Dynamic routing of information flow between brain areas
Possible roles for the prefrontal cortex and the pulvinar nucleus of the thalamusMarrit Zuure University of Amsterdam
March 19, 2014
Literature thesis for the master Brain & Cognitive Sciences Track Cognitive Neuroscience
Student ID 6030823 Supervisor Tobias Donner
University of Amsterdam Co-asssessor Markus Siegel
University of T¨ubingen Word count 7208
Abstract
Human can respond flexibly to their environment. This is achieved through a
flexible mapping from input to output in the brain (also known as dynamic routing),
influenced by internal states, desires, and other top-down information. In order to
organize dynamic routing across the brain in pursuit of a single goal, a coordinating
structure is needed. Such a structure should meet three requirements: it should
have wide-spread anatomical connections so that its neural activity can precisely bias
activity in other areas, its neurons should encode the relevant top-down information,
and it should entrain coherent oscillatory neural activity in multiple other areas to
enhance communication between them. I compared two candidate coordinating
areas on these points: the prefrontal cortex (PFC) and the pulvinar nucleus of the
thalamus. The PFC meets more of the requirements, but several key experiments
still need to be done for a strong conclusion. Last, I propose the existence of multiple
coordinating “network hubs” as an alternative to the idea of a single coordinating
Contents
1 Introduction 4
2 Oscillatory dynamics in the brain 5
2.1 Frequency band characteristics . . . 7
2.2 Coherence through communication . . . 8
2.3 Selective communication . . . 9
2.4 Top-down coordination of oscillatory dynamics . . . 11
3 The prefrontal cortex 12 3.1 Anatomical connections . . . 13
3.2 Function and informational content . . . 13
3.3 Coordination of oscillatory activity . . . 15
3.4 The PFC’s role in dynamic routing . . . 17
4 The pulvinar nucleus of the thalamus 17 4.1 Anatomical connections . . . 18
4.2 Function and informational content . . . 19
4.3 Coordination of oscillatory activity . . . 20
4.4 The pulvinar’s role in dynamic routing . . . 21
5 Conclusion and future directions 22 5.1 The current state of research . . . 23
5.2 A new view: distributed network hubs . . . 24
5.3 Closing remarks . . . 27
1
Introduction
Humans, as complex organisms, can perform a broad range of actions. In nearly any situation, we can select from a large range of appropriate – and less appropriate – responses. Being flexible in our action decisions is a vital component of interacting adequately with the world around us. If we rigidly responded to the same input with the same output, our chances of survival would be negatively impacted.
The key to this flexibility is that humans act on more information than is present in the external world. We also consider our internal world: intentions, motivations, knowledge, emotions, goals, and intuitions. Using this information to guide our actions is part of a process called top-down control, or executive control. The concept of top-down control can be understood as the restrictions and biases imposed by higher-order abstract concepts (internal states) on lower-order sensory input or motor output. Essentially, any time a direct response to the environment is suppressed or bypassed in favor of a response mediated by internal states, top-down control is in play.
What is it about the brain that facilitates this? At its core, flexible action selection results from a dynamic mapping of input to output in the brain. Neurons pass elec-trical pulses that represent information through the brain. Abstract goals, memories, emotions, and other internal states control and modulate the flow of this information, by altering the functional connectivity between brain areas. This allows these areas to exchange information either more effectively or less effectively than before. As such, the neural representations of internal states serve to flexibly guide information flow in the brain. This principle is called dynamic routing.
As an example of top-down control through dynamic routing, consider the following scenario. Imagine you have been instructed to pay particular attention to a certain location on a computer screen, because an image will appear there. You now have the abstract goal of attending that location. The representation of that goal increases the
effective connectivity between the part of the primary visual cortex that is receptive to that particular location, and the areas that will process the visual information once the image is shown. In this way, top-down control guides and enhances information processing in accordance with internal states.
This explanation of dynamic routing through altering functional connectivity raises the question how neural representations can alter connectivity in the first place. Re-cent research suggests that oscillatory brain dynamics play a major role. This will be discussed in the next section.
2
Oscillatory dynamics in the brain
Neurons in the brain have a basic principle of operation: when their membrane potential reaches a certain threshold, they discharge quickly. A neuron that fires like this affects the neurons that it is connected to, by releasing neurotransmitters into the synaptic cleft. When these neurotransmitters bind to the postsynaptic receptors, a postsynaptic potential is generated. Depending on the neurotransmitters in question, this potential can make the receiving neuron more likely to fire (excitatory postsynaptic potential, EPSP) or less likely to fire (inhibitory postsynaptic potential, IPSP).
Neural spikes are not typically measurable outside the neuron. However, the EPSPs and IPSPs affect the charge of the extracellular fluid around the neuron. When multiple neurons are excited or inhibited simultaneously, the small influences on the extracellular potential (local field potential, LFP) sum together, resulting in a signal that can be measured from some distance if the neural populations are large and synchronous enough. Box 1 describes several methods used to measure the LFP.
Neurons tend to fire rhythmically, and interactions between inhibitory and excita-tory neurons can entrain even stronger rhythmic synchrony, causing oscillations in the extracellular potential. LFPs therefore typically consist of summed oscillatory signals.
Using mathematical techniques like Fourier analysis, the LFP can be decomposed into the “pure” oscillations that make up the signal. These oscillations are colloquially called brain waves.
Box 1: Measurement methods
Local field potentials can be measured over coarse or fine areas. A way to measure the LFP very locally is through the use of microelectrodes that are inserted into the brain. Depending on the exact placement, the size of the electrode, and the signal filter used, they can record LFPs or spikes from one or multiple neurons. In the latter case the technique is referred to as single-unit or multi-unit recording. Because it is so invasive, it is only ever used with human subjects if they have already been outfitted with a microelectrode for medical reasons. Most microelectrode recording studies are performed with animals.
Another measuring method is electrocorticography (EcoG). With this technique, electrodes are placed directly on the brain to detect summed LFPs over larger dis-tances. As it is invasive, human subjects of ECoG studies are almost always patients who are about to undergo surgery for epilepsy and who have been equipped with an ECoG rig to determine the source of their seizures.
