Subsets of the COmPLETE-Health (n = 116, mean age 70.8 ± 8.7 years old, 65% male) and COmPLETE-Heart samples (n = 54, mean age 67.0 ± 11.5 years old, 91% male) were investigated. As reported in the study protocol, only clinically healthy participants from the Basel area (Switzerland), who had no exercise-limiting chronic diseases and were non-smokers (or had quit at least 10 years previously) were included in the COmPLETE-Health sample. This excluded participants with a history of CAD, stroke, heart failure, lower-extremity artery disease, any kind of malignant tumor, diabetes, obesity, clinically apparent kidney failure, severe liver disease, chronic obstructive pulmonary disease GOLD stages two to four, arterial hypertension grades two and three, drug or alcohol abuse, exercise-limiting osteoporosis or musculoskeletal conditions and clinically manifest Alzheimer's disease or dementia. The investigated subset of the COmPLETE-Heart sample consisted exclusively of patients suffering from CAD, diagnosis of which was confirmed by senior cardiologists. Patients with unstable angina pectoris, uncontrolled brady- or tachyarrythmia, permanent atrial fibrillation, severe valvular disease, acute myocardial infarction, transient ischemic attack, or stroke in the last 3 months were excluded. The exact recruitment procedure and the full list of inclusion and exclusion criteria can be found in the COmPLETE study protocol (49). The COmPLETE study was funded by the Swiss National Science Foundation (Grant No. 182815) and approved by the Ethics Committee of North-Western and Central Switzerland (EKNZ 2017-01451). A written informed consent document was obtained from all participants prior to inclusion.

Data collection was carried out between January 2018 and June 2019. Medical history and medication were reviewed by a physician using a standardized questionnaire. Participants were randomly allocated to one of five time slots (08:00, 10:00, 12:00, 14:00 and 16:00) for the measurements, which took around 4 h in total. They were instructed not to diverge from habitual eating behavior (for the previous 72 h), to avoid exercising, drinking alcohol (for the previous 24 h) and drinking caffeinated beverages (for the previous 4 h). On the day of sample collection, participants took the prescribed medication as usual. Fasting blood samples (at least 3 h fasting time) were collected before any kind of measurements involving physical activity. Trained medical staff collected serum samples (2 × 7.5 mL serum-gel, Monovette®, Sarstedt, Nümbrecht, Germany) by venipuncture in the cubital fossa. Serum samples were slightly shaken for 30 min, centrifuged (3,000 rpm; 10 min; 20–23°C) and aliquoted before being frozen at −80°C.

Height and body mass were measured to the nearest 0.5 cm and 0.1 kg, respectively. Body mass index was calculated as kg/m². Body composition was analyzed using a four-segment bioelectrical impedance analysis (Inbody 720, Inbody Co. Ltd., Seoul, South Korea). Using InBody 720, -as opposed to dual-energy x-ray absorptiometry analysis, to measure appendicular muscle mass was judged to be acceptable (50). After a rest phase of 10 min, resting systolic and diastolic blood pressures and resting heart rate were measured in supine position using a non-invasive vascular screening system (VaSera VS-1500 N; Fukuda Denshi, Tokyo, Japan). The peak oxygen uptake (VO₂ peak) was used as marker for the cardiorespiratory fitness and was determined during an exercise test until maximal exertion (i.e., volitional exertion, dyspnea, or fatigue) using an electromagnetically braked cycle ergometer (Ergoselect 200; Ergoline, Bitz, Germany) and a computer-based system (MetaMax 3B; Cortex Biophysik GmbH, Leipzig, Germany). VO₂ peak was defined as the highest 30 s average of VO₂ at any point of the test.

Serum samples were analyzed for triglyceride, total cholesterol, high-density lipoprotein (HDL) and low-density lipoprotein (LDL) cholesterol concentrations using an Olympus AU680 automatic analyzer (Beckman Coulter, Brea, CA, USA), enzymatic reagents (DiaSys, Holzheim, Germany) and secondary standards (Roche Diagnostics, Mannheim, Germany). For glycated hemoglobin (HbA1c), whole blood was analyzed by high pressure liquid chromatography (HPLC) using D-10 (Bio-Rad, Hercules, CA, USA). NT-proBNP was determined using a chemiluminescent microparticle immunoassay (Architect, Abott, IL, United States).

