ORIGINAL CONTRIBUTION

Transcription

1 ORIGINAL CONTRIBUTION High Cerebrospinal Fluid Tau and Low Amyloid 42 Levels in the Clinical Diagnosis of Alzheimer Disease and Relation to Apolipoprotein E Genotype D. Galasko, MD; L. Chang, MD; R. Motter, MPH; C. M. Clark, MD; J. Kaye, MD; D. Knopman, MD; R. Thomas, PhD; D. Kholodenko, MS; D. Schenk, PhD; I. Lieberburg, PhD, MD; B. Miller, MD; R. Green, MD; R. Basherad, MS; L. Kertiles, MS; M. A. Boss, PhD; P. Seubert, PhD Objective: To evaluate cerebrospinal fluid (CSF) levels of amyloid protein ending at amino acid 42 (A 42) and tau as markers for Alzheimer disease (AD) and to determine whether clinical variables influence these levels. Design: Cohort study. Setting: Six academic research centers with expertise in dementia. Subjects: Eighty-two patients with probable AD, including 24 with very mild dementia (Mini-Mental State Examination score 23/30) (AD group); 60 cognitively normal elderly control subjects (NC group); and 74 subjects with neurological disorders, including dementia (ND group). Main Outcome Measures: Levels of A 42 and tau were compared among AD, NC, and ND groups. Relationships of age, sex, Mini-Mental State Examination score, and apolipoprotein E (Apo E) genotype with these levels were examined using multiple linear regression. Classification tree models were developed to optimize distinguishing AD from NC groups. Results: Levels of 42 were significantly lower, and levels of tau were significantly higher, in the AD group than in the NC or ND group. In the AD group, 42 level was inversely associated with Apo E 4 allele dose and weakly related to Mini-Mental State Examination score; tau level was associated with male sex and 1 Apo E 4 allele. Classification tree analysis, comparing the AD and NC subjects, was 90% sensitive and 80% specific. With specificity set at greater than 90%, the tree was 77% sensitive for AD. This tree classified 26 of 74 members of the ND group as having AD. They had diagnoses difficult to distinguish from AD clinically and a high Apo E 4 allele frequency. Markers in CSF were used to correctly classify 12 of 13 patients who later underwent autopsy, including 1 with AD not diagnosed clinically. Conclusions: Levels of CSF 42 decrease and levels of CSF tau increase in AD. Apolipoprotein E 4 had a dosedependent relationship with CSF levels of 42, but not tau. Other covariates influenced CSF markers minimally. Combined analysis of CSF 42 and tau levels discriminated patients with AD, including patients with mild dementia, from the NC group, supporting use of these proteins to identify AD and to distinguish early AD from aging. In subjects in the ND group with an AD CSF profile, autopsy follow-up will be required to decide whether CSF results are false positive, or whether AD is a primary or concomitant cause of dementia. Arch Neurol. 1998;55: The affiliations of the authors appear in the Acknowledgment section at the end of the article. ACLINICAL diagnosis of Alzheimer disease (AD) typically requires a history from the patient and ideally an informant and results of mental status testing, physical and neurological examinations, a panel of laboratory tests, and a neuroimaging study to rule out other causes of dementia. The workup may include additional laboratory tests in atypical cases and more detailed psychometric testing and follow-up to increase the confidence of the diagnosis, especially when a patient presents with mild dementia. Current guidelines recognize an element of uncertainty in the clinical diagnosis with the terms probable or possible 1 ; definite AD is reserved for patients for whom results of a brain biopsy or autopsy show the requisite neuropathological features of AD. In recent series of patients with clinically diagnosed probable AD who underwent evaluation at research centers and were followed up to autopsy, the rates of agreement between the clinical and pathological diagnosis averaged about 85% to 90%. 2-7 These studies generally involve an extensive clinical workup and longitudinal monitoring of the progressive cognitive decline in subjects to strengthen the diagnosis of AD. Such studies define the maxi- 937

2 SUBJECTS AND METHODS SUBJECTS Subjects were enrolled through AD research clinics and neurology services at 6 academic medical centers. Informed consent was obtained from all participants or their guardians as appropriate. All procedures were performed under institutional guidelines for studies involving human subjects. The evaluation included history from the subject (and an informant if cognitive problems or dementia were apparent), neurological examination, and cognitive screening using the Mini-Mental State Examination (MMSE). 26 The Hachinski Ischemic Index 27 was used to screen all patients with dementia. In subjects suspected of having AD or other dementia, a neuroimaging study and dementia screening blood tests were required. A standardized case report form that documented this information and the diagnosis was recorded for all subjects. In addition, clinical vignettes were provided for all patients thought to have dementia due to reasons other than AD or with atypical dementia syndromes. Diagnoses were made by the neurologist investigator(s) at each site (D.G., L.C., B.M., C.M.C., J.K., D.K., and R.G.), and vignettes were reviewed by a coordinating neurologist (D.G.) to verify subject eligibility and to standardize diagnostic classifications. AD Group National Institute of Neurological and Communicative Disorders and Stroke Alzheimer s Disease and Related Disorders Association criteria for probable AD were required. Patients thought to have very early dementia were included, provided that this diagnosis was supported by results of additional psychometric testing and that follow-up subsequent to the CSF collection showed progression to probable AD. Recruitment was aimed toward mild to moderate degrees of dementia. Only 7 (9%) of 82 patients with AD had MMSE scores below 10 (of a possible 30), whereas 24 (29%) had scores above 23. NC Group Subjects were 50 years of age or older, without significant cognitive or neurological symptoms, and had normal results of a neurological examination. A score of at least 28 on the MMSE was required for eligibility. Sixty subjects were included in this group. ND Group In addition to patients with diverse neurological conditions, patients with dementia disorders that typically enter into the differential diagnosis with AD were recruited. The neurological diagnoses were based on best clinical judgment. Dementia was diagnosed according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition. 