The most common method of measuring LFPs is electro-encephalography (EEG). This non-invasive method records potentials using electrodes placed on the scalp. Because of the distance and the dense bone matter between the electrodes and the brain tissue that generates the oscillatory signals, EEG is fairly imprecise and sums signals over a relatively large area, resulting in a smoothed recording of the LFP. Source reconstruction techniques are often used to improve signal localization. Magneto-encephalography (MEG) is a fairly similar technique with some advantages, but is less commonly used because of its cost.
2.1 Frequency band characteristics
Oscillatory neural activity in the brain can occur at various temporal scales, ranging from very slow to very fast. It can take on frequencies from 1 Hz (so-called “slow wave” oscillations) up to 600 Hz (“ultra-fast” oscillations) (Schnitzler & Gross, 2005). As seen in Table 1, brain waves are commonly subdivided into various frequency bands, based on the fact that they seem to cluster around certain frequencies. Not all of these bands are equally well-studied, though research is ongoing.
Oscillatory activity also occurs at different spatial scales. In some cases, a small pop-ulation of neurons produces a very local oscillatory signal; in other cases, large amounts of neurons in wide-spread brain areas engage in the oscillations. As a general rule, high-frequency oscillations tend to be relatively local, and low-frequency oscillations are more global (von Stein & Sarnthein, 2000; Buzs´aki & Draguhn, 2004; Plankar, Breˇzan, & Jerman, 2013).
Resulting in part from their predisposition to certain spatial scales and from the resonant properties of the neural populations involved, neural oscillations of differing frequencies tend to indicate distinct processes within the brain. For example, alpha band oscillations have been linked to top-down processing, specifically selective inhibition in sensory areas, while gamma band oscillations appear to reflect the bottom-up processing of stimuli (Jensen & Mazaheri, 2010). Beta band oscillations appear to be involved in both the sensory and motor components of sensorimotor decision tasks, and may link sensory processing to motor processing (Siegel et al., 2008; Haegens et al., 2011). Theta band oscillations have been implicated in memory-related mental processes (Liebe et al., 2012). In fact, measuring oscillatory activity during tasks could be used to dissociate between different types of processing, even if the behavioral outcome for the task is exactly the same (Siegel et al., 2012).
Frequency range Frequency band name 1-3 Hz delta 4-7 Hz theta 8-14 Hz alpha 14-30 Hz beta 30-80 Hz gamma 80-200 Hz fast 200-600 Hz ultra fast
Table 1: The common definitions of oscillatory frequency bands in the brain. Data from Schnitzler & Gross (2005).
2.2 Coherence through communication
Neural oscillations consist of summed EPSPs and IPSPs. They therefore reflect fluctua-tions in neuronal excitability, meaning that neurons fire more easily at one phase of the oscillation than at another. To put it differently: the neural population involved in the oscillation is sensitized or desensitized to input, whether from itself or from more distant neural groups, during the different phases of the oscillation.
It follows that neural populations communicate most effectively when the neural spikes of the sending population arrive during the periods of optimal excitability of the receiving population. This means that efficiently communicating neural populations tend to produce oscillatory signals in the same rhythm, i.e. phase-locked and at the same frequency. Such groups of neurons are said to oscillate coherently. Accordingly, the principle that information transfer between coherent neural populations is facilitated is called communication through coherence (CTC) (Fries, 2005). The strength of coherence can vary; more coherent populations are more precisely phase-locked and will likely communicate more efficiently.
communication and effective information exchange. It is therefore an excellent mech-anism to implement dynamic routing and the guiding of information flow within the brain. Fries (2005) provides a clarifying example that illustrates both CTC and dy-namic routing. If a monkey perceives two stimuli, of which one is attended and one is not attended, local gamma-band synchronization within visual cortex will be stronger for the attended than for the unattended one. The activation patterns corresponding to these stimuli, which are not synchronous with each other (as explained in the next subsection), provide competing input to the next area in the visual processing stream. Because the attended stimulus representation is more precisely synchronized than the unattended one, the neurons in the next processing stage are more likely to start os-cillating in coherence with the attended input than with the unattended input. This results in enhanced communication between the first and next processing areas for that particular representation.
While this is a small-scale example of CTC, there are indications of large-scale ef-fects as well. Various studies link oscillatory coherence between task-relevant areas to improved behavioral performance, presumably as a result of improved communication between those areas. For example, Palva et al. (2010) studied coherence among a visual working memory network while subjects performed a visual working memory task. They found coherence in multiple frequency bands. The strength of coherence per individual was predictive for visual working memory capacity, suggesting a role for CTC in overall task performance.
2.3 Selective communication
At any given moment, many neural assemblies in the brain are busy firing and producing oscillatory signals, at many different locations, and at many different frequencies. With such a tumult of activity going on, it is important that signal transmission can be established in a selective manner. CTC is selective in various ways, allowing for effective
communication at various temporal and spatial scales at once.
Fries (2005) proposed two possible ways that the brain could implement selectivity in CTC: inhibited communication through mistimed coherence, or non-communication through non-coherence. In the case of the former, spikes from the sending population would always arrive at the moment that the receiving population is inhibited, and thus the information would not be properly transmitted. However, there is more evidence for non-communication through non-coherence: the sending and receiving population are not coherently oscillating, and the relation between their respective signals is essentially random. Because the receiving neurons are not reliably excitable when they receive the transmitted spikes, communication between such populations fails, and information exchange remains restricted to those areas that engage in synchronous oscillations.
Akam & Kullmann (2012) simulated CTC in a neural network and identified an additional caveat to non-communication through non-coherence. If an interfering input was oscillating incoherently, but in the same frequency as the signal input, the system’s capability to distinguish signal from noise was greatly degraded. They concluded that signal and noise needed to be distinguished by amplitude, phase, or frequency in order for sufficient signal-noise separation to occur. In the previously discussed example regarding monkey visual cortex, the representations of different stimuli - both providing input to the same upstream neural population - would need to vary on at least one of these components in order to be dissociated in further signal processing.