A large panel of carnitine related metabolites, including free carnitine, deoxycarnitine and 36 acylcarnitine species (Supplementary Table 1) were targeted in serum samples. BCAAs were also measured, as C3- and C5-acylcarnitine species are byproducts of the BCAA catabolism (43–45). Analysis was conducted at the Metabolomics Platform, Faculty of Biology and Medicine, University of Lausanne (Switzerland). A detailed description of the method used is available (32).

Metabolite concentrations were log2-transformed and z-standardized prior to statistical analysis. Multiple linear regressions were run to assess associations between metabolites and CAD phenotype. To determine which confounders required adjustment for regressions, directed acyclic graph (DAG) were drawn (Supplementary Figure 1) (51, 52). Metabolites were used as dependent variables, while CAD phenotype and confounders served as independent variables. Regressions using acylcarnitines or carnitine as dependent variable were adjusted for the following confounders age (53, 54), sex (55), HbA1c (%) (11, 56, 57), body fat (%) (11, 58, 59), smoking habits (60), antihypertensive and lipid lowering medication (61–64) as well as fasting and sampling time (Supplementary Figure 1) (65–67). Regressions using BCAAs as dependent variable were adjusted for the following confounders: age, sex (67–70), skeletal muscle mass (71, 72), smoking habits (60, 73), sampling time and fasting time (Supplementary Figure 2) (65–67).

Two sets of multiple linear regressions were run. In the first set, CAD was defined as a categorical two-level variable opposing sickness vs. health. In the second set, CAD was defined as the number of stenosed coronary arteries (0, 1, 2, or 3). As the concentrations of deoxycarnitine, hydroxyhexanoylcarnitine (C6:0-OH), hydroxydodecanoylcarnitine (C12:0-OH) and arachidonylcarnitine (C20:4) were below the quantification limit (0.003 μM) of the HILIC-HRMS method employed for some participants (deoxycarnitine n=1, hydroxyhexanoylcarnitine n = 2, hydroxydodecanoylcarnitine n = 1, arachidonylcarnitine n = 1), a Tobit regression using the CensReg R package was applied for these metabolites to estimate regression coefficients in the presence of left censored values (74).

Graphical methods were used to assess linearity, normal distribution, and homoscedasticity of data. P-values were adjusted using the Benjamini-Hochberg (BH) method, separately within each set of multiple linear regression (75). Adjusted p ≤ 0.05 were considered as significant. Statistical analyses were carried out using R (version 4.0.2) (76). Rain plots were computed using a previously published R-code (77).

The investigated subjects consisted of 116 healthy individuals (70.8 ± 8.7 years, 65% male) and 54 patients with confirmed CAD (67 ± 11.5 years, 91% male) (Table 1). The clinically healthy individuals had normal body mass index (BMI) and HbA1c values. (78). Mean blood pressure values were in the high normal range, with 20% of participants treated for arterial hypertension grade one (79). Sixty-six percent of the clinically healthy individuals had never smoked, whereas 34% had quit smoking more than 10 years previously. Both clinically healthy controls and CAD patients were characterized by elevated LDL-cholesterol according to the 2019 ESC/EAS guidelines on primary (clinically healthy participants) and secondary (CAD patients) prevention (2). Eighty nine percentage of the CAD patients were under statin, compared to only 8% of the clinically healthy participants, which could explain the lower levels of LDL-cholesterol in the CAD patients. CAD patients displayed elevated NT-proBNP levels and a mean BMI value in the overweight range, while HbA1c, triglyceride and HDL-cholesterol levels as well as systolic and diastolic blood pressure values were normal (2, 78–80). It is worth noting that all CAD patients were on antihypertensive medications. Additionally, 50% of CAD patients were non-smokers, while 20% ceased smoking at least 10 years ago. Fasting duration was of at least 3 h with mean of 8.5 ± 5.3 h for those with CAD and 6.7 ± 3.0 h for clinically healthy individuals.

