Mora‑Ortiz et al.

BMC Medicine (2022) 20:373

https://doi.org/10.1186/s12916-022-02566-z

RESEARCH ARTICLE Open Access

Metabolomics analysis of type 2

diabetes remission identifies 12 metabolites
with predictive capacity: a CORDIOPREV clinical
trial study
Marina Mora‑Ortiz1,2,3,4†, Juan F. Alcala‑Diaz1,2,3,4†, Oriol Alberto Rangel‑Zuñiga1,2,3,4,
Antonio Pablo Arenas‑de Larriva1,2,3,4, Fernando Abollo‑Jimenez3, Diego Luque‑Cordoba3,5,6,
Feliciano Priego‑Capote3,5,6, Maria M. Malagon3,7, Javier Delgado‑Lista1,2,3,4, Jose M. Ordovas8,9,10,
Pablo Perez‑Martinez1,2,3,4, Antonio Camargo1,2,3,4*† and Jose Lopez‑Miranda1,2,3,4*†   

Abstract
Background: Type 2 diabetes mellitus (T2DM) is one of the most widely spread diseases, affecting around 90% of
the patients with diabetes. Metabolomics has proven useful in diabetes research discovering new biomarkers to assist
in therapeutical studies and elucidating pathways of interest. However, this technique has not yet been applied to a
cohort of patients that have remitted from T2DM.
Methods: All patients with a newly diagnosed T2DM at baseline (n = 190) were included. An untargeted metabo‑
lomics approach was employed to identify metabolic differences between individuals who remitted (RE), and those
who did not (non-RE) from T2DM, during a 5-year study of dietary intervention. The biostatistical pipeline consisted of
an orthogonal projection on the latent structure discriminant analysis (O-PLS DA), a generalized linear model (GLM), a
receiver operating characteristic (ROC), a DeLong test, a Cox regression, and pathway analyses.
Results: The model identified a significant increase in 12 metabolites in the non-RE group compared to the RE group.
Cox proportional hazard models, calculated using these 12 metabolites, showed that patients in the high-score tercile
had significantly (p-value < 0.001) higher remission probabilities (Hazard Ratio, HR, high versus low = 2.70) than those in the
lowest tercile. The predictive power of these metabolites was further studied using GLMs and ROCs. The area under
the curve (AUC) of the clinical variables alone is 0.61, but this increases up to 0.72 if the 12 metabolites are considered.
A DeLong test shows that this difference is statistically significant (p-value = 0.01).
Conclusions: Our study identified 12 endogenous metabolites with the potential to predict T2DM remission fol‑
lowing a dietary intervention. These metabolites, combined with clinical variables, can be used to provide, in clinical
practice, a more precise therapy.

†
Marina Mora-Ortiz and Juan F. Alcala-Diaz contributed equally to this work.
Antonio Camargo and Jose Lopez-Miranda contributed equally to this work.
*Correspondence: [email protected]; [email protected]
1
Lipids and Atherosclerosis Unit, Internal Medicine Unit, Reina Sofia University
Hospital, 14004 Cordoba, Spain
Full list of author information is available at the end of the article

© The Author(s) 2022. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. The Creative Commons Public Domain Dedication waiver (http://​creat​iveco​
mmons.​org/​publi​cdoma​in/​zero/1.​0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 2 of 14

Trial registration: ClinicalTrials.gov, NCT00924937.

Keywords: Diabetes, Insulin resistance, Prospective human study, Metabolomics

Background and identify biomarkers of interest to assist in the predic-

T2DM is a metabolic disorder widely identified by a gen- tion of T2DM recovery.
eralized hyperglycaemia and insulin resistance [1]. Now-
adays, approximately 422 million adults are diagnosed Study design and participants
with diabetes [2]; T2DM is the most prevalent form of the This study was developed within the framework of the
disease, affecting 90% of the patients (circa 380 million CORDIOPREV (CORonary Diet Intervention with Olive
worldwide). Common co-morbidities associated with oil and cardiovascular PREVention) study, registered at
T2DM include cardiovascular diseases, blindness, nerve Clinicaltrials.gov (number NCT00924937). This study
damage, and kidney failure [3–5]. The co-occurrence of is an ongoing controlled, single-blind, and randomized
coronary heart disease (CHD) along with T2DM mark- trial, with 1002 CHD patients. The trial protocol and
edly increases the risk of macrovascular complications subsequent revisions were approved by the Reina Sofia
and mortality. Indeed, macrovascular events represent University Hospital Ethics Committee, following the Hel-
circa 80% of all deaths in these patients [1]. This current sinki Declaration and good clinical practices. All patients
scenario urges us to find new approaches to diagnose signed a written informed consent to participate in the
and treat these patients. Metabolic profiling, or metabo- study.
lomics, allows for the characterization of hundreds of Patients’ recruitment took place between November
compounds (i.e. metabolites) facilitating functional infor- 2009 and February 2012, mostly at Reina Sofia Univer-
mation of the metabolism at the time when the sample sity Hospital, Córdoba, Andalusia, Spain, with contribu-
is taken [6–8] and has proven useful to understand the tions from other hospitals in Córdoba and Jaen areas, in
metabolism in the context of (i) diagnostic and prognos- Andalusia, Spain. Complete details of the study meth-
tic biomarker discovery, (ii) therapy research, and (iii) ods, rationale, inclusion criteria, cardiovascular risk fac-
pathways determination within T2DM [8]. tors, and baseline characteristics are found elsewhere
Recent evidence has shown that T2DM is reversible [13]. In brief, eligible participants, in the age range of 20
when tackled in an early phase following two different to 75 years, had established CHD with no clinical events
strategies: calorie restriction and bariatric surgery. The in the previous 6 months. They all had at least a 5-year
former is linked to weight loss, gut permeability, and life expectancy and no other concurrent major diseases
reduction in inflammatory and endotoxemia biomarkers and were willing to participate in a long-term monitoring
[9]. The latter leads to normalizing plasma glucose levels study [13].
and a significant weight loss [10–12]. In this work, 183 patients, from the CORDIOPREV
Here, we analyse the metabolomics modulations dif- study (https://​www.​cordi​oprev.​es/​index.​php/​es/) diag-
ferences between individuals where T2DM has remitted nosed with diabetes, underwent a dietary interventional
(RE) and those who did not recover and consequently study where participants were offered either an LF or
remained as diabetics (non-RE) after 5 years of dietary MED diet for 5 years (Fig. 1). In our study, blood sam-
intervention. The dietary intervention consisted of two ples were taken during fasting (time 0) and 120 min after
different types of diets: low-fat diet (LF) and Mediterra- a glucose boost.
nean diet (MED).
Oral glucose tolerance test
The patients underwent an OGTT at the baseline and,
Methods once a year, every year during the dietary intervention.
Aim and objectives Before the test, patients had fasted (from food/drugs)
The aim is to characterize the metabolic profile of non- for 12 h and were asked to refrain from smoking and
RE and RE patients in serum samples at the baseline to alcohol intake during the preceding 7 days. They were
identify biomarkers of interest to assist in the diagnosis, also asked to avoid strenuous physical activity, a day
monitoring, and treatment of the disease. These biomark- before the test. At 8:00 A.M., patients were admitted
ers can be particularly useful to predict who will recover to the laboratory to perform the oral glucose toler-
from T2DM following a dietary intervention. ance test (OGTT) (75 g flavoured glucose load, Trutol
The specific objectives of our study are to analyse the 75; Custom Laboratories, Baltimore, MD, USA). Blood
metabolomics differences before the dietary intervention samples were taken at times corresponding to 0, 30, 60,
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 3 of 14

