measuring cortical connectivity in alzheimer...

Journal of the International Neuropsychological Society (2016), 22, 138–163.Copyright © INS. Published by Cambridge University Press, 2016.doi:10.1017/S1355617715000995

Measuring Cortical Connectivity in Alzheimer’sDisease as a Brain Neural Network Pathology:Toward Clinical Applications

Stefan Teipel,1,2 Michel J. Grothe,2 Juan Zhou,3 Jorge Sepulcre,4 Martin Dyrba,2 Christian Sorg,5 AND Claudio Babiloni61Department of Psychosomatic Medicine, University of Rostock, Rostock, Germany2DZNE, German Center for Neurodegenerative Diseases, Rostock, Germany3Center for Cognitive Neuroscience, Neuroscience and Behavioral Disorders Program, Duke-NUS Graduate Medical School, Singapore4Division of Nuclear Medicine and Molecular Imaging, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts5Department of Psychiatry and Neuroradiology, TUM-NIC Neuroimaging Center, Technische Universität München, Munich, Germany6Department of Physiology and Pharmacology “V. Erspamer”, University of Rome “La Sapienza”, Rome, Italy; IRCCS San Raffaele Pisana of Rome, Italy

(RECEIVED April 2, 2015; FINAL REVISION July 31, 2015; ACCEPTED September 20, 2015)

Abstract

Objectives: The objective was to review the literature on diffusion tensor imaging as well as resting-state functional magneticresonance imaging and electroencephalography (EEG) to unveil neuroanatomical and neurophysiological substrates ofAlzheimer’s disease (AD) as a brain neural network pathology affecting structural and functional cortical connectivityunderlying human cognition. Methods: We reviewed papers registered in PubMed and other scientific repositories on theuse of these techniques in amnesic mild cognitive impairment (MCI) and clinically mild AD dementia patients compared tocognitively intact elderly individuals (Controls). Results: Hundreds of peer-reviewed (cross-sectional and longitudinal) papershave shown in patients with MCI and mild AD compared to Controls (1) impairment of callosal (splenium), thalamic,and anterior–posterior white matter bundles; (2) reduced correlation of resting state blood oxygen level-dependent activityacross several intrinsic brain circuits including default mode and attention-related networks; and (3) abnormal powerand functional coupling of resting state cortical EEG rhythms. Clinical applications of these measures are still limited.Conclusions: Structural and functional (in vivo) cortical connectivity measures represent a reliable marker of cerebralreserve capacity and should be used to predict and monitor the evolution of AD and its relative impact on cognitive domainsin pre-clinical, prodromal, and dementia stages of AD. (JINS, 2016, 22, 138–163)

Keywords: Dementia diagnosis, Prognosis, PET, MRI, EEG, Treatment trials

INTRODUCTION

Alzheimer’s disease (AD) is the most frequent neurodegen-erative disorder causing cognitive impairment, disabilities,and finally dementia in aged people. This disease is related toan extra-cellular brain accumulation of beta-amyloid (Aβ)and intracellular tangles of hyperphosphorylated tau peptidesthat affect cortical neuronal networks related to cognitivefunctions (Pievani, de Haan, Wu, Seeley, & Frisoni, 2011).Soluble Aβ elicits a toxic signaling cascade by receptorsleading to synaptic impairments, intraneuronal Aβ42 aggregates,and correlated cognitive deficits (Dziewczapolski, Glogowski,Masliah, & Heinemann, 2009). This compromised signaling

possibly leads to or aggravates aggregation of hyperpho-sphorylated tau protein and formation of neurofibrillary tangles(NFTs). Preclinical data suggest a non-linear relationship of Aβpeptide levels with synaptic plasticity (Parihar & Brewer, 2010).Aβ at lower levels may play a physiological role in synapticplasticity (Puzzo et al., 2008), whereas higher levels of Aβ mayimpair synaptic activity (Shankar et al., 2007) or may even beinduced by synaptic activity (Cirrito et al., 2008) in metabolicactive cortical regions (Buckner et al., 2009). The bidirectionalinteraction between local amyloid accumulation and metabolicactivity may be an important determinant for the cognitiveeffects of AD pathological changes and represent a potentialdeterminant for brain reserve capacity, that is, the ability of thebrain to maintain function despite an increasing load of neuro-degenerative lesions.Structural, molecular, and functional neuroimaging studies

have replicated the findings of a systematic spread of AD

INS is approved by the AmericanPsychological Association to sponsor Continuing Education for psychologists.INS maintains responsibility for thisprogram and its content.

Correspondence and reprint requests to: Stefan Teipel, DZNE, GermanCenter for Neurodegenerative Diseases, Gehlsheimer Str. 20, 18147Rostock. E-mail: [email protected]

138

https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S1355617715000995Downloaded from https:/www.cambridge.org/core. National University of Singapore (NUS), on 10 Jul 2017 at 10:53:03, subject to the Cambridge Core terms of use, available at

https:/www.cambridge.org/core/termshttps://doi.org/10.1017/S1355617715000995https:/www.cambridge.org/core

pathology markers. First, patterns of brain atrophy, asrevealed by structural magnetic resonance imaging (MRI),most closely resemble the distribution of neurofibrillary tan-gles across different clinical stages of disease (Chetelat et al.,2002; Whitwell et al., 2007). Second, positron emissiontomography (PET) of amyloid in the brain suggests asequence of amyloid accumulation similar to stages derivedfrom autopsy studies (Thal, Attems, & Ewers, 2014). Third,18FDG-PET detection of cortical glucose consumptionuncovers a sequence of metabolic changes that overlapsbetween both, pattern of brain atrophy/tangle accumulation andamyloid deposition (Choo et al., 2007). An emergent regionalpattern from these multimodal-imaging studies suggests acharacteristic network of key brain regions that are involved ina specific temporal sequence across the clinical stages of AD.Compared to the mentioned neuroimaging techniques,

human brain neural networks have been more directly probedby diffusion tensor imaging (DTI) and resting-state func-tional MRI (rs-fMRI). These techniques have identifiedconsistent structurally and functionally connected brain net-works in the human brain (Fox et al., 2005; van den Heuvel,Mandl, Kahn, & Hulshoff Pol, 2009). The destruction of keyhubs of these networks may mediate the effect of molecularpathology on cognitive performance in AD (Koch et al.,2014); along the same line, brain reserve may act throughmodulation of such networks to preserve cognitive functionin the presence of molecular pathology (Bozzali et al., 2014).One step further, it has even been implicated that thefunctional connectivity within a network and the strengths ofpositive functional associations between intrinsic networksdetermines the regional spread of different types of neuro-degenerative changes, such as neurofibrillary tangles,amyloid accumulation, Lewy bodies, and TDP43 deposition,which in turn lead to distinct clinical disease entities, such asAD (tau and amyloid), Lewy body dementia (alpha-synuclein),or frontotemporal dementia (tau and TDP43) (Zhou, Gennatas,Kramer, Miller, & Seeley, 2012).On this basis, an understanding of the functional and

structural organization of brain networks may further ourunderstanding of neurodegenerative disease pathogenesisand brain reserve. In a back-translation approach, imagingcan provide evidence to support the hypothesis of networkspecificity of AD. These findings inform basic sciencestudies on potential molecular mechanisms that account forthe network specificity of pathological features of AD.Complementary, hypotheses on molecular mechanisms, suchas oxidative stress of highly connected network hubs(Buckner et al., 2009) or prion-like spread of pathogenicprotein conformations along strong anatomical connections(Braak & Del Tredici, 2011), can be tested in the humanin vivo framework using multimodal imaging. Associationsbetween metabolic characteristics of a cortical hub region andmolecular and atrophic changes, both in large cross-sectionalsamples across clinically and biomarker-based disease stagesand in longitudinal cohorts spanning the time of conversionfrom cognitively healthy to early dementia stages, can helpto test the pathogenetic validity of specific molecular

mechanisms in humans. In a complementary perspective, brainfunctional and structural connectivity represents the potentialsubstrate of brain reserve capacity in the presence of significantAD pathology and comorbid pathologies such as cere-brovascular disease. Understandingmechanisms of brain reserveprovides both potential targets for preventive interventions andin vivo surrogate endpoints to test themechanistic mode of actionof a specific intervention. From a clinical point of view, the studyof network connectivity may provide a diagnostic marker ofearly disease as well as a prognostic marker at an individual levelwhere the integrity of key functional networks will influence thelikelihood of cognitive decline at a given level of molecularpathology (Teipel et al., 2013).In the following sections, we will introduce established as

well as emerging methods to determine functional and struc-tural cortical connectivity in the living human brain, coveringacquisition and analysis of such data, and describe the maincontributions of these methods to our current understandingof pathogenesis, diagnosis, and disease monitoring in AD.A specific emphasis is put on multimodal approaches. In thisframework, limitations and perspectives will be outlined thatare related to the validity of a method (in respect to assumedunderlying neurobiological substrates) as well as its imple-mentation into a clinical setting in the middle to the far future.

