Journal of Alzheimer’s Disease 17 (2009) 661–680 IOS...

Journal of Alzheimer’s Disease 17 (2009) 661–680 661DOI 10.3233/JAD-2009-1087IOS Press

Caffeine Reverses Cognitive Impairment andDecreases Brain Amyloid-β Levels in AgedAlzheimer’s Disease Mice

Gary W. Arendasha,b,∗, Takashi Moric, Chuanhai Caoa,b,d, Malgorzata Mamcarza,b,Melissa Runfeldta,b, Alexander Dicksona,b, Kavon Rezai-Zadehe, Jun Tane, Bruce A. Citronf,g,Xiaoyang Lina,d, Valentina Echeverriaf,g and Huntington Pottera,b,d,h

aFlorida Alzheimer’s Disease Research Center, University of South Florida, Tampa, FL, USAbDepartment of Cell Biology, Microbiology, and Molecular Biology, University of South Florida, Tampa, FL, USAcDepartments of Medical Science and Pathology, Saitama Medical Center and Saitama Medical University,Kawagoe, Saitama, JapandThe Byrd Alzheimer’s Center and Research Institute, Tampa, FL, USAeDepartment of Psychiatry and Behavioral Medicine, University of South Florida, Tampa, FL, USAfDepartment of Molecular Medicine, University of South Florida, Tampa, FL, USAgBay Pines VA Healthcare System, Bay Pines, FL, USAhSuncoast Gerontology and Alzheimer’s Center, University of South Florida College of Medicine, Tampa, FL, USA

Accepted 20 January 2009

Abstract. We have recently shown that Alzheimer’s disease (AD) transgenic mice given a moderate level of caffeine intake (thehuman equivalent of 5 cups of coffee per day) are protected from development of otherwise certain cognitive impairment and havedecreased hippocampal amyloid-β (Aβ) levels due to suppression of both β-secretase (BACE1) and presenilin 1 (PS1)/γ-secretaseexpression. To determine if caffeine intake can have beneficial effects in “aged” APPsw mice already demonstrating cognitiveimpairment, we administered caffeine in the drinking water of 18–19 month old APPsw mice that were impaired in workingmemory. At 4–5 weeks into caffeine treatment, those impaired transgenic mice given caffeine (Tg/Caff) exhibited vastly superiorworking memory compared to the continuing impairment of control transgenic mice. In addition, Tg/Caff mice had substantiallyreduced Aβ deposition in hippocampus (↓40%) and entorhinal cortex (↓46%), as well as correlated decreases in brain soluble Aβlevels. Mechanistically, evidence is provided that caffeine suppression of BACE1 involves the cRaf-1/NFκB pathway. We alsodetermined that caffeine concentrations within human physiological range effectively reduce active and total glycogen synthasekinase 3 levels in SweAPP N2a cells. Even with pre-existing and substantial Aβ burden, aged APPsw mice exhibited memoryrestoration and reversal of AD pathology, suggesting a treatment potential of caffeine in cases of established AD.

Keywords: Alzheimer’s disease, Alzheimer’s transgenic mice, amyloid-β, caffeine, cognitive impairment, memory, treatment

∗Address for correspondence: Gary W. Arendash, Ph.D., Depart-ment of Cell Biology, Microbiology, & Molecular Biology, BSF218,University of South Florida, Tampa, Florida 33620, USA. Tel.:+1 813 974 1584; Fax: +1 813 974 1614; E-mail: [email protected].

INTRODUCTION

Caffeine is the most widely consumed psychoactivesubstance known to man, being primarily consumedin coffee, tea, and soft drinks. As a brain stimu-lant, its primary actions are to increase cerebral energymetabolism, cortical activity, and extracellular levels ofacetylcholine, collectively resulting in enhanced alert-

ISSN 1387-2877/09/$17.00 2009 – IOS Press and the authors. All rights reserved

UNDER EMBARGO UNTIL JULY 6, 2009, 00:00 CET

662 G.W. Arendash et al. / Caffeine Reverses Memory Impairment in AD Mice

ness [1]. In addition to these well-documented stim-ulatory effects of caffeine, epidemiology/longitudinalstudies in humans have suggested that caffeine is pro-tective against a variety of diseases associated with ag-ing including Parkinson’s disease, Type 2 Diabetes, andliver disease [2–5]. More recently, several longitudinalstudies have reported that daily caffeine intake, equiv-alent to 3 or more cups of coffee, reduces cognitivedecline in “non-demented” elderly men and women [6,7]. Of direct relation to Alzheimer’s disease (AD), an-other recent study involving a 21 year follow-up periodfound that midlife intake of 3–5 cups of coffee substan-tially reduced the risk of AD in later life [8]. Moreover,AD patients were found to have consumed markedlyless caffeine during the 20 years preceding AD diag-nosis when compared to age-matched individuals with-out AD [9]. These cognition-based human studies sug-gest that long-term caffeine intake may protect againstaged-associated memory impairment and AD [10].

Although human epidemiologic/longitudinal studiesare insightful, they cannot definitively isolate caffeineintake from other lifestyle choices that potentially affectcognition. Therefore, we recently performed a high-ly controlled study in AD transgenic (APPsw) mice,which develop high levels of brain amyloid-β (Aβ) andwidespread cognitive impairment as they age [11,12].Long-term caffeine administration that began in youngadulthood protected these APPsw mice against other-wise certain cognitive impairment, while also limitingtheir brain Aβ levels [13]. We further found that thislong-term caffeine administration suppressed brain Aβproduction by reducing expression of both Presenilin1 (PS1) and β-secretase (BACE1). Because the bene-ficial effects were achieved by a moderate daily levelof caffeine in drinking water (the human equivalent of5 cups of coffee or 500 mg caffeine per day), we havesuggested that a modest amount of daily caffeine maydelay or reduce the risk of AD [13].

In the present study, we first aimed to determine ifcaffeine intake can have beneficial effects in “aged”APPsw mice already exhibiting cognitive impairment.Therefore, at several months into caffeine treatment,were-evaluated cognitive performance in aged APPsw anddetermined the impact on both Aβ levels/Aβ deposi-tion and synaptophysin immunostaining in cognitivelyimportant brain areas.

Second, we wished to develop greater insight into themechanism(s) of the ability of caffeine to reduce Aβand thus explored the potential involvement of the Raf-1 inflammatory pathway and glycogen synthase kinase3 (GSK-3) in caffeine’s mechanism of action in APPsw

mice. Raf-1 is hyperactivated in AD brains [14,15],evidently due to an Aβ-induced decrease in the activityof proteins (e.g., Ras, PKA) that regulate Raf-1 activ-ity. GSK-3 is a constitutively active serine/threoninekinase that is highly expressed in the central nervoussystem (CNS) [16–18]. Dysregulation of GSK-3 is anessential component of AD pathogenesis, playing anintegral role in PS1/γ-secretase activity, spatial learn-ing deficits, Aβ-induced toxicity, and tau phosphoryla-tion [19–21].

Third, we administered caffeine throughout adult lifeto non-transgenic (NT) mice,with cognitive assessmentin older age to determine if: 1) life-long caffeine intakecan provide cognitive benefits to NT (normal) miceand consequently, 2) generalized, non-amyloidogenicmechanisms of caffeine action (such as increased cere-bral blood flow or enhanced brain glucose utilization)could be important for its therapeutic effects in APPswmice.

MATERIAL AND METHODS

Animals

A total of 55 mice were included in these studies.Each mouse had a mixed background of 56.25% C57,12.5% B6, 18.75% SJL, and 12.5% Swiss-Webster.All mice were derived from a cross between heterozy-gous mice carrying the mutant APPK670N, M671Lgene (APPsw) with heterozygous PS1 (Tg line 6.2)mice to obtain F4 and F5 generation mice consistingof APP/PS1, APPsw, PS1, and NT genotypes. Afterweaning and genotyping, only APPsw and NT micewere selected for behavioral testing and/or caffeine ad-ministration. All mice were maintained on a 12 h darkand 12 h light cycle with ad libitum access to rodentchow and water or caffeinated water. All animal proce-dures were performed in AAALAC-certified facilitiesunder protocols approved by Institutional Animal Careand Use Committees at the University of South Florida.

General protocol

Three separate in vivo studies were performed asfollows:

Study 1: At 18–19 months of age, APPsw and NTcontrols were pre-tested in the radial arm water maze(RAWM) task, wherein APPsw mice were shown tobe impaired in working memory. Impaired transgenic(Tg) mice were divided into two groups, equally bal-


G.W. Arendash et al. / Caffeine Reverses Memory Impairment in AD Mice 663

anced by behavioral performance, with half of Tg micestarting on caffeine administered in their drinking water(0.3 mg/ml; Sigma, St. Louis, MO) and the other halfremaining on standard tap water. At 4–5 weeks intocaffeine treatment, Tg caffeine-treated (Tg/Caff; n =6), as well as Tg control (Tg; n = 7), and NT control(NT; n = 8) mice were re-tested in the RAWM, as wellas in several other behavioral tasks. All animals wereeuthanatized immediately following completion of be-havioral testing at 20–21 months of age. Their brainswere processed for determination of total Aβ deposi-tion, soluble Aβ1−40 and Aβ1−42, and synaptophysinimmunochemistry.

