Patient recruitment

This metabolomics study was performed as a secondary analysis on plasma samples obtained from 22 hip fracture patients of both sexes undergoing surgery. Briefly, patients aged over 70 years with a proximal hip fracture of the femur undergoing surgery were recruited from November 2017 to March 2019. Patients were excluded if they suffered from specific neuromuscular diseases (myasthenia gravis, muscular dystrophy, ALS, polio), severe dementia, chronic inflammatory disease (e.g., Crohn’s disease, ulcerative colitis, rheumatoid arthritis with systemic anti-inflammatory therapy), or have been subject to systemic corticosteroid therapy (above 7.5mg per day), or cancer therapy in the last 5 years. All participants provided written informed consent before enrolment. Informed consent was taken before surgery with enough time to think about participation. In case of concerns of the patient, the patient was not included.

Patient data

The collected data includes information on the demographic, family, and socioeconomic characteristics, alcohol intake, smoking, and comorbidities as well as anthropometry (weight, height, BMI, fat mass (FM), and fat mass index (FMI). The anthropometric measurements were obtained by physical examination of the study participants by trained study personnel. Bio-Impedance Analysis (BIA; BIA 101, Akern, Florence, Italy) was performed after surgery and used for measuring lean mass. Measurements were taken under standard conditions, with the patient in a supine position and surface electrodes placed on the wrist and ankle contralateral to the side of the fracture. Appendicular lean mass (aLM) was estimated using the equation developed by Sergi et al. [10]. The skeletal muscle index [SMI, (kg/m2)] was calculated by dividing aLM by body height squared. Assigned cutoffs of 7 kg/m2 in men and 5.5 kg/m2 in women were used to define low SMI. Handgrip strength was assessed with a Saehan DHD-1 Digital Hand Dynamometer, with the patient lying supine. The maximal value of three consecutive measurements of both hands was used for the analysis. Similar to SMI, handgrip strength was defined as low, if below 27 kg and 16 kg, in men and women, respectively [7]. A z score combining handgrip strength and muscle mass was calculated separately for men [z score sarcopeniamen = [(27–handgrip strength)/SD (handgrip strength)] + [(7.0–SMI)/SD (SMI)] and women [z score sarcopeniawomen = [(16–handgrip strength)/SD (handgrip strength)] + [(5.5–SMI)/SD (SMI)]. Z-transformation of both values represents the degree of sarcopenia on a metric scale. The higher the z-score, the more sarcopenic the patient. Data were provided on the insulin growth factor (IGF) axis parameters including IGF-1, insulin growth factor binding protein 3 (IGFBP3) and IGF1/IGFBP3 ratio.

IGF-I and IGFBP3 measurement

Blood samples for measurement of serum concentrations of IGF-I and IGFBP3 were centrifuged and serum was stored at −80°C until analysis. Serum hormone concentrations (ng/ml) of IGF-I and IGFBP3 were measured at the Endocrine Laboratory of the University Hospital Munich using the IDS-iSYS automated chemiluminescent assay system (Immunodiagnostic System Ltd., Boldon, England, UK). Validation data for all assays and reference intervals have been published elsewhere [11, 12]. The assays are calibrated against the latest recombinant standards (02/254 for IGF-I).

Metabolomic measurements

For the metabolomic measurements, plasma samples were obtained after centrifugation of blood samples collected from patients in EDTA tubes, then stored at −80°C until analysis. Overall, a targeted metabolomics approach was applied for measuring a total of 300 metabolites in the patients’ samples at the Dept. of Paediatrics, LMU Munich. Concentrations were calculated in μmol/l. The measured metabolites belonged to the following classes:

Amino acids

Twenty-two amino acids including alanine (Ala), arginine (Arg), asparagine (Asn), aspartic acid (Asp), glutamine (Gln), glutamic acid (Glu), glycine (Gly), histidine (His), isoleucine (Ile), leucine (Leu), lysine (Lys), methionine (Met), phenylalanine (Phe), serine (Ser), threonine (Thr), tryptophan (Trp), tyrosine (Tyr), valine (Val), citrulline (Cit), ornithine (Orn), and proline (Pro) were analyzed in plasma samples obtained from patients by ion-pair liquid chromatography with tandem mass spectrometry (HPLC-MS/MS ) as previously described by Harder et al. [13].

Acyl carnitines

Carnitines (free carnitine (Carn) and acylcarnitine (Carn.a)) were analyzed using a modified method from Giesbertz et al. [14]. Briefly, proteins of 50 μL plasma samples were precipitated by a tenfold amount of methanol including isotopic labeled internal standards D3-Carnitine C2 (DLM-754-PK, Cambridge Isotope Laboratories), D3Carnitine C8 (DLM-755-0.01, Cambridge Isotope Laboratories), and D3-Carnitine C16 (DLM-1263-0.01, Cambridge Isotope Laboratories). After centrifugation, 50 μL of the supernatant was evaporated to dryness under a gentle stream of nitrogen at 40 °C. The residuals were re-dissolved in 50 μL hydrogen chloride-1-butanol solution, and derivatization was conducted at 60 °C for 10 min shaking at 600 rpm. Thereafter, the hydrogen chloride-1-butanol solution was evaporated to dryness and the residuals re-dissolved in 50 μL methanol. The butylated acylcarnitines were separated on a 1200-SL HPLC system (Agilent Technologies, Waldbronn, Germany) equipped with a degasser, pump, autosampler, column oven, and a 150 × 2.1 mm Kinetex® reversed-phase column with 2.6 μm particles (Phenomenex, Torrance, USA). Mobile phase A consisted of 5mM ammonium acetate in water and mobile phase B consisted of 333 μL 7.5 M ammonia acetate in 1 L methanol/ acetonitrile/isopropanol (1:4:5). The mass spectrometric detection was performed on a hybrid triple quadrupole mass spectrometer (4000 QTRAP, AB Sciex, Darmstadt, Germany) with a Turbo Ion source operating in negative ESI mode.