Methodology of Connectivity Analysis

Functional MRI provides a unique window to study AD’simpact on coherent slowly fluctuating brain activity, that is,intrinsic brain networks (Fox & Raichle, 2007). At rest,macroscopic brain activity fluctuates slowly at frequenciesbelow 0.1 Hz; such fluctuations are detectable by rs-fMRI.Slowly fluctuating activity is coherent or synchronized acrossbrain regions (i.e., functional connectivity), constituting acouple of intrinsic brain networks such as the default mode ordorsal attention network (Fox et al., 2005). Intrinsic brainnetworks represent a highly conserved and robust form oforganized macroscopic brain activity, that is, comparablenetworks are observed in distinct species such as mice,monkeys, and humans (Vincent et al., 2007), in distinctstages of ontogeny such as after preterm birth, babies,children, adults, and elder persons (Doria et al., 2010), and atdistinct stages of awareness from sleep to different domainsof goal-directed behavior (Smith et al., 2009). The rs-fMRIsignal is of special relevance for many analytical methodsassessing functional connectivity. It provides spontaneouslow-frequency fluctuations of blood oxygen level-dependent(BOLD) signals (Figure 1) and makes possible investigationsof the network architecture of brain systems (Biswal, Yetkin,Haughton, &Hyde, 1995; Biswal et al., 2010; Fox & Raichle,2007; Lu et al., 2011). Compared to conventional fMRIstudies, rs-fMRI is a task-free and data-driven neuroimagingtechnique that can be easily acquired in cognitively impairedpopulations. There are four main methods in which rs-fMRIhas been frequently applied. Seed-based analysis uses cor-relations of rs-fMRI spontaneous low-frequency fluctuationsbetween a seed region and the rest of the brain (Figure 1-I)

Cortical connectivity in Alzheimer’s disease 139



(Biswal et al., 1995). It describes relatively simple patterns offunctional connectivity of single brain areas. More advancedapproaches take advantage of multiple simultaneous braininteractions to pull apart coherent brain networks. Forinstance, independent component analysis (ICA) and clus-tering methods have been extremely useful to isolate modulesof the brain and to create subdivisions and parcellationschemes of the cerebral cortex (Figure 1-II and 1-III)(Beckmann, DeLuca, Devlin, & Smith, 2005; Yeo et al.,2011). Third, graph theory has facilitated the comprehensionand visualization of complex brain interactions by trans-forming connectivity data to nodes (vertices) and links(edges) (Figure 1-IV) (Bullmore & Sporns, 2009; Rubinov &Sporns, 2010). Several graph theoretical metrics quantify“network hubs” of the brain, that is central regions that act asintegration stations for connecting otherwise segregated brainnetworks (Figure 1-IV, Basic Metrics in Graph Theory)

(Buckner et al., 2009; Crossley et al., 2013; Sporns, Honey,& Kotter, 2007; van den Heuvel & Sporns, 2011; Zuo et al.,2012), while other metrics—such as clustering coefficient,path length, small-worldness, or rich-club organization—emphasize modularity or efficient communications. Finally,diffusion graph theory algorithms have been proposed tostudy putative pathways for the spread of pathology throughinterconnected brain systems (Figure 1-IV, Diffusion/Spreading Metrics in Graph Theory) (Raj, Kuceyeski, &Weiner, 2012; Sepulcre, Sabuncu, Becker, Sperling, &Johnson, 2013; Sepulcre, Sabuncu, Yeo, Liu, & Johnson,2012). More advanced analysis methods additionally incor-porate the phase lag information of the time signal, such asGranger causality analysis (Goebel, Roebroeck, Kim, &Formisano, 2003; Granger, 1969) or dynamic causal modeling(Friston, Harrison, & Penny, 2003), to derive a causalrelationship between brain regions. Those techniques have

Fig. 1. Overview of functional connectivity analysis methods for resting-state functional MRI. Several neuroimaging techniques use low-frequency spontaneous fluctuations in brain activity to analyze functional connections in the human brain. (I) Seed-based correlationanalysis is a widely used approach to characterize functional connectivity patterns of seed regions or voxels of interest. (II) Independentcomponent analysis is a signal processing method that is able to separate independent sources from mixed signals of the rs-fMRI data.(III) Clustering techniques, such as k-means or hierarchical clustering, are useful approaches to generate spatial partitions based onfunctional connectivity profiles. (IV) Graph theory refers to a wide field of research that focuses on the analysis of graphs, defined bypairwise associations of nodes, and network structures. In neuroimaging, graph theoretical metrics have been used to describe multiplenetwork properties of the human brain. Several examples of basic measures and a diffusion/spreading method are displayed in the figure.

140 S. Teipel et al.



been applied less frequently in the context of AD (Agostaet al., 2010; Dauwels, Vialatte, Musha, & Cichocki, 2010;Liu et al., 2012; Miao, Wu, Li, Chen, & Yao, 2011), as theyare prone to provide different results caused by randomvariance in the data (Daunizeau, David, & Stephan, 2011).Further in-depth discussion on methodology of fMRIdata analysis can be found in two recent reviews by

Krajcovicova, Marecek, Mikl, & Rektorova (2014) andDennis & Thompson (2014).The term structural connectivity refers to the interconnec-

tion between neurons or brain regions by nerve fibers. Theintegrity of fiber tracts can be assessed in vivo usingdiffusion-weighted imaging techniques (Le Bihan, Turner,Douek, & Patronas, 1992). These allow the mapping of the

Fig. 2. Overview of structural connectivity analysis methods for diffusion tensor imaging. Diffusion-weighted imaging assesses thediffusion of water molecules that is restricted by the tissue structure. In diffusion tensor imaging the diffusion process is modeled as atensor, which is estimated from the non-diffusion image (B0) and the diffusion-weighted scans. The tensor model can be represented as anellipsoid with three principal axes (λ1, λ2, λ3), the length of which reflects the diffusion tendency along each direction. (I) Fiber trackingalgorithms use the shape and the direction of the ellipsoid to trace the most likely fiber pathways. (II) Scalar tissue integrity measures, suchas the fractional anisotropy (FA) or mean diffusivity (MD), characterize the shape of the ellipsoid. In large tracts with mainly parallelorientation of the fibers, for example, in the corpus callosum, the ellipsoid is cigar-shaped such that FA reaches its largest values whileMD is relatively low. In the liquor, the water is not restricted in any direction leading to a ball-shaped ellipsoid, indicated by high MD andlow FA. Both measures have intermediate values in gray matter regions as well as crossing fiber areas where the ellipsoid may be moreoblate-shaped. Statistical analysis approaches can be categorized in hypothesis-based region of interest analysis and data-driven voxel-based analysis methods.




diffusion process of water molecules and rely on the principlethat the diffusion of water is restricted by the tissue structure.In the context of AD, diffusion tensor imaging (DTI) is beingused most frequently. The minimum scan protocol comprisesone non-diffusion image, often referred to as B0 image,and six diffusion-weighted images for opposing gradientdirections (Basser, Mattiello, & LeBihan, 1994). From thesedata, a tensor model is estimated that can be represented as anellipsoid with three principal axes, the length of whichreflects the diffusion tendency along each direction (λ1, λ2,λ3; Figure 2, Diffusion Tensor Modeling). Scalar tissueintegrity measures characterizing the shape of the ellipsoidinclude the fractional anisotropy (FA), mode of anisotropy(MO), as well as axial (aD), radial (rD), and mean diffusivity(MD), providing complementary information about theconfiguration of the ellipsoid axes (Le Bihan et al., 2001)(Figure 2-II). In extension to characterizing single voxels,one can perform fiber tracking to infer the white matterpathways within the brain, with the two most popularapproaches being streamline and probabilistic tractography(Mori, Crain, Chacko, & Van Zijl, 1999; Mori & Zhang,2006) (Figure 2-I). Both methods do not only use theshape but also the principal direction of the ellipsoid to tracethe most likely fiber pathways. Patient’s characteristics,group differences, or statistical associations betweenfiber tract integrity measures and, for instance, measures ofcognitive functioning can be assessed using hypothesis-based region of interest analysis (Figure 2, AnalysisMethods). Complementary data-driven voxel-based analysisallows the evaluation of group differences or statisticalassociations on the level of each single voxel (Figure 2,Analysis Methods).

CONTRIBUTION OF CORTICALDISCONNECTION TO AD PHENOTYPEAND DEVELOPMENT

Functional Connectivity Changes in the Courseof AD

Neuroimaging approaches including rs-fMRI have produceda tide of direct support for the network-based neurodegen-eration hypothesis in living humans (Buckner et al., 2005;Greicius, Srivastava, Reiss, & Menon, 2004; Raj, Kuceyeski,& Weiner, 2012; Seeley, Crawford, Zhou, Miller, &Greicius, 2009; Zhou, Gennatas, Kramer, Miller, & Seeley,2012). AD, the most common neurodegenerative disorder,begins with dysfunction in episodic memory before pro-gressing to involve posterior cortical cognitive functions suchas word retrieval, visuospatial function, arithmetic, andpraxis. In parallel to the symptoms, AD is associated withatrophy and hypometabolism predominantly in posteriorhippocampal, cingulate, temporal, and parietal regions,which collectively resemble the default mode network(DMN) as mapped in healthy subjects with task-free fMRI(Greicius, Krasnow, Reiss, & Menon, 2003). The DMN is

typically found deactivated during cognitive tasks requiringexternally focused attention and activated during internallyfocused mental tasks, such as episodic memory retrieval,mental state attribution, and visual imagery (Buckner,Andrews-Hanna, & Schacter, 2008; Mason et al., 2007;Raichle et al., 2001; Shulman et al., 1997).In addition to the regional atrophy and neuronal hypome-

