The Brain's Default Network Anatomy,Function,and Relevance to Disease RANDY L.BUCKNER,JESSICA R.ANDREWS-HANNA, AND DANIEL L.SCHACTER -Department of Psycholog,Harvard University,Cambridge,Massachusetts,USA Center for Brain Science,Harvard University,Cambridge,Massachusetts,USA Athinoula A.Martinos Center for Biomedical Imaging.Massachusetts General Hospital Boston,Massachusetts,USA tment of Radiology,Harvard Medical School,Boston,Massachusetts,USA "Howard Hughes Medical Institute,Chevy Chase,Maryland 20815,USA Thirty years of brain imaging research has con ed to define the brain's default nety ed brain system i int modes of co ly defined b a pre y activ are ports the ginsight in fun r the futu ing t he I of oth omy The m dial te al lobe su inf tal s the ant me The two sub of the al obs L C igat inte twork for understanding mental disorders including autism,sc phrenia.and alzheimer's rdefauit mode default system;default netor MRI;PET;hippocampus;memoryi zophrenia;Alzheime A common observation in brain imaging research within the default network (Buckner Carroll 2007) is that a specific set of brain regionsreferred to as ations prompt one to ask such question ecngagrdtcndidala mmon?and what is the significanice of t al 2001 Raichle this network to adaptive function?The default net ing this phenomenon further reveals that other kinds of situations,beyond freethinking,engage the default net- non to c work.For example,remembering the past,envisioning be important to understanding discases of the mind (eg,Lustig et al.2003,Greicius et al.2004,Kennedy et a 2006.Bluhm et al.2007). Address for
The Brain’s Default Network Anatomy, Function, and Relevance to Disease RANDY L. BUCKNER, a,b,c,d,e JESSICA R. ANDREWS-HANNA, a,b,c AND DANIEL L. SCHACTERa aDepartment of Psychology, Harvard University, Cambridge, Massachusetts, USA bCenter for Brain Science, Harvard University, Cambridge, Massachusetts, USA c Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Boston, Massachusetts, USA dDepartment of Radiology, Harvard Medical School, Boston, Massachusetts, USA eHoward Hughes Medical Institute, Chevy Chase, Maryland 20815, USA Thirty years of brain imaging research has converged to define the brain’s default network—a novel and only recently appreciated brain system that participates in internal modes of cognition. Here we synthesize past observations to provide strong evidence that the default network is a specific, anatomically defined brain system preferentially active when individuals are not focused on the external environment. Analysis of connectional anatomy in the monkey supports the presence of an interconnected brain system. Providing insight into function, the default network is active when individuals are engaged in internally focused tasks including autobiographical memory retrieval, envisioning the future, and conceiving the perspectives of others. Probing the functional anatomy of the network in detail reveals that it is best understood as multiple interacting subsystems. The medial temporal lobe subsystem provides information from prior experiences in the form of memories and associations that are the building blocks of mental simulation. The medial prefrontal subsystem facilitates the flexible use of this information during the construction of self-relevant mental simulations. These two subsystems converge on important nodes of integration including the posterior cingulate cortex. The implications of these functional and anatomical observations are discussed in relation to possible adaptive roles of the default network for using past experiences to plan for the future, navigate social interactions, and maximize the utility of moments when we are not otherwise engaged by the external world. We conclude by discussing the relevance of the default network for understanding mental disorders including autism, schizophrenia, and Alzheimer’s disease. Key words: default mode; default system; default network; fMRI; PET; hippocampus; memory; schizophrenia; Alzheimer Introduction A common observation in brain imaging research is that a specific set of brain regions—referred to as the default network—is engaged when individuals are left to think to themselves undisturbed (Shulman et al. 1997, Mazoyer et al. 2001, Raichle et al. 2001). Probing this phenomenon furtherreveals that other kinds of situations, beyond freethinking, engage the default network. For example, remembering the past, envisioning Address for correspondence: Dr. Randy Buckner, Harvard University, William James Hall, 33 Kirkland Drive, Cambridge, MA 02148. rbuckner@wjh.harvard.edu future events, and considering the thoughts and perspectives of other people all activate multiple regions within the default network (Buckner & Carroll 2007). These observations prompt one to ask such questions as: What do these tasks and spontaneous cognition share in common? and what is the significance of this network to adaptive function? The default network is also disrupted in autism, schizophrenia, and Alzheimer’s disease, further encouraging one to consider how the functions of the default network might be important to understanding diseases of the mind (e.g., Lustig et al. 2003, Greicius et al. 2004, Kennedy et al. 2006, Bluhm et al. 2007). Motivated by these questions, we provide a comprehensive review and synthesis of findings about the Ann. N.Y. Acad. Sci. 1124: 1–38 (2008). !C 2008 New York Academy of Sciences. doi: 10.1196/annals.1440.011 1
2 Annals of the New York Academy of Science brain's default network.This review covers both ba sic science and clinical observations,with its content organized across five sections.We begin with a brief nalysis of the anatomy rn the role of the default rk in task settings(section IV).While recognizing alterna- tive possibilities,we hypothesize that the fundamental an 1 isto faciltate GuREl.Aheotinegeotfregonalcerebralblbod ents before they ha The final s gevidence that relates the default network to cognitive disorders,including ged over e the possibility that activity in the default network aug gvar'sonian ments a metabol cascade that is conducive to the development of Alzheimer's disease (section V). 7ATg7gh98gmesdcuadinthieiewege I.A Brief History mcta the brain's defaul suggests th e rest state conta when individuals solve externally administered math problems. brain activity in humans during undire ted menta The Swedish brain physiologist David Ingvar was states.Even though no early studies were explicitly de- the first to aggregate imaging findings from rest task signed to explore such unconstrained states,relevant tes and note the importance ot data were nonetheless acquired because of the com- 1974, rest o CBE I nd hi s observed that frontal activity reached high levels during rest states(FIG.1).To explain this une most goal-direeted tasks.In almost all cascs.the lo pected phenomenon,Ingvar proposed that the "h ration of activity during the control states occurred as perfrontal"pattern of activity corresponded"to und an afterthought- -as part of reviews and meta-analyses tancous,con 3101 performed subsequent to the original reports,which a one un fr urbe focused on the goal-directed task 1031 Early Observations his work established that the brain is not idle when lef A clue that brain activity persists during undirected undirected.Rather,brain activity persists in the ab sence of external task direction.Second,Ingvar's ob wa 19 vations suggested that increased activity during res /1055 sed the ket-Schmidt nitr Schmidt 1948)to ask whether cerebral metabolism The Era of Task-Induced Deactivation changes globally when one goes from a quiet rest state Ingvar's ideas about resting brain activity remained largely unexplored for the next decade until positron on tomography (PE 1) ds for ain imag ing gained promi had finer resolution an
