Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Oct 1.
Published in final edited form as: Curr Opin Neurobiol. 2022 Jul 20;76:102606. doi: 10.1016/j.conb.2022.102606

A Goldilocks Theory of Cognitive Control: Balancing Precision and Efficiency with Low-dimensional Control States

Camden J MacDowell 1,2, Sina Tafazoli 1, Timothy J Buschman 1,3,*
PMCID: PMC9653176  NIHMSID: NIHMS1846644  PMID: 35870301

Abstract

Cognitive control orchestrates interactions between brain regions, guiding the transformation of information to support contextually-appropriate and goal-directed behaviors. In this review, we propose a hierarchical model of cognitive control where low-dimensional control states direct the flow of high-dimensional representations between regions. This allows cognitive control to flexibly adapt to new environments and maintain the representational capacity to capture the richness of the world.

Introduction

The world is rich; we are faced with a wide variety of sensory inputs and are capable of performing a myriad of different actions. Cognitive control brings order to this chaos, using our goals to focus on relevant information and guide our actions. To accomplish this, cognitive control creates a control state that directs the flow of information through the brain in a way that supports the desired behavior [1]. For example, both hitting a baseball and studying for a test require sensory processing. The control state defines how these sensory representations interact with other regions; hitting a baseball requires routing sensory information about the velocity and spin of the ball to and from motor regions, while studying requires routing orthographic sensory information to and from associative and memory regions. Several neural mechanisms have been proposed to direct the flow of neural activity through the brain, including gain modulation, synchronous oscillations, neuromodulators, and cortico-basal ganglia-thalamo-cortical loops (see Box 1 for details). In this way, cognitive control can ensure goal-relevant information propagates between functionally diverse brain regions, supporting behavior.

Box 1: Mechanisms of cognitive control.

Cognitive control is the ability to direct one’s behavior towards a goal. Goal representations in prefrontal and parietal cortex, as well as physiological drives represented in midbrain regions, act on the rest of the brain to alter how information propagates through (and is processed by) the brain. This defines the control state that supports the desired behavior (Fig. 1A). In this box, we highlight several neural mechanisms that have been proposed to support cognitive control by flexibly routing information between brain regions.

Synchronous oscillations:

Synchronous oscillations are thought to modulate the effective connectivity of two neural populations [49]. When two populations are synchronized, neural activity in one population is more effective at driving responses in the other region. In this way, synchrony can facilitate the communication of information between the two populations. When unsynchronized, communication is reduced. Importantly, because synchrony only requires changing the pattern of neural activity (and not restructuring synaptic weights), it could provide a mechanism for flexibly changing how information is routed through the brain. In this way, different control states could be established as different patterns of synchrony across the brain (Fig. 1B). In support of this hypothesis, attending to a stimulus (i.e., selectively processing it), increases the synchrony of neurons encoding the stimulus, both within and between regions [49].

Neuromodulation:

By altering how neural populations respond to inputs, neuromodulators may provide a mechanism for guiding the flow of neural activity through the brain (Fig. 1C). For example, top-down modulation of acetylcholine (ACh) is thought to support the selective enhancement of attended stimuli (and the suppression of distractors) [50]. Mechanistically, this is thought to be due to ACh increasing feed-forward connections while suppressing recurrent/feedback connections. In this way, the distribution of neuromodulators across cortex could insatiate the control state. Indeed, recent work using fluorescent indicators of ACh concentration has shown spatiotemporal fluctuations in cholinergic signaling that are associated with distinct behavioral and cortical states [44••]. Dopamine (DA) and Norepinephrine (NE) have also been hypothesized to allow the network to explore a broader diversity of states [51].

Cortico-basal ganglia-thalamo-cortical loops:

Recurrent connections between cortex and thalamus (Th) help sustain representations within cortex, and facilitate the transfer of information between regions [52]. The basal ganglia (BG) are anatomically well situated to control these cortico-thalamic loops; BG receives convergent projections from cortex and provides inhibitory control over the associated thalamus. In this way, neural representations in the basal ganglia may encode the control state, acting to ‘gate’ what information is represented in cortex and how that information flows between regions (Fig. 1D). For example, such gating is thought to control what representations are held in working memory[53–55].