tabolism affecting DMN nodes, disruptions in functionalconnectivity of the DMN in AD dementia have been widelyreplicated (Agosta et al., 2012; Binnewijzend et al., 2012;Greicius, Srivastava, Reiss, & Menon, 2004), and have beenlinked to core memory and visuospatial deficits (Greiciuset al., 2004; Supekar, Menon, Rubin, Musen, & Greicius,2008; Zhang et al., 2010). Intriguingly, connectivity disrup-tion and impaired task-related down regulation of the DMNmay already emerge during the presymptomatic phase of ADas modeled cross-sectionally on the basis of imaging evi-dence of cortical amyloid pathology (Hedden et al., 2009;Sperling et al., 2009) or an apolipoprotein E4 (APOE4)positive genotype, which is a major genetic risk factor for lateonset AD (Damoiseaux et al., 2012; Machulda et al., 2011;Persson et al., 2008). Particularly the functional isolation ofthe posterior cingulate from its main interaction sites in themedial temporal lobe and the medial prefrontal cortexappears to emerge early in the disease process and was foundto be related to worsening episodic memory function in MCIsubjects (Bai et al., 2011). Of interest, the posterior cingulatecortex showed reduced connectivity in MCI patients even inthe absence of gray matter atrophy, which was only detectableat the stage of fully developed AD (Gili et al., 2011). ADpathology, however, also presents with non-memory symp-toms such as impairments in executive function, language, andvisuospatial abilities, particularly at the clinically manifestdementia stage of the disease. While there is also evidence ofimpaired executive function network connectivity (includingthe dorsal lateral prefrontal cortex and superior parietal lobe) inAD and MCI (Brier et al., 2012; Liang, Wang, Yang, Jia, &Li, 2011; Sorg et al., 2007), the relation of these changes tothe emergence of respective clinical symptoms still requiresfurther investigation.In the task-free setting, DMN activity correlates inversely

with activity in multiple brain regions in health, including thesalience network (SN) (Fox et al., 2005; Greicius & Menon,2004; Seeley et al., 2007). Many forms of emotional saliencerequire a focusing of attention toward homeostatic demandsand behavioral responses (“here and now”), creating aneed to de-prioritize attention to internal (“there and then”)ruminations about one’s personal past or future, functionsattributed to the DMN (Seeley et al., 2007). Such opposingnetwork functions might engender between-network com-petition for brain resources (Deco & Corbetta, 2011), shiftsbetween “binary brain configurations” (Jones et al., 2012), ordirect reciprocal suppression of one network in favor of theother, orchestrated by nodes within the two networks or by anodal “switch” positioned elsewhere to reconfigure networkdynamics in response to shifting conditions (Menon &Uddin, 2010). Questionnaire- and laboratory-based studies




suggest that patients with AD show retained or enhancedinterpersonal warmth and empathy, mutual gaze, and emo-tional morality (Mendez & Shapira, 2009; Rankin et al.,2006; Sollberger et al., 2009; Sturm et al., 2011). Emotionalcontagion (sharing emotional states with others) appears toincrease linearly across the healthy through MCI to ADdementia spectrum (Sturm et al., 2013). In line with thesefindings, AD patients were shown to exhibit increased SNconnectivity compared to controls, which was associatedwith decreased DMN connectivity (Zhou et al., 2010). SNenhancement has been widely replicated in the growing ADtask-free fMRI literature. Evidence to date suggests that SNhub connectivity escalates in genetic at risk groups (APOE4carriers) and prodromal stages of AD (Bai et al., 2009;Brier et al., 2012; Machulda et al., 2011), correlates withemotion intensification symptoms (Balthazar et al., 2013), isaccompanied by SN hyperperfusion (Hsieh, Kao, Huang, &Chou, 2010), and may wane in later disease stages (Brieret al., 2012).

Multimodal imaging of functional connectivity andamyloid load

In addition to the connectivity changes within and acrossparticular networks, emerging graph theoretical approachesalso detected changes in general network topology in AD,characterized by a lower clustering coefficient or an increasedcharacteristic path length, which renders the whole-brainnetwork metrics closer to the theoretical values of randomnetworks and largely supports the hypothesis of disruptedglobal information integration in AD (Sanz-Arigita et al.,2010; Wang et al., 2007). Impaired parallel informationtransmission efficiency and reduced intra- and inter-modularconnectivity of the posterior DMN and executive controlnetwork were also detected in healthy APOE4 carrierscompared to non-carriers (Wang et al., 2015). Future work isneeded to develop network imaging methods equippedto handle both intra- and inter-network connectivity andtopology profiles, corresponding to the broad range ofclinico-anatomical presentations associated with the disease.Aberrant functional connectivity of intrinsic networks is

intimately linked with AD’s amyloid pathology (Drzezgaet al., 2011; Sheline, Raichle, et al., 2010; Sperling et al.,2009). Using PiB-PET (i.e., Pittsburgh Compound B PET) todetect in vivo amyloid-β plaques in combination withrs-fMRI in asymptomatic and mildly impaired elderly withamyloid positivity, Drzezga and colleagues found that thehigher the amyloid plaque load the more the global centralityin the parietal cortex is reduced (centrality measures for eachvoxel its degree of functional connectivity with all othervoxels of the brain) (Drzezga et al., 2011). More specifically,for several networks, such as default mode and differentattention networks, the spatial distributions of plaques andnetwork functional connectivity were highly correspondentin individuals with prodromal AD, suggesting that plaquespread is linked with a networks’ connectivity (Myers et al.,2014). Furthermore, in network centers of high connectivity

and high plaque load, this relationship changes, that is, themore plaques the more connectivity is impaired, demon-strating the detrimental effect of amyloid pathology onintrinsic functional connectivity when certain levels ofpathology are overstepped. These studies demonstrate therelevance of intrinsic brain networks for pathophysiology andpathogenesis of AD particularly in early stages of the disease.Of interest, while APOE4 genotype has been consistentlyassociated with increased amyloid load (Morris et al., 2010),detrimental effects of this genotype on functional brain con-nectivity have also been observed independently of amyloidpathology (Sheline, Morris, et al., 2010). Although geneticeffects on functional connectivity disruptions are best studiedfor the APOE4 genotype, there are now also initial reports ofconnectivity-altering effects of other risk genes for AD, suchas polymorphisms in tau- or KIBRA-related genes (Bai et al.,2014; Wang et al., 2013).

Structural Disconnection in the Course of AD

Several cross-sectional DTI studies revealed white matterintegrity changes in AD dementia patients compared tohealthy controls in wide spread commissural, association,and limbic fiber tracts, whereas extracortical projecting fibertracts were found to be relatively preserved until advancedstages of the disease (Bozzali et al., 2002, 2001; Fellgiebelet al., 2005; Friese et al., 2010; Huang, Friedland, & Auchus,2007; Medina et al., 2006; Naggara et al., 2006; Stahl et al.,2007; Xie et al., 2006; Zhang et al., 2007). In MCI subjects,disruptions of white matter integrity were mainly reported forlimbic fiber tracts with direct connections to medial temporallobe structures, including the posterior and parahippocampalcingulum, the perforant path, the fornix, and the uncinatefasciculus (Fellgiebel et al., 2005; Kalus et al., 2006; Sextonet al., 2010; Zhang et al., 2007).In an attempt to better characterize the earliest white matter

changes and their regional progression in the course of ADpathogenesis, recent DTI studies have more and morefocused on asymptomatic at-risk populations, such as healthysubjects carrying AD-susceptibility genes, most notably theAPOE4 allele (Bendlin et al., 2010; Kljajevic et al., 2014;Westlye, Reinvang, Rootwelt, & Espeseth, 2012; Xionget al., 2011), but also other risk-associated candidate genes(Braskie et al., 2012, 2011; Forde et al., 2014; Liang, Li,et al., 2015; Lyall et al., 2014; Voineskos et al., 2011), orcompletely dominant familial AD mutations (Ringman et al.,2007; Ryan et al., 2013). Other recent studies examined whitematter changes in asymptomatic individuals showingbiomarker evidence of amyloid or tau pathology (Bendlinet al., 2012; Chao et al., 2013; Gold et al., 2014; Kantarciet al., 2014; Molinuevo et al., 2014; Racine et al., 2014;Stenset et al., 2011). These cross-sectional studies of at-riskpopulations are now increasingly being complemented bylongitudinal follow-up studies that allow relating the detectedimaging abnormalities to future clinical outcomes (Douaudet al., 2013; Fletcher et al., 2013; Mielke et al., 2012; Teipel,Meindl, et al., 2010; Zhuang et al., 2012). Together the




findings converge on a pattern of microstructural whitematter changes in AD that begin and are most severe inlimbic tracts, including the fornix, uncinate fasciculus, andposterior and parahippocampal fibers of the cingulum, andsequentially extend to include more lateral temporoparietalassociation fibers, commissural fibers of the splenium, andfinally long-ranging association tracts involving the frontalwhite matter. Importantly, first microstructural alterations ofthe limbic tracts were already detectable in presymptomaticsubjects, years before they developed cognitive deficits and ata time the macrostructural gray matter volume was stillwidely preserved (Fletcher et al., 2013; Ringman et al., 2007;Zhuang et al., 2012).In accordance with models of the representation of

