Hex Parameter Assignment Satisfaction


Even simple sensory stimuli evoke neural responses that are dynamic and complex. Are the temporally patterned neural activities important for controlling the behavioral output? Here, we investigated this issue. Our results reveal that in the insect antennal lobe, due to circuit interactions, distinct neural ensembles are activated during and immediately following the termination of every odorant. Such non-overlapping response patterns are not observed even when the stimulus intensity or identities were changed. In addition, we find that ON and OFF ensemble neural activities differ in their ability to recruit recurrent inhibition, entrain field-potential oscillations and more importantly in their relevance to behaviour (initiate versus reset conditioned responses). Notably, we find that a strikingly similar strategy is also used for encoding sound onsets and offsets in the marmoset auditory cortex. In sum, our results suggest a general approach where recurrent inhibition is associated with stimulus ‘recognition' and ‘derecognition'.

Sensory systems can rapidly signal the presence of a visual1,2, auditory3,4,5 or an olfactory6,7,8 cue encountered by an animal. In addition to being rapid, the stimulus-evoked neural responses are usually elaborate, temporally patterned and tend to outlast the duration of the triggering stimulus9. The need for such dynamical neural responses is puzzling, especially considering that the behavioral response initiations can be equally fast, and delayed only by few hundreds of milliseconds after stimulus onset6. Further, another bout of strong spiking activities usually occurs after the termination of the stimulus and the behavioral relevance of this ‘OFF response' also is not understood. This apparent mismatch between the complexity in the neural encoding and the behavioral decoding raises the following fundamental question: how do neural response dynamics regulate the behavioral responses over time? More importantly, are there general rules of signal processing that are conserved across sensory systems?

A comparison of electrophysiological results reported across sensory systems of different modalities, reveal that there are striking similarities between stimulus-evoked temporally patterned neural responses9,10,11,12,13. For example, in the olfactory system, sensory input from olfactory receptor neurons (ORNs) drive spatiotemporal patterns of neural activity in the downstream neural circuits (invertebrate antennal lobe or vertebrate olfactory bulb) that are quite dynamic and information rich at the stimulus onsets and offsets9,14,15,16,17. In between these transient response epochs, when chemical cues are sustained, the ensemble neural activities in the peripheral and central regions tend to settle down to stable spiking activity patterns, and are often referred to as steady-state responses9.

Likewise, auditory stimuli elicit onset, steady-state and offset responses from the earliest brainstem nuclei through the auditory cortex. Distinct response patterns exist for specific cell types in subcortical nuclei, where the neuronal dynamics are the fastest18,19. Following several synapses of processing, auditory cortical responses also exhibit both transient and steady-state behaviour with a time scale slower than lower processing stations, but closer to the natural dynamics of common sound sources20,21,22,23.

Could these similarities observed in different sensory systems indicate a general framework for encoding/decoding information over time? Here we investigated this issue using two different sensory systems: an invertebrate olfactory system and a primate auditory system. We show that the same sensory circuit can use nearly non-overlapping sets of neurons, and different encoding formats (oscillatory versus non-oscillatory) to represent equivalent information about the identity and intensity of sensory stimulus during different response epochs (at onsets and offsets). Further, our results reveal that switching between distinct neural ensembles over time is temporally correlated with the behavioral dynamics evoked by a stimulus. Notably, our results suggest that such representations provide a potential mechanism for sensory neural networks to meet the evolving demands on the behavioral output during these epochs.


Odour-evoked ON versus OFF responses

We began by examining stimulus-evoked responses of projection neurons (PNs) in the locust antennal lobe (AL) circuit that receive direct sensory input from the olfactory receptor neurons. We used lengthy pulses of odorants (4 s in duration) in order to decouple, and examine the neural responses elicited following the stimulus onset and offset. The stimulus-evoked PN responses could be categorized into two major classes24,25 (Fig. 1a; also refer Supplementary Fig. 1a–c): increase in spiking activity limited to the periods of odour presentation (ON response), or excitatory responses that occur only in epochs following stimulus termination (OFF response). Consistent with previous findings14,26, we note that within each PN response category the temporal spiking patterns were heterogeneous.

Figure 1

Odour-evoked ON versus OFF neural responses are flexible orthogonal set.

ON and OFF responses have also been reported in both vertebrate and invertebrate visual systems27,28,29. However, a major difference between visual and olfactory ON and OFF responses is worth pointing out. In the visual system, whether a neuron responds with a light ON or light OFF type response is fixed and the ‘cell tuning' does not change in a stimulus-dependent manner28,29,30. On the other hand, in the antennal lobe circuit, we found that an individual PN can respond with either an ON or an OFF response depending on the odour identity and intensity (Fig. 1a). In addition, a comparison of neural firing rates at different processing levels reveals that these OFF responses are weak to non-existent at the level of sensory neurons, but become significant and comparable to the ON responses at the projection neuron level (Supplementary Fig. 2). Therefore, we conclude that the PN response types are not cell-specific, but arise as a result of stimulus-specific circuit interactions within the antennal lobe.

