2.1. Amygdala

Chapter 2 discusses the ever-important amygdala and its role in brain function. Based on rodent and human data, I describe how the amygdala's functions go beyond emotion as traditionally conceived, reflecting a trend toward viewing this structure not simply in terms of “fear.”

A key function of the amygdala is to shape selective information processing. Selection of information for further analysis is, of course, a central problem that needs to be solved for effective behavior (Grossberg & Levine Reference Grossberg and Levine1987). The amygdala is a core structure in a system involved in “What is it?” processing and thus contributes to highlighting what is of significance to the organism (Pribram & McGuinness Reference Pribram and McGuinness1975). However, the functions of the amygdala also involve “What's to be done?” A key reason for this is that the amygdala participates in the representation of value (including positive value) and in decision making. For example, amygdala lesions impair behavior on the Iowa Gambling Task in humans and alter delay-based decision making in rats (e.g., they become more impulsive). The amygdala thus takes part in an impressive array of processes that far exceed some of its proposed functions, such as vigilance, arousal, salience detection, novelty detection, and relevance detection. “Information gathering” (Whalen Reference Whalen1998) better captures several of its functions but comes short, too. In the end, it is better simply to refrain from overly summarizing its functional repertoire so as to better appreciate the wide scope of the amygdala's contributions to brain mechanisms and behavior.

2.2. Subcortical “low road” pathway and emotional processing

A purported division of labor between cortical and subcortical regions has been present from the time of the earliest circuit models of emotion (e.g., Papez Reference Papez1937). Many versions of this type of dual processing model exist, including some variants that have captured the popular imagination, such as the “triune brain” (MacLean Reference MacLean and Schmitt1970; Reference MacLean1990).

In the case of vision, it has been suggested that a subcortical pathway from the retina to superior colliculus to pulvinar to amygdala that entirely bypasses cortex enables the processing of emotion-laden visual stimuli to be fast, automatic, and nonconscious. In chapter 3, I argue against this notion on several general grounds: (1) Affective visual information is not handled qualitatively faster than other visual information; (2) processing of affective visual stimuli involves both low- and high-spatial frequency information; and (3) the amygdala is not required for rapid, nonconscious detection of affective information. For these and many other reasons, Ralph Adolphs and I proposed the “multiple waves” model as an alternative to the low-road pathway scheme (Pessoa & Adolphs Reference Pessoa and Adolphs2010). The model shifts the debate away from whether there is a unique subcortical pathway to whether a processing architecture exists that is capable of rapidly transmitting information via multiple pathways. The resulting multiple waves model emphasizes the role of the pulvinar in coordinating and regulating the flow of multimodal information, which is accomplished via a series of thalamo-cortical loops. In this role, the pulvinar moves from being a passive relay station of the “standard hypothesis” to being an active element of information processing.

2.3. What kind of unawareness matters?

The research literature is replete with paradigms such as backward masking and the attentional blink that challenge the visual system so that awareness can be studied. At times, much is made about neuroimaging responses observed in the amygdala for very brief stimuli (e.g., 15–30 ms). In such cases, subjects may report not seeing them (“subjective unawareness”). In the book, I argue that this type of “subliminal” unawareness is not the most relevant one to understand the impact of affective content on behavior and on clinical conditions such as anxiety. A more important sense is associated with the idea of unintentional processing, which may prove to be more important to the understanding of human behavior. Whether the unintentional unconscious is sophisticated and flexible, as argued by social psychologists (see Bargh & Morsella Reference Bargh and Morsella2008), is a matter of debate. But there can be no doubt that it is qualitatively different from the type of subliminal unconscious sometimes emphasized in the emotion literature (for evidence that the “subliminal unconscious” may be quite “dumb,” see Loftus & Klinger Reference Loftus and Klinger1992 – if at all present; Pessoa Reference Pessoa2005).

2.4. Why is the amygdala important?

In the broader neuroscience literature, the amygdala is viewed as a central node in emotional processing in part because of the “low-level” properties ascribed to the subcortical pathway. Defects in the amygdala system are said to underlie phobias, mood disorders, and post-traumatic stress syndrome, and variability in its functioning to reflect individual differences at the genotypic and personality level.

Although in chapter 3 I challenge many of the properties typically ascribed to the subcortical pathway, the amygdala is indeed important for behavior and mental health. First, together with the hypothalamus and medial PFC, the amygdala has extensive projections to downstream regions in the brainstem that are capable of mobilizing the body; indeed, its central nucleus is at times described as a “controller of the brainstem.” The autonomic and neuroendocrine connections of these brain regions are part of sympathetic and parasympathetic networks that coordinate bodily responses in the face of challenges to the organism. Second, the amygdala, hypothalamus, medial PFC, and related regions, being among the most extensively connected parts of the brain, are optimally positioned to influence information processing. As hubs through which evaluative signals are communicated, they are thought to have widespread effects on mental function and to play a significant role in affective and cognitive impairments observed in mood disorders. Metaphorically speaking, as one of these hubs, the amygdala is strategically positioned to “ignite” both body and brain.

2.5. Processing of emotion-laden information and automaticity

Shiffrin and Schneider (Reference Shiffrin and Schneider1977, pp. 155–156) defined an “automatic process … as a sequence of nodes that nearly always becomes active in response to a particular input configuration.” Because automatic and controlled processes appear to be qualitatively opposed, it is natural to dichotomize mental phenomena into these two classes. But such a dichotomy has simply not held up in the face of data. Reports of automaticity have invariably been countered by reports of capacity limitation; behavioral effects assumed to operate automatically are influenced in ways that belie that assumption.

The argument that I make in chapter 3 is that a better framework is one where performance is always considered capacity limited and described as a performance-resource function (Norman & Bobrow Reference Norman and Bobrow1975). Some behaviors will exhibit shallower performance-resource relationships, where performance only rises slowly based on the mental effort exerted – these behaviors are hence “controlled.” Other behaviors exhibit steeper relationships, and ceiling performance is reached even when conditions are degraded (e.g., under short exposure) – these behaviors are hence “efficient.” Although the performance-resource function may seem to be an abstract construct when little is known about the task at hand, it forces researchers to consider a spectrum of scenarios when studying how a behavior depends on multiple factors that influence performance.

