroup (Fig. S8a). Conversely, the APP 180/67 integrity index exhibited a weak positive correlation with age in CG–NON men (r = 0.11), followed by a decrease from NON to CEN men (r = − 0.23), although neither correlation reached statistical significance (Fig. S8b).

Global ccfDNA methylation do not vary in nonagenarians and centenarians

Global ccfDNA methylation was calculated as the average of methylation level of the eight CpG sites targeted in Line1 repeat obtained by bisulfite pyrosequencing. Our results did not show any variation in global ccfDNA methylation within our cohort, with no significant correlation with age nor differences between age groups (Fig. 2c). However, when analyzing each CpG individually (Fig. S9), CpG 7 exhibited significant age differences, with NON men showing lower methylation levels compared to both CG men (p = 0.021) and CEN men (p = 0.030). Additionally, we also observed a significant decrease of Line1 CpG 1 methylation in NON women compared to NON men (p = 0.043).

Levels of ccfmRNA increase in nonagenarians and decrease in centenarians

Plasma ccfRNA quantity and integrity were evaluated using RT-qPCR assays targeting 18S rRNA and four mRNAs (Fig. 1b). The ubiquitously expressed housekeeping genes GAPDH and B2M were used as proxies for total ccfmRNA load, while ccfmRNA levels of GSTA1/GSTA2, genes specifically expressed in the liver, kidney, and adrenal glands [36, 37], were analyzed to quantify ccfmRNA originating from minor contributing organs.

ccfRNA levels showed large variations between individuals, ranging from 1.23 × 10⁶ to 3.92 × 10⁷ (RSD = 81.5%), 3.63 × 10³ to 2.02 × 10⁵ (RSD = 85.8%), and 6.19 × 10³ to 2.19 × 10⁶ (RSD = 160.8%) copies per mL of plasma using the 18S 59-bp, GAPDH, and B2M assays, respectively (Fig. 3a). Using the four 18S rRNA assays amplifying fragments between 59 and 495 bp, our results revealed a non-significant decrease with age, and no significant differences in ccfrRNA levels between age or sex groups (Figs. 3a and S10). We did not observe any significant correlation of ccfmRNA levels with age using GAPDH and B2M assays (Fig. S10). However, investigating each age group, both assays displayed significantly lower ccfmRNA levels in CEN men compared to CG men (p = 0.028 with GAPDH and p = 0.019 with B2M) (Fig. 3a). Moreover, B2M ccfmRNA levels were also significantly decreased in all CEN compared to both CG (p = 0.0047) and NON (p = 0.031), as well as in CEN men compared to NON men (p = 0.021). In men, B2M ccfRNA showed a maximum mean value in NON, with 3.74 × 10⁵ copies per mL of plasma, compared to 1.86 × 10⁵ and 1.22 × 10⁵ for CG and CEN (Fig. 3a). Stratified correlation analyses revealed a clear but not significant trend that changed direction across age groups for B2M ccfmRNA (Fig. 3a). In men, an increase was observed between CG and NON (r = 0.28), followed by a decrease between NON and CEN (r = −0.36). This pattern is similar to that observed for ccfmtDNA quantification, but in opposition to the trend seen in ccfnDNA quantification (Figs. 2 and S5). Additionally, the GSTA1/GSTA2 assay detected ccfmRNA from minor contributing organs in 24.4% of plasma samples, and we observed a slight but not significant decrease in NON and CEN compared to CG, whether relative expression level or copies/mL is considered (Fig. 3b).

Fig. 3.