cognitive function across neuronal networks in the humanbrain, the clinical consequences of microstructural whitematter changes were found to depend on the specific whitematter tracts affected. Thus, disruptions of limbic fiber tracts,most notably the fornix and the parahippocampal whitematter, show associations with impaired episodic memoryfunction (Fellgiebel et al., 2008; Huang & Auchus, 2007;Sexton et al., 2010), whereas executive function deficitsappear to be more closely associated with fiber disruptions inlong-ranging association tracts connected to the frontal lobe(Chen et al., 2009; Grambaite et al., 2011; Huang & Auchus,2007). Of interest, memory performance in healthy elderlysubjects was found to depend primarily on fornix integrity,whereas memory performance in MCI subjects showedgreater dependence on parahippocampal white matterintegrity (Metzler-Baddeley et al., 2012). This cognitiverealignment from the more severely damaged fornix tothe parahippocampal white matter in MCI was found to bebeneficial for residual memory function in this conditionand depended on a relatively spared structure of the basalforebrain cholinergic system, thought to be implicated inplastic brain responses (Ray et al., 2015).While most studies focused on decreases in FA or increases

in MD as scalar diffusion markers of microstructural whitematter damage, simultaneous assessment of the full range oftensor-derived diffusion indices, including FA and MD, butalso axial (aD) and radial (rD) diffusivities, and the MO, mayprovide more detailed information about the specifics ofwhite matter degeneration in AD. Thus, in experimentalstudies on animal models increases in rD have beenspecifically associated with myelin degeneration, whereaschanges in aD were more reflective of direct axonal damage(Song et al., 2003). In AD, the type of microstructuralchanges as reflected by the distinct diffusivity indices wasfound to differ between limbic, commissural, and associationfiber tracts, indicating differing processes of tissue disruptionamong fiber populations (Huang et al., 2012). In general,increases in absolute diffusivities (i.e., MD, aD, and rD) werefound to be more sensitive markers of AD-related whitematter changes than decreases in FA, particularly in earlyand prodromal stages of the disease (Acosta-Cabronero,Williams, Pengas, & Nestor, 2010; Bosch et al., 2012). Therather counterintuitive observation of relatively increased FA

values along cortico-fugal and cortico-petal fiber tracts in ADand MCI (Douaud et al., 2011) may likely be explained bythe loss of intracortically projecting crossing fiber tracts, andhence a more linear shape of the resulting diffusion tensor, ashas been illustrated by a parallel increase of FA and MO inmotor related tracts (Douaud et al., 2011; Teipel, Grothe,et al., 2014).In addition to the regional analysis of scalar diffusion

indices, complementary information on structural networkorganization can be obtained from the graph theoreticalanalysis of individual whole-brain connectivity networksderived from tractography-based reconstructions. In ADdementia, structural connectivity networks exhibit alteredtopological network metrics, such as increased shortest pathlengths, decreased local and global efficiency, and decreasednumber of rich-club hub nodes (Daianu et al., 2013, 2015; Loet al., 2010; Shao et al., 2012). These topological networkchanges were also shown to account for core memory andexecutive function deficits (Reijmer et al., 2013), and werealready detectable in cognitively normal individuals withhigh amyloid burden (Fischer, Wolf, Scheurich, & Fellgiebel,2015) and asymptomatic APOE4 carriers (Brown et al.,2011). However, current DTI-based fiber tracking algorithmsare limited in their ability to resolve crossing and touchingfiber bundles, which are highly prevalent fiber configurationsin the human white matter (Jeurissen, Leemans, Tournier,Jones, & Sijbers, 2013). Recently developed model freereconstruction techniques based on high angular resolutiondiffusion data or diffusion spectrum imaging allow a moreaccurate reconstruction of crossing fiber tracts (Dell’Acqua &Catani, 2012; Wedeen et al., 2008), and may be used in thefuture to study age- and AD-related structural connectivitychanges in greater detail (Reijmer et al., 2012; Teipel, Lerche,et al., 2014).

Multimodal Imaging of Structural and FunctionalConnectivity Changes

Multimodal brain connectome approaches using bothrs-fMRI and DTI could provide exciting new insights onstructure-function relationships and how these are affected bydisease. In health, the presence of a direct fiber connection isalmost always correlated with functional connectivity in thecorresponding brain regions (Damoiseaux & Greicius, 2009).However, the presence of functional connectivity betweendistinct brain regions is not necessarily suggesting the pre-sence of a direct fiber connection, and the dependence ondirect structural connections between network nodes variesamong the different large-scale functional networks (Horn,Ostwald, Reisert, & Blankenburg, 2014). Table 1 lists keystudies on combined fMRI and DTI in AD. Within the DMNit has been found that functional connectivity strength islargely predefined by the structural integrity of fiber tractsconnecting the key nodes of this network, that is, the dorsalcingulum bundle connecting medial frontal and parietalnodes, and the ventral/parahippocampal cingulum connectingthe posterior cingulate with the medial temporal lobe




(Greicius, Supekar,Menon,&Dougherty, 2009; Teipel, Bodke,et al., 2010; van den Heuvel, Mandl, Luigjes, & HulshoffPol, 2008). Accordingly, the functional disconnection ofthe DMN in MCI and AD is paralleled by a decliningintegrity of these underlying fiber tracts (Hahn et al., 2013;Soldner et al., 2012; Weiler et al., 2014). Similarly, workingmemory related network breakdown occurred along structu-rally defined networks when evaluated using task fMRI andDTI in healthy aging and MCI individuals (Teipel et al.,2015). Further evidence for a close relation between whitematter degeneration and cortical dysfunction in aging andAD comes also from studies showing distinct associationsbetween decreasing fiber integrity and declining regionalhypometabolism in connected cortical areas (Bozoki,Korolev, Davis, Hoisington, &, Berger, 2012; Cross et al.,2013; Glodzik et al., 2014; Kuczynski et al., 2010; Villainet al., 2010). Of interest, a recent multimodal MRI studyfound that integrity of the fornix was positively correlatedwith hippocampal-thalamic functional connectivity strengthin normally aging subjects, but not in subjects with MCI,indicating that pathologic aging processes may alter therelationship between functional and structural connectivitycharacteristics within specific brain systems (Kehoe et al., 2015).Finally, combined assessment of structural and functional

abnormalities in asymptomatic young adult and middle-agedAPOE4 carriers indicated that abnormalities in functionalnetwork communication may precede the breakdown ofstructural white matter connections in the pathogenesis of AD(Matura et al., 2014; Patel et al., 2013). However, other stu-dies in asymptomatic APOE4 carriers have found paralleldecreases in functional and structural connectivity (Heiseet al., 2014) or even more pronounced structural networkchanges (Chen et al., 2014). Longitudinal studies usingmultimodal imaging assessments are needed to provide moredetail on the specific sequence of structural and functionalconnectivity alterations in the course from presymptomatic toclinically manifest AD.

Multimodal Imaging of Regional Distribution andProgression Patterns of Pathologic Alterations onthe Basis of the Brain’s Connectivity Architecture:MRI, fMRI, and PET

Besides its utility in studying functional and structuralconnectivity alterations in the course of normal and patho-logical aging, connectivity information from rs-fMRI andDTI may also be used to study the underlying mechanism ofthe distinct regional distribution and progression patterns ofpathologic alterations in AD and other neurodegenerativedementias. Thus, different neurodegenerative dementiaforms, such as AD, frontotemporal dementia, or semanticdementia, are not only characterized by distinct clinical pre-sentations, but also by different patterns of regional brainatrophy, which show only partial overlap or no overlap at allbetween dementia syndromes in the early clinical stages.Using functional connectivity information from healthy

individuals, it could be demonstrated that the specific atrophypatterns observed in distinct neurodegenerative diseasesresemble specific functional connectivity networks in thehuman brain, which largely correspond to the respectiveclinical presentation (Seeley et al., 2009; Zhou et al., 2012).Intriguingly, one recent study assessed functional con-nectivity patterns of the most atrophic regions in three distinctclinical variants of AD, namely early-onset AD, logopenicaphasia, and posterior cortical atrophy. In accordance withthe syndrome-specific clinical presentations, they found thatthe functional connectivity pattern of the most atrophicregion in early-onset AD resembled anterior salience andright executive-control networks, in logopenic aphasia itresembled the language network, and the functional con-nectivity pattern of the most atrophic region in posteriorcortical atrophy corresponded to the higher visual network.These findings suggest that, although degeneration inAD dementia generally targets the DMN, deviations fromthe typical regional atrophy pattern in the form ofsyndrome-specific neurodegenerative variants are driven bythe involvement of specific networks outside the DMN(Lehmann et al., 2013).Information about the functional and structural con-

nectivity architecture of the healthy human brain has alsobeen used to construct predictive models of the regionaldistribution of pathologic changes in AD and other neuro-degenerative dementias. Table 1 lists key studies on fMRIand DTI together with molecular imaging modalities.A region’s total amount of functional connections with otherregions in the healthy brain (i.e., its “functional hub”character) was shown to be predictive of the regional amountof amyloid accumulation in AD as measured with amyloid-sensitive PET imaging (Buckner et al., 2009; Myers et al.,2014). Moreover, an “epidemic spreading model” that con-sidered axonal propagation of amyloid proteins along thehealthy structural connectome in combination with regionalclearance mechanisms was able to explain approximately50% of the variance in real amyloid deposition patterns asobserved by amyloid-sensitive PET (Iturria-Medina, Sotero,Toussaint, & Evans, 2014). Thus, this model strongly supportsthe hypothesis that regional amyloid deposition likelihood isexplained to a large extent by the effective (i.e., connectional)anatomical distance from specific outbreak regions estimatedto lie in the anterior and posterior cingulate cortex. Similarapproaches have also been used to successfully predictregional atrophy severity and progression of atrophy based onstructural (Crossley et al., 2014; Raj et al., 2012, 2015) andfunctional (Zhou et al., 2012) connectomic brain features, suchas a region’s total connectivity in the healthy brain or theconnectional (rather than Euclidean) distance of this region tothe site where atrophy first manifests.These data provide first evidence in humans for hypotheses

on molecular disease mechanisms derived from preclinicalstudies, including increased vulnerability of highly con-nected network hubs due to increased amyloid accumulationand oxidative stress (Cirrito et al., 2005; Spires-Jones &Hyman, 2014), or prion-like spread of pathogenic protein




Table 1. Multimodal imaging studies relevant for AD

Studies Modalities Main findings Comments

DTI + (fMRI/MRI)(Dyrba, Grothe, et al.,2015)

DTI, rsfMRI,anatomical MRI

Group separation in AD dementia (n = 28)vs. controls (n = 25) structural MRI >DTI > rsfMRI using multiple kernelSVM.

