VU Research Portal Computational modelling and analysis of brain metabolism in neurodegenerative diseases Supandi, F.B. 2016 document version Publisher's PDF, also known as Version of record Link to publication in VU Research Portal citation for published version (APA) Supandi, F. B. (2016). Computational modelling and analysis of brain metabolism in neurodegenerative diseases. General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal ? Take down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. E-mail address: [email protected]Download date: 22. Jan. 2021
Transcript
VU Research Portal
Computational modelling and analysis of brain metabolism in neurodegenerativediseasesSupandi, F.B.
2016
document versionPublisher's PDF, also known as Version of record
Link to publication in VU Research Portal
citation for published version (APA)Supandi, F. B. (2016). Computational modelling and analysis of brain metabolism in neurodegenerativediseases.
General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal ?
Take down policyIf you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.
Alzheimer’sdisease (AD) is themostwidespreadneurodegenerativedisorder.Studieshave linked metabolic impairment, particularly glucose hypometabolism andmitochondrialdysfunction,withADpathology.Althoughdecreases incerebralglucoseand oxygen uptake have been measured in AD patients, little is known about theinternal changes in the network of energy metabolism. Here we used an algorithmtermedLsei-FBA,whichpredictshowthemetabolicfluxdistributionincentralcarbonmetabolismisaffectedduringthediseaseprocess.ThismakesuseofmeasurementsofgeneexpressioninbraintissueofADpatients.Wepredictadecreaseinglycolyticrateandoxygenconsumption from thegeneexpressionmeasurements that is in linewiththeexperimentalmeasurements(reductionsoftheorderof25%).TheanalysispredictsthatATPsynthesisisreducedmorestrongly,byabout50%.Areductionispredictedforreaction fluxes in the first section of the TCA cycle, and a stronger reductiondownstream of alpha-ketoglutarate. The latter reduction is partially compensated byupregulationoftheGABAshunt.Thetransportofreducingequivalentsfromcytosolintomitochondriaviathemalate-aspartateshuttleispredictedtobesubstantiallyreduced,accompaniedbya small increase in lactateproduction,butwith littleupregulationoftheglycerolphosphateshuttle.Wealso foundthatchanges inmetabolic fluxescanbeassociated with specific brain regions: changes in the middle temporal gyrus, thehippocampus,posteriorcingulatecortexandentorhinalcortexarelarge,whilechangesin the frontal gyrus and primary visual cortex remain smaller. We propose that thecalculationofdetailedmetabolicfluxchangesfromgeneexpressionchangesindiseasedtissuemayresultinusefulpredictions.
97
Introduction
Alzheimer’sdisease(AD)isatypeofneurodegenerativedisease,whichcausesthemostcommon type of dementia among the elderly worldwide. AD neuropathology ischaracterized by accumulation of extra-cellular β-amyloid (Aβ) plaques and intra-cellularneurofibrillarytangles(NFT),thelattercausedbyhyperphosphorylationoftauprotein (Hardy & Selkoe 2002; Braak & Braak 1991). The precise etiologies andpathogenesis of AD still need to be understood. The amyloid cascade hypothesis hasstimulated AD research in past decades, but recently it has been argued that thishypothesis should be rejected (Herrup 2015). Several alternative hypothesesmay beconsidered.Someofthesehypothesesinvolveimpairmentofenergymetabolism.Alossof mitochondrial functionmay precede Alzheimer pathology and is the basis for the‘mitochondrial cascade hypothesis’ for AD (Swerdlow et al. 2014). GlucosehypometabolisminthebrainmayalsoprecedepathologicalsignsofAD,evenbymanyyears, and deteriorating glucosemetabolism has been proposed to contribute to thedevelopmentofAD.
Althoughmeasurements of brainmetabolic rates for glucose and oxygenwithpositronemissiontomography(PET)arecommon(Chen&Zhong2013;Cunnaneetal.2011), little is knownon thedetailsof changes in theenergymetabolicnetworkas awhole. The expression of genes in brain tissue of peoplewhodiedwithADhas beenmeasured formanybrain regions in several studies.This includedmanygeneswhichcode for theenzymes for central energymetabolism.Thegoalof thepresent study istherefore to predict the distribution of fluxes in the metabolic network for energymetabolisminthebrainfromtheavailablegeneexpressionmeasurements.AlthoughanattemptwasmadetopredictchangesinmetabolicfluxesinADbasedonalargemodelofbrainmetabolism(Lewisetal.2010),thiswasbasedoncapacitymeasurementsforone enzyme. It therefore appears useful to investigate whether gene expressionmeasured for many enzymes in the network can be applied to predict changes inmetabolic fluxes. To this end we will apply a recently developed algorithm for thepredictionoffluxchangesfrommeasuredchangesintheexpressionofthegenes(Gavaietal.2015).