Non-esterified fatty acids (NEFA)

Sixty-three non-esterified fatty acids were measured in patients’ samples, using HPLC-MS/MS run in negative ESI mode as described previously by Hellmuth et al. [15]. The same formula CX:Y was used to indicate the chain length as well as the number of double bonds.

Bile acids

A new method for bile acids analysis was developed and validated using HPLC-MS/MS. The method description is presented in the supplementary material (Supplementary M1). Briefly, seventeen bile acids were measured, including cholic acid (CA), chenodeoxycholic acid (CDCA), deoxycholic acid (DCA), lithocholic acid (LCA), glycocholic acid (GCA), taurocholic acid (TCA), glycochenodeoxycholic acid (GCDCA), taurochenodeoxycholic acid (TCDCA), glycodeoxycholic acid (GDCA), taurodeoxycholic acid (TDCA), glycolithocholic acid (GLCA), taurolithocholic acid (T LCA), taurocholic acid 3-sulfate (TCA-3S), and taurolithocholic acid 3-sulfate (TLCA-3S).

Quality control

Due to the limited number of samples, only one batch was used in all analyses and quality control samples (QC) were used to check the within-batch variations (intra-batch CV% = 20%). Six QC prepared by pooling aliquots of all available study samples were consistently measured at regular intervals within the batch at the beginning, middle, and end of the batch. Measurements greater than 1.5 standard deviations (SD) away from the next closest measurement were considered as outliers and subsequently set to NA (not available). Measurements with >50% missing values were excluded.

Statistical analysisDescriptive statistics

The demographic and phenotypic characteristics of the study participants, including age, sex, BMI, comorbidities (diabetes mellitus, rheumatoid arthritis, thyroid, and parathyroid dysfunction, spine diseases, chronic lung diseases, kidney diseases, cancer, diarrhea intolerance), smoking, alcohol and drug intake (proton pump inhibitors (PPI), corticosteroids, anti-estrogenic therapy, tranquilizers), mobility problems, dizziness, stumbling, falls during the preceding year to the study, and activities as sports, daily outdoor activities, were summarized as mean (SD) and proportions for continuous and categorical variables, respectively. Wilcoxon rank-sum test and Fisher’s exact test were used to investigate the differences between groups for numerical and categorical variables, respectively. Results are shown in Table 1.

Table 1 Study population characteristicsData analysis

For the metabolomics data, after normalization and scaling, linear regression models were used to study the associations between plasma metabolite levels and different measures of sarcopenia (sarcopenia z scores and SMI) using the sarcopenia measures as the outcome and the plasma metabolites as the independent variables. Models were initially adjusted using potential confounders including age, sex, BMI, smoking, alcohol intake, and comorbidities such as cancer; however, it was noticed that the associations between the metabolite levels and the sarcopenia measures were not appreciably influenced by the inclusion of these confounders, hence they were not included in the models, especially considering the small sample size. Volcano plots were used to depict the results of the models with β on x-axis and |log10 (P)| values on y-axis indicating the sign, magnitude, and strength of the association, respectively. False discovery rate (FDR) [16] was used to minimize the occurrence of false positives, a common issue in multiple testing. Nevertheless, we also inspected associations with uncorrected P values for further interpretation, because of the exploratory nature of the analysis and regarded some as potentially meaningful differences, principally if they are common among the same metabolite class or subclass and share the same tendencies. These associations were referred to as trends albeit not significant after FDR correction due to the low statistical power. The cutoff for uncorrected P values was depicted as a red dotted line in the volcano plot. Additionally, sarcopenic and non-sarcopenic patient groups were defined using the sarcopenia z score cutoff values. To explore the group differences, unadjusted comparisons using multiple univariate tests were performed within the Metaboanalyst 5.0 software which includes fold change (FC) analysis and Wilcoxon rank-sum test. Then, a combination of both tests was used to produce a volcano plot using a cut-off of 1.5 and 0.05 for FC and P value, respectively [17]. Concomitantly, for class discrimination and identification of metabolites responsible for group separation, multivariate analyses were also conducted including principal component analysis (PCA), partial least squares–discriminant analysis (PLS-DA), and orthogonal partial least squares–discriminant analysis (oPLS-DA).

Metabolite set enrichment analysis (MSEA) was also performed using the metabolomics data sets, and the pathway was considered significantly enriched if P values were smaller than 0.05 and those significant after FDR correction were inspected. Both multivariate analysis and pathway enrichment analysis were carried out also using Metaboanalyst 5.0. Causal effect relationships involved in sarcopenia were investigated using Mediation Analysis by PLSSEM [18,19,20] using SmartPLS software. First, factor analysis was performed for the selection of indicator variables most associated with the relevant latent variables for each of the metabolite class (AA, NEFA, BA, Carn.a, TCA). Then bias-corrected and accelerated bootstrap was conducted to test the statistical significance of the investigated pathways using 0.05 as a significance level and the total effects, total indirect effects, and specific indirect effects were calculated.