A computational framework for temporal sharpening of stimulus input in the olfactory system

Abstract

The olfactory bulb glomerulus is a dense amalgamation of many unique and interconnected cell types. The mechanisms by which these neurons transform incoming information from the sensory periphery have been extensively studied but often with conflicting findings. A recent study by Carey et al. (J Neurophysiol 113: 3112–3129, 2015) details the computational framework for parallel modes of temporal refinement of stimulus input to the olfactory system mediated by local neurons within individual glomeruli.

a key feature of the olfactory system is its ability to extract stimulus information in a temporally phasic manner that is constrained by the respiratory cycle. This arrangement allows for independent sampling of the stimulus environment with each new sniff. Because of the sinusoidal nature of olfactory input, it is important that the activity of bulbar output neurons is tightly coupled with the respiration cycle to avoid cross-contamination between sampling events. Indeed, experimental evidence suggests that the activity of olfactory receptor neurons (ORNs) far outlasts the activity of downstream olfactory bulb output neurons (mitral cells, MCs; Carey and Wachowiak 2011). How the spike output of MCs is temporally refined with respect to sensory input remains largely unresolved, but it is postulated to result from both excitatory and inhibitory signaling from local neurons that also comprise the glomerular neuropil (Hayar et al. 2004; Murphy et al. 2005; Najac et al. 2015). The mechanisms by which small groups of neurons permutate input arising from naturalistic stimuli is an important component of stimulus feature extraction (Vizcay et al. 2015) and have broad applications to systems neuroscience.

In a recent publication, Carey et al. (2015) explored a series of model olfactory bulb glomerulus circuit configurations with respect to sensory input and temporal refinement of MC spike output. A major goal of this study was to better understand the cellular elements and mechanisms that transform incoming sensory signals within the earliest stages of the olfactory system. The series of models implemented within this study were highly constrained by experimental measurements, and results were compared with experimentally observed in vivo MC spike responses in rats, resultant from naturalistic odor stimuli. The primary findings of these studies reveal that local, glomerulus-specific, GABAergic and glutamatergic neurons are independently capable of temporally sharpening MC spike output. Furthermore, within the model environment, a more precise match to experimentally observed data is achieved when inhibitory and excitatory neurons operate in parallel. These findings provide novel insight into information transfer in the olfactory system that may prove to have commonalities among other sensory modalities utilizing repeated phasic sampling events including both the visual and somatosensory systems. In this Neuro Forum article, I discuss the model circuit configurations that were found to be capable of temporally refining MC output as well as functional implications of these findings. I also describe potential future in vivo studies to test the predictions made by these models.

As a starting point, Carey et al. (2015) designed a model that consisted solely of direct ORN input to MCs. This circuit was designed to test the ability of intrinsic MC conductances to temporally constrain MC output. The input to the model throughout this study was supplied in the form of ORN-derived excitatory currents based on direct experimental measurements of ORN output as assayed through presynaptic Ca²⁺-mediated fluorescence signals. Traditionally, MCs were thought, in large part, to receive direct synaptic excitation from the axon terminals of ORNs terminating within the glomerular neuropil. This connectivity could provide a direct linkage between sensory input and glomerular output; however, recent studies have identified a major source of MC excitation as arising through a secondary class of glutamatergic neurons known as external tufted (ET) cells (De Saint Jan et al. 2009; Gire et al. 2012; Najac et al. 2011). Furthermore, several classes of GABAergic neurons, both within the glomerulus itself (periglomerular, PG cells) and deeper in the olfactory bulb (granule cells, GCs), provide strong inhibition to MCs. To determine the capacity in which secondary neuronal inputs to MCs may contribute to temporal transformation of the MC input-output relationship, it is crucial first to understand how intrinsic MC conductances may similarly perform such a function.

The initial model implemented by Carey et al. (2015) examined how MCs might behave when directly coupled to ORNs in the absence of secondary synaptic inputs (ORN-MC model). Unsurprisingly, the ORN-MC model failed to exhibit any temporal sharpening of MC output with respect to the duration of input from ORNs in the absence of conductances generated by secondary neurons (Fig. 1A). The findings of this initial model circuit configuration further validate the critical role of secondary neurons in shaping the activity of MCs and, moreover, strongly indicate that intrinsic MC conductances are insufficient to constrain their output time course.

Fig. 1.

Multiple circuits are independently capable of sharpening input from the sensory periphery. A: direct synaptic linkage between olfactory receptor neurons (ORNs) and mitral cells (MCs) across a wide range of adjustments to intrinsic MC conductances failed to generate temporal refinement of MC output. B: feedforward inhibition from periglomerular (PG) cells with slow kinetics reduced both the duration and spike number generated from MCs in response to ORN input. C: excitation supplied to MCs by external tufted (ET) cells acting as a relay neuron similarly refined output generated from MCs. D: when the inhibitory and excitatory circuits operate in parallel, including additional elements not depicted, the output of MCs closely matches experimentally obtained data.

An obvious candidate to perform the function of temporal sharpening of MC output is the aforementioned inhibitory GABAergic neurons within the olfactory bulb. To this end, the authors next explored a range of circuit configurations including both recurrent and feedforward inhibitory pathways. Interestingly, their analyses revealed only one such model circuit configuration in which the predicted data matched experimentally observed temporal sharpening of MC output. Initially, a model consisting of recurrent inhibition derived from GABAergic PG cells as well as input from GCs. This configuration, however, was found to have little effect on the overall excitability of MCs in terms of spiking duration and overall spike number.

The subsequent iteration of the model consisted of a feedforward circuit whereby PG cells received direct ORN input in parallel with MCs. To begin, the model parameters included relatively fast PG cell rise and decay kinetics (τ_rise = 7 ms and τ_decay = 14 ms); however, in this configuration, inhibition from PG cells was highly potent and prevented the development of strong MC responses. Interestingly, when the kinetics of PG cell excitation were considerably slowed (τ_rise = 14 ms and τ_decay = 170 ms), it allowed for the development of MC spike responses while limiting the overall duration and spike number (Fig. 1B). The output of this model produced data that were highly similar to experimentally observed MC spike responses, thus indicating a potential role for PG cells in temporally refining MC output.

An interesting question raised by these findings is, What is the nature of the slow PG cell kinetics required for MC output sharpening? As discussed by Carey et al. (2015), a large fraction of PG cells receive their primary excitation through ET cells, whereas a smaller population receive direct ORN input. One possibility is that PG cells receiving indirect excitation (slow kinetics) function to provide inhibition to MCs that is critical for the refinement of their output, wherein ORN-driven PG cells (fast kinetics) serve an alternative function, perhaps to provide inhibition to ON terminals, which are known to be modulated by GABA_B receptors (Nickell et al. 1994).

Recent in vivo studies suggest that PG cells do indeed provide a great deal of inhibition to at least a subset of MCs within a glomerulus in response to naturalistic odor stimulation (Fukunaga et al. 2014); however, these studies focused on inhibition-dominated MC responses, rather than output modulation. Recently developed transgenic mouse models have allowed for cell type-specific expression of light-activated, hyperpolarizing proton pumps within the glomerular layer of the olfactory bulb. One such model, archaerhodopsin (ArchT), driven by PG cell-specific promoters including glutamate decarboxylase 65 or 67, could dampen or eliminate the activity of PG cells following light illumination. Light delivery to the glomerular layer to reduce PG cell activity while at the same time recording the in vivo spike activity of MCs following a naturalistic odor stimulus may provide a means to validate the predictions made by these modeling studies. If the model is predictive of experimental data, an expected outcome is that the output of MCs will be broadened in both spike number and spiking duration when light is delivered to the glomerular layer concurrent with odor presentation versus trials where odor alone has been delivered. Additional and important considerations for interpreting such data include the extent to which PG cells supply inhibition to ET cells, a known driver of MC activity, or even other PG cells, as has been demonstrated by in vitro slice recordings (Murphy et al. 2005).

A second component focused on the ability of local excitation within glomeruli to similarly perform the function of temporally sharpening MC activity following sensory input. Quite surprisingly, when the model incorporated ET cells (ORN-ET-MC model) as a relay between ORN input and MC excitation, the spike output from model MCs closely resembled experimentally derived observations regarding both spike number and duration. Interestingly, this temporal refinement of MC activity persisted in the absence of any inhibitory circuitry within the model environment. In response to excitatory input, ET cells typically generate a barrage of action potentials that are limited in temporal duration by a large potassium conductance (Liu and Shipley 2008). This temporally narrow window of excitation supplied by ET cells, at least in the model, limited the duration of output generated from MCs, thereby temporally sharpening the glomerular input/output relationship (Fig. 1C).

An obvious disadvantage of ORN-ET-MC configuration is the potential for the loss of sensory information fidelity. However, advantages of this configuration include the elimination of transmission of small signals, thus providing the potential for a broader range of dynamic outputs from MCs following stronger incoming signals. Recent experimental evidence supports the notion that ET cells provide the majority of excitation to MCs (De Saint Jan et al. 2009; Gire et al. 2012; Najac et al. 2011), and the studies by Carey et al. (2015) provide a previously unexplored role for ET cells in temporally constraining MC excitation. These results provide evidence that feedforward excitation arising from glutamatergic ET cells may also, or perhaps independently, contribute to the temporal refinement of MC output with respect to sensory input. Furthermore, it appears that ET cells may serve a dual function in providing excitation to MCs as well as constraining their excitation window.

Based on these findings, there are several potential experimental avenues to pursue. Similar to the experiments described above, these studies combine optogenetic and in vivo electrophysiological recordings from MCs. By following a method similar to that previously described, the ability of ET cells to sharpen MC output generation could be tested. For example, the cholecystokinin promoter allows for ET cell-specific expression of ArchT, which allows ET cell-specific optogenetic silencing. Recording from MCs while delivering light to the glomerular layer could serve as an effective means to reduce feedforward excitation from ET cells to MCs. Expected outcomes in terms of MC responses to odor stimuli are that 1) MC activity may be expanded in duration as well as overall spike number when compared with trials where light is omitted, more closely matching the activity pattern of ORNs. Such a result may be indicative of MCs receiving a large degree of direct ORN input in addition to ET cell-derived excitation. An alternate and tempting interpretation is that the inhibitory circuit is insufficient to constrain MC output; however, ET cells provide excitatory input to a large fraction of PG cells, and these cells will remain inactive following light illumination of the glomerular layer. Alternatively, 2) a truncated response following light delivery may be observed, indicating that MCs do indeed rely heavily on ET cells for their excitation as well as temporal refinement of their output.

The experiments described in this article may not only provide insight as to input-output transformations within olfactory bulb glomeruli, but also further clarify native glomerular circuit configurations, namely, how much excitation is supplied to MCs via direct ORN input versus indirect ET cell-mediated feedforward excitation, as well as the role ET cells play in constraining MC activity.

Last, building on the findings described above, Carey et al. (2015) incorporated both the ORN-ET-MC and the ORN-PG-MC models in a larger model environment that also consisted of circuit configurations that failed to generate MC temporal sharpening. When all configurations were operating in parallel, the model produced an MC output profile that most closely resembled experimentally obtained data (Fig. 1D). A largely unresolved question with regard to the optimal circuit configuration for temporal sharpening of sensory inputs is whether the inhibitory and excitatory circuits discussed above optimally operate in parallel, as was demonstrated by the model, or if one circuit configuration is more well suited to perform the function. A second consideration is whether the two distinct circuits provide a critical redundancy within the olfactory circuitry. These questions can begin to be addressed as described above by using a combination of electrophysiological and optogenetic techniques.

Carey et al. (2015) provided a series of predictions regarding the ability of distinct circuit elements within olfactory bulb glomeruli to temporally constrain MC spike output in response to naturalistic sensory stimuli using a modeling based approach. Understandably, not all possible circuit configurations were tested by their model; however, their studies identified two independent circuit configurations that were indeed sufficient to temporally sharpen the output profile of MCs. In addition to providing evidence that GABAergic interactions within the glomerulus can alter the time course of MC activity, the authors described a potentially novel role for glutamatergic ET cells in serving the same function. These predictions will serve as a valuable reference as information transfer in the olfactory system is further investigated on the physiological level. In addition, the experiments detailed here provide a means to test the model outcomes and further clarify the mechanisms by which input to the olfactory system is temporally refined.

GRANTS

This work was supported by National Institute on Deafness and Other Communication Disorders Grant F31 DC-013480.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author.

AUTHOR CONTRIBUTIONS

J.D.Z. prepared figures; J.D.Z. drafted manuscript; J.D.Z. edited and revised manuscript; J.D.Z. approved final version of manuscript.