Analysis of plasma ccfRNA variation in quantity and integrity (n = 86). a, Variation of plasma levels of 18S ccfrRNA, B2M and GAPDH ccfmRNAs, and b, GSTA1/GSTA2 ccfmRNAs. c, Assessment of ccfrRNA integrity. Mann–Whitney U tests were performed in box-plots between each age group and between men and women of the same age group. The ‘ + ’ symbol indicates the mean value obtained for each group. Pearson r coefficients and associated p-values, as well as linear regression lines, are indicated in scatterplots. p-values < 0.05 are considered significant; * < 0.05, ** < 0.01

In contrast to ccfDNA, integrity analysis of ccfRNA using 18S rRNA-based 115/59, 270/59, and 495/59 indexes did not show any significant variation with age or differences between age groups (Figs. 3c and S11). Stratified correlation analyses also revealed no significant variation with age, although a weak decrease in integrity between CG and NON (− 0.20 < r < − 0.10), followed by a weak increase from NON to CEN (0.07 < r < 0.30) was observed when considering all individuals or men only (Fig. S11).

miR-93-5p, miR-126-3p and miR-122-5p levels decrease in plasma samples of centenarians

We evaluated the expression of nine ccfmiRNAs by RT-qPCR, including housekeeping miRNAs (miR-93-5p, miR-22-3p and miR-320d), age-associated miRNAs (miR-126-3p, miR-21-5p, miR-483-5p), the liver-specific miR-122-5p, and the hemolysis markers miR-23a-3p and miR-451a, also analyzed in our previous study [28]. No sample was hemolyzed and all samples were therefore included in the analysis.

Absolute quantification analysis showed stable levels of housekeeping miRNAs miR-22-3p and miR-320d across age groups, as expected [38]. However, miR-93-5p displayed a significant decrease with age (r = − 0.25, p = 0.018), with lower expression observed in NON and CEN compared to CG (0.0009 < p < 0.047) (Fig. S12a). As a result, only miR-22-3p and miR-320d were selected as normalizers to calculate the relative expression of other miRNAs, including miR-93-5p (Fig. 4). miR-93-5p relative expression decreased significantly in all samples (r = − 0.43, p = 4.35 × 10⁻⁵), and similarly in men (r = − 0.43, p = 0.0058) and women (r = − 0.42, p = 0.0034. In age group comparisons, all CEN showed lower miR-93-5p expression than CG (p = 1.57 × 10⁻⁵), and NON (p = 0.0058).

Fig. 4.

Variation of mir-93-5p, age-associated (miR-126-3p, miR-21-5p, miR-483-5p) and liver-specific miR-122-5p relative expression levels in CG, NON and CEN (n = 86). Mann–Whitney U tests were performed in box-plots between each age group and between men and women of the same age group. The ‘ + ’ symbol indicates the mean value obtained for each group. Pearson r coefficients and associated p-values, as well as linear regression lines, are shown in scatterplots. p-values < 0.05 are considered significant; * < 0.05, ** < 0.01, *** < 0.001, **** < 0.0001

Regarding age-associated miRNAs, miR-126-3p relative expression showed a nearly significant decrease with age (r = − 0.21, p = 0.051), which was strongly significant in women (r = − 0.43, p = 0.0023) with CEN women showing lower miR-126-3p expression compared to CG women (p = 0.041) (Fig. 4). Similar observations were also found with absolute concentrations, sometimes reaching or even improving statistical significance (Fig. S12b). In contrast, miR-483-5p and miR-21-5p relative expression showed no significant correlation with age or differences between age groups were identified (Fig. 4). However, for miR-483-5p, a few significant differences were detected when considering absolute quantification (Fig. S12b).

Liver-specific miR-122-5p was detected in all plasma samples, and relative expression analysis revealed a significant decrease with age, considering all samples or only men (r = − 0.27, p = 0.012, and r = − 0.33, p = 0.042, respectively) (Fig. 4). Significantly lower levels were detected in both CEN and NON compared to CG in the full cohort (p = 8.8 × 10⁻⁶, p = 1.2 × 10⁻⁵), in men (p = 0.0009, p = 0.0031), and in women (p = 0.0034, p = 0.0012) (Fig. 4). These findings were also observed using absolute quantification (Fig. S12c).

