Biochemical and Biophysical Research Communications 375 (2008) 356–361
Contents lists available at ScienceDirect
Biochemical and Biophysical Research Communications journal homepage: www.elsevier.com/locate/ybbrc
A multi-metabolite analysis of serum by 1H NMR spectroscopy: Early systemic signs of Alzheimer’s disease Taru Tukiainen a, Tuulia Tynkkynen b, Ville-Petteri Mäkinen a,c,d, Pasi Jylänki a, Antti Kangas a, Johanna Hokkanen a, Aki Vehtari a, Olli Gröhn e, Merja Hallikainen f, Hilkka Soininen g, Miia Kivipelto g, Per-Henrik Groop c,d, Kimmo Kaski a, Reino Laatikainen b, Pasi Soininen b, Tuula Pirttilä g, Mika Ala-Korpela a,c,d,* a
Computational Medicine Research Group, Department of Biomedical Engineering and Computational Science, Helsinki University of Technology, P.O. Box 9203, FI-02015 HUT, Finland Laboratory of Chemistry, Department of Biosciences, University of Kuopio, Finland Folkhälsan Institute of Genetics, Folkhälsan Research Center, Helsinki, Finland d Division of Nephrology, Department of Medicine, Helsinki University Central Hospital, Finland e A.I. Virtanen Institute, University of Kuopio, Finland f Brain Research Unit, University of Kuopio, Finland g Department of Clinical Science, Neurology Unit, University of Kuopio and Kuopio University Hospital, Finland b c
a r t i c l e
i n f o
Article history: Received 16 July 2008 Available online 26 August 2008
Keywords: Alzheimer’s disease Lipid extracts Lipoproteins Metabolomics Metabonomics Mild cognitive impairment NMR spectroscopy Self-organising map Serum .
a b s t r a c t A three-molecular-window approach for 1H NMR spectroscopy of serum is presented to obtain speciﬁc molecular data on lipoproteins, various low-molecular-weight metabolites, and individual lipid molecules together with their degree of (poly)(un)saturation. The multiple data were analysed with selforganising maps, illustrating the strength of the approach as a holistic metabonomics framework in solely data-driven metabolic phenotyping. We studied 180 serum samples of which 30% were related to mild cognitive impairment (MCI), a neuropsychological diagnosis with severely increased risk for Alzheimer’s disease (AD). The results underline the association between MCI and the metabolic syndrome (MetS). Additionally, the low relative amount of x-3 fatty acids appears more indicative of MCI than low serum x-3 or polyunsaturated fatty acid concentration as such. The analyses also feature the role of elevated glycoproteins in the risk for AD, supporting the view that coexistence of inﬂammation and the MetS forms a high risk condition for cognitive decline. Ó 2008 Elsevier Inc. All rights reserved.
Metabonomics applications relying on 1H NMR spectroscopy are of increasing interest in the molecular characterisation of body ﬂuids for disease risk assessment and diagnostics [1–5]. While 1H NMR per se allows fast and reliable detection of a large number of metabolites, the molecular variety and multiple environments, particularly in serum, hamper the molecular identiﬁcation and quantiﬁcation. Here, we present a new three-molecular-window approach that gives speciﬁc molecular data on macromolecular lipid–protein aggregates such as lipoprotein particles, on various low-molecular-weight metabolites, and also on individual lipid molecules together with their degree of (poly)(un)saturation. The new approach was applied to a set of 180 serum samples from elderly people. According to neuropsychological diagnosis,
30% of the samples related to mild cognitive impairment (MCI), a transitional state with considerably increased risk for Alzheimer’s disease (AD) . The initiation and development of AD are poorly understood and there are no distinct biomarkers allowing early detection in order to facilitate preventive treatment. Recent evidence suggest that systemic metabolite and lipid levels associate more strongly with the development of AD than previously believed [7,8]. Thus, it was anticipated that the 1H NMR-based multi-metabolite analysis would provide new molecular insights on the potential early changes in systemic metabolism that relate to MCI and thereby to high risk for AD.
