Measurement of Phosphorylated Tau Epitopes in the Differential Diagnosisof Alzheimer Disease: Title and subTitle BreakA Comparative Cerebrospinal Fluid Study FREE

Abnormal hyperphosphorylation of the microtubule-associated proteintau and its incorporation into neurofibrillary tangles are major componentsof the pathogenesis of Alzheimer disease (AD).

Using monoclonal antibodies specific for different phosphorylated epitopesof tau, enzyme-linked immunosorbent assays have been developed that sensitivelymeasure concentrations of phosphorylated tau protein (p-tau) in cerebrospinalfluid (CSF). Statistically significant increases in CSF p-tau concentrationsin patients with AD have recently been demonstrated in independent pilot studies,mainly using 3 different immunoassays specific for the phosphorylated epitopesthreonine 231 (p-tau₂₃₁),^{1 - 2} threonine181 (p-tau₁₈₁),^{3 - 4} andserine 199 (p-tau₁₉₉).⁵ Evidencefrom these pilot studies indicates that quantification of tau phosphorylatedat these specific sites may improve early detection, differential diagnosis,and tracking of disease progression in AD (for a review see Blennow et al^{6 - 7} ).

In a recent study,¹ CSF p-tau₂₃₁ distinguished between patients with AD and those with other neurologicdisorders (ONDs) with a sensitivity of 85% and a specificity of 97%. Furthermore,p-tau₂₃₁ significantly improved differential diagnosis betweenAD and other non-AD groups, particularly frontotemporal dementia (FTD).⁸ In AD vs FTD, p-tau₂₃₁ correctly allocated91% of patients vs only 66% using total tau.⁸ Itohand colleagues⁵ showed that CSF p-tau₁₉₉ discriminates between AD and the combined non-AD groups with a sensitivityand a specificity of 85%. The level of CSF p-tau₁₈₁ was elevatedin patients with AD compared with patients with other dementias and controls,⁹ and p-tau₁₈₁has been proposed as a potentialmarker for discriminating patients with AD from those with dementia with Lewybodies (DLB).¹⁰ Furthermore, CSF p-tau₂₃₁ concentrations declined over time during the clinical progressionof AD and correlated with cognitive performance at baseline.¹¹ Resultsof these first studies suggested that CSF p-tau proteins are promising biomarkercandidates for AD.

Levels of CSF p-tau and the discriminative power of the 3 differentp-tau assays among patients with AD, nondemented controls, and patients withother dementias, however, have not yet been assessed in the same set of patients.Investigation of the 3 p-tau assays would allow for direct comparison of thediagnostic performance of the 3 markers to differentiate AD from other relevantdiseases. In the advent of large international multicenter trials (such asthe National Institute on Aging Initiative on Neuroimaging in Alzheimer'sDisease and the studies of the European Alzheimer's Disease Consortium) onthe value of biological markers and neuroimaging in the diagnosis of AD, thisis the first study to address the important clinical issue regarding the differentialdiagnostic performances of p-tau proteins as core biological marker candidates.

In particular, we tested the diagnostic accuracy of the 3 p-tau assaysaccording to the recommendations of a consensus group¹² suggestinga sensitivity level of 85% or greater and a specificity level of at least75% for useful biomarkers of AD. In addition, we asked whether use of a combinationof the 3 markers might be superior to the application of single markers. Toour knowledge, this is the first comparative study applying 3 developed andpublished immunoassays detecting different p-tau epitopes on the same setof controls and patients to compare individual and combined diagnostic accuracy.Concentrations of p-tau₂₃₁, p-tau₁₈₁, and p-tau₁₉₉ in the CSF were studied in the same group of patients with AD, DLB,FTD, or vascular dementia (VaD); patients with ONDs; and controls.