Finally, stratified correlation analyses based on relative expression displayed opposite trends between the two subgroups only for miR-126-3p and miR-483-5p, but no correlations were significant (Fig. S13). In men, miR-126-3p increased between CG and NON (r = 0.18), and decreased between NON and CEN (r = − 0.26). Considering all individuals or each sex separately, miR-483-5p decreased in the first subgroup (− 0.25 < r < − 0.19), then increased in the second one (0.12 < r < 0.25).

ccfmtDNA and ccfRNA are strongly correlated

We investigated potential correlations between ccfDNA and ccfRNA levels across our study cohort. Using the Kpn I repeat 149-bp assay, ccfnDNA concentrations showed a significant but weak correlation with ccfmRNA levels measured with the GAPDH assay (r = 0.22, p = 0.044), and a moderate correlation with ccfrRNA levels assessed by the 18S rRNA 59-bp assay (r = 0.43, p = 3.1 × 10⁻⁵). No significant correlation was found between ccfnDNA and the B2M ccfmRNA assay. Additionally, ccfnDNA correlated weakly but significantly with several miRNAs, i.e. miR-93-5p, miR-22-3p, miR-320d, and miR-21-5p (0.24 < r < 0.32, 0.0023 < p < 0.025) (Fig. S14). In contrast, the strongest correlations were observed between ccfmtDNA and ccfRNA. Using the mtDNA 100-bp assay, ccfmtDNA concentrations showed a strong and highly significant correlation with B2M ccfmRNA levels (r = 0.83, p = 4.7 × 10⁻²³), and moderate correlations with GAPDH ccfmRNA (r = 0.59, p = 3.3 × 10⁻⁹) and 18S ccfrRNA measured using the 59-bp assay (r = 0.54, p = 7.1 × 10⁻⁸). Finally, no significant correlations were observed between ccfmtDNA and targeted ccfmiRNAs (Fig. S14).

We finally performed principal component analysis (PCA) and heatmap clustering analysis using the majority of the parameters measured in our study (Fig. S15). Neither PCA nor heatmap analysis revealed any clear separation of samples based on age or sex (Fig. S15a and b). In the variable correlation plots from PCA, we were able to identify two main clusters: one located in the lower right quadrant, composed primarily of qPCR assays for ccfnDNA quantification, and a second cluster in the upper right quadrant, containing qPCR-based quantifications of ccfmtDNA, ccfmRNA, and ccfrRNA (Fig. S15c). The same two clusters were found in the heatmap clustering analysis (Fig. S15a).

Discussion

In this manuscript, we investigated multiparametric changes in ccfNAs associated with extreme aging, and longevity using a cohort of nonagenarians (90 to 98 years) and centenarians (100 to 109 years), along with a middle-aged control group (35 to 67 years). Our study is based on the optimized multiparametric ccfNA analysis workflow recently published, conducted on a cohort of 139 healthy donors aged 19 to 66 years [28]. This workflow enables the comprehensive analysis of ccfNA types, including nuclear and mitochondrial DNA, as well as messenger, ribosomal, and micro- RNA, focusing on quantity, integrity and DNA methylation. Thus, we were able to uncover age-related differences associated with longevity. To our knowledge, this is the first study to simultaneously investigate both ccfDNA and ccfRNA in nonagenarians and centenarians, offering new hypotheses regarding the molecular features of ccfNAs in long-lived individuals.