We examined the relationship between the sets of PNs that were activated during stimulus ON and OFF periods (Fig. 1b–e). We found that in general, PNs that were activated during stimulus exposure period were inhibited following stimulus termination with the firing activity reaching below baseline levels (Supplementary Fig. 1a,c). Similarly, the PNs that were activated following stimulus termination were inhibited during stimulus ON period (Supplementary Fig. 1b,c). Therefore, at an ensemble level distinct sets of PNs were activated during odour ON and OFF periods (Fig. 1b–e). In addition, we found that the OFF responses were more distributed over time rather than ON responses that had shorter response latencies (Supplementary Fig. 1d). However, it is worth noting that both in terms of the total number of spikes (across all PNs), and distribution of information rate across neurons (Supplementary Figs 1e,f and 3), both ON and OFF responses were statistically indistinguishable (two-sample Kolmogorov–Smirnov test, k=0.1625, P<0.05, n=80).

Next, we visualized odour-evoked neural activities at an ensemble level by pooling neurons across experiments9,14,15,17. Responses were aligned and binned with respect to the odour onset. Subsequently, high-dimensional response vectors were constructed, where each vector element corresponded to the spike count of a single PN in a given time bin (see Methods). To visualize the ensemble neural activity, we performed dimensionality reduction with principal component analysis. We found that each odorant generated two distinct trajectories in the neural response space (Fig. 1b–e): one during stimulus presentation (‘ON response' trajectory), and the other following stimulus termination (‘OFF response' trajectory). Plots revealing how these trajectories evolve over time are highlighted in Supplementary Fig. 4a. For all odorants examined, we found that the ON and the OFF response trajectories spanned sub-spaces that were nearly orthogonal to each other (that is, ∼90°). These qualitative results were independently confirmed by computing angular distance between high-dimensional response vectors (Fig. 1f). The generality of these results is shown using a larger odorant panel in Supplementary Fig. 5a,b.

Comparative analyses of stimulus-evoked ensemble responses

Our results clearly indicate that responses following the stimulus onset and termination are quite distinct from one another. Therefore, we next sought to examine whether OFF responses have the same specificity as the ON responses. Consistent with previous results9,14, we found that both ON and OFF response trajectories changed directions depending on odour identity (Fig. 2a). In comparison, changes in odorant intensity altered the directions of the ON and OFF trajectories only subtly14, but predominantly lengthened or shrank them (Fig. 2b; Supplementary Fig. 5c). While both the trajectories' span and length increased as odour concentration increased for some odorants (hex, 2oct and iaa), the opposite was true for bzald. These results are consistent with previous findings that both ON and OFF ensemble responses vary with, and therefore contain information about both stimulus identity and intensity9,14.

Figure 2

OFF responses vary with identity and intensity.

How different are the PN combinations activated during the ON and OFF epochs of the same stimulus when compared to PN ensembles activated by different odors or the same odour, but presented at different intensities? Since these neural circuits have been hypothesized to play a pivotal role in discriminating odorants31,32,33, we expected different stimuli to activate more distinct combinations of neurons. To understand this and quantitatively compare the similarity between ensemble responses generated in different time bins, we performed a correlation analysis14 (Fig. 3). As can be noted, the diagonal high-correlation blocks indicate that the ensemble neural activities evoked during the stimulus ON periods remained highly similar throughout the stimulus ON period. These high-correlation blocks persisted, albeit to varying levels, even when the comparisons were made between ensemble ON responses evoked by the same odorant but presented at different intensities (Fig. 3c), or between different odorants (Fig. 3d). Similarly, the ensemble neural activities evoked during stimulus OFF periods were highly correlated only among themselves (that is, the lower half of the high-correlation diagonal blocks). The off-diagonal blocks, comparing the ON and the OFF responses were the least correlated in all plots (that is, comparisons between ON and OFF responses of same odorant, different intensities and between different odorants). Furthermore, our results indicate that the combinatorial variations due to stimulus intensity or identity were less drastic, when compared to the differences in the ensemble activities at the onset and termination of the same stimulus (that is, ON versus OFF responses; Fig. 3e,f; Supplementary Fig. 5d). These results, therefore, reveal that the antennal lobe circuit emphasizes difference between stimulus onsets and offsets better than the dissimilarities between odorants.

Figure 3

ON versus OFF response similarities.

ON versus OFF responses in odour mixtures

Similar to monomolecular chemicals examined so far, we found that a binary mixture of two odorants also produced ON and OFF responses that were orthogonal to each other (Fig. 4). Predictably, the mixture trajectories appeared to be some combination of the individual odorant responses, both during the stimulus presentation, as well as after the mixture termination. Therefore, the mixture ON trajectories occupied the region between the component ON responses, and the mixture OFF trajectories projected onto the space between the OFF responses elicited by each component. These results combined with the results for complex mixtures, such as apple and mint (Supplementary Fig. 5), corroborate our conclusion that these observations regarding ON and OFF responses are general features of odour-evoked neural activities and are not limited to monomolecular chemicals.

Robustness of OFF responses

Apart from identity and intensity, naturally encountered odorant plumes also tend to vary in stimulus length34. We next examined how invariant were the OFF responses that followed the same stimulus delivered for different durations. We found that the orthogonal relationship between the ON and OFF responses was maintained independent of the stimulus pulse duration (Fig. 5a). Furthermore, consistent with prior results9,35, the odour response trajectories for different stimulus durations were well aligned during both the response onsets and offsets. Therefore, the ON and OFF response templates obtained for one odour pulse duration (4 s; see Methods), pattern-matched with ensemble neural activities evoked by the same odorant presented for different durations (Fig. 5b,c). Note that the ensemble response vectors during the entire odour presentation period pattern-matched only with the ON responses, but the response switched and gained similarity with the OFF response template after stimulus termination. On the other hand, different odorants evoked response patterns that were distinct from each other during both ON and OFF response periods (Fig. 5; hex versus bzald).

Figure 5

Classification analyses of ensemble neural activity.

In sum, these results suggest that OFF responses are as consistent as the ON responses, and they actively convey information about the termination of a particular stimulus at a specific intensity. Further, our results also show that neuronal networks can use two minimally overlapping sets of neurons to represent equivalent information about a stimulus during different epochs.

Engaging and disengaging recurrent inhibition

Apart from using different neural ensembles, are there other differences that distinguish the ON and the OFF responses? To understand this, and to gain mechanistic insights, we made intracellular recordings from GABAergic local neurons (LNs) and cholinergic PNs in the antennal lobe, while simultaneously monitoring the local field potential activity in the mushroom body (the neural circuit downstream to the antennal lobe). Although ON and OFF responsive local neurons have been reported in other model systems36, consistent with published results in locusts37, we found that most local neuron responses were limited to the odour onset period. We also found that odour exposures entrained oscillatory activity both in individual local neurons and in the local field potential38 (Fig. 6a,b). Further, these local neuron responses and field potential oscillatory responses were limited to the odour presentation period (that is, only during ON response epochs; Fig. 6b,c; Supplementary Fig. 6a). Notably, we found that the local neuron activity remained phase locked with field potential activity only when the stimulus was presented (Fig. 6d). Intracellular recordings from individual PNs were largely consistent with what we had observed in our extracellular datasets and most PN spikes occurred either during or after the stimulus duration (Fig. 6e–g).

Figure 6

Engaging and disengaging recurrent inhibitory network.

We pondered if the recruitment of inhibition could simply arise due to differences in the strength of the odour-evoked responses observed during these epochs. Therefore, we first compared the average spike counts across all recorded PNs during the ON and OFF epochs (Fig. 6h). We found that the PN responses had two distinct peaks, one following odour onset and the other following odour offset. PN spiking activity weakened considerably between these two transient response periods (that is, the sustained/steady-state responses). Interestingly, this weak sustained response was still sufficient to evoke local neuron activity and local field potential oscillations. In comparison, even though the OFF responses were considerably stronger than the neural activity just before the end of the odour pulse, it failed to entrain coherent field potential oscillatory activity (Fig. 6g). Furthermore, as mentioned before, a comparison of cumulative spike counts during the ON and OFF epochs revealed that the spike counts during these time periods were comparable (Supplementary Fig. 1e,f). Therefore, the overall strength of spiking activities across PNs alone appears to be a poor indicator of whether or not the local field potential oscillations are generated by the AL circuitry.

In our earlier work17, we found that the sensory input from ORNs did not have a strong bout of spiking activities after termination of the odorant as was observed in the PNs. Could this difference in the presence/absence of sensory input alone explain the limited entrainment of field potential oscillations during the odour exposure period? First, we note that the presence of strong input from sensory neurons was a good indicator of whether LFP oscillations were present in the AL. However, consistent with the existing data37, we found that application of picrotoxin, a GABAA antagonist, alone can reversibly abolish the field potential oscillations (Supplementary Fig. 6b–e). Note that this pharmacological manipulation did not impact the sensory input to the antennal lobe circuits, but rather blocked the fast inhibition from the local neurons onto the projections neurons. Combining these two observations, we conclude that input from sensory neurons is necessary for recruitment of inhibition from local neurons that then allow generation of oscillatory field potential activity.

In sum, these results indicate that the ON and the OFF responses significantly differ in their ability to engage the local inhibitory circuits, which are necessary for oscillatory synchronization of PN responses. Hence, we conclude that although the ON and the OFF responses have qualitatively similar information content, their neural encoding formats vary significantly.

Generating ON and OFF responses in a computational model

To further understand the mechanisms, we developed a well-constrained computational model of the early olfactory circuits (see Methods, Fig. 7; Supplementary Fig. 7). The AL model had the following components: (i) feed-forward input from ORNs onto PNs and LNs (ii) recurrent connections between LNs and PNs (iii) a bi-directional adaptive mechanism in individual PNs. Consistent with published results26,39, we found that recurrent inhibition from local neurons was the essential and sufficient component to generate results similar to our in vivo observations: 20 Hz field potential oscillations and phase locking of excitatory and inhibitory ensemble activities only during stimulus ON epoch (Supplementary Fig. 7e,f). Without these recurrent inhibitory connections from LNs onto PNs, the model did not generate any oscillatory field potential activity (Fig. 7c). This dependence of local neuron activity on stimulus-evoked input limited the recruitment of recurrent inhibition and therefore entrainment of the field potential oscillations to the duration of the stimulus exposure.

Figure 7

Modelling of ON–OFF neural activity.

While the model with recurrent inhibition alone was sufficient to generate LFP oscillations, as can be noted, the spiking activities in PNs were limited to the epochs, when strong ORN input was available (that is, no OFF responses). Therefore, a bidirectional adaptation mechanism was added to individual PNs that reduced the excitability following high-firing epochs, and at the same time increased the excitability following periods of hyperpolarization. Such adaptive control of neural excitability was necessary to generate orthogonal ensemble activities during stimulus onset and offset (Fig. 7f). This allowed the model to generate the PN OFF responses even when the sensory inputs were decaying back to baseline (that is, weak). We note that to keep the simulations simple and consistent with our electrophysiology data (Supplementary Fig. 2), we did not include any OFF-responsive ORNs in the simulations shown here. In addition, note that the LN inhibition was absent during these periods, as a strong ORN input was necessary in the model to recruit recurrent inhibition. Therefore, the PN OFF responses in the model were also desynchronized, thereby reducing power in the oscillatory field potential activity (Fig. 7d). Bifurcation analyses (Fig. 7e,f) indicated that the strength of local neuron inhibition was the only parameter that controlled the power of the entrained field potential oscillations in the model, whereas a strong inhibition from local neurons and the bi-directional adaptation of PN excitability were both necessary for generating distinct ON and OFF responses.

Hence, our modelling study suggests that stimulus-dependent engagement and disengagement of recurrent inhibition in the antennal lobe circuits provides a simple mechanism for generating distinct ON and OFF neural activities with differing response formats (oscillatory versus non-oscillatory).

Behavioral relevance of ON and OFF responses

Are the response patterns observed at the odour onset and offset relevant to odour-evoked behaviour? Earlier studies in rodents and insects have shown that odour recognition can be rapid and usually happens within a few hundred milliseconds of stimulus onset7,17,40. On the basis of these results, the early portions of only the ON responses can be expected to play a role in stimulus recognition. What then is the need for another round of stimulus-specific neural activity after odour termination? We sought to examine this issue using an appetitive-conditioning assay. During the training phase, starved locusts were presented with an odorant (conditioned stimulus) followed by a reward (wheat grass; see Methods). We found that locusts reached their asymptotic performance levels after six training trials17,41. Following training, locusts were tested in an unrewarded test phase. Locusts that learned the association between the odorant and the reward opened their maxillary palps following the presentation of the conditioned stimulus in anticipation of the reward. Consistent with previous studies, locusts retained the learned association even when tested multiple times in the unrewarded test phase17,41. To quantify the behavioral palp-opening response, we painted the distal end of the locust palps with a non-odorous green paint and tracked their whereabouts with fine spatial and temporal resolution (Fig. 8a; see Methods).

Figure 8

Stimulus-evoked OFF responses are required for behavioral reset.

We found that the palp-opening responses were quick to start and the palps were kept open, as long as the conditioned stimulus persisted (Fig. 8a). The behavioral responses generalized independent of the duration of the conditioned stimulus (note only a 10 s hexanol pulse was used to train the locusts; see Methods). More importantly, we found that the periods during which the ensemble neural activities pattern-matched with the ON responses corresponded to epochs when the palps were opened and usually kept open (Figs 5c and ​8a). In contrast, time segments when the palps closed correlated with those epochs, when neural activity gained similarity with the OFF responses (Figs 5c and ​8a). In sum, these results suggest two possible models for translating population neural activity into palp-opening and palp-closing responses: (i) ON model: gaining or losing pattern-match with ON responses underlies palp-opening and palp-closing responses, respectively; and (ii) ON–OFF model: pattern-match with the ON responses triggers behavioral response onset, whereas pattern-match with OFF responses is necessary for terminating the behavioral responses. We found that both the ON and the ON–OFF model could generate predictions consistent with the observed palp opening and closing responses for hexanol presentations of different durations (Fig. 8b,c). Note that we also explored two other model variants for completeness (Supplementary Fig. 8a,b).

ON–OFF model is a better predictor of behavioral output

We sought to test these models for translating neural activity to behavioral output by perturbing the pattern-match with the ON responses. To achieve this, we first presented the trained odorant (hex) and a distractant (bzald), as a binary mixture whose components were delivered synchronously (Supplementary Fig. 9). We found that the PN response to this mixture stimulus was dominated by a single component (hex); however, the pattern match with hex ON response templates (solitary presentations) was diminished (Supplementary Fig. 9a,b). Matching these classification analyses results, we found that locusts trained with hex, responded to the binary mixture of hex and bzald with a similar reduction in POR (Supplementary Fig. 9c). These results clearly demonstrate that reduction of pattern match with the ON response template of conditioned stimulus diminishes the behavioral POR responses.

Next, we presented the same two odorants in series such that onset of the distractant (bzald; untrained odorant) happened 500 ms before the termination of the conditioned stimulus (hex; Fig. 8d). Consistent with previous findings17, and unlike the synchronously presented binary mixture case, the neural activity remapped to gain pattern-match with the second odorant in the sequence (that is, bzald) following its onset. However, following the termination of the conditioned stimulus (hex), we found that the ensemble neural activity again remapped to gain similarity with the OFF response of the first odorant (that is, hex; Fig. 8d; Supplementary Fig. 10a).

Interestingly, we found that the palp-opening response to hexanol (the conditioned stimulus) did not end, when a distractant was introduced (Fig. 8e). Rather the closing of palps began after the termination of the conditioned stimulus following epochs, when pattern-match with hexanol OFF responses was observed (Fig. 8e). Therefore, the amount of time it took for closing palps following a mismatch with the ON responses doubled across the two conditions tested (Fig. 8e,f).

Two other observations are worth pointing out here. First, the distractant presented solitarily did not evoke a significant palp-opening response (Fig. 8a). So the prolongation of the POR after introduction of bzald cannot be explained based on the ongoing PN activity during this epoch, as its pattern-matched with the bzald ON template (Fig. 8d). Second, it is worth noting that in the overlapping sequence, the degree of pattern match with the hex OFF response template was diminished due to the presence of the distractant. Matching this physiological result, we found that the POR response termination was also slower than that observed in the case of solitary hex presentations (Supplementary Fig. 10b,c). Therefore, these results suggest that after the palps have been opened, a pattern-match with the ON responses may not be necessary for sustaining the behavioral response. More importantly, gaining pattern-match with the OFF responses of the conditioned stimulus is a good indicator of the palp-closing response dynamics. This interpretation is supported by the modelling results, which revealed that only the ON–OFF model was able to generate consistent palp-closing behaviour across conditions (Fig. 8g, Supplementary Fig. 8c). Furthermore, we found that POR to the hex-0.5s overlap-bzald stimulus sequence can be better predicted using results from the classification analysis than those directly made using the POR data. In other words, the time series of ensemble neural activities was Granger causal with the behavioral POR evolution over time (fNeural->POR=7.37, P<0.05, n=80; see Methods).

Finally, to further confirm this hypothesis, we presented the conditioned stimulus (hex) in a pulsatile fashion (Fig. 9). We found that the ON and OFF responses precisely encoded the presence and absence of the hexanol puffs. As can be predicted, trained locusts opened or closed palps during epochs when hex ON and OFF responses were observed, respectively. Taken together, these results strongly support our hypothesis that orthogonal neural activities may underlie opposing behavioral responses in this olfactory system.

Figure 9

Neural and behavioral responses to odour pulses.

ON versus OFF responses in the marmoset auditory cortex

Finally, we wondered how general are these signal processing features observed in the insect olfactory system? To understand this, we examined the response of cortical neurons in the marmoset monkey auditory cortex (A1 area) to monotones (0.5 s duration) (Fig. 10). Analogous to results in the olfactory system, we found that the same neuron could respond with an ON or an OFF response to monotones depending on the monotone frequency (Fig. 10b). In addition, it might be worth noting that primary auditory neurons in mammalian auditory system do not have the ON and OFF opponent responses observed in the visual system11,42,43. Therefore, we conclude that A1 neurons are ‘network tuned', as in olfaction and not ‘cell tuned' as in vision.

Figure 10

Sound-evoked ON versus OFF neural responses in the marmoset auditory cortex.

At the ensemble level, we found that cortical neuron responses to prolonged monotones were not sparse and a large fraction of the recorded neurons had a statistically significant response, when both ON and OFF response epochs were considered (Fig. 10c–h, left panels). Minimally overlapping ensembles of neurons were activated by the same sound following its onset and offset. This again resulted in two orthogonal neural response trajectories for each monotone, one following the sound onset to encode its presence and the other following sound offset to indicate its termination (Fig. 10c–h, right panels, Fig. 10i). In sum, these results reveal that locust olfactory circuits and marmoset auditory cortical circuits may employ conserved processing principles to actively encode stimulus presence and termination.


A behavioral response initiated by any sensory stimulus has to be reset after its termination. In most cases, the response onset (deviation from baseline) following the stimulus introduction, and the reset (return to baseline) following its termination are by necessity opposites of one another. Is the behavioral response reset actively brought about by the neural circuitry, or is it a result of a passive return of stimulus-evoked activity to the spontaneous level? Two lines of evidences appear to suggest that a more direct representation of the stimulus absence will be necessary in most sensory systems. First, sensory memory following stimulus encounters may persist even after the termination of the stimulus26,44,45,46. Second, in natural settings, sensory cues are mostly encountered in overlapping sequences, and a passive return to baseline may not happen until after all of the succeeding stimuli terminate. Furthermore, in sensory systems, absence of a stimulus can be as informative as their presence (light versus dark47,48 or heat versus cold in temperature sensing49,50). Taking into account that most sensory stimuli generate another round of transient activity following stimulus termination9,11,16, it would appear that an active signal regarding the absence of stimulus is available in many sensory systems.

For the OFF responses to encode stimulus absence, the neural activities during this epoch must be different from the ON responses and exclusively encode for each stimulus. Our results indicate that the ensemble neural activities at sound and odorant offsets were nearly orthogonal to (that is, independent from) the ON responses. Nevertheless, both these neural activities during stimulus onsets and offsets were able to uniquely encode for identity and intensity of a sensory cue. Importantly, while the onset responses were necessary for the initiation of the behavioral response (‘the palp-opening response'), our results reveal that the offset responses are necessary to actively terminate it (‘the palp-closing response'). Thus, orthogonal neural activities encoded for the presence and absence of a stimulus, and were translated to generate behavioral responses that were opposites of one another (start/onset versus stop/reset). Such mapping of distinct neural activities to generate behavioral responses that are opposites of one another have indeed been shown in a number of neural systems27,49,50,51,52,53,54,55,56. Our work is the first to show that a single sensory stimulus can activate unique and independent sets of neurons during and after its presentation in order to meet opposing behavioral output demands during these epochs.

Significantly, our results from the invertebrate olfactory circuit and the primate auditory cortical circuit are strikingly similar. Considering that these two sensory circuits differ in their modality, complexity, evolutionary origin, and their position in the sensory information processing hierarchy, we speculate that our results reveal a conserved information processing approach to signal stimulus presence and absence in a wide variety of sensory systems.


Odour stimulation

Neat odour solutions (Sigma-Aldrich) were diluted in mineral oil to their 1 or 0.1% concentrations by volume (v/v). The diluted odour solution was placed in a 60 ml sealed glass bottle with separate inlet and outlet lines. A pneumatic pico-pump (WPI Inc., PV-820) was used to displace a measured volume of the odour-bottle headspace (0.1 l min−1) that was then injected into a desiccated and filtered carrier air stream (0.75 l min−1) directed towards the locust antenna. A vacuum funnel was placed right behind the locust antenna to provide a constant flux and ensure removal of the delivered odour vapours. Each odorant was presented in a pseudorandom manner (blocks of 10 or 25 trials) with 60 s inter-trial intervals and 15 min inter-block intervals. The following odorants were used in this study: hexanol (hex), 2-octanol (2oct), isoamyl acetate (iaa), benzaldehyde (bzald) and binary mixtures of hexanol–2-octanol and isoamyl acetate–benzaldehyde. Each odorant was delivered at their 0.1% or 1% dilutions (v/v). The pre-mixed binary mixture contained vapours of individual components each at 0.5% (v/v) dilution levels.

Olfactory electrophysiology

Young locusts (Schistocerca americana) with fully developed wings (post fifth instar) of either sex were selected from a crowded colony. Locusts were immobilized with both antennae intact, and then the brain was exposed, desheathed, and continually perfused with locust saline as demonstrated previously38,57,58. In vivo extracellular recordings from the antennal lobe were performed using 16-channel, 4 × 4 silicon probes (NeuroNexus). Electrodes were gold plated such that their impedances were in the 200–300 kΩ range. The extracellular signals were acquired using a LabView data acquisition system. Raw extracellular signals were collected at 15 kHz sampling rate, amplified at 10 k gain using a custom made 16-channel amplifier (Biology Electronics Shop; Caltech, Pasadena, CA), filtered between 0.3 and 6 kHz before spike sorting.

Intracellular recordings were performed as previously described35. Briefly, sharp glass micropipettes were filled with 0.5 M potassium acetate solution to achieve impedance in the range of 50–150 MΩ. Voltage signals were amplified (Axoclamp-2B, Molecular Devices) and saved at 5 kHz sampling rate using Labview. Local field potentials were recorded simultaneously using saline-filled glass microelectrodes (4–10 MΩ) and low-pass filtered using a DC amplifier (<100 Hz, Brown-Lee Model 440)59.

Olfactory neural datasets

Recordings for different pairs of odorants (dataset1: hex–2oct and dataset2: iaa–bzald, n=25 trials each) were made separately (that is, different sets of PNs). In addition, PN recordings to monitor the responses to different durations of hexanol and to characterize the response patterns evoked by overlapping presentations of hexanol and benzaldehyde were collected separately (dataset3: hex durations and hex-bazld overlap, n=10 trials). A fourth PN response was collected to compare ensemble responses to a six puffs of hex and bzald, a binary mixture of those two chemicals, and a 2s ON—2s OFF—2s ON pulsatile delivery of hexanol (dataset4: hex and bzald, n=10 trials). In all the neural datasets, the delivery of odorants was pseudo randomized in blocks of 25 or 10 trials.

To test the generality of our results, we reanalyzed a previously published dataset17 of PN responses to a wide range of analytes (only used in Supplementary Fig. 2).

PN spike sorting

Spike sorting was done using a conservative approach described in earlier works8,17,60. In brief, the following criteria were used for the single-unit identification: cluster separation >5 noise s.d., number of spikes within 20 ms<6.5%, and spike waveform variance <6.5 noise s.d. Using this approach, a total of 329 PNs were identified from 41 locusts.

Sorting of PN response

PN responses were sorted in Fig. 1b–e based on the following metric:

Nonresponsive neurons were identified and moved to the bottom of these plots. Nonresponsive neuron criterion was similar to that used before17 with only exception that here the entire time window involving both the ON and the OFF responses was taken into consideration.

Dimensionality reduction analysis for PN responses

We used the linear principal component analysis technique for the purpose of visualizing high-dimensional neural response trajectories. For this analysis, we binned the PN responses in 50 ms non-overlapping time bins and averaged the responses in a given time bin across trials. This resulted in a time series matrix of n neurons (rows or dimensions) and m time steps (columns) for each odorant. When comparing trajectories elicited by different odorants, data matrices obtained for different odorants were concatenated to increase the number of columns. The resulting high-dimensional vector in each time bin was projected along the eigenvectors of the n × n response covariance matrix. Low-dimensional data points that represented response vectors in adjacent time bins were connected to generate low-dimensional response trajectories. Finally, the response trajectories were smoothed using a three-point moving average filter.

A time window that included both ON and OFF responses (twice the length of the odour pulse) was used while comparing response trajectories evoked following stimulus onset and termination (Figs 1b–e, ​4a and ​5a, Supplementary Figs 4,5a). For comparison of ON or OFF responses evoked by different odorants and by the same odorant at different concentrations, only a 4 s window comprising of either the ON or the OFF responses were exclusively used (Figs 2a,b and ​4b, Supplementary Fig. 5c). This was required as differences between the ON and OFF responses of the same stimulus was the dominant source of variance in the data set.

Note that this dimensionality reduction analysis was used only for qualitative purposes. All quantitative analyses were performed using the high-dimensional PN response vectors.

Angle between mean ON and OFF projection neuron responses

High-dimensional PN response vectors were generated using all recorded neurons. The mean baseline response during a 2 s pre-stimulus period, immediately preceding stimulus onset, was subtracted from all response vectors. The high-dimensional response vectors were averaged over the entire duration of the odour pulse (4 s) to generate the mean ON response template (WON). Similarly, the high-dimensional vectors were averaged for a 4 s period following the odour pulse termination to generate the mean OFF response template (WOFF). The angle between the mean ON and OFF responses were computed as follows:

Different analysis windows (2 s and 4 s) were used to ensure that the orthogonal relationship between these two response templates were insensitive with respect to the time bin size (Fig. 1f; Supplementary Fig. 5b).

For Fig. 3e,f and Supplementary Fig. 5d, the comparisons (cosine of the angle obtained from equation 1) were made either between mean ensemble activities (2 s window) during different epochs of a single stimulus (ON versus OFF), or between the ON and OFF responses evoked by two different stimuli.

Classification analysis

A bin-by-bin, trial-by-trial classification analysis8,17 was used to determine the pattern-match between PN responses observed in a particular time bin with the ON and OFF response templates of a particular odorant (Figs 5b, ​8d and ​9b, Supplementary Fig. 9b). Note that the ON and OFF templates were generated using solitary hexanol or benzaldehyde exposures using spike counts in the 2 s time windows immediately following stimulus onset and offset, respectively.

An angular distance metric was used to determine the nearest reference template. Each time bin in a test trial was classified as belonging to one of the following response categories: hexanol ON, hexanol OFF, benzaldehyde ON, benzaldehyde OFF or as an unclassified response. Those time bins that were not within a certain angular distance threshold (within 63° of the nearest reference template for Figs 5b and ​8d and within 67° for Fig. 9b; Supplementary Fig. 9b) were categorized as unclassifiable responses. This threshold was chosen such that <10% of the ensemble neural activities in the pre-stimulus period were misclassified as being similar to the hexanol or benzaldehyde response templates.

Information rate estimation

We estimated the information content carried by the neural spike trains during ON and OFF response windows by computing the mutual information rate between odour stimulus and the neural response61. We used the ‘direct method' approach by finding the difference between the total and conditional entropy rates of the responses62,63.

The total entropy rate (HTotal) was estimated using PN responses to five unique stimuli, and the conditional entropy rate (Hnoise) was obtained from 25 repeated presentations of the same odorant (Supplementary Fig. 3). The unique stimuli used were hexanol 1%, 2-octanol 1%, hexanol 0.1%, 2-octanol 0.1% and the binary mixture of hexanol 1% and 2-octanol 1%.