28 Wherever possible, research criteria were applied, eg, in cases of suspected frontotemporal dementia, 29 dementia with Lewy bodies, 30 and vascular dementia. 31 Subjects with depression accompanied by symptoms or signs of impaired cognition (n = 7) were included in this group of 74 patients, as the examining physicians thought that AD was unlikely to be the primary diagnosis. Depression was rated using the Hamilton Depression Scale, and a score of at least 10 was required. 32 CSF COLLECTION Cerebrospinal fluid samples were collected using lumbar puncture and tested for protein and glucose levels and cell count. Samples with greater than red blood cells per liter or white blood cells per liter were excluded. Samples were stored at 4 C for up to 24 hours or frozen if longer storage times were needed, before being shipped to Athena Diagnostics, Inc, Worcester, Mass, for evaluation. Our study included 81 subjects from a previous study by Motter et al. 11 The original CSF samples from these subjects were stored at 80 C until reassay in our study. The concentrations of 42 and tau in CSF do not change appreciably after storage for up to 2 years under these conditions (data not shown). mum in clinical accuracy but require resources that exceed those of routine clinical practice. Diagnosis at early stages of dementia is likely to be less accurate than after prolonged follow-up. The advent of cholinergic-based treatments for AD emphasizes the need for early and accurate diagnosis to allow initiation of therapy when it will be of most benefit to the patient. There are several reasons to consider biological markers for AD, including strengthening the certainty of the clinical diagnosis, distinguishing AD from cognitive symptoms attributable to aging and from other dementias, allowing early and even presymptomatic diagnosis, and providing an index of disease activity. The efficacy of future therapies may in fact be best monitored by using surrogate biological markers, as has been shown strikingly in recent years for disorders such as acquired immunodeficiency syndrome. Such an approach may complement the evaluation of cognitive changes in AD and could streamline the evaluation of future drugs. Biochemical markers for AD have been sought for many years. 8 Recent progress has focused on analytes in cerebrospinal fluid (CSF) that are based on the pathologically altered proteins found in the brains of patients with AD. Specifically, studies have measured CSF levels of the microtubule-associated protein tau, the primary constituent of neurofibrillary tangles, and of a form of amyloid protein ending at amino acid 42 ( 42), the major component of the parenchymal amyloid deposits. Several groups have now confirmed that tau levels are increased 9-19 and 42 levels are decreased 11,20,21 in the CSF of patients with AD, compared with nondemented elderly control subjects. Clinical or demographic factors have not been shown to have a strong effect on variation in levels of these CSF markers, although contradictory effects of dementia severity on CSF tau levels have been reported ,16 Apolipoprotein E (Apo E) genotype has a major effect on the risk for development of AD and the age of onset, with the Apo E 4 allele increasing the risk and the 938

3 42 ENZYME-LINKED IMMUNOSORBENT ASSAY Levels of 42 were quantified as described previously. 11 Briefly, the capture antibody 266, which recognizes residues 13 through 28, and the 42 specific reporter antibody were used as the sandwich enzyme-linked immunosorbent assay (ELISA) pair. Insignificant crossreactivity with 40 occurs using this assay. The 42 peptide (Bachem, Torrance, Calif) used as the standard was dissolved in dimethyl sulfoxide, sonicated, filtered, and immediately diluted in specimen diluent to final working concentrations before freezing. Standards were stable for more than 1 year when prepared in this manner. Despite the fact that amino acid analysis was used in this study and the study by Motter et al 11 to establish the peptide stock concentrations, we observed a difference of approximately 2-fold in standard quantitation. Our absolute results were thus not directly comparable to those of Motter et al. 11 Although the 42 values quantitated relatively higher than the original determinations, the paired values of the original and repeated assays were strongly correlated (Pearson r 2 = 0.68; n = 77). Total levels of in CSF, which consist primarily of peptides ending at position 40, were not routinely analyzed in this study, as the study by Motter et al 11 showed that this measure, used alone or in a ratio with other measures, did not increase ability to discriminate the AD from the NC groups. 11 TAU ELISA Tau levels in CSF were measured as previously described. 10,11 The tau standard, prepared from human brain as previously described, is a different lot than used in the previous studies, yielding slightly different quantitation of samples, with higher levels than previously reported. Reassayed samples again showed a strong correlation with the original measures (Pearson r 2 = 0.91). Apo E GENOTYPING Apolipoprotein E genotypes were determined using standard polymerase chain reaction methods with DNA derived from blood samples obtained from subjects. 33 Subjects consenting to blood donation included 76 in the AD, 53 in the NC, and 62 in the ND groups. DATA ANALYSIS Data were analyzed using commercially available software (Splus, version 3.4; MathSoft, Inc, Seattle, Wash). For betweengroupcomparisonsofdemographic,clinical,andcsfvariables, Student t tests were used, with the Holm adjustment for multiple comparisons of post hoc tests. Discrete variables such as Apo E allele frequencies were compared using 2 tests. To determinetheextenttowhichage, sex, dementiaseverity, orapoe genotypeinfluencedcsf 42ortaulevels,weusedmultiple linearregressionmodelstoexaminethead,nc,andndgroups, followedbybackwardselection,todefinewhichpredictorswere significant at a the level of P.05. To examine the extent to which CSF tau and 42 levels individually distinguished the AD from the NC groups, we first used receiver operating characteristic curve analysis to determine cutoffs for 42 and tau levels that maximized overall classification. To develop predictive cutoffs that optimized the combined use of 42 and tau levels, we used classification tree statistics in the software, specifically the RPART routines developed by Therneau and Atkinson. 