2 Annals of the New York Academy of Sciences brain’s default network. This review covers both basic science and clinical observations, with its content organized across five sections. We begin with a brief history of our understanding of the default network (section I). Next, a detailed analysis of the anatomy of the default network is provided including evidence from humans and monkeys (section II). The following sections concern the role of the default network in spontaneous cognition, as commonly occurs in passive task settings(section III), aswell asitsfunctionsin active task settings (section IV). While recognizing alternative possibilities, we hypothesize that the fundamental function of the default network is to facilitate flexible self-relevant mental explorations—simulations— that provide a means to anticipate and evaluate upcoming events before they happen. The final section of the review discusses emerging evidence that relates the default network to cognitive disorders, including the possibility that activity in the default network augments a metabolic cascade that is conducive to the development of Alzheimer’s disease (section V). I. A Brief History The discovery of the brain’s default network was entirely accidental. Evidence for the default network began accumulating when researchers first measured brain activity in humans during undirected mental states. Even though no early studies were explicitly designed to explore such unconstrained states, relevant data were nonetheless acquired because of the common practice of using rest or other types of passive conditions as an experimental control. These studies revealed that activity in specific brain regions increased during passive control states as compared to most goal-directed tasks. In almost all cases, the exploration of activity during the control states occurred as an afterthought—as part of reviews and meta-analyses performed subsequent to the original reports, which focused on the goal-directed tasks. Early Observations A clue that brain activity persists during undirected mentation emerged from early studies of cerebral metabolism. It was already known by the late 19th century that mental activity modulated local blood flow(James 1890). Louis Sokoloff and colleagues(1955) used the Kety-Schmidt nitrous oxide technique (Kety & Schmidt 1948) to ask whether cerebral metabolism changes globally when one goes from a quiet rest state to performing a challenging arithmetic problem—a task that demands focused cognitive effort. To their surprise, metabolism remained constant. While not FIGURE 1. An early image of regional cerebral blood flow (rCBF) at rest made by David Ingvar and colleagues using the nitrous oxide technique. The image shows data averaged over eight individuals to reveal a “hyperfrontal” activity pattern that Ingvar proposed reflected “spontaneous, conscious mentation” (Ingvar 1979). Ingvar’s ideas anticipate many of the themes discussed in this review (see Ingvar 1974, 1979, 1985). their initial conclusion, the unchanged global rate of metabolism suggests that the rest state contains persistent brain activity that is as vigorous as that when individuals solve externally administered math problems. The Swedish brain physiologist David Ingvar was the first to aggregate imaging findings from rest task states and note the importance of consistent, regionally specific activity patterns(Ingvar 1974, 1979, 1985). Using the xenon 133 inhalation technique to measure regional cerebral blood flow (rCBF), Ingvar and his colleagues observed that frontal activity reached high levels during rest states (FIG. 1). To explain this unexpected phenomenon, Ingvar proposed that the “hyperfrontal” pattern of activity corresponded “to undirected, spontaneous, conscious mentation, the ‘brain work,’whichwe carry outwhen left alone undisturbed” (Ingvar 1974). Two lasting insights emerged from Ingvar’s work. First, echoing ideas of Hans Berger (1931), his work established that the brain is not idle when left undirected. Rather, brain activity persists in the absence of external task direction. Second, Ingvar’s observations suggested that increased activity during rest is localized to specific brain regions that prominently include prefrontal cortex. The Era of Task-Induced Deactivation Ingvar’s ideas about resting brain activity remained largely unexplored for the next decade until positron emission tomography (PET) methods for brain imaging gained prominence. PET had finer resolution and
Buckner et al.:The Brain's Default Network uctures tha (Raichle 1987) ed tha cluded ma sk and co ed to iso.By the mid-vdomgin nonmemory control.In addition.to better understand were completed that examined perception,language, the cognitive processes associated with the rest state attention,and memory.Sca ns of rest-state brain ac they informally asked their participants to subjectively tivity were often acquired across these studies for a describe their mental experiences and re ongiated from this work tha what a the tim in fact at nd c onsists of a mixt The term "deactivation"was used because analyses of freely wandering past recollection,future plans,and and image visualization were referenced to the target, other personal thoughts and experiences."Second,the experimental task.Within this nomenclature,region more active in the target condi ton (c.g..rca onsas well a a distin post g d"n ation nia h stcm at isc nsistently activated in humans durin csent and often the most robust efTect in many carly ndirected mental states Broad awareness of the common regions that be interest emerged was activity reductions in unattended come active during passive task states emerged with of its a pair of meta-analyses that pooled extensive data to 1994.Buc (e.g 19 unction 1g0 sis oftask-induced de active as to explicitly determine if there were com on hrain re pared to passive task conditions.There was no initial rions active during undirected (passive)mental states They pooled data from 132 normal adults for which an r et al.19 part was cor ctc.)coul lirectly compar I to a pa pres he sa e state for an autobic phical men ctal.2001) ted data a ory task Andreasen and colleagues (1995)expl mal adults that included both visually and aurally cued the possibility that spontancous cognition makes an active tasks as compared to passive rest conditions important contribution to rest states.Much like other These two analyses revealed a remarkably consis studies at the time,the researchers included a rest con tent set of brain regions that were more actrve during n to their target co con ditio h rimental targct of the stud ns)The sults of the Shulman t al (1997)me ternally directed cognition much like the ntaneous analysis are shown in This image displays cognition that occurs during"rest"states.For this rea the full cortical extent of the brain's default network son.Andreasen and colleagues explored both the rest The broad generality of the rest activity pattern across any diverse studies reinforced the intriguing pos ty that a c this idea.ag the tes.Mo ed by ct2001) tor in the b thought by asking pants to describe their musings following the scanned y rest periods.Paralleling the informal observations by