Why is a continuous framework better than a dichotomous one? For one thing, it fits the empirical data better: Researchers have repeatedly found capacity limitations for “automatic” phenomena (e.g., Pashler Reference Pashler1998). For another, the dichotomous framework is plagued by serious conceptual issues (Moors & De Houwer Reference Moors and De Houwer2006). Another reason a continuous framework is better is that we still have an incipient understanding of competition – and, hence, of whether interference will result when multiple items are involved. The notion of competition, as accepted by most researchers, goes roughly as follows. Because processing capacity is limited, competition is proposed to “select” the most relevant information at any given time (Desimone & Duncan Reference Desimone and Duncan1995; Grossberg Reference Grossberg1980); when resources are not fully consumed, spare capacity is used to process task-irrelevant items (Lavie Reference Lavie1995). The problem is that we do not always know whether interference will occur in any given situation. Generally, multiple factors determine how information competes in visual cortex and beyond, including task difficulty, set size, spatial arrangement, cuing, and the like. Finally, a continuous framework demystifies the processing of certain complex features. For example, processes such as reading and the perception of elaborate emotional images are at times depicted as “automatic” in a sense that is almost magical (for a cogent in-depth discussion, see Pourtois et al. 2012). Indeed, the underlying mechanisms of abilities such as proficient reading and the perception of emotional scenes are remarkably fast. That we do not understand why they are so fast, however, simply means that we are still quite some way from a better mechanistic description of these processes.

2.6. Dual process models

The discussion of automatic versus controlled processes is also pertinent to dual process models. Common to these models is the strong assumption of the existence of two qualitatively different mental systems, for example, “intuition” and “reasoning” (see Keren and Schul Reference Keren and Schul2009). A popular trend is to call the two components “system 1” and “system 2,” where the first is automatic/heuristic/reflexive and the second is controlled/analytic/reflective (Evans Reference Evans2008). But as others have expressed in the past, the idea of a dual system model is both slippery and conceptually unclear (see Keren & Schul Reference Keren and Schul2009). For one, nearly all dual process models have as a central component the automatic versus controlled dichotomy, which as discussed above is not a viable distinction. In fact, as with the question of automatic versus controlled processing of emotion-laden stimuli, the question of whether there are two systems in dual process models is not an entirely empirical one. This is because no single critical experiment can provide a final, definitive answer. In the end, however irresistible dichotomies are to the human mind (Kelso & Engstrøm Reference Kelso and Engstrøm2006; Newell Reference Newell and Chase1973), dichotomizing implies oversimplifying (Keren & Schul Reference Keren and Schul2009; Kruglanski et al. Reference Kruglanski, Erbs, Pierro, Mannetti and Chun2006). A continuous framework is better, albeit more complex (Kruglanski et al. Reference Kruglanski, Erbs, Pierro, Mannetti and Chun2006).

3.1. The “Classical” view: Emotion-cognition push-pull

In an important paper, Drevets and Raichle (Reference Drevets and Raichle1998) noted that regional blood flow during attentionally demanding cognitive tasks decreased in regions such as the amygdala, orbitofrontal cortex, and ventral-medial PFC, whereas blood flow increased in these regions during specific emotion-related tasks. Conversely, blood flow during experimentally induced and pathological emotional states (Mayberg et al. Reference Mayberg, Liotti, Brannan, McGinnis, Mahurin, Jerabek, Silva, Tekell, Martin, Lancaster and Fox1999) decreased in regions such as the dorsal-medial and dorsal-lateral PFC, whereas blood flow increased in these regions during cognitive tasks. These reciprocal patterns of activation suggested to Drevets & Raichle (Reference Drevets and Raichle1998) that emotion and cognition engage in competitive interactions.

This insight led to a wealth of studies pursuing the notion of a dorsal-cognition versus ventral-emotion axis of organization in the human brain. For example, Dolcos and colleagues investigated emotional distraction during working memory tasks (see also Anticevic & colleagues Reference Anticevic, Repovs and Barch2010). Subjects were shown sample stimuli that had to be remembered during a subsequent delay period during which they saw distracting stimuli, including neutral and emotional pictures. The findings of one of their studies (Dolcos & McCarthy Reference Dolcos and McCarthy2006) are illustrated in Figure 2. During the delay period, responses in dorsal-lateral PFC (Fig. 2B) were highest for the “scrambled” (digitally scrambled versions of pictures), intermediate for neutral, and lowest for emotional distractors – a pattern of responses also observed in parietal cortex. Behavioral performance mirrored this and was worst for emotional distractors. Viewing emotional distractors during the delay period appeared to interfere with neural activity normally observed in these sites – activity that supports working memory performance (e.g., Pessoa et al. Reference Pessoa, Gutierrez, Bandettini and Ungerleider2002). Responses in the ventral-lateral PFC (Fig. 2C) followed the opposite pattern, namely, the strongest responses were observed during the viewing of emotional distractors, suggesting that ventral-lateral PFC contributed to inhibiting the distracting effects of stimuli presented during the delay period. Overall, several studies are consistent with the dorsal-cognition versus ventral-emotion segregation (both along the lateral surface of the brain and its medial sector), including those probing emotional distraction, emotional conflict, and emotion regulation (Ch. 5).

Figure 2.

Emotional distraction during a working memory task. Subjects were shown scrambled, negative, or neutral distractor images during the delay period of the task. (A) Schematic representation of differential responses in brain. Regions where responses were stronger to scrambled than to emotional images are shown in light gray; regions where they were stronger to emotional than to scrambled images, in dark gray. (B) Time course data for dorsal-lateral prefrontal cortex. (C) Time course data for ventral-lateral prefrontal cortex. Horizontal bars in panels B and C correspond to onset and duration of sample stimuli, distractors, and probes, respectively. Time series plots kindly provided by Florin Dolcos, adapted with permission from Dolcos and McCarthy (Reference Singer, Critchley and Preuschoff2006).

The organization of the medial PFC, a complex brain region involved in diverse functions (Vogt Reference Vogt2008), has strongly fueled the dorsal versus ventral view of emotion and cognition organization in the brain, particularly following another influential paper (Bush et al. Reference Bush, Luu and Posner2000; see also Devinsky et al. Reference Devinsky, Morrell and Vogt1995). In the next section, I argue against the dorsal versus ventral framework in the medial PFC in particular, and in the subsequent section against the dorsal versus ventral view in the PFC more generally.

3.2. Beyond the dorsal versus ventral-medial dichotomy in the prefrontal cortex

Results from several individual studies challenge the dichotomy. For example, Mobbs and colleagues (Reference Mobbs, Yu, Rowe, Eich, FeldmanHall and Dalgleish2010) examined how brain responses vary as a function of perceived threat proximity. In an unusual experimental manipulation, each participant inside the MRI scanner placed a foot into a custom-built box containing multiple compartments, while watching a video of a live tarantula placed into one of the compartments at varying distances from the foot (actually prerecorded). Increases in responses as a function of proximity were observed in several brain regions; notably in the dorsal-medial PFC.

