Studying Alzheimer’s disease through an integrative serum metabolomic and lipoproteomic approach
- Open Access
- 01.12.2025
- Research
Erschienen in:
Abstract
Background
Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline and behavioral changes. It is the most common neurodegenerative disorder, affecting about 5–7% of the population over 60 years of age [1].
Alzheimer’s disease typically advances through clinical stages, starting with subtle cognitive decline, progressing to mild cognitive impairment (MCI) and evolving to dementia with severe cognitive dysfunction and loss of self-sufficiency [2]. Prognosis at the individual level is challenging due to the variable rate of clinical progression among individuals, which is influenced by factors such as age, cognitive reserve, genetic predisposition, and comorbidities. The incomplete understanding of molecular mechanisms underlying AD might play a relevant role in such variability [3]. Therefore, accurate prediction of disease trajectory remains difficult.
Anzeige
The heterogeneity in how this disease progresses to dementia among patients presents a significant challenge for clinicians, caregivers, and patients in making informed decisions about treatment and long-term care planning, as well as for designing clinical trials that can effectively identify substantial benefits of candidate treatments [4].
Omics sciences represent a very promising instrument to perform the analysis of patients and their biological characteristics within the dynamic context of disease evolution, thus enabling the molecular characterization of a disease onset and evolution, and providing insight into individual susceptibility to drug treatments [5‐10]. Given these premises, metabolomics and lipoproteomics present themselves as compelling approaches for investigating alterations of multiple biochemical networks throughout the entire course of AD [11‐18].
Here we proposed an integrated metabolomics and lipoproteomics approach to study via nuclear magnetic resonance (NMR) spectroscopy serum samples of a consecutive series of AD patients at different cognitive/functional stages, i.e. subjects with AD at dementia stage (AD-dem) and subjects with MCI due to AD (MCI-AD). As control group, we considered 31 subjects with mild cognitive impairment in whom AD and other neurodegenerative disorders were excluded (MCI). The aims of this pilot study were: (i) to evaluate the diagnostic accuracy of metabolomics and lipoproteomics approaches in identifying AD patients, (ii) to investigate the potential prognostic role of these approaches to track cognitive progression of MCI-AD patients.
Methods
Study population
For the aim of this pilot, hypothesis-generating study, we retrospectively considered a consecutive series of 133 patients referring to the Center for Memory Disturbances, Section of Neurology of the University of Perugia. At baseline all patients underwent clinical neurological examination, neuropsychological assessment including Mini-Mental State Examination (MMSE) and Clinical Dementia Rating scale, cerebrospinal fluid (CSF) and blood collections. CSF was analyzed for AD core biomarkers including Aβ42, Aβ40, p-tau, and t-tau assessed with Lumipulse fully-automated platform in the Laboratory of Clinical Neurochemistry, Section of Neurology (University of Perugia) [19]. According to clinical/neuropsychological characteristics and CSF biomarker profile (concomitant positivity for amyloidosis and tauopathy biomarkers, A+/T+, defining AD [20]), patients were subdivided in: subjects with MCI-AD (n. 45), patients with AD-dem (n. 57) and patients with MCI not associated to AD (MCI, n. 31). The latter group was composed of subjects who underwent lumbar puncture for cognitive deficits. AD was excluded by means of CSF analysis (none of them exhibited a CSF A+/T + profile) and other neurodegenerative disorders were ruled out based on clinical/neuropsychological evaluation and imaging assessment (18F-FDG PET). Exclusion criteria for all patients enrolled included: severe vascular encephalopathy, recent major stroke and brain injury, major systemic disorders.
Anzeige
Clinical and neuropsychological assessments were available longitudinally, once a year (follow-up ranging from 1 to 5 years). Difference in MMSE scores at the last and at the first follow up divided by the time between the two visits was used to evaluate progressive cognitive deterioration. Patients who displayed a decrease in MMSE lower than 1.5 point per year were considered at lower progression rate, following this criterion we obtained a division in 18 MCI-AD at lower progression rate (MCI-AD LR) and 27 at higher progression rate (MCI-AD HR).
All patients (or their legal representatives) gave their written informed consent for the study participation.
