Limits...
The stomatogastric nervous system as a model for studying sensorimotor interactions in real-time closed-loop conditions.

Daur N, Diehl F, Mader W, Stein W - Front Comput Neurosci (2012)

Bottom Line: The resulting motor output of a gastric mill motor neuron is then recorded intracellularly and fed into a simple muscle model consisting of a series of low-pass filters.Model properties were either hand tuned to achieve the best match with data from semi-intact muscle preparations, or an exhaustive search was performed to determine the best set of parameters.We report the real-time capabilities of our models, its performance and its interaction with the biological motor system.

View Article: PubMed Central - PubMed

Affiliation: Institute of Neurobiology, Ulm University Ulm, Germany.

ABSTRACT
The perception of proprioceptive signals that report the internal state of the body is one of the essential tasks of the nervous system and helps to continuously adapt body movements to changing circumstances. Despite the impact of proprioceptive feedback on motor activity it has rarely been studied in conditions in which motor output and sensory activity interact as they do in behaving animals, i.e., in closed-loop conditions. The interaction of motor and sensory activities, however, can create emergent properties that may govern the functional characteristics of the system. We here demonstrate a method to use a well-characterized model system for central pattern generation, the stomatogastric nervous system, for studying these properties in vitro. We created a real-time computer model of a single-cell muscle tendon organ in the gastric mill of the crab foregut that uses intracellular current injections to control the activity of the biological proprioceptor. The resulting motor output of a gastric mill motor neuron is then recorded intracellularly and fed into a simple muscle model consisting of a series of low-pass filters. The muscle output is used to activate a one-dimensional Hodgkin-Huxley type model of the muscle tendon organ in real-time, allowing closed-loop conditions. Model properties were either hand tuned to achieve the best match with data from semi-intact muscle preparations, or an exhaustive search was performed to determine the best set of parameters. We report the real-time capabilities of our models, its performance and its interaction with the biological motor system.

No MeSH data available.


Related in: MedlinePlus

Muscle transfer function. (A) Bottom: representation of GM motor neuron activity recorded in a semi-intact preparation. Above are the results of the muscle transfer function (a series of low-pass filters), calculated for the sequence of GM spikes shown in the bottom trace but with different filter time constants. Δx and Δy (see text for details) are given for the different filter settings. The AGR burst is indicated by the gray box. Arrows mark its beginning and end. At a time constant of 320 ms Δx and Δy were smallest. (B) Left: plot showing Δx and Δy for recordings from seven semi-intact preparations, for increasing filter time constants. The values closest to zero for Δx and Δy for the seven experiments were found at a time constant of 320 ms (yellow plane). Right: 2D representation of Δx and Δy at 640 ms (top, orange) and 320 ms (bottom, yellow). Δx and Δy for all animals were smallest at τ = 320 ms. Please note differences in scaling.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3303146&req=5

Figure 6: Muscle transfer function. (A) Bottom: representation of GM motor neuron activity recorded in a semi-intact preparation. Above are the results of the muscle transfer function (a series of low-pass filters), calculated for the sequence of GM spikes shown in the bottom trace but with different filter time constants. Δx and Δy (see text for details) are given for the different filter settings. The AGR burst is indicated by the gray box. Arrows mark its beginning and end. At a time constant of 320 ms Δx and Δy were smallest. (B) Left: plot showing Δx and Δy for recordings from seven semi-intact preparations, for increasing filter time constants. The values closest to zero for Δx and Δy for the seven experiments were found at a time constant of 320 ms (yellow plane). Right: 2D representation of Δx and Δy at 640 ms (top, orange) and 320 ms (bottom, yellow). Δx and Δy for all animals were smallest at τ = 320 ms. Please note differences in scaling.

Mentions: We used a series of three low-pass filters with identical filter time constants to calculate the muscle response in Spike2 (kindly provided by C. Geier and S. Hooper; Ohio University; see also Geier et al., 2002 and Supplementary Material). The timing of the GM motor neuron action potentials is used as input for these filters, i.e., action potentials are treated as time events. We were, however, confronted with the question which time constant would generate the appropriate muscle response. We thus compared the muscle responses calculated at different filter settings (iterated between 10 and 3,000 ms in 10 ms increments) to the AGR bursts that were recorded in the experiments. Naturally, longer time constants resulted in slower muscle responses (Figure 6A from bottom to top). To determine the adequate time constant for any given recording we first measured the values of the transfer function at the beginning (t1 in Figure 6A) and the end (t2 in Figure 6A) of the AGR burst (indicated as gray box) for all calculated filter setting. Assuming that AGR activity starts and ends at the same value of the muscle transfer function only a particular time constant will give a minimum deviation between both values. The difference between these two values is shown in Figure 6 as Δy. As a second measure we used the time at which the muscle response returned to its initial value (the value at the beginning of the AGR burst). The deviation between this time and the time at which the AGR burst ended (Δx) should be small at the optimum muscle time constant. Figure 6A shows these calculations for a series of filters with different time constants. Note that very high time constant values (Figure 6A, top) caused the muscle response to be very slow which clearly did not resemble the condition in the animal. Similarly, very short time constants elicited muscle responses that were by far too short to have caused the observed AGR activity.


The stomatogastric nervous system as a model for studying sensorimotor interactions in real-time closed-loop conditions.

Daur N, Diehl F, Mader W, Stein W - Front Comput Neurosci (2012)

Muscle transfer function. (A) Bottom: representation of GM motor neuron activity recorded in a semi-intact preparation. Above are the results of the muscle transfer function (a series of low-pass filters), calculated for the sequence of GM spikes shown in the bottom trace but with different filter time constants. Δx and Δy (see text for details) are given for the different filter settings. The AGR burst is indicated by the gray box. Arrows mark its beginning and end. At a time constant of 320 ms Δx and Δy were smallest. (B) Left: plot showing Δx and Δy for recordings from seven semi-intact preparations, for increasing filter time constants. The values closest to zero for Δx and Δy for the seven experiments were found at a time constant of 320 ms (yellow plane). Right: 2D representation of Δx and Δy at 640 ms (top, orange) and 320 ms (bottom, yellow). Δx and Δy for all animals were smallest at τ = 320 ms. Please note differences in scaling.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3303146&req=5

Figure 6: Muscle transfer function. (A) Bottom: representation of GM motor neuron activity recorded in a semi-intact preparation. Above are the results of the muscle transfer function (a series of low-pass filters), calculated for the sequence of GM spikes shown in the bottom trace but with different filter time constants. Δx and Δy (see text for details) are given for the different filter settings. The AGR burst is indicated by the gray box. Arrows mark its beginning and end. At a time constant of 320 ms Δx and Δy were smallest. (B) Left: plot showing Δx and Δy for recordings from seven semi-intact preparations, for increasing filter time constants. The values closest to zero for Δx and Δy for the seven experiments were found at a time constant of 320 ms (yellow plane). Right: 2D representation of Δx and Δy at 640 ms (top, orange) and 320 ms (bottom, yellow). Δx and Δy for all animals were smallest at τ = 320 ms. Please note differences in scaling.
Mentions: We used a series of three low-pass filters with identical filter time constants to calculate the muscle response in Spike2 (kindly provided by C. Geier and S. Hooper; Ohio University; see also Geier et al., 2002 and Supplementary Material). The timing of the GM motor neuron action potentials is used as input for these filters, i.e., action potentials are treated as time events. We were, however, confronted with the question which time constant would generate the appropriate muscle response. We thus compared the muscle responses calculated at different filter settings (iterated between 10 and 3,000 ms in 10 ms increments) to the AGR bursts that were recorded in the experiments. Naturally, longer time constants resulted in slower muscle responses (Figure 6A from bottom to top). To determine the adequate time constant for any given recording we first measured the values of the transfer function at the beginning (t1 in Figure 6A) and the end (t2 in Figure 6A) of the AGR burst (indicated as gray box) for all calculated filter setting. Assuming that AGR activity starts and ends at the same value of the muscle transfer function only a particular time constant will give a minimum deviation between both values. The difference between these two values is shown in Figure 6 as Δy. As a second measure we used the time at which the muscle response returned to its initial value (the value at the beginning of the AGR burst). The deviation between this time and the time at which the AGR burst ended (Δx) should be small at the optimum muscle time constant. Figure 6A shows these calculations for a series of filters with different time constants. Note that very high time constant values (Figure 6A, top) caused the muscle response to be very slow which clearly did not resemble the condition in the animal. Similarly, very short time constants elicited muscle responses that were by far too short to have caused the observed AGR activity.

Bottom Line: The resulting motor output of a gastric mill motor neuron is then recorded intracellularly and fed into a simple muscle model consisting of a series of low-pass filters.Model properties were either hand tuned to achieve the best match with data from semi-intact muscle preparations, or an exhaustive search was performed to determine the best set of parameters.We report the real-time capabilities of our models, its performance and its interaction with the biological motor system.

View Article: PubMed Central - PubMed

Affiliation: Institute of Neurobiology, Ulm University Ulm, Germany.

ABSTRACT
The perception of proprioceptive signals that report the internal state of the body is one of the essential tasks of the nervous system and helps to continuously adapt body movements to changing circumstances. Despite the impact of proprioceptive feedback on motor activity it has rarely been studied in conditions in which motor output and sensory activity interact as they do in behaving animals, i.e., in closed-loop conditions. The interaction of motor and sensory activities, however, can create emergent properties that may govern the functional characteristics of the system. We here demonstrate a method to use a well-characterized model system for central pattern generation, the stomatogastric nervous system, for studying these properties in vitro. We created a real-time computer model of a single-cell muscle tendon organ in the gastric mill of the crab foregut that uses intracellular current injections to control the activity of the biological proprioceptor. The resulting motor output of a gastric mill motor neuron is then recorded intracellularly and fed into a simple muscle model consisting of a series of low-pass filters. The muscle output is used to activate a one-dimensional Hodgkin-Huxley type model of the muscle tendon organ in real-time, allowing closed-loop conditions. Model properties were either hand tuned to achieve the best match with data from semi-intact muscle preparations, or an exhaustive search was performed to determine the best set of parameters. We report the real-time capabilities of our models, its performance and its interaction with the biological motor system.

No MeSH data available.


Related in: MedlinePlus