You are here

Article Text

Article menu

Download PDFPDF

Neurodegeneration
Research paper
Diagnosis of dementias by high-field 1H MRS of cerebrospinal fluid
  1. M P Laakso1,2,3,
  2. N M Jukarainen4,
  3. J Vepsäläinen4
  1. 1Department of Neurology, Kuopio University Hospital, Kuopio, Finland
  2. 2The Vanha Vaasa Hospital, Vaasa, Finland
  3. 3The Niuvanniemi Hospital, Kuopio, Finland
  4. 4The School of Pharmacy, Biocenter Kuopio, University of Eastern Finland, Kuopio, Finland
  1. Correspondence to Dr Mikko P Laakso, Mikko Laakso Niuvanniemi Hospital 70240 Kuopio Finland; mikko.laakso{at}uef.fi

Abstract

Objective To test whether the information obtained from cerebrospinal fluid (CSF) and analysed with high-field proton (1H) MR spectroscopy (MRS) would help the diagnosis of most common forms of dementia.

Setting A total of 31 metabolites from CSF from 222 controls and patients suffering from various dementias (Alzheimer’s disease (AD), vascular dementia, Lewy body disease (LBD) and frontotemporal dementia (FTD)) were quantified using 1H MRS.

Main outcome measure Clinical diagnosis.

Results AD was classified with an accuracy of 85.5%. For a group of very early stage patients with AD, the result was significantly higher, 92.3%. Vascular dementia, LBD and FTD were all diagnosed with 100% accuracy in controls and from AD with an accuracy ranging between 85.5% and 93.4%.

Conclusions The results indicate that the composition of CSF contains enough information of the neurological state of a given patient with a given dementia to be diagnosed with extremely high accuracy. This approach might provide potentially a very powerful diagnostic tool to help the diagnostic process of dementias.

https://doi.org/10.1136/jnnp-2015-311171

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Introduction

    All biofluids carry measurable information about the biochemical status of a living organism. These include, but are not limited to, saliva, urine and blood. Cerebrospinal fluid (CSF) in particular provides a window onto the biochemical characteristics of the central nervous system. Therefore, investigation of the composition of CSF might well prove valuable in assessing the normal or pathological metabolic state of the brain. Ideally, such information might assist in clinical, even preclinical diagnostics of various brain disorders, which currently are clinically diagnosed at their relatively advanced stage.

    MR spectroscopy (MRS; sometimes referred as to nuclear MR (NMR) method) allows harvesting of small molecule metabolite concentrations easily such as without complex sample treatment and instrumental calibration. Even though there have been some attempts based on a straightforward diagnostic analysis of metabolites by MRS,1–4 currently a metabolic approach for Alzheimer’s disease (AD) and other dementing disorders is virtually non-existent. In the present study, MRS was used to analyse the CSF of patients with the four most common forms of dementia: AD, vascular dementia (VaD), Lewy body disease (LBD) and frontotemporal dementia (FTD). Our aim was to use the information derived from the CSF metabolites to diagnose these different forms of dementia.