Number of subjects was low, relative noveltywas multimodal machine learningapproach. Strict cross-validation.

(Kehoe et al., 2015) DTI, rsfMRI Association of functional connectivitybetween hippocampus and thalamusseeds and linear diffusion coefficient offornix tract was significantly positive in22 healthy controls, but was notsignificantly different from zero in 19MCI individuals.

Several diffusion parameters were assessed,but only effects of one of these wasreported as significant.

(Balachandar et al.,2015)

DTI, rsfMRI 15 AD patients showed decreasedfunctional connectivity within the DMN,and significantly increased functionalconnectivity in the executive networkcompared with 15 healthy controls, but noalterations in diffusion markers in anyregion.

No direct comparison between functional andstructural connectivity changes. Lack of FAchanges in AD patients is not consistentwith the majority of previous findings.

(Liang, Chen, et al.,2015)

DTI, rsfMRI In 24 MCI individuals, decreased FA ofventral cingulum was associated withdecreased functional connectivitybetween ventral and dorsal anteriorcingulum.

Hypothesis driven approach for assessment ofanterior cingulum connectivity, which isfunctionally connected with the saliencenetwork.

(Dyrba, Chen, et al.,2015)

DTI, anatomicalMRI

DTI was superior to anatomical MRI topredict amyloid positivity in MCI(n = 70 MCI, 25 controls) using SVM.

Medium number of subjects in a multicentreDTI dataset. Strict cross-validation.

(Ray et al., 2015) DTI, anatomicalMRI

Memory performance depended primarilyon fornix integrity in healthy elderlysubjects (n = 20) and onparahippocampal white matter integrity inMCI (n = 25). This beneficialrealignment in MCI depended oncholinergic basal forebrain integrity.

Potential method to probe plastic brainresponses in neurodegenerative diseasesand their relation to cholinergic systemintegrity. Only indirect volumetric markerof cholinergic system integrity was used.

(Jacobs et al., 2015) DTI, task fMRI 18 MCI individuals showed decreaseddeactivation in areas with decreaseddiffusion, and increased activation inareas with increased diffusion duringobject recognition compared to 18healthy controls.

Functional activation pattern was used tosearch for white matter regional FAchanges. No assessment of shape ofdiffusion tensor in areas with FA increasesin MCI.

(Teipel et al., 2015) DTI, task fMRI 12 MCI individuals showed decreasedantero-posterior functional and structuralconnectivity during working memoryperformance compared with 12 controlsusing three way joint independentcomponent analysis.

Assessing FA and mode of anisotropy asmeasures of fiber directionality and shapeof the diffusion tensor. Data drivenapproach without cross-validation, limitingthe generalizability of the findings.

(Vidal-Pineiro et al.,2014)

DTI, rsfMRI,anatomical MRI,perfusion MRI(ASL)

Functional disconnection between anteriorand posterior DMN nodes in healthyaging (n = 116 elders) correlated withatrophy in DMN areas and the cingulumbundle, but not with cerebral blood flow.

Comprehensive multimodal approach tostudy brain wide correlates of antero-posterior DMN disconnection as a sensitivemeasure of brain aging. Associations werenot controlled for age as a possible indirectdriver of the correlations.

(Salami, Pudas, &Nyberg, 2014)

DTI, rsfMRI, taskfMRI

Age-related increases in intra-hippocampalfunctional connectivity were associatedwith declining memory function, as wellas structural and functional corticaldisconnection of the hippocampal system.

Very large sample of healthy subjectsspanning the whole adult age range(n = 339; 25–80 y). Evidence for a malignnature of age-related hippocampalhyperconnectivity, probably related toa disinhibition effect.




Table 1: (Continued )


(Hahn et al., 2013) DTI, rsfMRI 23 AD dementia patients, 28 MCIindividuals and 26 healthy controlstractography originated from seed pointsfrom 8 intrinsic connectivity networkswith nodes derived from the rsfMRI data.Edge distributions for the whole graphand for selected networks (i.e., defaultmode and attention network) revealedprogressively impaired structuralconnectivity in patients.

Pattern of structural connectivity was drivenby functional connectivity pattern, but nodirect comparison between both measures.

(Wee et al., 2012) DTI, rsfMRI Group separation of 10 MCI individualsfrom 17 controls based on combined DTIand rsfMRI using multiple kernel SVM.

Strict cross-validation, the very low numberof subjects limits the generalizability of thefindings.

(Teipel, Meindl,et al., 2011)

DTI, anatomicalMRI

In separate analyses on 21 AD patients, 16MCI individuals and 20 controls, only inthe AD group atrophy of cholinergic basalforebrain nuclei was associated withreduced FA in intracortical projectingfiber tracts.

Potential approach to explore cholinergicfiber tracts in the human brain in vivo.

(Teipel, Bokde, et al.,2010)

DTI, rsfMRI Positive association between functional andstructural connectivity of the defaultmode network in 20 healthy oldercontrols using joint independentcomponent analysis.

Data driven approach without cross-validation, limiting the generalizability ofthe findings.

(Greicius et al., 2009) DTI, rsfMRI Rs-fMRI derived DMN nodes served asseeds for fiber tractography and led tostable fiber tract estimates between thesenodes in 23 healthy young controls.

Widely descriptive approach providing nostatistical inference on associationsbetween resting state functionalconnectivity and fiber tract integritybetween DMN nodes.

(van den Heuvelet al., 2008)

DTI, rsfMRI Positive association between functional andstructural connectivity of the cingulumtract in 45 healthy young controls, usingpartial correlation coefficients betweennodal time series and mean tract-basedFA.

Hypothesis driven approach; nodes ofconnectivity maps were determined fromthe group data, limiting the generalizabilityof the findings.

DTI + (Amyloid-PET/FDG-PET)(Kuczynski et al.,2010)

DTI, FDG-PET In 16 individuals ranging from cognitivelyhealthy to mild dementia regional FAreductions were associated with voxel-wise reductions of cortical metabolismwithout regional preference.

Effects were controlled for MMSE score,reducing the confounding effect of overallcognitive decline on associations. Numberof subjects too low to assess the specificityof these effects.

(Yakushev et al.,2011)

DTI, FDG-PET 21 AD dementia patients exhibitedsignificant associations betweenhippocampus mean diffusivity andmetabolism in hippocampus,parahippocampus and posterior cingulategyrus.

Data support the idea that disconnection fromdownstream areas contributes to posteriorcingulate hypometabolism in AD.

(Bozoki et al., 2012) DTI, FDG-PET Posterior cingulate metabolism wasassociated with cingulate bundle FA incombined group of 23 MCI individualsand 21 AD patients.

Effect only significant within combinedAD-MCI group, raising the question ofoverall disease severity as confound.

(Cross et al., 2013) DTI, FDG-PET Olfactory tract FA was significantlycorrelated with FDG-PET signal inolfactory processing structures andbeyond when combining the data of 12MCI individuals and 23 healthy controls.

Effect only significant within combined MCI-control group, raising the question of globalcognitive status as confound.






(Chetelat et al., 2013) DTI, rsfMRI,FDG-PET

Metabolism decline in healthy older(n = 57) compared to young adults(n = 40) predominated in the left inferiorfrontal junction and was associated withdisruption of fronto-temporo-occipitalfibers, but not with functionalconnectivity.

Insights into the association betweenage-related metabolism and connectivitychanges. Association with connectivityonly assessed for a priori seed region ofhighest metabolic decline.

(Glodzik et al., 2014) DTI, Amyloid-PET,FDG-PET,T2-MRI

In 72 healthy elderly, white matterhyperintensities on T2-MRI were relatedto decreased amyloid load and glucosemetabolism in gray matter regions that arestructurally connected with the whitematter lesion.

Human structural connectome informs the“change in connectivity” score: defined foreach GM region as the percentage of fibertracts connecting to that region that passthrough the white matter lesion.

(Kantarci et al., 2014) DTI, Amyloid-PET,FDG-PET

Fiber tract integrity was assessed in 570cognitively normal subjects and 131 MCIindividuals stratified by amyloid load andneurodegeneration status. FA of thefornix body decreased withneurodegeneration and was significantlyassociated with cognitive performance.

Loss of white matter integrity was moreassociated with gray matterneurodegeneration modulated by amyloidload. Amyloid biomarker positivity withabsence of gray matter neurodegenerationor cognitive impairment showed no effecton whole brain FA.

(Racine et al., 2014) DTI, Amyloid-PET FA was associated with whole brainamyloid load in long association fibersand intracortical projecting fibers in 139cognitively healthy subjects.

