HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Supplemental Data All Versions of this Article: 92/1/236    most recent 01132.2003v1 Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (16) Citing Articles via Google Scholar Google Scholar Articles by Daly, K. C. Articles by Smith, B. H. Search for Related Content PubMed PubMed Citation Articles by Daly, K. C. Articles by Smith, B. H.

Molecular Features of Odorants Systematically Influence Slow Temporal Responses Across Clusters of Coordinated Antennal Lobe Units in the Moth Manduca sexta

Department of Entomology, Ohio State University, Ohio 43210

Submitted 24 November 2003; accepted in final form 18 February 2004

	ABSTRACT

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
Behavioral studies of olfactory discrimination and stimulus generalization in many species indicate that the molecular features of monomolecular odorants are important for odor discrimination. Here we evaluate how features, such as carbon chain length and functional group, are represented in the first level of synaptic processing. We recorded antennal lobe ensemble responses in the moth Manduca sexta to repeated 100-ms pulses of monomolecular alcohols and ketones. Most units exhibited a significant change in spike rate in response to most odorants that outlasted the duration of the stimulus. Peristimulus data were then sampled over 780 ms for each pulse of all odorants. Factor analysis was used to assess whether there were groups of units with common response patterns. We found that factors identified and represented activity for clusters of units with common temporal response characteristics. These temporally patterned responses typically spanned 780 ms and were often dependent on carbon chain length and functional group. Furthermore, cross-correlation analysis frequently indicated significant coincident spiking even during spontaneous activity. However, this synchrony occurred mainly between units recorded on the same tetrode. In a final analysis, the Euclidean distance between odor responses was calculated for each pair of odorants using factors as dimensions. The distance between responses for any two odorants was maximized by 240 ms. This time course corresponded to the brief sequence of coordinated bursts across the recorded population. The distance during this period was also a function of systematic differences in molecular features. Results of this Euclidian analysis thus directly correlate to previous behavioral studies of stimulus generalization in M. sexta.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
Behavioral studies with a phylogenetically diverse group of animals have demonstrated that systematic changes in either molecular features or in blends of odorants directly correlate to odor discriminability (Cleland et al. 2002; Daly and Smith 2000; Daly et al. 2001a,b; Fine-Levy et al. 1988; Laska et al. 2000; Smith and Menzel 1989). It has been proposed that olfactory systems discriminate odors based on the interaction of specific molecular features, or odotopes, of odorant molecules with relatively few types of olfactory receptor neurons (ORN) (Buck and Axel 1991; Shepherd 1987; Vosshall et al. 2000). Individual ORNs that express the same seven-trans-membrane receptor protein converge on one or a limited number of glomeruli in the olfactory bulb (OB) of vertebrates or antennal lobe (AL) of insects (Bhalerao et al. 2003; Mombaerts 1999; Ressler et al. 1994; Vassar et al. 1994; Vosshall et al. 2000). Furthermore because each ORN responds to a range of odorants possessing similar molecular features (Firestein 2001; Hildebrand and Shepherd 1997; Malnic et al. 1999; Shields and Hildebrand 2001), each glomerulus may thus participate in the processing of several odorants. This "combinatorial" sensory input produces a multiglomerular spatial pattern of activation in the OB and AL (Galizia et al. 1999; Hildebrand 1996; Hildebrand and Shepherd 1997; Johnson et al. 1998; Kashiwadani et al. 1999; Rubin and Katz 1999; Sachse et al. 1999) that is correlated to an animal's ability to discriminate among many odorants (Linster et al. 2001).

ORN input is commonly described as phasic-tonic, rising and falling with the onset and offset of odor presentation (e.g., Friedrich and Laurent 2001). This initial input is then transformed into a temporally complex pattern of AL output (Christensen et al. 1998; Friedrich and Laurent 2001; Stopfer and Laurent 1999; Wehr and Laurent 1996) that could enhance odor discrimination (Stopfer et al. 1997). Glomeruli are interconnected by a network of inhibitory local interneurons (LN) in the AL and OB (Homberg et al. 1989; Shipley and Ennis 1996), which mediate at least two different types of temporal responses. First, responses of AL projection neurons (PN) to prolonged odor stimulation consist of slow, reproducible patterns of bursting and silence mediated by the spread of inhibition among glomeruli (Christensen et al. 1998; Friedrich and Laurent 2001; Wehr and Laurent 1996). Bursting patterns have been shown to be synchronous among pairs of output neurons that arborize the same glomerulus (Lei et al. 2002) and may indicate a functional coupling of these neurons (Schoppa and Westbrook 2001, 2002).

A second type of temporal response, which is also LN mediated and may enhance odor discrimination, are 20- to 40-Hz local field oscillations. These oscillations have been observed in the AL and OB and/or in their projection fields (e.g., Kashiwadani et al. 1999; Laurent 2002; Laurent and Davidowitz 1994; Laurent and Naraghi 1994; Macleod and Laurent 1996; Wehr and Laurent 1996). In the honeybee, disruption of GABA_A transmission in the AL increased generalization from a conditioned odorant to a closely related testodorant and abolished local field oscillations in AL projection fields (Stopfer et al. 1997). These researchers thus concluded that transient oscillatory synchrony was the mechanism through which fine odor discrimination was mediated. However, imaging studies using the same pharmacological techniques also show disruption of spatial patterns of glomerular activation in honeybee AL (Sachse and Galizia 2002). These two studies therefore suggest that the spatial and temporal aspects of an odor-driven response are confounded.