Buckner et al.: The Brain’s Default Network 3 sensitivity to deep-brain structures than earlier methods and, owing to the development of isotopes with short half-lives (Raichle 1987), typical PET studies included many task and control conditions for comparison. By the mid-1990s several dozen imaging studies were completed that examined perception, language, attention, and memory. Scans of rest-state brain activitya were often acquired across these studies for a control comparison, and researchers began routinely noticing brain regions more active in the passive control conditions than the active target tasks—what at the time was referred to as “deactivation.” The term “deactivation” was used because analyses and image visualization were referenced to the target, experimental task. Within this nomenclature, regions relatively more active in the target condition (e.g., reading, classifying pictures) compared to the control task (e.g., passive fixation, rest) were labeled “activations”; regions less active in the target condition than the control were labeled “deactivations.” Deactivations were present and often the most robust effect in many early PET studies. One form of deactivation for which early interest emerged was activity reductions in unattended sensory modalities because of its theoretical relevance to mechanisms of attention (e.g., Haxby et al. 1994, Kawashima et al. 1994, Buckner et al. 1996). A second form of commonly observed deactivationwas along the frontal and posterior midline during active, as compared to passive, task conditions. There was no initial explanation for these mysterious midline deactivations (e.g., Ghatan et al. 1995, Baker et al. 1996). A particularly informative early study was conducted while exploring brain regions supporting episodic memory. Confronted with the difficult issue of defining a baseline state for an autobiographical memory task, Andreasen and colleagues (1995) explored the possibility that spontaneous cognition makes an important contribution to rest states. Much like other studies at the time, the researchers included a rest condition as a baseline for comparison to their target conditions. However, unlike other contemporary studies, they hypothesized that autobiographical memory (the experimental target of the study) inherently involves internally directed cognition, much like the spontaneous cognition that occurs during “rest” states. For this reason, Andreasen and colleagues explored both the rest aPET and functional MRI (fMRI) both measure neural activity indirectly through local vascular (blood flow) changes that accompany neuronal activity. PET is sensitive to changes in blood flow directly (Raichle 1987). fMRI is sensitive to changes in oxygen concentration in the blood which tracks blood flow (Heeger and Ress 2002). For simplicity, we refer to these methods as measuring brain activity in this review. and memory tasks referenced to a third control condition that involved neither rest nor episodic memory. Their results showed that similar brain regions were engaged during rest and memory as compared to the nonmemory control. In addition, to better understand the cognitive processes associated with the rest state, they informally asked their participants to subjectively describe their mental experiences. Two insights originated from this work that foreshadow much of the present review’s content. First, Andreasen et al. (1995) noted that the resting state “is in fact quite vigorous and consists of a mixture of freely wandering past recollection, future plans, and other personal thoughts and experiences.” Second, the analysis of brain activity during the rest state revealed prefrontal midline regions as well as a distinct posterior pattern that included the posterior cingulate and retrosplenial cortex. As later studies would confirm, these regions are central components of the core brain system that is consistently activated in humans during undirected mental states. Broad awareness of the common regions that become active during passive task states emerged with a pair of meta-analyses that pooled extensive data to reveal the functional anatomy of unconstrained cognition. In the first study, Shulman and colleagues (1997) conducted meta-analysis of task-induced deactivations to explicitly determine if there were common brain regions active during undirected (passive) mental states. They pooled data from 132 normal adults for which an active task (word reading, active stimulus classification, etc.) could be directly compared to a passive task that presented the same visual words or pictures but contained no directed task goals. Using a similar approach, Mazoyer et al. (2001) aggregated data across 63 normal adults that included both visually and aurally cued active tasks as compared to passive rest conditions. These two analyses revealed a remarkably consistent set of brain regions that were more active during passive task conditions than during numerous goaldirected task conditions (spanning both verbal and nonverbal domains and visual and auditory conditions). The results of the Shulman et al. (1997) metaanalysis are shown in FIGURE 2. This image displays the full cortical extent of the brain’s default network. The broad generality of the rest activity pattern across so many diverse studies reinforced the intriguing possibility that a common set of cognitive processes was used spontaneously during the passive-task states. Motivated by this idea, Mazoyer et al. (2001) explored the content of spontaneous thought by asking participants to describe their musings following the scanned rest periods. Paralleling the informal observations by
Annals of the New York Academy of Sciences conditions were simply too unconstrained to be useful as control states.Richard Frackowiak summarized this widely held concern:"To call a'free-wheeling'state, or even a state where you are hxating on a cross and is to my mind quite wrongr and Fletcher 2007.Buckner Vincent 2007.Raichle Snyder 2007).As a result of this uneasiness in inter preting passive task conditions,beyond the few earlier studies mentioned,there was a general trend not to thoroughly report or discuss the meaning of rest state tically ard et al.2001) Their papers directly considered the empirical and theoretical implications of defining baseline states and what the specific pattern of activity in the default net- work might represent.Several lasting consequences on dE. FIGURE 2.The brain's default network was originally default network from other forr cluding attenuation of activity in unattended sensory The areas).Second,they compiled a considerable array of egions lin ork and what thei alzed in Buckner e al)mges show the st a n.A key ir ulation-ave tthe m the defa 20051.Blue regions m sing (Gusnard ct al.2001 task seltings. Gusnard Raichle 2001).Most importantly,the pa- Ingvar and Andreasen et al..they noted that the im. aged rest state is associated with lively mental activity nding its nan me,which,as oflate 2007 that includes "generation and manipulation of men es).u 1.1 tal images,reminiscence of past experiences based on ng plan ntially reported to be studied as a fundamental neurobiological system with physiological and cognitive properties that distin- Emergence of the Default Network as Its Own Research Area work is a brain system much like the ations by Raichle Gu 2001).A dominant theme in the field during the pre- in the liter vious decade concerned how to define an appropriate baseline condition for ncuroimaging studies This focus r(1995). on the ving con-