The reduction of brain cerebral metabolic rate of glucose (CMRglc) is moreprominent in specific brain areas (Mosconi 2005). PET studies with 18F-fluorodeoxyglucose (FDG) in AD have demonstrated reductions of CMRglc in theparieto-temporal cortex and posterior cingulate cortex.While the disease progresses,thefrontalcortexbecomesinvolved,whilethecerebellum,striatum,basalgangliaandprimaryandsensorimotorcorticesremainrelativelyunaffected(Mosconi2005;Braak& Braak 1991). Gene expression measurements are also available for specific brainregions,allowinganatomicallydetailedpredictionsofmetabolicchanges.
98
Geneexpression studiesofADhavebeen reported for several regions thataredifferentially affected. Liang at al. (Liang et al. 2007; 2008) analysed six regionsincluding entorhinal cortex (EC), hippocampus (HIP), medial temporal gyrus (MTG),posteriorcingulatecortex(PCC),superiorfrontalgyrus(SFG)andprimaryvisualcortex(VCX).Berchtoldatal.(Berchtoldetal.2013)analysedfourregionsincludingentorhinalcortex,hippocampus,post-centralgyrus(PCG)andsuperiorfrontalgyrus.Blalockandhis colleagues (Blalock et al. 2004; Blalock et al. 2011) analysed hippocampal geneexpressionofcontrolsandpatientsacrossdifferent levelsofADseveritybasedontheMini-MentalStateExamination(MMSE)scoreandNFT.Thelastdatasetweconsider,byDunkleyatal.(Dunckleyetal.2006), is fromtheentorhinalcortex.Detailedstatisticaltesting of these transcriptional datasets on the single gene level has already beendiscussed in the original publications.We do not repeat such statistical analysis butpredictthechangeinfluxdistribution.
The Lsei-FBA approach is a recently proposed method to predict changes inmetabolicnetworkfunctionbasedongeneexpressionmeasurement(Gavaietal.2015).The assumption underlying thismethod is that, on average, fluxes catalysed by eachenzymetendtochangetothesamerelativeextentasexpressionofthegenescodingfor(subunits of) that enzyme. This assumption is not expected to be true at the level ofeachenzymeindividually,butmaybepredictivewhenanentirenetworkisconsidered.Because of the correlative approach which is not based on detailed and preciseregulatory mechanisms and enzyme kinetic effects, it cannot be expected that thisalgorithmwillyieldveryaccuratepredictions,butweproposethatthealgorithmmaybeusefultoindicatedirectionandorderofmagnitudeofchangesinbrainmetabolism.
TheLsei-FBAapproachhasbeendiscussedandtestedinpreviousstudies.Theinitial description of analysing flux distribution using thismethodwas illustrated byapplying it to one dataset from AD patients (Gavai et al. 2015). Lsei-FBA was alsoapplied to a largenumber of datasets onParkinson’s disease (Chapter 4, this thesis).Here, we present an analysis involving more datasets for AD obtained for severaldistinct brain regions. The predicted changes in fluxes will be compared withmeasurementswhichareavailableforsomefluxes,whilepredictionsforseveralotherunknownfluxesarederived.
Materialsandmethods
Datasets
The microarray data used in this study were obtained from the Gene ExpressionOmnibus (GEO) (Edgar et al. 2002) and are summarized in Supplementary Table 1.DatasetsofLiangatal.andDunkleyatal.containedgeneexpressionprofilesfromlaser-capturedmicrodissectedneurons,whiledatasetsfromBerchtoldatal.containsprofiles
99
fromwholetissuedissectionofseveralbrainregionsthatareknowntobeaffectedbythedisease.Blalockatal.,intheirdatasetof2004and2011,comparedstagesofADindifferent severity - incipient,moderate and severe. The first study useswhole tissuedissected from the CA1 hippocampal region, while the latter uses microdissectedneuronsofthesamearea.ThegeneexpressionwasmeasuredonAffymetrixchips(seeSupplementary Table 1 for the type of chip). The Affymetrix CEL files were pre-processed and normalized using the RMA method (Irizarry et al. 2003). Log2transformed values were used to calculate differences in expression levels of ADpatientsagainsthealthycontrolsmeasuredinthesamelaboratoryonthesametypeofchip.
Metabolicnetworkmodelofbrainmetabolism
Themetabolic networkmodel used in this study consists of themajor pathways forcentral carbon metabolism. These pathways include glycolysis, pentose phosphatepathway (PPP), TCA cycle, oxidative phosphorylation (OxPhos), reducing equivalentshuttle mechanisms, glutamate-glutamine cycling, gamma-aminobutyric acid (GABA)shuntandtransportofmetabolitesacrossthemembraneswhichseparateintracellularcompartments.Theselectedreactionswereimportedfromthereconstructionofhumanmetabolism in the BiGG database (Schellenberger et al. 2010). A complete list of thereactionsinthenetwork,alongwiththelistofmetabolitesaregiveninSupplementaryTable 2 and 3. Supplementary Figure shows the network scheme. This model is thesameasusedinbySupandi&vanBeek(Chapter4,thisthesis)andisdiscussedthereindetail. We prefer to use this core model that has been accurately curated manually,rather than a ‘genome-scale’ model where each reaction has not been checkedindividuallyandwhichisoftennotfreefromerrors.