Study 2: Beginning at 9 months of age, APPsw mice(n = 6) were gavaged with caffeine (1.5 mg/200 µl)twice daily for two weeks. Gavage administration ofcaffeine was done in this study because: 1) the treat-ment period was relatively short-term compared to thatof Studies 1 and 3, and 2) gavage treatment is more pre-cise in providing exactly the amount of caffeine desiredcompared to ad libitum administration of caffeine indrinking water. Control APPsw mice (n = 5) and NTmice (n = 4) of the same age received gavage with wa-ter vehicle instead. All mice were sacrificed immedi-ately following the treatment period, their hippocampusdissected out and stored at −80◦C until neurochemicalassay of pcRaf-1 and PKA activity analyses.

Study 3: In a “behavior-only” study, NT (normal)mice began receiving caffeine in their drinking water(0.3 mg/ml) at 51/2 months of age (NT/Caff; n = 8)or were continued on standard water (NT; n = 11).Between 15–16 months of age, and at 10 months intocaffeine treatment, all animals were tested in our com-prehensive 6-week behavioral battery that included fivecognitive-based tasks.

Caffeine treatment

For Studies 1 and 3, caffeine was administered inthe drinking water (0.3 mg/ml) of Tg or NT mice. Onaverage, mice drink 5 ml per day, yielding a daily doseof 1.5 mg caffeine to each mouse. According to ourprior study’s calculations accommodating for size andmetabolic rate differences [13], this daily dose in amouse is equivalent to approximately 500 mg (5 cupsof coffee) daily caffeine intake by a human. The caf-feinated water was changed two times a week to en-sure caffeine remained fully dissolved at the appropri-ate concentration. Control APPsw and NT mice weregiven ad libitum access to untreated tap water that wasalso changed twice weekly to ensure freshness. Caf-

feine treatment was continued throughout behavioraltesting and until euthanasia. All groups consisted ofapproximate equal numbers of male and female mice.

Behavioral testing

In Study 1, aged 18–19 month mice were pre-treatment tested in the RAWM task, and then behav-iorally tested again at 4–5 weeks into treatment inthe following tasks: open field activity, balance beam,string agility, Y-maze, elevated plus maze, RAWM, andplatform recognition. In Study 3, NT mice were test-ed in our full behavioral battery beginning at 15–16months of age (e.g., at 10 months into treatment). Al-though detailed protocols for all behavioral tasks havebeen described previously [22–24], a brief descriptionof each task is provided below:

Open Field Activity (Sensorimotor-based task; 1day). As a test of activity and exploratory behavior,each animal was placed in an open black box paintedwith perpendicular lines demarcating 16 squares. Thetotal number of line crossings over a single 5-minutetrial was recorded through visual observation.

Balance Beam (Sensorimotor-based task; 1 day). Asa test of vestibular and general motor function, eachanimal was placed at the center of a suspended beamand released. The average balance time from threesuccessive trials was recorded. The surface below thebeam had a thick sponge cushion to protect mice thatfall off the beam.

String Agility (Sensorimotor-based task; 1 day). Asa test of agility and grip capacity, animals were permit-ted to grasp a suspended string only by their forepawsand then released. During a single 60-second trial, eachanimal was assessed using a 0–5 rating system (0 =animal unable to remain on string; 1 = hangs by twoforepaws; 2 = attempts to climb onto string; 3 = twoforepaws and one or both hind paws around string; 4= four paws and tail around string, with lateral move-ment; 5 = escape to one of the two poles suspendingthe string).

Y-maze spontaneous alternation (Sensorimotor- andCognitive-based task; 1 day). To measure generalactivity and basic hippocampal-dependent mnemonicfunction, mice were allowed to explore a black Y-mazewith 3 arms. Each arm measures 21 × 4 cm with40 cm high walls. Mice were placed in the center of themaze facing the center area and allowed to explore for5 minutes, with the number and sequence of arm choic-es being recorded. General activity was measured asthe total number of arm entries, while basic mnemonic



function was determined as a percent spontaneous al-ternation (the ratio of arm choices differing from theprevious two choices divided by the total number ofentries).

Elevated Plus Maze (Anxiety-based task; 1 day). Tomeasure anxiety/emotionality, mice were placed in thecenter of an elevated plus maze 82 cm above the floor.The maze consists of two opposite “open” and two op-posite “closed” arms, each 30 × 5 cm; 15 cm highblack aluminum walls surround the closed arms. Themice were placed in the 5 × 5 cm maze center, facinga closed arm, and were allowed to explore for 5 min-utes. The total number of closed arm entries, open armentries and total time (seconds) spent in the open armswas recorded. We have found that time spend in openarms is the purest measure of anxiety in this task [22].

Morris Water Maze (Cognitive-based task measur-ing spatial learning and reference memory; 15 days).The floor of a 100 cm circular pool was divided intofour quadrants, with a 9 cm platform submerged in thegoal quadrant. Surrounding the pool were a variety of3-D and 2-D visual cues. Fourteen consecutive daysof acquisition were performed, with each of four trialsper day beginning with the animal being placed intoeach of the four quadrants to initiate a 60-second tri-al. Thus, a single training session consisted of foursuccessive trials, with a 30 second stay period (e.g.,the animal is allowed to remain on the platform for 30seconds after finding it or being guided to it at the endof the trial). Average latency to find the submergedplatform was obtained through visual observation. Onthe day following acquisition testing, memory reten-tion was evaluated in a single 60-second probe trial inwhich the submerged platform was removed and theanimal released from the quadrant opposite the formerplatform-containing quadrant. Percent of time spent ineach quadrant, number of annulus crossings, and swimspeed were all determined from video tracking records.

Circular Platform (Cognitive-based task measuringspatial reference learning/memory; 8 days). A walled69 cm circular platform, with 16 equidistantly spacedholes along its periphery, was encircled by a black cur-tain. Visual cues, located on the black curtain and plat-form walls can be used by the animal to find the onehole through which it can escape the platform surfaceto avoid the aversive stimuli of bright lights and fanwind. During a single 5-minute maximum daily trial,the total number of errors (head pokes into non-escapeholes) and latency to find the escape hole were record-ed. Although the escape hole remained constant forany given animal over the 8 days of testing, it was re-

located after each animal’s trial to control for olfactorycues. Also, to control for olfactory cues, the maze wascleaned with a dilute vinegar solution following eachanimal’s trial.

Platform Recognition (Cognitive-based task measur-ing identification and recognition; 4 days). The plat-form recognition task measures the ability to search forand identify/recognize a variably placed elevated plat-form. It was run after the Morris Maze task (Study2), or immediately following the RAWM task (Study1) and in the same pool as these two tasks. Thus, theplatform recognition task requires animals to ignore thespatial cues present around the pool and switch froma spatial to an identification/recognition strategy – it isnot a task of visual acuity alone in our paradigm, withcognitive-based performance directly linked to levelsof brain Aβ [25]. Mice were given four successivetrials per day over a 4-day period. Latencies to find anelevated platform (9 cm diameter), bearing a prominentcone-shaped styrofoam ensign on a wire pole, were de-termined. For each trial (60 second maximum), ani-mals were placed into the pool at the same location andthe platform was moved to a different one of four possi-ble locations. For statistical analysis, escape latenciesfor all four daily trials were averaged.

Radial arm water maze (RAWM) (Cognitively-basedtask measuring working memory). For the radial armwater maze task of spatial working memory, an alu-minum insert was placed into a 100 cm circular pool tocreate 6 radially distributed swim arms emanating froma central circular swim area. An assortment of 2-Dand 3-D visual cues surrounded the pool. For Study1 involving aged Tg mice, the number of errors priorto locating which one of the 6 swim arms contained asubmerged escape platform (9 cm diameter) was deter-mined for 5 trials/day over 6 days of pre-treatment test-ing and 9 days of post-treatment testing; 3-day blockswere used to facilitate statistical analysis. For Study 3(involving NT mice), 6 days of testing were evaluatedover two 3-day blocks. There was a 30-min time delaybetween the 4th trial (T4; final acquisition trial) and5th trial (T5; memory retention trial). The platformlocation was changed daily to a different arm, with dif-ferent start arms for each of the 5 trials semi-randomlyselected from the remaining 5 swim arms. During eachtrial (60 second maximum), the mouse was returned tothat trial’s start arm upon swimming into an incorrectarm and the number of seconds required to locate thesubmerged platform was recorded. If the mouse did notfind the platform within a 60-second trial, it was guidedto the platform for the 30-second stay. The numbers



of errors during trials 4 and 5 are both considered in-dices of working memory and are temporally similar tothe standard registration/recall testing of specific itemsused clinically in evaluating AD patients. Analysis ofthe number of seconds taken per arm choice (an indexof swim speed) over all days of post-treatment testingrevealed no differences in swim speed between groupsin both Studies 1 and 2.

Brain tissue preparation

Following completion of behavioral testing in Study1, 20–21 month-old mice were deeply anesthetized withsodium pentobarbital and transcardially perfused with100 ml of 0.9% saline. Postmortem brains were im-mediately removed and bisected sagitally. The hip-pocampus and cerebral cortex were dissected from theright hemisphere and processed for soluble Aβ1−40 andAβ1−42 determinations by ELISA. The left hemispherewas placed in 4% paraformaldehydein 0.1M phosphatebuffer (pH 7.4) overnight, wherein tissues remained un-til routine paraffin embedding for immunohistochemi-cal analyses.