4 Annals of the New York Academy of Sciences FIGURE 2. The brain’s default network was originally identified in a meta-analysis that mapped brain regions more active in passive as compared to active tasks (often referred to as task-induced deactivation). The displayed positron emission tomography (PET) data include nine studies (132 participants) from Shulman et al. (1997; reanalyzed in Buckner et al. 2005). Images show the medial and lateral surface of the left hemisphere using a population-averaged surface representation to take into account between-subject variability in sulcal anatomy (Van Essen 2005). Blue represents regions most active in passive task settings. Ingvar and Andreasen et al., they noted that the imaged rest state is associated with lively mental activity that includes “generation and manipulation of mental images, reminiscence of past experiences based on episodic memory, and making plans” and further noted that the subjects of their study “preferentially reported autobiographical episodes.” Emergence of the Default Network as Its Own Research Area The definitive recent event in the explication of the default network came with the a series of publications by Raichle, Gusnard, and colleagues (Raichle et al. 2001, Gusnard & Raichle 2001, Gusnard et al. 2001). A dominant theme in the field during the previous decade concerned how to define an appropriate baseline condition for neuroimaging studies.Thisfocus on the baseline state was central to the evolving concept of a default network. Many argued that passive conditions were simply too unconstrained to be useful as control states. Richard Frackowiak summarized this widely held concern: “To call a ‘free-wheeling’ state, or even a state where you are fixating on a cross and dreaming about anything you like, a ‘control’ state, is to my mind quite wrong” (Frackowiak 1991). (For recent discussion of this ongoing debate see Morcom and Fletcher 2007, Buckner & Vincent 2007, Raichle & Snyder 2007). As a result of this uneasiness in interpreting passive task conditions, beyond the few earlier studies mentioned, there was a general trend not to thoroughly report or discuss the meaning of rest state activity. Raichle,Gusnard, and colleaguesreversed thistrend dramatically with three papers in 2001 (Raichle et al. 2001, Gusnard & Raichle 2001, Gusnard et al. 2001). Their papers directly considered the empirical and theoretical implications of defining baseline states and what the specific pattern of activity in the default network might represent. Several lasting consequences on the study of the default network emerged. First, they distinguished between various forms of task-induced deactivation and separated deactivations defining the default network from other forms of deactivation (including attenuation of activity in unattended sensory areas). Second, they compiled a considerable array of findings that drew attention to the specific anatomic regions linked to the default network and what their presence might suggest about its function. A key insightwasthat the medial prefrontalregions consistently identified as part of the default network are associated with self-referential processing (Gusnard et al. 2001, Gusnard & Raichle 2001). Most importantly, the papers brought to the forefront the exploration of the default network as its own area of study (including providing its name, which, as of late 2007, has appeared as a keyword in 237 articles). Our use of the label “default network” in this review stems directly from their labeling the baseline rest condition as the “default mode.”b Their reviews made clear that the default network is to be studied as a fundamental neurobiological system with physiological and cognitive properties that distinguish it from other systems. The default network is a brain system much like the motor system or the visual system. It contains a set of interacting brain areas that are tightly functionally bReferences to the default mode appear in the literature on cognition prior to the introduction of the concept as an explanation for neural and metabolic phenomena. Giambra (1995), for example, noted that “Taskunrelated images and thoughts may represent the normal default mode of operation of the self-aware.” Thus, the concept of a default mode is converged upon from both cognitive and neurobiological perspectives
Buckner eta:The Brain's Default Network 5 TABLE 1.Core regions associated with the brain's default network REGION ABREV INCLUDED BRAIN AREAS n( connected and distinct from other systems within the comprises multiple interacting hubs and subsystems brain.In the remainder of this review,we dehne the These anatomic observations provide the foundation default network in more detail,speculate on its func- on which the upcoming sections explore the functions of the default network suggest th ork has important Blocked Task-Induced Deactivation Because PET imaging requires about a minute of ll.Anatomy of the Default Network was The anatomy of the brain's default r characterized using multiple a paches.The defaul network was originally identified by its consistent ac ivity was averaged over blocks of multiple sequential tivity increases during passive task states as compared ask trials ence the label "blocked."Shulman et al. to a wide range of active tasks (e.g,Shulman et al. (1997)and Mazoyer et al.(2001)published two semi- more recent ap- poac中tha n regions co work (Greicius et al.2003,2004).More broadly the Shulman et al 1907)and auditory and visual modal. default network is hypothesized to represent a brain ities(Mazoyer et al.2001).In total,data from 195 system (or closely interacting subsystems)involving subjects were aggregated across 18 studies in the two interacting brain areas meta-analyse should be critically informed by original data ofShulman et al omy fre of n data m)are highly similar.FIGURE3 shov aches a third meta-analysis of blocked task data from a se to defining the default network and consider the spe ries of 4 MRI data sets from 92 young-adult subjects cific anatomy that arises from these approaches in the (Shannon 2006).In this meta-analysis of fMRI data the passive tasks were all visual fixation and the active the monk y.We highlg ht two asks involved approa ork th 004Ac all th is largely consistent with available infor ation abou istent set o广rgi ns in reases activity during connectional anatomy (TABLE 1).Second,the intrin- sive tasks when individuals are left undirected to think sic architecture of the default network suggests that it to themselves