We quantified ccfnDNA using four qPCR assays targeting Kpn I repeats, and revealed a significant decrease in concentration in NON men compared to CG men (Figs. 2a and S5a). These findings contrast with those of previous studies, reporting a significant increase in NON compared to CG [17, 23]. However, these studies included only [17] or predominantly [23] women, which could explain the discrepancy. Interestingly, we observed opposite trends of ccfnDNA in stratified correlation analyses. While ccfnDNA levels decreased with age in CG and NON, it increased with age when considering NON and CEN using all Kpn I and APP qPCR assays. Although an increase of ccfnDNA was observed with age between 19 and 66 years in our previous work [28], we report here the lowest levels of ccfnDNA in NON, and no significant differences between CG and CEN, both groups displaying close mean values. While ccfnDNA levels are associated with poor health [39, 40], notably frailty [17], and inflammation [22, 23], the lack of difference between CEN and CG could potentially be considered as a biomarker of healthy aging. CEN are known to have a better health status than most elderly individuals under 100 years old, with reduced burden of inflammaging and lower levels of inflammatory markers [41, 42], less morbidities such as cardiovascular diseases, and less chronic conditions [41, 43, 44]. Although not previously reported for ccfDNA levels, the non-monotonic trends observed have been described for other ccfNAs during aging, particularly for ccfmiRNAs. Thus, miR-200c and miR-200b were shown to decrease from 20-year-old individuals to octogenarians, followed by an increase reaching peak levels in CEN [16]. Moreover, quantification results in CG are consistent with previous findings, highlighting higher ccfnDNA concentrations in men than in women in younger (18–46 years) and middle-aged (47–70 years) populations [13, 45]. This could explain why the significant decrease observed in NON compared to CG was only displayed by men, as no significant differences between NON men and women were found in our study. Other quantification methods showed similar non-monotonic trends but no significant age-related differences in total ccfDNA amount. Moreover, an increased dispersion of ccfDNA levels was visible in NON and CEN compared to CG. It suggests an increased variability of ccfDNA amounts in long-lived individuals. (E)LLIs can exhibit highly variable health status, with frailty prevalence reaching up to 60% [46], while the remainder are considered robust. Health status was shown to influence ccfDNA levels and may explain the highest variance observed in NON and centenarians [7, 47].

Integrity analyses revealed a significantly lower ccfnDNA integrity in NON and CEN compared to CG, as measured by the Kpn I 291/149 and 421/149 integrity indexes (Fig. 2b). In addition, NON and CEN men showed a lower integrity index with the 291/149 ratio. These results are in line with those of our previous study on individuals aged 19–66 [28], which showed a decrease in integrity during aging. The increased fragmentation could be explained by an increased contribution of apoptosis with advancing age, resulting in smaller 166 bp fragments [48, 49]. Methylation analysis by pyrosequencing did not reveal any changes in global ccfDNA methylation in NON or CEN (Fig. 2c). This observation is consistent with our previous findings targeting the same CpGs, which reported no significant variation in ccfDNA methylation in individuals aged 19 to 66 years [28]. In our study, only one CpG showed a significant hypomethylation in NON. However, our approach was limited by the narrow scope of the analysis, which targeted only eight CpGs. Future studies should employ broader methods, such as methylome sequencing, to identify ccfDNA methylation variations in NON and CEN [26].

Unlike ccfnDNA, but consistent with our previous study [28], ccfmtDNA levels remained stable across all age and sex groups, whether quantified per milliliter of plasma or per genome equivalent (Figs. 2a, S5c). However, with total ccfmtDNA and with both circular and linear ccfmtDNA, a non-significant increase was observed between CG and NON, followed by a decrease in CEN reaching the lowest values (Figs. S5c and S7). Previous studies investigating age-related variations in ccfmtDNA have reported conflicting results. Our findings aligns with another study [23], which did not find significant differences in ccfmtDNA levels between NON and CG. While we observed no significant increase with age, our results showed the highest mean ccfmtDNA levels in NON, in line with a previous study quantifying ccfmtDNA concentrations in individuals aged 1 to 104 years [21]. Moreover, ccfmtDNA integrity appeared to be preserved in NON and CEN (Fig. 2b), consistent with our previous findings in individuals aged 19 to 66 years [28]. This may be due to protective mechanisms that maintain ccfmtDNA integrity, as ccfmtDNA can be present in blood encapsulated in extracellular vesicles or inside circulating mitochondria [5, 28, 50, 51], which could preserve ccfmtDNA.

