Introduction

Frailty is a complex age-related clinical syndrome characterized by a progressive decline in functioning across multiple physiological systems, with consequent increased vulnerability to stressors and high risk of hospitalization and mortality1. Multiple clinical tools to objectively quantify this multifaceted condition have been developed. The frailty phenotype defines frailty as a clinical syndrome, and classifies individuals as non-frail, pre-frail or frail on the basis of five physical criteria: weakness, physical activity, gait speed, exhaustion, and unintentional weight loss2. Frailty is a dynamic condition, meaning that an individual can fluctuate between states of severity of frailty in both worsening and improving directions. This evidence suggests that the factors implicated in frailty pathogenesis may be targeted with preventive interventions especially in its early stage, i.e., pre-frailty, which holds better chances to revert to non-frail status compared to frailty3,4.

Despite recent research efforts, validated biomarkers able to distinguish between non-frail, pre-frail, and frail subjects are still lacking5. We recently investigated by High-Performance Liquid Chromatography (HPLC) the concentration of a pool of neuroactive L- and D-amino acids which modulate the activation of glutamatergic receptors (L-glutamate, L-aspartate, glycine, D-serine and their precursors L-glutamine, L-asparagine, L-serine) in the blood of a cohort of elderly subjects encompassing the entire fit-to-frail continuum6. This study showed that D-serine levels independently predicted frailty severity, and that increased glycine/L-serine ratio correlated with worse cognitive decline and depressive symptoms in frail individuals. Although these findings suggest a putative role for these molecules in frailty, the employed HPLC approach did not allow us to evaluate the systemic metabolic framework on which these specific biochemical abnormalities are grounded. In keeping with this, in the present work we sought to explore the serum metabolome of the same well-characterized cohort of older subjects6 using a state-of-the-art untargeted 1H-NMR-based spectroscopy.

Blood sera of 96 subjects classified as frail (n = 37), pre-frail (n = 20), and non-frail (n = 39) were studied for their metabolomic profile using NMR spectroscopy analysis, as previously reported7. The clinical-demographic characteristics of participants are reported in Table 1. 1H-NMR assignment detected the presence of 45 metabolites for each spectrum (Fig. S1). Partial least-squares discriminant analysis (PLS-DA) score plots comparing frail, pre-frail, and non-frail groups (Fig. 1a) showed a clear separation of the three clusters with different metabolomic profiles reporting significant validation indexes (0.93 and 0.97 accuracy PC1 and PC2, respectively, with positive 0.78 and 0.79 Q2 indices). The main metabolites responsible for the metabolomic differences in the three groups were identified using VIP score analysis (Fig. 1b). In particular, increased malonate, leucine, and succinate characterized the blood metabolomic profile of frail subjects, while betaine, histidine, serine, glutamate and ornithine were remarkably upregulated in the serum of the pre-frail group. Most importantly, the discriminatory power of betaine, glutamate, histidine, and malonate was also confirmed in the comparison between frail and pre-frail participants (Fig. 1c, d).

Fig. 1: Multivariate analysis reveals different serum metabolomic profiles between frail, pre-frail and non-frail participants.
figure 1

a PLS-DA score scatter plots related to serum from pre-frail (n = 20), frail (n = 37) and non-frail participants (n = 39). The cluster analyses are reported in the Cartesian space described by the main components PC1:17.9% and PC2:7.4%. PLS-DA was evaluated using cross-validation (CV) analysis. CV tests performed according to the PLS-DA statistical protocol show a significant cluster separation (0.93 and 0.97 accuracy PC1 and PC2, respectively, with positive 0.78 and 0.79 Q2 indices). b VIP score graphs of metabolites discriminating the three clusters. c PLS-DA score scatter plots related to serum from pre-frail (n = 20), and frail subjects (n = 37). The cluster analyses are reported in the Cartesian space described by the main components PC1:17.4% and PC2:17.5%. PLS-DA was evaluated using cross-validation (CV) analysis. CV tests performed according to the PLS-DA statistical protocol show a significant cluster separation (0.76 and 0.87 accuracy PC1 and PC2 respectively, with positive 0.69 and 0.75 Q2 indices). d VIP score graphs of metabolites discriminating the two clusters.