34 The approach uses recursive partitioning, a nonparametric computation that considers all possible binary splits of the data in pursuit of optimal classification. 35 Classification trees typically achieve a similar degree of accuracy as logistic regression in classification analyses and, in addition, provide an intuitive classification process that can be readily applied to clinical diagnosis. The analysis considered 42 and tau levels simultaneously to create cutoff values that maximized heterogeneity in the classification of the AD and NC groups, generating a decision tree. To justify branches and prevent overfitting of the data in the classification model, an algorithm in RPART was used to find the most parsimonious model that optimized predictive ability. The cost of adding each successive branch to the model was weighed against its value in improving classification in a validation process that analyzed 10 successive subsets of 90% of the data set, each of which omitted a different 10% of the subjects. Apo E 2 allele protecting against it, relative to the Apo E 3 allele Protein interactions between Apo E and tau or have been proposed as mechanisms that could explain this genetic effect. 24,25 Previous reports have not found an effect of Apo E genotype on CSF tau and 42 levels, but the number of subjects in each genotype category was fairly small. 11,12 We herein extend observations on the use of CSF tau and 42 levels to distinguish patients with AD (AD group) from normal elderly controls (NC group) and patients with other neurological and dementing disorders (ND group). In addition, we explore the relationships between CSF tau and 42 levels and clinical variables such as age, sex, dementia severity, and Apo E genotype. RESULTS The analysis included data from a total of 82 subjects in the AD group, 60 in the NC group, and 74 in the ND group. Of these, 31 subjects in the AD group, 19 in the NC group, and 31 in the ND group had been included in the previous study by Motter et al 11 and had sufficient stored CSF for reassay; updated clinical information was taken into account to verify clinical diagnoses for these subjects. Table 1 shows that the AD and NC groups did not differ significantly regarding age and sex. All 3 groups covered a wide age range (42-85 years [median age, 73 years] for the AD group; years [median age, 70 years] for the NC group; and years [(median age, 65 years] for the ND group). The ND group was slightly younger than the NC and AD groups. The AD group had a mean ± SD MMSE score of 19.3 ± 6.8 and included 24 subjects with very mild dementia, defined as an MMSE score of at least 24. Follow-up information and results of detailed psychometric testing on these mildly demented subjects indicated progressive cognitive and functional decline consistent with AD. The Apo E 4 allele frequencies of 41% in the AD group and 17% in the NC group were similar to those of published clinical series 36,37 and showed the expected in- 939

4 Table 1. Demographic Data* Groups AD Group NC Group AD (n = 82) crease in Apo E 4 allele frequency associated with AD. However, the Apo E allele distribution in the ND group differed significantly from that of the NC group ( 2, P =.04) and included 7 patients homozygous for the Apo E 4 allele. This raised the possibility that unrecognized cases of AD existed among the ND group. The ND group included patients with the following clinical diagnoses: frontotemporal dementia (n = 19), Parkinson disease (PD) (with or without dementia) (n = 10, of whom 5 were clinically demented), vascular dementia (n = 7), depression (n = 7), progressive aphasia (n = 6), Lewy body dementia (n = 3), amyotrophic lateral sclerosis (n = 3), progressive supranuclear palsy (n = 4), cortical-basal ganglia degeneration (n = 2), cerebellar ataxia (n = 3), mild cognitive impairment (n = 3), nonprogressive amnesia after head trauma (n = 1), normal-pressure hydrocephalus (n = 1), dementia with motor neuron disease (n = 1), multiple sclerosis with dementia (n = 1), cognitive symptoms after treatment of Lyme disease (n = 1), ganglioglioma with partial complex seizures (n = 1), and myotonic dystrophy (n = 1). 42 LEVELS NC (n = 60) ND (n = 74) Age, y 71.7 ± ± ± 9.3 Sex, M:F 41:41 32:28 44:30 Mini-Mental State 19.3 ± ± ± 8.0 Examination score No. of subjects with 0 Apo E 4 alleles Apo E 4 alleles Apo E 4 alleles Apo E allele # ** comparisons 4 frequency CSF A 42, pg/ml 833 ± ± ± 464 CSF tau, pg/ml 663 ± ± ± 383 *Ranged values are mean ± SD. Apo E indicates apolipoprotein E; CSF, cerebrospinal fluid; and A 42, amyloid protein ending at amino acid 42. Groups are described in the Subjects subsection of the Subjects and Methods section. To convert picograms per milliliter to grams per liter, multiply by P =.05 vs NC group. P.001 vs ND group. P =.05 vs ND group. P.001 vs NC group. P =.02 vs ND group. #P =.10 vs ND group. **P =.04 vs ND group. P =.06 vs ND group. P =.005 vs ND group. P =.19 vs ND group. Aβ42 Levels, pg/ml No. of Apo ε4 Alleles Figure 1. Relationship of apolipoprotein E (Apo E) genotype to levels of amyloid protein ending at amino acid 42 ( 42) in cerebrospinal fluid of patients with Alzheimer disease (AD group) and normal control subjects (NC group). Levels of A 42 are given by Apo E 4 allele copy number. Boxes indicate 25th to 75th percentiles of subjects; lines within boxes, the median values; whiskers, range from the 10th to 90th percentiles; and open circles, individuals falling outside this range. The groups had the following numbers of members: 30 in the AD and 35 in the NC groups with no Apo E 4 alleles; 30 in the AD and 18 in the NC groups with 1 Apo E 4 allele; and 16 in the AD and 0 in the NC groups with 2 Apo E 4 alleles. Levels of 42 did not differ between members of the NC group with 0 or 1 Apo E 4 allele. Significant differences were found between members of the AD group homozygous for the Apo E 4 allele and members with 1 ( P =.009) or no Apo E 4 