Regarding ccfRNA, we found that 18S ccfrRNA levels and integrity remained relatively stable across all age and sex groups (Figs. S10 and S11), suggesting protective mechanisms against degradation in the circulation [28, 52]. However, all ccfrRNA assays showed a clear but non-significant decrease with age (Fig. 3a). Moreover, significant age-related decreases were observed in ccfmRNA levels using GAPDH and B2M qPCR assays (Fig. 3a). Notably, B2M-targeted analysis revealed significantly lower levels of ccfmRNA in CEN, particularly in men, compared to NON and CG, and NON showed the highest levels, in complete contrast with what we observed in ccfnDNA quantification. In stratified correlations, ccfmRNA increased from CG to NON, then declined from NON to CEN, aligning with ccfmtDNA but in opposition to ccfnDNA. As aging is associated with a decline in transcriptomic activity [28, 53–55], the observed decrease in ccfmRNA, and to a lesser extent in ccfrRNA, may reflect the reduced transcriptomic activity, a potential characteristic of longevity. Compared to our previous work, showing an age-related increase between 19 and 66 years old in ccfRNA levels [28], the decrease observed here may be explained by a decline of transcriptomic activity at extreme old age. Thus, low levels of ccfRNA in CEN display a potential as a biomarker of longevity. Correlation analyses revealed moderate to strong positive correlations between ccfRNA and ccfmtDNA, while correlations between ccfRNA and ccfnDNA were weak, or absent when using ccfmRNA B2M assay (Fig. S14). PCA and heatmap hierarchical clustering further distinguished the different ccfNA species, grouping ccfmtDNA and ccfRNA in one cluster, and ccfnDNA in another one (Fig. S15). These similar results and patterns observed in this cohort therefore confirm our previous findings, reinforcing the hypothesis of a shared release, maintenance and/or clearance mechanism for ccfmRNA, ccfrRNA and ccfmtDNA, distinct from those of ccfnDNA [28].

Of the candidate age-related miRNAs analyzed by RT-qPCR, miR-126-3p and miR-122-5p exhibited significant age-related changes (Fig. 4). First, miR-126-3p displayed the lowest levels in CEN women, echoing the results of a previous study conducted on a Sicilian cohort, which also showed the lowest levels in CEN [18]. miR-126-3p was described as a modulator of inflammation and immune response that targets the NF-κB pathway [18, 56], and several targets of miR-126-3p were identified with enriched KEGG pathways, including one longevity regulating pathway involving the FOXO3 gene [18]. It appears to be a promising biomarker of healthy aging and longevity. Interestingly, and similarly to our previous study in an independent younger cohort (19–66 years) [28], we did not observe significant age-related changes in miR-21-5p. Although described as age-associated in some studies [16, 18, 57], our results do not support an association of this miRNA with age. Finally, miR-93-5p showed the lowest expression in CEN, and decreased levels both in NON and CEN compared to CG. In serum, miR-93-5p expression was shown to be higher in frail subjects than in robust individuals [58]. Combined with our results, it suggests that low levels of miR-93-5p could be linked to healthy aging and a higher robustness in the elderly. Moreover, miR-93-5p, miR-22-3p, miR-320d and miR-21-5p all correlated significantly and positively with ccfDNA levels (Fig. S14). Although never described before, these correlations suggest that the release of ccfnDNA and several miRNAs into the bloodstream may result from a global phenomenon, potentially involving shared release mechanisms, which differ from those releasing ccfmRNA, ccfrRNA and ccfmtDNA [28]. Leukocytes being one of the major contributors of both ccfDNA and ccfmiRNAs [7, 59, 60], a simultaneous release of both type of ccfNAs by the same source could explain the correlations.

Liver-specific miR-122-5p showed the most significant decrease at extreme ages, in both NON and CEN compared to CG (Fig. 4). Circulating miR-122-5p was identified as a potential biomarker for liver diseases, showing significantly higher levels in metabolic dysfunction-associated steatotic liver disease (MASLD), and a positive correlation with hepatic fat content, liver stiffness and steatosis degree [61, 62]. Inhibition of miR-122 has also been linked to reduced cholesterol synthesis [63], lowering the risk of Alzheimer’s disease [64], and was already proposed as a therapeutic strategy against AD [65]. Thus, lower levels of miR-122-5p in NON and CEN could constitute a promising biomarker of healthy aging and longevity, preventing liver diseases and age-related diseases.