    Methods

    Participants

    Our clinic has focused on memory disorders since the late 1970s. We practically recruit all the patients who come to our memory clinic to studies such as this. In addition, we have access to data from various epidemiological studies such as the WHO FINMONICA study. The control group of 45 patients consisted of individuals examined at the Kuopio University Hospital for various neuropsychiatric symptoms, such as depression or headache, but who did not have cognitive impairment as determined by comprehensive neuropsychological testing. These recruited controls were further divided into two subgroups: patients with an AD marker (β-amyloid42 and/or τ/phosphotau protein) profile present in CSF (controls with AD pathology in CSF: C/ADP) and patients who did not have an AD marker profile in CSF (normal controls: C/NRM). The C/ADP group consisted of 11 individuals and the C/NRM group consisted of 34 individuals. The C/NRM group without AD pathology was used in all the further analyses. The groups with memory impairment were recruited after having been referred to examinations at the Kuopio University Hospital of the University of Kuopio. The patient groups included consisted of 76 patients with probable AD, 26 patients with an early stage AD (EAD), 33 patients with mild cognitive impairment (MCI), 16 patients with VaD, 10 patients with LBD and 16 patients with FTD. We had the change to follow-up 68 of the participants to verify or alter their diagnostic status (1–14 years, mean of 5±3 years). The diagnosis of probable AD was made according to the NINCDS-ADRDA criteria.5 Individuals with MCI were diagnosed based on criteria presented by the Mayo Clinic AD Research Center.6 The description of our EAD group deserves some elaboration. It can be considered to be a valuable group because we had an opportunity to follow-up individuals with MCI and the development of their cognitive state for quite some time (up to 168 months). The concept of MCI provides an economical manner in which to observe potential conversion to dementia bearing in mind that individuals with MCI are some 10 times more likely to develop a dementia than their cognitively normal peers.6 For our study, we managed to isolate 22 individuals who converted to AD during the follow-up and EAD refers to these converters. VaD was diagnosed according to the NINDS-AIREN criteria,7 FTD was diagnosed by using the Lund-Manchester criteria,8 and LBD was diagnosed according to the consensus criteria for LBD9 (figure 1).

    Figure 1
    Figure 1

    A flow chart of patient requirement.

    This study was approved by the ethics committee of the Kuopio University Hospital, and informed consent for participation was obtained from all the participants and the caregivers of the patients.

    Sample preparation

    Sample preparations and MRS measurements were performed as described previously.4 Lumbar CSF samples were stored at −70°C until use. Then 1800 µL of each sample was subjected to an identical lyophilisation protocol for 40 h. The freeze-dried samples were then stored at −20°C in sealed vials until analysis. Prior to the measurements, the samples were reconstituted in 600 μL of D2O (99.98%-D, Merck) and 450 μL of this liquid was transferred to a separate vial followed by addition of 50 μL of 21.5 mM TSP-d4 in D2O to be used as an internal standard of known concentration. The pH of the samples was not adjusted, being typically around 7.00±0.05. This pH can be defined as pH*, which is the reading of the pH metre as measured with a standard pH electrode. The pD value is ca. 0.4 units higher than pH*.

    NMR spectroscopy and quantification

    The metabolic profiling was based on a standard 1D 1H NMR spectrum. All spectra were measured by using a Bruker AVANCE DRX 500 instrument operating at 11.4 T (500.13 MHz) (Bruker-Biospin GmbH, Karlsruhe, Germany), equipped with a quadronuclear probe. The Bruker XWIN-NMR software V.3.5pl5 running on a standard PC was used for acquisition of all spectra. The relevant parameters used in the 1D experiments are as follows: recycling delay 45 s, acquisition time 6.5 s, number of scans 128, and a sweep width of 9.5 ppm. A calibrated 90° pulse was used for all spectra and all acquisitions were performed on non-spinning samples. To assess the use of relaxation editing in spectral simplification, T2 edited 1D NMR spectra were measured. For the three spectra (not edited, minimally and heavily edited) measured, a standard 1D Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence with a 40 or 320 ms (for minimally edited and fully edited, respectively) T2-filter using a fixed echo delay of 400 μs that eliminates diffusion and J-modulation effects was used. The quantitative descriptors representing the metabolite concentrations were obtained from the spectra using the same 31 metabolites and signal areas as previously reported.4 All spectral operations were performed using PERCH NMR Software V.2008.1 (PERCH Solutions Ltd, Kuopio, Finland) (figure 2).