NMR analysis
Serum samples analysis was carried out at the metabolomics laboratory of CERM/CIRMMP (University of Florence). Serum samples were previously stored at -80 °C. They were thawed at room temperature and homogenized by shaking. A total of 350 µL of sodium phosphate buffer (75 mM Na2HPO4 × 7H2O; 20% (v/v) 2H2O, 4.6 mM 3-(Trimethylsilyl)propionate-2,2,3,3-d4; 6.1 mM NaN3; pH 7.4) was added to 350 µl of each sample. After vortex mixing for 30 s, 600 µL of each mixture was transferred into a 5 mm NMR tube. All NMR spectra were acquired using a 600 MHz spectrometer (Bruker BioSpin) operating at 600.13 MHz proton Larmor frequency provided with an automatic and refrigerated (6 °C) sample changer (SampleJet, Bruker BioSpin), and with a BTO 2000 thermocouple utilized for temperature stabilization at 310 K (~ 0.1 K at the sample). To ensure high spectral quality and reproducibility, the spectrometer was calibrated daily following strict standard operating procedures. For each serum sample, a standard nuclear Overhauser effect spectroscopy (NOESY) sequence using 32 scans, 98,304 data points, a spectral width of 18,028 Hz, an acquisition time of 2.7 s, a relaxation delay of 4 s and a mixing time of 0.01 s [21] was applied to detect the NMR signals of both high (e.g., lipoproteins, proteins, lipids) and low (metabolites) molecular weight molecules in concentrations above the NMR detection limit. Before applying Fourier transform, free induction decays were multiplied by an exponential function equivalent to a 0.3 Hz line-broadening factor. Transformed spectra were automatically corrected for phase and baseline distortions and calibrated to the anomeric glucose doubled at δ 5.24 ppm.
A panel of 41 metabolites was unambiguously identified and quantified using the Plasma/Serum B.I.Quant-PS platform™ (version 2.0.0). Metabolites with more than 20% of observation under the limit of quantification were excluded from the present analysis, thus the system was reduced to 24 metabolites. In addition, the signals of the glycoproteins GlycA at δ 2.04 and GlycB at δ 2.08 ppm were quantified by integration using an R script developed in-house. The identification and quantification of 112 lipoprotein-related parameters were performed utilizing the B.I. LISA platform™ (version 1.0.0). This platform provides information about the concentrations of triglycerides, cholesterol, free cholesterol, phospholipids, Apo-A1, Apo-A2 and Apo-B100 of the main fractions and subfractions of VLDL, IDL, LDL and HDL classes.
Statistical analysis
All data analyses were performed in the “R” statistical environment. All pairwise ratios of metabolites and lipoproteins were calculated by means of the function “calc.rapports”, included in the R package “SARP.compo”. The matrix of metabolites, lipoproteins and their ratios was used as input for the following analyses.
The logistic LASSO regression algorithm was employed to identify the optimal combination of metabolic ratios to discriminate the groups of interest. This method can select a more concise and interpretable set of predictors from a large set of variables in the regression. LASSO regression models were calculated using the “glm” function of “glmnet” R package. The cut-off for discriminating the groups was optimized maximizing the sum of sensitivity and specificity (R package “cutpointr”).
To discriminate AD-dem and MCI patients a logistic regression model based on phenylalanine to triglycerides (TG) LDL 4 ratio, citrate to cholesterol (Chol) VLDL 2 ratio, VLDL 2 to TG LDL 4 ratio, age and sex was computed. This model was calculated on a training set constituted by 43 AD-dem and 17 MCI patients and the area under the receiver characteristic curve (AUC) of the model was internally validated by means of leave-one-out cross-validation (LOO-CV) scheme. To discriminate between AD-dem and MCI, the cut-off for the combined score was optimized at 0.73. This model was further validated in an external group of 14 AD-dem and 14 MCI patients.
To discriminate MCI-AD at faster and slower progression to dementia a logistic regression model based on histidine, N,N-Dimethylglycine, lactate and Chol VLDL-5 was computed. The AUC of this model was internally validated by means of LOO-CV scheme. To delineate high risk of faster progression to AD, the cut-off for the combined score was optimized (as previously described) at 0.59.
Anzeige
Results
Overall population characteristics
The baseline characteristics of the included cohort are summarized in Table 1. The groups were matched in age (no statistically significant difference) but not in sex, i.e. male subjects were more prevalent in the MCI group with respect to the MCI-AD group (p-value = 0.036). A/T/N CSF profile prevalences (A refers to amyloidosis, T refers to tauopathy and N refers to neurodegeneration) and MMSE scores were in line with the inclusion criteria for each clinical group (reduced MMSE in ADdem, and A/T/N positivity for MCI-AD and ADdem).