Full size image
Table 1 Demographic features of the elderly cohort stratified according to Fried’s frailty phenotype
Full size table

We subsequently used enrichment analyses to identify the distinct biochemical pathways modulated by the frailty metabotype. Pairwise comparisons revealed the significant enrichment of 19 and 28 pathways in pre-frail and frail participants, respectively, compared to non-frail subjects. The dysregulated pathways were mainly related to the metabolism of amino acids, ammonia recycling, energy metabolism and lipids metabolism (Fig. 2a, b and Table S1). The comparison between pre-frail and frail subjects showed six enriched pathways, mostly associated with amino acids metabolism (Fig. 2c). Interestingly, glycine-serine metabolism emerged as one of the most significantly enriched pathways in all three comparisons (Fig. 2a–c). In addition, to disclose potential metabolomic signatures specifically associated with pre-frail and frail phenotypes, we extracted from enrichment analyses the pathways revealed by the comparisons with the non-frail group and represented them in a Venn diagram (Fig. 2d). Of note, betaine metabolism emerged as the exclusive dysregulated pathway of pre-frail compared to non-frail subjects. Conversely, frail patients showed a specific enrichment in 10 pathways related to lipid metabolism (ketone bodies, oxidation of branched-chain fatty acids, phosphatidylethanolamine biosynthesis, sphingolipids metabolism, butyrate metabolism and propanoate metabolism), energy and mitochondrial metabolism (carnitine metabolism and citric acid cycle), histidine metabolism and folate metabolism. Finally, frail and pre-frail groups revealed a common dysregulation of several amino acid pathways (Fig. 2d).

Fig. 2: Analysis of enrichment pinpoints shared and unique deregulated biochemical pathways among frail, pre-frail, and non-frail individuals.
figure 2

Enrichment pathways analysis performed comparing a non-frail vs pre-frail, b non-frail vs frail and c frail vs pre-frail subjects. The discriminative pathways are ranked according to p-value and number of hits reported in the bars. d Venn diagram displaying the disrupted pathways emerged from the comparisons of frail and pre-frail subjects with non-frail controls. Blue box reports the unique pathway dysregulated in pre-frail but not in frail subjects; light yellow box reports the pathways enriched in frail but not in pre-frail subjects; dark yellow box reports the common pathways dysregulated in both frail and pre-frail participants.

Full size image

Next, we performed univariate analysis using robust volcano plot, which showed (i) increased serine, histidine, glutamate, ornithine, and betaine and (ii) decreased glutamine, aspartate, phenylalanine, tyrosine, alanine, valine and formate levels in pre-frail subjects compared to non-frail subjects (Fig. 3a). Moreover, robust volcano plot analysis revealed (i) decreased levels of the amino acids aspartate, valine, phenylalanine, tyrosine and (ii) increased malonate, arginine and serine levels in frail subjects compared to non-frail controls (Fig. 3b). Finally, we observed higher concentrations of glutamate, betaine, and histidine and lower concentrations of arginine and malonate in the sera of pre-frail subjects compared to the frail group (Fig. 3c).

Fig. 3: Robust volcano plots highlighting the upregulated and downregulated metabolites in the serum of frail and pre-frail subjects.
figure 3

a–c Volcano plot analyses of metabolic changes in pre-frail vs non-frail, frail vs non-frail and pre-frail vs frail subjects’ serum. Each point on the volcano plot was based on p- and fold-change values, set at 0.05 and 1.0, respectively. Red and blue circles identify upregulated and downregulated metabolites, respectively. Variations are expressed as follows: panels a-b are graphed as a function of pathological group; panel c is graphed as a function of pre-frail group.