    Figure 2
    Figure 2

    A presentation of a 1H nuclear MR spectrum of hCSF at 11.4 T (500 MHz). (A) Aromatic region: h=histidine, f=phenylalanine, w=tryptophan + other metabolites underlying its aromatic signal, y=tyrosine, fm=formate, o=others, X1 and X2=unknown aromatic metabolites. (B) Middle region: la=lactate, cre=creatine, crn=creatinine, glc=glucose (α protons not shown), mi=myo-inositol, A1 through A3=metabolite areas 1–3, X3=unknown metabolite. (C) Lower field aliphatic region: e=glutamate, ac=acetate, py=pyruvate, ci=citrate, A4=several overlapping metabolites, X4=unknown metabolite. (D) Higher field aliphatic region, with metabolite markings as follows: la=lactate, ahi=α-hydroxyisovalerate, ahb=α-hydroxybutyrate, bhb=β-hydroxybutyrate, A5 through A7=metabolite areas 5–7, X5 through X7=unknown metabolites. Metabolite areas and unknown signals: X1 and X2=unknown aromatic signals, A1=several overlapping amino acids, A2=metabolites in the 3.70–3.56 ppm range, not including myo-inositol, A3=A hump of signals (3.37–3.34 ppm), X3=a triplet + underlying signals at 3.16 ppm, A4=several overlapping metabolites in the range of 2.37–2.24 ppm, X4=unknown signal at 2.13 ppm, A5=the spectral region 1.56–1.40 ppm, A6=the spectral region 1.32–126 ppm, X5=unknown doublet at 1.13 ppm, X6=unknown doublet at 1.07 ppm, X7=unknown doublet at 1.03 ppm, A7=the spectral region 1.02–0.93 ppm.

    Statistical analyses

    An SPSS software, V.21.0.0.0, was used for data analyses with the default settings. To provide an overview of differences between the study groups, a two-way ANOVA with Tukey’s post hoc analysis was performed. A χ2 test with two-tailed significance was performed to study differences in sex distribution. To study the classification, a discriminant function analysis was performed first stepwise, then with all the available metabolites entering the model.

    Results

    Group demographics

    The control patients were significantly younger than the patient groups, save for FTD (F=6.8, p<0.05). The patients with FTD were significantly younger than all the other patients (F=61.7, p<0.001). This is not unexpected because it is typical for the patients with FTD to have an earlier onset of dementia.

    Controls whose CSF was free from AD pathology were used in the analyses. The groups were comparable for sex, except for AD and VaD. The AD group contained significantly more women than men (which is typical for AD) (χ2=4.6, p<0.05) and, conversely, the VaD patient group contained significantly more men than women (which is typical for VaD) (χ2=5.3, p<0.05; table 1).

    View this table:
    Table 1

    Demographics (±SD) of study participants displaying group sizes, sex and MMSE score

    Classification of the groups

    The assigned spectral models included 17 known metabolites, 7 unknown metabolites and 7 areas with several overlapping signals (for complete list of metabolites, please see our previous study where we verified the accuracy of the method).4

    The classification results are presented in table 2. They range from very reasonable accuracy to several cases of perfect classification. We tried to classify patients in two ways, with a stepwise analysis and with an analysis where all the available metabolites entered the model. The latter produced better classification. Those results are used in the rest of the work. However, in table 2, we also present results from the stepwise analysis to present the metabolites contributing to the diagnosis.

    View this table:
    Table 2

    Classification results

    Discussion

    Our results show that the CSF metabolite profiles can be used to discriminate different types of dementia. Individual analyses yielded excellent diagnostic classification: 100% correct for all dementia groups, save for AD, which was classified correctly in 85.5% of cases at the stage of probable AD and, better yet, at 92.3% accuracy in the preclinical state.

    In spite of the long history of MRS of CSF,10–12 to the best of our knowledge, this is only the second study to have used 1H MRS of CSF to study different types of dementia, and the first study to study other forms of dementia than AD and VaD. The first study on the subject, in general, stems back to 1996, describing that it is possible to study metabolic alterations in dementia. The study, however, used analytic methods not used in this study; the concept of dementia was vaguely defined, generic and didn’t even try to classify the groups.11 Even with the passage of time, despite the potential of the approach, the area remains neglected. Only 8 years later, 1H MRS of CSF was used to study dementia again. This time, both AD and VaD were studied. The authors found an unknown peak not only in AD, but also in the same spot in VaD, thus providing no help in distinguishing AD from VaD. Their unknown peak, coined as a ‘dementia associated factor’, was located at 3.16 ppm.12 Interestingly, we find a similar tiny, unknown peak, ‘X3’, at 3.16 ppm. Only in our study, X3 contributes to the distinction between AD and VaD but not with controls versus patients with AD, nor with patients with VaD.