No group differences were found for axial andradial diffusivity. MD was associated withamyloid load in a right fronto-lateral clusteronly.

(Iturria-Medina et al.,2014)

DTI, Amyloid-PET A model for amyloid-β spreading wasdeveloped using the information ofstructural connections derived from DTI.Main contributors for amyloid depositionwere the parameters amyloid clearancedeficiency and early amyloid onset age,both strongly associated with the diseaseseverity and APOE e4 genotype.

The model identified the posterior andanterior cingulate cortices as startinglocation for amyloid deposition. Localamyloid deposition patterns of 733 subjects(AD dementia, MCI and controls) could bereproduced explaining approximately 50%of variance in regional amyloid load.Regions less connected were less likely toaggregate amyloid.

(Fischer et al., 2015) DTI, Amyloid-PET Graph-theoretical measures of structuralconnectivity revealed significant groupdifferences between 31 amyloid-negativecognitively healthy subjects vs. 12amyloid-positive cognitively healthysubjects.

No differences were found in hippocampalvolume, whole-brain metabolism, or whitematter FA and MD.

(Kim et al., 2015) DTI, Amyloid-PET,T2-MRI

In 232 elderly patients with cognitiveimpairment, effects of small vesseldisease on cortical atrophy and cognitivedysfunction were mediated by whitematter network disruptions. Amyloidaffected cortical atrophy and cognitiveimpairment without being mediated bywhite matter network integrity.

Amyloid and small vessel disease can giverise to brain atrophy and cognitiveimpairments in regionally and domain-specific patterns. Effect of small vesseldisease, but not of amyloid, is mediated bywhite matter disruption.

(Raj et al., 2015) DTI, FDG-PET,anatomical MRI

Progression patterns of atrophic andhypometabolic changes on longitudinalMRI/PET scans can be accuratelypredicted by considering inter-regionaldiffusion of pathology along the humanwhite matter network.

Multimodal longitudinal imaging data oflarge sample (n = 418), including CN,MCI and AD. White matter network basedon advanced tractography (HARDI) inyoung adults (n = 73). “Network diffusionmodel” is promising prognostic biomarker.

fMRI + (Amyloid-PET/FDG-PET)(Hedden et al., 2009) rsfMRI, Amyloid-

PETAmyloid accumulation was correlated withDMN functional connectivity in clinicallynormal participants, controlled for ageand structural atrophy.

Data suggest DMN network breakdown in thepreclinical phase of AD. A priori DMNregions included posterior cingulate, lateralparietal and medial prefrontal cortices.






(Sperling et al., 2009) Task fMRI,Amyloid-PET

High levels of amyloid deposition wereassociated with aberrant DMN activity(reduced deactivation during successfulmemory encoding) in asymptomatic andminimally impaired older individuals.

High amyloid burden might be responsiblefor dysfunction of DMN supportingmemory function. Controlled for age andperformance but not atrophy.

(Sheline, Raichle,et al., 2010)

rsfMRI, Amyloid-PET

The PIB + cognitively normal group (andAD) had reduced functional connectivityof the precuneus to (para) hippocampus,anterior/dorsal cingulate, gyrus rectus,and superior precuneus compared to PIB-group.

No correlation between amyloid deposition inthe precuneus and the precuneus-regionalfunctional connectivity in the PIB + group.

(Mormino et al.,2011)

rsfMRI, Amyloid-PET

With increasing levels of global PIB uptakein healthy controls, functionalconnectivity decreases were identified inregions implicated in episodic memoryprocessing while connectivity increaseswere detected in dorsal and anteriormedial prefrontal and lateral temporalcortices.

Data suggest heightened vulnerability ofepisodic memory-related brain regions inAD whereas the observed increases infunctional connectivity may reflect acompensatory response.

(Drzezga et al., 2011) rsfMRI, Amyloid-PET, FDG-PET

PIB + MCI and asymptomatic participantshad disruption of whole-brainconnectivity in cortical hubs (posteriorcingulate cortex/precuneus), overlappingwith regional hypometabolism. Amyloidburden showed a negative correlationwith whole-brain connectivity andmetabolism.

Data suggest the possible high susceptibilityof cortical hubs in terms ofhypometabolism and disruption ofconnectivity in early AD and the possiblelink between synaptic dysfunction andfunctional disconnection. Results werecontrolled for structural atrophy.

(Oh & Jagust, 2013) rsfMRI, task fMRI,Amyloid-PET

Amyloid-positive elders showed increasedregional brain activation and decreasedtask-related connectivity during memory-related fMRI compared to amyloid-negative individuals. In the latter,increased task-related connectivityrelated to better memory performance.

Findings highlight the importance of networkconnectivity for compensating for reducedregional activity during successful memoryencoding in aging, while in those with Aβthis network compensation fails and isaccompanied by inefficient regionalhyperactivation.

(Lehmann et al.,2013)

rsfMRI, Amyloid-PET, FDG-PET

Hypometabolism patterns differed acrossAD variants, reflecting involvement ofspecific functional networks, whereasamyloid patterns were diffuse and similaracross variants.

First study to compare in vivo glucosemetabolism and amyloid depositionpatterns across three clinical variants ofAD, as well as their relation to functionalbrain networks.

(Arenaza-Urquijoet al., 2013)

rsfMRI, anatomicalMRI, FDG-PET

In 36 healthy elders, higher levels ofeducation were related to increased graymatter volume, glucose metabolism, andfunctional connectivity of the anteriorcingulate cortex. Increased connectivityalso correlated with better cognitiveperformance.

Structural, metabolic, and connectivitycharacteristics of the anterior cingulatecortex may underlie education-relatedreserve in healthy elders. Connectivityanalysis was limited to anterior cingulateseed, based on volumetric and metabolicfindings.

(Adriaanse et al.,2014)

rsfMRI, Amyloid-PET

No association between mean DMNfunctional connectivity and mean DMNamyloid binding was found across allsubjects or within each group (AD, MCI,healthy elderly). Voxel-wise regressionrevealed that reduced functionalconnectivity in posterior cingulate cortexwas associated with higher average DMNPIB binding across all subjects orPIB- subjects.

Data confirmed the DMN functionalconnectivity difference between PIB + andPIB- subjects. The lack of associationbetween mean DMN functionalconnectivity and amyloid binding might beinfluenced by small sample size andaveraging effect. The PIB- associationpoints to the need of longitudinal studies.






(Koch et al., 2014) rsfMRI, task fMRI,Amyloid-PET

Reduced rsfMRI functional connectivity inthe default mode and an attention networkwas associated with both regionalamyloid plaque load and aberrantfunctional connectivity during attentiontask. It was further linked to impairedattention, suggesting reduced rsfMRIconnectivity to link amyloid pathologyand impaired cognition.

Changes in functional connectivity areregionally consistent across task and reststates, and linked with regional amyloidpathology.

(La Joie et al., 2014) rsfMRI, FDG-PET,anatomical MRI

Atrophy in AD (n = 18) and semanticdementia (n = 13), overlaps in thehippocampus, but corticalhypometabolism in AD reflects moreclosely the functional connectivity patternof posterior compared to anteriorhippocampus seeds, and vice versa for SD.

Disease associated cognitive profiles seem toreflect the disruption of targeted networksmore than atrophy in specific brain regions.Direct correlations between memory scoresand disease-specific network disruptionswithin the patient groups were not assessed.

(Myers et al., 2014) rsfMRI, Amyloid-PET

Patterns of resting-state functionalconnectivity are associated with patternsof amyloid plaque deposition inindividual subjects.

Plaque deposition and functional connectivityof intrinsic networks interact.

(Klupp et al., 2015) rsfMRI, Amyloid-PET, FDG-PET

Increasing hypometabolism in a region notaffected by amyloid plaques is associatedwith increase of plaque load in afunctionally connected area, suggestingan amyloid-facilitated spread ofhypometabolim along functionalnetworks.

Longitudinal study reinforcing the idea ofdistant functional responses to regionalpathology mediated by brain connectivity.Presence of hypometabolic cortical regionsnot affected by plaque load in AD is at oddswith some previous multitracer PETstudies.

(Tahmasian et al.,2015)

rsfMRI, FDG-PET The more functional connectivity betweenprecuneus and hippocampus is impairedin AD patients, the higher hippocampusmetabolism, suggesting disinhibition-likeeffects of hippocampus dysconnectivity.

Integrating cortical dysconnectivity andhippocampus local activity.

(Perrotin et al., 2015) rsfMRI, FDG-PET Anosognosia in AD patients (n = 23)correlated with hypometabolism inorbitofrontal and posterior cingulatecortices, but also with reduced functionalconnectivity between these regions andthe medial temporal lobe.

In addition to local functional deficits withinself-related cortical midline regions, lack ofawareness of memory deficits in AD mayresult from connectivity disruptionsbetween self-related and memory-relatedbrain networks.

EEG + (MRI/DTI/PET)(Teipel et al., 2009) EEG, DTI Reduced fiber tracts (MD) in anterior corpus

callosum, frontal lobe white matter,thalamus, pons, and cerebellum werespecifically associated with decrease ofresting state frontal EEG (alpha) spectralcoherence across 16 MCI individuals.

Peterson’s clinical criteria for MCI diagnosis.Future studies should test if this effect isspecific for MCI due to prodromal AD asrevealed by amyloid biomarkers.