The “attentional network” involves fronto-parietal regions, including the dorsal-medial PFC (Corbetta & Shulman Reference Corbetta and Shulman2002; Kastner & Ungerleider Reference Kastner and Ungerleider2000). To assess brain regions that are sensitive to high levels of threat, I reviewed activation sites reported in aversive conditioning studies (Pessoa Reference Pessoa2009). Surprisingly, activation was repeatedly reported not only in the amygdala but also in frontal sites overlapping with those in the attentional network, including the dorsal-medial PFC – consistent with findings from formal meta-analyses (Etkin & Wager Reference Etkin and Wager2007; Mechias et al. Reference Mechias, Etkin and Kalisch2010). To understand the organization of the medial PFC and its role in emotion, Etkin and colleagues (Reference Etkin, Egner and Kalisch2011) reviewed both the human and nonhuman animal literatures. They surmise that sites in both dorsal- and ventral-medial PFC make prominent contributions to emotional processing. Finally, an extensive formal meta-analysis of human neuroimaging studies (Shackman et al. Reference Shackman, Salomons, Slagter, Fox, Winter and Davidson2011) further demonstrates the considerable overlap of sites in the medial PFC engaged during negative affect and cognitive control (Fig. 3).

Figure 3.

Cognition and emotion in medial frontal cortex. Foci of activation across studies of negative affect and cognitive control. Extensive overlap between emotion and cognition was observed in dorsal-medial prefrontal cortex. Figure kindly provided by Alex Shackman and adapted with permission from Shackman et al. (Reference Shackman, Salomons, Slagter, Fox, Winter and Davidson2011).

In summary, although it is still influential, the segregation model of medial PFC organization is no longer viable, as different research groups now argue (e.g., Etkin et al. Reference Etkin, Egner and Kalisch2011; Pessoa Reference Pessoa2009; Shackman et al. Reference Shackman, Salomons, Slagter, Fox, Winter and Davidson2011). Large portions of the PFC are engaged during emotional processing, including both dorsal and ventral portions of the medial PFC. Indeed, when large numbers of studies are considered jointly, the weight of their findings strongly favors an organization of the medial PFC that is not segregated into affective and cognitive compartments but instead is shared by cognitive and affective domains in a way that allows the medial PFC to support the adaptive control of complex behaviors (Pessoa Reference Pessoa2008; Shackman et al. Reference Shackman, Salomons, Slagter, Fox, Winter and Davidson2011).

3.3. Beyond push-pull: When emotion and cognition work together

Now, I will turn to the broader issue of the frequently held view of emotion-cognition organized as push-pull, or antagonistic, systems. Consider once more the study by Dolcos and McCarthy (Reference Dolcos and McCarthy2006) that showed that emotional distractors produced decreased responses in parts of the dorsal-lateral PFC that are important for cognitive tasks. This type of response, which favors the antagonistic organization, is far from universal, however. For example, also during conditions of emotional distraction, Erk et al. (Reference Erk, Kleczar and Walter2007) observed increased responses to emotional stimuli in the dorsal-lateral PFC. They also observed increased responses when they increased the load of a separate nonemotional working memory task. In other words, both the emotional and cognitive manipulations produced enhanced responses in the dorsal-lateral PFC. Conversely, emotional manipulations do not always generate decreased responses in frontal-parietal areas that are recruited by effortful, cognitive tasks. For example, in one of our studies, when subjects viewed a “threat cue” that signaled a potential upcoming shock, deactivation was observed in emotion-related regions (Choi et al. Reference Choi, Padmala and Pessoa2012).

In all, cognitive-emotional interactions take diverse forms that go beyond a straightforward antagonistic relationship (Ch. 5). Instead, I suggest that lateral PFC, in particular, is a convergence site for cognitive and emotional signals where they are integrated.

4.1. Energizing force versus selective effects

Traditional accounts describing motivation as a global activation independent of particular control demands have been echoed by a functional MRI study in which Kouneiher and colleagues (Reference Kouneiher, Charron and Koechlin2009) argue that motivation and cognitive control can be regarded as two separate and additive – instead of interactive – factors. Although there is little question that motivation can have generalized, activating contributions to behavior (see Robbins & Everitt Reference Robbins and Everitt2007; Salamone et al. Reference Salamone, Correa, Farrar, Nunes and Pardo2009), current findings (Ch. 6) underscore the ability of motivation to shape behavior selectively, whether by reducing response conflict or task-switch costs, via selective effects on working memory, or by improving long-term memory (for the latter, see the work of Adcock and colleagues; e.g., Adcock et al. Reference Adcock, Thangavel, Whitfield-Gabrieli, Knutson and Gabrieli2006). Another body of research demonstrating selective effects of motivation has investigated attentional effort, as described by Sarter and colleagues (e.g., Sarter et al. Reference Sarter, Gehring and Kozak2006).

5.1. Emotion

5.2. Executive competition

Because emotion can either enhance or impair cognitive performance, to see how emotional content impacts executive control, we must consider at least two factors: the strength or arousal of the stimulus (or manipulation) and task relevance (see also Mather & Sutherland Reference Mather and Sutherland2011). When arousal is “low” and affective significance is task irrelevant, some interference with the main task may be observed and the behavioral effect will be typically small. When, however, arousal is “high” and the stimulus/manipulation is task irrelevant, resources are more fully diverted toward the processing of the emotional item and, because the mobilization of resources is more pronounced, the effects on behavior are greater (Lang et al. Reference Lang, Davis and Ohman2000; Panksepp Reference Panksepp1998). For example, in our investigation of cognitive-emotional interactions, Choi, Padmala, and I (Reference Choi, Padmala and Pessoa2012) observed that response conflict increased on trials with the possibility of shock, suggesting that the impact of emotion on behavior comes in part from the more vigorous recruitment of attentional/effortful control required to prioritize the processing of high-arousal items. Naturally, attentional/effortful control involves executive control resources and, because situations associated with high levels of arousal are expected to recruit some of these resources (see also Bishop Reference Bishop2007; Eysenck et al. Reference Eysenck, Derakshan, Santos and Calvo2007; Mathews & Mackinstosh Reference Mathews and Mackinstosh1998), interference with executive functions will ensue (Fig. 7A). The impact of emotion on performance thus occurs because of limited processing capacity and competition for common-pool resources.

Figure 7.

