Buckner a:The Brain's Default Network 11 MONKEY HUMAN 240 32* 25 within medial(MPFC)themonk d human.The h elative to the mo monkey. expansion gue is homolog ith grea 32pl inhe human.Area 24 region activated parallel areas implicated as components of the default ysis of the functional correlations between regions. network. FIGURE7 plots maps of the intrinsic correlations as- At the broadest level,an important principle emerges from considering these anatomic details:the It network is no ture vMPFC.andIPL aled as the regions sho ge on ke y "hubs,.”in articular the PCC.that complete ove ecorcCoihthtemcdiicmporaliobecmcmony ss the maps.HF+forms a sub- svstem that is distinct from other major components system.In the next section,we explore evidence for of default network including the dMPFC:both are these subsystems from functional connectivity analysis strongly linked to the core hubs of the default network in humans. but not to each The Default Network Co prises Interacting The default network c sa set ofbrain regions original map of Shulman et al.(1997,FIG.2)and up that are coactivated during sive task states sho in. dates the description of the network to show that it trinsic functional correlation with one another and are comprises at least two interacting subsystems. connected via direct and indirect anatomic projections Normative estimates of the correlation strengths default net are pro mbedding algorithm Onnfunther e arpntia sition weakly correlated regions away from each other. tion of the default network is through detailed anal- This graphical representation illustrates the separation
Buckner et al.: The Brain’s Default Network 11 FIGURE 6. Architectonic areas within medial prefrontal cortex (MPFC) are illustrated for the monkey and human. The human MPFC is greatly expanded relative to the macaque monkey. This expansion is depicted by the triangle and asterisk that plot putative homologous areas between species based on Ong ¨ ur ¨ et al. (2003). Area 32 in the macaque is homologous with area 32pl in the human. Area 24c is expanded and homologous to the caudal part of area 32ac in human. The MPFC region activated within the human default network likely corresponds to frontalpolar cortex and its rostral expansion (areas 10m, 10r, and 10p), anterior cingulate (areas 24 and 32ac), and the rostral portion of prefrontal area 9. Because of differences in functional properties, we sometimes differentiate in this review between dorsal and ventral portions of MPFC (dMPFC and vMPFC). Adapted with permission from Ong ¨ ur ¨ et al. (2003). parallel areas implicated as components of the default network. At the broadest level, an important principle emerges from considering these anatomic details: the default network is not made up of a single monosynaptically connected brain system. Rather, the architecture reveals a series of interconnected subsystems that converge on key “hubs,” in particular the PCC, that are connected with the medial temporal lobe memory system. In the next section, we explore evidence for these subsystems from functional connectivity analysis in humans. The Default Network Comprises Interacting Subsystems The default network comprises a set of brain regions that are coactivated during passive task states, show intrinsic functional correlation with one another, and are connected via direct and indirect anatomic projections as estimated from comparison to monkey anatomy. However, there is also clear evidence that the brain regions within the default network contribute specialized functions that are organized into subsystems that converge on hubs. One way to gain further insight into the organization of the default network is through detailed analysis of the functional correlations between regions. FIGURE 7 plots maps of the intrinsic correlations associated with three separate seed regions within the default network in humans: the hippocampal formation including a portion of parahippocampal cortex (HF+), dMPFC, and vMPFC. The hubs—PCC/Rsp, vMPFC, and IPL—are revealed asthe regionsshowing complete overlap across the maps. HF+ forms a subsystem that is distinct from other major components of default network including the dMPFC: both are strongly linked to the core hubs of the default network but not to each other. We suspect further analyses will reveal more subtle organizational properties. Of note, the map of the default network’s hubs and subsystems shown in FIGURE 7 bears a striking resemblance to the original map of Shulman et al. (1997, FIG. 2) and updates the description of the network to show that it comprises at least two interacting subsystems. Normative estimates of the correlation strengths between regions within the default network are provided in FIGURE 8. The bottom panel of FIGURE 8 is a graph analytic visualization of the correlation strengths using a spring-embedding algorithm to cluster strongly correlated regions near each other and position weakly correlated regions away from each other. This graphical representation illustrates the separation
12 Annals of the New York Academy of Sciences PFC om em lobe.Functional understanding of the default network should seek to explain both the distinct contrbutions LTC spectrum,it has recently been shown that advanced aging is associated with disrupted correlations across large-scale brain networks including the default net- in pres adults An interesting arch will be to understand the developmental course of the default network as well as the functional implications of its late HE+ life disruption. ew照ee IGURE 7.Hubs are map using functional。 ult Net ons the epar of the ant question to ask is whether the pattern can be accounted for by some al- ach map is「 ternative explanation that is not linked to neural archi- tecture.One possibility is that the observed anatomy nale)co at a vas 2007b).Three obse ons le First the com mdp is remark d flow in ing”wh PET othe (see FIG.2).Second,PC /Rsp.IPL,and vMPFC resent of vascular regulation.The methods that have revealed the default network are based on hemodynamic mea sures of blood flow that are indirectly linked to neu- ed on in 23/31 urc.Wise ct al.(2004)recently measured fMRI correlations with the slow fluctua tions in the partial pressure of end-tidal carbon diox ide that accompany breathing Their results convinc ngly demonstrat correlate o revea ing that I While the iated with res it is important to note that the correlat ional strengthe piration do not dosely resemble the default netwprk associated with the medial temporal lobe are gener- the results of Wise and colleagues are a reminder that ally weaker than those observed for the distributed a vascular account should be explored further. One reason to be skeptical of a count is that the default network is also overap the default network. It is presently unclear using measures of resting glucose mctabolism.In a