Buckner et al.: The Brain’s Default Network 5 TABLE 1. Core regions associated with the brain’s default network REGION ABREV INCLUDED BRAIN AREAS Ventral medial prefrontal cortex vMPFC 24, 10 m/10 r/10 p, 32ac Posterior cingulate/retosplenial cortex PCC/Rsp 29/30, 23/31 Inferior parietal lobule IPL 39, 40 Lateral temporal cortex† LTC 21 Dorsal medial prefrontal cortex dMPFC 24, 32ac, 10p, 9 Hippocampal formation†† HF+ Hippocampus proper,EC, PH Notes: Region, abbreviation, and approximate area labels for the core regions associated with the default network in humans. Labels correspond to those originally used by Brodmann for humans with updates by Petrides and Pandya (1994), Vogt et al. (1995), Morris et al. (2000), and Ong ¨ ur¨ et al. (2003). Labels should be considered approximate because of the uncertain boundaries of the areas and the activation patterns. †LTC is particularly poorly characterized in humans and is therefore the most tentative estimate. ††HF+ includes entorhinal cortex (EC) and surrounding cortex (e.g., parahippocampal cortex; PH). connected and distinct from other systems within the brain. In the remainder of this review, we define the default network in more detail, speculate on its function both during passive and active cognitive states, and evaluate accumulating data that suggest that understanding the default network has important clinical implications for brain disease. II. Anatomy of the Default Network The anatomy of the brain’s default network has been characterized using multiple approaches. The default network was originally identified by its consistent activity increases during passive task states as compared to a wide range of active tasks (e.g., Shulman et al. 1997, Mazoyer et al. 2001, FIG. 2). A more recent approach that identifies brain systems via intrinsic activity correlations (e.g., Biswal et al. 1995) has also revealed a similar estimate of the anatomy of the default network (Greicius et al. 2003, 2004). More broadly, the default network is hypothesized to represent a brain system (or closely interacting subsystems) involving anatomically connected and interacting brain areas. Thus, its architecture should be critically informed by studies of connectional anatomy from nonhuman primates and other relevant sources of neurobiological data. In this section, we review the multiple approaches to defining the default network and consider the specific anatomy that arises from these approaches in the context of architectonic and connectional anatomy in the monkey. We highlight two observations. First, all neuroimaging approaches converge on a similar estimate of the anatomy of the default network that is largely consistent with available information about connectional anatomy (TABLE 1). Second, the intrinsic architecture of the default network suggests that it comprises multiple interacting hubs and subsystems. These anatomic observations provide the foundation on which the upcoming sections explore the functions of the default network. Blocked Task-Induced Deactivation Because PET imaging requires about a minute of data accumulation to construct a stable image, the brain’s default network was initially characterized using blocked task paradigms. Within these paradigms, extended epochs of active and passive tasks were compared to one another. During these epochs brain activity was averaged over blocks of multiple sequential task trials—hence the label “blocked.” Shulman et al. (1997) and Mazoyer et al. (2001) published two seminal meta-analyses based on blocked PET methods to identify brain regions consistently more active during passive tasks as compared to a wide range of active tasks. Tasks spanned verbal and nonverbal domains (Shulman et al. 1997) and auditory and visual modalities (Mazoyer et al. 2001). In total, data from 195 subjects were aggregated across 18 studies in the two meta-analyses. FIGURE 2 displays the original data of Shulman et al. visualized on the cortical surface to illustrate the topography of the default network; the data from Mazoyer et al. (not shown) are highly similar. FIGURE 3 shows a third meta-analysis of blocked task data from a series of 4 fMRI data sets from 92 young-adult subjects (Shannon 2006). In this meta-analysis of fMRI data, the passive tasks were all visual fixation and the active tasks involved making semantic decisions on visually presented words (data from Gold & Buckner 2002, Lustig & Buckner 2004). Across all the variations, a consistent set of regions increases activity during passive tasks when individuals are left undirected to think to themselves
Annals of the New York Academy of Sciences BLOCKED TASK-INDUCED DEACTIVATIONS 214 10 HIPPOCAMPAL FUNCTIONAL CONNECTIVITY COVERGENCE ACROSS APPROACHES ■BLOCK+ER☐ER+HFC☐BLOCK+HFC☐ALL FIGURE 3.The brain's default ne vork is po (A)oc ke y with the h to the right)(B)The co regi networ 98gaomedo3Copegh8mh9hano2o0 h an
6 Annals of the New York Academy of Sciences FIGURE 3. The brain’s default network is converged upon by multiple, distinct fMRI approaches. (A) Each row of images shows a different fMRI approach for defining the default network: blocked task-induced deactivation (top row), event-related task-induced deactivation (middle row), and functional connectivity with the hippocampal formation (bottom row). Within each approach, the maps represent a meta-analysis of multiple data sets thereby providing a conservative estimate of the default network (see text). Colors reflect the number of data sets showing a significant effect within each image (color scales to the right). (B) The convergence across approaches reveals the core regions within the default network (legend at the bottom). Z labels correspond to the transverse level in the atlas of Talairach and Tournoux (1988). Left is plotted on the left. Adapted from Shannon (2006)
Buckner eta:The Brain's Default Network Event-Related,Task-Induced Deactivation An alternative to defining the anatomy of the de- 0m2005.D2 Vincent et al.2006).Functional con it ana related (MRI make is particularly informative because it provides a means senting task trials at randomly jittered time int to assess locations of interacting brain regions within typically 2 to 10seconds apart.The reason to perform the default network in a manner that is independent such an analysis is the possibility that extended epochs of task-induced deactivation.In their initial studics are required to clicit activity during passive cpochs,as Greicius ct al.measured spontancous activityom th might be the case ifblocked task a core reg the de from sl cd ta a rapi :(c.g. ogether.Their map of the default network.based on FICURE 3 illustrates the results of a metacanalsis of intrinsic functional correlations,is remarkably similar tudies from Shannon(2006)that uses event-related to that originally generated by Shulman ct al.