Finally, it is important to note that these different mechanisms are not mutually exclusive. They likely all play a role in establishing a control state and supporting cognitive behaviors.

In addition to being rich, the world is dynamic – our environment and goals can change rapidly, and cognitive control must adapt to these changes. When in a familiar environment, cognitive control can use a previously learned control state to reinstate an optimal flow of neural activity. However, when faced with a new situation, cognitive control must learn the control state that is most appropriate for the situation.

In a rich and dynamic world, cognitive control must balance precision and flexibility. As we review below, the richness of the world is captured in high-dimensional neural representations. Given this, one might expect control to be similarly high-dimensional, with a large number of control states that can precisely guide neural activity in support of a diverse set of behaviors. However, learning such high-dimensional, precise, control states is difficult, making it hard to adapt to new situations. Instead, we propose the brain uses a limited, low-dimensional, set of control states. Each control state defines the general flow of neural activity between regions, carrying along the high-dimensional representations that represent information relevant to a particular task. Limiting the number of control states comes at the cost of imperfect, imprecise, control but facilitates learning in new environments. In this way, cognitive control finds the ‘just right’ balance between being able to control the detailed representations that capture the richness of the world and maintaining the flexibility needed to adapt to a changing world.

Neural populations have the capacity to represent detailed information in high-dimensional representations

Recent advances in large-scale electrophysiology and two-photon imaging has allowed researchers to record the activity of large populations of neurons. This has shown representations within neural populations are high-dimensional, capturing the details of stimuli and motor actions [2•–5••]. For example, populations of neurons in rodent cortex were found to contain hundreds of dimensions, many of which represented the minutiae of movements, such as facial twitches, nose position, locomotion, and patterns in whisker position [2•]. High-dimensional representations of movement were also found in subcortical areas, including striatum, thalamus, and hippocampus; both during spontaneous behavior and while viewing visual stimuli.

Similarly, high-dimensional representations encode the details of sensory information. Recordings from the somatosensory cortex of non-human primates found that individual neurons exhibited idiosyncratic responses to the specifics of tactile stimuli, such as its roughness, hardness, stickiness, and coolness [5••]. Across the neural population, this created a high-dimensional representation that could distinguish the details of natural textures.

Likewise, prefrontal cortex uses a high-dimensional representation to encode arbitrary combinations of task-relevant information [6]. Individual prefrontal neurons show ‘non-linear mixed selectivity’, responding to a conjunction of stimuli, task cues, and responses. This results in a high-dimensional representation across the neural population, which is useful for decoding arbitrary task variables [7]. While originally observed in prefrontal cortex, complex representations of task details have also been observed in the basolateral amygdala, prelimbic, and medial prefrontal cortex in mice [8–10].

It is important to note that these analyses likely underestimate the true dimensionality of representations. The measured dimensionality of a neural population scales with number of neurons recorded [2•] and the number of unique stimuli presented [3], suggesting that constraints on the reported dimensionality of neural populations are often a consequence of small stimulus sets and limited experimental time. Similarly, the spatial scale of neural recording can constrain the estimated dimensionality; recording methods such as fMRI and bulk calcium imaging capture the summed activity of large population of neurons which could place an upper bound on the estimated dimensionality [11].

Altogether, these results suggest neural representations are high-dimensional, providing the representational capacity to represent the rich diversity of stimuli, actions, and goals [7]. However, the high-dimensional, precise, nature of these representations could create a problem for cognitive control. If cognitive control must engage a specific detailed representation, then the large number of high-dimensional representations requires a similarly large number of control states.

The tradeoff between dimensionality, precision, and efficiency of control

To illustrate this problem, consider the challenge of entering a novel situation in which the correct behavior is not known a priori. In this case, the brain must discover the control state that supports the behavior most appropriate for the situation. Similarly, even when one has previously learned the control state for a situation, the dynamic nature of the world means it is advantageous to occasionally explore alternative behaviors that may yield improved outcomes [12].