Immunohistochemistry and image analysis

At the level of the hippocampus (bregma −2.92 mmto −3.64 mm), five 5 µm sections (150 µm apart) weremade from each mouse brain using a sliding microtome.Immunohistochemical staining was performed follow-ing the manufacturer’s protocol using a Vectastain ABCElite kit (Vector Laboratories, Burlingame, CA) cou-pled with the diaminobenzidine reaction, except thatthe biothinylated secondary antibody step was omittedfor Aβ immunohistochemical staining. The followingprimary antibodies were used for immunohistochemi-cal staining: a biothinylated human Aβ monoclonal an-tibody (clone 4G8; 1:200, Covance Research Products,Emeryville, CA) and rabbit synaptophysin polyclon-al antibody (undiluted, DAKO, Carpinteria, CA). ForAβ immunohistochemical staining, brain sections weretreated with 70% formic acid prior to the pre-blockingstep. Phosphate-buffered saline (0.1 mM, pH 7.4) ornormal rabbit serum (isotype control) was used insteadof primary antibody or ABC reagent as a negative con-trol.

Quantitative image analysis was done based on pre-vious methods with modifications [26,27]. Imageswere acquired using an Olympus BX60 microscopewith an attached digital camera system (DP-70, Olym-pus, Tokyo, Japan), and the digital image was rout-

ed into a Windows PC for quantitative analysis usingSimplePCI software (Compix Inc., Imaging Systems,Cranberry Township,PA). Images of five 5-µm sections(150 µm apart) through both anatomic regions of inter-est (hippocampus and entorhinal cortex) were capturedfrom each animal, and a threshold optical density wasobtained that discriminated staining from background.Each region of interest was manually edited to eliminateartifacts. For Aβ burden analysis, data are reported aspercentage of immunolabeled area captured (positivepixels) relative to the full area captured (total pixels).To evaluate synaptophysin immunoreactivity, after themode of all images was converted to gray scale, the av-erage intensity of positive signals from each image wasquantified in the CA1 and CA3 regions of hippocampusas a relative number from zero (white) to 255 (black).Each analysis was done by a single examiner blindedto sample identities.

Aβ ELISA

Hippocampal and cortical levels of human solubleAβ1−40 and Aβ1−42 were measured by ELISA. Briefly,30 mg brain samples were homogenized in 400 µl RI-PA buffer (100 mM Tris [pH 8.0], 150 mM NaCl, 0.5%DOC, 1% NP-40, 0.2% SDS, and 1 tablet proteinaseinhibitor per 100 ml (S8820, Sigma, St. Louis, MO),and sonicated for 20 seconds on ice. Samples werethen centrifuged for 30 min at 27,000 g at 4 ◦C, andsupernatants were transferred into new screw cap tube.The supernatants obtained from this protocol were thenstored at −80◦C for determination of soluble Aβ levelsusing ELISA kits (KHB3482 for 40, KHB3442 for 42,Invitrogen, Carlsbad, CA). Standard and samples weremixed with detection antibody and loaded on the anti-body pre-coated plate as the designated wells. HRP-conjugated antibody was added after wash, and sub-strates were added for colorimetric reaction, and thenstopped with sulfuric acid. Optical density was ob-tained and concentrations were calculated according astandard curve.

Analysis of pcRaf-1 and Protein Kinase A (PKA)

For hippocampal tissue from Study 2, equal amountsof protein were separated by sodium dodecyl sulphate–polyacrylamide gel electrophoresis (SDS-PAGE) andtransferred to nitrocellulose membranes (BA83 0.2 µm;Bio-Rad, Richmond, CA). The membranes wereblocked in buffered saline with 0.05% Tween 20 (PB-ST) containing 10% dry milk. Membranes were incu-



bated with primary antibodies in PBST with 3% drymilk (Sigma) overnight at 4◦C and with secondary an-tibodies for 2 h. Rabbit polyclonal antibodies (CellSignaling Technology, Inc., Danvers, MA) were direct-ed against pcRaf-1(Ser338; 1:250) or pcRaf-1(Ser259;1:500). A monoclonal mouse antibody anti α-tubulin(Promega, Madison, WI) was used as a control of pro-tein loading. The bands were detected using ECL de-tection kits (ECL, Pharmacia Biotech, Piscataway, NJ),visualized using the Kodak Image Station 440CF, andanalyzed using the molecular Imaging Software, ver-sion 4.0 (Rochester, NY) and NIH image J software.All data was normalized against tubulin immunoreac-tivity and expressed as percent of control values. PKAactivity from the same hippocampal tissue lysates weremeasured using a non-Radioactive PKA activity assaykit (Assay Designs, Ann Arbor, MI), which is based ona solid phase enzyme-linked immuno-absorbent assay(ELISA) using a synthetic peptide as a substrate forPKA and a polyclonal antibody against the phospho-rylated form of the substrate. All determinations ofPKA activity were performed at least in quintuplicate,repeated twice in separate experiment, and expressedas percentage of NT control values.

N2a cell cultures and caffeine treatment

Following dibutyryl cAMP-induced neuronal differ-entiation, cultured SweAPP N2a cells were treated withcaffeine in either a concentration- or time-dependentmanner. For the concentration-dependent treatment,cells were administered a serial dilution of caffeineranging from 0 µM-1 mM to establish the optimal rangeof caffeine action on GSK-3 activity. After Westernblot analysis established this range as 0–20 µM, cellswere treated with a serial dilution in this range. Cul-tures were incubated at 37◦C for 1 h prior to lysis withice-cold lysis buffer. For the time-dependent applica-tion, cells were given a 20 µM concentration of caffeineand incubated at time points ranging from 0–180 minprior to lysis.

For Western blotting, aliquots containing up to 50 µgof total protein were electrophoretically separated.Electrophoresed proteins were then transferred to nitro-cellulose membranes and blocked for 1 h at room tem-perature in Tris-buffered saline. After blocking, mem-branes were hybridized overnight at 4◦C with prima-ry antibodies (Sigma-Aldrich, St. Louis, MO) againsttotal GSK-3α/β or active GSK-3α/β [(pTyr279/216

residues); diluted 1:500 or 1:1000,respectively]. Mem-branes were then washed 3 × for 5 min each and in-

cubated for 1 h at room temperature with the appropri-ate HRP-conjugated secondary antibody. Blots weredeveloped using a luminol reagent, and densitometricanalysis of the blots ensued.

Statistical analysis

Behavioral performance was statistically evaluatedto determine any group difference based on transgenic-ity or caffeine treatment. For the single day tasks,one-way ANOVAs were used. For multi-day tasks,both one-way ANOVAs and two-way repeated mea-sure ANOVAs were performed. Prior to analysis, Mor-ris water maze (MWM) data was broken down into 2-day blocks and the RAWM data was divided into 3-day blocks, to aid in data presentation and analysis.After ANOVA analysis, post hoc pair-by-pair differ-ences between groups (planned comparisons) were re-solved using the Fisher LSD test. Pre-treatment ver-sus during-treatment RAWM performance was eval-uated for each group separately, using paired t-tests.All group comparisons were considered significant atp < 0.05. While very few in number, any outliersor non-performers (e.g., repeated circulars, consistentfloaters) in any given task were eliminated from statis-tical analysis of that task. Data analyses involving Aβburden, Aβ levels, pcRaf-1 levels, and synaptophysinimmunoreactivity were performed using ANOVA. Allother statistical analyses are as indicated in the text.In order to test if relationships were present betweenneurohistologic, neurochemical, and behavioral mea-sures, correlation analysis was performed using the Sy-stat analytical software package (Systat Software, Inc.,Chicago, IL).

To determine if the three groups of Study 1 (NT, Tg,Tg/Caff) or the two groups of Study 2 (NT, NT/Caff)could be distinguishable from one another based on“overall” behavioral performance, discriminant func-tion analysis (DFA) was performed using Systat soft-ware. For Study 1, direct entry DFAs were performedusing seven behavioral measures collected from thethree cognitive-based tasks administered during treat-ment. For Study 2, direct entry DFAs were also per-formed, using 19 behavioral measures collected fromthe entire behavioral battery. A complete descriptionof DFA multi-metric statistical analyses relevant to ADtransgenic mouse studies can be found in Leighty etal. [25,28].



Fig. 1. Caffeine treatment reverses memory impairment in aged Tg mice. (A) Prior to caffeine administration, aged Tg mice were impaired in theradial arm water maze task of working memory vs. non-transgenic (NT) controls. ∗p < 0.000005. (B) During the final block of RAWM testingat 4–5 weeks into caffeine treatment, aged Tg mice given caffeine (Tg/Caff) performed to the level of NT mice on working memory trials T4and T5. ∗∗p < 0.025 for Tg control vs. other two groups. (C) Comparing pre-treatment vs. during-treatment trial T5 performance, Tg controlsfailed to improve. However, Tg/Caff mice exhibited a dramatic improvement in working memory (∗∗p < 0.005; paired t-test), with NT controlsalso improving (∗p < 0.05; paired t-test). (D) During the final two days of platform recognition testing for strategy switching, Tg controls wereimpaired, while Tg/Caff mice performed comparable to NT controls (∗p < 0.02 for Tg vs. NT). Group symbols defined in (D). (Colours arevisible in the electronic version of the article at www.iospress.nl.