    However, to get back to this study, while it is apparent that every given diagnostic group presented with unique differences in their metabolite concentrations, even just controls, with or without AD pathology, have their own distinct metabolic fingerprints. The same thing with stable MCI versus converting MCI (EAD), where the spectra differ enough to result in excellent overall diagnostic accuracy. There was a minor overlap between the groups but it is not unexpected. The metabolite variations are as complex as the disease processes. Neuropathological studies have shown that patients with AD often have heterogeneous brain pathology and a considerable number of patients also exhibit brain changes indicating the presence of other concomitant neurodegenerative disorders.13–16 This is likely to explain the least accurate diagnostic accuracy of patients with AD in this study: AD pathology is common in the elderly regardless of whether they have the disease or not and, after the onset of clinical AD, multiple brain systems are involved, probably complicating the achievement of a more accurate diagnosis.13–16 From this point of view, it is extremely interesting to notice that before the course of AD has proceeded to multiple system pathology, as we expect the case to be in our relatively unique sample of patients with very early EAD, the classification was at least a degree better than in those with more advanced disease. In these patients with EAD, we assume the AD pathology to be substantially in more ‘pure’ AD than in an advanced AD. Therefore, the method we present here may be an extremely valuable tool to reveal very early AD, perhaps even long before the clinical diagnosis can be established. The implications of identifying patients with preclinical AD are manifold with obviously the most important reason, early, even preclinical AD being probably the best time window for intervention.

    As far as the rare dementias are considered, there is hardly anything to discuss. Each non-AD form of dementia—VaD, LBD and FTD—was diagnosed from controls with 100% accuracy. Non-AD dementias were diagnosed from AD with accuracies ranging up from 85.5%.

    The final issue we wish to address is the issue a challenge for future. Provided that these results are replicated and established as valid, how could this approach be converted into clinical practice to study individual patients presenting with memory disorders? There is nothing in this approach that would render it inapplicable in various tertiary care centres. One needs teaching material (CSF), and MRS hardware is relatively widely available and, suffice it to say, highly competitive costwise. Indeed, in terms of availability and cost, this method perhaps leaves behind many analytic methods such as positron emission tomography. In terms of neurobiology of disease, this approach outperforms many other approaches, such as CSF τ or amyloid, by harvesting much more clinical information than tests which need antibodies or other studies which reveal information from only one variable. In fact, depending on the hardware and software, harvesting of 31 metabolites may be but a minimum. The number can quite easily be increased. Also other probes than proton, such as phosphorus, can be used to scan the data to get more profound picture of the disease with the use of the very same CSF sample. In this work, we cannot make an in-depth analysis of individual metabolites. This is dictated by necessity. The information obtained from 31 metabolites in eight different groups is huge, 248 figures, to be exact, and this is something which cannot possibly be dealt with here in any significant manner, no matter how interesting the underlying metabolites. We, thus, have to settle the focus this work into the diagnostic aspects. Moreover, when one looks at the data contributing to the diagnoses, a majority of the metabolites are unknown, representing signals from A1-A7 and X3-X7. This makes an in-depth analysis vain. Thus, when interpreting the data, we believe it is safe to simply conclude that there are signals from 11 metabolites contributing to the diagnosis which just should not be there.