In our study, the most significant age-related differences were observed in ccfmiRNA levels, with lower p-values than those obtained for ccfDNA or ccfmRNA quantification. This strongly suggests that ccfmiRNAs are the most suitable type of ccfNAs to detect age-associated differences and to serve as biomarkers of aging and longevity. This is in agreement with a previous study describing ccfmiRNAs as superior biomarkers compared to ccfDNA for diseases diagnosis [66], stating several advantages of ccfmiRNAs compared to other biomarkers, including their high stability and robustness in various biological fluids, not prone to degradation unlike ccfDNA. Combined with their high sensitivity, it makes them powerful tools for the detection of age-related and pathological conditions [3, 66].

Finally, we would like to highlight some limitations of our study. First, the use of ccfDNA and ccfRNA as biomarkers of longevity could be limited by the high variability we observed among individuals of the same age group, that was also reported in other studies, including our previous work focused on younger donors aged 19–66 years [13, 28]. Our study is also limited by our cohort size, with only 86 individuals after the removal of samples that failed quality control due to over 50% high-molecular-weight DNA, leading to atypical Femto Pulse profiles (Fig. S1). Even though these samples represented 17% of the initial cohort, their exclusion was essential, as high-molecular weight DNA primarily results from cell lysis caused by preanalytical factors, such as the time between blood collection and plasma isolation, which can prevent detecting genuine biological differences [67, 68]. Although some studies used similar or smaller cohorts, the size of the age and sex groups was limited, and could have prevented the detection of significant differences between age groups, or significant correlations with age in global and stratified analyses. Moreover, our study did not formally account for the health status of the participants, which could act as a confounding factor. Another limitation is the use of long-lived individual’s offspring as younger controls (38–67 years), which was needed to compare samples with the same preanalytical processes. However, NCO form a healthier group than age-matched non-NCO [16, 69, 70], and they could also have prevented the identification of some age and sex-related changes. Lastly, we mostly used target-based method to assess ccfNA quantity and integrity. Sequencing the genome, methylome and transcriptome, including miRNome, will be needed to identify additional age-related ccfNA changes. We also did not assess the tissue of origin or extracellular carriers of ccfNA either. It would have been valuable to identify those encapsulated in extracellular vesicles, given their role in intercellular communication and their potential to reflect the physiological or pathological conditions of the originating cells, tissues, or organs [28, 71, 72]. A final limitation of our study is the absence of correlation analyses between ccfNA levels and routine biochemical and cellular parameters, including complete blood count, albumin, and CRP, which can influence ccfDNA and ccfRNA. Although previous studies have reported significant correlations of ccfDNA with albumin, CRP, and leukocyte and neutrophil counts in patients and/or disease-free individuals [22, 73–75], we were unable to assess these associations in long-lived individuals due to the lack of available data in our cohort.

Conclusion

In conclusion, the multiparametric analysis enabled us to highlight ccfDNA and ccfRNA changes during extreme aging, as well as similarities between NCO and CEN. These potential biomarkers of longevity could also be used as biomarkers of healthy aging, given the close relationship between longevity and healthy aging [7, 76]. Further studies will be necessary to explore the potential of these longevity biomarkers in clinical applications related to age-associated diseases.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

N.P.T., L.M.H., A.D., C.D. and C.H. performed the experiments. H.B. provided the plasma samples. N.P.T., A.D., M.S. and A.H–K. analyzed the data, made the figures and tables. A.H–K. conceived and supervised the project. N.P.T. drafted the first version of the manuscript. All authors read, edited and approved the final version of the submitted manuscript.

Funding

We acknowledge the financial support from the Agenomics project (Les Mutuelles AXA, programme de mécénat santé 2022 – Axe grand âge) and the GENMED Laboratory of Excellence on Medical Genomics [ANR-10-LABX-0013], for the project and/or publication fees. N.P.T. is financially supported by the Agenomics project.

Declarations

Conflict of interest

The authors declare no competing interests.