Cluster analysis for PN responses

For clustering PN responses, we first binned each PN spiking response in 50 ms non-overlapping time bins (smoothed with a five-point average moving average filter). The PN responses over an 8 s period starting at the odour onset (160-dimensional vector) were then trial-averaged. All PNs with a statistically significant response (excitatory or inhibitory) were used for this cluster analysis. Responses recorded for the following four odorants were analysed: hex, 2oct, iaa and bzald (at 1% concentration v/v). PN responses were clustered such that the furthest pairwise distance between any two samples assigned to an individual cluster was minimized. A correlation metric was used as a measure of similarity:

where xi and yi are ith vector elements of two different PN response vectors, and denote the mean firing rate for each PN over the entire 8 s window, and represent their s.d. The optimum number of clusters required to represent the entire data set was chosen based on the mean-squared error (Supplementary Fig. 1a–c). Peak latency was calculated for the ON and OFF responses by finding the time bin with maximum firing rate after baseline subtraction (Supplementary Fig. 1d).

Computational modelling of the locust antennal lobe

Odour representation in the antenna was modelled with a repertoire of 50 ORNs. A subset of ORNs was activated by the stimulus, as shown in Supplementary Fig. 7a. Note that the sensory neuron response time constants for the rise, adaptation and fall phases were heterogeneous as found in vivo.

Next, the modelled sensory neuron responses (ORN responses) were input to a realistic computational model of the antennal lobe circuits with 50 excitatory projection neurons (PNs) and 25 inhibitory local neurons (LNs). Each PN was modelled as a regular spiking neuron and inhibitory local neuron as a fast-spiking neuron using a reduced Hodgkin–Huxley model64.

if v=30 mV then v=c, u=u+d

PN model parameters: a=0.02, b=0.2, c=−65, d=8.

LN model parameters: a=0.1, b=0.2, c=−65, d=2.

I is the total input to the each neuron from both sensory neurons, as well as summed contributions of other antennal lobe neurons. Note that the adaptive parameters (vthresh and vmemory) were limited to PNs only. The update rule for these two parameters is as follows:

where inc=0.3 and τthresh=2,500 ms for all PNs, ts is the time when the neuron last fired an action potential, and is the Dirac delta function. Integration time step is 1 ms.

where v(t) is the membrane potential of the neuron at time t and h is a one-sided Gaussian kernel with s.d. uniformly distributed in the range (120, 320 ms).

Model connectivity

We modelled each PN to receive input from a single sensory neuron. LNs received input from nearly two-thirds of all sensory neurons. Further, since each LNs arborized extensively throughout the antennal lobe65, each local neuron received excitatory input from roughly 30% of PNs, and provided feedback inhibition to ∼30% of non-identical combination of PNs. Note that there were no excitatory lateral interactions between PNs or self-inhibition in the model. These connection probabilities and other network parameters including the type of synaptic currents were constrained based on estimates from locust antennal lobe circuits65,66,67,68. The connectivity matrix used in Fig. 7 is shown in Supplementary Fig. 7b–d.

The post-synaptic current generated by a pre-synaptic neuron i following a spike at time t was defined as follows:

where z(.) and g(.) are low pass filters of the form exp(−t/τsyn) and t × exp(−t/τsyn), respectively, τsyn is the synaptic time constant, gnorm is the peak synaptic conductance (a constant), and spk(i,t) marks the occurrence of a spike in neuron i at time t. Synaptic parameters used were the following: peak synaptic conductance (excitatory synapse)=0.1 nS, excitatory synapse response time constant=5 ms, peak synaptic conductance (inhibitory synapse)=0.3 nS, inhibitory synapse response time constant=6 ms.

Therefore, the total synaptic current received by neuron k from all other neurons in the network is given by:

The Professional Military Network

Can anyone explain the Army Satisfaction Key Background Tab?

In ASK, I have this in my On Assignment tab (I'm on orders to be a DA-Select Recruiter):

Report Date: 20170710
Location: FT KNOX

But in the Background tab, I have this:

Duty: Recruiter
Type Code: CLEARED
Agency: None
Location: N/A
Date: 11/9/2017

My specific questions:

1) What exactly does "CLEARED" mean for the Type Code?
2) What does the Date mean? Whether that's November 9, 2017, or 11 September 2017, neither makes sense to me. I'm scheduled for Recruiter School 10 April - 26 May 2017, with a report date to my gaining unit on 10 July 2017. I'm in the process of requesting a deferment because I'm in the middle of having multiple surgeries, but the paperwork hasn't gone past my company yet, so it can't be that.

Any ideas would be very much appreciated!
Categories: 1

0 Replies to “Hex Parameter Assignment Satisfaction”

Leave a comment

L'indirizzo email non verrĂ  pubblicato. I campi obbligatori sono contrassegnati *