allele ( P.001) or the NC group ( P.001). Members of the AD group heterozygous for the Apo E 4 allele differed from those with none ( P =.008) and the NC group ( P.001); those with no Apo E 4 allele also differed from the NC group ( P.001). To convert picograms per milliliter to grams per liter, multiply by Cerebrospinal fluid levels of 42 were significantly lower in the AD group compared with the other groups (Table 1). Mean levels of 42 were lowest in the AD group, intermediate in the ND group, and highest in the NC group. We analyzed the effects of age, sex, severity of dementia, and Apo E 4 allele on 42 levels using multiple linear regression. In the AD group, age and sex were not significant covariates, while MMSE scores showed a slight positive association (higher MMSE scores correlated with higher 42 levels; r 2 = 0.12; P =.007). The strongest covariate was Apo E 4, which showed a highly significant negative correlation with 42 level dependent on the number of Apo E 4 alleles. All of these covariates accounted for only a small amount of the variance of 42 level, since overall, r 2 = 0.33 (P.001). The extent of the correlation due to Apo E 4 copy number alone was As shown in Figure 1, CSF 42 levels in the AD group were highest in subjects with no Apo E 4 alleles ( ± g/l [1067 ± 449 pg/ml]), intermediate in patients with a single Apo E 4 copy ( ± g/l [772 ± 265 pg/ml]), and lowest in patients homozygous for the Apo E 4 allele ( g/l ± g/l [566 ± 201 pg/ ml]). Levels of CSF 42 in the subgroup of the AD group with no Apo E 4 alleles were significantly lower than in the NC group (P.001). For 42 levels among the NC group, multiple linear regression showed a minimal decline with age (r 2 = 0.11; P =.04), whereas sex and Apo E 4 allele count (0 or 1) had no significant effects; the model accounted for a very small amount of variance (overall, r 2 = 0.16; P =.04). Among the ND group, MMSE score was positively associated (P =.02) and Apo E 4 allele count was negatively associated (P =.02) with 42 levels, whereas age and sex were not significant for the overall model (r 2 = 0.24; P.001). 940

5 Aβ pg/ml AD 82 NC 60 Aβ42 >1125 pg/ml 2750 NC Group AD Group Non AD Undefined AD Aβ pg/ml AD 42 NC 01 AD 69 NC 14 AD 21 NC 03 Aβ42 >761 pg/ml Tau 503 pg/ml AD 27 NC 13 Tau <503 pg/ml NC 10 AD 06 Tau 558 pg/ml AD 6 NC 4 NC 46 AD 13 Tau <558 pg/ml NC 42 AD 07 Aβ42 Levels, pg/ml Figure 2. Classification tree generated using RPART software 34 and considering levels of tau and amyloid protein ending at amino acid 42 (A 42) in cerebrospinal fluid simultaneously in patients with Alzheimer disease (AD; n = 82) and normal control subjects (NC; n = 60). Results of successive partitioning steps (nodes) are indicated by circles; squares represent terminal nodes. Unshaded squares constitute the undefined zone described in the Differentiating AD From NC Groups subsection of the Results section. The number of subjects classified by each node is indicated within each circle or square. The shaded squares in which the number of patients with AD is much greater than that of the NC group constitute the AD zone, whereas the shaded node consisting mainly of the NC group constitutes the non-ad zone. To convert picograms per milliliter to grams per liter, multiply by TAU LEVELS Levels of CSF tau differed significantly among all 3 groups (overall analysis of variance, Table 1) and significantly between AD vs NC groups and AD vs ND groups based on results of post hoc tests. Mean levels of tau showed the opposite gradient to 42, ie, highest in the AD group, intermediate in the ND group, and lowest in the NC group. In the AD group, multiple linear regression models examined the covariates of age, sex, Apo E genotype, and MMSE score in relation to tau levels (overall, r 2 = 0.26; P.001). Significant predictors were sex (male more than female; P =.03), 1 Apo E 4 allele (P.001), and 2 Apo E 4 alleles (P =.04). The Apo E effects were in opposite directions, however, with increased tau levels in patients heterozygous for the Apo E 4 allele compared with those without the Apo E 4 allele, and decreased tau levels in patients homozygous for the Apo E 4 allele compared with those without the Apo E 4 allele. The MMSE scores and age were not significant predictors of tau levels. Each predictor contributed to a small extent (eg, partial r 2 = 0.12, 1 Apo E 4 allele). Similar models examining CSF tau levels among the NC group showed no significant effects of age, sex, or Apo E (overall, r 2 = 0.12). Among the ND group, the overall correlation coefficient was only 0.09, and no predictors attained statistical significance. DIFFERENTIATING AD FROM NC GROUPS Tau Levels, pg/ml Figure 3. Plot of high-specificity, 3-zone classification derived from the classification tree. Cerebrospinal fluid (CSF) data from subjects with Alzheimer disease (AD group) and normal control subjects (NC group) are indicated in relation to zones derived from the nodes of highest specificity (shaded nodes in Figure 2). Terminal nodes with less than 85% specificity for AD or NC group were undefined. The AD zone included 63 patients in the AD group (sensitivity, 77%) and 4 in the NC group (specificity, 93%). The non-ad zone included 42 subjects in the NC group (sensitivity, 70%) and 7 patients in the AD group (specificity, 91%). For clarity of presentation, 1 patient with AD with CSF tau measure of g/l (3717 pg/ml) and A 42 measure of g/l (636 pg/ml) is not included in the plot. Asterisk indicates the subject whose clinical diagnosis was normal at the time of CSF testing but in whom symptoms of early AD subsequently developed. This subject is considered as part of the NC group for the sensitivity and specificity calculations. To convert picograms per milliliter to grams per liter, multiply by We examined CSF 42 and tau levels individually for their ability to discriminate between the AD and NC groups. For 42 levels, the optimal cutoff for differential classification was g/l ( pg/ ml), which was used to classify correctly 115 (81.0%) of 142 subjects in the AD and NC groups, overall, with sensitivity of 78% and specificity of 83% for the diagnosis of AD. For tau levels alone, the best cutoff was g/l (502.5 pg/ml), which was used to classify correctly 97 (68.3%) of 142 subjects in the AD and NC groups, overall, with sensitivity of 57% and