A total of 206 individuals were studied. One hundred eight patientshad probable AD (National Institute of Neurological and Communicative Disordersand Stroke–Alzheimer's Disease and Related Disorders Association criteria)¹³ and 53 had other dementia disorders (24 patientswith FTD,¹⁴ 22 with DLB,¹⁵ and7 with VaD).¹⁶ Structural and functional imagingresults were consistent with the diagnoses in these patients. Twenty-two patientswith ONDs were diagnosed as having mild psychiatric (eg, depressed mood) orneurologic (eg, dizziness) symptoms. One patient with OND experienced a bulbarsyndrome of unknown etiology. We also studied 23 controls. Study participantswere recruited at 2 academic expert centers: the Dementia Research Sectionand Memory Clinic, Alzheimer Memorial Center and Geriatric Psychiatry Branch,Department of Psychiatry, Ludwig-Maximilian University (22 patients with AD,6 with FTD, 7 with VaD, 9 with DLB, 1 with OND, and 13 controls), and theDepartment of Clinical Neuroscience, University of Göteborg (86 patientswith AD, 18 with FTD, 13 with DLB, 21 with OND, and 10 controls). Participantsfrom the former center had been studied previously with a different objective,and the results of this study have been published.⁸ Characteristicsof the patients and controls are given in Table 1. The protocol was approved by the local ethical committeesand the institutional review boards of the 2 participating medical centers.Informed consent was obtained from all participants.

Examination of the controls included medical history, physical examination,routine blood tests (blood cell count; international normalized ratio; partialthromboplastin time; and sodium, potassium, creatinine, urea, and blood glucoselevels), and a cognitive test using the Consortium to Establish a Registryfor Alzheimer's Disease battery.¹⁷ Ten of the23 controls were volunteers without any medical, neurologic, or psychiatricdisorders. Samples of CSF were collected from 13 controls while they underwentspinal anesthesia for surgery of the urinary tract or lower extremities. Theywere cognitively normal according to the Consortium to Establish a Registryfor Alzheimer's Disease battery (results within ±1 SD in all subtests).Three of these control subjects had diabetes mellitus as a somatic comorbidity.

Samples of CSF were acquired via lumbar puncture between 9 and 11 AM according to a routine protocol (established by the 2 participatingmedical centers). Samples of CSF were collected in polypropylene tubes onice in 0.5-mL aliquots. For this study, a total of 1 mL was taken. Aliquotswere centrifuged at 4°C at 10 000g for 10minutes and stored at −80°C until analysis. The same procedureswere performed at the 2 sites involved in the study. There was no effect ofmedical centers on the variance of measured protein levels and no proteingradient in the CSF column for the markers (K.B., unpublished data, 2002).

Assay operators were masked to the diagnostic category of the samples.Levels of p-tau₂₃₁ were measured using an enzyme-linked immunosorbentassay (Applied NeuroSolutions Inc).¹ This assayuses a combination of CP27 (which recognizes amino acids 130-150 in normaltau and p-tau), Tau-1 (which recognizes amino acids 196-205 in nonphosphorylatedtau), and CP9 (which recognizes phosphothreonine 231). Experiments describingthe specificity of the detection antibody CP9 for phosphothreonine 231 havebeen reported previously.¹ Full-length recombinanttau (441 amino acids) phosphorylated at threonine 231 was used to producea standard curve. Levels of p-tau₂₃₁ in our patients were calculatedfrom the standard curve and expressed as CSF p-tau₂₃₁ in picogramsper milliliter.

Levels of p-tau₁₈₁ were measured using a sandwich enzyme-linkedimmunosorbent assay method (prototype version of Innotest Phospho-Tau [181p];Innogenetics), using a combination of monoclonal antibody HT7 (which recognizesamino acids 159-163 in normal tau and p-tau) and biotinylated monoclonal antibodyAT270 (which recognizes p-tau containing the phosphorylated threonine 181residue).³ A synthetic phosphopeptide was usedfor standardization.

Levels of p-tau₁₉₉ were measured using a previously reportedsandwich enzyme-linked immunosorbent assay method (Mitsubishi Chemical Corp,Shinagawa, Japan), using a combination of monoclonal antibody HT7 and polyclonalantibody anti-PS199 (specific for tau phosphorylated at serine 199).⁵

Differences among groups regarding age were assessed using the Mann-Whitney(M-W) test and regarding sex distribution using the χ² test.