Fig. 1 CORDIOPREV study design

90, and 120 min to determine the glucose and insulin Randomization and masking
concentrations [14]. The insulin sensitivity index (ISI) The process of randomization has been reported else-
was calculated from the OGTT using the following for- where [13]. Briefly, this is based on the following vari-
mula: ISI = 10.000 ÷ √([fasting plasma glucose X fast- ables: sex (male, female), age (under and over 60 years
ing plasma insulin] × [mean glucose in OGTT × mean old), and previous myocardial infarction (yes, no). Eight
insulin in OGTT]) [14]. HOMA-IR was calculated as different groups were created to represent all the possi-
described by Song et al. [15]. Beta-cell function was ble combinations of the above factors. Therefore, eight
calculated using the disposition index (DI) as follows: different blocks were created to assign the diets (bloc
DI = ISI × [AUC30 min insulin/AUC30 min glucose], randomization). Dietitians were the only members of
where AUC30 min is the area under the curve between the intervention team to be aware of the dietary group of
baseline and that at 30 min of the OGTT for insulin each participant.
(pmol/l) and glucose (mmol/l) measurements, calcu-
lated by the trapezoidal method [16]. The indices used Dietary assessment
to determine tissue-specific insulin resistance (IR) The participants were randomized to consume two diets:
were the hepatic insulin resistance index (HIRI, fasting the Med diet or an LF diet [13]. The LF diet consists
plasma insulin × fasting plasma glucose) and the muscle of < 30% total fat (< 10% saturated fat, 12–14% MUFA
insulin sensitivity index (MISI, (dG/dt)/mean of plasma fat, and 6–8% PUFA fat), 15% protein, and a minimum of
insulin) [17]. Insulinogenic Index (IGI) was calculated 55% carbohydrates. The Med diet consists of a minimum
by measuring plasma insulin at 30 min − fasting plasma of 35% of calories as fat (22% MUFA fat, 6% PUFA fat,
insulin (mU/L)/(plasma glucose at 30 min − fasting and < 10% saturated fat), 15% proteins, and a maximum
plasma glucose(mg/dL) [18]. of 50% carbohydrates [20]. Neither energy restriction,
The adipose tissue (AT) insulin resistance index (Adipo-IR) nor physical activity was specifically encouraged. In both
was determined according to the formula: Adipo-IR = fast- diets, the cholesterol content was adjusted to < 300 mg/d.
ing plasma NEFA (mM) × fasting plasma insulin (pmol/L), The Mediterranean and low-fat diets were designed
which has been found as a suitable and useful method in to provide a wide variety of foods, including vegeta-
clinical practice to estimate AT insulin sensitivity [19]. bles, fruit, cereals, potatoes, legumes, dairy products,
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 4 of 14

meat, and fish. The participants in both intervention 6 min. Five minutes post-run was included to equilibrate
groups received the same intensive dietary counselling. the column. The flow rate was maintained at 0.4 mL/min.
The nutritionists administered personalized individual The injected sample volume was 5.0 μL and the injector
advice every 6 months. In addition, quarterly group edu- needle was washed 10 times with 70% methanol between
cation sessions were held with up to 20 participants per injections. Therefore, the needle seat was flushed for 15 s
session; separate group sessions were performed every at a flow rate of 4 mL/min, with 70% methanol, to avoid
3 months, and dietary counselling by phone was done cross-contamination between samples. The autosampler
every 2 months [20]. At the beginning of the study, and was kept at 4 °C to increase sample stability. The settings
every 6 months afterwards, each patient had a face-to- of the electrospray ionization source, which was oper-
face interview with a nutritionist to complete a 137-item ated in negative and positive ionization modes, were as
semi-quantitative food frequency questionnaire (vali- follows: capillary voltage ± 3.5 kV, Q1 voltage 130 V, N2
dated in Spain [21]). The dietary evaluation was calcu- pressure in the nebulizer 35 psi; N2 flow rate and temper-
lated by the 14-item Med Diet Adherence Screener, ature as drying gas 10 L min–1 and 325 °C, respectively.
which was used for measuring adherence to the Med diet MS/MS data were acquired in both polarities, using the
[22]. Moreover, a 9-item dietary adherence screener was centroid mode at a rate of 2.5 spectra s–1 in extended
used to measure adherence to the LF diet guidelines. A dynamic range mode (2 GHz). Accurate mass spectra in
more detailed report on dietary adherence has been pub- the MS scan were acquired in the m/z range 40–1100 and
lished recently by our research group [20]. the MS/MS mode in the m/z range 30–1100. The instru-
ment gave a typical resolution of 18,000 full width at half
Diabetes remission criteria maximum (FWHM) at m/z 118.0862 and 35,000 FWHM
Remission required the following: (i) the absence of at m/z 922.0098. The instrument was calibrated and
glucose-lowering treatment and was defined by levels tuned as recommended by the manufacturer. To assure
of HbA1c < 6·5%, (ii) a fasting plasma glucose < 126 mg/ the desired mass resolution, continuous internal calibra-
dl, and (iii) a 2-h plasma glucose in the 75 g tion was performed during analyses by using the signals
OGTT < 200 mg/dl maintained for at least 2 years. This at m/z 121.0509 (protonated purine) and m/z 922.0098
agrees with the American Diabetes Association (ADA) [protonated hexakis(1H,1H,3H-tetrafluoropropoxy) phos-
diagnosis criteria [23]. phazine or HP-921] in the positive ion mode, while in
the negative ion mode, ions with m/z 119.0362 (proton
Sample preparation abstracted purine) and m/z 966.0007 (formate adduct of
Plasma samples (100 μL) were immersed in bath ice HP-921) were used. The collision energy was set at 20 V
and treated with 300 μL of 3:1 (v/v) methanol–acetoni- for the whole run. The analytical samples were injected in
trile (MeOH–ACN). The treated samples were vortexed auto MS/MS acquisition mode to obtain fragmentation
for 2 min and subsequently cooled at − 20 °C for 3 min. information from a maximum of two precursors selected
Centrifugation was carried out for 15 min at 4 °C and per cycle with an exclusion window of 0.1 min after 2 con-
13,800 × g in a thermostatic centrifuge Thermo Sorvall secutive selections of the same precursor.
Legend Micro 21 R from Thermo (Thermo Fisher Sci-
entific, Bremen, Germany), and the supernatant phase Data processing
was isolated. This phase was dried by evaporation and The MassHunter Workstation software (version B7.00
reconstituted with 60 μL of 3:1 (v/v) MeOH–ACN. All Qualitative Analysis, Agilent Technologies, Santa Clara,
samples were processed in a 1200 Series LC system (Agi- CA, USA) was used to process all the data obtained by
lent Technologies, Waldbronn, Germany) coupled to an LC–QTOF in data-dependent acquisition MS/MS mode.
Agilent 6530 high-resolution QTOF mass spectrometer Treatment of raw data files started with the extraction of
equipped with a dual electrospray ionization source. potential molecular features (MFs) with the suited algo-
rithm included in the software. For this aim, the extrac-
LC–QTOF MS/MS analysis tion algorithm considered all ions exceeding 500 counts
A Poroshell 120 EC-C18 column (50 mm × 2.1 mm i.d., for both polarities with a maximum charge state of 2 for
2.7 μm particle size, from Agilent), kept at 25 °C, was used the obtained chromatograms. The count cut-off value
to carry out the chromatographic division. The mobile was established considering the chromatographic back-
phases consisted of (A) 0.1% formic acid in deionized ground noise. Additionally, only MFs defined by two or
water and (B) 0.1% formic acid in acetonitrile. The pro- more ions were considered, with a tolerance for the iso-
tocol used for the elution consisted of 0–2 min, 5% B; topic distribution of 0.0025 m/z for peak spacing toler-
2–11 min and the percentage of mobile phase B was mod- ance, plus 7.0 ppm in mass accuracy. Only the following
ified from 0 to 100%. The final percentage was held for potential ions and adducts were considered in positive
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 5 of 14