(1997) MRI data to define the default network.In total,data based on PET vations. from 49 subjects were pooled for this analysis. The mpor from anal 0 c and p og Kirchho in the emn yolved scmantic classification (Shannon 2006:n=21). are associated with episodic mem- ory function (Greicius et al.2004).In fact,many of As can be anpreciated visually the default network de the major neocortical regi ons constituting the default fined based on event-related data is higbly similar to network can be revealed by placing a seed region in that previously reported using blocked data.I hus,the the hippocampal formation and mapping thos c cort differential activity in the default network between pas- rstates can ntancous on( cmerge rapidly,on the ed fron tions with the hipp ampal formation in four inde- Functional Connectivity Analysis pendent data sets. A final approach to defining the functional anatomy of the default network is based on the measurement of Convergence across Approache the brain's intrinsic Def ining the Default Networ Is th en the of the de brain systems (Biswal ct al.1995,De Luc a1.2006 ork desc there exists spontancous activity that tracks the func ribed abo dan amd tional and anatomic organization of the brain.The fault network anatomy is displayed on the bottom panel patterns of spontaneous activity are believed to re- of FIGURE3.The convergence reveals that the default network comprises a distributed set of regions that et al onal contributions ncludes ex an sory and ex(PCC/Rsp),and the inferior arictal lobule IPL) orain with fRI and can he used to characterize the intrinsic architecture of large-scale brain systems, an approach often referred to as functional connec Several more specifc observations are apparent tivity MRI(Biswal et al.1995,Haughton Biswal from this analysis of overlap.First,the hippocampal a.1995200 nation (HF)issho wn t H al 2006) pproach the de aL2006 ns have been characterizcd usine func nalysis)but.relative to the robust tional connectivity analysis(see also De Luca et al. 2006). nent using the approach of task-induced deactivations
Buckner et al.: The Brain’s Default Network 7 Event-Related, Task-Induced Deactivation An alternative to defining the anatomy of the default network based on blocked tasks is to perform a similar analysis on individual task events. Rapid eventrelated fMRI makes possible such an analysis by presenting task trials at randomly jittered time intervals, typically 2 to 10 seconds apart. The reason to perform such an analysis is the possibility that extended epochs are required to elicit activity during passive epochs, as might be the case if blocked task-induced deactivations arise from slowly evolving signals or sustained task sets that are not modulated on a rapid time frame (e.g., Dosenbach et al. 2006). FIGURE 3 illustrates the results of a meta-analysis of studies from Shannon (2006) that uses event-related fMRI data to define the default network. In total, data from 49 subjects were pooled for this analysis. The data are based on semantic and phonological classifi- cation tasks from Kirchhoff et al. (2005; n = 28) as well as a second sample of event-related data that also involved semantic classification (Shannon 2006; n = 21). As can be appreciated visually, the default network de- fined based on event-related data is highly similar to that previously reported using blocked data. Thus, the differential activity in the default network between passive and active task states can emerge rapidly, on the order of seconds or less. Functional Connectivity Analysis A final approach to defining the functional anatomy of the default network is based on the measurement of the brain’s intrinsic activity. At all levels of the nervous system from individual neurons (Tsodyks et al. 1999) and cortical columns (Arieli et al. 1995) towholebrain systems (Biswal et al. 1995, De Luca et al. 2006), there exists spontaneous activity that tracks the functional and anatomic organization of the brain. The patterns of spontaneous activity are believed to re- flect direct and indirect anatomic connectivity (Vincent et al. 2007a) although additional contributions may arise from spontaneous cognitive processes (as will be described in a latersection). In humans, low-frequency, spontaneous correlations are detectable across the brain with fMRI and can be used to characterize the intrinsic architecture of large-scale brain systems, an approach often referred to as functional connectivity MRI (Biswal et al. 1995, Haughton & Biswal 1998; see Fox & Raichle 2007 for a recent review). Motor (Biswal et al. 1995), visual (Nir et al. 2006), auditory (Hunter et al. 2006), and attention (Fox et al. 2006) systems have been characterized using functional connectivity analysis (see also De Luca et al. 2006). Greicius and colleagues (2003, 2004) used such an analysisto map the brain’s default network (see also Fox et al. 2005, Fransson 2005, Damoiseaux et al. 2006, Vincent et al. 2006). Functional connectivity analysis is particularly informative because it provides a means to assess locations of interacting brain regions within the default network in a manner that is independent of task-induced deactivation. In their initial studies, Greicius et al. measured spontaneous activity from the posterior cingulate cortex, a core region in the default network, and showed that activity levels in the remaining distributed regions of the system are all correlated together. Their map of the default network, based on intrinsic functional correlations, is remarkably similar to that originally generated by Shulman et al. (1997) based on PET deactivations. An important further observation from analyses of intrinsic activity is that the default network includes the hippocampus and adjacent areas in the medial temporal lobe that are associated with episodic memory function (Greicius et al. 2004). In fact, many of the major neocortical regions constituting the default network can be revealed by placing a seed region in the hippocampal formation and mapping those cortical regions that show spontaneous correlation (Vincent et al. 2006). FIGURE 3 shows a map of the default network as generated from intrinsic functional correlations with the hippocampal formation in four independent data sets. Convergence across Approaches for Defining the Default Network Is there convergence between the three distinct approaches for defining the anatomy of the default network described above? To answer this question, the overlap among the multiple methods for defining default network anatomy is displayed on the bottom panel of FIGURE 3. The convergence reveals that the default network comprises a distributed set of regions that includes association cortex and spares sensory and motor cortex. In particular, medial prefrontal cortex (MPFC), posterior cingulate cortex/retrosplenial cortex (PCC/Rsp), and the inferior parietal lobule (IPL) show nearly complete convergence across the 18 data sets. Several more specific observations are apparent from this analysis of overlap. First, the hippocampal formation (HF) is shown to be involved in the default network regardless of which approach is used (task-induced deactivation or functional connectivity analysis) but, relative to the robust posterior midline and prefrontal regions, the HF is less prominent using the approach of task-induced deactivations