RESULTS

Caffeine treatment reverses working memoryimpairment and normalizes identification/recognitionability in aged Tg mice

We have previously shown that long-term caffeinetreatment in young adult APPsw Tg mice protectedthem against otherwise inevitable cognitive impairmentin older age [13]. To determine if the same caffeinetreatment could benefit aged Tg mice already demon-

strating cognitive impairment, caffeine treatment wasgiven in drinking water to 19 month-old APPsw mice.Prior to caffeine administration, mice were evaluatedin the RAWM of working memory and verified to becognitive-impaired, as evidenced by many more errorsthan NT littermate controls during delayed retentionTrial 5 (T5) for the last block of pre-treatment testing(Fig. 1A). At 4–5 weeks into caffeine treatment, how-ever, those impaired Tg mice given caffeine (Tg/Caff)exhibited vastly superior working memory on Trials 4and 5 compared to the continuing impairment of control



Tg mice (Fig. 1B). Indeed, the performance of Tg/Caffmice was identical to that of NT controls during thelast block of testing. We further characterized the cog-nitive benefits of long-term caffeine treatment by com-paring pre-treatment RAWM performance to “duringtreatment” performance (Fig. 1C). In sharp contrast tothe inability of Tg controls to improve upon their pooroverall T5 working memory, Tg/Caff mice showed aprecipitous drop in errors during RAWM testing at 4–5weeks into treatment. Even NT controls significantlyimproved upon their pre-treatment performance level,although not to an error level lower than Tg/Caff mice.

We have consistently found that AD Tg mice are im-paired in their ability to switch from the spatial strategyof RAWM to the identification/recognition strategy re-quired for platform recognition testing [13,29,30]. Todetermine if caffeine treatment can improve the “strat-egy switching” ability of Tg mice, we evaluated micein the platform recognition task immediately followingcompletion of RAWM testing. As expected, Tg con-trols were impaired overall (p < 0.01), and particularlyduring the last two days of testing (Fig. 1D). However,Tg/Caff mice were not different from NT mice on anyof the four test days or overall, signifying that caffeineimproved strategy switching ability in aged Tg mice(Fig. 1D).

There were no group differences in tasks of sen-sorimotor skill (open field activity, balance beam,string agility, Y-maze choices) and anxiety (elevated-plus maze), indicating that neither genotype or caf-feine treatment impacted these non-cognitive measures(data not shown). Additionally, there were no groupdifferences in swim speed, as indexed by time takenper choice during RAWM testing.

Aged Tg mice show “overall” cognitive benefit withcaffeine treatment

Although evaluation of the effect of caffeine on in-dividual cognitive measures is informative and impor-tant, we have found that a collective evaluation acrossmultiple cognitive measures/tasks is a more insight-ful approach for determining if a given AD therapeu-tic has broad-based value in multiple cognitive do-mains [22–24]. In this context, we have previous-ly utilized discriminant function analysis (DFA) in re-porting that caffeine treatment from young adulthoodthrough older age “protects” APPsw mice from over-all cognitive impairment across multiple cognitive do-mains [13,28]. To elucidate whether caffeine treat-ment in aged Tg mice provides “overall” cognitive ben-

Fig. 2. A Canonical scores plot of discriminant function analysisinvolving 7 cognitive measures collected from aged Tg mice at 4–5weeks into caffeine treatment. Each symbol represents the cogni-tive performance of one animal graphed from the two linear func-tions derived in the discriminant function analysis (DFA). Group dif-ferences in overall cognitive performance across multiple cognitivemeasures/domains are evident, with Tg/Caff mice having superioroverall performance compared to Tg controls. Symbols: O = NTmice; X = Tg mice; + = Tg/Caff mice. (Colours are visible in theelectronic version of the article at www.iospress.nl.

efit spanning multiple measures/domains (e.g., work-ing/spatial memory, identification/recognition, gener-al mnemonic function), we collectively evaluated allseven of the cognitive measures attained “during treat-ment” in the present study, utilizing direct entry DFA.This approach was very effective in revealing that therewere clear differences in overall cognitive performancebetween groups (Wilk’s λ = 0.126; p < 0.02) andthat the Tg/Caff group was significantly better in over-all cognitive performance compared to the Tg controlgroup (p < 0.05). Figure 2 shows the canonical scoresplot of this DFA for individual mice included in thethree groups of Study 1.

Caffeine treatment markedly reduces brain Aβdeposition in aged Tg mice

Our prior work has demonstrated that oral caffeineadministration in young adult Tg mice for 51/2 monthssuppresses levels of both BACE1 and PS1 in hip-pocampus, resulting in lower hippocampal levels ofboth soluble and insoluble Aβ [13]. This suppressionof both major enzymes involved in brain Aβ produc-tion would suggest that caffeine administration, evento aged Tg mice exhibiting robust Aβ pathology, couldreduce/reverse that pathology. Therefore, we nextperformed a neurohistologic analysis of total Aβ im-



Fig. 3. Caffeine treatment reduces the robust Aβ deposition present in aged Tg mice. Total Aβ deposition was strikingly reduced in boththe entorhinal cortex and hippocampus in aged Tg mice following two months of oral caffeine treatment. (Upper) Photomicrographs of Aβdeposition staining in the entorhinal cortex and hippocampus exemplifying the reduction in Aβ deposition induced by caffeine treatment. Scalebar = 50 µm (Lower) Caffeine induced reductions of 46% and 40% in Aβ deposition within the entorhinal cortex and hippocampus, respectively.∗p < 0.001. (Colours are visible in the electronic version of the article at www.iospress.nl.

munostaining in the same aged Tg mice wherein caf-feine administration reversed memory impairment. Forboth entorhinal cortex and hippocampus, total Aβ de-position was strikingly reduced in aged Tg mice fol-lowing two months of oral caffeine treatment comparedto Tg controls (Fig. 3). Reductions of 46% and 40%in Aβ deposition were evident within entorhinal cor-tex and hippocampus, respectively, at 20–21 months ofage. These caffeine-induced reductions in Aβ deposi-tion are clearly evident in representative photomicro-graphs of Aβ immunostaining from entorhinal cortexand hippocampus (Fig. 3).

Caffeine treatment reduces brain levels of “soluble”Aβ in aged Tg mice, which correlate with extent of Aβdeposition

We have previously reported that long-term oral ad-ministration of caffeine to young adulthood Tg miceinduces a significant 37% reduction in hippocampalAβ1−40 levels, as measured by ELISA [13]. In thepresent study, we wished to determine if brain solubleAβ levels could also be reduced by caffeine treatment inaged Tg mice bearing high brain levels of both Aβ1−40

and Aβ1−42. In the same aged Tg mice wherein rever-



Fig. 4. Long-term caffeine treatment to aged Tg mice reduces soluble Aβ levels in both cortex and hippocampus; Hippocampal soluble Aβ levelscorrelate with hippocampal Aβ deposition. (A) At two months into oral caffeine treatment, hippocampal and cortical levels of both Aβ1−40

and Aβ1−42 were reduced in comparison to values from Tg controls. ∗p < 0.05 versus Tg control mice. Abbreviations: N.D., non-detectable.(B) In hippocampus (but not in cortex), significant correlations were evident between soluble Aβ and deposited Aβ, indicative of a dynamicequilibrium between these two pools of Aβ in the hippocampus. Each group consisted of 5–8 mice. (Colours are visible in the electronic versionof the article at www.iospress.nl.)

sal of cognitive impairment was exhibited, measure-ment of brain soluble Aβ levels at 20–21 months ofage (approximately 2 months into treatment) revealedsignificant decreases in both cortical and hippocampalAβ levels for caffeine-treated mice (Fig. 4A). Aβ1−42

levels were significantly reduced in both cortex (↓51%)and hippocampus (↓59%) of caffeine-treated Tg micecompared to Tg controls. Similarly, Aβ1−40 levels inTg mice were significantly reduced in cortex (↓25%)and nearly reduced (p = 0.09) in hippocampus (↓37%)

by caffeine treatment.To determine if an association exists between solu-

ble and deposited Aβ pools in the brains of these agedTg mice, correlation analysis was performed betweenextent of hippocampal Aβ deposition and Aβ1−40 orAβ1−42 levels, irrespective of treatment. The resultsof these correlation analyses indicate a clear relation-ship between extent of hippocampal Aβ deposition andhippocampal levels of Aβ1−40 (Fig. 4B) and Aβ1−42

(Fig. 4C). These correlations underscore a dynamic re-



Fig. 5. Long-term caffeine treatment to aged APPsw mice did not affect synaptophysin immunostaining in the hippocampal CA1 and CA3regions. Both Tg and Tg/Caff mice exhibited aberrantly high synaptophysin staining, typically seen in aged APPsw mice due to intense stainingof dystrophic nerve terminals. Scale bar = 20 µm. ∗p < 0.001 versus NT group. (Colours are visible in the electronic version of the article atwww.iospress.nl.

lationship between deposited and soluble forms of Aβin the brain, which is involved in the modulation ofbrain Aβ by caffeine (see diagrams in Fig. 10).

Synaptophysin immunostaining in hippocampus ofaged Tg mice is not affected by caffeine treatment

Synaptophysin immunostaining/immunoreactivity isa marker for total synaptic area – not number of synap-tic terminals – in a given brain region. As such, wehave observed increases in hippocampal synaptophysinimmunoreactivity following several therapeutics thatprovide cognitive benefit to AD Tg mice such as envi-ronmental enrichment [11,24] and T-cell immunother-apy [30]. Therefore, we subsequently wished to de-termine if the beneficial cognitive effects of caffeine

in aged Tg mice was associated with changes in hip-pocampal synaptophysin immunostaining. As shownin Fig. 5, caffeine treatment did not impact synapto-physin immunostaining in the hippocampal CA1 andCA3 regions of Tg mice. Irrespective of treatment, Tgmice exhibited enhanced synaptophysin immunostain-ing in comparison to NT controls (Fig. 5), which isconsistent with our prior work [31] and probably in-dicative of the intense synaptophysin immunostainingassociated with dystrophic nerve terminals in Tg mice.