(H + , Na + , K + , NH4 +) and negative ionization (H − , (2021–03-31, https://​www.r-​proje​ct.​org) using the pack-

HCOO + , Cl +) modes. Furthermore, a potential neutral ages ‘caret’ and ‘pROC’. Unadjusted Cox proportional-
loss by dehydration was also included to identify features hazard models calculated the hazard ratio (HR) of every
corresponding to the same potential metabolite. metabolite previously identified in the O-PLS DA model
Identification of metabolites was supported on MS and within a 95% confidence interval (CI). This unadjusted
MS/MS information by using METLIN MS and MS/MS Cox allowed the identification of the betas for every
databases (http://​metlin.​scrip​ps.​edu), the Human Metab- metabolite. This information was used to calculate the
olome Database (HMDB, 3.6 version), and the LIPID patient’s likelihood of recovering from diabetes by run-
MAPS website ((http://​www.​lipid​maps.​org); in all cases, ning a Cox analysis adjusted for sex, age, body mass index
the MFs obtained in the previous step were used. A data- (BMI), HDL, triglycerides, and intensity of statin therapy
base with all identified metabolites was used to perform a based on the tertiles resulting from the multiplication of
targeted compound extraction analysis using a tolerance the betas previously obtained by the abundance of each
window of 0.8 min and 5 ppm mass accuracy. This step metabolite for every patient. Finally, generalized linear
was performed with Profinder Analysis (version B8.00, models were calculated for (i) all the clinical variables
Agilent Technologies, Santa Clara, CA, USA). A table (i.e. sex, age, BMI, HDL, triglycerides, and intensity statin
with the peak area of all identified compounds in the dif- therapy), (ii) all the metabolites, (iii) the glycated haemo-
ferent samples injected was obtained as a result. globin, (iv) the clinical variables and the 12 metabolites,
and (v) the clinical variables and the glycated haemo-
Statistical analysis
globin. ROC analyses were carried out for these three
Metabolites showing in at least 80% of the samples were models, and AUC, sensitivity, specificity, accuracy, and
selected for further analysis. To allow predictive model- threshold were estimated for the models. Finally, DeLong
ling, imputation was carried out when needed, substi- analysis was used to compare whether the AUCs of these
tuting missing values by half the smallest value of the models were or not significantly different between them.
appropriate metabolite.
LC–MS data (polar and apolar) was imported into Results
Matlab (R2015a, Mathworks UK) and analysed using Baseline characteristic
the statistic toolbox and algorithms from Korrigan Tool- BMI, waist circumference, body weight, glucose, glycated
box version 0.1 (Korrigan Sciences Ltd, UK). In Matlab, haemoglobin (HbA1c), insulin, HIRI, and homeostatic
matrices were log 10 normalized. The biostatistical pipe- model assessment of insulin resistance (HOMA-IR) were
line for the multivariate statistical analysis considered a statistically significantly higher at baseline in the non-RE
preliminary unsupervised principal component analysis group than in the RE group. Conversely, ISI and DI val-
(PCA), followed by a supervised pairwise O-PLS DA [24, ues were statistically significantly lower at baseline in the
25], which identifies the specific modulations driven by RE group than in the non-RE group (p < 0.05) (Additional
the appropriate predictor (i.e. individuals who returned file 1: Table S1).
from T2DM versus those who did not).
To assess the predictive power of the O-PLS DA mod- O‑PLS DA results from the comparison between individuals
els, R2 (the explained variance) was calculated. This who remitted from T2DM and those who did not
parameter evaluates the model maximizing variance The O-PLS DA was examined based on R2Y, Q2Y, and
given by the endogenous variables. The Q2, or goodness overfit parameters obtained in every pairwise compari-
of prediction, assesses the predictive relevance of the son (see material and methods). The O-PLS DA analysis
model and is based on a matrix partition technique that identified differences between RE and DM individuals
ignores part of the data (in our case a seventh part each during fasting (R2Y = 0.1420, Q2Y = 0.0018), but not after
time), estimates the model parameters, and predicts the a glucose overdose (Fig. 2). DM individuals had higher
omitted parts using the estimates obtained previously. levels of sphingosine (d18:2), docosenamide, oxo-tricosa-
Q2 greater than 0 means the model has predictive value. noic acid, tetracosahexaenoic acid, ketodeoxycholic acid,
In addition, the overfitting of the model (the difference stearoylcarnitine, diglyceride (33:4), creatine, tridecanoic
between R2Y and Q2Y) was also considered, and only acid, monoacylglycerol (22:6), dihydroxycholesterol, and
models with less than 50% overfitting were further con- biliverdin. These metabolites, as well as a ranking with
sidered. Model parameters and associated metabolites the degree of statistically significant association of each
were reported and used for a Cox proportional hazard metabolite, are presented in further detail in Additional
model, GLM, and ROC calculations in R (version 4.0.5 file 2: Fig. S1.
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 6 of 14

Fig. 2 O-PLS DA analysis loading and score plot calculated using all spectra as a matrix of independent variables and diabetic remission as the
predictor

Cox proportional hazard models and the 12 metabolites 0.721, and for the combination of
Adjusted and unadjusted Cox proportional hazard mod- the clinical variables and the glycated haemoglobin 0.667.
els were calculated for every metabolite previously iden- A DeLong test was carried out comparing the AUC from
tified in the O-PLS DA model. These metabolites were the GLM of the clinical variables alone with the GLM
grouped in terciles for the Cox analysis. The results are with the clinical variables and the 12 metabolites or the
shown in Table 1. The unadjusted beta values were used GLM with the clinical variables and the glycated hae-
to calculate the patient risk score (beta × metabolite moglobin (Fig. 4). There were not significant differences
abundance) for all the patients and metabolites. This between the model with the clinical variables alone and
patient risk score grouped by terciles was used to carry the addition of the glycated haemoglobin. However, the
out a new Cox model adjusted for sex, age, BMI, HDL, p-value (0.01265) resulting from the DeLong test com-
TG, and intensity statin therapy (see Fig. 3). Results from paring the model with the clinical variables alone and
the Cox analysis of the patient’s risk score showed that the clinical variables with the 12 metabolites indicated
individuals in the high tercile (luckily to recover from that these two models were significantly different and
T2DM) had an HR of 2.70 compared to individuals in showed that the addition of the metabolites significantly
the low tercile; this difference was statistically significant improved the prediction capacity of the model.
(p-value = 0.003). Moreover, we also grouped the patients
by ascending terciles of glycated haemoglobin, to carry Discussion
out a Cox model adjusted for sex, age, BMI, HDL, TG, Our study identified 12 plasma metabolites by O-PLS
and intensity statin therapy (Additional file 2: Fig. S2). DA differing between RE and non-RE patients at the
Results from the Cox analysis of the glycated haemoglo- baseline of the study; these metabolites were further
bin showed that individuals in the low and medium ter- used to build a score to assess regression probability.
cile had an HR of 3.80 and 4.31 respectively, compared to This score was significantly associated with a higher
individuals in the High tercile (both p-value = 0.001). probability of T2DM remission. These 12 metabolites,
together with the clinical variables previously described,
Generalized linear models and receiving operating significantly improved the T2DM remission prediction
characteristics power of the model. However, the prediction capacity of
GLMs and ROCs were run for (i) clinical variables alone, the model with the clinical variables did not significantly
(ii) the 12 metabolites of interest, (iii) glycated haemoglo- improve when the glycated haemoglobin was added. In
bin, (iv) the clinical variables and the 12 metabolites, and practice, this could be achieved by analysing a plasma
(v) the clinical variables and the glycated haemoglobin. sample from the patients and determining the concen-
AUC, sensitivity, specificity, accuracy, and threshold were trations of these molecules.
calculated in all cases. The results are shown in Tables 2 Lifestyle modifications, including the implementation
and 3 and Fig. 4. The AUC for the clinical variables was of healthy diets, result in a beneficial effect on T2DM
0.610, for the metabolites was 0.701, for the glycated hae- prevention and management [26]. Recent studies indi-
moglobin 0.618, for the combination of clinical variables cated that it was possible to induce T2DM remission by
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 7 of 14