Analysisoffluxdistribution
Themethod toanalyse fluxdistribution in thehealthynormalanddiseasedbrainhasbeendiscussedindetailpreviously(Gavaietal.,2015;Chapter4,thisthesis).Here,wewilldiscussthemethodbriefly.Themetabolicsystemisassumedtobein(orverycloseto)steadystateinthesensethatchangesinconcentrationsofmetabolitesinsidethecellperunit timearenegligible relative to the fluxes into andout of themetabolitepool.Substrateuptakeandreleasemeasurementsareusedasconstraints inthemodel.Forthehealthyelderlyhumanbrain(55-65years),theuptakeratesofglucose,andreleaseof lactate, glutamineandpyruvate for thebrainwere reported tobe0.203, -0.0092, -0.011and -0.0024µmolgwetweightofbrain-1min-1 respectively (Lying-Tunelletal.1980).Asmallfluxwhichamountsto6.9%ofglycolysisismeasuredinthePPPinthenormal brain (Dusick et al. 2007). Pyruvate carboxylation is 13% and glutamate-glutamine cycling fluxes is 62% of the value of the total glucose uptake in the brain(Hyderetal.2006).TheGABAshuntfluxis32%oftheglucoseuptakevalue(Pateletal.
100
2005). In the healthy brain, a cost function which maximizes ATP synthesis wasassumed, as discussed extensively in (Gavai et al. 2015), The flux distribution wassubsequentlysolvedusingthelinearprogrammingroutineLinpfromthepackageLIM(Soetaert&vanOevelen2009)fortheRprogrammingenvironment.
ForADpatients, theLsei-FBAmethod isused toestimate the fluxdistribution,basedonthechangesingeneexpressiondataandthefluxdistributioninnormalbraincalculated above (Gavai et al. 2015). For each reaction, the average fold change fromcontrolswascomputedfortheexpressionofeachgeneassociatedwiththebiochemicalreactions in the model (Supplementary Table 4). If more than one gene-proteincombinationisassociatedwiththeenzymereaction,thecontributionisaggregated.Aninitial rough estimate of the flux for every reaction in the model is calculated bymultiplying the fold change for gene expression in the AD patients and the fluxestimated for theassociatedbiochemical reaction for thehealthybrain.Next, the fluxestimatewasrefinedbasedonfluxbalanceinthemodel(SupplementaryFigure).
Undertheassumptionofabsolutefluxbalanceoftheinternalmetabolitesinthemodel and of zero backflux for the irreversible reactions which are given inSupplementary Table 2, a cost function was minimized consisting of the sum of thesquareddeviationsbetweenfinalestimatedfluxandinitialroughestimateofthefluxascalculated above. This procedure was described in detail in Gavai et al. (2015) andSupandi & van Beek (Chapter 4, this thesis). The equations of this problem of leastsquareswithequalities(balancedfluxes)andinequalities(irreversiblereactions)weresolved using the least squares with equality and inequality conditions (lsei) methodfromthelimSolvepackage(Soetaertetal.2009).Thisalgorithm,termedLsei-FBA,hasalreadybeendescribedindetailin(Gavaietal.2015).
Results
ADgeneexpressionpatternacrossbrainregions
Foreachreaction, theaverage foldchangesofmRNAexpression forpatientswithADrelative to healthy controls are shown in Supplementary Table 4 for various brainregions.Downregulatedgenesareshowningreen,whileupregulatedgenesareshowninred.FoldchangesobservedbyLiangatal.arelargercomparedtotheBerchtoldatal.andBlalockat al.datasets,while in thedataset fromDunckleyat al., changes inmostreactionsaresmall.Inthisrespectthelastdatasetisanoutlier.
Overall, fold changes formost of the reactions in the glycolytic pathway, TCAcycle,malate-aspartateshuttleandoxidativephosphorylationtendtoshowreductionofgene expression in all regions. Increased expression can be observed in nucleotidediphosphatekinasereaction(NDPK1m)andglutaminesynthetasereaction(GLNS).The
101
glucose-6-phosphate dehydrogenase (G6PDH2r) and 6-phosphoglucolactonase (PGL)reactions at the entrance of the pentose phosphate pathway are also upregulated inmostoftheregions,withtheentorhinalcortexintheLiangetal.datasetasconspicuousexception forall fourenzymes.Pleasenote thatstatisticalanalysesof thechangesarefoundintheoriginalpublicationsonthesedatasetsandarenotrepeatedhere.