specificity of 83% for the diagnosis of AD. Better resolution of the groups was realized by applying the binary tree structured classification system, in which tau and 42 levels were considered simultaneously. This produced a tree (Figure 2) that correctly classified 121 (85.2%) of 142 subjects, with sensitivity of 90% and specificity of 80% for the diagnosis of AD. Because Apo E 4 influenced 42 levels, a separate classification tree analysis used Apo E 4 allele count and CSF 42 and tau levels as possible predictors. The Apo E 4 allele count did not enter into the optimal tree, which remained identical to that in Figure 2. For a biomarker test to be a useful adjunct in the clinical workup of AD, it should have high sensitivity. In addition, given the consequences of a positive diagnosis, very high specificity is desirable. To achieve a useful balance of sensitivity and very high specificity, we selected terminal nodes of the classification tree analysis as being diagnostically useful only if they contained rela- 941

6 Table 2. Classification of ND Group by 3-Zone Decision Tree* CSF Diagnosis Subgroup All Subjects AD Undefined Non-AD All subjects, No allele frequency 0.27 (62) 0.45 (22) 0.21 (14) 0.17 (26) Nondementia, No. of subjects allele frequency 0.21 (14) 0.67 (3) 0.00 (3) 0.07 (8) Dementia and cognitive impairment, No. of subjects allele frequency 0.31 (48) 0.42 (19) 0.27 (11) 0.22 (18) *ND indicates patients with neurological and dementing disorders other than AD; CSF, cerebrospinal fluid; and AD, Alzheimer disease. The CSF diagnosis classification is described in Figure 3. The numbers in parentheses represent the number of subjects in that category for which genotype information is available. Diagnoses include depression (n = 5), Parkinson disease (n = 4), amyotrophic lateral sclerosis (n = 3), cerebellar ataxia (n = 2), progressive supranuclear palsy (n = 2), myotonic dystrophy (n = 1), and seizure (n = 1). tively pure AD or NC groups (specificity of 85% or higher). Terminal nodes of the tree with specificities below 85% were considered to be undefined. This created 3 classification zones as shown in Figure 3. Applying this system to the cohorts yielded an AD zone with 76.8% sensitivity and 93.3% specificity and a non-ad zone with 70% sensitivity and 91% specificity. Overall, using the 3-zone classification system, 63 subjects in the AD group and 42 subjects in the NC group (73.9%) were correctly classified, and 11 (7.7%) were incorrectly classified. Application of the zonal classification system to the 24 patients with very mild AD (MMSE scores 24 of 30) resulted in 15 classified in the AD zone, 3 in the non-ad zone, and 6 in the undefined zone. One interesting footnote to this group concerned a subject whose test results were positive for AD using CSF markers, but who had cognitively normal test results at the time of lumbar puncture. In the 30 months after the CSF test was performed, symptoms consistent with a diagnosis of mild AD, including decline in cognitive test scores, developed in this subject. CLASSIFICATION OF ND GROUP The ND group was also classified by applying the highspecificity 3-zone scheme used to differentiate the AD from the NC groups (Figure 3). Twenty-six subjects were in the AD zone; 29, in the non-ad zone; and 19, in the undefined zone (Table 2). This suggests that the AD CSF profile is not specific for AD (false positive) and may occur in a variety of conditions that damage the brain, or that AD neuropathological features may be present in many subjects with dementia and that their contribution to the dementia was underrecognized by clinicians required to make a single diagnosis (in these cases, the CSF tests yield true positive results, at variance with the clinical diagnoses). To examine these possibilities, we considered the specific clinical diagnoses and Apo E genotypes among the ND group. First, classification as AD or non-ad zone varied according to whether patients were Table 3. CSF Diagnosis of Demented or Cognitively Impaired Subjects in ND Group* CSF Diagnosis Disorder AD Undefined Non-AD Frontotemporal dementia Vascular dementia Progressive aphasia Parkinson disease dementia Lewy body dementia Depression, memory decrease Mild cognitive impairment Miscellaneous *The CSF diagnoses and ND group are described in Table 2. Data are given as number of patients. Miscellaneous disorders classified as AD include dementia after central nervous system Lyme disease, multiple sclerosis with dementia, normal pressure hydrocephalus, and temporal lobe ganglioglioma and seizures; classified as undefined, atypical dementia with amyotrophy and progressive supranuclear palsy (n = 2); and classified as non-ad, cortical basal ganglia degeneration (n = 2), cerebellar ataxia after meningitis, and posttraumatic amnesia. demented. Very few patients without dementia (MMSE score 27 and no significant memory or cognitive complaints) were classified as having AD based on CSF test results (3 of 18), compared with patients with conditions associated with dementia or memory problems (23 of 56). The breakdown into specific diagnoses (Table 3) shows that within each diagnostic group, fewer of the ND group were classified as having AD using CSF levels of 42 and tau. Among these disorders were dementias that are clinically difficult to distinguish from AD, ie, frontotemporal dementia, vascular dementia, progressive aphasia, and PD with dementia. Two of 3 patients with mild cognitive impairment, a possible prodrome of AD, had an AD profile. Further evidence that subjects in the ND group with an AD profile actually may have AD was provided by the Apo E 4 allele frequencies. Overall, the ND group had a slightly elevated Apo E 4 allele frequency (27%, including 7 patients homozygous for the Apo E 4 allele) compared with the NC group. In the ND subgroup classified in the AD zone, the Apo E 4 allele frequency was 45%, similar to the 41% of the AD group. Of the 7 patients homozygous for the Apo E 4 allele among the ND group, 5 were classified in the AD zone. Conversely, in the ND subgroup in the non-ad zone, the Apo E 4 allele frequency was 22%, similar to the NC group value of 17%. Whereas presence of the Apo E 4 allele alone is not synonymous with AD, its increased frequency in this subset of the ND group strongly increased the likelihood of an admixture of AD. Only neuropathological examination will reveal whether the CSF tests or the clinician is correct in these cases where the 2 sets of results disagree. In the course of our study, 2 subjects in the ND group classified in the AD zone using CSF test results have undergone autopsy. One had rapidly progressive dementia accompanied by amyotrophy with onset at 81 years of age, clinically diagnosed as atypical amyotrophic lateral sclerosis dementia complex, possibly frontal lobe dementia; autopsy 6 months later revealed plaques and tangles in suffi- 942