Full size image

Given the potential role of sex in modulating frailty-related phenotypes8 and serum metabolomic profile9, we further carried out robust volcano plots between sexes. Interestingly, our data showed that the increase in betaine and glutamate were characteristic of the pre-frail females when compared to non-frail females (Fig. S2a). Furthermore, the comparison between pre-frail and frail females revealed the influence of sex in modulating betaine and serine blood levels, which were specifically upregulated in pre-frail females (Fig. S2c, Fig. S4). Conversely, frail and pre-frail males showed a peculiar up-regulation of malonate compared to non-frail males (Fig. S3a). The discriminating role of malonate in prefrail males is further confirmed when compared its occurrence with frail males (Fig. S3c).

Previous studies investigating blood metabolomics in frail subjects showed heterogeneous and often inconsistent results (reviewed in Shekarchian et al.10 and Table S2). This inconsistency is likely attributable to differences in the clinical tools used to classify frailty, analytical platforms employed, comorbidities, ethnicity, exercise levels and diet regimens. However, few classes of metabolites often emerged as dysregulated in frail populations, including antioxidants, amino acids11 and mitochondria-related12 molecules. Here, we attempted to shed light on this matter by evaluating the serum metabolome in a well-characterized elderly cohort whose clinical and serum amino acid profiles were already known6. Our present findings confirm a complex dysregulation of amino acids, lipids, and energy metabolism-related molecules in frail and pre-frail patients, with glycine-serine metabolism emerging among the most significant discriminating pathways. Strikingly, glycine-serine metabolism was the pathway featuring the highest number of metabolite hits altered in both frail and pre-frail subjects (n = 11, Fig. 2a–c). Multivariate and univariate analyses further supported this result by showing upregulated serine levels in frail and pre-frail compared to non-frail subjects, nicely mirroring our previous findings6. Moreover, we highlighted a remarkable upregulation of glutamate and downregulated glutamine levels as biochemical signatures specifically associated with pre-frailty. These findings are in line with previous metabolomics studies showing dysregulated glycine-serine metabolism13,14,15,16 and increased blood glutamate levels15,17,18,19,20,21,22 in frail and pre-frail subjects compared to non-frail controls. Interestingly, increased serum aspartate concentration was reported in frail compared to non-frail individuals15,16,17,19,20, while we found downregulated aspartate levels in both pre-frail and frail patients. Beyond differences in study design, this discrepancy may be due to the lower proportion of sarcopenic patients in our pre-frail and frail groups (5%; see Table 1) compared to previous studies (100% in frail group17,19, 13% and 50% in pre-frail and frail groups, respectively15). Since high blood aspartate levels were previously associated with sarcopenic trait23,24, our findings suggest that serum aspartate may be differentially modulated among frail subjects with and without sarcopenia. Overall, these results entirely support our previous HPLC findings6 further confirming a role of amino acids acting on glutamatergic transmission as putative biochemical signature of frail aging. In this regard, besides their pivotal neuroactive role, these biomolecules are directly involved in the metabolism of several peripheral organs, including the liver, kidney, skeletal muscle and immune system25,26. Therefore, the disrupted amino acids homeostasis highlighted in this and previous studies most likely signals a decline in functioning across multiple organs and physiological systems that characterizes frailty, rather than a specific fingerprint of brain health. In addition, frail and pre-frail subjects showed dysregulated levels of several metabolites related to lipids, urea cycle and mitochondrial metabolism, including the tricarboxylic acid cycle. These observations are consistent with previous studies10,11,12,27 and, in turn, strengthen the hypothesis that energy metabolism and mitochondrial dysfunction play a key etiopathogenetic role in frailty28.