PredictedfluxdistributionduringAlzheimer’sdisease
WepredictchangesinmetabolicfluxesdistributionbetweenthenormalhealthybrainandADbrainwith theLsei-FBAalgorithm.The fluxdistribution for thenormalbrainhasbeenpredictedanddiscussedin(Gavaietal.2015),andisgiveninFigure1A.Thisdistributionhasalsobeencomparedextensively(Gavaietal.2015)withthepredictedfluxdistributioninalargemodelofbrainmetabolism(Lewisetal.2010).Notethattheoxygenuptakemeasuredby(Lying-Tunelletal.1980),1.68µmolgwetweightofbrain-1min-1,ishigherthanpredicted.Whenwehadsetthemeasuredoxygenconsumptionasa constraint, the flux balance analysis did not yield a feasible solution,which reflectsthat themeasured oxygen uptake is higher than can be accounted for by the carbonsubstratestakenupbythebrain.
ComparisonoffluxpredictioninalldatasetsforADshowsapatternofreductionin the glucose and oxygen uptake, glycolytic pathway, TCA cycle and oxidativephosphorylation, while increases in flux are predicted for the PPP pathway, malicenzyme (ME) and pyruvate carboxylase (PC) reaction. In the Liang et al. dataset fordifferentbrainregions,glycolyticfluxispredictedtobereduced28.1%inHIP,28%inEC,26.9%inPCC,25.7%inMTG,13.8%inVCXand10.7%inSFG.ATPsynthesisinthemitochondriaispredictedtobereducedby51.7%inPCC,43.5%inHIP,39%inMTG,26.8%inVCX,25.7%inECand17.6%inSFG.Oxygenuptakeispredictedtobereduced51.4%inPCC,42.8%inHIP,38.4%inMTG,26.6%inVCX,25.2%inECand17.5%inSFG(SupplementaryTable5).NotethattherelativereductioninoxygenconsumptionandATPsynthesisisinseveralcasessubstantiallylargerthanthereductioninglucoseuptake.Figure1B-1DshowsfluxespredictedbasedontheLiangatal.dataset.Figure1B and 1C shows predicted flux distributions during AD in the hippocampus andposterior cingulate cortex, respectively. Figure 1D shows the flux distribution in theSFG,theregionwhichshowedthesmallestfluxdeclinesintheLiangatal.study.
Smaller changes were found for predicted fluxes across brain regions in theBerchtoldatal.dataset.Glycolyticpathwayfluxeswerereducedbyabout10.7%(EC)to6.8%(PCG),ATPsynthesiswasreducedbyabout14.4%(HIP) to7.9%(PCG),oxygenuptakereducedbyabout14%(HIP)to7.7%(PCG).
Accordingtotheprediction,fluxesinthepentosephosphatepathway(PPP)areincreased inmost of the regions ranging from 18% to 155% increase. No consistentincrease or decrease in flux is found in the glutamine-glutamate cycle and the GABAshunt.Thesubstantialpredictedchangesinfluxformalicenzyme(ME2m)andpyruvate
102
carboxylase(PCm)inthemitochondriaarenoteworthy.Thesetwoenzymespotentiallyform a bypass ofMDHm; however, this result needs to be interpretedwith care (seeDiscussion).Theglycerolphosphateshuttletotransportreducingequivalentsfromthecytosol into themitochondria is predicted to be slightly activated for somedata sets.Predictedfluxes inthemalate-aspartateshuttleareconsistently,althoughmoderately,decreased.
In many of the datasets the relative reduction in the AKDGm reaction ispredicted to be considerably stronger than in the section between citrate and alpha-ketoglutaratewhichprecedesitintheTCAcycle.Thisiscompensatedtoasmalldegreebyan increase intheGABAshunt flux.Aconsiderabledecrease inalpha-ketoglutarateproductionfromglutamateviaaspartatetransaminaseisalsopredicted.However,thisdecrease is exactly balanced by the change in alpha-ketoglutarate export from themitochondriainexchangeformalatewhichispartofthemalate-aspartateshuttle.IfwelookattheHIPdatasetofLiang(SupplementaryTable5)asanexample,weseethatthedecrease inalpha-ketoglutarate(AKG)uptakebytheAKGDmreaction is0.169µmolgwetweightofbrain-1min-1whichisbalancedbya0.097µmolgww-1min-1decreaseinAKGproductionbytheICDHxmreaction,a0.028largeruptakebytheGLUDymreactionanda0.045largeruptakebytheABTArmreaction.
Comparisons across stages of the diseasewith different severity can bemadebased on the Blalock at al. datasets for the hippocampal region. The flux predictionsshowlargerchangesinfluxesduringthemoreseverestageofthedisease,exceptfortheseverestageintheBlalockatal.2011datasetmeasuredinexcisedneuronsratherthanwhole tissue, where the fluxes are relatively unchanged. In the Blalock 2004 at al.dataset, fluxes inglycolyticpathwayarereducedby8.7%(MOD)and19%(SEV),ATPsynthesisisreducedby19.6%(MOD)and31.4%(SEV),andoxygenuptakeisreducedby19.4%(MOD)and31%(SEV).