Caffeine treatment down-regulates the active form ofcRaf-1 in hippocampus from Tg mice: Possiblelinkage to BACE1 suppression

Because we have previously reported that long-term caffeine treatment decreases BACE1 expression



in brains of AD Tg mice [13] and also found caffeine-induced reductions in brain Aβ levels in the presentstudy, we wished to develop insight into possible mech-anism(s) involved in BACE1 suppression by caffeine.The cRaf-1 is a cytosolic protein kinase and is an inte-gral part of the Ras/cRaf/ERK pathway that regulatescell proliferation, survival, and senescence. When per-sistently activated through phosphorylation at its Ser-ine 338 site, cRaf-1 stimulates the transcription factornuclear factor kappa B (NFκB) [32], which then canstimulate expression of BACE1 and other AD-linkedproteins [33]. Phosphorylation of cRaf-1 at its Serine259 site stabilizes cRaf-1 in an inactive conformation,thus suppressing BACE1 expression. These relation-ships are diagramed in Fig. 10.

In a second caffeine treatment study, we thereforeexamined the hippocampus of 9.5 month-old Tg micethat had been given oral caffeine treatment via gavagefor two weeks prior to euthanization. Analysis of hip-pocampal levels of the inactive form of cRaf-1 (pcRaf-1[Ser259]) by Western blot revealed a significant down-regulation of this inactive form in Tg controls to 42%of NT control levels (Fig. 6A). In sharp contrast, Tgmice that had been treated with caffeine for only twoweeks exhibited active cRaf-1 (pcRaf-1[Ser338]) levelsthat were significantly lower than Tg controls (↓27%;Fig. 6B) and inactive cRaf-1 (pcRaf1[Ser259]) thatwere significantly higher than Tg controls (↑141%;Fig. 6A). Moreover, we also determined from the samehippocampal tissue lysates that a significant decrease(↓29%) in protein kinase A (PKA) activity was presentin Tg controls and that caffeine treatment significant-ly increased PKA activity (↑25%) in Tg mice to lev-els exhibited by NT mice (Fig. 6C). Since PKA con-verts active to inactive cRaf-1, these results are consis-tent with a PKA/Raf-1 mediated mechanism of BACE1suppression by caffeine (see Fig. 10).

Caffeine induces concentration- and time-dependentdecreases in both active and total GSK-3 levels inSweAPP N2a cells

We have reported that chronic caffeine treatment re-duces PS1 expression/γ-secretase activity in AD Tgmice, presumably contributing to the decreased Aβproduction also observed [13]. In the present study,we next examined the potential involvement of GSK-3 in caffeine’s reduction of γ-secretase-mediated Aβproduction, as GSK-3 is reported to regulate Aβ pro-duction by interfering with γ-secretase cleavage [20,34]. To identify the possible actions of caffeine on

this essential component of the AD pathogenic path-way, Western blot analysis was performed on SweAPPN2a cell lysates using antibodies against both activeand total GSK-3. A concentration-dependent decreasein both GSK-3α and GSK-3β was evident, with sig-nificant reductions achieved at caffeine concentrationsof 10 and 20 µM (Fig. 7). This effect was observedon active GSK-3 (pTyr216/279) protein levels (A), aswell as on GSK-3 holo protein levels (C). Caffeine al-so exerted a time-dependent decrease in both GSK-3isoforms, as shown in Fig. 7. Significant reductionsin active protein levels were seen in both GSK3 iso-forms at time-points exceeding 120 minutes (B). How-ever, significant alterations in holo protein levels wereisoform-specific, with GSK-3β reduction by 30 min-utes and GSK-3α reduction by 90 minutes (D). It isnoteworthy that, while active GSK-3 levels declined ascaffeine concentration was raised from 0 to 16 µM, ac-tive GSK-3 levels increased as caffeine concentrationwas elevated from 31.2 µM to 1 mM (data not shown).Thus, approximately 20 µM was determined to be theoptimal caffeine concentration for active GSK-3α andGSK-β inhibition, which corresponds to that typicallyobserved in the blood and brain of humans following1–2 cups of coffee [1].

Long-term caffeine treatment from adulthood throughold age does not provide cognitive benefit to normalmice

It is possible that the caffeine-induced reversal ofcognitive impairment that we observed in aged Tg miceinvolved non-specific, generalized mechanisms such asincreased glucose utilization [35], resulting in the well-known ability of caffeine to increase alertness/attention.If such generalized, non-amyloidogenic mechanismswere critical for the cognitive benefits seen in Tg mice,we reasoned that NT mice should also benefit fromlong-term caffeine treatment. Therefore, in a third caf-feine treatment study, we began providing 51/2 month-old NT (normal) mice with caffeine in their drinkingwater and behaviorally evaluated them at 10 monthsinto caffeine treatment (e.g., 15–16 months of age).Parenthetically, no truly long-term caffeine administra-tion studies (over months) have been done previouslyin rodents that might relate to habitual caffeine use inhumans.

As shown in Fig. 8, life-long caffeine administrationhad no effect on cognitive performance of NT miceacross 13 measures in five cognitively-based tasks (Y-maze percent alternation, standard Morris water maze,



Fig. 6. In the hippocampus of Tg mice, caffeine treatment increases the “inactive” form (A) and decreases the active form (B) of cRaf-1, whilealso increasing protein kinese A (PKA) activity back to normal levels. Two weeks of oral caffeine treatment was given to 9 month-old Tg miceand their hippocampus analyzed. (A) The significant decrease in inactive cRaf-1 (pcRaf-1[Ser259]) that was evident in Tg controls was reversedby caffeine treatment to the level of NT controls. ∗p < 0.05 versus other two groups (B) Caffeine treatment decreased the levels of active cRaf-1(pcRaf-1[Ser338] in Tg mice treated with caffeine compared to Tg control mice. Tubulin or GADPH served as non-specific protein controls.∗p < 0.05 versus Tg group. (C) The significant decrease in hippocampal PKA activity present in Tg mice was reversed by caffeine treatmentback to levels present in NT mice. ∗p < 0.02 for Tg controls versus other two groups. (Colours are visible in the electronic version of the articleat www.iospress.nl.

circular platform, platform recognition, and radial armwater maze). It should be noted that cognitive perfor-mance of the control NT mice at 15–16 months of agewas not outstanding, thus allowing for possible thera-

peutic improvement. Chronic caffeine administrationdid not affect sensorimotor (open field activity, balancebeam, string agility) or anxiety (elevated-plus maze)measures, nor did it affect swim speed in water-based



Fig. 7. Caffeine induces concentration- and time-dependent decreases in both GSK-3α and GSK-β expression in vitro. SweAPP N2a cellswere treated with a range of caffeine concentrations for 1 h (A,C) or with a 20 µM concentration of caffeine over a range of time points(B,D). Prepared lysates were subjected to Western blot (WB) analysis using antibodies against active GSK-3α (pTyr216 residue) and GSK-3β(pTyr279 residue) isoforms (A,B) or against total GSK-3α and GSK-β (C and D). Anti-actin antibodies were used as an internal referencecontrol for each WB. Histograms depict densitometric analysis of three independent experiments, with a representative WB of each experimentdisplayed. Isoform-specific band to actin ratios are delineated on histogram y-axes, and concentration- or time-points are depicted on x-axes.Asterisks denote significant differences (p < 0.05) between specific concentration- or time-group and the control group (0 µM or 0 minutes),using a Mann-Whitney test for pair wise comparisons of non-parametric data. (Colours are visible in the electronic version of the article atwww.iospress.nl.

tasks (data not shown). A direct entry DFA performedacross all 19 behavioral measures of the test batteryalso indicated no significance discrimination betweencaffeine-treated and control NT mice in overall behav-ioral performance.

DISCUSSION

This study provides the first evidence that caffeinetreatment can reverse cognitive impairment and ADneuropathology in a model for the disease. Aged,



Fig. 8. Long-term caffeine treatment does not provide cognitive benefits to normal (non-transgenic) mice in multiple tasks. A total of 13 measureswere evaluated in 5 cognitive-based tasks: (A) Y-maze spontaneous alternation, (B and C) Morris water maze acquisition and memory retention,(D) Circular platform, (E) platform recognition, and (F) radial arm water maze. (Colours are visible in the electronic version of the article atwww.iospress.nl.

cognitively-impaired APPsw mice given a moderateamount of daily caffeine (the human equivalent of 5cups of coffee) exhibited a restoration of working mem-ory to the level of normal, aged mice. In these sameaged AD mice, which had pre-existing and substantialAβ burdens, caffeine treatment reduced both solubleand deposited (insoluble) brain Aβ levels. Caffeine-induced cognitive restoration is likely due to its pro-found ability to reduce Aβ production by reducing ex-pression of both BACE1 and PS1/γ-secretase [13]. Thepresent study provides mechanistic insight into both ofthese caffeinergic actions by showing that 1) caffeineadministration to APPsw mice was able to correct theirdysregulation of the Raf-1/NFκB inflammatory path-way, which is linked to BACE expression, and 2) caf-feine suppressed both α and β isoforms of GSK-3 invitro, which are linked to PS1/γ-secretase activity andtau hyperphosphorylation, respectively. The extent to

which these caffeinergic effects involve adenosine re-ceptor antagonism are unknown and currently beinginvestigated.