Table 1 Cox proportional hazard models were calculated for every metabolite. The hazard ratios (HR) between groups were
calculated with Tertile 3 as a reference and adjusted by age, gender, diet, body mass index, HDL-c, and triglycerides
95.0% CI for HR
Metabolite Tertile Model type Sig HR Lower Upper

Sphingosine (d18:2) Tertile 3 (ref.)

Tertile 2 Unadjusted .283 1.414 .751 2.664
Adjusted .324 1.389 .723 2.667
Tertile 1 Unadjusted .013 2.111 1.171 3.806
Adjusted .018 2.059 1.129 3.753
Docosenamide Tertile 3 (ref.)
Tertile 2 Unadjusted .704 1.122 .621 2.027
Adjusted .670 1.141 .622 2.094
Tertile 1 Unadjusted .320 1.336 .755 2.364
Adjusted .404 1.278 .718 2.275
Oxo-tricosanoic acid Tertile 3 (ref.)
Tertile 2 Unadjusted .522 1.222 .661 2.257
Adjusted .458 1.265 .680 2.352
Tertile 1 Unadjusted .046 1.794 1.009 3.189
Adjusted .044 1.832 1.017 3.300
Tetracosahexaenoic acid Tertile 3 (ref.)
Tertile 2 Unadjusted .008 2.440 1.258 4.732
Adjusted .004 2.758 1.384 5.496
Tertile 1 Unadjusted .002 2.829 1.479 5.413
Adjusted .001 3.155 1.571 6.336
Ketodeoxycholic acid Tertile 3 (ref.)
Tertile 2 Unadjusted .612 .850 .453 1.595
Adjusted .677 .871 .456 1.665
Tertile 1 Unadjusted .065 1.682 .969 2.920
Adjusted .065 1.717 .967 3.046
Stearoylcarnitine Tertile 3 (ref.)
Tertile 2 Unadjusted .039 1.921 1.034 3.568
Adjusted .050 1.890 1.001 3.570
Tertile 1 Unadjusted .039 1.911 1.033 3.533
Adjusted .063 1.820 .967 3.426
Diglyceride (33:4) Tertile 3 (ref.)
Tertile 2 Unadjusted .569 1.207 .632 2.304
Adjusted .464 1.279 .661 2.474
Tertile 1 Unadjusted .004 2.380 1.328 4.266
Adjusted .004 2.423 1.321 4.444
Creatine Tertile 3 (ref.)
Tertile 2 Unadjusted .115 1.635 .887 3.014
Adjusted .123 1.632 .876 3.042
Tertile 1 Unadjusted .061 1.778 .973 3.249
Adjusted .093 1.703 .916 3.166
Tridecanoic acid Tertile 3 (ref.)
Tertile 2 Unadjusted .327 1.372 .729 2.585
Adjusted .394 1.323 .695 2.518
Tertile 1 Unadjusted .013 2.101 1.166 3.786
Adjusted .033 1.946 1.053 3.594
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 8 of 14

Table 1 (continued)
95.0% CI for HR
Metabolite Tertile Model type Sig HR Lower Upper

Monoacylglycerol (22:6) Tertile 3 (ref.)

Tertile 2 Unadjusted .315 1.358 .748 2.466
Adjusted .406 1.311 .692 2.485
Tertile 1 Unadjusted .131 1.571 .873 2.827
Adjusted .109 1.634 .896 2.979
Dihydroxycholesterol Tertile 3 (ref.)
Tertile 2 Unadjusted .388 .744 .381 1.454
Adjusted .373 .736 .375 1.445
Tertile 1 Unadjusted .008 2.099 1.214 3.631
Adjusted .019 1.987 1.119 3.529
Biliverdin Tertile 3 (ref.)
Tertile 2 Unadjusted .842 1.064 .577 1.964
Adjusted .961 .984 .517 1.872
Tertile 1 Unadjusted .116 1.574 .894 2.772
Adjusted .168 1.497 .843 2.657

weight loss with calorie restriction interventions [27]. induce the remission of T2DM, with the main differ-
Previous studies showed that bariatric surgeries can ence between the diets in the amount and type of dietary
revert T2DM regulating the glucose in plasma before a sources of fat and the amount of carbohydrates [20]. Both
significant weight reduction is obtained [9–11]. diets were ethically appropriate for this profile of patients
The importance of the early identification of diabetic and no further energy restriction or physical activity was
patients with a probability of achieving T2DM remission implemented.
lies in the ability of the β cells to recover long-term func- This profile was characterized by a reduction in 12
tionality after T2DM diagnosis, and before an irreversible metabolites across RE patients identified by O-PLS DA.
stage of β cell dysfunction [9]. Moreover, an effective and This model also showed which metabolites found have
efficient therapeutic action focused on disease remission the strongest influence. These metabolites ranked with a
is especially important in individuals with co-occurring different degree of significance in the model as shown in
acute myocardial infarction and T2DM, who have a Additional file 2: Fig. S1. These metabolites, according to
higher risk of developing a new cardiovascular event than the results from the DeLong test, significantly improved
those without T2DM [28]. the predictive capacity of the clinical variables alone.
In terms of predicting T2DM remission, a variety of These metabolites are included in several insulin-related
scores and variables were previously used to identify pathways such as sphingolipid metabolism or alpha-lino-
subjects with a probability of remission. Classical clini- lenic acid and linoleic acid metabolism. Moreover, taken
cal variables have shown a reduced prediction capacity together, these metabolites may be linked to the lipid
[29, 30]. Conversely, we have previously reported how alterations associated with T2DM. In fact, the derange-
the use of miRNAs or gut microbiota proxies improve the ment of the lipid metabolism is a common complica-
predictive power of the clinical variables alone, improv- tion in T2DM due to the inadequate functioning of key
ing the estimation of the T2DM remission probability enzymes and pathways as well as the insulin resistance
induced by the dietary intervention [31–33]. However, to prevalent in these patients [34]. In addition, the associa-
the best of our knowledge, the plasma metabolite profile tion between dyslipidaemia and atherosclerosis is well
has not been applied to such end. established, and the composition of lipid particles in dia-
This study has shown a metabolic profile associated betic dyslipidaemia has a stronger atherogenic impact on
with T2DM remission in CHD patients. Here, TOF/ the disease compared to other kinds of dyslipidaemia [35,
LS-MS metabolomics at baseline was used to iden- 36]. Thus, it would be also expected the relationship of
tify which newly diagnosed T2DM patients will benefit these metabolites with lipid alterations considering that
from a dietary intervention (a Med or a low-fat diet) to the individuals included in this study were CHD patients.
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 9 of 14

Fig. 3 Adjusted Cox for the analysis of the patient risk scored grouped in tertiles. a Survival probability chart overtime (expressed in months). b
Hazard ratio of the risk score for the three tertiles and the covariables