Table 1
Demographic and clinical characteristics of the study cohort
MCI (31) | MCI-AD (46) | AD-dem (57) | MCI vs. MCI-AD p-value | MCI vs. AD-dem p-value | MCI-AD vs. AD-dem p-value | |
|---|---|---|---|---|---|---|
Sex, n M/n F | 18/13 | 15/31 | 22/35 | 0.036 | n.s. | n.s. |
Age (years), median ± IQR | 72 ± 12 | 72 ± 9 | 72 ± 9 | n.s. | n.s. | n.s. |
MMSE, median ± IQR | 24 ± 4 | 24 ± 4 | 17 ± 5 | n.s. | < 0.001 | < 0.001 |
A+ (%) | 19% | 100% | 100% | < 0.001 | < 0.001 | n.s. |
T+ (%) | 17% | 100% | 100% | < 0.001 | < 0.001 | n.s. |
N+ (%) | 25% | 93% | 97% | < 0.001 | < 0.001 | n.s. |
A+/T+ (%) | 0% | 100% | 100% | < 0.001 | < 0.001 | n.s. |
Discrimination of AD-dem and MCI
To determine whether it is possible to discriminate AD-dem and MCI patients, we looked for possible combinations of metabolites-lipoproteins and their ratios. Phenylalanine to TG LDL 4 ratio, citrate Chol VLDL 2 ratio, Chol VLDL 2 to TG LDL 4 ratio coupled with age and sex were used to build the model (Table 2). To ensure that the calculated model was reliable, we split the dataset into training and validation sets. The AUC of the training set was also internally validated using a LOO-CV approach. In the training set our model discriminates AD-dem and MCI with sensitivity 83.7%, specificity 76.5% and accuracy 81.7% (Fig. 1A). These results were reproduced in the validation set: sensitivity 78.6%, specificity 71.4%, and accuracy 75.0% (Fig. 1B).
Fig. 1
Discrimination between AD-dem and MCI patients using phenylalanine to TG LDL 4 ratio, citrate Chol VLDL 2 ratio, Chol VLDL 2 to TG LDL 4 ratio, age and sex in the (A) training and (B) validation sets. The receiver operator curves (ROC) and the area under the curve (AUC) scores are presented. Sensitivity, specificity and accuracy are also reported
Anzeige
Predicting MCI-AD disease trajectory to dementia
The study population comprises a group of 45 patients affected by AD at pre-dementia stage (MCI-AD). Evolution of these patients to overt dementia was evaluated over time. Patients who displayed a decrease in MMSE < 1.5 point per year were considered at a lower progression rate, following this criterion we obtained a division in 18 MCI-AD at lower progression rate (MCI-AD LR) and 27 at higher progression rate (MCI-AD HR).
Histidine, N,N-Dimethylglycine, lactate and Chol VLDL-5 (Table 2; Fig. 2A), selected by a LASSO regression approach, were used to build the model to discriminate MCI-AD LR and MCI-AD HR. This model was internally validated using a LOO-CV approach and cross-validated results are reported in Fig. 2B. The model discriminated MCI-AD LR and MCI-AD HR with sensitivity 74.2%, specificity 72.2%, and accuracy 73.3%.
Table 2
Metabolic features selected by LASSO in training predictive models
Metabolic features included in the models | LOO-CV AUC | |
|---|---|---|
MCI vs. AD-dem | Phenylalanine to TG LDL 4 ratio, citrate Chol VLDL 2 ratio, Chol VLDL 2 to TG LDL 4 ratio coupled with age and sex | 0.811 |
MCI-AD LR vs. MCI-AD HR | Histidine, N,N-Dimethylglycine, lactate and Chol VLDL-5 | 0.772 |
Fig. 2
(A) Boxplot reporting the levels of the metabolic variables included in the model discriminating MCI-AD LR (yellow) and MCI-AD HR (orange), p-values < 0.05 are indicated with an asterisk. (B) Predictive discrimination between MCI-AD LR and MCI-AD HR. The ROC curve and the AUC score are presented. Sensitivity, specificity and accuracy are also reported.
Anzeige
Discussion
The complexity in understanding the molecular underpinnings of AD, combined with the heterogeneity in clinical progression, has made predicting the course of the disease a key focus of research [3]. There is a critical need for less invasive and more accessible biomarkers to predict AD progression both for clinical management of patients and for selecting patients and monitoring treatment response in clinical trials.
