The Selective Muscarinic M1 Agonist AF102B Decreases Levels of Total A in Cerebrospinal Fluid of Patients with Alzheimers Disease

Roger M. Nitsch, MD,* Meihua Deng, PhD, Marsha Tennis, RN, David Schoenfeld, PhD, and John H. Growdon, MD
-Amyloid (A ) deposits in diffuse and compact senile plaques in the brain are one of the defining histopathological features of Alzheimers disease (AD). Preventing A deposition is a goal of drug therapy for AD, because excessive amounts of A may be toxic to neurons. In preclinical studies, activation of the muscarinic M1 receptor subtype inhibited A secretion from cultured cells. To determine whether a similar sequence occurs in human beings, we administered the selective M1 agonist AF102B to 19 AD patients and measured total A (A total) levels in cerebrospinal fluid (CSF) before and during treatment. A total levels in CSF decreased in 14 patients by 22%, increased in 3 patients, and were unchanged in 2 patients; the overall decrease in the group as a whole was statistically significant. To test the specificity of the M1 effect, we also measured the relative changes in A total levels in CSF during treatments in separate sets of AD patients with the acetylcholinesterase inhibitor physostigmine or the anti-inflammatory drug hydroxychloroquine. CSF A total levels did not change significantly in the 9 AD patients in the physostigmine protocol or in the 10 AD patients in the hydroxychloroquine study. These data provide evidence that the activation of M1 receptors reduces A levels in the CSF of AD patients. If this effect also occurs in brain, M1 agonists may have long-term therapeutic benefits by lowering amyloid in AD. Nitsch RM, Deng M, Tennis M, Schoenfeld D, Growdon JH. The selective muscarinic M1 agonist AF102B decreases levels of total A in cerebrospinal fluid of patients with Alzheimers disease. Ann Neurol 2000;48:913918

-Amyloid deposits in the brain are one of the defining histopathological features of Alzheimers disease (AD). According to the amyloid cascade hypothesis, early -amyloid deposits trigger a series of neurotoxic events that eventually lead to synapse loss and neuronal degeneration. Preventing -amyloid deposition is thus a leading target for drug development in AD. Brain -amyloid is composed of a 40-42amino acid peptide called A , which is cleaved from a large amyloid precursor protein (APP) by the action of - and -secretases. Activation of another enzyme, -secretase, splits APP within its A domain and prevents the formation of A . In a series of cell culture and brain slice experiments, we showed that APP processing via the -secretase pathway can be activated by cell surface receptors that are linked to a phospholipase C-diacylglycerol second messenger system.1 4 Cell surface receptors that mediate this re-

sponse include muscarinic M1 and M3 receptors, serotonergic 5-HT 2b and 2c receptors, metabotropic glutamate mGlu R1 receptors, and certain neuropeptide receptors. Hung and co-workers5 subsequently found that cholinergic stimulation of cells transfected with the M1 subtype simultaneously increased the soluble N-terminal APP ectodomain generated by -secretase cleavage while inhibiting A production. If such a sequence were to occur in human beings, administration of an M1 agonist would be expected to decrease levels of A in the central nervous system and possibly to slow the course of AD dementia. To test this hypothesis, we administered the selective M1 and M3 agonist AF102B to 19 patients with a clinical diagnosis of AD and measured cerebrospinal fluid (CSF) levels of soluble A before and during drug administration.

From the *Division of Psychiatry Research, University of Zurich, Zurich, Switzerland; and Department of Neurology and Department of Medicine, Biostatistics Center, Massachusetts General Hospital, Boston, MA. Received Apr 10, 2000, and in revised form Jul 19. Accepted for publication Jul 21, 2000.

Address correspondence to Dr Growdon, WACC 830, Massachusetts General Hospital, Fruit Street, Boston, MA 02114.