(Babiloni, Pievani,et al., 2009)

EEG, anatomicalMRI

Higher load of white matter lesions alongsuspected cholinergic fiber tracts asdetermined from T2-weighted MRI scanswas associated with reduced power ofoccipital, parietal, temporal, and limbicalpha 1 and theta frequency sources in 57MCI individuals and 28 age-matchedcontrols.

Interesting approach toward estimatingimpairment of cholinergic system integrity.Limited, however, by the limited evidenceon localization of cholinergic fiber tracts inthe cerebral white matter and the unclearspecificity of the suspected locations forcholinergic fibers.

(Moretti, Paternico,Binetti, Zanetti, &Frisoni, 2012)

EEG, anatomicalMRI

Association of a higher ratio of high to lowalpha band power, and of higher ratio oftheta to gamma band power in EEG withlarger volumes of basal ganglia andthalamus in 74 MCI individuals.

Largely exploratory study that generated ahypothesis on the involvement of ventralstriatum and pulvinar in different subtypesof MCI to be tested in subsequent studies.Very liberal level of significance.




conformations (such as misfolded tau and amyloid proteins)along synaptic connections (Ahmed et al., 2014; Braak & DelTredici, 2011; Nath et al., 2012).

Multimodal Imaging of Brain ConnectivityMeasurements: Resting State EEG Rhythms, MRI,FDG-PET, and rs-fMRI Activity

It is well-known that low spatial resolution (centimeters) ofthe electroencephalography (EEG) techniques prevents areliable and precise spatial estimation of the cortical sourcesand of the functional coupling of the EEG rhythms. On theother hand, rs-fMRI has an insufficient temporal resolution(seconds) for the study of the brain rhythms but a veryhigh spatial resolution (millimeters). For this reason, thecombination of the EEG and fMRI techniques has beenperformed in the past years to exploit the informationcontents of both methodologies. In this line, several multi-modal studies have investigated the correlation betweenEEG rhythms in the resting state and low-frequency(approximately 0.1 Hz) fluctuations of the blood oxygena-tion signal (BOLD) in healthy subjects, showing that thesefluctuations are temporally correlated across large-scale

distributed networks. In the resting-state eyes-closedcondition, some studies have reported a positive correlationbetween the alpha power and the BOLD signal time series inthe DMN (Mantini, Perrucci, Del Gratta, Romani, &Corbetta, 2007). Other evidence pointed to negative or mixedcorrelations (Goncalves et al., 2006; Laufs et al., 2003). Incontrast, the alpha power was negatively correlated withactivity in the Dorsal Attention Network (DAN) during theresting state condition (Laufs et al., 2003; Mantini et al.,2007; Sadaghiani et al., 2010). This is a set of control regionsrecruited during goal-driven behavior and perceptualselection (Corbetta & Shulman, 2002). The same negativecorrelation is observed between the alpha power and theventral fronto-parietal cortical network (VAN; Corbetta &Shulman, 2002). Finally, the resting state alpha power alsocorrelated to BOLD activity in a cingulo-insular-thalamicsubnetwork of the VAN, the so-called Salience network(Goncalves et al., 2006; Sadaghiani et al., 2010).Correlation between the resting state EEG power and the

brain BOLD activity was not limited to alpha rhythms. It hasbeen shown that the power of several EEG bands (i.e., delta,theta, alpha, beta, and gamma) correlated to fMRI timecourses within the resting state networks identified by the useof independent component analysis (Mantini et al., 2007).



(Gonzalez-Escamilla,Atienza, Garcia-Solis, & Cantero,2014)

EEG, anatomicalMRI, FDG-PET

Association between the inter-hemisphericcoupling of fronto-occipital regions in thealpha band of resting state EEG and graymatter volume of a thalamic region andbetween glucose consumption of rightparietal lobe and alpha coupling of a rightparietal region in 26 healthy oldercontrols, but not in 29 MCI individuals.

Hypothesis driven study, connectingstructural and metabolic integrity ofpotential EEG source regions with restingstate EEG connectivity. Due to multipletesting these data need independentconfirmation.

(Babiloni et al., 2015) EEG, anatomicalMRI

Atrophy in occipital gray matter wascorrelated to lower resting state EEG(alpha) occipital source activity across 45healthy elderly, 100 MCI subjects and 90patients with mild to moderate ADdementia.

Effect was significant within the combinedAD-MCI-control group. Future studiesshould test if this effect is specific for ADneurodegenerative process or is merelyrelated to global cognitive status

(Vecchio et al., 2015) EEG, DTI Reduced fiber tract integrity (FA) inposterior corpus callosum wassignificantly associated with resting stateEEG source activity path length (decreasein alpha, increase in delta) across 9healthy controls, 10 MCI individuals and21 patients with mild to moderate ADdementia.

Effect was significant within the combinedAD-MCI-control group, raising thequestion of global cognitive status asconfound.

(Mander et al., 2015) EEG, Amyloid-PET In 26 healthy older controls, prefrontalcortical amyloid levels were associatedwith Non REM slow wave activities(0.6 – 1 Hz range) in EEG; thisassociation mediated the associationbetween amyloid levels and memoryconsolidation.

Hypothesis driven study, linking amyloidpathology with impaired hippocampusrelated memory consolidation vianon-REM sleep disruption.




Analogously to the alpha power, the beta power waspositively correlated to the BOLD activity in the DMN andself-referential networks, and was negatively correlatedwith the BOLD activity observed in the DAN (Mantiniet al., 2007).The correlation between the resting state alpha power and

the BOLD signal in the DMN and attentional networksunveils the functional role of brain EEG oscillatory activityfor the functional connectivity and neurotransmission withinlong-range cortical networks, as a possible basis of theregulation of spontaneous cortical arousal in wakefulness(Fox et al., 2005). Keeping in mind these data, we think thatthe study of correlation between the resting state alpha powerand the BOLD signal in the DMN, DAN, and VAN representa new avenue for a better understanding of the clinicalneurophysiology of AD patients and for the definition andvalidation of instrumental markers for diagnostic, prognostic,and therapy monitoring purposes.Another important avenue is the study of the relationship

between resting state EEG rhythms and structural MRImarkers of AD neurodegeneration co-registered in the sameanatomical space. For this purpose, a promising approach isthe estimation of the cortical sources of the resting state EEGrhythms by low-resolution brain electromagnetic tomo-graphy (LORETA) (http://www.uzh.ch/keyinst/loreta.htm).With this goal in mind, LORETA has been repeatedly used tostudy cortical sources of the resting state EEG rhythms inMCI and AD subjects (Figure 3). In these studies, occipitalsources of the resting state alpha rhythms were the mostpromising EEG marker of prodromal AD in MCI subjects.Specifically, the magnitude of occipital sources of alpharhythms was related to MRI markers including white matterlesions and atrophy of the hippocampal and global corticalgray matter in MCI and AD subjects (Babiloni et al., 2013,2015; Babiloni, Ferri, et al., 2009; Babiloni et al., 2006;Babiloni, Frisoni, et al., 2009; Babiloni, Pievani, et al., 2009).Further evidence for a close relation between white matterdegeneration and cortical dysfunction in aging and AD comesalso from studies showing distinct associations betweendecreasing fiber integrity and declining trans-hemisphericcoherence in resting-state EEG rhythms (Teipel et al., 2009).Key studies on EEG combined with other imaging modalitiesare listed in Table 1.

Contribution of Multimodal Imaging ofConnectivity to Diagnostic Accuracy in AD: fMRIand DTI

First studies used the critical role of intrinsic functionalconnectivity in AD’s pathophysiology to apply rs-fMRI andfunctional connectivity in a diagnostic context. In general,due to low signal-to-noise ratio and its non-quantitativenature, fMRI signal (i.e., the BOLD signal) is highlyproblematic for individual reliable diagnostics (Fox &Greicius, 2010). Quantitative BOLD imaging and new dataacquisition techniques producing massively more data in

comparable time to increase the power of data analysis (e.g.,multi-band fMRI) might be helpful for future approaches(Smith et al., 2013). Furthermore, detection of individualintrinsic functional connectivity might be confounded bysystematic center and scanner effects. For example Biswaland colleagues demonstrated significant center effects onongoing BOLD activity and coherence in a huge sample ofmore than 1400 subjects collected across 35 centers (Biswalet al., 2010). The authors identified several factors whichunderlie such center effects and which have to be carefullycontrolled for across subject and center comparisonsin a diagnostic context, including scanner type, sequencespecifications, instructions to participants, and degree ofparticipant’s wakefulness. Finally, since intrinsic functionalconnectivity reflects individual wakefulness and ongoingcognitive activity such asmindwandering (Mason et al., 2007),intra-individual reliability and consistency of rs-fMRI–baseddiagnostic markers is a challenge (Damoiseaux et al., 2006;Patriat et al., 2013). Nevertheless, some studies demonstratedacceptable diagnostic potential of resting-state functionalconnectivity maps. For example, independently from eachother Dyrba and colleagues and Wee and colleaguesdemonstrated that pattern classification of individualfunctional connectivity matrices of whole brain connectivityseparates patients with AD or MCI from healthy controlswith accuracy and specificity rates of approximately 70%(Dyrba, Grothe, Kirste, & Teipel, 2015; Wee et al., 2012).However, both studies also found that combining functionalund structural connectivity (based on DTI data) substantiallyincreases these rates above 90%, suggesting that multimodalconnectivity measures might help in future diagnosticapproaches.The diagnostic use of DTI has been assessed in few studies