In this pilot study, we proposed an integrated metabolomics and lipoproteomics approach to unravel metabolic alterations at the systemic level, as reflected in the blood serum, in subjects with mild cognitive impairment, mild cognitive impairment due to AD and AD dementia.
The model calculated using phenylalanine to TG LDL 4 ratio, citrate to Chol VLDL 2 ratio, Chol VLDL 2 to TG LDL 4 ratio, age and sex, which discriminates between AD-dem and MCI patients with an accuracy of about 82% in the training set and 75% in the validation set, is consistent with prior findings suggesting that circulating metabolic markers could provide a window into brain pathology [11, 12, 22]. Phenylalanine decrease in serum, cerebrospinal fluid and brain from AD patients has been already described [23, 24]. Phenylalanine, as well as tyrosine, participates in catecholamine metabolism for dopamine synthesis, and alterations of the dopaminergic system have been frequently reported in Alzheimer’s disease and associated with cognitive decline [25, 26]. The three selected ratios all included lipoprotein-related parameters. This finding is not surprising given that there is growing biological and epidemiological evidence suggesting a link between the lipoprotein metabolism and the development of AD [11, 27, 28].
A particularly valuable and novel aspect of our study is the stratification of MCI-AD patients based on the rate of cognitive worsening. MCI due to AD is a heterogeneous condition, with some patients progressing to AD relatively quickly, while others may remain cognitively stable for years, making particularly difficult to predict when a patient will enter the dementia stage [29]. Our model, calculated using histidine, N,N-dimethylglycine, lactate and cholesterol VLDL-5, can distinguish between these two groups with a cross-validated accuracy of 73.3%. Among these four metabolic features, histidine and cholesterol VLDL 5 showed to be significantly decreased in MCI-AD patients showing the higher worsening rate. Decrease of histidine levels has been associated with cognitive decline and dementia, postulating possible antioxidant actions of this amino acid [12, 30]. Recently, a pilot study provides evidence that lower levels of VLDL cholesterol may be linked to amyloid pathology [31]. The results obtained using our model, although in a small cohort, add an objective and quantitative dimension to the prediction of AD progression, providing a promising base for future studies and validations. Identifying which MCI-AD patients are at lower/higher risk of rapid progression is crucial for optimizing treatment plans, patient counseling, and clinical trial design. Indeed, AD trials often struggle with the heterogeneity of patient populations, which can dilute treatment effects and make it difficult to achieve statistically significant results [32‐34]. By identifying fast- and slow-progressing patients, clinical trials could be enriched with patients who are more likely to experience measurable cognitive decline within the study period, increasing the likelihood of detecting a treatment effect.
Anzeige
While the results of the present study are promising and novel, there are some limitations to consider. First, the findings of this pilot study need to be validated in larger and independent cohorts of patients. Secondly, future studies should include additional information, such as genetic risk factors (e.g., APOE genotype), lifestyle factors, and other clinical measures. Finally, the small sample size prevented us from performing sex-based analyses.
Conclusions
In conclusion, this study has enabled the identification of a combination of metabolite-lipoprotein ratios and demographic variables capable of discriminating between MCI and AD-dem patients, as well as predicting which MCI-AD patients will worsen faster or slower. The analysis was conducted using serum samples, thus exploring a minimally invasive approach to patient assessment. While further validation and refinement of the models in more numerous cohorts are needed, the results already provide strong evidence that systemic metabolic changes are closely tied to the neurodegenerative processes of AD. The results of this pilot retrospective study constitute a basis for future research towards the development of more effective strategies for early detection, risk stratification, and personalized treatment of AD.
Acknowledgements
The authors acknowledge the support and the use of resources of Instruct-ERIC, a landmark ESFRI project, and specifically the CERM/CIRMMP Italy center, as well as the project “Potentiating the Italian Capacity for Structural Biology Services in Instruct-ERIC, Acronym “ITACA. SB” (Project no. IR0000009) within the call MUR 3264/2021 PNRR M4/C2/L3.1.1, funded by the European Union – NextGenerationEU.
Declarations
Ethics approval and consent to participate
The present study has received the approval from the local Ethics Committee (Comitato Etico Aziende Sanitarie Regione Umbria 19369/AV and 20942/21/OV), and all patients and/or their legal representatives gave informed written consent to participate in the study.
Anzeige
Consent for publication
Not applicable.
Competing interests
The authors declare that the research was conducted without any commercial or financial relationships that could be construed as potential conflicts of interest.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, 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 you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. 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://creativecommons.org/licenses/by-nc-nd/4.0/.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