Thus it remains unclear whether temporally patterned olfactory output, generated by the intrinsic properties of the AL or OB circuitry, contributes to the code for odor identity (Friedrich and Laurent 2001; Laurent et al. 2001; Wehr and Laurent 1996). Extrinsic properties of a given stimulus other than molecular identity, such as its concentration, duration, or intermittency, can also affect both slow and fast temporal response patterns in the AL (Christensen et al. 1993, 1998, 2003; Lei et al. 2002; Stopfer et al. 2003; Vickers et al. 2001). Therefore our objective was to evaluate whether temporal response patterns among populations of neurons in the AL of the moth, Manduca sexta, were correlated to molecular features of odorants, while keeping constant, these extrinsic properties of the stimulus (i.e., concentration, duration and intermittency; Christensen et al. 1998, 2000; Lei et al. 2002; Vickers et al. 2001).

Our analyses test two basic hypotheses. First, we test whether information about odor is encoded in the AL by one or more subgroups of neurons, which produce correlated response patterns (Lei et al. 2002; Schoppa and Westbrook 2001, 2002). Our prediction is that by virtue of their correlated responses over time, subgroups of units will produce a common factor using factor analysis. Factor analysis should therefore account for a fraction of the variance across a large number of recorded units. Alternatively, units might function independently of one another, in which case factor analysis will fail to identify subgroups.

Our second hypothesis is based in part on our prior behavioral results in the moth M. sexta (Daly et al. 2001a). This work demonstrated that a conditioned behavioral response generalized from the conditioning odorant to test odorants as a function of differences in carbon chain length and functional group. We now hypothesize that details of the slow temporal response patterns, produced by either individual units or subgroups of coordinated units, will also vary as a function of changes in the molecular features of odorants. This hypothesis will be tested in two ways. First, we test whether temporal response patterns, produced by either individual units or identified subgroups of coordinated units, differ as a function of the molecular features of the presented odorant. This can be tested statistically by analysis of the individual units or factors for time by odor interactions. Second, across subgroups of coordinated units, which respond in a significantly odor-dependent manner, we can characterize differences in their collective responses as a function of odor features. Differences in population responses can be characterized by calculating the Euclidian distance between population response trajectories. These methods allow us to assess the extent to which the population response varies as a function of odorant features. In addition, these methods will also allow us to characterize when, in response time, odor-dependent population responses diverge.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
Animals

Larvae of M. sexta (Lepidoptera: Sphingidae) were reared on artificial diet (Bell and Joachim 1976) under a long-day photoperiod (17/7 L/D). Pupae were shipped during late-stage development from Arizona Research Laboratories Division of Neurobiology to the Rothenbuhler Honeybee Research Laboratory. On arrival, pupae were isolated in paper bags and stored in an environmental chamber held at 28°C and 90% relative humidity on a 16/8 L/D cycle. Pupae were checked daily at the start of the dark cycle. A date was recorded for each newly emerged adult to track age. Experiments were performed on adults within 5-7 days posteclosion to ensure full development of the olfactory system.

Preparation and treatment of subjects

Subjects were placed into individual copper preparation tubes and restrained with molten soft wax. This method securely positioned the head for dissection and stable extracellular recording. To expose the ALs, cuts were made in the cuticle of the head capsule 1 mm on either side of the sagittal midline. These cuts extended forward from the posterior coronal seam just under 4 mm. There is no muscle attached to this portion of the cuticle, and thus it was easily removed. Next, a cut around the perimeter of the pharyngeal dilator muscle attachment site on the head capsule was made. This region extends from just behind the posterior coronal seam to approximately two-thirds the distance to the extreme posterior end of the head capsule (Eaton 1971). The freed portion of cuticle with attached muscle was slid forward into the previously opened notch and attached with superglue. Additional cuts were made to remove the remaining portion of cuticle at the very posterior end of the head capsule. All occluding tracheal and vascular material was removed to expose the protocerebrum and ALs on both sides. Finally, the completed preparation was fixed to a stage, the opening in the head was perfused with physiological saline (Christensen et al. 1993), and a 16-channel silicon microprobe (see following text) was inserted into the AL with a micromanipulator.

Odorant selection and delivery

Our previous behavioral analysis of odor discrimination in M. sexta used a restricted range of alcohols and ketones that differed in chain length (Daly and Smith 2000; Daly et al. 2001a,b). We used these same odorants (1-hexanol, 1-heptanol, 1-octanol, 1-nonanol, 1-decanol, 2-hexanone, 2-heptanone 2-octanone, and 2-decanone), which have also been used in other studies of odor coding (e.g., Smith and Menzel 1989; Stopfer et al. 1997).

Responses of neurons within the AL frequently revealed both mechanosensory and chemosensory components. Multimodal cells have been described previously in M. sexta (Matsumoto and Hildebrand 1981) as well as in other species (e.g., Schmidt and Ache 1996). This necessitated standardization of the mechanosensory aspect of the stimulus-evoked response. To accomplish this, odorant cartridges were attached to the left arm of a T valve. At the bottom of the T was a glass syringe the inside of which had been sleeved with a length of Teflon tubing. This reduced the inside diameter of the syringe to more closely conform to the diameter of the moth antenna. One entire antenna was inserted into this sleeve. Under normal circumstances, clean air, driven by an aquarium pump, flowed through a bypass in a computer-controlled shunt-valve (Lee Valves model LFAA12001) to the right arm of the T valve and then through the Teflon sleeve and over the antenna. When the shunt valve was activated, airflow was diverted from the right arm to the left arm of the T valve, moving air through an odorant cartridge and over the antenna. Expelled odorant was then drawn into an exhaust vent.