    One of the strengths of this study is the multitude of groups. Starting with controls with and without AD pathology in the CSF; continuing to patients with MCI who remain the same or convert to AD; and down to the availability of non-AD dementias in this study. Moreover, the study participants are invaluble, most of whom we were able to follow-up for quite some time to establish the diagnoses properly, and to follow people turning to AD from MCI. As far as shortcomings are concerned, they seem to be relatively few. One statistical obstacle worth mentioning is a potential limitation which theoretically may influence the results. That is overfitting. Overfitting is a statistical problem which may occur when a model is excessively complex, such as having many parameters relative to the number of observations. In the case of overfitting, a statistical model describes a random error or noise instead of an underlying relationship. However, a model which has been overfit generally results into poor performance as it exaggerates minor fluctuations in data. Instead, in this study, the diagnostic performance was rather far from being poor.

    The objective of this study was the development of a test for CSF to be used in the diagnosis of the most common forms of dementia. We feel that this objective was satisfactorily met. On the basis of our results, we can conclude that the large differences of the metabolite profiles in CSF, hidden under complex relationships of the partly unidentified components, indeed reflect the real significant variance of the neurological states in the patients under scrutiny. As a conclusion, we believe that the profiling of dementias by using 1H MRS of CSF might offer an invaluable means in the diagnostic workup of the dementias, not to mention to offer a platform for further studies to understand and probe neural system biology and metabolomic intricacies of any given form of dementia.

    Acknowledgments

    The authors recognise the effort of the clinicians of the Kuopio University Hospital and University of Kuopio in recruiting and performing the clinical examinations of the study participants. The authors recognise the help of the laboratory staff in the handling of the specimens. The authors also acknowledge the input of Professor Laatikainen to be extraordinary. The authors also recognise the kindness and willingness of the patients to be examined as study participants, and without whom none of this would have been possible.