Moreover, these disruptions in the metabolite profile In line with this, it is worth mentioning that THA and
in non-RE patients seem to be additional to the dys- tridecanoic acid are rare fatty acids. Current literature is
lipidaemia associated with CHD, which includes hyper- contradictory about the beneficial or detrimental effects
triglyceridemia, hypercholesterolemia, and elevated of rare fatty acids in CHD or T2DM patients [38–40]. In
LDL cholesterol [37]. This present study brings insights our study, higher levels of THA and tridecanoic acid were
into the metabolomics modulations occurring during observed in non-RE patients. This could be because our
dyslipidaemia. population comprises patients with established CHD in
The O-PLS DA model showed that out of the 12 which may exist certain alterations in their lipid species
metabolites identified, tetracosahexaenoic acid (THA), profile patients due to the cardiovascular disease.
oxo-tricosanoic acid, dihydrocholesterol, and tridecanoic Dihydroxycholesterol is a metabolite involved in the
acid were the most discriminant between RE and non- primary bile acid biosynthesis pathway and derived
RE; hence, high levels were observed in non-RE patients. from cholesterol. This metabolite was identified as an
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 10 of 14

Table 2 Results were obtained for the generalized linear models used with the (i) clinical variables, (ii) metabolites, and (iii) clinical
variables and metabolites. * means statistically significant
Estimate Std. error z value Pr( >|z|)

Clinical variables Intercept − 0.43249 0.18301 − 2.363 0.0181*

Sex − 0.08264 0.16283 − 0.508 0.6118
Age baseline 0.09660 0.16102 0.600 0.5486
HDL Imputed 0.10651 0.16663 0.639 0.5227
TG Imputed − 0.27225 0.17275 − 1.576 0.1150
BMI baseline 0.44999 131.228 0.343 0.7317
Intensity statin therapy − 0.11911 0.15505 − 0.768 0.4424
Metabolites Intercept − 0.459189 0.704513 − 0.652 0.5145
Sphingosine (d18:2) − 0.107924 0.260230 − 0.415 0.6783
Docosenamide 0.004636 0.004572 1.014 0.3106
Oxo-tricosanoic acid − 0.002283 0.003195 − 0.714 0.4750
Tetracosahexaenoic acid − 0.003083 0.003112 − 0.991 0.3219
Ketodeoxycholic acid − 0.293914 0.290994 − 1.010 0.3125
Stearoylcarnitine − 0.436030 0.230849 − 1.889 0.0589
Diglyceride (33:4) 0.194136 0.198864 0.976 0.3290
Creatine − 0.001211 0.003222 − 0.376 0.7071
Tridecanoic acid − 0.001950 0.003332 − 0.585 0.5584
Monoacylglycerol (22:6) 0.003391 0.003096 1.095 0.2733
Dihydroxycholesterol − 0.327887 0.177616 − 1.846 0.0649
Biliverdin − 0.362871 0.179566 − 2.021 0.0433*
Clinical variables and Intercept − 0.330027 0.726670 − 0.454 0.6497
metabolites Sphingosine (d18:2) − 0.042235 0.271197 − 0.156 0.8762
Docosenamide 0.004359 0.004724 0.923 0.3561
Oxo-tricosanoic acid − 0.002246 0.003257 − 0.689 0.4905
Tetracosahexaenoic acid − 0.004089 0.003290 − 1.243 0.2139
Ketodeoxycholic acid − 0.321368 0.302763 − 1.061 0.2885
Stearoylcarnitine − 0.461926 0.236503 − 1.953 0.0508
Diglyceride (33:4) 0.184457 0.208359 0.885 0.3760
Creatine − 0.001613 0.003291 − 0.490 0.6241
Tridecanoic acid − 0.001681 0.003458 − 0.486 0.6268
Monoacylglycerol (22:6) 0.003439 0.003158 1.089 0.2761
Dihydroxycholesterol − 0.307896 0.182783 − 1.684 0.0921
Biliverdin − 0.348474 0.181466 − 1.920 0.0548
Sex − 0.091159 0.178688 − 0.510 0.6099
Age baseline 0.142588 0.177915 0.801 0.4229
HDL Imputed 0.053527 0.188053 0.285 0.7759
TG Imputed − 0.223763 0.185640 − 1.205 0.2281
BMI baseline 0.402986 1.322.298 0.305 0.7605
Intensity Statin Therapy − 0.214802 0.174998 − 1.227 0.2197

intermediate in C21-Steroid hormone metabolism [41]. those individuals who did not return from T2DM with
Previous studies have shown that plasma concentrations the dietary intervention, which suggests that the lipid
of cholesterol oxidation (ChOx) products (including dif- metabolism of those individuals is more impaired in non-
ferent forms of hydroxycholesterol) are elevated in DM1 RE than in RE patients. The relationship between cho-
and DM2 patients compared to age-matched subjects lesterol oxidized metabolites and lipid metabolism lies
without diabetes [42]. In our study, significantly higher in the fact that high levels of glucose seem to promote
levels of dihydroxycholesterol were identified across lipidic accumulation via the DNA CpG methylation in the
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 11 of 14

Table 3 Results obtained for the general lineal models (GLM) DNA methyltransferase-1 (DNMT1) promoter region,
carried out using the clinical variables alone (GLM1), the where cholesterol oxidized metabolites act as mediators
metabolites alone (GLM2), the glycated hemoglobin as reference [43]. The high levels of glucose increase a nuclear form
(GLM3), or the combination of the clinical variables and the of oxidized cholesterol (25-hydroxycholesterol) which
metabolites (GLM4), and the combination of the clinical variables activates DNMT1 regulating the expression of several
and the glycated hemoglobin (GLM5)
genes implicated in the intracellular lipid metabolism in
Model Sensitivity Specificity Accuracy Threshold the liver. This process results in the hypermethylation
of genes directly implicated in carbohydrate and lipid
GLM 1 0.5633 0.6454 0.6132 0.4107
metabolism of PI3K, cAMP, insulin, insulin secretion,
GLM 2 0.7887 0.5818 0.6630 0.3642
and diabetic and non-alcoholic fatty liver disease signal-
GLM 3 0.6761 0.7636 0.7293 0.4429
ling pathways. This suggests that the high hydroxycho-
GLM 4 0.8028 0.5545 0.6519 0.3173
lesterol plasma levels found in our study, in the non-RE
GLM 5 0.6197 0.6909 0.6630 0.4223
group, might be altering the expression of these genes
towards a deleterious gene expression profile. This epige-
netic regulation of hepatic cell metabolism seems to also
have relevance in non-alcoholic, fatty liver disease, and
metabolic syndrome [44, 45].