One way cognitive control could optimize behavior is to have a high-dimensional set of control states that can precisely guide neural activity (Fig. 2A). The advantage of such precise control states is that they can optimally guide information flow, maximizing behavioral outcomes. However, precision also carries significant computational and physiological costs. In particular, precise control states take longer to learn. The high-dimensional nature of precise control states means, by definition, there is a large library of potential control states from which to choose. Therefore, finding the best control state in this set requires repeatedly sampling different potential control states, estimating their fit to the situation, and then deciding whether to continue searching or use the current state. This sampling process takes time, meaning there is an opportunity cost when searching for the best control state from a library of high-dimensional, precise, control states.

Figure 2. The tradeoff between precision and cost of control.

Figure 2.

Open in a new tab

(A) Schematic of high-dimensional control in which precise control representations recruit specific neurons (circles) within distributed neural populations (colors) in order to support a specific behavior. (B) Schematic of low-dimensional control in which control mechanisms broadly engage neural populations to facilitate behavior. In this model, control is agnostic to the nuances of the detailed representation. Rather, control facilitates the broader flow of information across neural populations, carrying along the details of behavior in the local high-dimensional representations. (C) A simple model of control demonstrates how increasing the number of control states decreases the asymptotic error in control (taken as the difference between the best control state and the ideal control state). However, increasing the number of control states also slows the identification of the optimal control state.

In contrast, if control states are restricted to a low-dimensional set of possible patterns, then it will be easier to identify the best pattern in the restricted set. However, these low-dimensional control states lack the precision needed to optimally guide the flow of neural activity between regions (Fig. 2B). And so, the best control state will likely only roughly match the situation. This imprecision means that even the best control state from the set will be suboptimal, limiting how well one can perform a particular behavior (note: this is true, even when the set of control patterns is chosen in a way that maximizes the expected performance across all situations and behaviors).

This leads to a tension between high and low dimensional representations, as schematized in Figure 2C. A high-dimensional set of control states captures the richness of behavior, allowing one’s eventual behavioral output to be near perfect at the asymptote. But this comes with a high cost for finding the optimal pattern, meaning it takes longer to reduce the error between the current control state and an ideal control state (Fig. 2C, darker colors). In contrast, one can rapidly find the best control state from a low-dimensional set, but, on average, this is going to result in a greater asymptotic error (Fig. 2C, lighter colors). This argues for a goldilocks theory of cognitive control, such that the number of control states available reflects a ‘just right’ level of representational detail that minimizes the sum of the cost of suboptimal behaviors and the opportunity costs incurred when searching. Given the high opportunity costs of most environments, this suggests that the brain will tend to settle on a relatively low-dimensional set of control states.

Evidence for a limited dimensionality of control states in cortex-wide neural activity

Recent work has provided evidence for low dimensional control states. Using mesoscale calcium imaging (Fig. 3A; [13]), we recently showed that the moment-to-moment flow of neural activity across dorsal cortex can be explained by a surprisingly small set of ‘motifs’ ([11]; Fig. 3B). Each motif captured a unique spatiotemporal pattern of neural activity. For example, one motif captured a burst of neural activity in visual cortex associated with processing a visual stimulus (Fig. 3C). Another motif captured a traveling wave that propagated from anterior to posterior cortex (Fig. 3D; [14]), which is thought to reflect integrative processes across cortical areas and has been observed while at rest, in social contexts, and even when under anesthesia [11,15–20••]. Finally, a third motif captured a bilateral burst of activity in limb somatomotor and anterior parietal cortices, which has been seen in multiple studies, both during the preparatory period of a sensory-motor task and during spontaneous behavior (Fig. 3E; [11,21–23]).

Figure 3. Low-dimensional control states and cortical motifs.

Figure 3.