The fact that some neural populations, but not others, produce coherent oscilla-tions implies the existence of another form of selectivity: spatial selectivity. If neu-ral synchrony (and thus coherence) were present at a large scale, indiscriminate, and extended to irrelevant neural populations, the fact that coherence results in non-communication would not be enough to establish selective non-communication. In order to adequately guide information through task-relevant brain regions, certain groups of neurons - and no other irrelevant groups - need to precisely establish communication
through oscillatory coherence. Conveniently, oscillatory synchrony in the brain is spa-tially self-limiting. In healthy organisms, there is a balance between the propagation and the cessation of oscillatory signals that has the side effect of spatially restricting them. The fact that different frequencies of oscillations lend themselves to communication over different spatial scales (Plankar et al., 2013) may play a role in this restriction as well. In cases where this balance is disrupted, excessive phase-locking and synchrony arises in the brain; a pattern of activity that is typically seen during epileptic seizures (Stam & van Straaten, 2012). This illustrates that the spatial restriction of oscillations is not just serendipitous, but necessary for normal functioning.
To summarize, the CTC principle supports selective, point-to-point communication. It allows for many different oscillatory signals to be present simultaneously, yet operate independently and be engaged selectively, so that communication can occur at multiple temporal and spatial scales at once.
2.4 Top-down coordination of oscillatory dynamics
Neural synchrony has been established as a crucial factor in communication and the dynamic routing of information. In this section, we have discussed how communication through coherence is established, how it can affect task performance, and how selectivity is maintained. However, none of this explains how effective connectivity in the brain can vary based on internal goals, beliefs, or knowledge. The most likely explanation is that a structure exists that has access to representations of such goals and intentions, and coordinates the effective connectivity between task-relevant areas based on that information. In this way, it could guide the flow of information from sensory input to motor output.
If a singular brain structure like this exists, it has not yet been identified as such. It is nevertheless possible to define a number of characteristics that a structure of this type should possess. First, it should have wide-spread anatomical connections to other
brain areas. These connections are needed to influence neural activity in other areas. Second, its own neural activity patterns should encode the rules, goals, and other internal states that serve as the basis for dynamic routing. Third, it should be able to entrain oscillations in multiple other areas, so as to establish coherence between those areas.
In this thesis, I will evaluate two candidate areas - the prefrontal cortex and the pulvinar nucleus of the thalamus - on all three of these points, with the end goal of establishing whether one of them could reasonably be a source of coordination for the dynamic routing of information in the brain. Since the concepts of both dynamic routing and CTC are relatively new, this will be a challenge: there are many loosely connected studies that have to be unified into a coherent view, and even when combining informa-tion from many studies, evidence is at times still spread thin. This thesis will hopefully provide a useful overview of the current state of research into this topic.
3
The prefrontal cortex
The prefrontal cortex (PFC) is a large brain area that is generally thought to be respon-sible for top-down control and cognitive planning. In a landmark paper, Miller & Cohen (2001) presented a model of how the PFC could implement top-down control. In their view, the PFC integrates information from different neural sources; robustly maintains patterns of activity that represent goals, rules, or intentions; and uses these patterns to produce excitatory “bias signals” that influence information flow throughout the rest of the brain. As discussed in the previous section, dynamic routing through selective coherence could well be the underlying mechanism for such bias signals. Evidence for the PFC’s role as a coordinating source of synchrony will be reviewed in this section.
3.1 Anatomical connections
The PFC is anatomically well-connected for a coordinating role. Miller & Cohen (2001) note that the PFC is extensively interconnected with premotor areas, sensory areas, and a number of subcortical areas. Muhammad et al. (2006) describe the PFC as being “at an anatomical crossroad”. The PFC’s extensive connectivity means that it could exert influence over a large range of regions. It thus meets the first requirement for an area that could coordinate information flow through oscillatory coherence.
Additionally, the anatomical connections of the PFC suggest a role in the integration of information. The various subdivisions of the PFC vary in their exact connectivity, but most of them receive input from several primary sensory cortices. The PFC is also connected to sites of multimodal sensory integration. Additionally, there are ample in-terconnections between the different PFC areas, which allow for the local integration of inputs from various sources. It is therefore ideally connected to receive and process infor-mation concerning sensory input. The PFC is also connected to various limbic structures which process aspects of – among other things – memory, affect, and motivation (Miller & Cohen, 2001), supporting the idea that the PFC could combine information regard-ing internal states and the external world, and perhaps synthesize this information into abstract goals or rules.
3.2 Function and informational content
As noted, the PFC’s anatomical properties would hypothetically allow it to consolidate information about the external world (for example, the presence of two stimuli) and internal states (knowledge about which stimulus leads to reward), synthesizing it into an action plan, guideline, or behavioral goal (attend the left stimulus rather than the right stimulus).
Regardless of whether or not the PFC actually synthesizes such plans of action au-tonomously, its neural activity patterns need to vary with different goals or internal
states – in other words, encode top-down information – in order to alter cortical con-nectivity in a goal-based manner. The explanation for this is as follows: given that the effective connectivity needs to change every time the goal (and thus the top-down information) changes, the neural activity patterns in the structure controlling the con-nectivity also need to change every time the goal changes. Miller & Cohen (2001) note that their model relies on neural activity patterns within the PFC staying active for the duration of the goal, rule, or intention, biasing information flow in other areas while the goal remains in play.
Patients with various types of PFC damage show impaired performance on tasks that require top-down control, such as the Stroop task (Stuss, Floden, Alexander, Levine, & Katz, 2001) and the Wisconsin Card Sorting Task (WCST). In the WCST, subjects need to sort cards based on hidden rules that change intermittently. Patients with damage to the PFC make more errors, mostly due to their persistence in sorting cards according to a rule that is no longer in effect (Milner, 1963; Robinson, Heaton, Lehman, & Stilson, 1980). These findings lends credibility to the idea that the PFC engages in biasing information processing according to active rules, goals, or intentions.
Several studies have found that neural activity patterns within the PFC represent abstract rules: rules that are learned from particular cases, and then generalized to apply to similar scenarios (Muhammad, Wallis, & Miller, 2006). Wallis, Anderson, & Miller (2001) found that for a subset of neurons in macaque PFC, neural activity patterns recorded from individual neurons changed when the macaques switched between different task rules (classifying images as “same” or “different”). Muhammad et al. (2006) confirmed these results. The PFC thus meets the second of the requirements for a coordinating source of synchrony: part of its neurons code for the internal states that form the basis for guiding the cortical information flow.
To top it off, in accordance with the model proposed by Miller & Cohen (2001), the PFC appears to be capable of selectively maintaining its representations. Miller
et al. (1996) showed that the PFC is capable of robustly maintaining neural activity even in the face of distracting stimuli. Macaques performed a delayed stimulus-response task while activity from PFC neurons was recorded. Many neurons were selective to the specific target stimulus used, and many of these neurons showed increased activity during the delay period, regardless of the other stimuli presented. These results suggest that stimulus representations in the PFC can be robustly maintained despite interference. This robustness may extend to representations of abstract goals or rules within the PFC, supporting the idea that the PFC is capable of maintaining goal-representing patterns of activity – and using these to coordinate cortical connectivity - for as long as the goal is relevant. However, it should be noted that this robustness has not been confirmed for other types of representations than stimulus representations, so this conclusion is as of yet tentative.
The PFC is vital for the proper implementation of rule changes, its neurons encode representations of the rule or goal in effect, and these representations may be robust against interference. These findings are all compatible with the possibility that the PFC is a source of internal-state-based coordination of oscillatory synchrony, and thus communication, across the cortex. Nevertheless, there is another condition to be met, which will be discussed in the next section.
3.3 Coordination of oscillatory activity
In order to qualify as a coordinating source of cortical communication, the PFC needs to entrain coherent oscillatory activity in multiple other brain areas, resulting in enhanced communication between these areas. Unfortunately, research into this topic is severely lacking.
The PFC itself does exhibit neural oscillations that are coherent with those in other task-relevant areas and networks. As reviewed in Benchenane, Tiesinga, & Battaglia (2011), this coherent activity occurs in different frequency bands (notably theta, beta
and gamma), suggesting communication over multiple spatial and temporal scales. The frequencies and brain areas involved vary based on the task requirements. In the sole study reviewed in Benchenane et al. (2011) where the directionality of the activity was determined, gamma-band oscillations appeared earlier in PFC than in other areas dur-ing an attention task, suggestdur-ing that they originated within PFC (Gregoriou, Gotts, Zhou, & Desimone, 2009). A study by Zanto, Rubens, Thangavel, & Gazzaley (2011) investigated the causality more directly. They manipulated part of the PFC using tran-scranial magnetic stimulation (TMS) during an attention task, thus inducing a virtual lesion. They found that attention strengthened coherence between the PFC region and distant task-related regions, but applying a virtual PFC lesion weakened coherence and lead to decreased task performance. These studies together suggest that in at least some cases, the PFC intrinsically generates oscillatory neural activity, which then entrains and maintains synchronous activity in other task-relevant areas – thus leading to coherence between the PFC and those other areas.
Returning to the core question of this section – does the PFC coordinate oscilla-tory activity that is coherent between multiple other brain regions? – it appears that the critical research has simply not yet been done. The PFC appears able to entrain synchrony in other areas and establish coherence between itself and other regions. It may have a coordinating role in information processing, in the sense that entraining oscillations within distant neural populations – particularly gammaband oscillations -could enhance local information processing (Jensen & Mazaheri, 2010). However, cur-rent studies only indicate that the PFC maintains coherence between itself and one other neural population, and not between multiple other populations. To implement dynamic routing, a coordinating source would need to drive synchrony in multiple cortical areas, so as to set them up for reciprocal communication. There is as of yet no evidence that the PFC does so: in the few studies where PFC activity is coherent with activity in multiple other areas, the PFC is not the driving force behind the oscillations (Sehatpour
et al., 2008).
3.4 The PFC’s role in dynamic routing
The PFC meets many of the requirements that would be expected of a source that coordinates neural communication through oscillatory activity. It possesses extensive anatomical connections to other brain structures and areas, providing it with the means to affect neural activity in those areas. Lesion studies show that the PFC is necessary for proper top-down control and flexible behavioral responses. Its neural activity patterns encode abstract rules, and it is capable of maintaining activity patterns even in the face of distractions.
However, the role it plays in directly affecting cortical oscillatory synchrony is unclear. The PFC does exhibit neural oscillatory activity that is coherent with that in other brain areas, and drives this activity in at least some cases. Unfortunately, there is no experimental evidence either for or against the possibility that it coordinates CTC between multiple other areas besides itself. The critical experiments to either confirm or deny this have either not been performed yet or have not been published. Possible underlying reasons for the current state of research will be considered in the discussion section of this thesis. For now, we will simply have to refrain from drawing a conclusion about the PFC’s capability of coordinating cortical communication.
The PFC is nevertheless not the only candidate area for the coordination of dynamic routing. Evidence for the involvement of another brain structure – the pulvinar nucleus of the thalamus – will be reviewed in the following section.
4
The pulvinar nucleus of the thalamus
The pulvinar is a nucleus of the thalamus, a subcortical structure situated fairly centrally in the brain. Several authors have noted that it is anatomically suitably located to
modulate cortical activity and have argued that this could be its main function (Shipp, 2004; Sherman, 2007; Saalmann, Pinsk, Wang, Li, & Kastner, 2012). The pulvinar is generally identified as a spatial salience processing region, particularly with regard to the visual system, although Shipp (2003) notes that it also interacts with other sensory systems.
Since the pulvinar is a deep-seated structure, the oscillations that its neurons gen-erate can only be studied using invasive methods. The amount of available information regarding its oscillatory properties is therefore limited; even more so when it comes to studies in humans rather than animals. As such, the evidence reviewed in this section is relatively sparse.
4.1 Anatomical connections
The pulvinar is extensively interconnected with the cortex, in such a mapping that the cortico-pulvino-cortical connections generally mimic the existing cortico-cortical inter-connections (Shipp, 2003; Leh, Chakravarty, & Ptito, 2008). Shipp (2003) even notes that the pulvinar can, for the sake of illustration, be imagined as the seventh layer of the cortex. Ergo, for any two connected cortical areas, the pulvinar tends to connect to them both. Based on this connectivity, the pulvinar could plausibly function as an additional relay to the direct cortico-cortical communication structures. However, Shipp (2003) points out that this is improbable on account of the marked differences between the cortico-cortical and cortico-pulvino-cortical mapping and instead proposes that the pulvinar plays a coordinatory role for cortical activity. The pulvinar is ideally connected for a scenario in which it precisely biases oscillatory activity in multiple cortical regions, in order to establish enhanced cortico-cortical communication between those regions.