Annals of the New York Academy of Sciences MONKEY DEFAULT NETWORK was placed d.The ddleegionssho ving correla 9 Se nd the morant subcortical connce way an p0 th of more extensivr recruitment during passive tive states,including both in posterior parietal cortex of the activated regions,as defined based on human and in prefrontal c ortex.These details will be showr functional neuroimasing data.extends across multiple to be informative when subsystems within the default brain areas that have distinct architecture and conne ork are Third,lateral HE is lc an robust.Tog e obse able dat n initial analysis of the that it is provisional and incomplete (TABLE 1). Posterior cingulate cortex (PCC)and restrosple- nial cortex(Rsp)have been extensively studied in the Insights from Comparative Anatomy macaque monkey and recently so with focus on di man an my (c.g.Mor 0. 1 nat least three contig default network regions in po ureas:Rsp(areas29/30).PCC(areas 23/31).and pre- tative monkey homologues including PCC/Rsp,IPL cuneus (area 7m).Rsp is just posterior to the corpus and the HF (FIG.4,see also Rilling et al.2007).In callosum and,in humans,extends along the ventral addition,architectonic maps reveal many similarities nthe vicinity et al.2001).In caques, s.Pric s(Mo 00 v shi 2001)Motivated by the the main gyrus,is PCC.The precuneus,a region often cited as connectional anatomy of the default network.while being involved in the default network,comprises the recognizing that there may be fundamental differences C/ we lo and i ludes are a 7m (Cavanna cus on areas tha Rsp and Parvizi et al. thesc threest
8 Annals of the New York Academy of Sciences FIGURE 4. The default network in the monkey defined using functional connectivity analysis. A seed was placed in the posterior midline (indicated by asterisk) and the regions showing correlated activity were mapped. The left image shows the medial surface, the middle image a transverse section through parietal cortex, and the right image a coronal section through the hippocampal formation. Left is plotted on the left. Adapted from Vincent et al. (2007a). Second, multiple default network regions are functionally correlated with the HF, reinforcing the notion that the medial temporal lobe is included in the network. Overlap is not perfect, however, with some indications of more extensive recruitment during passive cognitive states, including both in posterior parietal cortex and in prefrontal cortex. These details will be shown to be informative when subsystems within the default network are discussed. Third, lateral temporal cortex (LTC) extending into the temporal pole is consistently observed across approaches but, like the HF, is less robust. Together these observations tentatively define the core anatomical components of the default network (TABLE 1). Insights from Comparative Anatomy Important insights into the organization of human brain systems have been provided by comparative studies in the monkey. Vincent et al. (2007a) recently used functional connectivity analysis to show that the major default network regions in posterior cortex have putative monkey homologues including PCC/Rsp, IPL, and the HF (FIG. 4, see also Rilling et al. 2007). In addition, architectonic maps reveal many similarities between human and monkey anatomy in the vicinity of the default network (e.g., Petrides & Pandya 1994, Morris et al. 2000, Ong ¨ ur¨ & Price 2000, Vogt et al. 2001). Motivated by these recent observations, we provide here a detailed analysis of the architectonics and connectional anatomy of the default network, while recognizing that there may be fundamental differences in humans. As a means to simplify our analysis, we focus on areas that fall within PCC/Rsp and MPFC and their anatomic relationships with other cortical regions and the HF. Potentially important subcortical connections, such as to the striatal reward pathway and the amygdala, are not covered. Even with this simplification, the details of the anatomy are complex and one is immediately confronted with the observation that each of the activated regions, as defined based on human functional neuroimaging data, extends across multiple brain areas that have distinct architecture and connectivity. Progress will require significantly more detailed analysis of the anatomic extent and locations of default network regions in humans. Nonetheless, using available data we provide an initial analysis of the anatomy recognizing that it is provisional and incomplete. Posterior cingulate cortex (PCC) and restrosplenial cortex (Rsp) have been extensively studied in the macaque monkey and recently so with focus on direct comparison to human anatomy (e.g., Morris et al. 2000, Vogt et al. 2001). The PCC and Rsp fall along the posterior midline and exist within a region that contains at least three contiguous, but distinct, sets of areas: Rsp (areas 29/30), PCC (areas 23/31), and precuneus (area 7m). Rsp is just posterior to the corpus callosum and, in humans, extends along the ventral bank of the cingulate gyrus (Morris et al. 2000, Vogt et al. 2001). In macaques, Rsp is much smaller and does not encroach onto the cingulate gyrus (Morris et al. 1999, Kobayashi & Amaral 2000). Just posterior to Rsp, along the main portion of the cingulate gyrus, is PCC. The precuneus, a region often cited as being involved in the default network, comprises the posterior and dorsal portion of the medial parietal lobe and includes area 7m (Cavanna & Trimble 2006, Parvizi et al. 2006). As an ensemble, these three structures are sometimes referred to as “posteriomedial
Buckner et al.:The Brain's Default Network 9 not a componen The predominant extrinsic conn ections to and from the posteriomedial cortex difter by area collectively includes PCC area 29/30.Precuncus area 7m pre the connections are widespread and,much like other dominantly connects with occipital and parietal areas association areas,are con stent with a role in infor- mation integration.Specifically,Rsp is heavily inter- oacR and par Leichnetz 2001).Moreover,medial tem poral lobe 12003 show na m.C Cand Rsp ashi&An Amrl 1004 Morris et al 1900) ca 7m and the PCC.which ma omoiects back to the medial temporal lobe as well as be the basis for the extensive activation patterns som prominently to multiple prefrontal regions(Kobayashi times observed along the posterior midline,but we Amaral 2007,FIG.5).PCC area 23 has reciprocal suspect that area 7m is not a core component of the e and robust forcing thisi se exam ion of 1 he the hur uman IPL (Koba maps tha Amaral 2003.2007.FIG.5).The medial te ent of the network usually does not encroach on the also has modest but consistent connections with arca dge of the parietal midline(where arca 7m is located 7a(Suzuki Amaral 1994,Cloweretal.2001,Lavenex Scheperjans et al.2007).This boundary is labeled ex et al.2002).Thus,PCC/Rsp provides a key hub for plicitly in FIGURE7 by an asterisk.The middlle panel tween the c mc of FiGURE 18 shows a particularly clear obe,an n the An unresoledissue is whether the lateral ectiot netuork and a zone of PCC/Rsp is restricted to area 7a in humans (2004;their Figure 2A versus 2B).For all these reasons, or extends to areas 39/40.Macaque PCC has recipro we provisionally conclude that area 7m in precuneus cal projections to superior temporal sulcus(STS)and is not part of the default network The second of th MPF areas t 1004 STG Vincent etal.2007a).Com 200 Huma plicating the picture.IPL is g atly expanded in hu ey (Onguir et al.2003,FIG.6).Two differences are no mans,including areas 39/40(Culham Kanwisher table.First.macaque area 32 is pushed ventrally and 2001,Simon et al.2002,Orban et al.2006)that are rostrally in humans to below the corpus callosum (a- closely localized to the lateral parietal regior beled ct al.as area 3pln the human based ing on Br ion be the brain based on ma ows the rostral path of anterior cingulate areas 24 crease(Van Essen Dieker 2007).Thus,these lateral and 32ac much like typical activation of MPFC in the default network.This is relevant because commonly referenced maps based on classic architectonic analy and pote n in h natel The conn ctional anat r of area 7m in the pre to cialization default network even though it is often included in 2001)
