Limits...
Using multi-compartment ensemble modeling as an investigative tool of spatially distributed biophysical balances: application to hippocampal oriens-lacunosum/moleculare (O-LM) cells.

Sekulić V, Lawrence JJ, Skinner FK - PLoS ONE (2014)

Bottom Line: Models were quantified and ranked based on minimal error compared to a dataset of O-LM cell electrophysiological properties.Co-regulatory balances between conductances were revealed, two of which were dependent on the presence of dendritic Ih.These findings inform future experiments that differentiate between somatic and dendritic Ih, thereby continuing a cycle between model and experiment.

View Article: PubMed Central - PubMed

Affiliation: Toronto Western Research Institute, University Health Network, Toronto, Ontario, Canada; Department of Physiology, University of Toronto, Toronto, Ontario, Canada.

ABSTRACT
Multi-compartmental models of neurons provide insight into the complex, integrative properties of dendrites. Because it is not feasible to experimentally determine the exact density and kinetics of each channel type in every neuronal compartment, an essential goal in developing models is to help characterize these properties. To address biological variability inherent in a given neuronal type, there has been a shift away from using hand-tuned models towards using ensembles or populations of models. In collectively capturing a neuron's output, ensemble modeling approaches uncover important conductance balances that control neuronal dynamics. However, conductances are never entirely known for a given neuron class in terms of its types, densities, kinetics and distributions. Thus, any multi-compartment model will always be incomplete. In this work, our main goal is to use ensemble modeling as an investigative tool of a neuron's biophysical balances, where the cycling between experiment and model is a design criterion from the start. We consider oriens-lacunosum/moleculare (O-LM) interneurons, a prominent interneuron subtype that plays an essential gating role of information flow in hippocampus. O-LM cells express the hyperpolarization-activated current (Ih). Although dendritic Ih could have a major influence on the integrative properties of O-LM cells, the compartmental distribution of Ih on O-LM dendrites is not known. Using a high-performance computing cluster, we generated a database of models that included those with or without dendritic Ih. A range of conductance values for nine different conductance types were used, and different morphologies explored. Models were quantified and ranked based on minimal error compared to a dataset of O-LM cell electrophysiological properties. Co-regulatory balances between conductances were revealed, two of which were dependent on the presence of dendritic Ih. These findings inform future experiments that differentiate between somatic and dendritic Ih, thereby continuing a cycle between model and experiment.

Show MeSH
Extracting subsets of appropriate O-LM models from the database.(A) Plot showing the general subset cutoff determined by visual examination of the derivative of the distance metric with respect to the model ranking in the database. Vertical dashed line shows cutoff point. (B) The firing frequency plotted as a function of model rank demonstrates one restricted subset of O-LM models. The arrow points to the first failure-to-fire model, thus marking the cutoff point for this restricted subset. (C) The voltage traces of the failure-to-fire model shown in (B). (D) The time constant of the hyperpolarization-induced sag plotted as a function of model rank. The vertical dashed line shows the point in the ranking at which the time constant starts to deviate from the experimentally observed time constants. (E) Histogram of hyperpolarization-induced sag time constants within the experimental O-LM cell dataset.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4215854&req=5

pone-0106567-g003: Extracting subsets of appropriate O-LM models from the database.(A) Plot showing the general subset cutoff determined by visual examination of the derivative of the distance metric with respect to the model ranking in the database. Vertical dashed line shows cutoff point. (B) The firing frequency plotted as a function of model rank demonstrates one restricted subset of O-LM models. The arrow points to the first failure-to-fire model, thus marking the cutoff point for this restricted subset. (C) The voltage traces of the failure-to-fire model shown in (B). (D) The time constant of the hyperpolarization-induced sag plotted as a function of model rank. The vertical dashed line shows the point in the ranking at which the time constant starts to deviate from the experimentally observed time constants. (E) Histogram of hyperpolarization-induced sag time constants within the experimental O-LM cell dataset.

Mentions: Although highly-ranked models better represented O-LM cells than poorly-ranked models (Fig. 2), it is not obvious where a cut-off point should be applied to distinguish the two sets as the distance metric considers all features in a ranked fashion. A principled criterion is therefore needed to extract appropriate models (Fig. 1, Step 2(ii)). We first considered determining a cut-off point from the distance measure itself, noting that the distance measure incorporates consideration of a multitude of electrophysiological features (see Methods and Supplementary Materials). We refer to this cut-off point as the general criterion. This was done by plotting the slopes of the distance measure with respect to the model rank, which is equivalent to the difference of distances between adjacent models in the ranking. In order to test whether there were morphology-specific differences in the intrinsic dynamics of the O-LM models, the models using each of the two reconstructed O-LM cell morphologies were ranked and analyzed separately. After plotting the slopes for models of both morphologies, it was clear that the distance values changed rapidly in the first few thousand models, after which they increased at a relatively steady pace. In other words, model errors accumulated at a constant rate (Fig. 3A, horizontal dashed line). Eventually, the distance values of the ranked models for both morphologies started changing again at a more rapid rate. We therefore set the point at which the ranked models started to rapidly increase their distance values as the cut-off (Fig. 3A, vertical dashed line), as chosen by eye. For models of morphology 1, this resulted in the first 60,000 highly-ranked models counting as appropriate O-LM cell representations (Fig. 3A); likewise, for models of morphology 2, the first 90,000 highly-ranked models were incorporated into the ensemble of appropriate O-LM cell representations (not shown). This total set of 150,000 models was considered the general subset of appropriate O-LM models.


Using multi-compartment ensemble modeling as an investigative tool of spatially distributed biophysical balances: application to hippocampal oriens-lacunosum/moleculare (O-LM) cells.

Sekulić V, Lawrence JJ, Skinner FK - PLoS ONE (2014)

Extracting subsets of appropriate O-LM models from the database.(A) Plot showing the general subset cutoff determined by visual examination of the derivative of the distance metric with respect to the model ranking in the database. Vertical dashed line shows cutoff point. (B) The firing frequency plotted as a function of model rank demonstrates one restricted subset of O-LM models. The arrow points to the first failure-to-fire model, thus marking the cutoff point for this restricted subset. (C) The voltage traces of the failure-to-fire model shown in (B). (D) The time constant of the hyperpolarization-induced sag plotted as a function of model rank. The vertical dashed line shows the point in the ranking at which the time constant starts to deviate from the experimentally observed time constants. (E) Histogram of hyperpolarization-induced sag time constants within the experimental O-LM cell dataset.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4215854&req=5

pone-0106567-g003: Extracting subsets of appropriate O-LM models from the database.(A) Plot showing the general subset cutoff determined by visual examination of the derivative of the distance metric with respect to the model ranking in the database. Vertical dashed line shows cutoff point. (B) The firing frequency plotted as a function of model rank demonstrates one restricted subset of O-LM models. The arrow points to the first failure-to-fire model, thus marking the cutoff point for this restricted subset. (C) The voltage traces of the failure-to-fire model shown in (B). (D) The time constant of the hyperpolarization-induced sag plotted as a function of model rank. The vertical dashed line shows the point in the ranking at which the time constant starts to deviate from the experimentally observed time constants. (E) Histogram of hyperpolarization-induced sag time constants within the experimental O-LM cell dataset.
Mentions: Although highly-ranked models better represented O-LM cells than poorly-ranked models (Fig. 2), it is not obvious where a cut-off point should be applied to distinguish the two sets as the distance metric considers all features in a ranked fashion. A principled criterion is therefore needed to extract appropriate models (Fig. 1, Step 2(ii)). We first considered determining a cut-off point from the distance measure itself, noting that the distance measure incorporates consideration of a multitude of electrophysiological features (see Methods and Supplementary Materials). We refer to this cut-off point as the general criterion. This was done by plotting the slopes of the distance measure with respect to the model rank, which is equivalent to the difference of distances between adjacent models in the ranking. In order to test whether there were morphology-specific differences in the intrinsic dynamics of the O-LM models, the models using each of the two reconstructed O-LM cell morphologies were ranked and analyzed separately. After plotting the slopes for models of both morphologies, it was clear that the distance values changed rapidly in the first few thousand models, after which they increased at a relatively steady pace. In other words, model errors accumulated at a constant rate (Fig. 3A, horizontal dashed line). Eventually, the distance values of the ranked models for both morphologies started changing again at a more rapid rate. We therefore set the point at which the ranked models started to rapidly increase their distance values as the cut-off (Fig. 3A, vertical dashed line), as chosen by eye. For models of morphology 1, this resulted in the first 60,000 highly-ranked models counting as appropriate O-LM cell representations (Fig. 3A); likewise, for models of morphology 2, the first 90,000 highly-ranked models were incorporated into the ensemble of appropriate O-LM cell representations (not shown). This total set of 150,000 models was considered the general subset of appropriate O-LM models.

Bottom Line: Models were quantified and ranked based on minimal error compared to a dataset of O-LM cell electrophysiological properties.Co-regulatory balances between conductances were revealed, two of which were dependent on the presence of dendritic Ih.These findings inform future experiments that differentiate between somatic and dendritic Ih, thereby continuing a cycle between model and experiment.

View Article: PubMed Central - PubMed

Affiliation: Toronto Western Research Institute, University Health Network, Toronto, Ontario, Canada; Department of Physiology, University of Toronto, Toronto, Ontario, Canada.

ABSTRACT
Multi-compartmental models of neurons provide insight into the complex, integrative properties of dendrites. Because it is not feasible to experimentally determine the exact density and kinetics of each channel type in every neuronal compartment, an essential goal in developing models is to help characterize these properties. To address biological variability inherent in a given neuronal type, there has been a shift away from using hand-tuned models towards using ensembles or populations of models. In collectively capturing a neuron's output, ensemble modeling approaches uncover important conductance balances that control neuronal dynamics. However, conductances are never entirely known for a given neuron class in terms of its types, densities, kinetics and distributions. Thus, any multi-compartment model will always be incomplete. In this work, our main goal is to use ensemble modeling as an investigative tool of a neuron's biophysical balances, where the cycling between experiment and model is a design criterion from the start. We consider oriens-lacunosum/moleculare (O-LM) interneurons, a prominent interneuron subtype that plays an essential gating role of information flow in hippocampus. O-LM cells express the hyperpolarization-activated current (Ih). Although dendritic Ih could have a major influence on the integrative properties of O-LM cells, the compartmental distribution of Ih on O-LM dendrites is not known. Using a high-performance computing cluster, we generated a database of models that included those with or without dendritic Ih. A range of conductance values for nine different conductance types were used, and different morphologies explored. Models were quantified and ranked based on minimal error compared to a dataset of O-LM cell electrophysiological properties. Co-regulatory balances between conductances were revealed, two of which were dependent on the presence of dendritic Ih. These findings inform future experiments that differentiate between somatic and dendritic Ih, thereby continuing a cycle between model and experiment.

Show MeSH