so far. In monocenter studies, the separation of AD patientsfrom healthy controls as well as MCI converters from healthycontrols reached 80% to 90% accuracy using FA or MDmaps in multivariate analysis based on principal componentanalysis (Friese et al., 2010), or support vector machineclassification (Grana et al., 2011; O’Dwyer et al., 2012; Shaoet al., 2012).Regional diffusion measures of white matter integrity,

most notably of the fornix, posterior cingulum, and para-hippocampal white matter, have also shown promisingaccuracies between 77% and 95% for the prediction of con-version from MCI to AD dementia over clinical follow-uptimes of 2 to 3 years (Douaud et al., 2013; Mielke et al., 2012;Scola et al., 2010; Selnes et al., 2013). Preliminary findingsfurther suggest that MD may be of higher predictive valuecompared to FA (Douaud et al., 2013), and that diffusionmetrics may be generally better predictors of conversion thanvolumetric measurements on structural MRI (Fellgiebel et al.,2006; Scola et al., 2010), particularly for the prediction offuture cognitive impairments in cognitively normal elderly(Fletcher et al., 2013; Zhuang et al., 2012). However, thesefirst monocentric studies are generally limited by relativelysmall sample sizes, and diagnostic and prognostic findingswithin the highly controlled experimental conditions of these



http://www.uzh.ch/keyinst/loreta.htmhttps:/www.cambridge.org/core/termshttps://doi.org/10.1017/S1355617715000995https:/www.cambridge.org/core

studies, such as uniform DTI acquisition protocols andselected patient populations, will probably not translateseamlessly into the broader clinical context.Only recently, multicenter acquisition and analysis of DTI

data have begun to be explored in the framework of theEuropean DTI Study on Dementia (EDSD) (Teipel, Reuter,et al., 2011). Using a physical and clinical phantom study, theEDSD showed an at least 50% increase of between scannervariability compared to anatomical MRI acquisitions (Teipel,Reuter, et al., 2011). Diagnostic accuracy for the comparisonof AD versus controls yielded only 70% accuracy usingunivariate voxel-based meta-analysis (Teipel et al., 2012) or

posterior cingulate tractography (Fischer et al., 2012), butwas increased to more than 80% accuracy using supportvector machine analysis (Dyrba et al., 2013). The morerelevant discrimination of prodromal MCI individuals fromhealthy controls and biomarker negative MCI individualswas more accurate using FA and MDmeasures from DTI in amachine learning framework than using gray matter andwhite matter volume, but yielded only approximately 70%accuracy (Dyrba, Barkhof, et al., 2015). The clinically mostrelevant question of predicting short to mid-term conversionwithin a group of MCI subjects recruited from multiplecenters is presently being explored in the EDSD framework.

Fig. 3. Overview of EEG spectral analysis. Several EEG techniques use brain electrical activity recorded during spontaneous fluctuationsof vigilance in the resting state eyes closed condition to analyze functional synchronization and functional coupling of cortical neuralactivity in normal elderly subjects and patients with Alzheimer’s disease (AD). On the whole, four main methodological stages can berecognized: (I) EEG recordings, typically from 19 scalp electrodes placed according to 10–20 system. This is the typical electrodemontage used in clinical context. A low spatial sampling of EEG signals is allowed when the spatial frequency of EEG activity isrelatively low as in the condition of resting state eyes-closed condition. (II) Preliminary EEG data analysis is a procedure aimed atselecting artifact-free EEG segments to be used for further analysis. In some cases, artifacts in the EEG segments can be corrected bymathematical procedures (e.g., correction of blinking artifacts). (III) Spectral EEG analysis is a procedure to compute EEG power spectraat scalp electrodes. This procedure aims at evaluating the general quality of EEG segments selected for the final analysis. In the case ofhealthy elderly subjects the EEG power spectra of posterior electrodes is dominated by a main peak of power density around 8–10 Hz.Power density at frequency lower than 4–6 Hz is typically higher in amplitude in the frontal than in the posterior electrodes. (IV) Corticalsources of resting state eyes closed EEG rhythms (free from artifacts) are typically estimated and compared among groups of healthyelderly subjects and patients with mild cognitive impairment and AD. For this purpose, a promising approach is the estimation of EEGcortical sources by low-resolution brain electromagnetic tomography (LORETA) (http://www.uzh.ch/keyinst/loreta.htm). These four basicstages are displayed in the figure.



http://www.uzh.ch/keyinst/loreta.htmhttps:/www.cambridge.org/core/termshttps://doi.org/10.1017/S1355617715000995https:/www.cambridge.org/core

SUMMARY

„Κάλχας Θεστορίδης οἰωνοπόλων ὄχ᾽ ἄριστος, ὃςᾔδη τά τ᾽ ἐόντα τά τ᾽ ἐσσόμενα πρό τ᾽ ἐόντα,”

“Calchas son of Thestor, far the best of bird-diviners,who knew the things that were, and that were to be, andthat had been before,” Iliad, first book, lines 69/70

In the Iliad, the seer Calchas is the one who knows what is(the present), what will be (the future), and what was beforehim (the past). The previous sections have shown that whatwas characteristic for the most well-known seer of old dayscharacterizes the potential utility of neuroimaging markerstoday. Measures of structural and functional connectivityhelp to resolve the three major challenges that are importantfor diagnosing and managing disease today as they wereimportant for the Mycenaean Greeks when they sieged thecity of Troy:

1. Looking into the past: structural and functional con-nectivity changes can explain how an endophenotype ofmolecular pathological changes, such as cortical amyloidand tau accumulation, has built up in an individual brain.The hub characteristics of a brain region and the degreeof their functional connectivity and structural integrationexplain why certain brain networks are more vulnerablethan others to brain diseases such as AD.

2. Understanding the presence: the presence of a certaindisease stage, such as predementia or dementia AD, candiagnostically be detected using measures of structuraland functional connectivity with reasonable accuracy inhighly selected patient samples. In addition, recentevidence from multicenter data suggests that structuralimaging methods may be robust diagnostic markers inthe context of less controlled clinical samples as well.

3. Predicting the future: the most important task in clinicalprognosis is to predict what will come next. This task hastwo aspects. First, for clinical prediction, measures ofstructural and functional connectivity need furtherexploration in large scale multicenter studies. Preliminaryevidence from such studies suggests that measures ofstructural connectivity may be less accurate than classicalmeasures of regional brain atrophy in predicting individualprogression from MCI to AD dementia (Brüggen et al., inrevision). Similar data from fMRI are still widely lacking.Resolving this question will provide guidance if thesemeasures will usefully be used in the selection ofindividuals into clinical trials on prevention strategiesgeared toward a specific molecular pathogenic mechanismof disease or will become relevant for individual prognosisof disease progression in MCI individuals in tertiary andprimary care settings.

Second, for predicting the spread of molecular orfunctional lesions of disease throughout the brain, thehub characteristics of a region could be used to predict thelikelihood of a region to accumulate amyloid or tau pathologyin the further course of the disease. Thus, the spread of a

molecular event, such as amyloid accumulation, can bepredicted with moderate accuracy on a group level, but is stillunresolved for an individual brain based on its networkconnectivity features. If this observation, however, is foundto be robust in larger and more heterogeneous samplesthis would help to bridge the gap between molecular brainchanges and clinical phenotype on the syndromal or even onthe symptomatological level.Several issues remain to be addressed: What will be the

role of functional and structural brain connectivity in the lightof international diagnostic guidelines? Summarizing, thecurrent model of typical (i.e., most frequent) presentation ofAD assumes that brain amyloidosis biomarkers (i.e., abnormaltracer retention on amyloid PET imaging and low Aβ42concentration in the CSF) turn abnormal earliest, and arediagnostic biomarkers of AD when associated to episodicmemory deficits as revealed by neuropsychological tests(Jack et al., 2010). During the evolution of the disease, thispicture would be followed by cortical and hippocampalhypometabolism (FDG-PET), and finally by massiveneuronal loss (i.e., brain atrophy on structural MRI). In thisline, an International Working Group [IWG, (Dubois et al.,2014)] differentiated clinical phenotypes of AD and mixedAD, and proposed to distinguish clinical use of diagnostic(CSF and ligand PET of Aβ42 and tau brain accumulation)from disease tracking (MRI, FDG-PET) biomarkers. Thesystematic collection and analysis of multicenter multimodalimaging data including biomarkers of functional and structuralcortical connectivity are an indispensable requirement for thefuture assessment of the diagnostic, prognostic, monitoring,and therapy response accuracy of these markers, both forclinical trials as well as health care applications, such as aradiological expert system. This also involves the analysis ofthe robustness or vulnerability of the markers to degradingimage quality or varying numbers of available imagingmodalities. Systematic studies need to explore the minimumimage quality and data dimensions that still yield diagnosticallyuseful information for an individual subject.In addition, in a turn of the perspective, the functional and

structural connectivity properties of an individual brain mayhelp to predict what will be the likelihood for an individual ata given level of molecular brain lesions to remain cognitivestable over 2 to 3 years. The positive predictive value of asignificant cortical amyloid accumulation in cognitively intactpersons is approximately 25% for subsequent development ofcognitive symptoms over 3 years (Villemagne et al., 2011);this points to mechanisms of cerebral reserve as an importantfactor for the individual resistance to brain lesions. As theassociations between education and brain structural andfunctional connectivity suggest, the degree of structural andfunctional connectivity may be an important proxy of brainreserve. The assessment of these markers in combinationwith the

measuring cortical connectivity in alzheimer...

Documents