Flux distribution in healthy brain (A) and during AD from the Liang et al. dataset inhippocampalregion(B),posteriorcingulatecortex(C)andsuperiorfrontalgyrus(D)inµmolg(wet)brain-1min-1.Blacknumbers,fluxduringnormalcondition;greennumbers,fluxdecreasedduringADandrednumbers, increasedfromthenormalcondition.Notethat forclaritynotallseparate biochemical steps are plotted: oxaloacetate is for instance first transaminated toaspartatebeforebeingtransportedacrossthemitochondrialmembraneaspartof themalate-aspartate shuttle. GLC, glucose; G3P, glyceraldehyde 3-phosphate; RU5PD, ribulose-5-phosphate;PYR,pyruvate;LAC,lactate;CIT,citrate;AKG,alpha-ketoglutarate;SUCC,succinate;MAL, malate; OAA, oxaloacetate; GLU, glutamate; GLN, glutamine, GABA, 4-aminobutanoate
Under normal physiological conditions the brain relies almost exclusively onglucose as its main substrate for energy production. ATP is mostly produced in themitochondriafromtheoxidationofglucoseunderaerobicconditions.Almostalloftheglucose is oxidized to water and CO2 (Clarke & Sokoloff 1999). Thus, glucosehypometabolismcan severely affectbrain energy supply.Alternative substratesotherthan glucose may be used under certain circumstances. Glycogen provides anemergency store of carbohydrates, but there is a limited amount (Clarke & Sokoloff1999)whichhastoberestoredfromglucose.Ketonebodies,intheformofacetoacetateandβ–hydroxybutyrate,mayreplaceglucosetoa largeextentduringextendedfastingor starvation. Compared to the substantial decrease in glucose uptake, brain ketoneuptake is not significantly affected during aging or AD (Lying-Tunell et al. 1981;Castellanoetal.2015).Nevertheless,supplyingketonebodieshasbeenproposedasatherapeuticmeasureduringAD(Cunnaneetal.2011).
Brainenergymetabolismissupportedbythreemajorprocesses-cerebralbloodflow,oxygenconsumptionandglucosemetabolism(Cunnaneetal.2011).Allof thesecanbemeasured,amongotherswithpositronemissiontomography(PET).Decreasesincerebral metabolic rate (CMR) for glucose (Minoshima et al. 1997; Li et al. 2008;Mosconi2005;Mosconietal.2008;Cunnaneetal.2011)andoxygen(Ishiietal.1996)have been reported in AD. The decrease in metabolic rate in the temporal lobemeasured by positron emission tomography (PET) was 23% for oxygen (Ishii et al.1996)and36%forglucose(Ishiietal.1997).SupplementaryTable6givesanoverviewof availablemeasurements and a comparisonwith thepresent predictions fromgeneexpression.Theglobalcerebralmetabolicrateforglucose(CMRglc)isreducedby20–25%inAD(reviewedinCunnaneetal.2011).Thedirectionandorderofmagnitudeoffluxchangespredictedinthepresentstudyagreeswithmeasuredglucoseandoxygenuptakeratesmeasuredexperimentally.
Decrease in cerebral glucose metabolism and severity of AD pathology bothdiffer acrossbrain regions (Mosconi2005).ThemostvulnerableareasduringADarethemedialtemporallobe,entorhinalandperirhinalcortex,andhippocampus(Mosconi2013). When compared with age-matched controls, AD individuals show metabolicreductionsinregionalglucosemetabolisminposteriorcingulatecortexduringdiseaseprogression (~21% compared to our prediction of 25%, Supplementary Table 6)
105
(Minoshimaetal.1997). Incontrast, theprimarymotorandvisualcortex,cerebellumandbasalganglianucleiarelessseverelyaffected(Mosconi2005).Thisisalsoseeninourprediction fromtheLiangatal.dataset,where fluxchangesare larger inPCC,EC,HIP andMTG,while SFG andVCX show smaller changes (See SupplementaryTable5and6).
ThereisevidencethatAβcanhaveaseveredetrimentaleffectonmitochondrialfunction through several pathways, such as impairment of oxidativephosphorylation,elevationof reactiveoxygen speciesproduction, alterationofmitochondrialdynamicsand interaction with mitochondrial proteins. This may subsequently leads to loss ofsynapticfunctionandneuronalcelldeath(reviewedinEckertetal.2011).Itisnotyetclearwhetherdecrease in function leads todecreaseofmetabolismor theotherwayaround.