Figure 1 exhibits the results of the first set of regression, in which CAD was defined as a dichotomous variable opposing CAD vs. absence of CAD. It shows that 20 out of 30 quantified carnitines (seven short-chain, eight medium-chain and five long-chain acylcarnitines) were significantly and positively associated with CAD phenotype after adjustment for confounders (Table 2, Supplementary Table 2, Supplementary Figure 3). Hexanoylcarnitine (C6:0) showed the strongest positive association with CAD (β-coefficient 1.02; BH p ≤ 0.004) followed by palmitoylcarnitine (C16:0) (β-coefficient 1.02; BH p-value 0.002), and hexadecenoylcarnitine (C16:1) (β-coefficient 0.96; BH p-value 0.003) (Supplementary Table 2). Figure 2 represents the results of the second set of regressions, in which CAD was defined as the number of stenosed coronary arteries (zero-, one-, two- and three-vessel coronary artery disease). It shows that the strength of association (β-coefficient) increased with increasing number of affected coronary arteries for 17 acylcarnitines (3 short-, 8 medium- and 6 long-chain acylcarnitine species) (Supplementary Table 2). Besides, individuals with CAD had significant elevated levels of two short-chain hydroxylated acylcarnitines (C4:0-OH and C5:0-OH) and dicarboxylic acylcarnitine suberoylcarnitine (C8:0-DC) (Supplementary Table 2).

Regarding the cardiometabolic risk factors used as confounders, age showed significant and positive associations with more than 50% of the measured acylcarnitine species (one short-, eight medium- and eight long-chain species, including two hydroxylated acylcarnitines). Six medium-chain and six long-chain acylcarnitines were found to be significantly and negatively associated with HbA1c, while no significant association was found between acylcarnitines and smoking status.

Figure 3 exhibits the results of the first set of regression, in which CAD was defined as a dichotomous variable opposing CAD vs. absence of CAD. It shows that valine and isoleucine were significantly and positively associated with CAD phenotype (valine: β-coefficient 0.55; BH p-value 0.046 and isoleucine: β-coefficient 0.52; BH p-value 0.046) (Supplementary Table 2). Conversely, no significant association was found between BCAAs and the number of stenosed coronary arteries (Figure 4). Concerning the cardiometabolic risk factors used as confounders, valine and leucine showed a significant negative association with age and a significant positive association with the male sex. No significant association was found between muscle mass and BCAAs.

In the present study, circulating medium- and long-chain acylcarnitines, especially C6:0, C8:0, C8:1, C12:1, C14:1, C16:0, C16:1, C18:1, and C20:4, were found to be elevated in CAD patients. This accumulation could be explained by a dysregulation in carnitine shuttle enzymes and by an inefficient beta-oxidation as previously demonstrated (34, 38, 40, 44, 56, 87–90). The main carnitine shuttle enzymes are carnitine palmitoyltransferase 1 (CPT1) and carnitine palmitoyltransferase 2 (CPT2), which are responsible for the conversion of acyl-CoA and carnitine to free CoA and acylcarnitine and the opposite reaction, respectively. The conversion of acyl-CoA to acylcarnitine allow fatty acids longer than 10 carbon atoms to be transported across the mitochondrial membrane for subsequent beta oxidation. In ischemic conditions, CPT1 activity is increased and CPT2 activity decreased, leading to an accumulation of medium- and long-chain acylcarnitines (Figure 5) (87). Furthermore, ischemia leads to an altered beta-oxidation, which may be attributed to impaired function of fatty acid oxidation enzymes or increased fatty acid oxidation relative to tricarboxylic acid (TCA) flux (12, 38), both leading to accumulation of acyl-CoA (12, 34, 47, 88, 91, 92). Excess acyl-CoA can be retroconverted to acylcarnitine, which can then be excreted via blood and urine, thus detoxifying mitochondria of excess carbons (Figure 5) (37, 39, 88, 93).

Our results are in line with previous findings. Within medium-chain acylcarnitines, hexanoylcarnitine (C6:0) was reported to be able to discriminate patients with cardiovascular diseases from clinically healthy controls (16). Likewise, octanoylcarnitine (C8:0) was associated with cardiovascular mortality and reduced heart function (17, 94). Within long-chain acylcarnitines, palmitoylcarnitine (C16:0) has been associated with heart failure (94), cardiovascular mortality in patients with stable angina pectoris (17) and cardiovascular events in very old individuals with previous history of CAD (15). Similarly, oleoylcarnitine (C18:1) was shown to be able to predict cardiovascular events in elderly individuals with previous history of CAD (15).

We found elevated levels of several C3- and C5-acylcarnitine species, as well as of their precursors valine and isoleucine, in patients with CAD. Our results are consistent with those of previous studies, which found that elevated levels of circulating BCAAs (19, 28, 29) and short-chain acylcarnitines (19, 28) were associated with CAD and stroke in a population at high cardiovascular risk (95). BCAAs and acylcarnitines seem to interplay at different levels. First, chronic cardiac ischemia can disrupt the BCAA catabolism, leading to increased BCAA catabolism derivatives such as C3- and C5-acylcarnitines (38, 96, 97). Excess BCAAs can then impair fatty acid oxidation, which results in the accumulation of incompletely oxidized lipid species and acylcarnitines (5, 98).

Acetylcarnitine (C2:0), which is the most abundant circulating acylcarnitine (99), plays a central role in detoxifying mitochondria from excessive acetyl-CoA, the universal degradation product of all metabolic substrates (Figure 5) (93). Interestingly, we found that acetylcarnitine (C2:0) was not significantly associated with CAD phenotype and does not accumulate in CAD patients. As the acetyl-CoA is the main substrate of the TCA cycle, this observation implies that the capacity of the TCA cycle is not necessarily exceeded as previously postulated (12, 38, 98). Therefore, the globally elevated levels of acylcarnitines are likely due to impaired fatty acid oxidation and carnitine shuttle enzymes, rather than to reduced TCA flux.

The short-chain hydroxylated acylcarnitine hydroxybutyrylcarnitine (C4:0-OH) was found to be elevated in CAD. Accumulation of plasma hydroxybutyrylcarnitine (C4:0-OH) is used for the diagnosis and screening of patients with an inherited defect in the short-chain hydroxyacyl-CoA dehydrogenase (SCHAD), an enzyme of the mitochondrial fatty acid oxidation (100). This finding further supports an impaired beta oxidation in CAD.

Suberoylcarnitine (C8:0-DC) was positively associated with CAD phenotype. This is in line with the findings of Shah et al., which showed that a signature composed of short- and medium-chain dicarboxylic acylcarnitines was predictive of cardiovascular events in individuals with CAD (19). In addition to an alteration in mitochondrial fatty acid oxidation, elevated dicarboxylic acylcarnitine levels in CAD could indicate increased fatty acid omega-oxidation (101). Dicarboxylic acylcarnitines are byproducts of medium-chain dicarboxylic acids. The latter are the final products of microsomal omega-oxidation and of the subsequent peroxisomal beta-oxidation (102). Both mitochondria and peroxisomes perform fatty acid beta-oxidation but with different aims. Short-, medium and long-chain fatty acids are predominantly oxidized in mitochondria, whereas peroxisomes oxidize specific carboxylic acids such as very long-chain fatty acids, branched-chain fatty acids, bile acids, and fatty dicarboxylic acids (DCAs) (103, 104). The carnitine shuttle is then used to transport the end-products (acetyl-CoA, propionyl-CoA, and medium-chain acyl-CoA) from the peroxisome to the mitochondria for complete oxidation via the TCA cycle (Figure 5) (37, 38, 105). While the mitochondrial beta-oxidation is essential for catabolism and energy production, peroxisomal beta-oxidation is mainly involved in biosynthesis pathways (106). Altogether, in addition to impaired mitochondrial beta-oxidation, CAD patients also seem to have an altered peroxisomal and microsomal fatty acid oxidation (105).

An accumulation of medium- and long-chain acylcarnitines can impair several regulatory mechanisms in mitochondria. First, fatty acids and glucose intermediates compete as metabolic substrates for energy production (Randle cycle) (42). An intramitochondrial accumulation of long-chain acylcarnitines inhibits pyruvate and lactate oxidation, leading to metabolic inflexibility (Figure 5) (42) or the incapacity to switch between substrate for energy production depending on their availability (41). Secondly, an excess of long-chain acylcarnitines compromises membrane function, induces electrophysiological alterations through modulation of calcium and potassium channels (contributing to cardiac arrhythmias), promotes insulin resistance and inflammation, inhibits oxidative phosphorylation and stimulates the production of reactive oxygen species (34, 37, 87, 88, 107).

A dysregulation in the BCAA catabolism, with a back-up of BCAAs and their byproducts, also has important consequences. The BCAA catabolism intermediates branched-chain keto acids (BCKAs) can be cytotoxic at high levels, promoting mitochondrial dysfunction, superoxide accumulation, and cardiomyocyte death, eventually leading to heart failure (96). Furthermore, an accumulation of BCAAs sensitizes the heart to ischemic injury (108) and contributes to cardiac dysfunction and remodeling following myocardial ischemia (109, 110). This can be explained by various mechanisms including inhibition of glucose metabolism (108) and activation of the mammalian target of rapamycin (mTOR) (109, 110). Altogether, BCAA catabolism seems to be disrupted in individuals with CAD, provoking an accumulation of BCAAs and their byproducts, favorizing mitochondrial dysfunction.

While an accumulation of BCAAs, medium- and long-chain acylcarnitines have deleterious consequences, research has indicated that some short-chain acylcarnitines could also have positive effects (37). Evidence has suggested that propionylcarnitine (C3) increases cellular carnitine content, stimulates pyruvate dehydrogenase activity and increases TCA cycle efficiency under hypoxia (111, 112). Therefore, supplementation in propionylcarnitine (C3) could be beneficial in the treatment of cardiovascular disorders (112). For L-carnitine, several studies have also reported a protective role on the myocardium by exerting anti-apoptotic effects in cardiomyocytes (94, 113). A protective role of L-carnitine within the myocardium is further supported by studies in individuals with genetic carnitine deficiency developing cardiomyopathies (114). Importantly, in our study, propionylcarnitine and L-carnitine were found to be elevated in CAD patients. Considering the potential beneficial effects of these carnitine species, elevated levels of circulating L-carnitine and proprionylcarnitine observed in CAD patients in our study could suggest an attempt of the body to adapt to chronic ischemia. This hypothesis remains to be further investigated.

Mitochondrial dysfunction is a major determinant of metabolic disease such as metabolic syndrome, non-alcoholic fatty liver disease and type 2 diabetes mellitus, conditions which are highly linked to increased risk of cardiovascular disease and myocardial infarction (115, 116). This supports the use of acyclarnitines as markers of mitochondrial function and cardio-metabolic risk. Additionally, understanding the acylcarnitine metabolism in CAD could lead to new treatment targets. Most research on the effects of ischemia on mitochondrial enzymes has been done in acute ischemic situations. This study showed that altered mitochondrial metabolism reflected in high levels of circulating carnitine, acylcarnitines and BCAAs is also a hallmark of CAD, a chronic ischemic situation. These findings should be further investigated at an enzymatic level and in model organisms. There has been emerging evidence that manipulating fuel supply and substrate consumption of the myocardium could have an impact on the development and progression of heart failure (117, 118), which often occurs in CAD. Therapeutic interventions with “metabolic” antianginal agents that suppress fatty acid oxidation and increase the oxidation of pyruvate in the mitochondria could reduce the ischemia-induced accumulation of long-chain fatty acid intermediates and other disruption in cardiac metabolism (119). Hypoxia, anoxia and ischemia might have different effects on the cardiac metabolism and these conditions should be clearly distinguished in future works (33).

While previous studies analyzed L-carnitine and/or a limited number of acylcarnitines in patients with cardiometabolic diseases, we examined a large panel of acylcarnitine species and related BCAAs. This resulted in a thorough phenotyping of patients with CAD at the molecular species level (35 metabolites) comparing them to clinically healthy controls, which was rarely the case in previous research. Additionally, we conducted a detailed analysis of possible confounders and adjusted the regressions for those variables. Given the exploratory nature of this study, several limitations should be taken into consideration. First, as multiple organs usually contribute to the circulating pool of metabolites, it is difficult to identify the cellular origin, destination or subcellular localization of these metabolites (120). Therefore, the findings of the present study should be interpreted with caution when it comes to mechanistic explanations. Fortunately, it was recently shown that circulating acylcarnitine levels reflect cardiac tissue content of acylcarnitines (121), whereas elevated plasma levels of BCAAs reflect impaired BCAA catabolism in cardiac cells (122). These two facts support the rationale of the present study. Secondly, the cross-sectional nature of this study only allows for the establishment of associations, and not causality, between metabolites and CAD (123). While it is currently unknown if elevated level of circulating acylcarnitines is a consequence or a cause of ischemia-related cardiac damages, this question is of crucial relevance for the therapy of CAD and should be further investigated. Thirdly, the extent to which the reported associations represent the acylcarnitine and BCAA signature of CAD in females is unclear, as most enrolled patients were male (90.7%). This study likely did not capture the sex-specific metabolic signature of CAD. Fourthly, although serum samples were collected in a fasting state and regression analyses were adjusted for fasting time, the fasting duration might have been too short (6.7 h ± 3.0 h for healthy controls and 8.5 h ± 5.3 h for CAD patients). However, while circulating acylcarnitine levels have been shown to decrease up to 3 h after food intake (124) and increase after 12 h of fasting (125–127), the effect of a fasting time comprised between 3 and 12 h is unknown to the authors' best knowledge. Finally, we did not control for the amount and type of food intake, which is known to influence the metabolome (67).