We recently reported that caffeine, when adminis-tered from young adulthood through older age, pro-tected APPsw mice from otherwise certain impairmentin multiple cognitive domains and reduced their brainAβ levels [13]. These results, suggestive that moderatedaily intake of caffeine could delay or reduce the riskof AD, are supportive of an earlier epidemiologic studyshowing that AD patients consumed much less caffeinein the 20 years prior to disease onset compared to same-aged individuals who were non-demented [9]. Morerecent longitudinal studies have also reported protec-tive effects of caffeine/coffee intake against cognitiveimpairment during normal aging [6,7] and AD [8]. Theevidence for the protective effects of caffeine againstaged-associated cognitive impairment and AD, which



Fig. 9. Diagrams depicting the inhibitory actions of caffeine on Aβ production and resultant lower brain Aβ levels/deposition. Unmodulated:Newly produced Aβ enters a dynamic equilibrium between soluble and deposited (insoluble) Aβ in the brain, with continual transport of solubleAβ into plasma. Caffeine: Long-Term: Caffeine suppression of both β- and γ-secretase activities reduces Aβ production, resulting in lowersoluble Aβ in the brain. These lower levels of brain soluble Aβ induce a flux of insoluble Aβ out of the deposited form and into the solubleform, which is then cleared into plasma via soluble Aβ transport. (Colours are visible in the electronic version of the article at www.iospress.nl.

is nicely reviewed by Rosso et al. [10], is in contrastto the reality that human intake of caffeine usually de-creases during aging – this, despite the fact that themetabolism and physiologic responses to caffeine arethe same in elderly and younger individuals [36].

Accompanying the caffeine-induced cognitive res-toration in aged APPsw mice was a reversal of Aβ de-position and decrease in soluble Aβ levels within theirbrains. These reductions in both soluble and deposited(insoluble) forms of Aβ would be expected, given thatcaffeine decreases Aβ production through suppressionof both BACE and PS1/γ-secretase expression [13].Newly produced Aβ enters a dynamic equilibrium be-tween soluble and deposited Aβ in the brain, with con-tinual transport of soluble Aβ out of the brain and intoplasma down a concentration gradient (Fig. 9, Unmod-ulated). We hypothesize that caffeine suppression ofAβ production results in lower brain levels of solubleAβ, which then induces a flux of insoluble Aβ out ofthe deposited form and into the soluble form (Fig. 9,Caffeine). This newly-solublized Aβ would then becleared into the plasma via Aβ transport. Supportiveof this dynamic equilibrium hypothesis are the strongcorrelations that we report in aged APPsw mice be-tween soluble Aβ (both 1–40 and 1–42) and depositedAβ in hippocampus. Moreover, the additional correla-

tions that we report between both soluble and depositedforms of Aβ versus cognitive performance clearly un-derscore a direct impact of this dynamic equilibrium oncognitive performance. Indeed, caffeine’s reversal ofcognitive dysfunction in aged APPsw mice most like-ly involves lowering of both soluble and deposited Aβthrough this equilibrium.

The present study presents new mechanistic in-sight into how caffeine may suppress both BACE1and PS1/γ-secretase expression to decrease Aβ pro-duction. Whether or not this caffeine-induced de-crease in Aβ production is direct or involves adeno-sine receptor blockade/mediation is currently unknown.There certainly exists considerable evidence that theactions of caffeine against Aβ are mediated throughantagonism of A2a and/or A1 adenosine receptors,particularly in the context that 1) caffeine and se-lective A2a receptor antagonists both provide protec-tion against Aβ neurotoxicity in vitro [37]; 2) caf-feine and/or selective A2a receptor antagonists protectagainst the cognitive impairment induced by intrac-erebral Aβ1−42 or Aβ25−35 infusion in mice [38,39];and 3) acetylcholine neurotransmission in hippocam-pus/cortex is increased through pre-synaptic A1 recep-tor blockade [40]. Although we have determined thatbrain adenosine receptor levels are unaffected by long-



Fig. 10. Diagram depicting 1) stimulatory actions of “active” cRaf-1 on NFκB activation, resulting in increased BACE1 expression and Aβproduction, and 2) caffeine’s activation of PKA, resulting in enhanced levels of ‘inactive” cRaf-1 and less Aβ production. (Colours are visible inthe electronic version of the article at www.iospress.nl.

term caffeine treatment in AD transgenic mice [13], arich body of scientific findings suggests that the pos-sibility of adenosine-receptor-mediated actions needsfurther investigation [41].

Regarding BACE1 suppression,we provide evidencethat caffeine is able to correct a dysregulation of thecRaf-1 inflammatory pathway in APPsw mice by stim-ulating PKA activity, thus increasing brain levels of“inactive” cRaf-1 (pcRaf-1[Ser259]) and reducing theactive form of cRaf-1 phosphorylated at Serine 338(pcRaf-1[Ser338]). Serine(s) 259 and S621 bind to 14-3-3, a protein that maintains cRaf-1 in an inhibited state(phosphorylated at Serine 259) in basal conditions. Af-ter stimulation, active Ras displaces 14-3-3 proteins,cRaf-1 is trans-located to the plasma membrane, de-phosphorylated by phosphatases at this site and phos-phorylated at the kinase domain at the Serine 338 ac-tivation site [42]. As shown in Fig. 10, “active” Rafstimulates NFκB intracellularly, resulting in increasedexpression of reporter genes such as BACE1 [43,44].It should be noted that cRaf-1 has been reported to beover-activated in AD brains [15,45].

In this study, we demonstrate that caffeine stimulatesPKA activity in the hippocampus of treated AD Tgmice, which is consistent with similar effects of caffeinein peripheral tissues [46,47]. This enhanced PKA activ-ity then inhibits (inactivate) cRaf-1 by phosphorylationat serine259 [48], decreasing NFκB activity and the

expression of NFκB-controlled genes such as BACE1;thus, decreased Aβ production results (Fig. 10). Sup-portive of the idea that the cRaf-1/NFκB pathway playsa key role in Aβ production and toxicity in the brain isour recent discovery that the Raf-1 inhibitor Sorafenibinhibits NFκB signaling and ameliorates memory im-pairment in aged APPsw mice [49]. Caffeine treatmenthas often been associated with the inhibition of phos-phodiesterases (PDEs); indeed, we and others have re-ported that treatment with the PDE inhibitor Rolipramimproves working memory of AD transgenic mice [50,51]. However, caffeine’s inhibition of PDE occurs atmM concentrations [1], well above blood levels of caf-feine that are achievable in rodents/humans. We there-fore believe that the profound ability of caffeine to re-duce brain Aβ pathology is due to its suppression ofboth β- and γ-secretases [13]. Consistent with thispremise, Rolipram was unable to decrease Aβ levels invivo [51].

A potential mechanism through which caffeine sup-presses PS1/γ-secretase (to reduce Aβ production)was also explored in the present study. Dysregula-tion of GSK-3 is a key feature of AD pathogene-sis, with the activity of both α and β isoforms play-ing critical roles [52]. While GSK-3β directly bindsPS1 [53], GSK-3α is most notable for enhancingPS1/γ-secretase activity and consequently increasingAβ production [20]. In vitro studies demonstrate



that siRNA inhibition of GSK-3α drastically reducesAβ production by interfering with γ-secretase cleav-age of APP [20], likely through disruption of APP-PS1 association [34]. Furthermore, conditional over-expression of GSK-3β results in increased tau phospho-rylation, neuronal apoptosis, reactive gliosis, and spa-tial learning deficits in the Tet/GSK-3β mouse modelfor AD [19,54]. This AD-like phenotype is completelyreversed once GSK-3β overexpression is ceased andGSK-3β activity returns to normal [55]. In the presentstudy, we report that caffeine suppresses GSK-3α andGSK-3β levels in SweAPP N2a cell cultures in both adose- and time-dependent fashion. Moreover, caffeineimpacted both active and holoprotein (total) GSK-3 lev-els, suggesting that caffeine does not diminish GSK-3activity through direct substrate-level phosphorylation,but rather through alternations in protein production ordegradation.

Because optimal GSK-3 inhibition was obtainedwith a 20 µM concentration of caffeine in vitro, actionsof caffeine that only occur at much higher concentra-tions are unlikely candidates to mechanistically explainboth our in vitro and in vivo observations [1]. Indeed, a10–20 µM plasma and brain concentration of caffeinewould be quickly reached (1–2 hours) in humans fol-lowing ingestion of 1–2 cups of coffee [1]. This under-scores that the effects of caffeine on GSK-3 are readi-ly attained at physiologically relevant concentrations –in fact, the same 10–20 µM concentration of caffeineeffectively inhibits both Aβ1−40 and Aβ1−42 produc-tion by SweAPP N2a cell cultures [13]. Reductionsin GSK-3 thus represent a further mechanism throughwhich caffeine may ameliorate AD pathology and re-stores cognitive function. Accordingly, reduction ofGSK-3 levels/activity in the brain is in itself a pursuedAD therapy [52].