Asaconsequenceofreducedglucosemetabolism,adecreaseinATPproductionhasbeenestimated,whichtendstoincreaseduringthecourseofthedisease,byupto50%inadvancedAD(Hoyer1992).Thisissimilartoourpresentpredictions.AnalysisfrommRNAgene expressiondata (Liang et al. 2008)with our algorithm leads to theprediction that energymetabolism in the brain is strongly compromised, particularlyaffecting reactions in the mitochondria. Some key enzymes in the citric acid cyclepathway-pyruvatedehydrogenase(PDHm),cytochromeoxidase(CYOOm3)andalphaketoglutarate dehydrogenase (AKGDHm) are reduced during AD (Blass 2001;Chandrasekaran et al. 1994). A decrease in AKGDm activity has been found bybiochemicalassay(Bubberetal.2005).Basedona57%decreaseinAKGDmactivity,asubstantial decrease inmetabolic rates in glutamatergic and cholinergic neuronswaspredicted in an analysiswith a largemodel of brainmetabolism, butmetabolic ratesweremaintainedinGABAergicneuronswheretheGABAshuntpathwaywasactivatedto compensate for reduced AKGDm activity (Lewis et al. 2010). Here we predictmetabolicfluxchangesforbraintissuebasedongeneexpressiondistributedacrossthewholemetabolic network. Thepredictions from themodel of Lewis et al. (2010) andourpresentanalysisunderlinemitochondrialdysfunctioninAD.
Ourmodel prediction is that lactate transport out of the brain is increased inmanybrainregions,whileincreasesinlactateandpyruvatelevelshavebeenmeasuredin the cerebrospinal fluid of AD patients (Liguori et al. 2014; Redjems-Bennani et al.1998).However,formanybrainregionsanincreaseduptakeofpyruvateispredictedinourpresentanalysis.IngeneralsimilartrendsinmetabolisminADpatientsareseenincalculated predictions and independent measurements for glucose and oxygenmetabolism.
While glucose and oxygen uptake and lactate and pyruvate levels have beenmeasured in AD and comparisons with predicted fluxes for these metabolites arepossible, the present analysis provides further predictions for fluxes internal to themetabolic network that have not yet been measured. Given the qualitative
106
correspondence between predictions and measurements where available, the modelpredictions for those parts of the metabolic network that are not yet accessible formeasurementsinpatientstudiesmaybeofinterestandcanamongothersbeusedforfurtherassessmentofthevalidityofthemodelifthesefluxesbecomemeasurableinthefuture.
The predictions suggest that malate-aspartate flux is substantiallydownregulatedinmanybrainregions.ThisconformswithlowerNADHproductionasaconsequence of lower glycolytic activity, and is enhanced further by NADH used forlactate production. In contrast, the glycerol phosphate shuttle flux is somewhatupregulatedinmanyregions.
Thecoremodelofbrainenergymetabolismusedinthepresentstudycontainsmalic enzyme andpyruvate carboxylase. Acting in concert these enzymes can bypassthe mitochondrial malate dehydrogenase. ME2m and PCm have been included inseveral models of brain energy metabolism (Calvetti & Somersalo 2013; Cakir et al.2007; Occhipinti et al. 2010; Lewis et al. 2010). Both ME and PC occur in the brainaccording to biochemical measurements. As such these are added to the model.However,ME is neuronal,mitochondrial andNADPdependent,while PC is astrocyticandmitochondrial(Hassel2001).Theselocationswouldsuggestthattheconsiderableincrease in flux via both enzymes is coupled via pyruvate diffusion between neuronsand astrocytes. Some flux is redistributed from neuronal MDHm to production ofpyruvate, CO2 and NADPH in the neuronal mitochondria. The decrease in NADHproduction by MDHm would lead to a reduction in predicted O2 uptake andmitochondrialATPproduction. In theastrocytesoneATPwouldthenbeconsumedtosynthesize oxaloacetate. The predicted effects of the loop formed byME2mandPCmmean a decrease in predicted net mitochondrial ATP production, while NADPH isformedintheneuron(potentiallyusefulforreactiveoxygenspeciesdetoxification),andoxaloacetateissynthesizedintheastrocyte,resultinginaspecialformofanaplerosis.
However,weemphasizethatwhiletheenzymesconnectedwithME2mandPCmarefoundinbraintissue,itispossiblethatregulatoryprocessespreventthatthiscycleis actually more active in brain tissue during AD. If this cycle is not upregulated inreality, thiswouldmeanthatmitochondrialATPproduction isunderestimatedbyourpresent calculations. It is remarkable that the predicted flux via ME2m and PCm isincreaseddespiteadecrease inmRNAexpressionlevelswhichis foundinmostcases.This is the consequence of the substantial reduction in MDHm expression which isgreaterthanthereductionofME2mandPCmexpression.