Materials and methods * Corresponding author. Address: Computational Medicine Research Group, Department of Biomedical Engineering and Computational Science, Helsinki University of Technology, P.O. Box 9203, FI-02015 HUT, Finland. E-mail address: [email protected]
ﬁ (M. Ala-Korpela). URL: http://www.computationalmedicine.ﬁ/ (M. Ala-Korpela). 0006-291X/$ - see front matter Ó 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.bbrc.2008.08.007
Study population. A population-based sample of 806 elderly individuals was recruited from the city of Kuopio, Finland, between October 1997 and November 1998 to determine the prevalence of MCI among elderly population . The participants free of demen-
T. Tukiainen et al. / Biochemical and Biophysical Research Communications 375 (2008) 356–361
tia were followed-up at approximately 3, 5, and 6 years after the baseline visit. Altogether 45 individuals completed the full 6 years of follow-up. This group of individuals was included in this pilot study (see Supplementary Table). On each visit the participants underwent a structured interview including demographic and medical information together with an extensive neuropsychological assessment (described in detail in ). For the diagnosis of MCI, the diagnostic criteria proposed by Mayo Clinic Alzheimer’s Disease Research Center were applied . At baseline two of the participants were classiﬁed as having MCI. On the following visits the number of MCI subjects was 18, 16, and 18. In addition, at the last visit one of the subjects received a dementia diagnosis. At each visit, blood samples were drawn after an overnight fast of 12 h, and weight, height, and blood pressure were also measured. Serum lipid extraction. Lipids were extracted from the serum samples using a standard protocol described in the literature [10,11] with some in-house modiﬁcations. Brieﬂy, 5 ml of methanol was added drop by drop into 0.5 ml of serum, after which 10 ml of dichloromethane and 15 ml of 0.15 M NaCl were added to the sample. The sample was vigorously vortexed for 45 s and centrifuged for 20 min at 2200g to separate the organic and water phase. The lower organic phase was recovered and the aqueous layer was extracted once again with 10 ml dichloromethane to standardise the yield. The separated organic layers were combined and evaporated for sufﬁcient dryness. To remove residual CH2Cl2 and H2O, 1 ml of CCl4 and 0.1 ml of CD3OD were added into the dried lipid extract and removed by evaporation. Finally, the lipids were redissolved into 0.6 ml of CDCl3 containing 0.09% of tetramethylsilane.
NMR spectroscopy. 1H NMR measurements for three molecular windows, namely lipoprotein lipids (LIPO), low-molecular-weight metabolites (LMWM) and lipid extracts (LIPID) were performed for all serum samples on a Bruker AVANCE 500 DRX spectrometer operating at 500.13 MHz (see Fig. 1). For the 64 K LIPO data, 128 transients were collected with a 90° ﬂip angle. Acquisition time of 6.2 s and a relaxation delay of 0.1 s were used. The 64 K LMWM data were collected with a standard 1D CPMG pulse sequence with a 325 ms T2-ﬁlter using a ﬁxed inter echo delay of 400 ls that eliminates diffusion and J-modulation effects. Acquisition time of 6.2 s and a relaxation delay of 8.7 s were used and 48 transients were collected after 16 dummy scans. Data were recorded at a physiological temperature of 310 K for LIPO and LMWM windows using a double tube system facilitating absolute metabolite quantiﬁcation. The external reference tube (o.d. 2 mm, supported by a Teﬂon adapter) containing the reference substance (sodium 3-trimethylsilyl[2,2,3,3-d4]propionate (TSP) 40 mmol/l, MnSO4 0.6 mmol/l in 99.8% D2O) was placed coaxially into the NMR sample tube (o.d. 5 mm) containing 430 ll of each sample. No water suppression was used. For the LIPID window 64 K data were measured at 293 K using a standard 5 mm tube. The acquisition time was 6.1 s, the spectral width 5340 Hz, the relaxation delay 3.9 s, and the total recycling time 10 s. Sixty-four transients were collected after 4 dummy scans. Data processing. The free induction decays (FIDs) with 65,536 data points were zero-ﬁlled and multiplied by an exponential window function with a 1.0 Hz line-broadening for the LIPO spectra and 0.5 Hz line-broadening for the LMWM spectra. The preprocess-
