The Neural Basis of Timing: Distributed Mechanisms for Diverse Functions
Read:: Project:: [] Print:: ❌
- print Zotero Link:: NA PDF:: NA Files:: Paton_Buonomano_2018_The Neural Basis of Timing.pdf; patonNeuralBasisTiming2018-zotero.md; PubMed entry; ScienceDirect Snapshot Reading Note:: Joseph J. Paton, Dean V. Buonomano 2018
Abstract
Timing is critical to most forms of learning, behavior, and sensory-motor processing. Converging evidence supports the notion that, precisely because of its importance across a wide range of brain functions, timing relies on intrinsic and general properties of neurons and neural circuits; that is, the brain uses its natural cellular and network dynamics to solve a diversity of temporal computations. Many circuits have been shown to encode elapsed time in dynamically changing patterns of neural activity—so-called population clocks. But temporal processing encompasses a wide range of different computations, and just as there are different circuits and mechanisms underlying computations about space, there are a multitude of circuits and mechanisms underlying the ability to tell time and generate temporal patterns.
Quick Reference
Top Comments
They are arguing, as I would, that the timing properties are emergent or intrinsic. See pg 2 exceprts
Topics
Tasks
—
Extracted Annotations and Comments
Annotations(6/21/2022, 8:45:10 AM)
“studies in rodents have confirmed the lack of any direct relationship between circadian timing and interval timing on the scale of seconds (Lewis et al., 2003; Cordes and Gallistel, 2008; Papachristos et al., 2011)—of course, because the circadian rhythms modulate a wide variety of cognitive and physiological functions, it can affect performance on a wide range of tasks, including timing tasks (Golombek et al., 2014)” (p. 687)
“the mechanisms underlying timing on the intermediate scale of tens of milliseconds to tens of seconds remains a mystery” (p. 687)-
==“The open question”==
“these models loosely fit into two broad classes: dedicated and intrinsic models (Ivry and Schlerf, 2008). Dedicated models propose that the brain has a more or less centralized set of circuits for timing that account for timing across modalities, tasks, and scales within the range of hundreds of milliseconds to many seconds. In these models, timing relies on dedicated or specialized neural mechanisms. Intrinsic models propose that timing is an intrinsic computation of most neural circuits, and timing per se emerges from general properties of neurons and the inherent dynamics of neural circuits.” (p. 688)-
==“So what I want to say is that timing is an emergent property of the timing menchanisms”==
“Here, we argue that converging data strongly support intrinsic models. Indeed, we suggest that given the importance and universality of temporal computations, dedicated models would not make computational sense. This does not imply that there are not some brain areas involved in a range of temporal tasks that share similar temporal processing requirements, but rather that distinct temporal computations, such as processing a Morse code message and anticipating when a traffic light will change, rely on distinct circuits and mechanisms. Under this view, areas that are consistently implicated in timing tasks should not be thought of as a central clock, but as areas that are involved in tasks that are inherently temporal in naturee.g., since preparing and producing motor responses are inherently temporal in nature, motor areas should be consistently implicated in timing.” (p. 688)-
==“And they argue the same”==
“How the brain processes information about space provides a useful analogy for the intrinsic timing perspective. Like the temporal dimension, the spatial dimension permeates much of what the brain must accomplish, from localizing the position of objects in space, to guiding movements to grasp objects, and creating large-scale maps for spatial navigation. Mammals have many different maps of external space, including those in the colliculi, auditory cortex, visual cortex, hippocampus, and parietal cortex (Knudsen et al., 1987; Kandel et al., 2013). The multitude of spatial representations within the brain can map onto each other and form more general polymodal maps in the parietal cortex. Furthermore, consistent with the intrinsic perspective of timing, different maps of external space are computed in different ways and make distinct contributions to sensori-motor processing and cognition.” (p. 688)-
==“overlap in representations of space and time”==
==“Fig 1”==
(p. 688)
“the timing field is to establish the correct taxonomy of time (Meck and Ivry, 2016).” (p. 688)-
“That is, to determine which of the many different forms of timing rely on the same circuits and mechanisms.” (p. 688)
*“As a first step toward a taxonomy of time, it is critical to distinguish between true timing tasks and time-dependent tasks.
Timing tasks refer to those that are directly based on interval or duration and that require some sort of timing device to solve. In contrast, some tasks are defined by their temporal properties but are not considered timing tasks, such as judging whether two sensory events occur simultaneously or not (asynchrony tasks) or which of two events came first (temporal-order tasks).
These tasks do not require a clock or timing device to solve. Standard examples of timing tasks include (Grondin, 2010):
Interval/duration discrimination. discriminating which of two presented durations (or intervals) is the longest, or making a judgment as to whether an event is short or long relative to a standard (e.g., bisection task).
Reproduction. reproducing the duration or temporal structure of a presented sensory stimulus—e.g., tapping an interval demarcated by two tones or reproducing the complex temporal structure of a presented Morse code pattern.
Production. production of a simple or complex temporal pattern in the absence of any recent sensory presentation of the relevant interval or pattern—e.g., human subjects asked to press a key for ‘‘1 second,’’ or a rodent that produces a timed anticipatory motor response (e.g., an eyeblink that precedes the US, or licking in anticipation of a predicted reward).”*
(p. 688)-
==“Timing Task”==*
“Figure 1. Taxonomy of Timing Tasks The continuum along at least two task dimensions are likely to be important for understanding the neural basis of timing: sensory versus motor and interval versus pattern timing. Some tasks (Interval Timing) require the discrimination (Sensory Timing) or production (Motor Timing) of simple durations or intervals (or anticipation of an external event). Other tasks (Pattern Timing) require the discrimination or production of complex temporal or spatiotemporal patterns—such as deciphering Morse code signals (Sensory timing) or tapping a complex temporal pattern (Motor Timing). Upper left: adapted from Gouveˆ a et al. (2015). Lower left: adapted from Kawai et al. (2015).”*
“Subsecond versus Suprasecond Timing. There is ample evidence that timing of very short and very long intervals relies on different mechanisms and areas; however, there is no clear boundary between what constitutes a short or long interval. Nevertheless, a loose distinction between sub- and supra-second timing is often made. Pharmacological (Rammsayer and Vogel, 1992; Rammsayer, 1999), psychophysical (Karmarkar and Buonomano, 2007; Spencer et al., 2009; Rammsayer et al., 2015), and imaging (Lewis and Miall, 2003) studies suggest that discriminating a short interval (e.g., 50–100 ms) recruits different circuits than the discrimination of longer (>1 s) intervals.” (p. 689)
*Interval versus Pattern Timing. Imaging studies suggest that tasks that require the production of simple intervals or specific patterns recruit different neural circuits (Grube et al., 2010; Teki et al., 2011).
Indeed, the distinction between simple and complex timing seems critical because these timing tasks can have fundamentally different computational requirements (Hardy and Buonomano, 2016).
Discriminating the duration of a single musical note or anticipating the arrival of a reward relies on the timing of isolated durations or intervals and can easily be solved with timing mechanisms analogous to a stopwatch.
In contrast, recognizing the tempo of a song, the prosody of speech, or producing Morse code are tasks that are defined by the duration and interval of components, as well as by the overall global temporal structure of a sequence of these components. Critically, when such patterns are scaled in time, they can be identified as the same pattern (a song played at different tempos is still the same song).* (p. 689)-
==“Interval Timing”==
*“David Marr distinguished between three levels of analyses:
(1) a computational level that essentially defined the problem being addressed from a computational or information processing perspective;
(2) an algorithmic level that sought to solve a problem algorithmically—that is, without regard to how the brain may actually implement such an algorithm; and
(3) an implementational level, which, in the case of neuroscience, seeks to develop models implemented at the level of synapses, neurons, and neural circuits.”* (p. 689)
“The first models of timing on the scale of hundreds of milliseconds and seconds were pacemaker-accumulator models (Creelman, 1962; Treisman, 1963)—and, by far, the most influential of these is referred to as scalar-expectancy theory (Gibbon, 1977).” (p. 689)
“pacemaker-accumulator models (Creelman, 1962; Treisman, 1963)” (p. 689)-
==“These seem important”==
“scalar-expectancy theory (Gibbon, 1977).” (p. 689)
“Like man-made clocks, pacemaker-accumulator models postulated a time-base or oscillator, and an accumulator or integrator that essentially provides a linear readout of elapsed time. Most pacemaker-accumulator models, however, concerned themselves with accounting for the behavioral data, such was whether Weber’s law was satisfied, and not with a biological implementation.” (p. 689)
“Weber’s law (or the scalar property) is a general feature of timing and represents an important benchmark for models of timing (Gibbon, 1977).” (p. 689)
“It refers to the observation that, for example, in motor timing tasks the SD of the response time across trials increases linearly with the mean time of the responses. While Weber’s law is robust, it is not universal, and it generally applies to restricted temporal ranges, e.g., the Weber fraction (s/t) can differ significantly for intervals of a few hundred milliseconds, seconds, and tens of seconds (Lewis and Miall, 2009; Grondin, 2014).” (p. 689)
“Ramping models (e.g., Durstewitz, 2003; Simen et al., 2011; Balci and Simen, 2016) propose that time is encoded in monotonic changes in firing rate and that actions are produced when the firing rates reaches a threshold value. Such ramping neurons have been observed in a wide range of brain areas during timing tasks.” (p. 689)-
==“Ramping Models”==
“An alternative to encoding time in the monotonic changes in firing rate is that the nervous system encodes time in the dynamically changing population of neurons (population” (p. 689)-
==“Population Models”==“clocks)—ranging from sequential chains of activity (Abeles, 1982) to complex patterns. This hypothesis, referred to as population clock, was first proposed in the context of the cerebellum (Buonomano and Mauk, 1994; Mauk and Donegan, 1997), and there is now a large amount of cumulative data supporting this hypothesis.”
*“a prototypical sensory timing task is interval (or duration) discrimination, whereas in animal studies the bisection task is often used. In a bisection task, subjects are trained to make one choice when presented with a stimulus of a long duration, and another choice when presented with a stimulus of a short duration.
“ interval-timing timing-task”
*“Although some studies demonstrated that musicians are superior at interval discrimination (Keele et al., 1985), other studies suggested interval timing does not improve with practice (Rammsayer, 1994).
Subsequent studies, however, revealed that interval learning undergoes robust learning—however, unlike some forms of perceptual learning, temporal perceptual learning is relatively slow and requires training across days (for a review, see Bueti and Buonomano, 2014).”* (p. 690)-
“One of the first studies to demonstrate temporal perceptual learning revealed that, after training subjects for 1 hr a day for 10 days, interval discrimination thresholds for a 100-ms interval improved from 24% to 12% (Wright et al., 1997). Importantly, despite the significant learning on the trained 100-ms interval, there was no detectable improvement on untrained 50-, 200-, and 500-ms intervals. This temporal specificity of temporal perceptual learning has been replicated in many studies and is now seen as a general characteristic of temporal perceptual learning (Nagarajan et al., 1998; Karmarkar and Buonomano, 2003; Buonomano et al., 2009; Wright et al., 2010; Bueti et al., 2012). Temporal specificity during interval-discrimination tasks constrain the neural mechanisms and models underlying sensory timing and argue against the notion of a single master clock.” (p. 690)
“Specifically, if the overall precision of a clock improved with practice, it would be expected to enhance performance across a range of intervals, not just the trained interval. Another critical question relates to ‘‘spatial’’ generalization of temporal learning—e.g., after training on a 100-ms interval demarcated by brief 1-kHz tones, do humans improve on their ability to discriminate that same interval now bounded by 4-kHz tone? Interestingly, most studies have reported robust spatial generalization, but the interpretation of this finding is complicated by the fact that spatial generalization lags temporal perceptual learning—suggesting that generalization to different tones may result from top-down mechanisms independent of the timing mechanisms per se (Wright et al., 2010).” (p. 690)
“Interval- and rate-tuned neurons have also been identified in the brainstem of weakly electric fish that use the temporal features of discharge from their electric organs to communicate (Figure 2A) (Carlson, 2009). The mechanism underlying temporal tuning in these cases is not fully understood, but it has been established that selectivity relies in part on dynamic changes in the balance of excitation and inhibition imposed by temporal summation and short-term synaptic plasticity (see below).” (p. 690)
“Temporally selective neurons have also been identified in the cortical circuits of birds and mammals” (p. 691)-
==“Except that these are probably circuits.Not the total result of one neutron,”==
“Basal Ganglia” (p. 691)
“The basal ganglia (BG), a collection of subcortical nuclei that receive input from almost the entire cortical mantle as well as multiple thalamic areas, are often implicated in sensory and motor timing on the scale of hundreds of milliseconds to seconds. This is perhaps not surprising given that the BG contribute to reinforcement learning—forming predictions about future reward and selecting actions that lead to rewarding outcomes.” (p. 691)-
==“#RL”==
“A fundamental aspect of learning to predict something is the ability to detect temporal contingencies (Balsam and Gallistel, 2009), the degree to which some event or action reduces uncertainty about another, and there is behavioral evidence that animals represent the temporal statistics of events required for performing probabilistic inference thought to underlie this manner of associative learning (Kheifets and Gallistel, 2012; Li and Dudman, 2013). In addition, execution of behavior often involves proper timing and sequencing of action.” (p. 691)-
==“Perhaps part of the discussion on the importance of time in DRL”==
“Thus, the BG should at the very least have access to representations of timing information for both learning predictions and producing proper behavior.” (p. 691)-
==“was the Ba part of the loop Gustavo mentioned?”==
==“Fig 2”==
!“Figure 2. Example of Interval-Tuned Neurons (A) Voltage traces from a neuron in the midbrain of an electric fish to trains of electrical pulses presented at intervals of 100 (left), 50 (center), and 10 ms (right). The rows represent three separate repetitions of each train. This neuron was tuned to pulses delivered at intervals of 50 ms (right). Adapted from Carlson (2009). (B) Rastergram of a neuron from rat auditory cortex in response to five different stimuli, each composed of a 200-ms 3-kHz tone followed by a 50-ms 7-kHz (characteristic frequency [CF]) tone with different stimulus-onset asynchronies. Numbers represent the facilitation index. Rats were trained to detect an interval of 100 ms between both tones (red arrow), and this was the spatiotemporal pattern that elicited the maximal response across the population (right). Error bars represent SEMs. Adapted from Zhou et al. (2010). (C) Model of how STP can generate an interval selective neuron in a disynaptic circuit composed of an excitatory (blue) and inhibitory (red) neuron (traces from three intervals are overlaid). Left, the input to both neurons exhibits pairedpulse facilitation. Right, by adjusting the weights onto both the Ex and Inh neurons, it is possible to create an Ex neuron that functions as a 50-, 100-, or 200-ms detector. Adapted from Buonomano (2000).”
“One piece of evidence that the BG play a causal role in sensory timing is data showing that inactivation via infusion of muscimol into the rat dorsal striatum impairs performance of a interval categorization task (Gouveˆ a et al., 2015). Recordings from single units around the site of muscimol infusions revealed rich and variable dynamics that, when viewed at the population level, encoded information about elapsed time during interval presentation. Furthermore, the timing information derived from simultaneously recorded ensembles of striatal neurons predicted the trial-to-trial variation in duration judgments produced by the animals (Figure 3). When population dynamics proceeded more quickly, rats were more likely to judge a given interval as being in the ‘‘long’’ category, and vice versa when population dynamics proceeded more slowly, indicating that striatal dynamics reflected the timing information that rats were using to guide their judgments (Gouveˆ a et al., 2015). These data demonstrate that the striatum was required, and striatal populations encoded information, for guiding what we would define as a sensory timing task.” (p. 692)
“comparing simultaneously recorded high-speed video and neural population activity revealed a clear asymmetry between when timing information appeared in neural activity and behavior, with neural activity leading behavior by 300 ms (Gouveˆ a et al., 2015). Thus, while time encoding by striatal neurons likely carries information about a plan for future action, it is unlikely to represent motor commands on their way out of the CNS, nor could it solely reflect the sensory consequences of action” (p. 692)
“Computational models of timing have not generally explicitly distinguished between sensory and motor timing. We argue that such a distinction is important, because the temporally selective neurons in the brainstem and sensory cortex seem to behave as temporal filters as opposed to timers, and are unlikely to be directly responsible for the production of timed motor patterns. Mechanistically, we can think of the sensory and motor timing distinction as relying on passive versus active neural mechanisms, respectively.” (p. 692)
“Passive neural mechanisms refer to those that react to the temporal structure of stimuli, but that are incapable of actively generating a timed response. A prototypical example of a passive mechanism is a band-pass temporal filter, which gates the information arriving at certain frequencies, but cannot actively produce a timed response. In contrast, motor and implicit timing require a circuit to actively generate a timed signal. We stress, however, that, while the distinction between sensory and motor timing is important, they can be overlapping, and indeed many models of motor timing can account for simple sensory timing (such as interval and duration discrimination).” (p. 692)
“Figure 3. Midbrain Dopamine Neurons and Striatal Dynamics May Interact to Regulate Timing (A) The speed with which striatal ensembles traverse neural space (top panel) predicts duration judgments (lower panel) in an interval-discrimination task. Colored schematic trajectories in top panel depict a quickly (red) or slowly (blue) evolving ensemble activity pattern during interval presentation in a space defined by the firing of simultaneously recorded striatal neurons. Psychometric curves for trials segregated on the basis of whether activity proceeded quickly or slowly during interval presentation. Adapted from Gouveˆ a et al. (2015). (B) Calcium signals collected from dopamine neurons in the SNc exhibited trial-to-trial variability during interval presentations (top panel) that predicted the timing judgments of mice during the same interval-discrimination task used during the data collected in (A) (adapted from Soares et al., 2016). Given the dense innervation of striatal networks (in black, center) by nigro-striatal dopamine neurons (in purple, center) and the fact that SNc dopamine neurons receive significant input from striatum, these data support a hypothesis where the two brain areas reciprocally influence each other’s timing functions.” (p. 692)
“Hardy, N.F., Goudar, V., Romero-Sosa, J.L., and Buonomano, D. (2017). A model of temporal scaling correctly predicts that Weber’s law is speed-dependent. bioRxiv. https://doi.org/10.1101/159590.” (p. 702)
“Meck, W.H., and Ivry, R.B. (2016). Editorial overview: Time in perception and action. Curr. Opin. Behav. Sci. 8, vi–x.” (p. 703)
==“Fig 1”==
(p. 688)