Fig. 4 ROCs were obtained for the clinical variables and in combination with the metabolites or glycated haemoglobin (HbA1c)
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 12 of 14

Hepatic lipogenesis can be suppressed by downregu- taking into account that the baseline characteristics
lating the gene SREBP1c (trigged by n3 supplemen- in the groups of patients analysed in the current study
tation), implicated in fatty acid biosynthesis. These according to the diet were similar.
inhibitory effects in hepatic lipogenesis are associated
with a reduction in plasma levels of THA [46]. In our Conclusions
study, we have observed high levels of THA amongst the This study showed the association of a specific metabolic
non-RE group, which suggests a lack of inhibitory feed- profile in plasma with T2DM remission in patients with
back in this pathway. This could explain the metabolic CHD in a dietary intervention. These metabolites, com-
resilience to respond to the dietary intervention in these bined with clinical variables, could be used to provide
patients, which could be due to the methylation of the in clinical practice more precise therapeutical advice.
promoter region DNMT1. Previous studies from Nagao This new approach will allow the possibility to discern
and colleagues (2003) showed that the depletion in THA between newly diagnosed T2DM luckily to remit from
was associated with omega 3-PUFAs supplementation. diabetes following a dietary intervention, from those
This suggests that the omega 3-PUFAs ingested in the who will need a more exhaustive treatment such us anti-
dietary intervention are effective only in patients with glycaemic drugs. That would improve the management
low THA levels at baseline. Therefore, the high levels of of these patients and represent a personalized medi-
THA amongst non-RE patients could predispose them cine approach. Moreover, our results suggest that lipid
towards a higher degree of metabolic impedance, pre- metabolism is implicated in the probability to remit
venting them to respond to the healthy diet adminis- from T2DM. The detailed pathway that allows for high
trated which contains 6–8% of total calories as PUFAs. glucose (or high dietary glucose) to produce lipids accu-
However, this hypothesis needs further evidence. mulation is not fully understood yet, but early down-
We also found elevated tridecanoic acid levels in the regulation of liver lipogenesis seems determinant for
DM group, a metabolite previously associated with hypo- the metabolism to recover from T2DM. Further inves-
glycaemia, fatty liver, and cardiomyopathies, that are tigations should interrogate the potential methylation of
implicated in functions such as oxidation, cell death, and the promoter region DNMT1, to unravel whether epige-
insulin resistance [47]. The latter is especially important netic changes may influence the capacity to return from
as high tridecanoic aid levels were found in the non-RE T2DM.
group, which included the patients who did not reduce
their insulin resistance after the ingestion of the healthy
Abbreviations
diet intervention. ADA: American Diabetes Association; Adipo-IR: Adipose tissue insulin resist‑
Predicting which patients with CHD can recover from ance index; AT: Adipose tissue; AUC​: Area under the curve; BMI: Body mass
T2DM is crucial since patients with co-occurring CHD index; CHD: Coronary heart disease; CI: Confidence interval; DI: Disposition
index; GLM: Generalized linear model; HIRI: Hepatic insulin resistance index;
and T2DM have a considerably higher risk of developing HMDB: Human Metabolome Database; HOMA-IR: Homeostatic model assess‑
a new cardiovascular event than those without T2DM. ment of insulin resistance; HR: Hazard ratio; IR: Insulin resistance; ISI: Insulin
Furthermore, some of the metabolites might be linked sensitivity index; LF: Low-fat diet; MED: Mediterranean diet; MISI: Muscle insu‑
lin sensitivity index; non-RE: Patients who remained as diabetics; O-PLS DA:
to other physiological processes that remain unclear and Orthogonal Projection on the Latent Structure Discriminant Analysis; OGTT​
future research in this field is needed. : Oral glucose tolerance test; PCA: Principal component analysis; RE: Patients
In summary, our study provides new plasma biomark- who remitted from T2DM; ROC: Receiver operating characteristic; T2DM: Type
2 diabetes mellitus; THA: Tetracosahexaenoic acid.
ers to predict, in combination with clinical variables, the
dietary remission capacity of T2DM patients with co-
Supplementary Information
occurring CHD. Indeed, the addition of the 12 metabo-
The online version contains supplementary material available at https://​doi.​
lites identified significantly improved the prediction org/​10.​1186/​s12916-​022-​02566-z.
power of the clinical variables alone.
It is also important to mention the limitations of this Additional file 1: Table S1. Baseline characteristics of the study
study. Firstly, this research is based on a long-term, well- population.
controlled dietary intervention, which despite ensur- Additional file 2: Figures S1, S2. Figure S1. Ranking with the 12 most
significant metabolites derived from the O-PLS DA model. Figure S2.
ing the quality of the study, may not reflect the level of Adjusted Cox for the analysis of the patient risk scored grouped in ascending
compliance in a free-living population. The second limi- tertiles of glycated haemoglobin.
tation is that the remission of T2DM was not the pri-
mary endpoint of the CORDIOPREV trial, although it Acknowledgements
was a secondary objective of this study. However, there The CIBEROBN is an initiative of the Instituto de Salud Carlos III, Madrid, Spain.
are no reasons to believe that this randomization would Antonio Camargo is supported by an ISCIII research grant (Programa Miguel-
Servet CP14/00114 and CPII19/00007). The funders had no role in the design
not have worked in such a large subset of participants,
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 13 of 14