Fig. 1. Representative 1H NMR spectra of serum via the three different molecular windows used. These spectra were used as inputs for the self-organising maps. The LIPO window is dominated by broad signals arising from macromolecules, mainly lipoprotein lipids and albumin. In the LMWM window a pulse sequence that suppresses the macromolecule signals is applied, thus, enhancing the detection of smaller solutes. The extraction procedure of serum lipids breaks down the lipoprotein particles yielding valuable information on the individual lipid species inside these particles, i.e., molecular data not captured in the NMR experiments of native serum. The water region in the LMWM window and the narrow spectral regions (marked with grey squares) with potential residual solvent peaks in the LIPID window were excluded from the analysis. EC, esteriﬁed cholesterol; FA, fatty acid; FC, free cholesterol; PC, phosphatidylcholine; PGLY, phosphoglycerides; PUFA, polyunsaturated fatty acid; sat, saturated; SM, sphingomyelin; TC, total serum cholesterol; TG, total serum triglycerides. The resonances marked x-6 may also contain a small contribution from x-7 fatty acids.
T. Tukiainen et al. / Biochemical and Biophysical Research Communications 375 (2008) 356–361
ing of the FIDs and the subsequent Fourier transformations were done on the PERCH NMR software platform (PERCH Solutions Ltd., Kuopio, Finland). Each LIPO and LMWM spectrum was scaled according to the area of the corresponding TSP reference signal at 0 ppm. Baseline corrections and chemical shift adjustments in the LIPO and LMWM windows were done as previously described in . All preprocessing was performed in the Matlab programming environment (The MathWorks Inc., Natick, MA, USA) using the statistical toolbox and in-house scripts. The FIDs with 65,536 data points for the LIPID window were zero-ﬁlled and multiplied by an exponential window function with a 0.3 Hz line-broadening. Visual inspection revealed ethanol contamination in four LMWM spectra (two controls and two MCI subjects), which, together with the related LIPO and LIPID spectra, were subsequently excluded from further analyses. Thus, the total number of spectra included in the ﬁnal analyses was 528 (instead of all the 540 measured). Lineshape ﬁtting of the LIPID window. The areas of the lipid resonances in the LIPID spectra were determined using lineshape ﬁtting analysis . A model lineshape based approach was adopted to enable quantitative analysis of severely overlapping peaks and to increase the quantiﬁcation accuracy [12,13]. Intensity ratios of some well-deﬁned multiplets were used as constraints, and, in some cases, the known coupling constants or relative line widths were also constrained (see Fig. 2). This kind of analysis is often termed as ‘‘the use of biochemical prior knowledge” and its use is recommended to decrease the mathematical uncertainties with overlapping resonances . It can be applied in situations where the molecular components within the spectral region of interest are known. This sophisticated methodology allowed us to get information on the amounts of several lipid components, e.g., sphingomyelin and x-3 fatty acids. Also, the average degree of (poly)(un)saturation could be calculated from these variables. Absolute quantiﬁcation is not established in the lipid extraction procedure. Thus, each LIPID spectrum was scaled, via the ﬁtted