Airflow over the antenna was measured at a constant rate of 290 cm/s. This rate was used to minimize lingering odorant around the antenna after each odor pulse. Thus within 500 ms, 145 cm of clean air would have passed over the antenna. Additionally, a 100-ms pulse would have been only 29 cm in length, not including spread from boundary layer effects. Furthermore, given that the length of the antennal flagellum is 2-3 cm, it would take 7-10 ms for the leading edge of the odor stimulus to move from the tip of the flagellum to the base. Although the boundary layer effects within the tubing undoubtedly stretched the odor pulse by an undetermined amount, this approach to odor delivery nevertheless provided highly consistent and accurately timed pulses across all odorants; this is essential for comparison of temporally patterned responses. By providing a continuous, fast moving stream of air and then diverting that air to the odor cartridge, we maintained flow rate and standardized any mechanosensory responses while quickly moving odor into and out of contact with the antennae. This approach also allowed us to standardize the distance from the cartridge to the antenna; this too was essential for maintaining consistent timing of odor delivery between odorants.

A randomized sequence of odorants was established and in turn each odorant cartridge, loaded with 3 µl of neat odorant on filter paper, was connected to the T valve and then pulsed 10 times. Each pulse was 100 ms in duration. For one animal, pulses were separated by a 2-s interpulse interval; all other animals received 5-s interpulse intervals. The longer pulse spacing was used after initial data collection and analysis indicated a pulse-dependent modulation of the odor-evoked responses (see following text). This modulation could have been due to a number of possible effects including decreasing concentration of the odor stimulus due to depletion of the odor cartridge. Pulse-dependent modulation may have also been related to sensory adaptation. Changing to the 5-s spacing reduced but did not completely eliminate this effect. Therefore a "pulse" variable was included in the statistical analyses of responses to account for this modulation (see following text). Once all nine odorants had been sequentially presented, the sequence was repeated to provide a total of 20 pulses for each odorant with 20 min between the first and second block of pulses. This allowed us to assess whether changes from odorant to odorant were attributable to subtle shifting of the preparation.

Recording physiological activity in the AL

The recording and spike sorting methods we used are optimal for dense neuropil such as the insect AL. Recording each spike from four positions greatly enhances the sorting process (Gray et al. 1995; McNaughton et al. 1983; Musial et al. 2002). Sixteen-channel silicon microprobes arrayed in four sets of tetrodes (model 2x2tet; generously supplied by the University of Michigan's Center for Neural Communication Technology) were used to simultaneously monitor the spike activity of 43 isolated units, or clusters, of neural activity within the AL. The four tetrodes were arrayed in a square across two shanks with a center-to-center distance of 150 µm. Probes were inserted directly posterior of the AL under visual control. We selected this position because there is an easily identifiable seam that demarks the posterior end of the AL, allowing relatively reproducible placement. Spikes were sampled across all electrodes within a tetrode if they crossed a minimum voltage threshold on any one of the electrodes (thresholds typically set at 100-150 µV). Data were acquired by a Neuralynx hardware/software system using a Data Translations 3010 digital I/O board for A/D conversion. Threshold neural events were sampled at 32 kHz for 1 ms. Although a 1-ms sampling window does not capture all of the spike waveform, it reduces the number of double spikes captured within a single sampling window (i.e., 2 spikes from different sources).

Off-line spike sorting

Spike sorting into separate unit records was performed off-line in a two-step, partially automated fashion. First, raw tetrode data were processed using the BubbleClust toolbox for Matlab (Neuralynx). BubbleClust uses a kth-nearest neighbor algorithm to assess events across the four recording channels of a tetrode using spike wave energy and the first two extracted principle components of the waveforms. BubbleClust produced a decision tree that segregates events from lowest to highest neighbor density. Each region of neighbor density was then assessed for unit fidelity and stability based on an array of summary statistics including peak plots, autocorrelations, cross correlations, inter-spike interval (ISI) histograms, peak by channel by time plots as well as waveform summary statistics. Clusters of events that appeared to represent valid and stable waveforms of single neuron activity, or well isolated poly-neuron activity, were then exported to the MClust toolbox for Matlab (A. D. Redish) for the second step of processing. In this second step, clusters were manually refined by defining cluster boundaries to exclude any events that appeared to deviate from the average waveform of the cluster.

Once this process was complete, clusters were reevaluated and selected based on the same summary statistics described in the preceding text. Of the 15 recordings made, each from an individual moth, we retained 4 recordings for further analysis. The criterion for selection of recordings was based solely on the quality of the recording made and whether the units were responsive to any of the odorants. If cluster waveforms shifted over time (an indication that the preparation was unstable), the recording was not used for further analyses.

Selected clusters from the four retained recordings were exported as time-stamped unit events. All four retained recordings were analyzed separately. Time-stamped data were then imported into NeuroExplorer (Plexon). As a final selection step power spectral density analysis was used to assess 60-Hz contamination. Any unit with a sharp 60 ± 1-Hz peak, representing more than 1% of the total power, was discarded.

Each retained cluster, or unit, most likely represented either LN or PN activity. Currently, there is no definitive method for discriminating between these two classes of AL neurons. M. sexta has on the order of 1,260 neurons within each AL; these are primarily composed of LNs (ca. 360) and PNs (ca. 900) in <500 µm³ (Homberg et al. 1989). These numbers indicate that there are far more PNs than LNs within the AL. Nevertheless, our recordings represent a combination of local processing and output.