and conduct of the study; the collection, management, analysis, or interpreta‑ Cordoba, Spain. 6 CIBER de Fragilidad Y Envejecimiento Saludable (CIBERFES),
tion of the data; preparation, review, or approval of the manuscript; or the Instituto de Salud Carlos III, Madrid, Spain. 7 Department of Cell Biology, Physi‑
decision to submit the manuscript for publication. We wish to acknowledge ology and Immunology, University of Cordoba, 14004 Cordoba, Spain. 8 Nutri‑
Biobank from the ‘Sistema Sanitario Público de Andalucía’ (Córdoba, Andalusia, tion and Genomics Laboratory, J.M.‑US Department of Agriculture Human
Spain) for providing the human biological samples. We also thank EASP Nutrition Research Center On Aging at, Tufts University, Boston, MA 02111,
(Escuela Andaluza de Salud Publica), in Granada Spain, for performing the USA. 9 IMDEA Alimentacion, Madrid, Spain. 10 CNIC, 28049 Madrid, Spain.
randomization process.
Received: 16 May 2022 Accepted: 12 September 2022
Authors’ contributions
MM-O and JFA-D contributed equally to this work. AC and JL-M contributed
equally to this work. MM-O, JFA-D, and AC wrote the draft manuscript.
MM-O, OAR-Z, FA-J, and JFA-D carried out the analysis. JFA-D, APA-deL,
OAR-Z, DL-C, and FP-C collected the data and performed the classification References
of participants. MM-O, DL-C, and FP-C performed the experiments. JFA-D, 1. DeFronzo RA. Insulin resistance, lipotoxicity, type 2 diabetes and athero‑
APAdeL, JD-L, and PP-M performed the medical revisions of participants sclerosis: the missing links. The claude bernard lecture 2009. Diabetologia.
and clinical databases. MM-O and FA-J performed the statistical analysis. 2010;53(7):1270–87.
MM-O, MMM, JMO, AC, and JL-M interpreted the data and contributed to 2. WHO. Diabetes, Health Topics. http://​www.​who.​int/​diabe​tes/​en/.
the discussion. OAR-Z, FP-C, MMM, JD-L, JMO, PP-M, and JL-M contributed Accessed 7 Feb 2022.
to the writing of the manuscript and revised it critically for important intel‑ 3. Amin AP, Spertus JA, Reid KJ, Lan X, Buchanan DM, Decker C, Masoudi
lectual content. AC and JL-M are the guarantors of this work and, as such, FA. The prognostic importance of worsening renal function during
had full access to all the data in the study and takes responsibility for the an acute myocardial infarction on long-term mortality. Am Heart J.
integrity of the data and the accuracy of the data analysis. The author(s) 2010;160(6):1065–71.
read and approved the final manuscript. 4. Anavekar NS, McMurray JJ, Velazquez EJ, Solomon SD, Kober L, Rouleau
JL, White HD, Nordlander R, Maggioni A, Dickstein K, et al. Relation
Funding between renal dysfunction and cardiovascular outcomes after myocar‑
The CORDIOPREV study is supported by the Ministerio de Economia y dial infarction. N Engl J Med. 2004;351(13):1285–95.
Competitividad, Spain, under the grants AGL2012/39615, PIE14/00005, 5. Trautner C, Icks A, Haastert B, Plum F, Berger M. Incidence of blind‑
and PIE14/00031 associated to J.L.-M.; AGL2015-67896-P to J.L.-M. and A.C.; ness in relation to diabetes. A population-based study. Diabetes Care.
CP14/00114 to A.C.; PI19/00299 to A.C.; DTS19/00007 to A.C.; FIS PI13/00023 1997;20(7):1147–53.
to J.D.-L., PI16/01777 to F.P.-J. and P.P.-M.; Antonio Camargo is supported by 6. Nicholson JK, Lindon JC, Holmes E. ‘Metabonomics’: understanding the
an ISCIII research contract (Programa Miguel-Servet CPII19/00007); Marina metabolic responses of living systems to pathophysiological stimuli via
Mora-Ortiz has received funding from the European Union’s Horizon 2020 multivariate statistical analysis of biological NMR spectroscopic data.
research and innovation programme under the Marie Skłodowska-Curie grant Xenobiotica. 1999;29(11):1181–9.
agreement No 847468; ‘Fundacion Patrimonio Comunal Olivarero’, Junta de 7. Pauling L, Robinson AB, Teranishi R, Cary P. Quantitative analysis of urine
Andalucía (Consejería de Salud, Consejeria de Agricultura y Pesca, Consejería vapor and breath by gas-liquid partition chromatography. Proc Natl Acad
de Innovacion, Ciencia y Empresa), ‘Diputaciones de Jaen y Cordoba’, ‘Centro Sci U S A. 1971;68(10):2374–6.
de Excelencia en Investigación sobre Aceite de Oliva y Salud’ and ‘Ministerio 8. Suhre K. Metabolic profiling in diabetes. J Endocrinol. 2014;221(3):R75-85.
de Medio Ambiente, Medio Rural y Marino’, Gobierno de España; ‘Consejeria 9. Taylor R, Al-Mrabeh A, Zhyzhneuskaya S, Peters C, Barnes AC, Aribisala BS,
de Innovación, Ciencia y Empresa, Proyectos de Investigación de Excelencia’, Hollingsworth KG, Mathers JC, Sattar N, Lean MEJ. Remission of human
Junta de Andalucía under the grant CVI-7450 obtaiend by J.L.-M.; and we type 2 diabetes requires decrease in liver and pancreas fat content
would also like to thank the ‘Fondo Europeo de Desarrollo Regional (FEDER)’. but is dependent upon capacity for beta cell recovery. Cell Metab.
2018;28(4):547-556 e543.
Availability of data and materials 10. Clemente-Postigo M, Roca-Rodriguez Mdel M, Camargo A, Ocana-
The data that support the findings of this study are available from the cor‑ Wilhelmi L, Cardona F, Tinahones FJ. Lipopolysaccharide and lipopolysac‑
responding author upon reasonable request. charide-binding protein levels and their relationship to early metabolic
improvement after bariatric surgery. Surg Obes Relat Dis. 2015;11(4):933–9.
11. Kirwan JP, Munzberg H, Berthoud HR. Mechanisms responsible for meta‑
Declarations bolic improvements of bariatric surgeries. Diabetes. 2018;67(6):1043–4.
12. Zhang L, Xue X, Zhai R, Yang X, Li H, Zhao L, Zhang C. Timing of calorie
Ethics approval and consent to participate restriction in mice impacts host metabolic phenotype with correlative
CORDIOPREV is registered at ClinicalTrials.gov (number NCT00924937). The changes in gut microbiota. mSystems. 2019;4(6):e00348-19.
trial protocol and subsequent revisions were approved by the Reina Sofia 13. Delgado-Lista J, Perez-Martinez P, Garcia-Rios A, Alcala-Diaz JF, Perez-
University Hospital Ethics Committee (ref. number 1496/27/03/2009), follow‑ Caballero AI, Gomez-Delgado F, Fuentes F, Quintana-Navarro G, Lopez-
ing the Helsinki Declaration and good clinical practices. All patients signed a Segura F, Ortiz-Morales AM, et al. CORonary Diet Intervention with Olive
written informed consent to participate in the study. oil and cardiovascular PREVention study (the CORDIOPREV study): ration‑
ale, methods, and baseline characteristics: a clinical trial comparing the
Consent for publication efficacy of a Mediterranean diet rich in olive oil versus a low-fat diet on
Not applicable. cardiovascular disease in coronary patients. Am Heart J. 2016;177:42–50.
14. Matsuda M, DeFronzo RA. Insulin sensitivity indices obtained from oral
Competing interests glucose tolerance testing: comparison with the euglycemic insulin
The authors declare that they have no competing interests. clamp. Diabetes Care. 1999;22(9):1462–70.
15. Song Y, Manson JE, Tinker L, Howard BV, Kuller LH, Nathan L, Rifai N, Liu
Author details S. Insulin sensitivity and insulin secretion determined by homeostasis
1
Lipids and Atherosclerosis Unit, Internal Medicine Unit, Reina Sofia University model assessment and risk of diabetes in a multiethnic cohort of women:
Hospital, 14004 Cordoba, Spain. 2 Department of Medical and Surgical Sci‑ the Women’s Health Initiative Observational Study. Diabetes Care.
ence, University of Cordoba, 14004 Córdoba, Spain. 3 Maimonides Biomedical 2007;30(7):1747–52.
Research Institute of Cordoba (IMIBIC), Av. Menendez Pidal, s/n, 14004 Cor‑ 16. Tang W, Fu Q, Zhang Q, Sun M, Gao Y, Liu X, Qian L, Shan S, Yang T. The
doba, Spain. 4 CIBER Fisiopatologia de la Obesidad y Nutricion (CIBEROBN), association between serum uric acid and residual beta -cell function in
Instituto de Salud Carlos III, 28029 Madrid, Spain. 5 Department of Analytical type 2 diabetes. J Diabetes Res. 2014;2014:709691.
Chemistry and Nanochemistry University Institute, Universidad de Cordoba,
Mora‑Ortiz et al. BMC Medicine (2022) 20:373 Page 14 of 14