cholesterol –C(27)H3 resonance at 0.8617 ppm, according to the serum total cholesterol as estimated from the corresponding LIPO spectrum (see below). The PERCH NMR software platform was used for all the lineshape ﬁtting analyses . Lineshape ﬁtting of the LMWM window. Some low-molecularweight metabolites, e.g., lactate, glucose, valine, and acetate, were quantiﬁed from the LMWM spectra using lineshape ﬁtting analysis [12,13]. The PERCH NMR software platform was used . Regression models for the LIPO spectra. Cross-validated regression models were applied to quantify serum triglycerides (TG) and serum cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C) . A simpliﬁed and computationally more efﬁcient modiﬁcation of the approach presented recently in  was used. Self-organising map analysis. The SOM analysis was based solely on the NMR spectra. Before constructing the SOM, the spectra were truncated to 0.3–3.3 ppm (LIPO), 0.6–5.8 ppm (LMWM) and 0.4– 6.1 ppm (LIPID), and the chemical shift resolution was reduced to 0.001 ppm (LIPO), 0.00087 ppm (LMWM), and 0.00095 ppm (LIPID). Data points from 4.1 to 5.1 ppm were omitted from the LMWM spectra due to the water peak. In the LIPID spectra chemical shift areas 2.17–2.19, 3.47–3.51, and 5.28–5.32 were omitted due to solvent signals. In the SOM analysis the spectra of all the three molecular windows were combined for each serum sample. To ensure equivalent contribution in the SOM analysis the total intensity variance of the spectral data points in each molecular window was set equal. The total number of samples used in the SOM analysis was 176 corresponding to 528 spectra. We chose a 7 9 map of hexagonal units (resulting in 2.8 samples per unit) and a Gaussian neighbourhood function. After the positions of the study subjects on the SOM were computed, the map was coloured according to the clinical and biochemical variables within different parts of the SOM with overall permutation estimations for the p-values for the statistical signiﬁcance of the patterns (see Supplementary Material 1 in ). The null distributions from the
Fig. 2. Illustration of the lineshape ﬁtting analysis for the selected key regions of the 1H NMR spectrum of serum lipid extracts (LIPID). The black line represents the observed spectrum (OBS) and the coloured lines refer to the ﬁtted signals. The green line at the bottom represents the difference (DIFF) between the observed spectrum and the ﬁtted signals. The coupling trees above the signals illustrate the used intensity ratio constraints of the well-deﬁned multiplets. In some cases known coupling constants (given next to a coupling tree) were used as an additional constraint for the ﬁtting. See the caption of Fig. 1 for the key for the abbreviations.
T. Tukiainen et al. / Biochemical and Biophysical Research Communications 375 (2008) 356–361
permutation analysis were also the basis of the colour scale in each component plane so that categorical and continuous variables can be compared visually while maintaining the statistical interpretation. In cases where the variables were derived from the spectra, i.e., the actual input data of the SOM analysis, the p-value estimation is no longer strictly valid. Consequently, q (instead of p) was used to denote the level of regional variability on the map. All the analyses were performed using in-house scripts in the MATLAB programming environment. An open source package (termed MeliKerion) for SOM analyses in the Matlab/Octave programming environment is freely available via http://www.computationalmedicine.ﬁ/software. Results and discussion We have recently developed a holistic metabonomics framework that combines 1H NMR spectroscopy and self-organising map (SOM) analysis in metabolic phenotyping . The same strategy was applied here to comprehensively analyse the 1H NMR spectra available from the three molecular windows. A visual summary of the multi-metabolite data obtained by the new 1H NMR spectroscopy approach, together with molecular identiﬁcations, is given in Fig. 1. It is notable that the data analysis via the SOM was based solely on the NMR spectral data and thereby provides a purely data-driven metabolic phenotyping. The clinical variables available as well as the metabolite concentrations estimated from the NMR spectra were used only to elucidate and interpret the observed structure of the SOM (Fig. 3 and the Supplementary Figure).