7 cient quantities to meet a neuropathological diagnosis of AD. The second subject had seizures and cognitive decline. A right temporal lobe ganglioglioma was found on biopsy, and examination of the temporal lobe tissue removed during surgery to resect the tumor did not indicate the presence of AD. Another 4 patients in the ND group who were not classified in the AD zone using CSF markers have since undergone autopsy, and all were confirmed not to have AD. Seven autopsy-confirmed AD cases were AD positive based on results of CSF testing. The 9 patients in the ND group with 42 levels below g/l (761 pg/ml), a classification tree zone occupied almost exclusively by patients with AD in the AD vs NC group comparison, consisted of 4 patients with frontotemporal dementia and 1 patient each for diagnoses of primary progressive aphasia (Apo E 4/ 4), depression with cognitive changes (MMSE score, 21; Apo E 4/ 4), vascular dementia, PD with depression and memory complaints (Apo E 4/ 4), and Lyme disease followed by cognitive complaints (MMSE score, 20; Apo E 3/ 4). All of these patients presented with dementia or cognitive impairment, while none of the cognitively normal subjects in the ND group had an 42 level of less than g/l (761 pg/ml). Alzheimer disease likely contributed to the clinical picture in some of these cognitively impaired patients. COMMENT Our study is consistent with and extends the results of a previous report showing that 42 levels are decreased and tau levels are increased in CSF in AD. 11 The larger number of subjects in our report, recruitment of very mildly demented individuals, and the larger number of subjects with Apo E genotype provided adequate statistical power to examine the effects of a number of covariates. Across an age range of 52 to 82 years, cognitively normal individuals showed only a small trend toward increased CSF tau levels and a minimal effect on 42 levels. Dementia severity did not influence tau levels, as we and others 11,12 have previously reported, and in fact tau levels have been shown to be elevated in very mild AD. 38,39 Interestingly, the extent of cognitive impairment slightly affected 42 levels. The decrease in this marker with increasing dementia is consistent with a reduction in viable neurons producing 42 or a larger fraction of 42 being sequestered in plaque or otherwise unable to enter the CSF. The fact that total levels do not decrease in the CSF of patients with AD 11 favors the latter argument. Apolipoprotein E genotype was the major significant covariate related to 42 levels in the AD group in an Apo E 4 allele dose-dependent fashion. Patients homozygous for the Apo E 4 allele had the lowest mean level of 42, while those lacking an Apo E 4 allele had the highest level. Regardless of whether subjects had 0, 1, or 2 Apo E 4 alleles, the AD group showed lower 42 mean levels than the NC group. One possible explanation for the genotype effect is that the frequency of misdiagnosis may be higher in subjects lacking an Apo E 4 allele, 37 artifactually raising the group mean. However, tau levels did not show a similar Apo E 4 dose dependence, and we would have expected both measures to be similarly distorted by misdiagnoses. Failure to find this genotypic effect on 42 levels in an earlier study 11 was likely due to the insufficient number of patients with AD lacking an Apo E 4 allele, as that study had an unusually high Apo E 4 frequency (58%). It was suggested originally that the decrease in CSF 42 levels was related to the specific deposition of this protein, as opposed to the bulk levels, in amyloid plaque. 11 Our data strengthen this idea by linking 42 levels inversely to Apo E 4 gene dosage. The plaque burden in AD has been well documented to increase with Apo E 4 gene dosage, 40,41 consistent with the lowering of CSF 42 levels being most pronounced in patients homozygous for the Apo E 4 allele. A link between CSF 42 levels and plaque burden would also explain the diagnostic specificity for AD associated with the Apo E genotype and CSF 42 levels. Ongoing studies will use neuropathological correlates to further test this idea. The undefined zone in Figure 3 included patients who showed high CSF tau and 42 levels. Numerous studies have shown elevation of CSF tau levels can occur not only in AD but also in other neurological and neurodegenerative disorders. The tau released into CSF has been shown, in patients with AD or other disorders, to be a fragment lacking much of the portion of tau thought to be critical in paired helical filament formation. 42 Only rarely have elevated CSF tau levels been observed in normal controls, suggesting that the finding of high CSF tau levels indicates an underlying neuropathological process. The ND group included the most diagnostically challenging patients, and CSF 42 and tau levels could not be fully interpreted in the absence of autopsy examination. Thus the patients with non-ad dementias enrolled in our study may not be typical of those seen in community practice settings. Although AD is by far the most common cause of dementia, its distinction from other conditions associated with dementia can be clinically difficult. One factor that may lead to misdiagnoses is the lack of robust clinical criteria for identifying patients with specific non-ad dementing conditions. Another confounding factor is that AD may coexist with other causes of dementia, making the clinical diagnosis of non-ad dementias more uncertain than that of probable AD. For example, studies of dementia conditions clinically diagnosed as frontotemporal dementia, 29,43 progressive aphasia, 44,45 progressive supranuclear palsy, 46 and cortical-basal ganglia degeneration 43 have all revealed a few patients who have AD lesions at autopsy, alone or in conjunction with another degenerative condition. In vascular dementia and dementia associated with PD, coexistence of a significant burden of AD lesions is a common finding. 47,48 In view of these reservations, interpretation of the CSF data in the ND group can only be tentative. Nonetheless, we identified several interesting observations and trends. First, there was a high Apo E 4 allele frequency among the ND group, especially those with an AD-like profile of CSF 42 and tau levels (Table 3). In clinical series, association between the Apo E 4 allele and disorders such as PD 49 and Pick disease 50,51 have not been 943