There was also a possibility of missed and contamination spikes as a result of the sorting techniques used (Harris et al. 2000; Henze et al. 2000). Thus "unit" does not perfectly equate to "neuron" although it is presumed that they are highly correlated. Nevertheless, the sorted units in the present study frequently contained information about how the olfactory system responded to an odor. Furthermore, given that "unit" and "neuron" records do not strictly equate, we must also recognize that discussion of above-chance coincident spiking (see following text) must be considered within the context of supporting intracellular and patch electrode results (Lei et al. 2002; Schoppa and Westbrook 2001, 2002).

There was a concern that the units we report as having correlated response patterns to odors might represent either subsamples of a single neuron or recording artifacts. Additional methods were therefore applied during the spike sorting process to identify and exclude these artifacts. For example, one artifact commonly referred to as overshoot occurs when spike waveforms produce an extracellular voltage that crosses the recording threshold twice, once during the initial spike and once during the refractory period thus triggering the recording of a second duplicate event. Overshoot artifacts are common when recording thresholds are set low and waveform sampling time is brief (i.e., 1 ms). However, this artifact has an easily identifiable "clam-shell" waveform. Furthermore, such clusters are always lagged 1 ms behind another larger amplitude waveform; this is evident in crosscorrelograms as a sharp peak at 1 ms.

To assess whether subgroups of coordinated units were actually subsamples of a single unit, waveform profiles across the tetrode sampling channels were again produced and used to reevaluate unit separation. Figure 1 displays a three-dimensional (3D) plot of waveform peaks for four units that were highly correlated in their responses. Also shown are average spike waveforms sampled across all four recording sites of the tetrode. Inspection of the raw clusters indicated that the sources of these four spike records were clearly separated across three recording channels based solely on waveform peak. However, to further test whether these records were from the same neuron, timestamps from all possible pairs of unit records were merged and ISI histograms were assessed for each merged pair. In all cases ISI histograms clearly indicated that when any two of these unit records were merged, the maximum firing rate of the merged record far exceeded 250 Hz. This is the maximum possible spike frequency for a single LN or PN, based on a normal refractory period (T. A. Christensen, personal communication). This provided additional evidence that sorted units were not subsets of a single neural record but rather unique sources of spikes.

View larger version (24K): [in this window] [in a new window]   FIG. 1. Three-dimensional (3D) plot of unit waveform peak by recording channels 1, 3, and 4 (axis scales are post amplified in mV). Each spike event was simultaneously sampled across 4 channels of a tetrode (inset). On the right are the average waveforms of each sorted cluster across all 4 tetrode sampling channels; waveforms are shown with SD bars. Inset: the relative position of each channel of the tetrode. Across all 4 recording channels there is a unique waveform profile for each unit indicating their spatial separation.

Each statistical analysis we used focused on assessing different aspects of the spatiotemporal responses of both individual units and the recorded population. Unless otherwise noted, analyses were performed on 780-ms peristimulus response windows. Data from these response windows were sampled using 10-, 20-, 30-, 40-, 60-, and 80-ms bin widths. Sample window length was based on prior results in zebrafish (Friedrich and Laurent 2001). Re-sampling data, using an array of different bin resolutions, was done to assess the robustness of the subsequent factor and statistical analyses as a function of temporal resolution. As described in Fig. 2, each sample window started when the odor valve opened.

View larger version (35K): [in this window] [in a new window]   FIG. 2. Example of sampling and orientation of data for factor analysis. A: shown here is a raster with 31 units (rows) and the 1st 5 100-ms odor stimulations from a block of 10. For each pulse, 780-ms peristimulus rasters were sampled from the recording. These are highlighted by the open rectangles superimposed on the rasters. Stimulus duration is represented by solid black bars on bottom left of each open rectangle. B: peristimulus rasters (sampled as histograms) are then spliced to form a single dataset of sampled bins. Each completed dataset contained the peri-stimulus samples from all stimulations of all odors which were analyzed using factor analysis. Finally we binned this data using 10-, 20-, 30-, 40-, 60-, and 80-ms bin widths. Subsequent factor analysis was reiterated to assess the effect of bin resolution on factor analysis results.

To test for significant change in unit spike rate in response to odor presentation, peristimulus time histograms averaged across all 20 pulses of each odorant were produced. Significant changes in spike rate in response to each odor were based on the peristimulus cumu-lative histogram crossing the 99% confidence limit. This statistical analysis was performed in NeuroExplorer (www.neuroexplorer.com) based on the method described by Abeles (1982). In cases where a significant change in firing rate was observed, peristimulus rasters were visually inspected to assess the nature of the change in firing rate. Each response was then scored based on three different response types. When the peristimulus spike rate significantly decreased with no visual evidence of rebound bursting, the response was scored as inhibited (–). An increased firing rate, with no evidence of an inhibitory epoch before or after the increase, was scored as excited (+). Finally, when the response of a unit showed consistent patterns of inhibition and excitation it was scored "P."

Analysis of common response patterns across units

Next, as shown in rat OB slice preparations, principle output neurons arborizing a given glomerulus can be coupled (Schoppa and Westbrook 2001, 2002). In M. sexta it has been shown that PNs of a given glomerulus produce synchronized bursts in response to odor (Lei et al. 2002). This leads to the prediction that there should be subgroups of coordinated units within individual recordings. Furthermore, these units should be in close physical proximity. To test this hypothesis, a factor analysis (FA), using the principal components method (SAS Institute 2001), was designed specifically to identify and represent groups of units with correlated patterns of responses across all pulses of all odors. Our use of factor analysis is unique from the approach employed by Friedrich and Laurent (2001) or Stopfer et al. (2003). In these cases, FA was used to correlate odorants (not units) at different periods in time as a function of a virtual population response (i.e., multiple neurons pooled from multiple animals). The general method described here has been used previously to describe the co-activity of populations of extracellularly recorded neurons (Kjaer et al. 1994; Laubach et al. 1999; McClurkin et al. 1991; Nicolelis and Chapin 1994) and is particularly well suited for sensory systems (Chapin and Nicolelis 1999). A separate and complete FA was performed for each of the six bin resolutions. Each retained factor was then statistically analyzed using general linear modeling (GLM).

Figure 2 shows how peristimulus data (of a given bin resolution) were sampled and concatenated into a single data matrix. The sampled data set was then oriented as described by Chapin and Nicolelis (1999). In this case, units (reoriented into columns) were treated as variables. Bins for each successive pulse of each odor were entered as separate rows and treated as a single measure or case. Thus for example, when binned at 30 ms, there were 26 bins for each of 20 pulses. Given nine odorants this produced a vector of 4,680 cases for each unit record; this represents 140.4 s of peristimulus recording time (i.e., all pulses of all odorants).

FA analysis was performed in SAS on the unit correlation matrices to normalize the variance across variables (Chapin and Nicolelis 1999). This approach ensured that the most active units did not bias the factor patterns. Each extracted factor represented groups of units with correlated activity patterns over the entire 140.4 s of peristimulus sample time. The analysis extracted all factors with an eigenvalue value >1. From those factors, we retained only the first five from each recording for detailed analysis. Although we retained the first five extracted factors for detailed analysis, it is typical that only the first two or three factors are retained (Chapin and Nicolelis 1999). We chose to retain more than three factors because in principle the number of meaningful factors, given this analytical approach, should be related to the number of groups of coactive units present in the recording. By extracting and analyzing more than the first three factors we could assess this possibility. Extracted factors were then rotated orthogonally using the Varimax algorithm (Laubach et al. 1999; SAS Institute 2001).

FA produced a factor score for each time bin, which represented the activity of the subpopulation of units that were most highly correlated to the factor. These factor scores are standardized to a z-score distri-bution. Thus time bins with a factor score of 0.0 indicate the average level of activity of the subpopulation of units. Scores >0.0 indicate above average activity, and scores <0.0 indicate below average activity. This does not imply that units are becoming more or less correlated, rather, the subpopulation of units represented by a factor are collectively becoming more or less active. The degree to which the factor represents each individual unit is reflected in the unit-factor correlation coefficient (Pearson's r, standardized using a z-score distribution) (Chapin and Nicolelis 1999; Gorsuch 1983).

The factor scores produced by the FA for each time bin were used in place of individual units to statistically characterize the relationship between the spiking activity of groups of units and features of the stimulus, such as odor identity and stimulus dynamics (i.e., the effect of repeated pulses). This analysis was performed using GLM, which is a multivariate analytical tool that hierarchically partitions variance components for both categorical and continuous variables based on the sequence in which they appear in the model (Cohen and Cohen 1983; SAS 2001). The main effect variables of odorant features, like chain length and functional group, were always partitioned after intrinsic variables (such as time) and variables relating to stimulus dynamics (such as pulse number). This provided the most stringent test of the hypothesis that odorant features accounted for unique variance of a given factor. All two- and three-way interactions were also included in the model.

Analysis of coincident spiking within and between factors

FA was employed to sort units based on coordinated or covarying responses. It is not, however, an appropriate method for discriminating synchronous activity from other possible sources of spike-time covariation (Aertsen et al. 1989; Baker and Gerstein 2001; Brody 1998, 1999). Cross-correlation analysis was also performed on pairs of units both within and between factors to explore whether groups of units identified by the FA contained above-chance synchrony.

Given the evidence in the raw rasters indicating nonstationary responses to odor (e.g., see Fig. 7), we performed crosscorrelational analysis on three distinct subsamples of data: spontaneous unit activity prior to each block of odor pulses, the 780-ms peristimulus data, and spontaneous unit activity after each block of odor pulses. This analysis assessed the presence of a standing level of above-chance coincident spiking between pairs of units. It also provided an indication of whether above-chance coincident spiking persisted during the odor responses. For spontaneous activity, the shuffle-corrected cross-correlogram was calculated based on all possible shuffles and was subtracted from the raw correlogram. The number of shuffles varied between 15 and 20 based on the number of available pre- and poststimulus intervals in each recording. The 99.9% confidence limit was used to indicate above-chance coincident spiking in the pre- and poststimulus intervals; this high significance criterion was selected to produce a low experiment wise error rate. The shuffle correction for the peristimulus data was calculated and subtracted from the raw correlogram for each odorant. Results from all odorants were then averaged. However, because there was also pulse-dependent modulation of the responses (i.e., nonstationary and patterned responses), the correction will be confounded, hence precluding statistical tests of significance (Aertsen et al. 1989; Baker and Gerstein 2001; Brody 1998, 1999). Nevertheless, analysis of the peristimulus crosscorrelograms should provide an indication that any above-chance coincident spiking, present during spontaneous activity, was also present during odor-driven responses. Finally, in all analyses, cross correlations were calculated using 2-ms bins and smoothed with a 6-ms (3 bins) Gaussian filter.

View larger version (51K): [in this window] [in a new window]   FIG. 7. The relationship between odor-dependent responses of all units that correlated to a single factor. Histograms and rasters for 4 units (columns) that correlated to the second factor shown in Fig. 6 (r = 0.20 or greater; see Table 3A). Each unit is designated by the tetrode it was recorded from (T) and cluster number (C). Their overall correlation coefficient (r) with the factor is listed to the right of this designation. Histograms were again normalized to a uniform probability distribution with a y-axis range of 0-1. Odorants, listed on the left, are designated by functional group (A = alcohol, K = ketone) and the number of carbon units on the chain. The superimposed line on each histogram displays the mean factor score by odor and bin (30-ms bin width). The same line is superimposed on all histograms of a row. The mean factor score does not extend to the 200 ms prior to odor stimulation as this time was not included in the analysis. This prestimulus data were included here to display spontaneous activity. , stimulus duration. The relationship between the average unit response (histogram) and the average factor score by bin is quite high even in cases where the magnitude of the response is low. This relationship is quantified as a Pearson's r inset in the upper right of each raster (). Note that these odor-specific correlations are much higher than the overall unit-factor correlation (listed at the top of each column) because the overall unit-factor correlation does not average out within odor response variability. Furthermore, each factor pattern closely matches the each histogram even when there is a pulse-related modulation or a nonstationary response (e.g., ). This occurs because each pulse was treated individually in the factor analysis and, more importantly, because all units shown here are modulating in precisely the same manner in spite of differences in spike rate.

In the final analysis, all factors identified by the GLM to contain significant odor-dependent responses (P < 0.001) were then treated as orthogonal response dimensions in Euclidean space. For each odor, an n-dimensional response trajectory was calculated where n was equal to the number of factors with significant odor-dependent effects. The number of factors with significant odor-dependent effects ranged between three and five per recording. Factors without significant odor-dependent responses were excluded from this analysis.

For each successive bin, the mean Euclidean distance of the factor-represented population response for each pair-wise comparison of odors was calculated. Calculations were performed on individual recordings, and the subsequent results were then averaged across all recordings (see RESULTS). Factors were orthogonally rotated; hence the Pythagorean theorem was used for distance calculations. Furthermore, because factor variance is standardized to a z-score distribution, all dimensions are in the same SD metric. The distance measure we report is thus expressed in SD units. This analysis allowed us to quantify when in time and by how much the population response trajectory to each odorant differed from the response trajectories of all other odorants.

Finally, a systematic change in Euclidian distance between trajectories, as a function of differences in chain length between comparison odors, will produce a gradient with a slope that can be expressed in degrees. Here slope angle was assessed by first calculating the linear regression slope for each time bin. This was done by plotting the distance score of each chain length comparison against the corresponding difference in carbon units for each comparison (ranging from 1 to 4; see Fig. 9C). A linear regression produced a beta weight for each bin, which was standardized to a range +1 to –1. To express results as an angle in degrees (i.e., +90 to –90°), each value was multiplied by 90. Results were then smoothed across bins using a three-bin moving average.

View larger version (49K): [in this window] [in a new window]   FIG. 9. Euclidean distance between population responses trajectories as a function of odorant features. Distance was calculated in n dimensional factor response space. Only factors with significant odor-dependent effects were included (see Table 3). A: distance by functional group differences. Analysis based on a 4-dimensional factor space. These 4 factors represented a total of 17 units (where unit-factor correlations were 0.20). Results are based on a single animal (see Table 3A). x axis is peristimulus time and the y axis represents distance in SD units. The gray bar indicates stimulus duration. AvA, the distance by time among all comparisons of alcohols with alcohols; KvK, the distance by time among all comparisons of ketones with ketones; KvA, the distance by time among all comparisons of ketones with alcohols. The maximum distance was achieved in the KvA group and occurred between 150 and 300 ms. B: distance in 4-dimensional factor response space between alcohols as a function of increasing differences in carbon chain length between comparison odors (factors again representing the same animal and 17 units in A). When the comparison odors differ by 4 carbon units the response distance was maximized, followed by comparisons of alcohols differing by 3 carbon units then 2 then 1. Furthermore, there was a time course whereby distance was maximized then reduced. C: summary of all 4 recordings based on an 18-dimensional factor response space (representing 72 total units where unit-factor correlations were 0.20). The mean population distance as a function of difference in the number of carbon units. Results were averaged across the 780-ms response time and broken down by alcohols and ketones. As the difference in odors increased, the distance in the population response trajectories also increased to produce a "discrimination slope" (overall slope angle = 16 and 14° for alcohols and ketones respectively; results based on a linear function of carbon chain length). D: measure of the discrimination slope angle as a function of time. Results based on all 4 recordings, which represented a total of 72 units (18 factors). x axis represents response time (s); the y axis represents the angle of the discrimination slope in degrees. Calculations were based on a moving average of 3 bins (90 ms). At 120-ms post stimulus onset, as the excitatory phase of the response develops across the population (see Figs. 2 and 3), the steepness of the slope angle rapidly increases for both alcohols and ketones maximizing at 240 ms at an angle of 17° for alcohols and 270 ms at an angle of 17° for ketones.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

Results were based on a total of 133 sorted units recorded from four moths. Alignment of the peristimulus spike trains revealed that individual units responded with excitatory, inhibitory or temporally patterned responses that in many cases depended on the specific test odorant (Fig. 3). The initiation of a unit's response to an odorant was a visually identifiable event that occurred consistently from pulse to pulse. The latency between odor onset and this initial phase of the response appeared to be unit specific and observable as early as 60-90 ms after odor onset (Fig. 3).

View larger version (60K): [in this window] [in a new window]   FIG. 3. Peristimulus rasters (980 ms) and histograms for 8 different but simultaneously recorded units (listed in columns) in response to 20 pulses of 9 different odorants (rows). Each odorant was presented in 2 blocks of 10 pulses. Each pulse was 100 ms in duration. Pulses were separated by 2 s (5 s in other recordings). Each block of 10 pulses was separated by 20 min. Rasters and histograms thus contain the 20 combined stimulus presentations of a single odorant. Two units are shown from each of the 4 recording tetrodes. Each unit was designated with a unit number (C) and tetrode number (T). Each 30-ms bin of the histogram represents the sum of all spikes across all 20 odor pulses. Histograms were normalized to a uniform probability distribution to produce a standardized y-axis scale that ranged from 0 to 1. The open boxes demark the pulse width of the odor stimuli relative to the activity of the neurons. Prestimulus time (200 ms) was included in the figure to show spontaneous activity levels; this prestimulus time was not included in the analyses.

For some units, responses to specific odorants were modulated over repeated pulses. In these cases, visual inspection of peristimulus rasters indicated that response patterns within the first and second blocks of pulses shifted systematically toward longer periods of silence and fewer spikes (Fig. 3); this is consistent with previous findings (Stopfer and Laurent 1999). This pulse-to-pulse modulation typically recovered between the first and second blocks of pulses. Pulse-dependent modulation and between block recovery effects are evident in Fig. 3, unit T2C10 in response to a number of odorants (see also Fig. 6).

View larger version (56K): [in this window] [in a new window]   FIG. 6. The relationship between factors and unit activity. A: list of units that had strong correlations for each of the 1st 5 extracted factors. Units are designated by the tetrode they were recorded from (T) and a unit number (C). To the right of each unit designation is that unit's standardized correlation to the factor. 22 of the 31 units in this recording correlated to the first 5 factors (r 0.20). Only 2 units correlated (r 0.20) to >1 factor (). B: peristimulus histograms and rasters from the unit (in columns) that best correlated to each factor (these are designated with * in A). Each histogram and raster in a column represents the compiled responses to 1 of 9 odorants. Odorants, listed on the left, are designated by functional group (A = alcohol, K = ketone) and the number of carbon units on the chain. Histograms were normalized to a uniform probability distribution with a y-axis range of 0-1. The superimposed line on each histogram represents the mean factor score at each 30-ms time bin for each odorant. Each odorant (row) elicited an average response pattern from each factor (columns) that was in many cases unique from all other factor responses to the given odor. Collectively, 1 row reflects the total simultaneous responses to a single odorant from all units that the 5 factors represented (in this case, 22/31 sorted units). Thus each row represents a 5-dimensional response trajectory, or signature, to each respective odorant based on a 22-unit population. C: the results of the post hoc statistical analysis of the 2-way interactions of chain length-by-time bin (CL) and functional group-by-time bin (FG; see also Table 3A). Results are displayed in the series of . Each row of represents a single bin and is aligned in time with the bins in the preceding histograms. , time bins contained statistically significant odor-dependent differences in factor score (P < 0.001 to correct for experiment-wise error rate). Analysis was based on a GLM dummy coding scheme.

View larger version (101K): [in this window] [in a new window]   FIG. 4. Results of statistical analysis of change in firing rate in response to each odorant for all units from 2 of the 4 recordings. Each box represents tests for change in firing rate for a single unit in response to 20 pulses of a single odor. , a statistically significant change in average firing rate (P < 0.01; see Abeles 2003; NeuroExplorer 2003). Most units responded with a significant change in firing rate to most odorants. Within each is an indication of the nature of the response: +, excitatory response; –, inhibition of activity; and P, the response appeared temporally patterned.

FA effectively sorted units that exhibited common responses properties across all responses to all odors. The analysis produced a matrix of correlations between each unit and each factor. Units most strongly correlated to a given factor remained strongly correlated to the same factor regardless of the temporal resolution (sic. bin widths). However, unit-factor correlations that were not the largest for a given unit tended to increase considerably with increasing bin width. For example, Fig. 5 A summarizes the mean largest, second largest, and third largest correlations to the factors. The first and second largest unit-factor correlations increased negligibly as a function of increasing bin width. The third strongest unit-factor correlations increased slightly as bin width increased, although this effect was still small. The highest mean unit-factor correlation for each unit shows much the same result (Fig. 5B), suggesting that increasing bin width had little effect on the primary unit-factor correlations. Finally, as bin width increased, the number of modest (0.10 < r < 0.20) and moderate (r 0.20) unit-factor correlations increased substantially (Fig. 5C).

View larger version (18K): [in this window] [in a new window]   FIG. 5. The relationship between sampling bin width and unit-factor correlations. Results are averaged across all of the 1st 5 extracted factors and across all 4 moths. A: the 3 largest unit-factor correlations averaged across factors as a function of sampling bin width. B: the average largest unit-factor correlation averaged across units for all recordings as a function of bin width; these include units that did not correlate well to any of the 1st 5 extracted factors. C: the mean number of unit-factor correlations per factor ranging between 0.10 and 0.20 () and equal to or >0.20 () as a function of increasing bin width. Data based on the 1st 5 extracted factors.

All subsequently reported results were based on a 30-ms bin resolution. Table 2 summarizes the corresponding FA for each recording. This table shows the number of sorted units per recording and the number of factors with an eigenvalue >1. By grouping together units with correlated response patterns, FA explained between 47-52% of the total variance. Residual unexplained variance occurred for a number of reasons, including: error biasing effects that occur with small bin sizes (see Chapin and Nicolelis 1999); units with low spike rates and probabilistic responses (i.e., units that occasionally responded with a small yet appropriately timed response); random variation occurring from trial-to-trial; and units that did not correlate to factors. Indeed, 21% of the recorded units failed to correlate to a factor (Table 2). The lack of correlation occurred in most cases because these units did not respond to most (if any) of the odorants (see Table 1A). Units that did not respond to any odorants were observed in three of four recordings and accounted for 8% of sorted units. Additionally, only 14% responded to three or fewer odorants (Table 1A). As described in the preceding text there were also units, which, from pulse-to-pulse, may have responded at the appropriate times (relative to the other units correlated to a given factor), but they tended to do so in a variable manner and with low firing rates. If a unit responded in some of the correct bins within a pulse, or on some but not all pulses, the result was a correlation between the unit and the factor that was lower, relative to more active and consistent units (see Fig. 3, T1C1 and T4C2). In all of these cases, the net effect was a decrease in variance explained by the FA.