The MCI samples cluster on the southeast part of the self-organising map (p < 0.05) as seen in Fig. 3. The SOM analysis performed is solely based on the combination of the three spectra (LIPO, LMWM, and LIPID in Fig. 1) for each serum sample. Thus, it appears that systemic metabolism, as indicated by the molecular composition of serum and revealed by the new 1H NMR approach, is indicative of MCI and thereby increased risk for Alzheimer’s disease (AD). However, the distribution of MCI samples is not inclusive, most likely reﬂecting the complex link between the serum biochemistry and cognitive decline as well as potentially various differing biochemical pathways behind MCI. This is also supported by the analyses at individual time points since none of the serum metabolites showed statistically signiﬁcant changes between the control and MCI groups or between the different time points. Nevertheless, the holistic SOM analysis based on the multi-metabolite information in the 1H NMR spectra of serum did deﬁne statistically signiﬁcant metabolic associations for the MCI, suggesting that the combined changes of several metabolites can be descriptive while the changes in the individuals metabolites are not. As expected, the distribution of samples in Fig. 3 also indicates that diabetes and obesity (high body mass index; BMI) are closely related. The strong association between these two conditions and MCI is also consistent with the current clinical knowledge . Thus, the NMR data as such reveals the link between vascular factors and cognitive decline and risk for dementia. Thereby, the new multi-metabolite molecular phenotyping of serum by 1H NMR spectroscopy uncovers the recently established clinical interrelationships . With the extensive molecular information available
Fig. 3. Statistical colourings of the clinical and biochemical variables in the SOM analysis of the combined LIPO, LMWM, and LIPID molecular windows from 1H NMR spectroscopy of 176 serum samples (corresponding to 528 spectra, see Fig. 1). Due to the SOM analysis the samples are positioned so that the multi-metabolite differences between nearby samples are minimised. Mild cognitive impairment (MCI) and diabetes status are shown as an example of the clinical diagnostics and BMI indicates the body mass index. TC refers to serum total cholesterol, TG to serum triglycerides, LDL-C and HDL-C to low and high-density lipoprotein cholesterol, respectively (estimated from the LIPO window). Acyl groups denotes the total amount of –CH3 groups in serum lipids. x-3 absolute refers to the absolute amount of serum x-3 fatty acids and x-3 relative (as well as x-6 relative) is the corresponding portion of all fatty acids in serum. SM and PC refer to sphingomyelin and phosphatidylcholine, respectively (estimated from the LIPID window). The rest of the metabolites shown are estimated from the LMWM window. For the details of the SOM analysis and the deﬁnition of p and q, see Materials and methods section.
T. Tukiainen et al. / Biochemical and Biophysical Research Communications 375 (2008) 356–361
via the 1H NMR spectra, this basis provides now a novel opportunity for a multi-metabolite characterisation of the risk for AD. Recently one of the key issues has been the role of serum cholesterol (TC) as a risk factor for AD [7,8]. Here, the highest TC values concentrate on the northeast corner of the SOM. The LDL-C values follow the TC with only subtle differences. Notably, the low HDL-C clearly associates with high triglycerides (TG) in the southeast corner; this indicates association of the metabolic syndrome (MetS) with MCI and the risk for AD. Thereby, within the current limited data set, it appears that MetS would be more associated with MCI than high TC as such. Line ﬁtting analysis was performed for all the 1H NMR spectra of the serum lipid extracts to speciﬁcally quantify the individual lipid constituents [12,13]. This allowed detailed molecular interpretation of the biomedical features as discussed above. As expected, the concentration of the main serum lipids, TC and TG, coincides with the highest amount of acyl groups on the eastern section of the map. The total amount of x-3 fatty acids peaks at the northeast corner forming a pattern with high resemblance to that of the total number of acyl groups. However, the pattern for the relative amount of x-3’s of all the fatty acids is quite different and the lowest relative concentration coincides remarkably well with MCI in the southeast corner of the map. These results suggest that a low relative amount of x-3 would be more indicative of high risk for AD than low serum x-3 or polyunsaturated fatty acid (PUFA) concentration as such . It has recently been suggested that serum sphingomyelin (SM) may be a good pre-clinical predictor of memory impairment . Here, SM and phosphatidylcholine (PC) behaved similarly to TC; the lowest values of SM and PC coinciding with those of TC and the southeast area thereby having severe overlap with MCI. See also the Supplementary Figure for the component planes for the amyloid-b 42 to 40 ratio and for more data on the neuropsychological tests. As indicated by the LMWM window, many of the smaller serum solutes, e.g., valine and lactate are elevated in the southeast corner of the SOM where the prevalence of MCI is the highest. This observation is in line with the ﬁnding of abnormal serum amino acid levels in individuals with MCI and AD . In accordance with previous ﬁndings , elevated lactate associates with diabetes and obesity, and, thereby also with MCI. Here, elevated 3-hydroxybutyrate also partly coincides with MCI, suggesting association with ketogenesis. Interestingly, the association of acetate and MCI is quite the opposite. There seems to be no previous ﬁndings on the potential link between low serum acetate and cognitive decline. Nevertheless, acetate, in the form of acetyl-coenzyme A, is linked to several metabolites, for example, cholesterol and acetylcholine, known to be related to AD [7,19]. The highest serum glucose levels overlap only partially with diagnosed diabetes. This is likely to relate to undiagnosed diabetes for some of the individuals situated in the eastern edge of the SOM in Fig. 3. Taking this into account in the global consideration will enhance the association of diabetes with MCI (as well as obesity and MetS). The pattern of glycoproteins in Fig. 3 is similar to that of TG, indicating that their elevation is related to MCI, MetS, and diabetes. Serum inﬂammatory proteins may relate to cognitive decline in older persons . Thus, it is likely that the elderly people with both inﬂammation and MetS are most susceptible to cognitive decline [21,22]. In conclusion, the experimental three-molecular-window 1H NMR approach together with the SOM analysis appears a generally useful methodology to characterise systemic metabolic phenotypes with respect to clinical diagnostics and molecular biochemistry. The application in cognitive impairment suggests a distinct role of systemic lipid metabolism, particularly the metabolic syndrome and the relative amount of serum x-3 fatty acids, in the risk
assessment of Alzheimer’s disease. The results from this pilot study also justify more extensive applications of serum NMR metabonomics in neurological disorders. Acknowledgments This work was supported by the Academy of Finland Centre of Excellence program (KK&MAK), the National Graduate School of Organic Chemistry and Chemical Biology (TTy) and the Jenny and Antti Wihuri Foundation (VPM). Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.bbrc.2008.08.007. References  M. Ala-Korpela, Potential role of body ﬂuid 1H NMR metabonomics as a prognostic and diagnostic tool, Expert Rev. Mol. Diagn. 7 (2007) 761–773.  M. Ala-Korpela, Critical evaluation of 1H NMR metabonomics of serum as a methodology for disease risk assessment and diagnostics, Clin. Chem. Lab. Med. 46 (2008) 27–42.  V.-P. Mäkinen, P. Soininen, C. Forsblom, M. Parkkonen, P. Ingman, K. Kaski, P.H. Groop, M. Ala-Korpela, 1H NMR metabonomics approach to the disease continuum of diabetic complications and premature death, Mol. Syst. Biol. 4 (2008) 167.  