8 found, and in autopsy series, the presence of the Apo E 4 allele in non-ad dementias can be related to coexisting AD abnormalities. 52 This is consistent with the possibility that patients with non-ad dementia and the Apo E 4 allele are more likely to harbor AD abnormalities, which fits with the striking increase in Apo E 4 allele frequency (45%) in such patients with an AD-CSF profile. Second, no particular subgroup of the ND group was classified consistently as having AD. Third, only 9 patients in the ND group were classified as having AD because of high CSF tau values ( 42, g/l [ pg/ml]; tau, g/l [ 503 pg/ml]), whereas 16 were classified as having AD because of 42 levels below g/l (761 pg/ml). Therefore, low CSF 42 levels were more important than tau levels in determining the classification of the ND group. As there is no evidence that deposition occurs nonspecifically in other neurological conditions besides AD, the CSF results hinted at undiagnosed AD rather than false positive results in the ND group, assuming the lowering of CSF 42 levels is related to plaque formation. Our study has confirmed and extended earlier findings that low 42 and high tau levels characterize the CSF of most patients with AD and supports several potential clinical applications of these measures. In patients suspected of having AD, but in whom the clinical diagnosis is uncertain, CSF 42 and tau levels can be used to strengthen the diagnosis. When more than 1 potential cause of dementia is suspected, these CSF markers may help to establish that AD is a contributory factor. With the advent of symptomatic treatment for cognitive symptoms of AD, and as drugs are developed to slow the progression of AD, very early diagnosis is important to define the earliest window of opportunity to apply such treatments. At this early stage of AD, it is more difficult to document a history of progressive cognitive and functional decline or to demonstrate deficits on cognitive testing. This often raises the clinical question of whether symptoms represent age-associated changes or AD, a decision that the biomarker tests may help to resolve. A major future focus of further CSF studies and studies of other biological markers will be subjects with mild cognitive impairment or with risk factors for dementia combined with minimal cognitive symptoms. In this regard, it is encouraging that most patients with AD and very mild dementia (MMSE score 24) were classified as having an AD CSF profile, indicating that measuring CSF 42 and tau levels may be helpful in diagnosing very early or mild AD. Accepted for publication December 2, From the Department of Neurosciences, University of California, San Diego (Drs Galasko and Thomas), and San Diego Veterans Affairs Medical Center (Drs Galasko and Thomas); the Department of Neurology, University of California, Los Angeles, Harbor Medical Center, Torrance (Drs Chang and Miller); Athena Neurosciences, Inc, South San Francisco, Calif (Mss Motter and Kholodenko and Drs Schenk, Lieberburg, and Seubert); the Department of Neurology, University of Pennsylvania, Philadelphia (Dr Clark); Aging and Alzheimer s Disease Center, Oregon Health Sciences University, Portland (Dr Kaye); the Department of Neurology, University of Minnesota Hospital, Minneapolis (Dr Knopman); the Memory Assessment Clinic and Alzheimer s Disease Program, College of Health and Human Sciences, Georgia State University, Atlanta (Dr Green); and Athena Diagnostics, Inc, Worcester, Mass (Messrs Basherad and Kertiles and Dr Boss). Supported in part by Athena Neurosciences, Inc, South San Francisco, Calif, and Athena Diagnostics, Inc. Reprints: Peter Seubert, PhD, Athena Neurosciences, Inc, 800 Gateway Blvd, South San Francisco, CA REFERENCES 1. McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM. 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: Tierney MC, Fisher RH, Lewin AJ, et al. The NINCDS-ADRDA Work Group criteria for the clinical diagnosis of probable Alzheimer s disease. Neurology. 1989; 38: Morris JC, McKeel JW, Fulling K, Torack RM, Berg L. Validation of clinical diagnostic criteria for Alzheimer s disease. Ann Neurol. 1988;24: Galasko D, Hansen LA, Katzman R, et al. Clinical-neuropathological correlations in Alzheimer s disease and related dementias. Arch Neurol. 1994;51: Gearing M, Mirra SS, Hedreen JC, Sumi SM, Hansen LA, Heyman A. Consortium to Establish a Registry for Alzheimer s Disease (CERAD), X. Neurology. 1995; 45: Klatka LA, Schiffer RB, Powers JM, Kazee AM. Incorrect diagnosis of Alzheimer s disease: a clinicopathologic study. Arch Neurol. 1996;53: Mendez M, Mastri AR, Sung JH, Frey WH. Clinically diagnosed Alzheimer s disease: neuropathologic findings in 650 cases. Alzheimer Dis Assoc Disord. 1992; 6: Seubert P, Galasko D, Boss MA. Use of CSF-based markers in the diagnosis of Alzheimer s disease. In: Brioni J, Decker M, eds. Pharmacological Treatment of Alzheimer s Disease: Molecular and Neurobiological Foundations. New York, NY: Wiley-Liss Inc; 1997: Vandermeeren M, Mercken M, Vanmechelen E, et al. Detection of proteins in normal and Alzheimer s disease cerebrospinal fluid with a sensitive sandwich enzyme-linked immunosorbent assay. J Neurochem. 1993;61: Vigo-Pelfry C, Seubert P, Barbour R, et al. Elevation of microtubule-associated protein tau in the cerebrospinal fluid of Alzheimer s patients. Neurology. 1995; 45: Motter R, Vigo-Pelfry C, Kholodenko D, et al. Reduction of -amyloid peptide 42 in the cerebrospinal fluid of patients with Alzheimer s disease. Ann Neurol. 1995; 38: Arai H, Terajima M, Miura M, et al. Tau in cerebrospinal fluid: a potential diagnostic marker in Alzheimer s disease. Ann Neurol. 