Open in a new tab

(A) Left: Schematic depicting the experimental setup for mesoscale imaging in an awake head-fixed mouse (adapted from [11] and [20••]). Fluorescence indicators of neural activity (i.e., calcium or voltage-indicators) in neurons of the dorsal cortex are excited by LEDs. The emitted fluorescence is captured by a CMOS camera at high spatial (~68μm) and temporal (10–30hz) resolution. Right: Mesoscale imaging captures population-level neural activity across functionally diverse regions of the dorsal cortex including sensory, motor, and associative regions. (B) Schematic illustrating how spatiotemporal deconvolution approaches can be used to decompose mesoscale recordings of cortex-wide neural activity into sequential activity over time (bottom) of recurring spatiotemporal ‘motifs’ (top). It is important to note that mesoscale imaging has the resolution to capture detailed neural representations within local cortical areas (e.g., relating to specific stimuli), however, these signal are distinct from the broader, cortex-wide spatiotemporal patterns captured by motifs [11]. (C-D) Example cortex-wide motifs from (adapted from [11]). Individual images on left show snapshots of a motif’s activity over 1 second. Schematic panels on right of how motifs may reflect different control states in the flow of information through the cortex (related to Fig. 2B). (C) Timecourse of motif that captures a burst of activity in visual regions associated with the general processing of visual stimuli. (D) Motif that captures a broad anterior-to-posterior traveling wave in neural activity that sequentially engages motor, sensory, associative, and visual regions. (E) Example comparison between one motif – a localized burst of activity in anterior parietal and limb somatomotor regions – and similar patterns in other studies. Left: the motif observed during spontaneous activity. Right: similar activity observed during the holding period of a sensory-motor task (adapted from [21]) and during motor action in lever-pull tasks (adapted from [27–28]).

Motifs appear to be a general phenomenon. The same motifs were seen across individual animals and in different environments [11,20••]. Furthermore, as noted above, motifs are similar to patterns of neural activity observed in other mesoscale studies (despite differences in experimental paradigms and analyses, [15–27]). This suggests that motifs form a ‘basis set’ of different ways in which neural activity can flow across cortex, with each motif supporting a different cognitive computation, such as sensory processing, sensorimotor integration, or motor action. Consistent with these observations in rodents, decades of work studying functional connectivity in human imaging and electroencephalographic data has found evidence for a limited set of ‘functional networks’ of brain regions. Each network couples together multiple brain regions and is associated with different cognitive functions, such as sensory-motor processing, attention, or executive functions [28–31].

Given these results, one hypothesis is that each motif reflects a different control state. If true, then changing behaviors should change which motifs are expressed [32,33]. This appears to be the case. Both humans and mice cycle between different functional networks of neural populations over time [27,34]. Human functional imaging has shown the brain dynamically transitions between different states of connectivity in a task-specific manner [31,35–37•]. In mice, the frequency with which different cortical motifs occur changes with different behavioral contexts [11], consistent with the idea that adapting to a new environment requires changing control states. Indeed, recent work suggests animals with behavioral dysfunctions show an inability to adapt what motifs they express when changing to a new environment [20••]. Altogether, these results are consistent with a low-dimensional set of control states, each of which establishes a unique motif in dynamics across brain areas to support behavior.

Balancing low-dimensional control states with high-dimensional neural representations

From a normative point of view, low-dimensional control states make sense: they allow the brain to minimize the cost of sub-optimal control states and the time spent searching for an optimal control state. However, high-dimensional representations allow the neural population to encode the richness of the world and one’s thoughts. This begs the question – how does the brain balance between these two needs?

One solution may be that cognitive control is hierarchical [38]. Low-dimensional control states coordinate the broad dynamics of how information is carried across brain regions to enable general cognitive processes (i.e., visual stimulus processing, motor action, or maintenance of task variables). The details of the behavior, such as the identity of a stimulus or the specifics of a task, are represented in the high-dimensional representations of the neural population within a brain region. In this framework, control states only guide the general flow of information between regions, agnostic to the details encoded in the precise, high-dimensional, representations. For example, a control state for sensory-response behaviors can direct the flow of neural activity from visual cortex to associative regions, and then onto motor regions. This broad flow would carry the detailed stimulus information encoded in sensory regions to prefrontal and parietal regions, where it can interact with detailed task representations. This transformed representation is then passed onto motor cortex to generate precise actions. Importantly, these same cortex-wide dynamics are broadly applicable to a wide variety of sensory-response behaviors; different stimuli, tasks, and actions will be carried by detailed representations riding on the same cortex-wide wave of activity. In this way, hierarchical structure would balance efficient and flexible control with preserving the representational capacity of neural populations.

Future Directions

Several questions remain open for future study. For example, what determines the structure of motifs of neural activity? One possibility is that they are defined by the anatomical connectivity of the cortex, which may constrain the flow of neural activity between areas [24,39,40]. Such constraints may be predetermined by evolution or be based on an organism’s experience. Related to this, it is unknown whether motifs can change, and at what timescale. Recent work has shown cortex-wide dynamics change during learning [22,41], which could reflect motifs changing to match a behavior. Yet, future work is needed to understand the mechanisms underlying these changes.

It is also not clear how the diverse neural mechanisms associated with cognitive control, such as changes in gain, synchronous oscillations, or neurotransmitters (Box 1), could support a hierarchical model of cognitive control. One intriguing hypothesis is that these mechanisms act at different scales. For example, synchronizing neural activity may help to bridge across scales, with higher-frequency oscillations supporting the precise modulation of specific representations within a region [42], while lower-frequency oscillations provide support for the broader routing of information across regions [43]. Similarly, recent work has shown neurotransmitters, such as acetylcholine, are released in heterogenous patterns across cortex [44••]. These patterns may reflect low-dimensional control states, engaging networks of brain regions, while also decorrelating neural responses within a region (effectively expanding the dimensionality of the representation).

Finally, research is needed to understand how different sources of control interact to establish a control state. Recent neurophysiological evidence has found that physiologic drives, such as thirst and hunger, are represented in subcortical regions and play a role in guiding dynamics of communication across brain areas [4,45]. How these drives interact with more ‘cerebral’ task-guided control represented in prefrontal and parietal cortex is unknown [46]. Advancements in our capacity to record neural activity across spatiotemporal scales [47] and for long durations [48] will enable deeper investigations into the explicit interplay of these systems, promising exciting new insight into the nature of cognitive control.

Figure 1. Mechanisms of cognitive control.

Figure 1.

Open in a new tab

(A) Schematic of how two different control states guide the flow of neural activity in different ways to support two different behaviors. Control states may act by (B) modulating the synchrony of regions; (C) changing the neuromodulatory tone; or (D) basal ganglia control of cortico-thalamic loops. See Box text for details.

Highlights.

  • Within cortical regions, neural populations use high-dimensional representations to capture the details of behavior

  • Precise control of detailed representations could allow for accurate control of behavior, but would be difficult to learn

  • The brain may use a low-dimensional set of control states that define how neural activity flows between brain regions

  • Using a low-dimensional set of control states may balance between accurate control and efficiency in learning

Acknowledgements

We thank the Buschman lab for their feedback and discussion during the writing of this manuscript. This work was funded by a grant from SFARI 670183 (T.J.B.) and NIH NCATS Award TL1TR003019 (C.J.M).

Footnotes

Conflict of interest statement

Nothing declared

Works Cited

  • 1.Miller EK, Cohen JD: An Integrative Theory of Prefrontal Cortex Function. Annual Review of Neuroscience 2001, 24:167–202. [DOI] [PubMed] [Google Scholar]
  • 2.•.Stringer C, Pachitariu M, Steinmetz N, Reddy CB, Carandini M, Harris KD: Spontaneous behaviors drive multidimensional, brainwide activity. Science 2019, 364:255. [DOI] [PMC free article] [PubMed] [Google Scholar]; Two-photon imaging of neurons in the primary visual cortex of mice coupled with a novel Shared Variance Components Analysis revealed that spontaneous neural population activity contains hundreds of reliable (i.e., non-noise) dimensions. Some of these dimensions represented facial motor activity; other dimensions were related to arousal; while others were reliable, but lacked any measured motor tie-in, suggesting that they may reflect internal cognitive variables. Electrophysiological results across a diversity of cortical and subcortical areas found similar high-dimensional representations in neural population activity.
  • 3.Stringer C, Pachitariu M, Steinmetz N, Carandini M, Harris KD: High-dimensional geometry of population responses in visual cortex. Nature 2019, doi: 10.1038/s41586-019-1346-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Allen WE, Chen MZ, Pichamoorthy N, Tien RH, Pachitariu M, Luo L, Deisseroth K: Thirst regulates motivated behavior through modulation of brainwide neural population dynamics. Science 2019, 364:eaav3932. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.••.Lieber JD, Bensmaia SJ: High-dimensional representation of texture in somatosensory cortex of primates. Proc Natl Acad Sci U S A 2019, 116:3268–3277. [DOI] [PMC free article] [PubMed] [Google Scholar]; Electrophysiological recordings from the somatosensory cortex of awake monkeys found neural population representations of unique tactile stimuli to be high-dimensional. Individual neurons exhibited unique patterns in response to different features of texture including, roughness, hardness, stickiness, and coolness. Across the population, this results in a high-dimensional representation that facilitates the discrimination of detailed and complex natural textures. More broadly, these results highlight the capacity of neural populations to represent a complex perceptual space.
  • 6.Fusi S, Miller EK, Rigotti M: Why neurons mix: high dimensionality for higher cognition. Current Opinion in Neurobiology 2016, 37:66–74. [DOI] [PubMed] [Google Scholar]
  • 7.Badre D, Bhandari A, Keglovits H, Kikumoto A: The dimensionality of neural representations for control. Current Opinion in Behavioral Sciences 2021, 38:20–28. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Kyriazi P, Headley DB, Paré D: Different Multidimensional Representations across the Amygdalo-Prefrontal Network during an Approach-Avoidance Task. Neuron 2020, 107:717–730.e5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Kyriazi P, Headley DB, Pare D: Multi-dimensional Coding by Basolateral Amygdala Neurons. Neuron 2018, 99:1315–1328.e5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Ma L, Hyman JM, Durstewitz D, Phillips AG, Seamans JK: A Quantitative Analysis of Context-Dependent Remapping of Medial Frontal Cortex Neurons and Ensembles. J Neurosci 2016, 36:8258–8272. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.MacDowell CJ, Buschman TJ: Low-Dimensional Spatiotemporal Dynamics Underlie Cortex-wide Neural Activity. Current Biology 2020, 30:2665–2680.e8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Cohen JD, McClure SM, Yu AJ: Should I stay or should I go? How the human brain manages the trade-off between exploitation and exploration. Philos Trans R Soc Lond B Biol Sci 2007, 362:933–942. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Cardin JA, Crair MC, Higley MJ: Mesoscopic Imaging: Shining a Wide Light on Large-Scale Neural Dynamics. Neuron 2020, 108:33–43. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Muller L, Chavane F, Reynolds J, Sejnowski TJ: Cortical travelling waves: mechanisms and computational principles. Nat Rev Neurosci 2018, 19:255–268. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Mitra A, Kraft A, Wright P, Acland B, Snyder AZ, Rosenthal Z, Czerniewski L, Bauer A, Snyder L, Culver J, et al. : Spontaneous Infra-slow Brain Activity Has Unique Spatiotemporal Dynamics and Laminar Structure. Neuron 2018, 98:297–305.e6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Greenberg A, Abadchi JK, Dickson CT, Mohajerani MH: New waves: Rhythmic electrical field stimulation systematically alters spontaneous slow dynamics across mouse neocortex. NeuroImage 2018, 174:328–339. [DOI] [PubMed] [Google Scholar]
  • 17.Matsui T, Murakami T, Ohki K: Transient neuronal coactivations embedded in globally propagating waves underlie resting-state functional connectivity. PNAS 2016, 113:6556–6561. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Matsui T, Murakami T, Ohki K: Neuronal Origin of the Temporal Dynamics of Spontaneous BOLD Activity Correlation. Cereb Cortex 2019, 29:1496–1508. [DOI] [PubMed] [Google Scholar]
  • 19.Vanni MP, Chan AW, Balbi M, Silasi G, Murphy TH: Mesoscale Mapping of Mouse Cortex Reveals Frequency-Dependent Cycling between Distinct Macroscale Functional Modules. J Neurosci 2017, 37:7513–7533. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.••.MacDowell CJ, Briones BA, Lenzi MJ, Gustison ML, Buschman TJ: Variability in sampling of cortex-wide neural dynamics explains individual differences in functional connectivity and behavioral phenotype. 2022, doi: 10.1101/2022.01.24.477572. [DOI] [Google Scholar]; Mice with unique behavioral profiles induced by in utero teratogen exposure and typical control mice all exhibit the same spatiotemporal motifs in cortical neural dynamics. However, how often each motif occurred differed across individuals and correlated with individual differences in behavioral profile and structure in cortical functional connectivity. This suggests that individual variability in the sampling of patterns in information flow across neural populations may underlie differences in behavior.
  • 21.Musall S, Kaufman MT, Juavinett AL, Gluf S, Churchland AK: Single-trial neural dynamics are dominated by richly varied movements. Nature Neuroscience 2019, 22:1677–1686. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Makino H, Ren C, Liu H, Kim AN, Kondapaneni N, Liu X, Kuzum D, Komiyama T: Transformation of Cortex-wide Emergent Properties during Motor Learning. Neuron 2017, 94:880–890 e8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Mitelut C, Zhang Y, Sekino Y, Boyd J, Bolanos F, Swindale NV, Silasi G, Saxena S, Murphy TH: Mesoscale cortex-wide neural dynamics predict self-initiated actions in mice several seconds prior to movement. 2021, doi: 10.1101/2021.12.17.473193. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Mohajerani MH, Chan AW, Mohsenvand M, LeDue J, Liu R, McVea DA, Boyd JD, Wang YT, Reimers M, Murphy TH: Spontaneous cortical activity alternates between motifs defined by regional axonal projections. Nat Neurosci 2013, 16:1426–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Linden NJ, Tabuena DR, Steinmetz NA, Moody WJ, Brunton SL, Brunton BW: Go with the FLOW: visualizing spatiotemporal dynamics in optical widefield calcium imaging. J R Soc Interface 2021, 18:20210523. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Xiao D, Forys BJ, Vanni MP, Murphy TH: MesoNet allows automated scaling and segmentation of mouse mesoscale cortical maps using machine learning. Nat Commun 2021, 12:5992. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Vanni MP, Murphy TH: Mesoscale Transcranial Spontaneous Activity Mapping in GCaMP3 Transgenic Mice Reveals Extensive Reciprocal Connections between Areas of Somatomotor Cortex. J Neurosci 2014, 34:15931–15946. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Yeo BT, Krienen FM, Sepulcre J, Sabuncu MR, Lashkari D, Hollinshead M, Roffman JL, Smoller JW, Zollei L, Polimeni JR, et al. : The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J Neurophysiol 2011, 106:1125–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Duncan J, Assem M, Shashidhara S: Integrated Intelligence from Distributed Brain Activity. Trends in Cognitive Sciences 2020, 24:838–852. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Parks EL, Madden DJ: Brain Connectivity and Visual Attention. Brain Connect 2013, 3:317–338. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Preti MG, Bolton TA, Van De Ville D: The dynamic functional connectome: State-of-the-art and perspectives. Neuroimage 2017, 160:41–54. [DOI] [PubMed] [Google Scholar]
  • 32.Wiltschko AB, Johnson MJ, Iurilli G, Peterson RE, Katon JM, Pashkovski SL, Abraira VE, Adams RP, Datta SR: Mapping Sub-Second Structure in Mouse Behavior. Neuron 2015, 88:1121–1135. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Mearns DS, Donovan JC, Fernandes AM, Semmelhack JL, Baier H: Deconstructing Hunting Behavior Reveals a Tightly Coupled Stimulus-Response Loop. Current Biology 2019, 0. [DOI] [PubMed] [Google Scholar]
  • 34.Ramirez-Mahaluf JP, Medel V, Tepper Á, Alliende LM, Sato JR, Ossandon T, Crossley NA: Transitions between human functional brain networks reveal complex, cost-efficient and behaviorally-relevant temporal paths. Neuroimage 2020, 219:117027. [DOI] [PubMed] [Google Scholar]
  • 35.•.Vidaurre D, Smith SM, Woolrich MW: Brain network dynamics are hierarchically organized in time. Proc Natl Acad Sci USA 2017, 114:12827–12832. [DOI] [PMC free article] [PubMed] [Google Scholar]; Functional imaging of the human brain in during resting state revealed hierarchical gradients in the temporal organization of ongoing activity across the cerebral cortex and subcortical structures. These timescale gradients related to functional connectivity with slower dynamics correlating to stronger functional connectivity across distributed brain areas. These results suggest a hierarchical organization in human brain dynamics over time.
  • 36.Baker AP, Brookes MJ, Rezek IA, Smith SM, Behrens T, Probert Smith PJ, Woolrich M: Fast transient networks in spontaneous human brain activity. eLife 2014, 3:e01867. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Raut RV, Snyder AZ, Raichle ME: Hierarchical dynamics as a macroscopic organizing principle of the human brain. PNAS 2020, 117:20890–20897. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Botvinick MM: Hierarchical models of behavior and prefrontal function. Trends Cogn Sci 2008, 12:201–208. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Melozzi F, Bergmann E, Harris JA, Kahn I, Jirsa V, Bernard C: Individual structural features constrain the mouse functional connectome. PNAS 2019, 116:26961–26969. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Levakov G, Faskowitz J, Avidan G, Sporns O: Mapping individual differences across brain network structure to function and behavior with connectome embedding. NeuroImage 2021, 242:118469. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Gilad A, Helmchen F: Spatiotemporal refinement of signal flow through association cortex during learning. Nature Communications 2020, 11:1744. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Buschman TJ, Denovellis EL, Diogo C, Bullock D, Miller EK: Synchronous oscillatory neural ensembles for rules in the prefrontal cortex. Neuron 2012, 76:838–846. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Bonnefond M, Kastner S, Jensen O: Communication between Brain Areas Based on Nested Oscillations. eNeuro 2017, 4:ENEURO.0153–16.2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.••.Lohani S, Moberly AH, Benisty H, Landa B, Jing M, Li Y, Higley MJ, Cardin JA: Dual color mesoscopic imaging reveals spatiotemporally heterogeneous coordination of cholinergic and neocortical activity. bioRxiv 2021, doi: 10.1101/2020.12.09.418632. [DOI] [PMC free article] [PubMed] [Google Scholar]; Simultaneous mesoscale imaging of cortex-wide neural activity and acetylcholine release with concomitant behavioral monitoring found that patterns in cortical neural activity and acetylcholine release are highly variable across different behavioral states. Patterns in acetylcholine release reflected facial movement and locomotion and were associated with both increased synchronization of neural activity across cortical areas and decorrelation within local neural populations.
  • 45.Sternson SM: Exploring internal state-coding across the rodent brain. Current Opinion in Neurobiology 2020, 65:20–26. [DOI] [PubMed] [Google Scholar]
  • 46.Eiselt A-K, Chen S, Chen J, Arnold J, Kim T, Pachitariu M, Sternson SM: Hunger or thirst state uncertainty is resolved by outcome evaluation in medial prefrontal cortex to guide decision-making. Nat Neurosci 2021, 24:907–912. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Barson D, Hamodi AS, Shen X, Lur G, Constable RT, Cardin JA, Crair MC, Higley MJ: Simultaneous mesoscopic and two-photon imaging of neuronal activity in cortical circuits. Nature Methods 2020, 17:107–113. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Luo TZ, Bondy AG, Gupta D, Elliott VA, Kopec CD, Brody CD: An approach for long-term, multi-probe Neuropixels recordings in unrestrained rats. bioRxiv 2020, doi: 10.1101/2020.04.13.039305. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 49.Fries P: Rhythms for Cognition: Communication through Coherence. Neuron 2015, 88:220–235. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50.Sarter M, Hasselmo ME, Bruno JP, Givens B: Unraveling the attentional functions of cortical cholinergic inputs: interactions between signal-driven and cognitive modulation of signal detection. Brain Research Reviews 2005, 48:98–111. [DOI] [PubMed] [Google Scholar]
  • 51.Ranjbar-Slamloo Y, Fazlali Z: Dopamine and Noradrenaline in the Brain; Overlapping or Dissociate Functions? Frontiers in Molecular Neuroscience 2020, 12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Sherman SM, Guillery RW: The role of the thalamus in the flow of information to the cortex. Philos Trans R Soc Lond B Biol Sci 2002, 357:1695–1708. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.O’Reilly RC, Frank MJ: Making working memory work: a computational model of learning in the prefrontal cortex and basal ganglia. Neural computation 2006, 18:283–328. [DOI] [PubMed] [Google Scholar]
  • 54.Pouzzner D: Control of Functional Connectivity in Cerebral Cortex by Basal Ganglia Mediated Synchronization. arXiv:170800779 [q-bio] 2020, [Google Scholar]
  • 55.Poulet JFA, Fernandez LMJ, Crochet S, Petersen CCH: Thalamic control of cortical states. Nat Neurosci 2012, 15:370–372. [DOI] [PubMed] [Google Scholar]

RESOURCES