Buckner et al.: The Brain’s Default Network 9 cortex,” and each structure is interconnected with the others (e.g., Parvizi et al. 2006, Kobayashi & Amaral 2003). The predominant extrinsic connections to and from the posteriomedial cortex differ by area. Collectively, the connections are widespread and, much like other association areas, are consistent with a role in information integration. Specifically, Rsp is heavily interconnected with the HF and parahippocampal cortex, receiving nearly 40% of its extrinsic input from the medial temporal lobe (Kobayashi & Amaral 2003, see also Suzuki & Amaral 1994, Morris et al. 1999). Rsp also projects back to the medial temporal lobe as well as prominently to multiple prefrontal regions (Kobayashi & Amaral 2007, FIG. 5). PCC area 23 has reciprocal connections with the medial temporal lobe and robust connections with prefrontal cortex and parietal cortex area 7a—an area at or near the putative homologue of the human default network region IPL (Kobayashi & Amaral 2003, 2007, FIG. 5). The medial temporal lobe also has modest, but consistent, connections with area 7a (Suzuki&Amaral 1994, Clower et al. 2001, Lavenex et al. 2002). Thus, PCC/Rsp provides a key hub for overlapping connections between themselves, the medial temporal lobe, and IPL—three of the distributed regions that constitute the major posterior extent of the default network. An unresolved issue is whether the lateral projection zone of PCC/Rsp is restricted to area 7a in humans or extends to areas 39/40. Macaque PCC has reciprocal projections to superior temporal sulcus (STS) and the superior temporal gyrus (STG; see also Kobayashi & Amaral 2003). Analysis of the default network in macaques provides indication that the network’s lateral extent includes STG (Vincent et al. 2007a). Complicating the picture, IPL is greatly expanded in humans, including areas 39/40 (Culham & Kanwisher 2001, Simon et al. 2002, Orban et al. 2006) that are closely localized to the lateral parietal region identified by human neuroimaging as being within the default network (see Caspers et al. 2006). A recent analysis of cortical expansion between the macaque and human brain based on mapping of 23 presumed homologies revealed that IPL is among the regions of greatest increase (Van Essen & Dieker 2007). Thus, these lateral parietal and temporo-parietal areas, which are not as well characterized as PCC/Rsp, are extremely interesting in light of their anatomic connections, involvement in the default network, and potential evolutionary expansion in humans. The connectional anatomy of area 7m in the precuneus is difficult to understand in relation to the default network even though it is often included in the default network. One possibility is that area 7m is simply not a component of the default network. References to precuneus in the neuroimaging literature are often used loosely to label the general region that includes PCC area 29/30. Precuneus area 7m predominantly connects with occipital and parietal areas linked to visual processing and frontal areas associated with motor planning (Cavada& Goldman-Rakic 1989, Leichnetz 2001). Moreover, medial temporal lobe regions that have extensive projections to PCC and Rsp show minimal connections to area 7m. Connections do exist between area 7m and the PCC, which may be the basis for the extensive activation patterns sometimes observed along the posterior midline, but we suspect that area 7m is not a core component of the network. Reinforcing this impression, close examination of the many maps that define the human default network in this review shows that the posterior medial extent of the network usually does not encroach on the edge of the parietal midline (where area 7m is located, Scheperjans et al. 2007). This boundary is labeled explicitly in FIGURE 7 by an asterisk. The middle panel of FIGURE 18 shows a particularly clear example of the separation between task-induced deactivation of PCC and its dissociation from the region at or near area 7m. Another example of dissociation between the default network and area 7m can be found in Vogeley et al. (2004; their Figure 2A versus 2B). For all these reasons, we provisionally conclude that area 7m in precuneus is not part of the default network. The second hub of the default network, MPFC, encompasses a set of areas that lie along the frontal midline (Petrides & Pandya 1994, Ong ¨ ur¨ & Price 2000). Human MPFC is greatly expanded relative to the monkey (Ong ¨ ur¨ et al. 2003, FIG. 6). Two differences are notable. First, macaque area 32 is pushed ventrally and rostrally in humans to below the corpus callosum (labeled by Ong ¨ ur¨ et al. as area 32pl in the human based on Brodmann’s original labeling of this area in monkey as the “prelimbic area”). Human area 32ac corresponds to Brodmann’s dorsal “anterior cingulate” area. Second, human area 10 is quite large and follows the rostral path of anterior cingulate areas 24 and 32ac much like typical activation of MPFC in the default network. This is relevant because commonly referenced maps based on classic architectonic analyses restrict this area to frontalpolar cortex (e.g., Petrides & Pandya 1994). Some evidence suggests that area 10 is disproportionately expanded in humans even when contrasted to great apes, suggesting specialization during recent hominid evolution (Semendeferi et al. 2001)
10 Annals of the New York Academy of Sciences RETROSPLENIAL CORTEX POSTERIOR CINGULATE CORTEX 24 PARAHIPPOCAMPAL CORTEX FIGURE 5.M ns t dis nts the co e.henecedwi obule (PL)exten Given these details,MPFC activation within the monkey-the medial prefrontal networkshow re default network is estimated to encompass human ciproc 100032ac to the areas in
10 Annals of the New York Academy of Sciences FIGURE 5. Monkey anatomy suggests that the default network includes multiple, distinct association areas, each of which is connected to other areas within the network. Illustrated are two examples of output (efferent) and input (afferent) connections for posterior cingulate/retrosplenial cortex (PCC/Rsp) and parahippocampal cortex (PH). (A) Output connections from Rsp (areas 29 and 30) and PCC (area 23) are displayed. Lines show connections to distributed areas; thickness represents the connection strength. Rsp and PCC are heavily connected with the medial temporal lobe (HF, hippocampal formation; PH, parahippocampal cortex), the inferior parietal lobule (IPL) extending into superior temporal gyrus (STG), and prefrontal cortex (PFC). Numbers in the diagram indicate brain areas. Adapted from Kobayashi and Amaral (2007). (B) Input and output connections to and from PH to medial prefrontal cortex (MPFC) are displayed. Adapted from Kondo et al. (2005). Given these details, MPFC activation within the default network is estimated to encompass human areas 10 (10 m, 10 r, and 10 p), anterior cingulate (area 24/32ac), and area 9 in prefrontal cortex. The closest homologues to these areas in the monkey—the medial prefrontal network—show reciprocal connections with the PCC, Rsp, STG, HF, and the perirhinal/parahippocampal cortex; sensory inputs are nearly absent (Barbas et al. 1999, Price 2007). These connectivity patterns closely