Executive control, competition, and processing resources. (A-C) Processes are proposed to share resources called “common-pool resources” (smaller ellipses in gray), such that the engagement of one will detract from the processing of the other. Common-pool resources are necessary for general functions of attentional/effortful control. (A) High-arousal emotional stimuli recruit common-pool resources that allow their processing to be prioritized, thus detracting from other mechanisms sharing those resources. (B) These stimuli also trigger executive functions, such as updating, shifting, and inhibition, to handle the challenges to the organism, as indicated by the arrows emanating from attentional/effortful control. (C) Competition for resources during cognitive and emotional manipulations can, at times, produce push-pull–like interactions. Reproduced with permission from Pessoa (Reference Pessoa2009).

What about the situation when the emotional stimulus is task relevant? Here, two outcomes are possible. If the affective intensity is “low,” task performance might improve because control will be mobilized in the service of handling the task at hand, and the executive functions needed for task completion will more effectively compete for resources. In all, task performance will be enhanced. If, however, the affective intensity is sufficiently high, task performance might be compromised. Thus, in a study of response inhibition, my colleagues and I asked participants to perform a simple discrimination task but to withhold responding when they saw a stop signal (Pessoa et al. Reference Pessoa, Padmala, Kenzer and Bauer2012). We found that, when we used both fearful and happy faces as low-arousal stop signals, response inhibition was enhanced relative to neutral faces, but when we employed high-arousal emotional stimuli (previously paired with mild shock) as stop signals, response inhibition was impaired relative to neutral stimuli. Thus, inhibition performance was degraded even though emotional content was task relevant. We conjectured that processing the emotional stimulus consumed resources needed for inhibition.

5.3. Processing resources

Although the concept of resources invoked in accounts of the limits of information processing has been criticized in the past (e.g., Logan Reference Logan1988; Navon Reference Navon1984; Neisser Reference Neisser1976) and has not been mechanistically specified, further insight into it can be gained by examining brain regions sensitive to changes in task load, including the attentional network. Accordingly, researchers have probed attentional bottlenecks observed during tasks such as the attentional blink and the phenomenon known as the “psychological refractory period.” Based on these paradigms, Marois and colleagues have proposed the existence of a “unified” attentional bottleneck that involves several regions of the fronto-parietal attentional network (Tombu et al. Reference Tombu, Asplund, Dux, Godwin, Martin and Marois2011). If robust emotional manipulations indeed consume processing resources, then they should engage sites implicated as “bottleneck areas.” As described in section3, a compilation of activation peaks in aversive conditioning functional MRI studies revealed sites throughout the lateral and medial PFC, in addition to the anterior insula (Pessoa Reference Pessoa2009). Thus, attentional bottleneck regions are consistently recruited during emotion processing. If this recruitment prevents them from being adequately engaged when neutral task-related processing is required, we should expect to see behavioral impairments (see also Bishop et al. Reference Bishop, Duncan, Brett and Lawrence2004).

5.4. Triggering additional functions

A distinct impact of emotion is the result of its influence on specific resources. Dealing with an emotional stimulus requires the types of behavioral adjustments that characterize executive functions. For example, to refresh the contents of working memory, to switch the current task set, and to cancel previously planned actions might require updating, shifting, and inhibition, respectively. Such adjustments recruit specific resources required for emotional processing (Fig. 7B) and, if these resources are temporarily unavailable for the task at hand, behavioral performance will be compromised – the more so, the stronger the emotional manipulation (see below). An example may help to illustrate. Suppose a subject is performing a cognitive task and a change in background color signals that she or he will receive a shock sometime in the next 30 seconds. The subject might update the contents of working memory to include the “shock possible” information, shift between the execution of the cognitive task and “monitoring for shock” every few seconds, and, if another cue indicated that the shock would be delivered in the next second, inhibit a response to the cognitive task to prepare for the shock. In other words, dealing with the emotional situation necessitates the same types of executive functions that are considered to be the hallmark of cognition.

5.5. Cognitive-emotional interactions versus push-pull

The dual competition framework suggests that brain regions important for executive control are actively engaged by emotion. In contrast, push-pull studies have demonstrated reduced signals in some of these regions when emotional stimuli are shown. Hence, the two frameworks appear to make opposite predictions. The findings of Anticevic and colleagues (Reference Anticevic, Repovs and Barch2010) provide a potential clue as to when we might expect antagonistic interactions. Whereas, relative to neutral, negative distractors decreased responses in the dorsal-lateral PFC during the delay period of the working memory task, task-related distractors (stimuli similar to items to be remembered) actually increased responses, in much the way increases in working memory demand would. What explains this difference?

Dealing with the negative stimuli during the delay period produced a momentary “neglect” of the memory maintenance (Anticevic et al. Reference Anticevic, Repovs and Barch2010). In contrast, because neutral task-related distractors were so similar to the to-be-remembered items, participants may in effect have also held them in memory so as to avoid matching the final probe stimulus to a distractor. Consequently, the distractors may actually have increased working memory load. I therefore suggest that cognitive-emotional push-pull interactions are related to a type of competition that directs processing away from the concurrently executed main task, thereby producing decreased activation (in relative terms) in some of the key frontal and parietal regions underlying the task at hand (Fig. 7C). Which is to say, deactivations are the result of competitive interactions between resources required for executive functions. As such, they should be understood not in terms of a mutually suppressive relationship between emotion and cognition, but in terms of executive competition.

5.6. Neural interactions

Cognitive-emotional interactions rely on the communication between “task networks” (e.g., the attentional network during attention tasks) and “valuation networks,” which involve both subcortical regions, such as hypothalamus and amygdala, and cortical ones, such as orbitofrontal cortex, anterior insula, and medial PFC. These interactions are suggested to take place via multiple forms of communication (Fig. 8).

Figure 8.

Modes of interaction between cognitive and emotion/motivation networks. (1) Interactions rely on hub regions, such as those in the dorsal-medial prefrontal cortex, which are part of both attentional and motivational networks (hub region in the slice and gray node in the cortical valuation network). (2) In addition, specific regions may link the two networks, either directly or via the thalamus. (3) Finally, motivational signals are further embedded within cognitive mechanisms through the action of diffuse neuromodulatory systems. Key: ant., anterior; NAcc, nucleus accumbens; OFC, orbitofrontal cortex; PCC, posterior cingulate cortex; PFC, prefrontal cortex; SN, substantia nigra; VTA, ventral tegmental area. Reproduced with permission from Pessoa and Engelmann (Reference Pessoa and Engelmann2010).