While processing visual salience is considered the pulvinar’s main role, its broad cortical connections suggest that it may do more than that. This idea is supported by the fact that there does not appear to be a non-visual analogue of the pulvinar.
4.2 Function and informational content
The pulvinar plays a role in various aspects of visual attention, such as distractor filtering and visual search (Strumpf et al., 2013). Pulvinar lesions typically result in impairments in these processes. In a study by Snow, Allen, Rafal, & Humphreys (2009), for example, lesions of the ventral pulvinar lead to deficits in distraction filtering in humans. Among pulvinar lesion patients, the more salient stimulus was processed better than the one that was more goal-relevant, leading to a drop in task performance. If the most goal-relevant stimulus was also the most salient one, task performance was normal. Results from this study illustrate a role for the pulvinar in the implementation of top-down spatial attention.
As with the PFC, in order for the pulvinar to guide information flow based on internal states, knowledge, beliefs, and so on – i.e., top-down information – its neural activity has to encode this information. Pulvinar neural responses have in monkeys been linked to such properties as stimulus motion, the presence of face-like stimuli, and confidence in visual decisions (Robinson & Petersen, 1985; Nguyen, Hori, Matsumoto, Tran, Ono, & Nishijo, 2013; Komura, Nikkuni, Hirashima, Uetake, & Miyamoto, 2013). Stimulus motion is an entirely sense-dependent, and thus bottom-up, quality. Sensitivity to the presence of face-like stimuli requires both top-down and bottom-up information, to determine first what a face looks like and second whether the presented stimulus matches the properties of a face. Confidence in visual decisions requires both the bottom-up evidence and the top-down notion of the decision to be made (e.g.“Are all these dots black?” will be answered with different confidence than “Are all these dots moving in the same direction?” for the same evidence). In addition, Saalmann, Pinsk, Wang, Li, & Kastner (2012) found that pulvinar neural responses encode the locus of visual attention, in the sense that synchrony increases among the neurons responding to the attended receptive field. This locus of visual attention, if not inherently salient, can only be extracted from top-down information.
These studies together provide evidence that the pulvinar encodes top-down infor-mation as well as bottom-up inforinfor-mation. The former is a prerequisite for a structure that biases neural communication based on internal states. However, all properties that it has been found to encode are related to spatial attention and visual stimulus qual-ities, and there is a marked lack of research into potential non-visual functions of the pulvinar. More research is necessary to investigate the possibility that the information that it encodes, and thus its range of functions, extends beyond the strictly visual and spatial.
4.3 Coordination of oscillatory activity
As is evident from the described lesion studies, the pulvinar is heavily involved in the proper direction of spatial attention. This spatial attention could arise through precise neural synchrony in the visual system, as in the familiar example of a monkey attending one stimulus in favor of another.
The thalamus – and by extension the pulvinar – seem suited to the generation of oscillatory activity. Shipp (2003) noted that cortico-thalamic loops could operate as a resonant oscillatory system, pointing to sleep research (Steriade, 2000) as a demonstra-tion of this system’s tendency to synchronize neural activity in other areas. The pulvinar could modulate spatial attention in visual areas by driving coherent oscillations there. Shipp (2004) suggested that the pulvinar specifically increases synchrony between areas that represent visual information at different spatial scales and levels of detail, so that the same locus is attended and enhanced in all visual maps.
Saalmann et al. (2012) set out to test this theory. They specifically considered the influence of pulvinar activity on the activity in areas V4 and the temporal-occipital area (TEO), which are interconnected and process visual information sequentially. These areas can be thought of as adjacent visual maps. The research team recorded the local field potential (LFP) within macaque pulvinar, visual map V4, and visual map TEO
during a cued selective attention task. When the attentional cue was shown, synchrony among pulvinar neurons sensitive to the cue location increased. This synchrony remained increased during the delay between cue and target, i.e. for the duration of the selec-tive attention to that location. During this time, synchrony between pulvinar–V4 and pulvinar–TEO increased for those neurons sensitive to the same location on the visual map. This also increased coherence between V4 and TEO, though Granger causality analysis showed that the pulvinar was the driving force of the oscillatory activity. The authors concluded that pulvino-cortical biasing of neural activity was responsible for sustained selective attention in the absence of visual stimuli.
The results found here indicate that the pulvinar indeed selectively synchronizes oscillatory activity across spatial maps in visual cortex, in an attention-driven manner. The structure therefore appears to implement dynamic routing of cortical activity to a degree. However, the study by Saalmann et al. (2012) is the only study so far to directly address this theory. It is as of yet unclear whether this extends to other cortical areas and other processes than selective visual attention. More research is needed to determine whether this is the case.
4.4 The pulvinar’s role in dynamic routing
The pulvinar is anatomically well-connected to bias information flow across the cortex, and it has been shown to do so in visual cortex under conditions of selective visual attention. It appears to tie together several maps of visual representation, so that the neural populations representing the locus of attention oscillate coherently and as such can exchange information effectively. Despite the pulvinar’s broad connectivity, it is commonly believed to engage in visual processing only; there is no evidence that it engages in dynamic routing in areas beyond the visual cortex; and there is no evidence that it makes use of more top-down information than simply the locus of attention and the stimulus properties to be attended. Although more research aimed at uncovering
potential non-visual roles for the pulvinar could change this, the PFC currently has an edge over the pulvinar as a candidate area for the coordination of dynamic routing, by virtue of having been shown to perform a broader range of functions.
In the next section, evidence for the roles of the PFC and the pulvinar will be combined, and a potential new view of the coordination of dynamic routing will be suggested.
5
Conclusion and future directions
Top-down control – the mediation of actions by internal states, goals, beliefs, knowledge, and intentions – is implemented by dynamic routing, i.e. the alteration of effective connectivity between brain areas. Since brain areas can’t autonomously “know” to establish effective communication with other areas in subservience of a larger goal, a coordinating brain structure is needed. Evidence for possible coordinating roles for the prefrontal cortex and the pulvinar nucleus of the thalamus has been presented in this thesis.