1995;38: Tato R, Frank A, Hernanz A. Tau protein concentrations in cerebrospinal fluid of patients with dementia of the Alzheimer type. J Neurol Neurosurg Psychiatry. 1995;59: Jensen M, Basun H, Lannfelt L. Increased cerebrospinal fluid tau in patients with Alzheimer s disease. Neurosci Lett. 1995;186: Mori H, Hosoda K, Matsubara E, et al. Tau in cerebrospinal fluids: establishment of the sandwich ELISA with antibody specific to the repeat sequence in tau. Neurosci Lett. 1995;186: Hock C, Golombowski S, Naser W, Muller-Spahn F. Increased levels of protein in cerebrospinal fluid of patients with Alzheimer s disease: correlation with degree of cognitive impairment. Ann Neurol. 1995;37: Munroe WA, Southwick PC, Chang L, et al. Tau protein in cerebrospinal fluid as an aid in the diagnosis of Alzheimer s disease. Ann Clin Lab Sci. 1995;25: Blennow K, Wallin A, Agren H, Spenger C, Siegfried J, Vanmechelen E. Tau protein in cerebrospinal fluid: a biochemical marker for axonal degeneration in Alzheimer disease? Mol Chem Neuropathol. 1995;26: Rosler N, Wichart I, Jellinger KA. Total tau protein immunoreactivity in lumbar cerebrospinal fluid of patients with Alzheimer s disease. J Neurol Neurosurg Psychiatry. 1996;60: Ida N, Hartmann T, Pantel J, et al. Analysis of heterogeneous A4 peptides in human cerebrospinal fluid by a newly developed sensitive Western blot assay. J Biol Chem. 1996;271: Tamaoka A, Sawamura N, Fukushima T, et al. Amyloid protein 42(43) in ce- 944

9 rebrospinal fluid of patients with Alzheimer s disease. J Neurol Sci. 1997;148: Corder EH, Saunders AM, Strittmatter WJ, et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer s disease in late onset families. Science. 1993;261: Weisgraber KH, Mahley RW. Human apolipoprotein E: the Alzheimer s disease connection. FASEB J. 1996;10: Strittmatter WJ, Saunders AM, Schmechel D, et al. Apolipoprotein E: highavidity binding to -amyloid and increased frequency of type 4 allele in lateonset familial Alzheimer disease. Proc Natl Acad Sci U S A.1993;90: Strittmater WJ, Weisgraber KH, Goedert M, et al. Hypothesis: microtubule instability and paired helical filament formation in the Alzheimer disease brain are related to apolipoprotein E genotype. Exp Neurol. 1994;125: Folstein MF, Folstein SE, McHugh PR. Mini-Mental State: a practical method for grading the cognitive status of patients for the clinician. J Psychiatr Res. 1975; 12: Hachinski V, Illif L, Zilhka E, et al. Cerebral blood flow in dementia. Arch Neurol. 1975;32: American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition. Washington, DC: American Psychiatric Association; Brun A, Mann DMA, Englund L, et al. Consensus on clinical and neuropathological criteria for frontotemporal dementia. J Neurol Neurosurg Psychiatry. 1994; 57: McKeith IG, Galasko D, Kosaka K, et al. Consensus criteria for the clinical and pathological diagnosis of dementia with Lewy bodies: report of the CDLB International Workshop. Neurology. 1996;47: Chui HC, Victoroff JI, Margolin DM, et al. Criteria for the diagnosis of ischemic vascular dementia proposed by the State of California Alzheimer s Disease Diagnostic and Treatment Centers. Neurology. 1992;42: Hamilton M. Development of a rating scale for primary depressive illness. Br J Soc Psychol. 1976;6: Wenham PR, Price WH, Blundell G. Apolipoprotein E genotyping by one-stage PCR. Lancet. 1991;337: Therneau TM, Atkinson EJ. An Introduction to Recursive Partitioning Using the RPART Routines. Rochester, Minn: Mayo Foundation; Breiman L, Friedman JH, Olshen RA, Stone CJ. Classification and Regression Trees. Belmont, Calif: Wadsworth International Group; Saunders AM, Strittmatter WJ, Schmechel D, et al. Association of apolipoprotein E 4 allele with late-onset familial and sporadic Alzheimer s disease. Neurology. 1993;43: Saunders AM, Hulette C, Welsh-Bohmer KA, et al. Specificity, sensitivity, and predictive value of apolipoprotein E genotyping for sporadic Alzheimer s disease. Lancet. 1996;348: Galasko D, Clark C, Chang L, et al. Cerebrospinal fluid levels of tau protein are increased early in the course of Alzheimer s disease. Neurology. 1997;48: Riemenschneider M, Buch K, Schmolke M, Kurz A, Guder WG. Cerebrospinal protein tau is elevated in early Alzheimer s disease. Neurosci Lett. 1996;212: Schmechel DE, Saunders AM, Strittmatter WJ, et al. Increased amyloid peptide deposition in cerebral cortex as a consequence of apolipoprotein E genotype in late-onset Alzheimer disease. Proc Natl Acad Sci USA.1993;90: Rebeck GW, Reiter JS, Strickland DK, Hyman BT. Apolipoprotein E in sporadic Alzheimer s disease: allelic variation and receptor interactions. Neuron. 1993; 11: Johnson GVW, Seubert P, Cox TM, Motter R, Brown JP, Galasko D. The tau protein in human cerebrospinal fluid in Alzheimer s disease consists of proteolytically derived fragments. J Neurochem. 1997;68: Feany MB, Dickson DW. Neurodegenerative disorders with extensive tau pathology: a comparative study and review. Ann Neurol. 1996;40: Green J, Morris JC, Sandson J, McKeel DW, Miller JW. Progressive aphasia: a precursor of global dementia? Neurology. 1990;40: Turner RS, Kenyon LC, Trojanowski JQ, Gonatas N, Grossman M. Clinical, neuroimaging, and pathologic features of progressive nonfluent aphasia. Ann Neurol. 1996;39: Gearing M, Olson DA, Watts RL, Mirra SS. Progressive supranuclear palsy: neuropathologic and clinical heterogeneity. Neurology. 1994;44: Jellinger K, Danielczyk W, Fischer P, Gabriel E. Clinicopathological analysis of dementia disorders in the elderly. J Neurol Sci. 1990;95: Bancher C, Braak H, Fischer P, Jellinger K. Neuropathological staging of Alzheimer lesions and intellectual status in Alzheimer s and Parkinson s disease. Neurosci Lett. 1993;162: Whitehead AS, Bertrandy S, Finnan F, Butler A, Davey Smith G, Ben-Shiomo Y. Frequency of the apolipoprotein E 4 allele in a case-control study of early-onset Parkinson s disease. J Neurol Neurosurg Psychiatry. 1996;61: Gomez-Isla T, West HL, Rebeck GW, et al. Clinical and pathological correlates of apolipoprotein E 4 in Alzheimer s disease. Ann Neurol. 1996;70: Farrer LA, Abraham CR, Volicer L, et al. Allele 4 of apolipoprotein E shows a dose effect on age of onset of Pick s disease. Exp Neurol. 1995;136: Galasko D, Saitoh T, Xia Y, et al. The apolipoprotein E allele 4 is overrepresented in patients with the Lewy body varient of Alzheimer s disease. Neurology. 1994;44: