The ability of insects to track and locate an odor source mid-flight is critical for finding food and mates, but this olfactory navigation task is a daunting one. While odors emanating from a source initially emerge in the form of continuously diffusing filaments, turbulent air currents rapidly fold, spin, and fragment these filaments into discontinuous odor strands of differing sizes and concentrations intermixed with clear media. Thus, insects rarely, if ever, are presented with an easily discernable concentration gradient, marking an unambiguous trail to an odor source, to follow; rather, insects, while tracking an odor source, encounter an odor plume comprised of brief, fragmented strands of odor, with odor pulses tending to occur with higher spatial frequency and increasing odor concentrations as the source is approached (Murlis and Jones, 1981; Murlis et al., 1992; Vickers, 2000; Carde and Willis, 2008). Insects have therefore evolved to perform olfactory navigation within the natural context of odor stimuli encountered in pulsatile fashion; indeed, behavioral experiments show that intermittent stimuli are more effective at prompting male moths to exhibit source-seeking behavior than continuous odor plumes (Baker et al., 1985, 1988; Baker, 1989; Kaissling and Kramer, 1990; Kramer, 1992), and that moths employ a source-seeking strategy in which they surge upwind upon encountering odor strands and cast across wind when losing contact with odors (Vickers and Baker, 1994).

While it is crucial for an insect to be able to detect individual encounters with odor strands embedded in a natural plume, it may be equally important to sense the duration, or temporal length, of stimulation (roughly corresponding to the size of an odor strand), for two reasons. On the one hand, the size of an odor strand contains information about distance to the odor source—at closer distances, the cloud of odor molecules is less fragmented by air turbulence (Yee et al., 1995; Connor et al., 2018; Pannunzi and Nowotny, 2019), and so animals are expected to receive relatively longer stimulation (due to increased odor strand size) when near an odor source; it is therefore plausible that animals also use stimulus duration (in addition to the frequency of odor pulses) as a measure to assess distance to the odor source (Celani et al., 2014). On the other hand, lengthier response durations may be important in allowing the animal sufficient time to evaluate a stimulus, as it is likely that longer response durations permit greater extraction of information about the odor. Responding rapidly and briefly at the onset of stimulation is an effective way to register the temporal dynamics of pulsatile encounters with odor strands, but the brevity of such responses may impair assessment of stimulus quality. It is possible that the moth olfactory system employs bimodality—sensory integration of olfactory and mechanosensory input—to resolve this conundrum.

Thus, the natural olfactory landscape for a flying insect consists of fragments of odor strands embedded in a windy medium exhibiting turbulently varying air speed, with robust stimulation of olfactory receptors likely to occur when odor fragments are delivered via packets of high speed air flow. The task of locating an odor source therefore requires integration of chemosensory input (encoding odor identity and concentration) and mechanosensory input (encoding change in wind velocity), since both pieces of information must intermingle in order to simultaneously classify the identity of a particular odor and rapidly resolve the spatiotemporal dynamics of pulsatile odor encounters (Mamiya and Dickinson, 2015). Accordingly, integration of olfactory and mechanical input has been observed in several regions of the nervous system (Jarman, 2002; Sane et al., 2007), including a subtype of trichoid sensilla on male hawkmoth antennae (Lee and Strausfeld, 1990) and sensilla chaetica on honeybee antennae (Whitehead and Larsen, 1976). Unfortunately, the details of such sensory integration—the physiological mechanisms by which these two modalities are interleaved within brain networks, the influence of both modalities on network dynamics, and the dynamical and behavioral consequences that ensue from such co-mingling—are as of yet poorly understood.

A highly promising brain region in which to explore chemosensory and mechanosensory integration is the antennal lobe (AL). The AL consists of excitatory projection neurons (PNs) and inhibitory local neurons (LNs), and is the first brain area to substantially process odor information arriving from olfactory receptor neurons (ORNs) in the sensory periphery. Although the AL has primarily been studied within the domain of odor (and CO₂) detection, there exist data in moths, cockroaches, tadpoles, and mice suggesting that AL neurons (in moths and cockroaches) or olfactory bulb neurons (in tadpoles and mice) are actually bimodal, and exhibit responses to both olfactory and mechanosensory stimuli (Walldow, 1975; Boeckh and Polz, 1983; Kanzaki et al., 1989; Zeiner and Tichy, 1998; Han et al., 2005; Brinkmann and Schild, 2016; Iwata et al., 2017; Baker et al., 2019). While a cohesive picture of the influence of mechanosensory input on AL dynamics has not yet emerged, the AL is, however, an ideal system for the study of sensory integration, as AL architecture is well understood, AL odor response dynamics are well-studied, and AL neurons are easy to measure individually and in aggregate. Additionally, the anatomy and physiology of the AL is analogous to that of the olfactory bulb (OB) in vertebrates (Hildebrand and Shepherd, 1997), and hence may provide insight into chemo- and mechano-sensory integration in a broad range of species.

In prior work (Tuckman et al., 2021), we employ a combined experimental and computational approach to study the AL as a structure that integrates input from multiple sensory modalities, rather than focusing solely on ambient olfactory stimuli. We present experimental evidence showing that, in addition to responding to odors, PNs within the moth AL also respond to mechanosensory signals arising from high-speed air flow across the antennae. Specifically, we show that chemosensory and mechanosensory stimuli induce remarkably differing response dynamics within the AL—olfactory input (in the absence of significant mechanical input, i.e., low wind speed) tends to induce long-lasting PN responses that lack temporal precision, mechanosensory input (high speed non-scented air puffs) leads to brief, temporally precise PN responses, and the two in combination (high speed odor-laden air puffs) leads to an approximate superposition of the two response patterns (i.e., PN responses with a large, temporally precise transient component along with a less intense longer-lasting component). We then develop a biophysically detailed model of the moth AL that captures many salient features of moth AL odor responses reported in the literature, and we employ our model to simulate both olfactory and mechanosensory input, with olfactory input represented as a focal signal delivered to a glomerular subset and mechanosensory input represented as a global signal impinging upon all glomeruli. We show that PN responses within our model closely mimic our experimental observations, and we suggest that a slow inhibitory current from LNs to PNs, coupled with the weak but widespread nature of mechanosensory input (in comparison to olfactory input) and a glomeruli-spanning LN network that widely distributes inhibition throughout the AL, may be largely responsible for the starkly disparate AL dynamics we observe experimentally in response to olfactory vs. mechanosensory signals. We further suggest, using our model, that mechanosensory input may actually somewhat diminish the ability of AL activity to parse and classify a set of ambient environmental odors, putting forth the hypothesis that the role of mechanosensory input may be to help the insect resolve the spatiotemporal structure of environmental odor plumes while possibly sacrificing some accuracy in odor classification in the process.

Our prior modeling work, however, only examines the dynamics of mechanosensory and olfactory responses within the AL in response to “static” stimuli—i.e., solitary and lengthy (1 s) stimuli. As described above, natural odor delivery, particularly during mid-flight tracking of an odor source, consists of rapid encounters with brief, high wind speed odor pulses. In this work, we extend our model to study the AL response to more naturally structured stimuli, and hence we examine the effect of mechanosensory input on the ability of AL activity to resolve the temporal dynamics of brief odor stimuli delivered in a pulsatile fashion.

The architecture of our model is based on known properties of a typical insect AL network, and in particular the AL of the hawkmoth Manduca sexta. In this moth, the axons of 167,000–251,000 ORNs on each antenna (Homberg et al., 1989) ipsilaterally innervate 64–70 glomeruli in the AL (Rospars and Hildebrand, 1992; Grosse-Wilde et al., 2011), suggesting that roughly 64–70 different types of olfactory receptors are distributed among the ORNs. ORN axons synapse onto a total of about 360 LNs and 1,200–1,300 PNs within the AL (Homberg et al., 1988, 1989), with these cells anatomically segregated into glomeruli and each glomerulus tending to receive input from ORNs expressing a common olfactory receptor. PN axons leave the AL in one of five or six antennocerebral tracts (Homberg et al., 1988; Ian et al., 2016). About 10% of PNs are multiglomerular, whereas most are uniglomerular (Homberg et al., 1988). Additionally, an estimate of 80–100 centrifugal neurons provide feedback to the AL (Homberg et al., 1988), at least one of which responds to both olfactory and mechanosensory stimuli (Zhao et al., 2013). Given the complexity of AL circuits and a lack of quantifiable physiological properties of many circuit elements in the AL, we simplify our model to six glomeruli with canonical circuit connections.

We construct a realistic, large-scale, spiking-network model of the moth AL consisting of PNs and LNs organized into six glomeruli, with 10 PNs and 6 LNs per glomerulus (Hildebrand et al., 1979; Matsumoto and Hildebrand, 1981; Kingan and Hildebrand, 1985; Hoskins et al., 1986; Christensen and Hildebrand, 1987; Homberg et al., 1988, 1989). Individual PNs and LNs are governed by integrate-and-fire spiking dynamics, with random but fixed network connectivity—LNs synapse onto other LNs within the same glomerulus and onto PNs both within and across glomeruli, while PNs synapse only onto PNs and LNs within the same glomerulus (Lei et al., 2004; Reisenman et al., 2004). LNs are GABAergic and exert their postsynaptic effects through fast GABA_A receptors (with several ms kinetics) as well as slower-acting metabotropic GABA receptors acting over ~500–1,000 ms time scales. PNs within the model are cholinergic and act synaptically through fast nicotinic acetylcholine receptors; PNs are also equipped with an intrinsic small-conductance calcium-dependent potassium (SK) current that activates following several PN spikes and serves to curtail further spiking (Mercer and Hildebrand, 2002a,b).

We simulate both chemosensory and mechanosensory input to the model, both in isolation and in conjunction. In accordance with the well-established combinatorial odor code employed by olfactory receptor neurons (ORNs) (Joerges et al., 1997; Vickers and Christensen, 1998; Vickers et al., 1998; Malnic et al., 1999; Ng et al., 2002; Wang et al., 2003; Ache and Young, 2005), an odor (in the absence of significant mechanosensory input, i.e., low wind speed) is simulated by delivering an excitatory stimulus current to all cells within a subset of model glomeruli (half of model glomeruli are designated to receive stimulus current); odor identity is equated with the glomerular subset receiving stimulus current. Mechanosensory input (in the absence of odor input) is modeled by sending stimulus current to all cells within the entire AL network. While our prior work is suggestive that PN responses to mechanosensory input tend to be more global than responses to an odor, this cannot be definitively concluded from our prior data (Tuckman et al., 2021), and other work suggests that not all PNs respond to mechanosensory input (Kanzaki et al., 1989); however, we note that the results presented in this paper do not depend on the global nature of mechanosensory input—the model simply requires mechanosensory input to be somewhat more widespread (which is biologically reasonable if ORNs deliver mechanosensory input) and weaker than odor input. If both of these features exist, then activation of the globally connected LN network allows rapid suppression of relatively weak PN responses, ensuring transient responses to mechanosensory input. In order to maintain similar net levels of excitation to the AL in the case of an olfactory vs. a mechanosensory stimulus, the strength of the stimulus current in the mechanosensory case is reduced by a factor of 12 relative to stimulated glomeruli in the olfactory case; this facilitates direct comparison of network dynamics in response to a global signal (mechanosensory input) vs. a focal signal (odor input). Figure 1 shows a schematic of the model network.

Thus, we employ three stimulus scenarios: (i) odor only, with no mechanosensory input, which simulates odor delivered at low wind speed; (ii) mechanosensory input only, with no odor input, which simulates a non-scented, high speed air puff; (iii) additive sensory integration, simulating high wind speed odor delivery, enacted via simple summation of the current pulses in the odor only and mechanosensory only scenarios. We note that in our prior work (Tuckman et al., 2021) we also employed another distinct paradigm of high wind speed odor delivery (the normalized paradigm), simulated by halving then summing the current pulses in the odor only and mechanosensory only scenarios. However, we found that the additive paradigm described in (iv) yielded moth AL dynamics in response to high wind speed odor presentation that were more consistent with our experimental data, and hence in this work we do not include the fourth paradigm. Model details can be found in the section 4. We also note that there is no “odor only” scenario in nature, but we explore this scenario as a means to isolate the effects of the two sensory modalities on the system.

Experiments show that a series of short odor pulses evokes a sequence of corresponding spike trains in activated PNs; each individual pulse produces an I₁ hyperpolarization followed by phase II depolarization (on top of which is superimposed a burst of spikes), with pulse offset eliciting abrupt truncation of spiking activity. The prolonged after-hyperpolarization (AHP) phase, however, does not fully manifest until the end of the final pulse in the stimulus train (Christensen et al., 1998; Lei et al., 2002). Our model PNs, in the stimulus scenarios that include odor, exhibit similar behavior (Figures 2A,C). Within our model, the onset of each pulse initially elicits a brief, synchronous burst of LN spikes that leads to a brief hyperpolarization of PNs (I₁), with PN spiking ensuing once LNs desynchronize and excitation to PNs ramps up (phase II). Pulse offset yields diminishment of PN spiking as excitation ramps down—LN spiking, on the other hand, does not diminish as rapidly between pulses, and hence the slow inhibitory current to PNs continues to climb during interpulse intervals. At the end of the final pulse in the stimulus train, stimulus-induced excitation to PNs and LNs rapidly vanishes and PN and LN spiking diminishes; however, the slow inhibitory current to PNs, due to its prolonged dynamics, decays over a longer time scale than the stimulus-induced excitation, leading to a lengthy (~1 s) period of PN hyperpolarization as the slow inhibition gradually vanishes (the AHP phase). We also note that within our model PN responses to mechanosensory stimulation alone tend to be briefer and less intense than responses to odor stimulation or combined odor and mechanosensory stimulation (Tuckman et al., 2021).

Furthermore, Figure 2 shows that, at a stimulus pulse frequency of 4 Hz, network PNs receiving stimulus current in the odor only (Figure 2A) and mechanosensory only (Figure 2B) stimulus scenarios appear capable of tracking the temporal structure of pulsatile stimulus encounters. Interestingly, in the additive sensory integration scenario (Figure 2C), PNs that receive both odor and mechanosensory stimulation appear to be less capable of tracking pulse frequency than PNs receiving mechanosensory stimulation alone. This latter result is of particular importance to the present work, as additive sensory integration is the scenario simulating high wind speed odor delivery and hence likely to be most relevant to tracking odor sources. The role played by mechanosensory vs. odor input in the ability of PNs to track the length and frequency of stimulus pulses, as well as the underlying network dynamics, will be explored in the remainder of this paper.

In this work, we study a model of the moth AL incorporating both mechanosensory input, as a relatively weak but more broadly distributed signal, as well as olfactory input, as a relatively strong signal but with focal glomerular targeting, and we assess the effects of mechanosensory input, along with various components, on the ability of PN activity to track the temporal length of a stimulus pulse or the temporal dynamics of a train of stimulus pulses. Notably, we find that when sensory integration of both stimulus modalities is taken into account, PNs receiving both olfactory and mechanosensory input are adept at tracking the temporal length of a single stimulus pulse but are inefficient at tracking the temporal dynamics of a train of stimulus pulses (i.e., at tracking pulse frequency), while PNs receiving mechanosesory input alone (and no odor input) are unable to monitor the length of a single stimulus pulse but are proficient at tracking pulse frequency during a train of stimulus pulses. We therefore hypothesize that the ratio of these two kinds of input plays a large role in determining if a PN mainly encodes pulse duration or pulse frequency. Furthermore, we find, in general, that the ability of PNs to monitor pulse length is enhanced when the strength of the intrinsic SK current or synaptic slow inhibition is relatively low, while the efficacy of pulse frequency tracking is augmented by relatively high strength values of the SK current or slow inhibition.

In particular, this suggests an intriguing connection to the variable phase II spiking patterns of PNs. Within the moth AL, shortly after stimulus onset and following a brief initial hyperpolarization, stimulus-induced spiking ensues (phase II of the response). Experimental data shows that during phase II, spike patterns can vary broadly across active PNs; some PNs spike in sporadic bursts, while others fire more continuously, with a range of spike patterns between the two extremes. Moreover, the phase II dynamics exhibited by a PN show no spatial dependency—in fact, a significant amount of intraglomerular variability has been observed during stimulation (Christensen et al., 2000; Lei et al., 2011). In prior work with the current model (Tuckman et al., 2021), we show that, within our model, phase II spiking variability arises as a consequence of the SK current—lower SK current strength tends to yield more continuous phase II activity, while higher SK current strength results in phase II activity consisting more of sporadic bursts. The above results suggest the tantalizing possibility that phase II spiking variability may be related to the role that a PN plays in pulse tracking—i.e., a low SK strength value entails continuous phase II firing and effective tracking of pulse length (but not pulse frequency), while a high SK strength value entails burst-like phase II firing and effective tracking of pulse frequency (but not pulse length).

Moreover, these effects of the SK current on phase II activity suggested by our model are supported by pharmacological experiments. Application of bicuculline methiodide onto the AL of the male moth M. sexta leads to pheromone-responsive PNs prolonging their phase II response well beyond the duration of a pheromone stimulus (Lei et al., 2009). Interestingly, one of the effects of bicuculline methiodide is believed to be blockage of SK channels (Khawaled et al., 1999). Additionally, another modeling study where SK channels were blocked also showed prolonged excitatory responses in PNs (Belmabrouk et al., 2011).

We construct a spiking model of the AL network that strives to attain enough architectural complexity to achieve the complex dynamics of the AL while maintaining enough simplicity to allow for investigation of core mechanistic components. Below, we elaborate the components and connectivity of our model as well as the details of our analyses of model dynamics. Details are similar to those in our prior work with this model (Tuckman et al., 2021).

Publicly available datasets were analyzed in this study. This data can be found at: https://gitlab.com/HarrisonTuckmanWM/antennal-lobe-model-2-0.

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

This research was partly funded by Award 2014217, a subaward to BHS as part of the NSF/CIHR/DFG/FRQ/UKRI-MRC Next Generation Networks for Neuroscience Program.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.