12 Annals of the New York Academy of Sciences FIGURE 7. Hubs and subsystems within the default network are mapped using functional connectivity analysis. This map was produced by seeding three separate regions (dMPFC, vMPFC, HF+) and plotting the overlap of the functional correlations across the three regions (legend is at bottom; threshold for each map is r = .07). Data are high-resolution rest data (2mm voxels) from 40 participants (mean age = 22 years; 16 male) collected at 3-Tesla using a 12-channel head coil (data from Andrews-Hanna et al. 2007b). Three observations are notable. First, the combined map is remarkably similar to the original estimate of the default network from PET task-induced deactivation (see FIG. 2). Second, PCC/Rsp, IPL, and vMPFC represent anatomic hubs in the default network to which all other regions are correlated. Third, dMPFC and HF+, which are both strongly correlated with the hub regions, are not correlated with each other, indicating that they are part of distinct subsystems. A further interesting feature is that area 7m within the precuneus (indicated by asterisk) is not part of the default network. The black line near the asterisk represents the approximate boundary between areas 7m and 23/31 (estimated boundary based on Vogt & Laureys 2005). of the medial temporal lobe subsystem. The analysis also reveals that the medial temporal subsystem is less strongly associated with the core of the default network that is centered on MPFC and PCC. However, it is important to note that the correlational strengths associated with the medial temporal lobe are generally weaker than those observed for the distributed neocortical regions. As shown in FIGURE 3, the most robust correlations linked to the medial temporal lobe overlap the default network. It is presently unclear how to interpret the quantitatively lower overall levels of correlations associated with the medial temporal lobe. Functional understanding of the default network should seek to explain both the distinct contributions of the interacting subsystems and the role of their close interaction. Of interest, infants do not show the structured interactions between the default network regions, suggesting that the network developsin toddlers or children (Fransson et al. 2007). At the other end of the age spectrum, it has recently been shown that advanced aging is associated with disrupted correlations across large-scale brain networks including the default network (Andrews-Hanna et al. 2007a, Damoiseaux et al. in press). Thus, the correlation strengths presented in FIGURE 8 are only representative of normal young adults. An interesting topic for future research will be to understand the developmental course of the default network as well as the functional implications of its late life disruption. Vascular and Other Alternative Explanations for the Anatomy of the Default Network Given the reproducibility of the specific anatomy of the default network, an important question to ask is whether the pattern can be accounted for by some alternative explanation that is not linked to neural architecture. One possibility is that the observed anatomy reflects a vascular pattern—either draining veins, a global form of “blood stealing” whereby active regions achieve blood flow increases at the expense of nearby regions, or some other poorly understood mechanism of vascular regulation. The methods that have revealed the default network are based on hemodynamic measures of blood flow that are indirectly linked to neural activity (Raichle 1987, Heeger & Ress 2002). This issue is particularly relevant for analyses based on intrinsic correlations because slow fluctuations in vascular properties track breathing as well as oscillations in intracranial pressure. Wise et al. (2004) recently measured fMRI correlations with the slow fluctuations in the partial pressure of end-tidal carbon dioxide that accompany breathing. Their results convincingly demonstrate correlated, spatially specific fMRI responses suggesting that fMRI patterns can reflect vascular responses to breathing (see also Birn et al. 2006). While the spatial patterns associated with respiration do not closely resemble the default network, the results of Wise and colleagues are a reminder that a vascular account should be explored further. One reason to be skeptical of a vascular account is that the default network is also identified using measures of resting glucose metabolism. In a
Buckner et al.:The Brain's Default Network 13 INTRINSIC CORRELATIONS WITHIN THE DEFAULT NETWORK R LT L HF 0.14 R LTC R PHC H IGURE 8.(p)Function regions efa ork.Each ofth to ach othe the d a core set c gions (red)that are m(blue)includes both the hip tion (HF)and al c tex (PHC)This subsysten ate y hubs particularly informative study,Vogt and colleagues cular coupling Vogt et al.first defined regions within (2006)used [F]flourodeoxyglucose (FDG)PET to the PCC(ventral PCC and dorsal PCC)and Rsp in postmortem human tissue samples.Ihey then mea m cach of thes regions across
Buckner et al.: The Brain’s Default Network 13 FIGURE 8. (top) Functional correlation strengths are listed for multiple regions within the default network. Each of the regions is displayed on top with the strengths of the region-to-region correlations indicated below (r-values were computed using procedures identical to Vincent et al. 2006). Regions are plotted on the averaged anatomy of the participant group (MNI/ICBM152 atlas with Z coordinates displayed). (bottom) The regions of the default network are graphically represented with lines depicting correlation strengths. The positioning of nodes is based on a spring-embedding algorithm that positions correlated nodes near each other. The structure of the default network has a core set of regions (red) that are all correlated with each other. LTC is distant because of its weaker correlation with the other structures. The medial temporal lobe subsystem (blue) includes both the hippocampal formation (HF) and parahippocampal cortex (PHC). This subsystem is correlated with key hubs of the default network including PCC/Rsp, vMPFC, and IPL. The dMPFC is negatively correlated with the medial temporal lobe subsystem suggesting functional dissociation. Graph analytic visualization provided by Alexander Cohen and Steven Petersen. particularly informative study, Vogt and colleagues (2006) used [18F]flourodeoxyglucose (FDG) PET to explore anatomy associated with the default network. Critically, FDG-PET measures neuronal activity through glucose metabolism independent of vascular coupling. Vogt et al. first defined regions within the PCC (ventral PCC and dorsal PCC) and Rsp in postmortem human tissue samples. They then measured resting state glucose metabolism in each of these regions across 163 healthy adults and correlated the