First, direct pathways connect task and valuation networks. One example is the pathway between orbitofrontal and lateral PFC (Barbas & Pandya Reference Barbas and Pandya1989). Other examples are the pathways between the extensively interconnected lateral surface of the PFC (including dorsal-lateral PFC) and all cingulate regions (Morecraft & Tanji Reference Morecraft, Tanji and Vogt2009). A second type of communication relies on “hub” regions at the intersection of task and valuation networks – hubs are highly connected and central regions that play a key role in information communication between different parts of a network.

What are some of the hub regions? Dorsal-medial PFC plays a prominent role as “common node” of executive and emotional networks because of its participation in integrating inputs from diverse sources, notably cognitive and affective ones (e.g., Devinsky et al. Reference Devinsky, Morrell and Vogt1995; Fig. 8). This region is involved in multiple executive functions, such as conflict detection, error likelihood processing, and error monitoring (Alexander & Brown Reference Alexander and Brown2011). As reviewed in section3, the dorsal-medial PFC is also reliably engaged during conditions involving negative affect (see Fig. 3), as are all sectors of the anterior-medial PFC.

A second hub region, the anterior insula, is important for interoception (Craig Reference Craig2002; Reference Craig2009). Moreover, threat, uncertainty, and risk are all factors that engage the anterior insula (Singer et al. Reference Singer, Critchley and Preuschoff2009), which is also reliably recruited by cognitive processes (Craig Reference Craig2009; Van Snellenberg & Wager Reference Van Snellenberg, Wager, Christensen, Goldberg and Bougakov2010). Indeed, in a recent analysis of the functional diversity of brain regions (see sect. 7.4 and Fig. 14), the anterior insula emerged as one of the most diverse in the brain (Anderson et al. Reference Anderson, Kinnison and Pessoa2013; see also Uddin et al. Reference Uddin, Kinnison, Pessoa and Anderson2013). In all, the dorsal-medial PFC and anterior insula provide substrates for ample cognitive-emotional integration that, in broad terms, include both bodily “input” and “output” signals (roughly, via anterior insula and dorsal-medial PFC, respectively). Of course, these regions do not work in isolation. During cognitive-emotional interactions, they interact with the lateral PFC and parietal cortex, for example (Fig. 8).

A third type of communication depends on the diffuse action of neuromodulatory systems, including the action of dopamine and norepinephrine. Widespread modulatory connections originating from these systems reach large portions of the cortical surface and multiple subcortical areas, from which they are able to rapidly influence brain responses during emotional situations (Arnsten Reference Arnsten2009; Panksepp Reference Panksepp1998).

6.1. Perceptual competition

How does motivational significance influence sensory processing? Several of the circuits described in the context of emotion operate in the case of motivation, too. Notably, the interactions between valuation networks and fronto-parietal regions important for attentional control are engaged by both emotion and motivation. An illustration of the latter was described in the response-conflict study reviewed previously (see Fig. 4 and Fig. 5). One of the differences between emotion and motivation is that at times the interactions will involve different valuation regions, say, the amygdala in the case of emotion and the accumbens in the case of motivation. Yet, the general form of the interaction is similar. Which is to say, items of affective/motivational significance will redirect the flow of signals such that their processing is favored. I further propose that mechanisms involving the basal forebrain and the pulvinar operate for both emotion and motivation. More generally, despite the considerable differences between basal forebrain, pulvinar, and fronto-parietal mechanisms, each shapes, say, visual perception by altering competition in visual cortex. Thus, the idea is that their respective pathways may be engaged both during emotional and motivational conditions. Once they are engaged, the downstream effects on visual processing (and elsewhere) may be the same for both types of manipulation. A corollary of this notion is that priority maps (Awh et al. Reference Awh, Belopolsky and Theeuwes2012; Baluch & Itti Reference Baluch and Itti2011; Fecteau & Munoz Reference Fecteau and Munoz2006; Serences & Yantis Reference Serences and Yantis2006; Wolfe Reference Wolfe1994) – containing representations of spatial locations that are behaviorally important – incorporate signals as a result of an item's affective and motivational significance.

6.2. Executive competition

Motivation influences executive competition, too, and section 4 described examples during response-conflict, task switching, and working memory. Two effects of motivation on executive function are proposed here. First, motivation sharpens executive functions by enhancing them or by making them more efficient (Fig. 9). An illustration of this effect was the working memory study by Kobayashi and colleagues (Reference Kobayashi, Lauwereyns, Koizumi, Sakagami and Hikosaka2002) in which reward increased the amount of transmitted information regarding the item being maintained in memory. Second, motivation reallocates resources available to executive functions, increasing the likelihood of reward attainment by improving performance (Fig. 9). For example, in the study by Jimura and colleagues (Reference Jimura, Locke and Braver2010) brain responses appeared to reflect a shift toward a proactive control strategy that was beneficial to performance. Motivation can thus be viewed, at times, as reallocating resources to prioritize implementation of the rewarded task component at the expense of unrewarded components (Fig. 7C) (which at times can lead to deleterious performance effects; Padmala & Pessoa Reference Padmala and Pessoa2010).

Figure 9.

Executive control and reward. Motivation is proposed to have two key effects on executive function: first, it fine-tunes executive functions that are important for the task at hand (represented by the change of shape of the updating function; see solid arrow); and, second, it redistributes the allocation of common-pool resources (gray ellipse; see dashed arrow), and thus modulates how executive processes compete with each other. Reproduced with permission from Pessoa (Reference Pessoa2009).

6.3. Neural interactions

The same general architecture for cognitive-emotional interactions is proposed to underlie cognitive-motivational interactions (Fig. 8). In particular, the interactions between valuation networks and fronto-parietal regions important for attention and executive control are suggested to be common to both emotion and motivation. Subcortical reward/valuation regions include the caudate (particularly more ventral portions), nucleus accumbens, midbrain, and the amygdala; and cortical regions include orbitofrontal cortex, anterior insula, medial PFC, and posterior cingulate cortex.

Hub regions also play a central function during interactions between cognition and motivation. For example, Mesulam and colleagues suggested that posterior cingulate cortex is important for the integration of motivational and spatial attention information (Mohanty et al. Reference Mohanty, Gitelman, Small and Mesulam2008; Small et al. Reference Small, Gitelman, Simmons, Bloise, Parrish and Mesulam2005; see also Platt & Huettel Reference Platt and Huettel2008). Another key hub region is medial the PFC (including the dorsal PFC), already discussed in the context of emotion. Indeed, multiple sources of evidence demonstrate that the medial PFC is a critical component of the motivational system (see Summerfield & Koechlin Reference Summerfield and Koechlin2009; Vogt Reference Vogt2008; Walton et al. Reference Walton, Rudebeck, Bannerman and Rushworth2007). Shackman and colleagues (Reference Shackman, Salomons, Slagter, Fox, Winter and Davidson2011) proposed that the dorsal-medial PFC implements domain-general processes of adaptive control, based on the region's extensive contributions to cognitive control, negative affect, and nociception. I suggest that the proposal should be extended to incorporate motivation as well, which is to say, that dorsal-medial prefrontal context implements motivated adaptive control – where “motivated” is understood to include emotional processing. The anterior insula has been repeatedly implicated during the processing of negative events (Paulus & Stein Reference Paulus and Stein2006; Simmons et al. Reference Simmons, Strigo, Matthews, Paulus and Stein2006). But a growing number of studies implicate it during appetitive conditions (Liu et al. Reference Liu, Hairston, Schrier and Fan2011a; Mizuhiki et al. Reference Mizuhiki, Richmond and Shidara2012; Naqvi & Bechara Reference Naqvi and Bechara2009; Padmala & Pessoa Reference Padmala and Pessoa2011; Samanez-Larkin et al. Reference Samanez-Larkin, Gibbs, Khanna, Nielsen, Carstensen and Knutson2007). Here, I propose that the anterior insula is a chief hub region for cognition-motivation interactions.

As in the case of emotion, a third mode of communication involves the widespread action of neuromodulatory signals, including those of dopamine and acetylcholine. It is possible that dopaminergic and cholinergic neuromodulation provide a key mechanism by which motivation sharpens executive control (and hence behavioral performance), for example, by improving the signal-to-noise ratio of relevant neurons (e.g., Goldman-Rakic et al. Reference Goldman-Rakic, Leranth, Williams, Mons and Geffard1989). Motivation thus enhances processing efficiency in target cortical and subcortical regions.

6.4. “Resources”: Linking human and animal literatures

The dual competition model employs the admittedly vague concept of “resources.” One way in which a more mechanistic account can be formulated is to build on the extensive literature of motivation in nonhuman animals. Redgrave and colleagues (Redgrave & Gurney Reference Redgrave and Gurney2006; Redgrave et al. Reference Redgrave, Prescott and Gurney1999) have proposed that dopamine-related circuits in the striatum facilitate the reallocation of limited processing capacity toward unexpected events of behavioral significance, including rewarding ones. Thus, instead of simply providing a “reward signal,” striatal activation drives the redistribution of available resources to salient events whose processing is then prioritized (see also Horvitz Reference Horvitz2000; Zink et al. Reference Zink, Pagnoni, Martin-Skurski, Chappelow and Berns2004). Furthermore, Sarter et al. (Reference Sarter, Gehring and Kozak2006) propose that increased prefrontal cholinergic activity contributes to the recruitment of goal-driven mechanisms (see also Sarter et al. Reference Sarter, Hasselmo, Bruno and Givens2005), which depend on fronto-parietal regions, act to enhance sensory processing and to attenuate interference effects.

6.5. Mechanisms of motivational effects: Conceptual issues

Disentangling the contributions of cognition and motivation to neural signals is far from easy, especially when experiments involve goal-directed task manipulations. For example, in human studies, subjects may be instructed that a potential reward will result following a cue stimulus if their performance is both fast and accurate. In such cases, increased brain signals may reflect enhanced attention because subjects are more likely to engage attention when a reward is at stake. But whether the increased signals actually reflect greater attention is another matter, an issue Maunsell described forcefully in the context of monkey physiology studies of attention:

Maunsell's point raises the broader issue of the relationship between motivation and cognition. One possibility is that motivation has effects that take place independently of cognition (Fig. 10A). A second is that motivation modulates behavior by engaging the same functions that are used by cognition, in which case, the impact of motivation on behavior could be described as “mediated by cognition” (Fig. 10B). This mediation could be partial only, such that both direct (motivation-to-behavior) and indirect (motivation-via-cognition-to-behavior) effects take place. A third possibility is that cognition and motivation are more intertwined, such that they jointly guide behavior (Fig. 10C), in which case, although certain processes could be described as “cognitive” and others as “motivational,” the interactions between them are sufficiently strong that their separation is more semantic than real. See Chelazzi et al. (Reference Chelazzi, Perlato, Santandrea and Della Libera2013) for a related discussion.

Figure 10.

Three models of the relationships between attention and motivation. (A) In the parallel model, attention and motivation have independent effects on behavior. (B) In the mediation model, the influence of motivation on behavior is mediated via attentional systems. (C) In the integration model, attentional and motivational systems interact so strongly they cannot be decomposed. Adapted with permission from Pessoa and Engelmann (Reference Pessoa and Engelmann2010).

The situation Maunsell describes thus could be portrayed in terms of the mediation model (B): Mechanistically, effects of attention are obtained via “attentional circuits.” Whereas this relationship would presumably indicate that such motivational effects are less interesting, l argue that how motivation recruits “cognitive” circuits is as important as which circuits it recruits. Indeed, I suggest that the major issue is conceptual, and that by using separate boxes for “attention” and “motivation,” the models of Figure 10 describe motivation in an impoverished way. As in the case of emotion and cognition (Pessoa Reference Pessoa2008), I propose that it is counterproductive to carve the brain into “attention” (or “cognition”) and “motivation.” Chapters 6 and 7 outline how motivational signals are embedded into cognition (and perception) through multiple mechanisms. In this manner, the “inextricably confounded” relationship described by Maunsell (Reference Maunsell2004) ceases to be a problem and can be seen as a property of brain organization (see also sect. 7.2).

7.1. Overlapping networks

Commonly, networks are described in terms of unique, nonoverlapping sets of brain regions (Fig. 11A). But this assumes that brain areas compute a specific function, one that is perhaps elementary and needs other regions to be “actualized,” but nonetheless is well defined. I propose that networks contain overlapping regions, such that specific areas will belong to several intersecting networks (Mesulam Reference Mesulam1990). In this manner, the processes carried out by an area will depend on its network affiliation at a given time. What determines a region's affiliation? For this, the importance of the context within which a brain region is operating must be considered (McIntosh Reference McIntosh2000). For example, in Figure 11B, region A_n will be part of network N₁ during a certain context C_k, but will be part of network N₂ during another context C_l. The existence of context-dependent, overlapping networks also means that from the perspective of structure-function mappings summarized in Figure 11B, a given region will participate in multiple processes. In addition, the importance of context emphasizes the need to consider dynamic aspects of structure-function relationships. A network needs to be understood in terms of the interactions between multiple brain regions as they unfold temporally. In the extreme, two networks may involve the exact same regions interacting with each other in distinct ways across time.