Copyright 2000 by the American Neurological Association

913

Subjects and Methods Subjects The 19 participants in the AF102B study all had a clinical diagnosis of AD.6 Of the 19 patients, 3 died from causes unrelated to the drug after the study was completed; AD was confirmed neuropathologically in all 3. There were 9 men and 10 women in this study (Table). Their mean age was 72.2 10.1 (mean SD) years, and the mean duration of their illness was 4.9 2.9 years. The severity of dementia was estimated by the Information, Memory, and Concentration (IMC) subscale of the Blessed Dementia Scale7 score, in which 0 to 2 is normal and rising scores up to a maximum of 37 indicate increasing dementia severity. The mean IMC score was 19.4 9.8. We determined each patients apolipoprotein E (ApoE) genotype by a polymerase chain reaction8 method: 5 had two E4 alleles, 10 had one E4 allele, 3 had no E4 alleles, and genotype analysis was not performed in 1. Medication AF102B is a selective muscarinic M1 and M3 agonist that is a conformationally rigid analogue of acetylcholine.9 In cell culture, AF102B activates -secretase APP processing as indexed by an increase in APP secretion (APPs).10 AF102B is active in the brain as judged by improved spatial learning in mice11 and reversal of impaired maze performance in aged rats.12 Peak effects occur within 60 minutes after oral administration; brain concentrations peak 30 minutes after administration. In human beings, AF102B plasma concentrations peak at 1 to 2 hours after oral administration; the plasma half-life ranges between 2 and 5 hours. Protocol Subjects (with spousal consent) agreed to participate in this study according to the provisions of the protocol approved by the Massachusetts General Hospital (MGH) Institutional Review Board. Patients were admitted to the MGH General Clinical Research Center for lumbar punctures on two occasions: at baseline before treatment and during the fourth week of maximal
Table. Clinical Characteristics of Patients in the Three Drug Trials Sex M AF102B (n 19) Hydroxychloroquine (n 10) Physostigmine (n 9)
a

treatment. AF102B was supplied by Snow Brand Milk Products Company (Tokyo, Japan) and was administered in escalating doses with 20 mg three times daily during week 1, 40 mg three times daily during week 2, 60 mg three times daily during week 3, and 80 mg three times daily during week 4. Spinal fluid was collected according to conventional lumbar puncture techniques. CSF was immediately divided into 500- l aliquots and frozen on dry ice at the bedside. Samples were stored at 80C until biochemical assay. We used a sensitive and specific sandwich enzyme-linked immunosorbent assay (ELISA) method to measure total A (A total A 40 A 42) in 25 l of CSF. The antibodies were 22C4, a monoclonal antibody raised against a KLH-conjugated peptide corresponding to the C-termini of A 40 and A 42 for capture and biotinylated 6E10 (purchased from Senetec, St Louis, MO) directed against their N-termini as the detection antibody. We used 25 l of CSF so as to fall into the linear range (0.6 20 ng) of the assay, which was reproducible with intra-assay variances of less than 10%. In 11 cases in which we had sufficient CSF, we also analyzed A total by Western blot analysis with 6E10 and densitometry. To determine the specificity of AF102B on CSF A levels, we administered two other drugs, hydroxychloroquine or physostigmine, to separate sets of AD patients. There were 10 AD patients (7 men and 3 women) in the hydroxychloroquine study. They had a mean age of 68.3 6.9 years, a mean duration of illness of 3.8 1.9 years, and a mean IMC score of 19.5 13.1 (see Table). There were 9 AD patients in the physostigmine protocol (3 men and 6 women). They had a mean age of 67.7 9.3, a mean duration of illness of 3.8 1.5 years, and a mean IMC score of 13.2 8.9. Hydroxychloroquine is an anti-malarial and anti-rheumatic drug that is also effective in lupus erythematosus and rheumatoid arthritis. The mechanism for its action in collagen diseases is not clearly known. We tested hydroxychloroquine because of its

F 10 3 6

Age, yr (mean SD) 72.2 68.3 67.7 10.1 6.9 9.3

Duration of illness (yr) (mean SD) 4.9 3.8 3.8 2.9 1.9 1.5

IMC Score (mean SD) 19.4 19.5 13.2 9.8 13.1 8.9

Baseline CSF A total (mean SD) 22.8 23.5 21.2 7.5 10.1 7.9

Treatment CSF A total (mean SD) 19.9 24.0 21.0 7.9a 8.1 6.1

9 7 3

0.004. male; F female; IMC Information, Memory, and Concentration subtest of the Blessed Dementia Scale7; CSF cerebrospinal fluid.