Despite several investigations explored the peripheral metabolomic profile of frailty, data focused on pre-frailty are scarce. Here, we identify a metabolomic signature able to distinguish pre-frail from frail and non-frail individuals. Specifically, betaine metabolism emerged as the distinctive pathway exclusively enriched in pre-frail subjects. Betaine, or trimethyl glycine, is an amino acid derivative that can be endogenously synthesized by the oxidation of choline and exogenously absorbed with diet29. Betaine serves as an alternative methyl donor in the methionine cycle, where it transfers a methyl group to homocysteine to synthesize methionine and dimethyl glycine. Dimethyl glycine is then catabolized to glycine and serine through multiple enzymatic reactions in the mitochondria of kidney and liver29. Remarkably, dysregulated methionine metabolism in both pre-frail and frail groups and elevated plasma homocysteine levels were previously associated with frailty in older adults30. Moreover, preclinical studies suggested that betaine supplementation counteracts oxidative stress and inflammation31, which have been proposed as key physiopathological mechanisms of frailty32. Although increased betaine levels could result from the mitochondrial dysfunction that characterizes frailty12 or be secondary to the elevated serine levels observed in pre-frail subjects (Fig. 1b), we speculate that the higher serum betaine concentration mainly found in the pre-frail females represents an adaptive mechanism to counteract an abnormal increase in homocysteine levels and protect against oxidative stress and inflammation. This protective metabolic mechanism may subsequently be lost during the transition from pre-frailty to overt frailty.

Due to the relatively limited sample size of the included cohort, this exploratory study requires further validation in independent larger cohorts of pre-frail and frail individuals. Future studies should investigate the role of age, BMI, diabetes, and cigarette smoking as putative confounding factors that could impact the serum metabolomic profile of pre-frail and frail subjects. However, our findings have practical implications for future clinical research aimed at evaluating amino acids and betaine metabolism as a potential source of suitable blood biomarkers and targeted interventions to counteract frailty in its early stages.

Methods

Participants enrollment and inclusion/exclusion criteria

Thirty-four consecutive hospitalized subjects were recruited at the Physical Medicine and Rehabilitation Unit of Istituto Santa Margherita, Pavia, Italy, between February 2019 and August 2021. Sixty-two additional outpatients were recruited at the Endocrinology and Nutrition Unit of the same institute. The patients were included if (1) admitted for functional loss secondary to a non-disabling disease; (2) aged 65 years or older. The following exclusion criteria were applied: (1) any disease that could directly affect muscle strength (including neurological diseases, hip fractures or amputations); (2) dementia according to DSM-5 criteria33; (3) any systemic condition potentially affecting serum amino acid levels, including kidney, liver, rheumatologic and neoplastic diseases, history of drug or alcohol abuse; (4) history of altered serum creatinine levels ( > 1.2 mg/dl) or liver function parameters (aspartate transaminase or alanine transaminase >50 U/l). Smoking status (current/former/never smoker) was assessed trough interview. The total number of drugs habitually taken by subjects was retrieved from medical records. This study was approved by the local ethics committee (protocol 20180097520, 09/11/2018) and was in conformity with the Helsinki Declaration. Written informed consent was obtained from all participants.

Cognitive evaluation

Each subject underwent a standardized examination including evaluation of global cognition, performed through the Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA)34.

Sarcopenia and visceral adiposity

Body composition (fat mass (FM) and fat-free mass (FFM)) was evaluated using fan-beam dual-energy X-ray absorptiometry (DXA) (Lunar Prodigy DXA, GE Medical Systems). The in vivo coefficients of variation were 0.89% and 0.48% for FM and FFM, respectively. Skeletal Muscle Index (SMI) was calculated as the sum of fat-free soft tissue mass of arms and legs divided for height squared. Sarcopenia was defined according to the European Working Group on Sarcopenia in Older People (EWGSOP2) definition35 as low muscle strength (handgrip below the proposed cut-off values) associated to low muscle quantity (SMI below the proposed cut-off points). Visceral adipose tissue (VAT) volume was estimated using a constant correction factor (0.94 g/cm3). The software automatically placed a quadrilateral box, representing the android region, outlined by the iliac crest and with a superior height equivalent to 20% of the distance from the top of the iliac crest to the base of the skull.

Functional performance and independence

Handgrip measurement was assessed on the dominant hand and was considered “strong” or “weak” based on sex and body mass index (BMI)-adjusted cut-off scores, as previously described6.

Nutritional status

Nutritional status was evaluated with Mini Nutritional Assessment (MNA), which is composed of 18 items divided in four categories: anthropometric assessment, general state, dietary assessment and self-assessment. A score ≥24 points indicates a good nutritional status; a score between 17 and 23.5 points indicates risk of malnutrition, while a score ≤17 points indicates malnutrition36.

Frailty

Frailty was evaluated with the frailty phenotype described by Fried and colleagues2, with slight modifications. The physical frailty phenotype contains 5 criteria, including weight loss, exhaustion, low physical activity, slow walking speed and low grip strength7. Participants who met 3 or more criteria were defined frail, those who met 2 criteria were classified as pre-frail and those who met 0 or 1 criteria were defined non-frail.

NMR sample preparation and acquisition

NMR samples were prepared as previously reported7. NMR samples were created by combining 250 μL of phosphate buffer (0.075 M Na2HPO4·7H2O, 4% NaN3, and H2O) with 250 μL of blood serum and transferred to a 5 mm NMR tube. To align and quantify NMR signals, trimethylsilyl propionic acid and sodium salt (0.1% TSP in D2O) were utilized as an internal reference. A Bruker DRX600 MHz spectrometer (Bruker, Karlsruhe, Germany) fitted with a 5 mm triple-resonance z-gradient TXI Probe was used for NMR investigations. TOPSPIN, version 3.2, was used for spectrometer control and data processing (Bruker Biospin, Fällanden, Switzerland). Carr-Purcell-Meiboom-Gill (CPMG) experiments were acquired with a spectral width of 7 kHz, 32 k data points; water presaturation was applied over a 5 s relaxation delay, and a spin-echo delay of 0.3 ms. A weighted Fourier transform was applied to the time domain data with a 0.5 Hz line-broadening, followed by a manual phase and baseline correction in preparation for targeted profiling analysis. Resonance assignment was performed with Chenomx software. The quantification of assigned metabolites was carried out using automated Bayesil software. (Fig. S1)

Statistical analysis

Clinical and demographic characteristics were described using, as summary statistics, median and the interquartile range (IQR) or absolute and relative frequencies. Comparisons of continuous variables between non-frail, pre-frail and frail subjects were evaluated using Kruskal-Wallis test or two-way ANCOVA with frailty status and sex as factors, age as covariate. Post-hoc tests were adjusted with Bonferroni correction. Likelihood-ratio test was used for categorical variables.

The matrix was analyzed using a univariate approach combining T-test and Fold-change through the robust volcano plot37 by setting the threshold of the fold change to a value of 1 and the p-value to less than 0.05 (Table S3). Partial least-squares discriminant analysis (PLS-DA) was performed with normalized metabolomics data using MetaboAnalyst 5.0 (http://www.metaboanalyst.ca/)38. The performance of the PLS-DA model was evaluated using the coefficient Q2 (10-fold internal cross-validation method) and the coefficient R2, defining the variance predicted and explained by the model, respectively. Both indicators are considered significant if they have positive values; in addition, the accuracy of the model was assessed. Clusters of discriminatory metabolites were ranked according to their variable influence on projection (VIP) scores. VIP scores are the weighted sums of squares of the PLS-DA weights, indicating the variable’s importance and is statistically significant if > 1. The analysis of the pathways was carried out using the Enrichment tool based on the global test39 and using the Small Molecules Pathways Database (SMPDB) for Homo sapiens as pathway enrichment. Only pathways with false discovery rate (FDR)-adjusted p values lower than 0.05 and hit values (i.e., the number of metabolites belonging to the pathways) greater than 1 were considered40.