In addition to the aforementioned mechanisms ofcaffeine action directly impacting Aβ production, thereare multiple “non-amyloidogenic” mechanisms thatcould contribute to the ability of caffeine to protectagainst or reverse memory impairment. Since inflam-mation is an essential part of the AD pathogenic path-way [56] and AD is characterized by increased oxida-tive stress, it is possible that the anti-inflammatory [46]and antioxidant [57] actions of caffeine contribute toits pro-cognitive affects. It is important to emphasizethat generalized mechanisms of caffeine action (e.g.,increased glucose utilization) may also be contributo-ry [35] to the cognitive reversal observed. Howev-er, our finding that life-long caffeine treatment to non-transgenic mice did not provide any cognitive bene-

fits in old age does not support this possibility – par-ticularly given that there was ample room for cogni-tive improvement when testing occurred in older age.Nonetheless, if the same caffeine treatment paradigmhad been given to aged “normal” mice as was givento aged Tg mice, beneficial cognitive effects may havebeen revealed. Indeed, generalized mechanisms (suchas increased glucose utilization) may play a role be-cause the substantial brain Aβ levels in aged APPswmice reduce their cerebral blood flow [58].

Although caffeine has been reported to have effectson dendritic morphology during post-natal develop-ment [59], we presently found no effect of long-termcaffeine treatment on synaptic area in hippocampusof aged APPsw mice. Regarding caffeine effects onsynaptic function, several electrophysiologic studiesinvolving hippocampal slice recordings from the hip-pocampal CA1 region have reported that caffeine in-creases the amplitude of EPSPs [60,61] and inducesa novel pre-synaptic form of long-term potentiation(LTP) [62]. However, because all such studies have in-volved caffeine concentrations well above physiologiclevels (e.g., 0.1–10 mM), electrophysiologic studies in-volving caffeine concentrations relevant to tissue levelsof caffeine achieved with human or rodent caffeine con-sumption (e.g., 10–30 µM) are necessary to determinethe impact of caffeine impact on synaptic transmission.Finally, caffeine has very recently been shown to blockdisruptions in the blood-brain barrier induced by a highcholesterol diet in rabbit [63]. Since breakdown ofthe blood-brain barrier has been reported in AD [64]and appears to precede Aβ plaque formation in APPswtransgenic mice [65], blood-brain barrier stabilizationby caffeine is a viable mechanism for consideration.

All of the aforementioned non-amyloidogenic andgeneralized mechanisms of caffeine action notwith-standing, the cognitive restoration provided to aged ADtransgenic mice by caffeine treatment in the presentstudy would appear to be mostly due to the profoundability of caffeine to reduce Aβ production via secre-tase suppression, and in the process result in brain clear-ance of both soluble and deposited/insoluble Aβ. Giv-en the safety, availability, and brain bioavailability ofcaffeine, its ability to suppress both β- and γ-secretaseswould suggest it is preferable to synthetic secretase in-hibitors currently under development. Based on the ro-bust protective and treatment effects of caffeine that wehave observed in AD transgenic mice, we have initiatedclinical trials with caffeine.



ACKNOWLEDGMENTS

This research was supported by grants to (G.W.A.)and (H.P.) within the NIA-designated Florida Alzhei-mer’s Disease Research Center (P50AG025711), aswell as by funds from the Byrd Alzheimer’s Center andResearch Institute. H.P. holds the Eric Pfeifer chair forAlzheimer’s research at the Suncoast Gerontology Cen-ter, University of South Florida College of Medicine.

REFERENCES

[1] Fredholm B, Battig K, Holmen J, Nehlig A, Zvartau E (1999)Actions of caffeine in the brain with special reference to factorsthat contribute to its widespread use. Pharmacol Rev 51, 83-133.

[2] Ruhl C, Everhart J (2005) Coffee and caffeine consumptionreduce the risk of elevated serum alanine aminotransferaseactivity in the United States. Gastroenterology 128, 24-32.

[3] Higdon J, Frei B (2006) Coffee and health: a review of recenthuman research. Crit Rev Food Sci Nutr 46, 101-123.

[4] Ross G, Abbott R, Petrovitch H, Morens D, Grandinetti A,Tung K, Tanner C, Masaki K, Blanchette P, Curb J, Popper J,White L (2000) Association of coffee and caffeine intake withthe risk of Parkinson disease. JAMA 283, 2674-2679.

[5] Hernan M, Takkouche B, Caamano-Isorna F, Gestal-Otero J(2002) A meta-analysis of coffee drinking, cigarette smoking,and the risk of Parkinson’s disease. Ann Neurol 52, 276-284.

[6] Ritchie K, Carriere I, de Mendonca A, Portet F, Dartigues J,Rouaud O, Barberger-Gateau P, Ancelin M (2007) The neuro-protective effects of caffeine: a prospective population study(the Three City Study). Neurology 69, 536-545.

[7] van Gelder B, Buijsse B, Tijhuis M, Kalmijn S, GiampaoliS, Nissinen A, Kromhout D (2007) Coffee consumption isinversely associated with cognitive decline in elderly Europeanmen: the FINE Study. Eur J Clin Nutr 61, 226-232.

[8] Eckelinen M, Ngandu T, Tuomilehto J, Soininen H, Kivipel-to M (2008) Midlife coffee and tea drinking and the riskof late-life dementia: A population-based CAIDE study. JAlzheimer’s Dis 16, 85-91

[9] Maia L, de Mendonca A (2002) Does caffeine intake protectfrom Alzheimer’s disease? Eur J Neurol 9, 377-382.

[10] Rosso A, Lippa C, Mossey J (2008) Caffeine: Neuroprotec-tive Functions in Cognition and Alzheimer’s Disease. Am JAlzheimers Dis Other Demen 23, 417-422.

[11] Arendash G, Garcia M, Costa D, Cracchiolo J, Wefes I, PotterH (2004) Environmental enrichment improves cognition inaged Alzheimer’s transgenic mice despite stable beta-amyloiddeposition. Neuroreport 15, 1751-1754.

[12] Arendash G, Lewis J, Leighty R, McGowan E, Cracchiolo J,Hutton M, Garcia M (2004) Multi-metric behavioral compar-ison of APPsw and P301L models for Alzheimer’s disease:linkage of poorer cognitive performance to tau pathology inforebrain. Brain Res 1012, 29-41.

[13] Arendash G, Schleif W, Rezai-Zadeh K, Jackson E, ZachariaL, Cracchiolo J, Shippy D, Tan J (2006) Caffeine protectsAlzheimer’s mice against cognitive impairment and reducesbrain beta-amyloid production. Neuroscience 142, 941-952.

[14] Keller E, Fu Z, Brennan M (2004) The role of Raf kinase in-hibitor protein (RKIP) in health and disease. Biochem Phar-macol 68, 1049-1053.

[15] Mei M, Su B, Harrison K, Chao M, Siedlak S, Previll L, Jack-son L, Cai D, Zhu X (2006) Distribution, levels and phos-phorylation of Raf-1 in Alzheimer’s disease. J Neurochem 99,1377-1388.

[16] Woodgett J (1990) Molecular cloning and expression of glyco-gen synthase kinase-3/factor A. EMBO J 9, 2431-2438.

[17] Grimes C, Jope R (2001) CREB DNA binding activity is in-hibited by glycogen synthase kinase-3 beta and facilitated bylithium. J Neurochem 78, 1219-1232.

[18] Grimes C, Jope R (2001) The multifaceted roles of glycogensynthase kinase 3beta in cellular signaling. Prog Neurobiol 65,391-426.

[19] Hernandez F, Borrell J, Guaza C, Avila J, Lucas J (2002) Spa-tial learning deficit in transgenic mice that conditionally over-express GSK-3beta in the brain but do not form tau filaments.J Neurochem 83, 1529-1533.

[20] Phiel C, Wilson C, Lee V, Klein P (2003) GSK-3alpha reg-ulates production of Alzheimer’s disease amyloid-beta pep-tides. Nature 423, 435-439.

[21] Takashima A (2006) GSK-3 is essential in the pathogenesis ofAlzheimer’s disease. J Alzheimers Dis 9, 309-317.

[22] Jensen M, Mottin M, Cracchiolo J, Leighty R, Arendash G(2005) Lifelong immunization with human beta-amyloid (1-42) protects Alzheimer’s transgenic mice against cognitiveimpairment throughout aging. Neuroscience 130, 667-684.

[23] Arendash G, Jensen M, Salem N, Hussein N, Cracchiolo J,Dickson A, Leighty R, Potter H (2007) A diet high in omega-3fatty acids does not improve or protect cognitive performancein Alzheimer’s transgenic mice. Neuroscience 149, 286-302.

[24] Cracchiolo J, Mori T, Nazian S, Tan J, Potter H, ArendashG (2007) Enhanced cognitive activity–over and above socialor physical activity–is required to protect Alzheimer’s miceagainst cognitive impairment, reduce Abeta deposition, andincrease synaptic immunoreactivity. Neurobiol Learn Mem 88,277-294.

[25] Leighty R, Nilsson L, Potter H, Costa D, Low M, Bales K, PaulS, Arendash G (2004) Use of multimetric statistical analysisto characterize and discriminate between the performance offour Alzheimer’s transgenic mouse lines differing in Abetadeposition. Behav Brain Res 153, 107-121.

[26] Mori T, Town T, Tan J, Yada N, Horikoshi Y, Yamamoto J,Shimoda T, Kamanaka Y, Tateishi N, Asano T (2006) Arundicacid ameliorates cerebral amyloidosis and gliosis in Alzheimertransgenic mice. J Pharmacol Exp Ther 318, 571-578.