    References

    1. ↵
      1. Holmes E,
      2. Tsang TM,
      3. Tabrizi SJ
      . The application of NMR-based metabonomics in neurological disorders. NeuroRx 2006;3:358–72. doi:10.1016/j.nurx.2006.05.004
      OpenUrlCrossRefPubMed
    2. ↵
      1. Holmes E,
      2. Tsang TM,
      3. Huang JTJ, et al
      . Metabolic profiling of CSF: evidence that early intervention may impact on disease progression and outcome in schizophrenia. PLoS Med 2006;3:e327. doi:10.1371/journal.pmed.0030327
      OpenUrlCrossRefPubMed
    3. ↵
      1. Weiss U,
      2. Bacher R,
      3. Vonbank H, et al
      . Cognitive impairment: assessment with brain magnetic resonance imaging and proton magnetic resonance spectroscopy. J Clin Psychiatry 2003;64:235–42. doi:10.4088/JCP.v64n0303
      OpenUrlCrossRefPubMedWeb of Science
    4. ↵
      1. Jukarainen NM,
      2. Korhonen SP,
      3. Laakso MP, et al
      . Quantification of 1H NMR spectra of human cerebrospinal fluid: a protocol based on constrained total-line-shape analysis. Metabolomics 2008;4:150–60. doi:10.1007/s11306-008-0106-6
      OpenUrlCrossRef
    5. ↵
      1. McKhann G,
      2. Drachman D,
      3. Folstein MF, et al
      . Clinical diagnosis of Alzheimer's disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's disease. Neurology 1984;34:939–44. doi:10.1212/WNL.34.7.939
      OpenUrlCrossRefPubMed
    6. ↵
      1. Petersen RC,
      2. Smith GE,
      3. Waring SC, et al
      . Aging, memory, and mild cognitive impairment. Int Psychogeriatr 1997;9(Suppl S1):65–9. doi:10.1017/S1041610297004717
      OpenUrlPubMed
    7. ↵
      1. Román GC,
      2. Tatemichi TK,
      3. Erkinjuntti T, et al
      . Vascular dementia: diagnostic criteria for research studies. Report of the NINDS-AIREN International Workshop. Neurology 1993;43:250–60. doi:10.1212/WNL.43.2.250
      OpenUrlCrossRefPubMed
    8. ↵
      1. Neary D,
      2. Snowden JS,
      3. Gustafson L, et al
      . Frontotemporal lobar degeneration: a consensus on clinical diagnostic criteria. Neurology 1998;51:1546–54. doi:10.1212/WNL.51.6.1546
      OpenUrlCrossRefPubMed
    9. ↵
      1. McKeith IG,
      2. Galasco D,
      3. Kosaka K, et al
      . Consensus guidelines for the clinical and pathologic diagnosis of dementia with Lewy bodies (DLB). Report of the consortium on DLB international workshop. Neurology 1996;47:1113–24. http://dx.doi.org/10.1212/WNL.47.5.1113
      OpenUrlCrossRefPubMed
    10. ↵
      1. Bell JD,
      2. Brown JCC,
      3. Sadler PJ, et al
      . High-resolution proton nuclear magnetic resonance studies of human cerebrospinal fluid. Clin Sci 1987;72:563–70.
      OpenUrlCrossRefPubMed
    11. ↵
      1. Nicoli F,
      2. Vion-Dury J,
      3. Confort-Gouny S, et al
      . Cerebrospinal fluid metabolic profiles in multiple sclerosis and degenerative dementias obtained by high resolution proton magnetic resonance spectroscopy. C R Acad Sci III 1996;319:623–31.
      OpenUrlPubMed
    12. ↵
      1. White LR,
      2. Gårseth M,
      3. Aasly J, et al
      . Cerebrospinal fluid from patients with dementia contains increased amount of an unknown factor. J Neurosci Res 2004;87:297–301. doi:10.1002/jnr.20241
      OpenUrl
    13. ↵
      1. Petroff OAC,
      2. Yu RK,
      3. Ogino T
      . High-resolution proton magnetic resonance analysis of human cerebropsinal fluid. J Neurochem 1986;47:1270–6. doi:10.1111/j.1471-4159.1986.tb00750.x
      OpenUrlCrossRefPubMedWeb of Science
    14. ↵
      1. Lopez OL,
      2. Becker JT,
      3. Kaufer DI, et al
      . Research evaluation and prospective diagnosis of dementia with Lewy bodies. Arch Neurol 2002;59:43–6. doi:10.1001/archneur.59.1.43
      OpenUrlCrossRefPubMedWeb of Science
    15. ↵
      1. Schneider JA,
      2. Arvanitakis Z,
      3. Bang W, et al
      . Mixed brain pathologies account for most dementia cases in community-dwelling older persons. Neurology 2007;69:2197–204. doi:10.1212/01.wnl.0000271090.28148.24
      OpenUrlCrossRefPubMed
    16. ↵
      1. Price JL,
      2. McKeel DW Jr.,
      3. Buckles VD, et al
      . Neuropathology of nondemented aging: presumptive evidence for preclinical Alzheimer disease. Neurobiol Aging 2009;30:1026–36. doi:10.1016/j.neurobiolaging.2009.04.002
      OpenUrlCrossRefPubMedWeb of Science

    Footnotes

    • Correction notice The results in the abstract have been updated since this paper was first published online.

    • Contributors MPL designed the study and primarily drafted the manuscript. The analyses were carried out by NMJ and JV. They drafted the manuscript from their part of the study and approved the final version of the manuscript. MPL had full access to all the data in the study, is the guarantor of the study and takes responsibility for the integrity of the study, the data and the accuracy of the data analysis.

    • Funding This work was financed by EVO grants (special government subsidies) to the Kuopio University Hospital, to the University of Kuopio, to the Vanha Vaasa Hospital, to the Niuvanniemi Hospital; and by a grant by the Maire Taponen Foundation (MPL).

    • Competing interests None declared.

    • Ethics approval This study was approved by the ethics committee of the Kuopio University Hospital, and informed consent for participation was obtained from all the participants and the caregivers of the patients.

    • Provenance and peer review Not commissioned; internally peer reviewed.

    Linked Articles