E. Holmes, R.L. Loo, J. Stamler, M. Bictash, I.K. Yap, Q. Chan, T. Ebbels, M. De Iorio, I.J. Brown, K.A. Veselkov, M.L. Daviglus, H. Kesteloot, H. Ueshima, L. Zhao, J.K. Nicholson, P. Elliott, Human metabolic phenotype diversity and its association with diet and blood pressure, Nature 453 (2008) 396–400.  J.L. Grifﬁn, A. Vidal-Puig, Current challenges in metabolomics for diabetes research: a vital functional genomic tool or just a ploy for gaining funding?, Physiol Genomics 34 (2008) 1–5.  S.K. Herukka, S. Helisalmi, M. Hallikainen, S. Tervo, H. Soininen, T. Pirttilä, CSF Abeta42, Tau and phosphorylated Tau, APOE epsilon4 allele and MCI type in progressive MCI, Neurobiol. Aging 28 (2007) 507–514.  L.A. Shobab, G.Y. Hsiung, H.H. Feldman, Cholesterol in Alzheimer’s disease, Lancet Neurol. 4 (2005) 841–852.  M. Kivipelto, A. Solomon, Cholesterol as a risk factor for Alzheimer’s disease— epidemiological evidence, Acta Neurol. Scand. Suppl. 185 (2006) 50–57.  T. Hänninen, M. Hallikainen, S. Tuomainen, M. Vanhanen, H. Soininen, Prevalence of mild cognitive impairment: a population-based study in elderly subjects, Acta Neurol. Scand. 106 (2002) 148–154.  J.S. Folch, M. Lees, H.S. Stanley, A simple method for the isolation and puriﬁcation of total lipids from animal tissues, J. Biol. Chem. 226 (1957) 497– 509.  R.K. Adosraku, G.T. Choi, V. Constantinou-Kokotos, M.M. Anderson, W.A. Gibbons, NMR lipid proﬁles of cells, tissues, and body ﬂuids: proton NMR analysis of human erythrocyte lipids, J. Lipid Res. 35 (1994) 1925–1931.  S. Mierisova, M. Ala-Korpela, MR spectroscopy quantitation: a review of frequency domain methods, NMR Biomed. 14 (2001) 247–259.  P. Soininen, J. Haarala, J. Vepsäläinen, M. Niemitz, R. Laatikainen, Strategies for organic impurity quantiﬁcation by 1H NMR spectroscopy: constrained totalline-shape ﬁtting, Anal. Chim. Acta 542 (2005) 178–185.  A. Vehtari, V.-P. Mäkinen, P. Soininen, P. Ingman, S.M. Mäkelä, M.J. Savolainen, M.L. Hannuksela, K. Kaski, M. Ala-Korpela, A novel Bayesian approach to quantify clinical variables and to determine their spectroscopic counterparts in 1H NMR metabonomic data, BMC Bioinform. 8S2 (2007) S8.  I.J. Martins, E. Hone, J.K. Foster, S.I. Sünram-Lea, A. Gnjec, S.J. Fuller, D. Nolan, S.E. Gandy, R.N. Martins, Apolipoprotein E, cholesterol metabolism, diabetes, and the convergence of risk factors for Alzheimer’s disease and cardiovascular disease, Mol. Psychiatry 11 (2006) 721–736.  M.M. Mielke, V.V. Bandaru, N.J. Haughey, P.V. Rabins, C.G. Lyketsos, M.C. Carlson, Serum sphingomyelins and ceramides are early predictors of memory impairment, Neurobiol. Aging, in press..  G. Ravaglia, P. Forti, F. Maioli, G. Bianchi, M. Martelli, T. Talerico, L. Servadei, M. Zoli, E. Mariani, Plasma amino acid concentrations in patients with amnestic mild cognitive impairment or Alzheimer disease, Am. J. Clin. Nutr. 80 (2004) 483–488.  T.M.E. Davis, D. Jackson, W.A. Davis, D.G. Bruce, P. Chubb, The relationship between metformin therapy and the fasting plasma lactate in type 2 diabetes: the fremantle diabetes study, Br. J. Clin. Pharmacol. 52 (2001) 137–144.  F. Amenta, S.K. Tayebati, Pathways of acetylcholine synthesis, transport and release as targets for treatment of adult-onset cognitive dysfunction, Curr. Med. Chem. 15 (2008) 488–498.  M.G. Dik, C. Jonker, C.E. Hack, J.H. Smit, H.C. Comijs, P. Eikelenboom, Serum inﬂammatory proteins and cognitive decline in older persons, Neurology 64 (2005) 1371–1377.
T. Tukiainen et al. / Biochemical and Biophysical Research Communications 375 (2008) 356–361  K. Yaffe, A. Kanaya, K. Lindquist, E.M. Simonsick, T. Harris, R.I. Shorr, F.A. Tylavsky, A.B. Newman, The metabolic syndrome, inﬂammation, and risk of cognitive decline, JAMA 292 (2004) 2237–2242.
 G.M. Torri, J. Torri, J.M. Gulian, J. Vion-Dury, P. Viout, P.J. Cozzone, Magnetic resonance spectroscopy of serum and acute-phase proteins revisited: a multiparametric statistical analysis of metabolite variations in inﬂammatory, infectious and miscellaneous diseases, Clin. Chim. Acta 279 (1999) 77–96.