Figure 11.

Structure-function mapping and networks. (A) The “landscape of behavior” depicts the multidimensional space of behaviors. A₁, A₂, A_n, B₁, and B_n=brain regions; N₁ and N₂=networks; P_i and P_j=processes. (B) Intersecting networks. The networks C_k and C_l (and the additional ones) intersect at node A_n. (C) Dynamic aspects. Because region A_n will have network affiliations that vary as a function of time, the processes carried out by the emerging networks will evolve across time and lead to dynamic “landscapes of behavior.” The four time points represented are such t₁ is close to t₂ but far from t₃ and t₄, which are close to each other. (D) Structure-function mappings in the case of networks. Two networks may instantiate similar processes, a case of many-to-one mapping. The reverse relationship is also suggested to apply to networks, namely, one-to-many mappings. Reproduced with permission from Pessoa (Reference Pessoa2013).

Though simple, the “multiple affiliation” point is sufficiently important to merit an example. Consider the case of the amygdala. Even a simplified view of its anatomical connectivity shows that, minimally, it belongs to three networks. The first is a “visual network,” as the amygdala receives fibers from anterior parts of temporal cortex. The amygdala, by its turn, influences visual processing via a set of projections that reach most of ventral occipito-temporal cortex. The second is the well-known “autonomic network,” as evidenced by connectivity with subcortical structures such as the hypothalamus and periaqueductal gray, among others. Via this network, the amygdala participates in the coordination of many complex autonomic mechanisms. The third is a “value network,” as evidenced by its connectivity with the orbitofrontal cortex and medial PFC. In total, the amygdala affiliates with different sets of regions (“networks”) in a highly flexible and context-dependent manner. Many other examples of this dynamic affiliation idea exist, including the fronto-parietal cortex, whose regions affiliate with others based on task demands (Cole et al. Reference Cole, Reynolds, Power, Repovs, Anticevic and Braver2013).

Two issues deserve further consideration here. First, when describing networks, the term process is preferable to function. One reason is that a process emerges from the interactions between regions – it is thus an emergent property (see Bressler & Menon Reference Bressler and Menon2010). Furthermore, a process is viewed as a useful external description of the operation of the network, and not necessarily as a fixed internal computation implemented by the network (Thompson Reference Thompson2007; Thompson & Varela Reference Thompson and Varela2001; Varela et al. 1992; cf. Lindquist & Barrett Reference Lindquist and Barrett2012).

A second – and critical – issue is whether utilizing networks solves the many-to-many mapping problem we face when considering regions as the unit of interest. In other words, does a description of structure-function relationships in terms of networks allow for a one-to-one mapping? For example, in the context of the salience network, Menon, Uddin, and colleagues note that “to determine whether this network indeed specifically performs this function will require testing and validation of a sequence of putative network mechanisms” (Bressler & Menon Reference Bressler and Menon2010, p. 285; see also Moussa et al. Reference Moussa, Vechlekar, Burdette, Steen, Hugenschmidt and Laurienti2011). The prospect of simpler structure-function relationships (hence less context dependent) is discussed by Buckner and colleagues when describing regions of high connectivity: “An alternative possibility is that the hubs reflect a stable property of cortical architecture that arises because of monosynaptic and polysynaptic connectivity. Within this alternative possibility, the same hubs would be expected to be present all of the time, independent of task state” (Buckner et al. Reference Buckner, Sepulcre, Talukdar, Krienen, Liu, Hedden, Andrews-Hanna, Sperling and Johnson2009 pp. 1867–68; emphasis mine).

Unfortunately, the attempt to map structure to function in a one-to-one manner in terms of networks will be fraught with some of the difficulties encountered when considering individual brain regions (Ch. 8). To be true, the problem is ameliorated, but the mapping is still highly complex. For example, two distinct networks may generate similar behavioral profiles (Fig. 9D; many-to-one); a given network will also participate in several behaviors (one-to-many). Broadly speaking, a network's operation will depend on several more global variables, namely an extended context that includes the state of several “neurotransmitter systems,” arousal, slow wave potentials, and so forth. In other words, a network that is solely defined as a “collection of regions” is insufficient to eliminate the one-to-many problem. What if we extend the concept of a network with these additional variables? For example, Cacioppo and Tassinary (Reference Cacioppo and Tassinary1990) suggest that psychological events can be mapped to physiological ones in a more regular manner by considering a spatiotemporal pattern of physiological events. The notion of a network is thus extended to incorporate other physiological events, for example, the state of a given neurotransmitter (as in the elegant work by Marder and colleagues; see Marder & Goaillard Reference Marder and Goaillard2006). How extensive does this state need to be? Clearly, the usefulness of this strategy in reducing the difficulties entailed by many-to-many mappings will depend on how broad the context must be (Thompson Reference Thompson2007).

7.2. An example: Cognitive-motivation interactions

Graph-theoretical analysis of functional neuroimaging data has focused almost exclusively on characterizing the large-scale properties of resting-state data (Bullmore & Sporns Reference Bullmore and Sporns2009; Wang et al. Reference Wang, Zuo and He2010). In a recent study, we sought instead to understand the network properties of a focused set of brain regions during task conditions engaging them (Kinnison et al. Reference Kinnison, Padmala, Choi and Pessoa2012). In particular, we analyzed the data of the response-conflict task discussed previously (Fig. 4 and Fig. 5; Padmala & Pessoa Reference Padmala and Pessoa2011). At the network level, global efficiency (a measure of integration) increased and decomposability (a measure of how easily a network can be divided in terms of smaller subnetworks or “communities”) decreased (Fig. 12). In other words, the network became less segregated with reward, revealing that one way in which a reward cue affects brain responses is by increasing functional connections across brain regions. From the vantage point of a single region, the changes in functional connectivity can be quite broad and can be characterized via a functional connectivity fingerprint (see Passingham et al. Reference Passingham, Stephan and Kotter2002). For example, the caudate (Fig. 6C) and the nucleus accumbens showed increases in functional connectivity to nearly all cortical regions that were driven by reward, reinforcing the notion of “embedding” described earlier. Finally, this example underscores the need to move beyond simple pairwise relationships between regions to a multivariate representation of the changes in functional connectivity that underlie network reorganization.

Figure 12.

Network structure and reward. (A) Community detection was applied to the set of brain regions that responded more strongly to reward than to no-reward context at the cue phase. Two communities were detected. (B) Comparison of the pattern of connectivity between reward and no-reward contexts revealed increases during the former, mostly between the two communities, reflecting increased integration with reward. Adapted with permission from Anderson et al. (Reference Bilder, Sabb, Parker, Kalar, Chu, Fox, Freimer and Poldrack2013).

7.3. Issues when considering networks

If we are to use networks to understand structure-function mappings, then we must consider several issues. I briefly describe them here (see also Pessoa Reference Pessoa2014).

7.4. Understanding a region's function via multidimensional profiles

If brain regions are engaged in many processes based on the networks they are affiliated with in particular contexts, they should be engaged by a range of tasks. Although this introduces outstanding problems, the availability of data repositories containing the results of thousands of neuroimaging studies provides novel opportunities for the investigation of human brain function (Yarkoni et al. Reference Yarkoni, Poldrack, Van Essen and Wager2010).

Like others (e.g., Robinson et al. Reference Robinson, Laird, Glahn, Blangero, Sanghera, Pessoa, Fox, Uecker, Friehs, Young, Griffin, Lovallo and Fox2012), my colleagues and I recently employed a data-driven approach to investigate the functional repertoire of brain regions based on a large set of human functional MRI studies (Anderson et al. Reference Anderson, Kinnison and Pessoa2013). We characterized the function of brain regions in a multidimensional manner via their functional fingerprint (Passingham et al. Reference Passingham, Stephan and Kotter2002), namely, the relative degree of engagement of the region across a range of task domains (Fig. 13, top); the approach was extended to networks, too (Fig. 13, bottom). Based on the fingerprints, we calculated a diversity index to summarize the degree of functional diversity; a brain region with high diversity would be one engaged by tasks in many domains, whereas a low-diversity region would be engaged by a few domains. We found that diversity varied considerably across the brain (Fig. 14).

Figure 13.

Functional fingerprints of regions and networks. (Top) Polar plots illustrate the fingerprints of three brain regions. Each vertex corresponds to one of the domains investigated. Both the left anterior insula and the left intraparietal sulcus exhibited diverse functional profiles. The superior temporal gyrus in the vicinity of auditory cortex was less diverse, although the fingerprint revealed its involvement in emotional processing, in addition to audition. (Bottom) Polar plots illustrate the fingerprints of two brain networks, which were defined by Toro and colleagues (Reference Toro, Fox and Paus2008) based on a meta-analysis of task activation data. The frontal-parietal “attention” network was a task-positive network generated by “seeding” the left intraparietal sulcus. The cingulate-parietal “resting-state” network was a task-negative network generated by “seeding” ventral-anterior medial prefrontal cortex. Although both networks are quite diverse, the analysis revealed that they are fairly complementary to one another. Adapted with permission from Anderson et al. (Reference Bilder, Sabb, Parker, Kalar, Chu, Fox, Freimer and Poldrack2013).

Figure 14.

Functional diversity map. Areas of higher functional diversity are shown in warm colors, and areas of lower diversity are shown in cool colors (color bar represents diversity Shannon entropy values). Locations without colors did not have sufficient findings for the estimation of diversity. Adapted with permission from Anderson et al. (Reference Bilder, Sabb, Parker, Kalar, Chu, Fox, Freimer and Poldrack2013).

Our findings suggest that brain regions are very diverse functionally, in line with the points raised by Poldrack (Reference Poldrack2006; Reference Poldrack2011). Beyond the descriptive aspects of the approach, it outlines a framework in which a region's function is viewed as inherently multidimensional: Avector defines the fingerprint of a region in the context of a specific domain structure. Although the domain that we explored used a task classification scheme from an existing database, it was not the only one possible. How should one define the domain structure? One hope is that cognitive ontologies can be defined that meaningfully carve the “mental” into stable categories (Bilder et al. Reference Bilder, Sabb, Parker, Kalar, Chu, Fox, Freimer and Poldrack2009; Price & Friston Reference Price and Friston2005). I contend, however, that no single ontology will be sufficient. Instead, it is better to conceive of several task domains that are useful and complementary in characterizing brain function and/or behavior. Thus, a region's functional fingerprint needs to be understood in terms of a family of (possibly related) domains. Finally, the framework can be extended to networks, provides a way to compare them as described next, and to advance our understanding of the properties of constituent nodes (see Anderson et al. Reference Anderson, Kinnison and Pessoa2013).

7.5. Comparing brain networks

In several instances, investigators have proposed closely related networks (for example, “dorsal attention” and “executive control”), raising the possibility that they could be closely related, or possibly the same except for a change in label. Thus, developing tools that help characterize and understand brain networks is of great relevance and could help reveal principles of organization.

With this in mind, we asked the following question (Anderson et al. Reference Anderson, Kinnison and Pessoa2013): What is the relationship of the functions of regions belonging to a given network? One approach is to evaluate how homogeneous fingerprints are in a network. In other words, are fingerprints from the regions of network X more similar to each other than to those of regions from network Y? In our investigation, we chose to not investigate a unique set of networks, but instead considered possibly related (or even closely related) networks defined by different research groups and approaches, including meta-analysis, resting-state, and task-based approaches. To contrast brain networks to each other in terms of the functional fingerprints of the component regions, we employed a multivariate test based on “statistical energy” (Aslan & Zech Reference Aslan and Zech2005). Interestingly, several network pairs were found to be only modestly distinct (e.g., dorsal and ventral attention networks). Moreover, some of the networks that have been distinguished from one another in the past were not strongly distinct (e.g., the fronto-parietal “adjust control” network and the cingulo-opercular “maintain task set” network described by Dosenbach and colleagues Reference Dosenbach, Fair, Cohen, Schlaggar and Petersen2008).

We also evaluated the assortativity of the regions within networks, where assortativity refers to the tendency of “like to connect with like” (e.g., Christakis & Fowler Reference Christakis and Fowler2007). Functional fingerprints within an assortative network would be relatively similar to each other and relatively dissimilar to fingerprints from other networks. Interestingly, we observed several levels of assortativity, suggesting that existing networks are composed of nodes whose functional repertoire varies in their homogeneity. In fact, one version of the task-negative network tended to be disassortative, namely, its regions tended to be more dissimilar to each other than to those of other networks, consistent with the notion that task-negative networks are relatively heterogeneous (Andrews-Hanna et al. Reference Andrews-Hanna, Reidler, Sepulcre, Poulin and Buckner2010).