The prefrontal cortex (PFC) meets a number of the requirements for such a coordi-nating area: it is extensively anatomically connected, and part of its neurons encode the top-down information necessary to inform actions. In addition, it possesses a number of non-critical features that would be useful for a structure with a coordinating function, such as the fact that its neural activity patterns are robust in the face of interference.
Its role in coordinating oscillatory coherence is less clear. It has been shown to drive coherence between itself and other areas, but not between other areas per se. In studies where activity within the PFC is coherent with multiple other areas, the PFC has not been identified as the originating source. This is not to say with certainty that the PFC cannot establish communication through coherence between two areas; the necessary studies to determine whether this is the case have simply not yet been performed. While
some studies do investigate the oscillatory coherence between PFC and multiple other brain regions, very few of them report directionality measures such as Granger causality; presumably because it is difficult to get robust results. A shift to from correlational to causal study designs, in which the task-relevant network is actively perturbed, could shed light on the PFC’s capability of driving oscillatory coherence in multiple areas.
Like the PFC, the pulvinar nucleus of the thalamus appears suited to a coordinating role in some aspects, but not others. It is broadly anatomically connected, it is ideally situated to drive coherence in multiple cortical areas to enhance direct cortico-cortical communication between those areas, and it has in fact been shown to do so in visual areas. However, the pulvinar is classically considered a spatial salience map rather than a region implementing top-down control, and the research reviewed here does nothing to disprove that: its neurons have only been shown to encode information regarding visual decision-making and salience, and it has only been found to drive oscillatory activity within visual areas. For these reasons, the PFC has an edge over the pulvinar as a potential coordinator of dynamic routing – although it should be noted that future research could uncover potential non-visual roles for the pulvinar.
5.1 The current state of research
The evidence reviewed in this thesis is not strongly conclusive in favor of either the PFC or the pulvinar as a coordinating source of oscillatory coherence. I believe that this can, in part, be traced to two particular problems with the current state of research into this topic.
One is that the majority of studies are correlational rather than causal; that is, they determine the relationship between two measurements (often oscillatory neural activity in two brain areas) rather than manipulating one and determining how it affects the other. In correlational studies, identifying the origin of oscillatory activity is complicated at best and impossible at worst. Nevertheless, many researchers opt for this type of
study. While there are entirely valid reasons to do so, this leaves the question of the origins of synchronous neural activity somewhat neglected. Such neglect is unjustified. Coherence across multiple regions in a goal-oriented, top-down-driven fashion – as seen in many of the studies here – cannot arise without coordination. The identification of sources of coherence can serve as a stepping stone for further understanding of cognition, in the sense that we could better explore the possibilities and limitations of information exchange and processing in brain networks. Hopefully, this question will gain greater visibility in neuroscientific research in the coming years, leading to a shift towards causal study designs.
A second concern is that there is little research that challenges the status quo. Re-search regarding the pulvinar is a prime example. The pulvinar has been established to play a role in mediating visuospatial attention. Therefore, its role in coordinating oscillatory coherence has predominantly been studied in visual areas, and the properties that its neural firing patterns encode have predominantly been investigated in terms of visual tasks and stimuli, despite its extensive cortical connectivity. Preciously little information regarding non-visual properties of the pulvinar has been published, making it extremely difficult to conclude whether its function is limited to what it is commonly believed to do. This could be an artifact of scientific journals’ tendency to publish pos-itive results in favor of negative results, or it could signify that the research in question is not being performed. To researchers interested in the field, there is no way to discern the difference, though both are problematic and should be rectified in the future.
5.2 A new view: distributed network hubs
As there is no time better than the present to challenge the status quo, I hereby present a new approach to the question discussed in this thesis. There are some points of interest that suggest that the search for an area that coordinates dynamic routing should be conducted in a different way. In fact, by looking for a single source of coherence, we may
be asking the wrong question altogether.
While both the pulvinar and the PFC meet some of the requirements, neither of them appear to qualify as the only source of oscillatory coherence across multiple brain areas. Neither the PFC nor the pulvinar are involved in all cases of interareal coherence, and there are cases where such coherence arises without involvement of either of these two structures, such as between the hippocampus and amygdala (Rutishauser et al., 2010). This raises the possibility that there are many more of these areas that orches-trate coherence in a limited section of the brain. In fact, such a role has not only been suggested for the PFC and pulvinar, but also for the reticular nucleus of the thalamus, the anterior insula, the superior colliculus, the amygdala, and the striatum (Shipp, 2004; Senkowski, Schneider, Foxe, & Engel, 2008; Bressler & Menon, 2010; Daitch, Sharma, Roland, Astafiev, Bundy, Gaona, Snyder, Shulman, Leuthardt, & Corbetta, 2013). Per-haps we should move away from the idea of strictly separate modules coordinating co-herence, and towards the idea that many interconnected “network nodes” or “network hubs” each bias coherence in part of the brain.
The suggestion that there are many different “hubs” that enhance effective connec-tivity in the brain fits well with the fact that the brain is organized like a small-world network (Rubinov & Sporns, 2010). This term, originating from network theory, indi-cates a network that has many short-range connections and few longer ones. This means that there are many extensively connected hub regions that, in turn, tend to connect to each other over longer ranges. Such a formation is frugal in its connectivity, meaning that it saves on wiring (in this case, axons and dendrites) and decreases transmission time as compared to a randomly connected network or various other types of networks. Both functional and anatomical networks within the brain exhibit small-world proper-ties (Rubinov & Sporns, 2010; Stam, 2010). If hubs found in the brain have a tendency to coordinate neural synchrony, this would be a very efficient way to dynamically route information across and between larger brain networks.
Bressler & Menon (2010) define a network as a set of interconnected brain areas that collectively perform a function. They review evidence for the existence of several largely dissociable brain networks, characterized by synchrony between task-relevant areas. They explicitly note that brain areas can contribute to the functioning of the network in different ways: through information processing, or through coordinating the engagement of other areas. They dub brain areas of the latter type “controllers”. It should be clear that the elusive sources of oscillatory synchrony discussed throughout this thesis would qualify as such controllers.