[27] Tan J, Town T, Crawford F, Mori T, DelleDonne A, CrescentiniR, Obregon D, Flavell R, Mullan M (2002) Role of CD40ligand in amyloidosis in transgenic Alzheimer’s mice. NatNeurosci 5, 1288-1293.

[28] Leighty R, Runfeldt M, Berndt D, Schleif W, Cracchiolo J,Potter H, Arendash G (2008) Use of artificial neural networksto determine cognitive impairment and therapeutic effective-ness in Alzheimer’s transgenic mice. J Neurosci Methods 167,358-366.

[29] Austin L, Arendash G, Gordon M, Diamond D, DiCarlo G,Dickey C, Ugen K, Morgan D (2003) Short-term beta-amyloidvaccinations do not improve cognitive performance in cogni-tively impaired APP + PS1 mice. Behav Neurosci 117, 478-484.

[30] Ethell D, Shippy D, Cao C, Cracchiolo J, Runfeldt M, BlakeB, Arendash G (2006) Abeta-specific T-cells reverse cognitivedecline and synaptic loss in Alzheimer’s mice. Neurobiol Dis23, 351-361.



[31] King D, Arendash G (2002) Behavioral characterization of theTg2576 transgenic model of Alzheimer’s disease through 19months. Physiol Behav 75, 627-642.

[32] Li S, Janosch P, Tanji M, Rosenfeld G, Waymire J, Mischak H,Kolch W, Sedivy J (1995) Regulation of Raf-1 kinase activityby the 14-3-3 family of proteins. EMBO J 14, 685-696.

[33] Bourne K, Ferrari D, Lange-Dohna C, Rossner S, Wood T,Perez-Polo J (2007) Differential regulation of BACE1 pro-moter activity by nuclear factor-kappaB in neurons and gliaupon exposure to beta-amyloid peptides. J Neurosci Res 85,1194-1204.

[34] Rezai-Zadeh K, Shytle R, Bai Y, Tian J, Hou H, MoriT, Zeng J, Obregon D, Town T, Tan J (2009) Flavonoid-mediated presenilin-1 phosphorylation reduces Alzheimer’sdisease beta-amyloid production. J Cell Mol Med 13, 574-588.

[35] Nehlig A, Daval J, Debry G (1992) Caffeine and the centralnervous system: mechanisms of action, biochemical, metabol-ic and psychostimulant effects. Brain Res Rev 17, 139-170.

[36] Massey L (1998) Caffeine and the elderly. Drugs Aging 13,43-50.

[37] Dall’Igna O, Porciuncula L, Souza D, Cunha R, Lara D,Dall’lgna O (2003) Neuroprotection by caffeine and adeno-sine A2A receptor blockade of beta-amyloid neurotoxicity. BrJ Pharmacol 138, 1207-1209.

[38] Cunha G, Canas P, Melo C, Hockemeyer J, Muller C, OliveiraC, Cunha R (2008) Adenosine A2A receptor blockade preventsmemory dysfunction caused by beta-amyloid peptides but notby scopolamine or MK-801. Exp Neurol 210, 776-781.

[39] Dall’Igna O, Fett P, Gomes M, Souza D, Cunha R, Lara D(2007) Caffeine and adenosine A(2a) receptor antagonists pre-vent beta-amyloid (25-35)-induced cognitive deficits in mice.Exp Neurol 203, 241-245.

[40] Fisone G, Borgkvist A, Usiello A (2004) Caffeine as a psy-chomotor stimulant: mechanism of action. Cell Mol Life Sci61, 857-872.

[41] Daly J (2007) Caffeine analogs: biomedical impact. Cell MolLife Sci 64, 2153-2169.

[42] Chong H, Lee J, Guan K (2001) Positive and negative regu-lation of Raf kinase activity and function by phosphorylation.EMBO J 20, 3716-3727.

[43] Finco T, Baldwin A (1993) Kappa B site-dependent inductionof gene expression by diverse inducers of nuclear factor kappaB requires Raf-1. J Biol Chem 268, 17676-17679.

[44] Gilmore T (2006) Introduction to NF-kappaB: players, path-ways, perspectives. Oncogene 25, 6680-6684.

[45] Valerio A, Boroni F, Benarese M, Sarnico I, Ghisi V, Bres-ciani L, Ferrario M, Borsani G, Spano P, Pizzi M (2006)NF-kappaB pathway: a target for preventing beta-amyloid(Abeta)-induced neuronal damage and Abeta42 production.Eur J Neurosci 23, 1711-1720.

[46] Horrigan L, Kelly J, ConnorT (2004) Caffeine suppress-es TNF-alpha production via activation of the cyclicAMP/protein kinase A pathway. Int Immunopharmacol 4,1409-1417.

[47] Al-Wadei H, Takahashi T, Schuller H (2006) Caffeine stim-ulates the proliferation of human lung adenocarcinoma cellsand small airway epithelial cells via activation of PKA, CREBand ERK1/2. Oncol Rep 15, 431-435.

[48] D’Angelo G, Lee H, Weiner R (1997) cAMP-dependent pro-tein kinase inhibits the mitogenic action of vascular endothe-lial growth factor and fibroblast growth factor in capillary en-dothelial cells by blocking Raf activation. J Cell Biochem 67,353-366.

[49] Echeverria V, Burgess S, Gamble-George J, Zeitlin R, LinX, Cao C, Arendash GW (2009) Sorafenib inhibits nu-clear factor kappa B, decreases inducible nitric-oxide syn-thase and cyclooxygenase-2 expression, and restores work-ing memory in APPswe mice. Neuroscience, doi: 10.1016/j.neuroscience.2009.05.019.

[50] Costa D, Cracchiolo J, Bachstetter A, Hughes T, Bales K, PaulS, Mervis R, Arendash G, Potter H (2007) Enrichment im-proves cognition in AD mice by amyloid-related and unrelatedmechanisms. Neurobiol Aging 28, 831-844.

[51] Gong B, Vitolo O, Trinchese F, Liu S, Shelanski M, Arancio O(2004) Persistent improvement in synaptic and cognitive func-tions in an Alzheimer mouse model after rolipram treatment.J Clin Invest 114, 1624-1634.

[52] Hooper C, Killick R, Lovestone S (2008) The GSK3 hypoth-esis of Alzheimer’s disease. J Neurochem 104, 1433-1439.

[53] Twomey C, McCarthy J (2006) Presenilin-1 is an unprimedglycogen synthase kinase-3beta substrate. FEBS Lett 580,4015-4020.

[54] Lucas J, Hernandez F, Gomez-Ramos P, Moran M, Hen R,Avila J (2001) Decreased nuclear beta-catenin, tau hyperphos-phorylation and neurodegeneration in GSK-3beta conditionaltransgenic mice. EMBO J 20, 27-39.

[55] Engel T, Hernandez F, Avila J, Lucas J (2006) Full reversalof Alzheimer’s disease-like phenotype in a mouse model withconditional overexpression of glycogen synthase kinase-3. JNeurosci 26, 5083-5090.

[56] Potter H, Wefes I, Nilsson L (2001) The inflammation-inducedpathological chaperones ACT and apo-E are necessary cata-lysts of Alzheimer amyloid formation. Neurobiol Aging 22,923-930.

[57] Azam S, Hadi N, Khan N, Hadi S (2003) Antioxidant andprooxidant properties of caffeine, theobromine and xanthine.Med Sci Monit 9, BR325-330.

[58] Paris D, Quadros A, Humphrey J, Patel N, Crescentini R,Crawford F, Mullan M (2004) Nilvadipine antagonizes bothAbeta vasoactivity in isolated arteries, and the reduced cerebralblood flow in APPsw transgenic mice. Brain Res 999, 53-61.

[59] Juarez-Mendez S, Carretero R, Martınez-Tellez R, Silva-Gomez A, Flores G (2006) Neonatal caffeine administrationcauses a permanent increase in the dendritic length of pre-frontal cortical neurons of rats. Synapse 60, 450-455.

[60] Lee W, Anwyl R, Rowan M (1987) Caffeine inhibits post-tetanic potentiation but does not alter long-term potentiationin the rat hippocampal slice. Brain Res 426, 250-256.

[61] Lu K, Wu S, Gean P (1999) Promotion of forskolin-inducedlong-term potentiation of synaptic transmission by caffeine inarea CA1 of the rat hippocampus. Chin J Physiol 42, 249-253.

[62] Martın E, Buno W (2003) Caffeine-mediated presynapticlong-term potentiation in hippocampal CA1 pyramidal neu-rons. J Neurophysiol 89, 3029-3038.

[63] Chen X, Gawryluk J, Wagener J, Ghribi O, Geiger J (2008)Caffeine blocks disruption of blood brain barrier in a rabbitmodel of Alzheimer’s disease. J Neuroinflammation 5, 12.

[64] Zipser B, Johanson C, Gonzalez L, Berzin T, Tavares R,Hulette C, Vitek M, Hovanesian V, Stopa E (2007) Microvas-cular injury and blood-brain barrier leakage in Alzheimer’sdisease. Neurobiol Aging 28, 977-986.

[65] Ujiie M, Dickstein D, Carlow D, Jefferies W (2003) Blood-brain barrier permeability precedes senile plaque formation inan Alzheimer disease model. Microcirculation 10, 463-470.


Journal of Alzheimer’s Disease 17 (2009) 661–680 IOS...

Documents

Transcript of Journal of Alzheimer’s Disease 17 (2009) 661–680 IOS...