There is a strong relationship of brain hypometabolism and AD. The questionarises whether deteriorating brain metabolism causes AD or vice versa. Brainhypometabolismmaybecausedbyimpairmentofglucoseuptakeand/ordownstreamstepsofglucoseutilizationandmayevencausallycontributetothedevelopmentofADneuropathology (Cunnane et al. 2011; Mosconi 2013). This may involve
107
neurodegenerationandaccumulationofAβplaques,developingtheclinicalsymptomsofADi.e.memorylossandbehaviouralchange.SuchchangesinbrainfunctionmayinturnaffectATPconsumptionandsynthesis.The reason thatbrainhypometabolism isconsidered an upstream event during AD development is because it is known to bealteredatveryearlystages(Cunnaneetal.2011).Moreover,whetherhypometabolismin brain specifically affects only glucosemetabolismalone or also other brain energysourcessuchasketonebodiesasglucosereplacementarenotyetresolved(Cunnaneetal.2011;Eckertetal.2011).
Geneexpressionchangesdifferwhencomparingdatasetsfrommeasurementsofa whole tissue and dissected neurons (Supplementary Table 5). Flux changes in theLiang at al. dataset on neurons tend to show larger changes as compared to theBerchtoldatal.datasetforwholebraintissue.Stempleratal.concludedthatmetabolicgenesshowhigheraccuracy forpredictingADprogressionandcognitivedecline fromhippocampalgeneexpression indissectedneurons than forwhole tissue(Stempleretal. 2012).However, gene expression changes are less consistentwhen comparing the2004wholetissuedatasetandthe2011microdissectedneurondatasetofBlalocketal.(Blalocketal.2004;Blalocketal.2011).
Limitationsofthisstudy
Ofcoursethereareseveralimportantlimitationstothisstudy.TheconceptthatatthelevelofametabolicnetworkfluxestendtochangeproportionallytotheexpressionoftheassociatedmetabolicgeneshasbeentestedonlyonbraintissuefromAlzheimerandParkinson’sdiseasepatients.Nevertheless,geneexpressionlevelshavebeenshowntogive good predictions of flux distribution in yeast (Lee et al. 2012). Other proposedalgorithmstopredictmetabolicfluxfromgeneexpression(changes)arelesswellsuitedfor the present situation where many genes are altered relatively modestly inexpression (Gavai et al. 2015). The results for PD and AD look promising, althoughrelativelyfewfluxmeasurementswereavailableforcomparison.
Ourmodeldoesonlyincludethecoreofenergymetabolism.Thesereactionsarealso connected with the broader metabolic network represented by genome-scalereconstructionsofmetabolism.However,comparison(Gavaietal.2015)ofresultsfromthepresentmodelwithalargemodelofbrainmetabolism(Lewisetal.2010)suggestedthattheomittedconnectionstobroadermetabolismcarrynegligiblefluxes.
Our model does not take into account the compartmentation of brainmetabolism in several cell types, such as neurons, glia and endothelium. This wasinspiredby thenatureof thedatawheregeneexpressionhasnotbeendifferentiatedbetweenthecelltypes.AsillustratedbytheexampleofthedifferentlocationsofME2mandPCm(seeabove)thismayhaveconsequencesfortheoutcome.Thepresentanalysisthereforeincorporatestheassumptionthatchangesintransportprocessesbetweenthevariouscelltypesdonothavealargeeffectonthemetabolicsystem.
108
Conclusion
Inthisstudy,wedescribetheapplicationofarecentlydevelopedalgorithmtopredictchangesinmetabolicfluxesbasedongeneexpressionchangesfromAlzheimer’sdiseasepatients.Fluxesthroughcentralcarbonmetabolismarepredictedtobereducedinmostregions.ThisreductiondiffersacrossregionsandtosomeextentparallelsdifferencesinAD pathology. Reduced metabolism via alpha ketoglutarate dehydrogenase in themiddle of the TCA cycle is partially compensated via the GABA shunt. Changes inmetabolic fluxes are associatedwith specific regional pattern: changes in themiddletemporalgyrusandthehippocampus,posteriorcingulatecortexandentorhinalcortexarelargerwhilechangesthefrontalgyrusandprimaryvisualcortexremainsminimum.ThisparallelstheseverityofADpathology.
Supplementarymaterials
SupplementaryTable1–SupplementaryofalldatasetsusedinthisstudySupplementaryTable2–List of reactions included in this model after extraction from the Recon1 knowledgebase for human metabolism Supplementary Table 3 – List of metabolites included in this model, extracted from the Recon 1 knowledgebaseSupplementaryTable4–Average fold changes of gene expression for each reactionSupplementaryTable5–Results for flux prediction for Alzheimer's disease datasetsSupplementaryTable6–Predicted flux in this study (between parentheses) compared to experimental results in Alzheimer's diseaseSupplementaryFigure–Reconstructedmetabolicreactionnetwork
References
Berchtold, N.C. et al., 2013. Synaptic genes are extensively downregualted acrossmultiple brain regions in normal human aging and Alzheimer’s disease.NeurobiologyofAging,34(6),pp.1653–1661.
Blalock,E.M.etal.,2004.IncipientAlzheimer’sdisease:microarraycorrelationanalysesrevealmajor transcriptional and tumor suppressor responses.Proceedingsof theNational Academy of Sciences of the United States of America, 101(7), pp.2173–2178.
Blalock, E.M. et al., 2011. Microarray analyses of laser-captured hippocampus revealdistinct gray and white matter signatures associated with incipient Alzheimer’sdisease.JournalofChemicalNeuroanatomy,42(2),pp.118–126.
Blass, J.P., 2001. Brain metabolism and brain disease: Is metabolic deficiency theproximate causeofAlzheimerdementia? JournalofNeuroscienceResearch, 66(5),
Bubber, P. et al., 2005. Mitochondrial abnormalities in Alzheimer brain: Mechanisticimplications.AnnalsofNeurology,57(5),pp.695–703.
Cakir, T. et al., 2007. Reconstruction and flux analysis of coupling betweenmetabolicpathways of astrocytes and neurons: application to cerebral hypoxia.TheoreticalBiology&MedicalModelling,4,p.48.
Calvetti, D. & Somersalo, E., 2013. Quantitative in silico Analysis of NeurotransmitterPathwaysUnderSteadyStateConditions.FrontiersinEndocrinology,4,p.137.
Castellano, C.A. et al., 2015. Lower brain 18F-fluorodeoxyglucose uptake but normal11C-acetoacetate metabolism in mild Alzheimer’s disease dementia. Journal ofAlzheimer’sDisease,43(4),pp.1343–1353.
Chen, Z. & Zhong, C., 2013. Decoding Alzheimer’s disease from perturbed cerebralglucose metabolism: Implications for diagnostic and therapeutic strategies.ProgressinNeurobiology,108,pp.21–43.
Clarke, D. & Sokoloff, L., 1999. Circulation and EnergyMetabolism of the Brain. In G.Siegeletal.,eds.BasicNeurochemistry:Molecular,CellularandMedicalAspects.NewYork:RavenPress,pp.637–669.
Dunckley, T. et al., 2006. Gene expression correlates of neurofibrillary tangles inAlzheimer’sdisease.NeurobiologyofAging,27(10),pp.1359–1371.
Dusick, J.R. et al., 2007. Increased pentose phosphate pathway flux after clinicaltraumatic brain injury: a [1,2-13C2]glucose labeling study in humans. Journal ofCerebralBloodFlowandMetabolism,27(9),pp.1593–1602.
Hyder, F. et al., 2006. Neuronal-glial glucose oxidation and glutamatergic-GABAergicfunction.JournalofCerebralBloodFlowandMetabolism,26(7),pp.865–77.
Irizarry, R.A. et al., 2003. Exploration, normalization, and summaries of high densityoligonucleotidearrayprobeleveldata.Biostatistics,4(2),pp.249–264.
Ishii, K. et al., 1996. Decreased medial temporal oxygen metabolism in Alzheimer’sdiseaseshownbyPET.JournalofNuclearMedicine,37(7),pp.1159–1165.
Ishii, K. et al., 1997. Reduction of cerebellar glucose metabolism in advancedAlzheimer’sdisease.JournalofNuclearMedicine,38(6),pp.925–928.
Li,Y.etal.,2008.RegionalanalysisofFDGandPIB-PET images innormalaging,mildcognitive impairment, and Alzheimer’s disease. European Journal of NuclearMedicineandMolecularImaging,35(12),pp.2169–2181.
Liang, W.S. et al., 2007. Gene expression profiles in anatomically and functionallydistinct regions of the normal aged human brain. Physiological Genomics, 28(3),pp.311–322.
Liang,W.S. et al., 2008. Alzheimer’s disease is associatedwith reduced expression ofenergy metabolism genes in posterior cingulate neurons. Proceedings of theNational Academy of Sciences of the United States of America, 105(11), pp.4441–4446.
Liguori, C. et al., 2014. CSF lactate levels, τ proteins, cognitive decline: a dynamicrelationship in Alzheimer’s disease. Journal of Neurology, Neurosurgery, andPsychiatry.
Lying-Tunell,U.etal.,1980.Cerebralbloodflowandmetabolicrateofoxygen,glucose,lactate, pyruvate, ketone bodies and amino acids.ActaNeurologicaScandinavica,62,pp.265–275.
Lying-Tunell,U.etal.,1981.Cerebralbloodflowandmetabolicrateofoxygen,glucose,lactate, pyruvate, ketone bodies and amino acids.ActaNeurologicaScandinavica,63,pp.337–350.
Mosconi, L., 2005. Brain glucose metabolism in the early and specific diagnosis ofAlzheimer’sdisease:FDG-PETstudiesinMCIandAD.EuropeanJournalofNuclearMedicineandMolecularImaging,32(4),pp.486–510.
Mosconi, L., 2013. Glucose metabolism in normal aging and Alzheimer’s disease:Methodological and physiological considerations for PET studies. Clinical and