Because communication through coherence (CTC) is selective, many different task-relevant networks in the brain could be engaged by “controllers” at the same time without interfering with each other. In fact, there are indications that different large-scale brain networks communicate at dissociable frequency bands (Ba¸sar, Ba¸sar-Eroglu, Karaka¸s, & Sch¨urmann, 2001; Fan, Byrne, Worden, Guise, McCandliss, Fossella, & Posner, 2007; Siegel, Donner, Oostenveld, Fries, & Engel, 2008; Benchenane, Tiesinga, & Battaglia, 2011; Hipp, Engel, & Siegel, 2011).
All of this suggests a new way of looking at the coordination of oscillatory coherence in the brain: the coordinating structure may not be a single entity, but rather a variety of hubs, possibly interconnected in a single network, and all interdependently driving neural coherence across other task-relevant networks. Although compelling in its efficiency, it is difficult to determine whether this is truly the case given the current state of research. Many studies consider coherence between preselected areas rather than across large-scale networks, and the studies that investigate large-scale networks tend to be correlational rather than casual, obscuring the sources that entrain neural synchrony.
In order to determine whether neural coherence across large-scale brain networks could be driven by multiple hubs, we first need to exclude the possibility that the brain contains a single coordinating structure for internal state-driven coherence. Unfortu-nately, there are many difficulties inherent in designing studies to identify a single source
of neural synchrony, let alone identify multiple possible hubs. At first consideration, it would seem relatively easy to introduce various internal states in a subject, keep the external world constant, and determine whether there is a single brain structure that drives the oscillations in task-relevant networks in all those states. However, it would be difficult to reliably manipulate internal states that go beyond basic goals and knowl-edge. Even if this could be done, it would be a challenge to select external stimuli that not only run the gamut of sensory modalities (making sure that we do not just iden-tify a visual or auditory coordinating structure of coherence), but that are repeatable from one experiment to the next. In addition, the measurement techniques used would either have to be invasive and spatially limited at pre-selected locations (using micro-electrodes, thereby precluding human experimentation), or result in spatially diffuse and entirely correlational data from which the origin of oscillatory activity cannot reliably be extracted (using EEG). While theoretically interesting, it seems currently practically unattainable to determine whether one brain structure or multiple distributed hubs are responsible for the coordination of neural coherence, given that we would have to test this in a range of situations that is representative for daily life.
5.3 Closing remarks
To conclude, this thesis provides an overview of current evidence regarding two suggested sources of oscillatory synchrony and the coordination of dynamic routing in the brain: the prefrontal cortex and the pulvinar nucleus of the thalamus. Based on current data, the prefrontal cortex seems slightly more suited as a coordinating structure, though evidence for both is lacking and there are several more key experiments to be performed to draw a strong conclusion. Understanding how neural coherence and thus communication is centrally coordinated could be vital in the understanding of both normal and disrupted patterns of coherence, the latter seen in psychopathologies such as autism, schizophrenia, and Alzheimer’s disease (Uhlhaas et al., 2006). I hereby express hope that the question
of coordination of coherence within the brain will gain more visibility and attention as time goes on.
References
Akam, T. E., & Kullmann, D. M. (2012). Efficient ”communication through coher-ence” requires oscillations structured to minimize interference between signals. PLoS Computational Biology, 8 (11), 1–15.
Ba¸sar, E., Ba¸sar-Eroglu, C., Karaka¸s, S., & Sch¨urmann, M. (2001). Gamma, alpha, delta, and theta oscillations govern cognitive processes. International Journal of Psy-chophysiology, 39 (2-3), 241–8.
Benchenane, K., Tiesinga, P. H., & Battaglia, F. P. (2011). Oscillations in the prefrontal cortex: a gateway to memory and attention. Current Opinion in Neurobiology, 21 (3), 475–85.
Bressler, S. L., & Menon, V. (2010). Large-scale brain networks in cognition: emerging methods and principles. Trends in Cognitive Sciences, 14 (6), 277–90.
Buzs´aki, G., & Draguhn, A. (2004). Neuronal oscillations in cortical networks. Science, 304 (5679), 1926–9.
Daitch, A. L., Sharma, M., Roland, J. L., Astafiev, S. V., Bundy, D. T., Gaona, C. M., Snyder, A. Z., Shulman, G. L., Leuthardt, E. C., & Corbetta, M. (2013). Frequency-specific mechanism links human brain networks for spatial attention. Proceedings of the National Academy of Sciences, 110 (48), 19585–90.
Fan, J., Byrne, J., Worden, M. S., Guise, K. G., McCandliss, B. D., Fossella, J., & Posner, M. I. (2007). The relation of brain oscillations to attentional networks. The Journal of Neuroscience, 27 (23), 6197–206.
Fries, P. (2005). A mechanism for cognitive dynamics: neuronal communication through neuronal coherence. Trends in Cognitive Sciences, 9 (10), 474–80.
Gregoriou, G. G., Gotts, S. J., Zhou, H., & Desimone, R. (2009). High-frequency, long-range coupling between prefrontal and visual cortex during attention. Science, 324 (5931), 1207–10.
Haegens, S., N´acher, V., Hern´andez, A., Luna, R., Jensen, O., & Romo, R. (2011). Beta oscillations in the monkey sensorimotor network reflect somatosensory decision making. Proceedings of the National Academy of Sciences, 108 (26), 1–6.
Hipp, J. F., Engel, A. K., & Siegel, M. (2011). Oscillatory synchronization in large-scale cortical networks predicts perception. Neuron, 69 (2), 387–96.
Jensen, O., & Mazaheri, A. (2010). Shaping functional architecture by oscillatory alpha activity: gating by inhibition. Frontiers in Human Neuroscience, 4 (186), 1–8. Komura, Y., Nikkuni, A., Hirashima, N., Uetake, T., & Miyamoto, A. (2013). Responses
of pulvinar neurons reflect a subject’s confidence in visual categorization. Nature Neuroscience, 16 (6), 749–55.
Leh, S. E., Chakravarty, M. M., & Ptito, A. (2008). The connectivity of the human pulvinar: a diffusion tensor imaging tractography study. International Journal of Biomedical Imaging, 2008 , 789539.
Liebe, S., Hoerzer, G. M., Logothetis, N. K., & Rainer, G. (2012). Theta coupling between V4 and prefrontal cortex predicts visual short-term memory performance. Nature Neuroscience, 15 (3), 456–62.
Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex function. Annual Review of Neuroscience, 24 , 167–202.
Miller, E. K., Erickson, C. A., & Desimone, R. (1996). Neural mechanisms of visual working memory in prefrontal cortex of the macaque. The Journal of Neuroscience, 16 (16), 5154–67.
Milner, B. (1963). Effects of Different Brain Lesions on Card Sorting.
Muhammad, R., Wallis, J. D., & Miller, E. K. (2006). A comparison of abstract rules in the prefrontal cortex, premotor cortex, inferior temporal cortex, and striatum. Journal of Cognitive Neuroscience, 18 (6), 974–89.
Nguyen, M. N., Hori, E., Matsumoto, J., Tran, A. H., Ono, T., & Nishijo, H. (2013). Neuronal responses to face-like stimuli in the monkey pulvinar. The European Journal of Neuroscience, 37 (1), 35–51.
Palva, J. M., Monto, S., Kulashekhar, S., & Palva, S. (2010). Neuronal synchrony reveals working memory networks and predicts individual memory capacity. Proceedings of the National Academy of Sciences, 107 (16), 7580–5.
Plankar, M., Breˇzan, S., & Jerman, I. (2013). The principle of coherence in multi-level brain information processing. Progress in Biophysics and Molecular Biology, 111 (1), 8–29.
Robinson, A. L., Heaton, R. K., Lehman, R. A., & Stilson, D. W. (1980). The utility of the Wisconsin Card Sorting Test in detecting and localizing frontal lobe lesions. Journal of Consulting and Clinical Psychology, 48 (5), 605–14.
Robinson, D. L., & Petersen, S. E. (1985). Responses of pulvinar neurons to real and self-induced stimulus movement. Brain Research, 338 (2), 392–4.
Rubinov, M., & Sporns, O. (2010). Complex network measures of brain connectivity: uses and interpretations. NeuroImage, 52 (3), 1059–69.
Rutishauser, U., Ross, I. B., Mamelak, A. N., & Schuman, E. M. (2010). Human memory strength is predicted by theta-frequency phase-locking of single neurons. Nature, 464 (7290), 903–7.
Saalmann, Y. B., Pinsk, M. a., Wang, L., Li, X., & Kastner, S. (2012). The pulvinar regulates information transmission between cortical areas based on attention demands. Science, 337 (6095), 753–6.
Schnitzler, A., & Gross, J. (2005). Normal and pathological oscillatory communication in the brain. Nature Reviews Neuroscience, 6 (4), 285–96.
Sehatpour, P., Molholm, S., Schwartz, T. H., Mahoney, J. R., Mehta, A. D., Javitt, D. C., Stanton, P. K., & Foxe, J. J. (2008). A human intracranial study of long-range oscillatory coherence across a frontal–occipital–hippocampal brain network during vi-sual object processing. Proceedings of the National Academy of Sciences, 105 (11), 4399–4404.
Senkowski, D., Schneider, T. R., Foxe, J. J., & Engel, A. K. (2008). Crossmodal bind-ing through neural coherence: implications for multisensory processbind-ing. Trends in Neurosciences, 31 (8), 401–9.
Sherman, S. M. (2007). The thalamus is more than just a relay. Current Opinion in Neurobiology, 17 (4), 417–22.
Shipp, S. (2003). The functional logic of cortico-pulvinar connections. Philosophical transactions of the Royal Society B: Biological Sciences, 358 (1438), 1605–24.
Shipp, S. (2004). The brain circuitry of attention. Trends in Cognitive Sciences, 8 (5), 223–30.
Siegel, M., Donner, T. H., & Engel, A. K. (2012). Spectral fingerprints of large-scale neuronal interactions. Nature Reviews Neuroscience, 13 (2), 121–34.
Siegel, M., Donner, T. H., Oostenveld, R., Fries, P., & Engel, A. K. (2008). Neuronal synchronization along the dorsal visual pathway reflects the focus of spatial attention. Neuron, 60 (4), 709–19.
Snow, J. C., Allen, H. A., Rafal, R. D., & Humphreys, G. W. (2009). Impaired attentional selection following lesions to human pulvinar: evidence for homology between human and monkey. Proceedings of the National Academy of Sciences, 106 (10), 4054–9.
Stam, C. J. (2010). Characterization of anatomical and functional connectivity in the brain: a complex networks perspective. International Journal of Psychophysiology , 77 (3), 186–94.
Stam, C. J., & van Straaten, E. C. W. (2012). The organization of physiological brain networks. Clinical Neurophysiology, 123 (6), 1067–87.
Steriade, M. (2000). Corticothalamic resonance, states of vigilance and mentation. Neu-roscience, 101 (2), 243–276.
Strumpf, H., Mangun, G. R., Boehler, C. N., Stoppel, C., Schoenfeld, M. A., Heinze, H.-J., & Hopf, J.-M. (2013). The role of the pulvinar in distractor processing and visual search. Human Brain Mapping, 34 (5), 1115–32.
Stuss, D. T., Floden, D., Alexander, M. P., Levine, B., & Katz, D. (2001). Stroop performance in focal lesion patients: dissociation of processes and frontal lobe lesion location. Neuropsychologia, 39 (8), 771–86.
Uhlhaas, P. J., Linden, D. E. J., Singer, W., Haenschel, C., Lindner, M., Maurer, K., & Rodriguez, E. (2006). Dysfunctional long-range coordination of neural activity during Gestalt perception in schizophrenia. The Journal of Neuroscience, 26 (31), 8168–75.
von Stein, A., & Sarnthein, J. (2000). Different frequencies for different scales of cortical integration: from local gamma to long range alpha/theta synchronization. Interna-tional Journal of Psychophysiology, 38 (3), 301–13.
Wallis, J. D., Anderson, K. C., & Miller, E. K. (2001). Single neurons in prefrontal cortex encode abstract rules. Nature, 411 (6840), 953–6.
Zanto, T. P., Rubens, M. T., Thangavel, A., & Gazzaley, A. (2011). Causal role of the prefrontal cortex in top-down modulation of visual processing and working memory. Nature Neuroscience, 14 (5), 656–61.