Distributions of p-tau values differed statistically significantly fromnormal as revealed by the Kolmogorov-Smirnov test. Differences in mean CSFlevels of the 3 p-tau subtypes among all groups were assessed usingthe Kruskal-Wallis test. Pairwise comparisons between patients with AD andthe other groups were performed using the M-W test. Correlations between p-tausubtypes were assessed using the Spearman rank correlation.

Cutoff values for p-tau proteins were determined such that 85% of thepatients with AD were correctly identified according to the recommendationsof the consensus conference¹² of 85% sensitivityfor an "excellent" biomarker. To determine differences in diagnostic accuracyamong markers, the specificity levels that correspond to the 85% sensitivitycutoff level for each marker in the comparison groups were compared betweenall possible pairs of markers using the McNemar test.

To develop predictive cutoff values that optimized the combined useof the different p-tau markers, we used classification tree analysis withSYSTAT 7.0 (SPSS Inc, Chicago, Ill). Classification tree analysis uses recursivepartitioning to consider all possible binary splits of the data in pursuitof optimal classification.¹⁸ The analysis consideredall 3 p-tau markers simultaneously to create cutoff values that maximize separationamong groups, resulting in a decision tree. The number of branches dependson the separation that has been achieved at the first split. To avoid overfittingof the data, generation of further branches was interrupted if sensitivityor specificity declined below 80%. A similar technique has been presentedin earlier CSF studies in AD.¹⁹

To show sensitivity and specificity levels over the entire range ofcutoff levels, we determined receiver operating characteristic curves.

As illustrated in Figure 1 and Table 1, levels of all p-tau subtypeswere significantly increased in patients with AD compared with the other groupsstudied. Because the controls were significantly younger than the patientswith AD, we investigated correlations between p-tau subtypes and age. In patientswith AD, there was no correlation between the 3 p-tau subtypes and age (ρ= –0.050 to 0.002; P = .61-.93). In controls,p-tau₁₈₁ (ρ = 0.61; P = .002) andp-tau₂₃₁ (ρ = 0.41; P = .05) correlatedwith age, but p-tau₁₉₉ did not (ρ = 0.05; P = .80). Therefore, we repeated the analyses in a group of 23 patientswith AD and 23 controls who were matched for age (M-W₁ = 208; P = .21) and sex (χ²₁<0.001; P>.99). Differences between the AD group and the controlgroup remained unchanged and statistically significant for all3 p-tau subtypes (p-tau₂₃₁: M-W₁ = 32; P<.001; p-tau₁₈₁: M-W₁ = 52.5; P<.001; p-tau₁₉₉: M-W₁ = 46; P<.001). Consequently, we included all of the patients with AD inour analyses. Correlations among the 3 p-tau subtypes are given in Table 2.

Specificity levels of single markers using p-tau proteins when sensitivitywas set at 85% or higher, as recommended by a consensus report, are givenin Table 3.¹² Inthe case of p-tau₁₉₉, several individuals in each comparisongroup had a value equal to the 85% sensitivity cutoff value. Therefore, forp-tau₁₉₉, a lower and an upper limit is given for the specificity,corresponding to 2 alternatives: (1) all patients with this specific valuewere allocated to the AD group and (2) all patients with this specific valuewere allocated to the comparison group. In the differentiation of patientswith AD from those in the non-AD group, p-tau₂₃₁ (85%) and p-tau₁₈₁ (81%) reached specificity levels of 75% or higher, but p-tau₁₉₉ did not (61%-72%). All 3 p-tau proteins showed excellent specificitylevels when patients with AD were compared with those with ONDs and controls.

Comparing AD to FTD, p-tau₂₃₁and p-tau₁₈₁ showedgood discriminative power, with specificity levels of 92% and 79%, respectively.For p-tau₁₉₉, specificity ranged from 42% to 54% in the discriminationof AD from FTD. When AD was compared with DLB, p-tau₁₈₁ showeda specificity of 68%, whereas for p-tau₂₃₁ the specificity was64%. For p-tau₁₉₉, specificity ranged between 50% and 64%. Valuesfor AD vs VaD are not separately given owing to the small sample size of patientswith VaD (n = 7).

We tested for differences in diagnostic accuracy among p-tau proteins(Table 3). Discriminative powerbetween AD and the combined non-AD groups was significantly higher for p-tau₂₃₁ (McNemar test P = .004 to P<.001) and p-tau₁₈₁ (McNemar test P = .10 to P<.001) compared with p-tau₁₉₉. Diagnostic accuracy was also significantly higher using p-tau₂₃₁ and p-tau₁₈₁ compared with p-tau₁₉₉ in differentiatingAD from FTD (p-tau₂₃₁ vs p-tau₁₉₉: McNemar test P = .004 to P<.001; p-tau₁₈₁ vs p-tau₁₉₉: McNemar test P =.07 to P = .01). There was no statistically significantdifference in diagnostic accuracy among p-tau proteins for discriminationbetween patients with AD vs ONDs and DLB. Receiver operating characteristiccurves for pairwise comparisons of marker levels between patients with ADand comparison groups are shown in Figure2.

We used classification tree analysis to investigate whether a combinationof p-tau markers would improve group discrimination (Table 4). For AD vs the combined non-AD groups, AD vs controls,AD vs ONDs, and AD vs FTD, we found that p-tau₂₃₁ accounted formaximal group discrimination, whereas the other p-tau subtypes added no additionaldiscriminatory power. Similarly, p-tau₁₈₁ alone maximized groupseparation between patients with AD and those with DLB and between the ADand VaD groups (data not shown). For discrimination between AD and non-ADdementias, a combination of p-tau₂₃₁ and p-tau₁₈₁resultedin a slight increase in sensitivity from 86% to 94% at the cost of decreasedspecificity from 75% to 66%, resulting in only a slight increase in correctclassification accuracy (from 83% to 85%) (Figure 3).

Levels of CSF p-tau₂₃₁ (ρ = −0.18; P = .06) and p-tau₁₈₁ (ρ = 0.002; P = .99) did not correlate with the Mini-Mental State Examination scorein patients with AD. Only for p-tau₁₉₉ did we found a correlationwith the Mini-Mental State Examination score (ρ = −0.25; P = .01). There was no significant effect of sex (p-tau₂₃₁: M-W₁₀₇ = 1302; P = .55; p-tau₁₈₁: M-W₁₀₇ = 1375; P = .89; p-tau₁₉₉: M-W₁₀₇ = 1203; P = .22) or age (p-tau₂₃₁: ρ = 0.002; P = .98; p-tau₁₈₁: ρ= −0.05; P = .61; p-tau₁₉₉: ρ= −0.013; P = .89) on levels of p-tau proteins.Levels of p-tau did not differ significantly between participating centers(p-tau₂₃₁: M-W₁₀₇ = 903; P =.74; p-tau₁₈₁: M-W₁₀₇ = 822; P =.34; p-tau₁₉₉: M-W₁₀₇ = 813; P =.31).

In the present study, we investigated the diagnostic performance of3 different pathophysiologically important CSF tau phosphorylation epitopes(p-tau₂₃₁, p-tau₁₈₁, and p-tau₁₉₉)^{20 - 22} to discriminate patientswith AD from those with other clinically important causes of dementia, thosewith ONDs, and nondemented controls. All 3 phosphorylation sites were studiedusing 3 recently developed and reported immunoassays.^{1 ,3 ,5} Sofar, independent early studies of the single phosphorylation sites have indicatedstatistically significant discriminative power between the AD and non-AD studygroups. Subsequently, CSF p-tau proteins in general were suggested as promisingbiological marker candidates for AD. In none of these pilot studies, however,were all 3 p-tau protein assays applied in the same set of subjects and patientsto look at differential diagnostic assay accuracy. With the advent of internationallarge-scale multicenter trials on putative biomarkers and neuroimaging inAD, the important issue has not yet been addressed of how p-tau assays performin general and whether there are potential relevant differences in assay performance,particularly in diagnostic sensitivity and specificity. Moreover, it is notyet known whether a combination of different p-tau epitopes or assays mightimprove diagnostic accuracy.

Our group showed that concentrations of all 3 p-tau proteins were equallysignificantly increased in patients with AD compared with the other groupsstudied. This finding is in strong agreement with all previously reportedresults.^{1 ,3 ,5 ,8 - 10}

In a next step, discriminative power of the individual p-tau proteinswas studied. To interpret the clinical significance of the determined diagnosticaccuracy, we followed the recommendations of a consensus report¹² foruseful (ideal) biomarkers of AD, determining specificity levels after thesensitivity level is 85% or higher. According to a consensus report on molecularand biochemical markers of AD, a useful (ideal) biomarker should yield a specificitylevel of at least 75% to 85%. In the differentiation of AD from non-AD, p-tau₂₃₁ and p-tau₁₈₁, but not p-tau₁₉₉, reached therecommended specificity level. All 3 p-tau proteins, however, showed excellentspecificity levels when patients with AD were compared with nondemented controls.

For the clinically relevant differential diagnosis of AD, it is essentialto have a marker or a set of markers that discriminates AD from other clinicallyrelevant dementias. Therefore, we included patients with FTD, DLB, and VaDas well. In the discrimination between AD and FTD, p-tau₂₃₁andp-tau₁₈₁ fulfilled the proposed biomarker criteria for sensitivityand specificity. Comparison of the markers revealed that p-tau₂₃₁andp-tau₁₈₁ discriminated better than p-tau₁₉₉ betweenthe AD group and the combined non-AD group and between AD and FTD.

The high level of discrimination between AD and FTD may originate indistinct differences in the biochemical and molecular signatures of tau-relatedpathophysiological changes between the 2 diseases.²³ Therewas an increase in the p-tau protein level in some patients with DLB and VaD.Concomitant AD-type neuropathological changes in the brain, including neurofibrillarytangles, has been described for many patients with VaD²⁴ andDLB²⁵ who are clinically indistinguishablefrom those with "pure" VaD and DLB, respectively. In a clinical setting, ithas to be assumed that patients with VaD and DLB are heterogeneous regardingunderlying AD characteristic neuropathological changes in the brain, resultingin an increase in the p-tau protein level in at least some patients with VaDand DLB.

Using marker combinations did not add discriminative power comparedwith applying single markers. This might be a consequence of the high intercorrelationof the markers and the accurate discrimination between groups applied as singlemarkers. Group separation was maximized between AD and FTD using p-tau₂₃₁and between AD and DLB using p-tau₁₈₁.

In addition, we considered the effect of potentially confounding factorson CSF p-tau levels to assess the clinical applicability of p-tau proteins.Mini-Mental State Examination score accounted for approximately 5% of thevariance in p-tau levels, being significant only for p-tau₁₉₉.This effect was not reported in previous studies,^{5 ,8} andit should be followed in independent samples. There was no effect of age andsex on levels of p-tau. In addition, different diagnostic centers did notaffect variance of p-tau levels. These findings indicate that p-tau proteinsmay be valuable markers for the clinical diagnosis of AD irrespective of age,sex, and diagnostic center.

To our knowledge, this is the first comparative study applying 3 developedand published immunoassays detecting different p-tau epitopes on the sameset of subjects and patients to compare the individual and combined diagnosticaccuracy. The results of this study indicate that all 3 p-tau assays performnearly equally well in discriminating patients with AD from nondemented controls.Both p-tau₂₃₁ and p-tau₁₈₁ fulfill the proposed criteriafor useful biomarkers in the differentiation of AD and non-AD and particularlyof AD and FTD.

Although there is no doubt that tau phosphorylation differs in AD, itis hard to speculate why. There have been few studies of phosphoserine 199and phosphothreonine 181 in the human brain. Except for one study,²² all we really know about these sites is that theyare phosphorylated in advanced AD neuropathological changes. The antibodiesused in the 181 and 199 assays have not been investigated to the same extentas the antibodies to phosphothreonine 231. It is well established that phosphorylationof threonine 231 is a very early event in AD, occurring before the formationof paired helical filaments in neurons of the hippocampus.²¹ Accordingto Augustinack and collegues,²² phosphorylationat threonine 181 and serine 199 occurs later, and these are only found toany appreciable extent in intracellular tangles. Reactivity to TG3, an antibodythat recognizes phosphothreonine 231, is found in pretangles, intracellulartangles, and extracellular tangles and so is present at all stages of thedisease. Augustinack and colleagues also suggest that several kinases canphosphorylate 199 and 231, but only extracellular regulated protein kinase2 phosphorylates 181. We suggest further investigations of the temporal sequenceof phosphorylations of the 3 sites, using the same antibodies as used in theCSF assays. Moreover, other potentially pathophysiologically relevant p-tauepitopes, such as serine 396 and serine 404, need to be further explored intheir ability to differentiate between relevant dementia disorders.²⁶

Another relevant issue is to distinguish patients with mild cognitiveimpairment (MCI) from controls and particularly to predict AD in MCI. We showedthat CSF p-tau₂₃₁ levels are elevated in patients with MCI comparedwith controls.²⁷ In this longitudinal study,high p-tau₂₃₁ levels at baseline correlated with the rate of cognitivedecline in Mini-Mental State Examination scores in patients with MCI. A subgroupof patients with MCI converted to AD. In agreement with the analysis of ratesof cognitive decline, increased levels of p-tau₂₃₁ correlated withconversion to AD. De Leon and colleagues²⁸ showeda longitudinal increase in p-tau₂₃₁ levels in patients with MCI.Elevated levels of CSF p-tau₁₈₁in patients with MCI compared withcontrols have also been shown.²⁹ Future studiesare warranted to further explore CSF p-tau proteins in MCI and to comparetheir diagnostic and prognostic value.

This study was conducted in an academic clinical setting. Diagnoseswere performed by experienced dementia experts according to National Instituteof Neurological and Communicative Disorders and Stroke–Alzheimer's Diseaseand Related Disorders Association criteria,¹³ withan estimated positive predictive value of 89% to 100%.³⁰ Partof our sample is enrolled in an ongoing neuropathological program designedto provide autopsy-confirmed diagnoses. In addition to autopsy-confirmed determinationof assay performance, population-based studies are warranted to establishCSF p-tau proteins as potential biomarkers for routine diagnostic use. Thesestudies are currently under way in international large-scale multicenter approaches.One large network, the National Institute on Aging Initiative on Neuroimagingin Alzheimer's Disease, will potentially start to evaluate neuroimaging thisyear, as well as an array of potential biomarkers in a 5-year longitudinalapproach in 650 individuals (patients with MCI, patients with AD, and controls).In their recent proceedings, a subconsortium, the Biological Marker WorkingGroup, has determined that measurement of CSF p-tau levels is a "feasiblecore marker" within the National Institute on Aging initiative.³¹

Reprints: Raymond Zinkowski, PhD, Applied NeuroSolutions Inc, 50Lakeview Pkwy, Vernon Hills, IL 60061 (e-mail: zinkowski@moleculargeriatrics.com).

Submitted for publication September 11, 2002; final revision receivedJune 24, 2003; accepted July 3, 2003.

This study was supported by grants from the Volkswagen-Foundation, Hannover,Germany (Dr Hampel); the Hirnliga e.V., Nürmbrecht, Germany (Drs Hampeland Buerger); a grant from the Medical Faculty, Ludwig-Maximilian University(Drs Buerger, Teipel, and Hampel); grants 11560 and 14002 from the SwedishMedical Research Council, Stockholm (Dr Blennow); and a grant from the Stiftelsenför Gamla Tjänarinnor, Stockholm (Dr Blennow).

We thank Felician Jancu, Bea Riemenschneider, Jenny Wagner, and OliverPogarell, MD, for clinical support; Thomas Nolde, PhD, and Heike Gluba fortechnical assistance; and Arun Lawrence Warren Bokde, PhD, and Jens Prüssner,PhD, for helpful discussion of the manuscript.

This study was presented in part at the 8th International Conferenceon Alzheimer's Disease and Related Disorders 2002, Alzheimer's Association;July 23, 2002; Stockholm, Sweden.

Corresponding authors: Katharina Buerger, MD, and Harald Hampel, MD,Dementia Research Section and Memory Clinic, Alzheimer Memorial Center andGeriatric Psychiatry Branch, Department of Psychiatry, Ludwig-Maximilian University,Nussbaumstrasse 7, 80336 Munich, Germany (e-mail: katharina.buerger@psy.med.uni-muenchen.de; hampel@psy.med.uni-muenchen.de).