914

Annals of Neurology

Vol 48

No 6

December 2000

anti-inflammatory property and also because it inhibits lysosomal protease function and thereby may interfere with APP processing.13 Subjects took 200 mg of hydroxychloroquine twice daily, which was supplied by Sanofi Pharmaceuticals (New York, NY). Physostigmine is an acetylcholine esterase inhibitor. The net effect of its administration is to increase cholinergic neurotransmission by inhibiting the enzymatic hydrolysis of endogenously released acetylcholine. Physostigmine was provided by Forest Laboratories (New York, NY) as a sustained-release tablet. Of the 9 patients, 2 took a total of 9 mg/day, 1 took 12 mg/day, and 6 took 15 mg/day. As in the AF102B protocol, each patient underwent a baseline lumbar puncture and then a repeat lumbar puncture during the fourth week of treatment. In all protocols, CSF samples were collected in 500- l aliquots, frozen immediately at the bedside, stored at 80C, and assayed according to identical procedures. The primary outcome measure of this study was a change in CSF A levels during treatment. In these experiments, changes in CSF A levels served as surrogate markers to monitor drug effects, because A levels in CSF remain stable over time14 in the untreated state. CSF A total levels were not used as biomarkers to diagnose AD, because mean values overlap with those of normal control subjects.15 A two-tailed t test was used to compare mean changes before and after drug administration. Analyses of covariates were employed to examine whether the changes correlated with age, gender, dementia severity, and ApoE genotype. Results Individual CSF levels of A at baseline ranged from 9.13 to 31.27 ng/ml (mean SD, 22.8 7.5 ng/ml). During treatment, these levels decreased by a median of 22% in 14 patients, increased in 3, and were unchanged in 2 (mean SD: 19.9 7.9 ng/ml) (Fig 1). The mean decrease of 2.9 0.9 ng/ml during treatment was statistically significant ( p 0.004). The change in A total was not related to the dose of medication; it did not correlate with age, sex, dementia severity, or ApoE genotype. We verified the identity of A total in CSF by Western immunoblot densitometry. When quantitated by densitometry, there was good correspondence (r 0.937) between the relative change in A total detected by the ELISA assay and by Western immunoblotting (Fig 2). In contrast to A total, levels of A 4216 and soluble N-terminal APP ectodomain derivatives measured by ELISA and Western immunoblots did not change in CSF in a consistent or statistically significant way during drug treatment (data not shown). The most common side effects of AF102B were excessive sweating, gastrointestinal complaints, and headache. Many of these symptoms may result from activating M3 receptors. In most cases, symptoms subsided

Fig 1. AF102B significantly (p 0.004) decreased cerebrospinal fluid levels of ABtotal in Alzheimers disease patients. Each bar represents the percentage of change from baseline during treatment in an individual patient.

several hours after medication ingestion. One patient withdrew from the study because of persistent fever, chills, and abdominal cramps. A second patient was hospitalized for persistent vomiting that began 27 days after completing the AF102B study. Gastric endoscopy revealed esophagitis that resolved with conservative medical treatment. CSF A total levels did not change significantly in AD patients who took hydroxychloroquine or physostigmine. Mean levels of A total in the hydroxychloroquine study at baseline were 23.5 10.1 and 24.0 8.1 ng/ml during the final week of treatment. Among the 10 hydroxychloroquine patients, CSF
Fig 2. The correlation between enzyme-linked immunosorbent assay and Western immunoblot methods for measuring A total levels in cerebrospinal fluid is excellent (r 0.937). Each treatment point represents the relative change (baseline A A ) in a single patient. Points below 1.0 are those in which levels are decreased, and points above 1.0 are those in which levels are increased.

Nitsch et al: AF102B Decreases Total A in CSF

915

A total levels increased in 5, decreased in 2, and were unchanged in 3. Mean CSF A total levels in the physostigmine study were 21.2 7.9 ng/ml at baseline and 21.0 6.1 ng/ml during treatment. Among the 9 physostigmine patients, CSF A total levels increased in 4, decreased in 4, and were unchanged in 1. Discussion The results of this study indicate that the selective M1 agonist AF102B decreased levels of A total in the CSF of most AD patients. This result is consistent with our underlying scientific rationale for administering M1 and M3 agonists, which was originally based on in vitro experiments with transfected cells.1,5 This result is also consistent with an in vivo study on cholinergic effects of APP processing in rats. Lin and co-workers17 administered the muscarinic agonist RS86 to normal rats, aged rats, and rats with basal forebrain cholinergic deficits and measured levels of cell-associated APP in brain tissue and the APPs in CSF. They found that treatment decreased APP levels in neocortex and hippocampus and increased levels of APPs in CSF in aged as well as cholinergically denervated rats. That study was the first in vivo demonstration that increased cholinergic neurotransmission can enhance nonamyloidogenic APP processing. Our finding with AF102B extends the preclinical results of cholinergic influences on APP processing to human beings and establishes that selective M1 and M3 agonists can decrease A levels in the CSF of AD patients. In contrast to the RS86 finding in animals,17 APPs levels did not increase during treatment with AF102B. An obvious explanation for decreased A with no associated increase in APPs- would be inhibition of -secretase. To our knowledge, the possibility that AF102B inhibits -secretase activity has never been examined. Clinical studies cannot establish biological mechanisms unequivocally, but our finding should prompt further investigations into the biochemical effects of AF102B on -secretase. For example, among the many signaling events coupled to muscarinic receptor stimulation is the phosphorylation of presenilin 1 (PS-1).18 PS-1 is either a -secretase or is closely linked to its activity.19 Phosphorylating PS-1 may well be involved in regulating its endoproteolysis, facilitating its assembly into the PS-1 enzymatic complex and leading ultimately to its biological activity. In our study, the main outcome measure was A total. The assay for A total captures and measures both A 40 and A 42 moieties. Most A circulating in the CSF is A 40, whereas A 42 accounts for only about 10%. The fact that A 42 levels did not change during treatment may reflect its relatively small contribution to the total A measurement. Alternatively, it is possible that AF102B exerts a selective drug effect on - and

-secretase processing of APP, with predominant inhibition of A 40 production. To determine whether the decrease in A total induced by AF102B is specific to M1 agonists or occurs with other pharmacological treatments, we tested two other drugs in separate sets of AD patients: the acetylcholine esterase inhibitor physostigmine and the antiinflammatory drug hydroxychloroquine. Physostigmine given in doses that improved cognition in clinical studies20,21 failed to change CSF A total levels in our study. This finding is consistent with our prior preclinical studies, in which we found that physostigmine did not alter APP processing.22 Although physostigmine increases cholinergic neurotransmission, the effect is nonspecific and extends to all muscarinic receptors. Whereas stimulation of M1 and M3 receptors increases -secretase APP processing, activation of M2 receptors does not1; simultaneous stimulation of M2 receptors blocks the M1 stimulation effect.22 Anti-inflammatory drugs are being evaluated as a treatment for AD following the observation of microglia activation surrounding amyloid plaques. Hydroxychloroquine administration did not significantly alter A total CSF levels in AD patients. The fact that neither of these drugs influenced CSF levels of A total emphasizes the specificity of the AF102B effect on APP processing. The goal of our study was to determine whether an M1 muscarinic agonist would alter APP processing in human beings as hypothesized from preclinical studies. The scientific rationale behind this study is therefore much different from prior clinical trials with M1 agonists that were based solely on a neurotransmitter replacement strategy.23 The initial rationale for developing muscarinic agonists was based on increasing acetylcholine neurotransmission according to the tenets of the cholinergic hypothesis of memory dysfunction.24 Because postsynaptic receptor sites, especially M1 receptors, are largely preserved in the face of significant decrements in presynaptic indices of cholinergic neurons,25 it was hoped that M1 agonists would correct the cholinergic neurotransmitter deficit in AD and thereby restore memory capabilities. To date, predictions of short-term symptomatic benefit have not been realized. Of several M1 agonists that have entered clinical development, xanomeline was the most carefully studied. In a large clinical trial, its administration nearly abolished psychotic signs in AD patients such as delusions and hallucinations but did not improve cognition.26 In our study, we sought changes in APP processing in CSF; we did not expect, or measure, shortterm changes in behavior or cognition. The results of our study indicate that compounds stimulating M1 and M3 receptors decrease A secretion in AD patients and suggest that this class of drugs should therefore be re-evaluated as a treatment for AD. In contrast to simple neurotransmitter replacement, our study raises the

916

Annals of Neurology

Vol 48

No 6

December 2000

possibility that muscarinic agonists that stimulate M1 or M3 receptors may prevent or slow neurodegeneration by lowering amyloid in AD patients. The development of drugs that either block A production, prevent its aggregation and deposition in brain tissue, or enhance its clearance is a major theme in AD therapeutics. In a transgenic mouse model, Schenk and co-workers27 discovered that immunization with the A 42amino acid peptide sequence prevented amyloid deposition in the brain. Moreover, drugs that inhibit - or -secretase processing of APP may also prevent A deposition. The development of this line of therapeutics was greatly facilitated by the identification of -secretase28 and by the possibility that PS-1 is a -secretase.19 In addition to these approaches, neurotransmitter agonists that activate -secretase processing may be useful to reduce A formation. As these preclinical experiments lead to drugs for clinical trials, it is important to verify that these treatments are acting as proposed in human beings. The usual outcomes for drug studies in AD rely on short-term changes in cognitive test performance29,30 or on long-term behavioral outcomes.31 In future trials, it would be advantageous to have an accepted biomarker of AD pathology used as a surrogate outcome for drug effect; direct measures of amyloid in the brain may be performed in mice but are impossible in human beings. Amyloid deposits in the brain do not correlate with the duration or severity of dementia32 and A levels in CSF apparently do not change across clinical estimates of dementia severity or over time.14 Our study with AF102B indicates that A measurements in CSF do change with treatment and can therefore be used as a reasonable correlate of drug action. Thus, drugs that are proposed to decrease A production such as - or -secretase inhibitors would be expected to decrease A levels in CSF, whereas drugs that increase A clearance from the brain might actually increase A levels in CSF or plasma. Our experience in this study argues in favor of measuring a full panel of APP derivatives, including p3, so as to capture unexpected complexities of drug effects on APP processing in human patients. Measuring a change in any APP derivative in CSF can provide supportive evidence of a drugs mode of action; druginduced changes in CSF may not accurately reflect amyloid formation in the brain, however. Whether amyloid-reducing therapies prevent or even slow the progressive course of AD dementia continues to rely on clinical and behavioral measures.
This study was supported by grants from the National Institutes of Health (P50 AG 05134) and the Snow Brand Milk Products Company. The MGH Clinical Research Center is supported by National Institutes of Health grant 5-MO1-01066-23. The authors are not employed by Snow Brand Milk Products Company, do not serve as consultants, and do not have any financial interest in the company.

We thank Dr Dominic Walsh for the A 42 measurements and Dr Manfred Brockhaus for the APPs ELISA measurements.

References
1. Nitsch RM, Slack BE, Wurtman RJ, Growdon JH. Release of Alzheimer amyloid precursor derivatives stimulated by activation of muscarinic acetylcholine receptors. Science 1992;258: 304 307 2. Nitsch RM, Farber SA, Growdon JH, Wurtman RJ. Release of amyloid -protein precursor derivatives from hippocampal slices of electrical depolarization. Proc Natl Acad Sci USA 1993;90:51915193 3. Nitsch RM, Deng A, Wurtman RJ, Growdon JH. Metabotropic glutamate receptor subtype mGluR1a stimulates the secretion of the amyloid beta protein precursor ectodomain. J Neurochem 1997;69:704 712 4. Nitsch RM, Kim C, Growdon JH. Vasopressin and bradykinin regulate secretory processing of the amyloid protein precursor of Alzheimers disease. Neurochem Res 1998;23:807 814 5. Hung AY, Haass C, Nitsch RM, et al. Activation of protein kinase C inhibits cellular production of the amyloid -protein. J Biol Chem 1993;268:22959 22962 6. McKhann G, Drachman D, Folstein M, et al. Clinical diagnosis of Alzheimers disease: report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimers Disease. Neurology 1984;34:939 944 7. Blessed G, Tomlinson BE, Roth M. The association between quantitative measures of dementia and of senile change in the grey matter of elderly subjects. Br J Psychiatry 1968;114:797 811 8. Rebeck GW, Reiter JS, Strickland DK, Hyman BT. Apolipoprotein E polymorphism and Alzheimers disease: allelic variation and receptor interactions. Neuron 1993;11:575580 9. Fisher A, Barak D. Progress and perspectives in new muscarinic agonists. Drug News and Perspectives 1994;7:453 464 10. Haring R, Gurwitz D, Barg J, et al. Amyloid precursor protein secretion via muscarinic receptors: reduced desensitization using the M1-selective agonist AF102B. Biochem Biophys Res Commun 1994;203:652 658 11. Vincent GP, Sepinwall J. AF102B, a novel M1 agonist, enhanced spatial learning in C57BL/10 mice with a long duration of action. Brain Res 1992;597:264 268 12. Brandeis R, Dachir S, Sapir M, et al. Reversal of age-related cognitive impairments by an M1 cholinergic agonist, AF102B. Pharmacol Biochem Behav 1990;36:89 95 13. Caporaso GL, Gandy SE, Buxbaum JD, Greengard P. Chloroquine inhibits intracellular degradation but not secretion of Alzheimer ( /A4 amyloid precursor protein. Proc Natl Acad Sci USA 1992;89:22522256 14. Kanai M, Matsubara E, Isoe K, et al. Longitudinal study of cerebrospinal fluid levels of tau, A 1-40, and A 1-42(43) in Alzheimers disease: a study in Japan. Ann Neurol 1998;44: 1726 15. Nitsch RM, Rebeck GW, Deng M, et al. Cerebrospinal fluid levels of amyloid -protein in Alzheimers disease: increased correlation with severity of dementia and effect of apolipoprotein E genotype. Ann Neurol 1995;37:512518 16. Johnson-Wood K, Lee M, Motter R, et al. Amyloid precursor processing and A 42 deposition in a transgenic mouse model of Alzheimers disease. Proc Natl Acad Sci USA 1997;94:1550 1555 17. Lin L, Georgievska B, Mattsson A, Isacson O. Cognitive changes and modified processing of amyloid precursor protein

Nitsch et al: AF102B Decreases Total A in CSF

917

18.

19.

20. 21.

22.

23. 24.

25.

in the cortical and hippocampal system after cholinergic synapse loss and muscarinic receptor activation. Proc Natl Acad Sci USA 1999;96:12108 12113 Walter J, Grunberg J, Capell A, et al. Proteolytic processing of the Alzheimer disease-associated presenilin-1 generates an in vivo substrate for protein kinase C. Proc Natl Acad Sci USA 1997;94:5349 5354 Wolfe MS, De Los Angeles J, Miller DD, et al. Are presenilins intramembrane-cleaving proteases? Implications for the molecular mechanism of Alzheimers disease. Biochemistry 1999;38: 1122311230 Stern Y, Sano M, Mayeux R. Effects of oral physostigmine in Alzheimers disease. Ann Neurol 1987;22:306 310 Thal LJ, Ferguson JM, Mintzer J, et al. A 24-week randomized trial of controlled-release physostigmine in patients with Alzheimers disease. Neurology 1999;52:1146 1152 Farber SA, Nitsch RM, Schultz JG, Wurtman RJ. Regulated secretion of -amyloid precursor protein in rat brain. J Neurosci 1995;15:74427450 Growdon JH. Muscarinic agonists in Alzheimers disease. Life Sci 1997;60:993998 Bartus RT, Dean RL III, Beer B, Lippa AS. The cholinergic hypothesis of geriatric memory dysfunction. Science 1982;217: 408 417 Davies P, Verth AH. Regional distribution of muscarinic ace-

26.

27.

28.

29.

30.

31.

32.

tylcholine receptors in normal and Alzheimers type dementia brains. Brain Res 1978;138:385392 Bodick NC, Offen WW, Levey AI, et al. The effects of xanomeline, a selective muscarinic receptor agonist, on cognitive function and behavioral symptoms of Alzheimer disease. Arch Neurol 1997;54:465 473 Schenk D, Barbour R, Dunn W, et al. Immunization with amyloid- attenuates Alzheimer-diseaselike pathology in the PDAPP mouse. Nature 1999;400:173177 Vassar R, Bennett BD, Safura B-K, et al. -secretase cleavage of Alzheimers precursor protein by the transmembrane aspartic protease BACE. Science 1999;286:735741 Knapp MJ, Knopman DS, Solomon PR, et al. A 30-week randomized controlled trial of high-dose tacrine in patients with Alzheimers disease. JAMA 1994;271:985991 Rogers SL, Friedhoff LT. The efficacy and safety of donepezil in patients with Alzheimers disease: results of a US multicenter, randomized, double-blind, placebo-controlled trial. Dementia 1996;7:293303 Sano M, Ernesto C, Thomas RG, et al. A controlled trial of selegiline, alpha-tocopherol, or both as treatment for Alzheimers disease. N Engl J Med 1997;336:1216 1222 Gomez-Isla T, West HL, Rebeck GW, et al. Clinical and pathological correlates of apolipoprotein E 4 in Alzheimers disease. Ann Neurol 1996;39:6270

918

Annals of Neurology

Vol 48

No 6

December 2000