17. Abdul-Ghani MA, Matsuda M, Balas B, DeFronzo RA. Muscle and liver in patients with type 2 diabetes: a meta-analysis. J Cardiovasc Pharmacol.
insulin resistance indexes derived from the oral glucose tolerance test. 2020;75(4):351–7.
Diabetes Care. 2007;30(1):89–94. 36. Sun Y, Yan D, Cui L, Li G, Sun Y, Hao Z. Sex-disparities in risk factors and
18. Hanson RL, Pratley RE, Bogardus C, Narayan KM, Roumain JM, Imperatore atherosclerosis cardiovascular disease in diabetic patients. Postgrad Med.
G, Fagot-Campagna A, Pettitt DJ, Bennett PH, Knowler WC. Evaluation 2021;133(8):860–4.
of simple indices of insulin sensitivity and insulin secretion for use in 37. Ozder A. Lipid profile abnormalities seen in T2DM patients in pri‑
epidemiologic studies. Am J Epidemiol. 2000;151(2):190–8. mary healthcare in Turkey: a cross-sectional study. Lipids Health Dis.
19. Ter Horst KW, van Galen KA, Gilijamse PW, Hartstra AV, de Groot PF, van 2014;13:183.
der Valk FM, Ackermans MT, Nieuwdorp M, Romijn JA, Serlie MJ. Methods 38. Li K, Sinclair AJ, Zhao F, Li D. Uncommon fatty acids and cardiometabolic
for quantifying adipose tissue insulin resistance in overweight/obese health. Nutrients. 2018;10(10):1559.
humans. Int J Obes (Lond). 2017;41(8):1288–94. 39. Villasanta-Gonzalez A, Alcala-Diaz JF, Vals-Delgado C, Arenas AP, Cardelo
20. Quintana-Navarro GM, Alcala-Diaz JF, Lopez-Moreno J, Perez-Corral I, MP, Romero-Cabrera JL, Rodriguez-Cantalejo F, Delgado-Lista J, Malagon
Leon-Acuna A, Torres-Pena JD, Rangel-Zuniga OA, Arenas de Larriva AP, MM, Perez-Martinez P, et al. A plasma fatty acid profile associated to
Corina A, Camargo A, et al. Long-term dietary adherence and changes in type 2 diabetes development: from the CORDIOPREV study. Eur J Nutr.
dietary intake in coronary patients after intervention with a Mediterra‑ 2022;61(2):843–57.
nean diet or a low-fat diet: the CORDIOPREV randomized trial. Eur J Nutr. 40. Wu JH, Micha R, Imamura F, Pan A, Biggs ML, Ajaz O, Djousse L, Hu FB,
2020;59(5):2099–110. Mozaffarian D. Omega-3 fatty acids and incident type 2 diabetes: a sys‑
21. Fernandez-Ballart JD, Pinol JL, Zazpe I, Corella D, Carrasco P, Toledo E, tematic review and meta-analysis. Br J Nutr. 2012;107(Suppl 2):S214-227.
Perez-Bauer M, Martinez-Gonzalez MA, Salas-Salvado J, Martin-Moreno 41. Wishart DS, Tzur D, Knox C, Eisner R, Guo AC, Young N, Cheng D, Jewell
JM. Relative validity of a semi-quantitative food-frequency question‑ K, Arndt D, Sawhney S, et al. HMDB: the Human Metabolome Database.
naire in an elderly Mediterranean population of Spain. Br J Nutr. Nucleic Acids Res. 2007;35(Database issue):D521-526.
2010;103(12):1808–16. 42. Ferderbar S, Pereira EC, Apolinario E, Bertolami MC, Faludi A, Monte O,
22. Martinez-Gonzalez MA, Fernandez-Jarne E, Serrano-Martinez M, Wright M, Calliari LE, Sales JE, Gagliardi AR, Xavier HT, et al. Cholesterol oxides as
Gomez-Gracia E. Development of a short dietary intake questionnaire for biomarkers of oxidative stress in type 1 and type 2 diabetes mellitus.
the quantitative estimation of adherence to a cardioprotective Mediter‑ Diabetes Metab Res Rev. 2007;23(1):35–42.
ranean diet. Eur J Clin Nutr. 2004;58(11):1550–2. 43. Moore LD, Le T, Fan G. DNA methylation and its basic function. Neuropsy‑
23. American Diabetes Association Professional Practice C. 2 Classification chopharmacology. 2013;38(1):23–38.
and diagnosis of diabetes: standards of medical care in diabetes-2022. 44. Wang Q, Tang SB, Song XB, Deng TF, Zhang TT, Yin S, Luo SM, Shen W,
Diabetes Care. 2022;45(Suppl 1):S17–38. Zhang CL, Ge ZJ. High-glucose concentrations change DNA methylation
24. Bylesjo M, Rantalainen M, Nicholson JK, Holmes E, Trygg J. K-OPLS pack‑ levels in human IVM oocytes. Hum Reprod. 2018;33(3):474–81.
age: kernel-based orthogonal projections to latent structures for predic‑ 45 Wang Y, Chen L, Pandak WM, Heuman D, Hylemon PB, Ren S. High
tion and interpretation in feature space. BMC Bioinformatics. 2008;9:106. glucose induces lipid accumulation via 25-hydroxycholesterol DNA-CpG
25. Cloarec O, Dumas ME, Craig A, Barton RH, Trygg J, Hudson J, Blancher methylation. iScience. 2020;23(5):101102.
C, Gauguier D, Lindon JC, Holmes E, et al. Statistical total correlation 46. Nagao K, Nakamitsu K, Ishida H, Yoshinaga K, Nagai T, Mizobe H, Kojima K,
spectroscopy: an exploratory approach for latent biomarker identification Yanagita T, Beppu F, Gotoh N. Comparison of the lipid-lowering effects of
from metabolic 1H NMR data sets. Anal Chem. 2005;77(5):1282–9. four different n-3 highly unsaturated fatty acids in HepG2 cells. J Oleo Sci.
26. Lifestyle Management. <em>Standards of Medical Care in Diabe‑ 2014;63(10):979–85.
tes—2018</em&gt. Diabetes Care. 2018;41(Supplement 1):S38. 47. Elsner M, Gehrmann W, Lenzen S. Peroxisome-generated hydrogen
27. Taylor R, Al-Mrabeh A, Zhyzhneuskaya S, Peters C, Barnes AC, Aribisala BS, peroxide as important mediator of lipotoxicity in insulin-producing cells.
Hollingsworth KG, Mathers JC, Sattar N, Lean MEJ. Remission of human Diabetes. 2011;60(1):200–8.
type 2 diabetes requires decrease in liver and pancreas fat content but is
dependent upon capacity for β cell recovery. Cell Metab. 2018;28(4):547-
556.e543. Publisher’s Note
28. Martin-Timon I, Sevillano-Collantes C, Segura-Galindo A, Del Canizo- Springer Nature remains neutral with regard to jurisdictional claims in pub‑
Gomez FJ. Type 2 diabetes and cardiovascular disease: have all risk factors lished maps and institutional affiliations.
the same strength? World J Diabetes. 2014;5(4):444–70.
29. Lee WJ, Hur KY, Lakadawala M, Kasama K, Wong SK, Chen SC, Lee YC,
Ser KH. Predicting success of metabolic surgery: age, body mass index,
C-peptide, and duration score. Surg Obes Relat Dis. 2013;9(3):379–84.
30. Still CD, Wood GC, Benotti P, Petrick AT, Gabrielsen J, Strodel WE, Ibele
A, Seiler J, Irving BA, Celaya MP, et al. Preoperative prediction of type 2
diabetes remission after Roux-en-Y gastric bypass surgery: a retrospective
cohort study. Lancet Diabetes Endocrinol. 2014;2(1):38–45.
31. Rangel-Zuniga OA, Vals-Delgado C, Alcala-Diaz JF, Quintana-Navarro GM,
Krylova Y, Leon-Acuna A, Luque RM, Gomez-Delgado F, Delgado-Lista J,
Ordovas JM, et al. A set of miRNAs predicts T2DM remission in patients
with coronary heart disease: from the CORDIOPREV study. Mol Ther
Nucleic Acids. 2021;23:255–63.
32. Sanchez-Ceinos J, Rangel-Zuniga OA, Clemente-Postigo M, Podadera-
Herreros A, Camargo A, Alcala-Diaz JF, Guzman-Ruiz R, Lopez-Miranda J,
Malagon MM. miR-223-3p as a potential biomarker and player for adipose
tissue dysfunction preceding type 2 diabetes onset. Mol Ther Nucleic
Acids. 2021;23:1035–52.
33. Vals-Delgado C, Alcala-Diaz JF, Molina-Abril H, Roncero-Ramos I, Caspers
MPM, Schuren FHJ, Van den Broek TJ, Luque R, Perez-Martinez P, Katsiki
N, et al. An altered microbiota pattern precedes type 2 diabetes mellitus
development: From the CORDIOPREV study. J Adv Res. 2022;35:99–108.
34. Taskinen MR. Diabetic dyslipidemia. Atheroscler Suppl. 2002;3(1):47–51.
35. Han CL, Qu CZ. Cardiovascular risk and safety evaluation of a dual peroxi‑
some proliferator-activated receptor-alpha/gamma agonist, aleglitazar,