Relationship between factor scores and unit activity

FA accurately identified groups of units that consistently responded in the same manner. Figure 6A lists all units, sorted by recording tetrode, with a standardized correlation of 0.20 to the corresponding factor. These correlations represent the overall relationship between each unit and the associated factor across all odors. Highlighted in each list are the two cases where units correlated to more than one factor. Figure 6B shows the relationship between the first five extracted factors from one moth (see Table 1A and Table 2, moth A) and the unit with the highest correlation to each factor. Each peristimulus raster and histogram represents all 20 responses to a single odorant. The line superimposed on each histogram represents the average factor score across the 20 pulses of each odorant at each time bin. The degree of correlation between unit and factor can be visualized by the degree of fit between the histogram and the superimposed line representing the mean factor scores.

The factors represented periods of silence and bursting that were observed in the responses of correlated units. The precise combination of bursting and silence again depended on the test odorant. In Fig. 6B, the initial response to odor for factors 1-4, and their corresponding units, was frequently a cessation of background spiking by the second or third bin. This period of silence ranged from relatively brief (e.g., 30-60 ms) to prolonged (e.g., 700ms). The pattern of bursting activity was equally diverse. The response to 2-octanone by units correlating to factor 1, for example, consisted of between one and three spikes within a single 30-ms bin at 120 ms after stimulus onset (Fig. 6B). The response of the same units to 2-hexanone, however, followed a pattern of inhibition, then a brief excitatory burst, followed by inhibition, and finally a strong and prolonged excitatory response that gradually decayed back to spontaneous spike rates. In both cases, the factor accurately reflected the responses of the displayed unit. In summary, despite the broad array of odor-driven responses, shown in Fig. 6, the retained factors nevertheless accurately reflected the activity of the most correlated unit.

Figure 7 shows that this unit-factor relationship can be consistent for all units that were correlated to a factor even when those correlations are moderate. This figure displays the mean factor score by odor and bin for factor 2 from Fig. 6B (see also Tables 1A and 2A). As before, the mean factor score by bin was superimposed as a line on each corresponding histogram. Unit T2C9 had the highest correlation (0.83) to the factor, and therefore the peristimulus time histograms show a very strong correspondence to the factor. In spite of the relatively lower firing rates and relatively moderate correlations of units T2C4 and T2C1, these units nevertheless produced histograms exhibiting a close relationship the mean factor scores. The overall correlation between each odor-specific histogram and the mean factor score by bin can be expressed as a correlation. These correlations are inset in the top right corner of each raster in Fig. 7. These results indicate that the average correlation between mean unit activity and the corresponding mean factor scores are typically high.

Thus units that correlated to a given factor showed a strong tendency to respond in a correlated manner. Even when responses pattern were complex, there was often a strong correlation between correlated units and the underlying factor. For example, Fig. 7 shows that the response of all four units to 2-octanone consisted of an initial epoch of silence followed by two distinct and brief bursts of activity at 180 and 300 ms after stimulus onset (though this shifts systematically by pulse). The lowest factor-histogram correlation for this odorant was 0.85 (unit T2C4). This tight relationship between factors and units was evident across all recordings and indicates that factors accurately represent the activity of their constituent units.

Statistical variation in factor patterns

We found that the temporal response patterns of factors were statistically dependent on odorant chain length and functional group. Separate GLM analyses performed for each factor included all two- and three-way interactions. The GLM results are listed in Table 3 by moth. All variables are listed in the order in which they were entered into the hierarchical GLM. Main effects of time (bin), pulse, chain length (CL), and functional group (FG; alcohol vs. ketone) were significant for most factors. Furthermore, two- and three-way interactions of time with chain length and functional group indicated that temporal response characteristics changed significantly as the molecular features of the stimulus changed. Thus the precise shape of the temporal response (lines compared down columns in Figs. 6B and 7) for a given factor with these interactions depended on the odorant. For the factor displayed in Fig. 7, these interactions were manifested as epochs of silence and bursting phases that were either present or absent, and furthermore, could shift in both latency and duration. Post hoc GLM analysis of significant time by chain length and time by functional group interactions (see Fig. 6C; P < 0.001), indicated which time bins contained odor-dependent effects.

Evidence of spike timing synchrony

Shuffle-corrected cross-correlograms were calculated between all pairs of units within a factor and between units correlated to different factors. Results of this analysis indicated that units within a factor tended to exhibit above-chance coincident spiking. This was true for spike train samples taken from recording periods prior to odor stimulation (Fig. 8 A), during odor responses (B), and after odor stimulation (C). Units recorded from different tetrodes but correlated to the same factor (e.g., factors 4 and 5 in Fig. 6), rarely showed evidence suggesting above-chance synchrony. Indeed, across all 4 recordings, only 1 case of above-chance synchrony occurred between pairs of units recorded from different tetrodes.

View larger version (26K): [in this window] [in a new window]   FIG. 8. All possible pair-wise crosscorrelograms and 99.99% confidence limits for the 4 units that correlated factor 2 (see Figs. 2, 6, factor 2, and 7). All correlograms are shuffle corrected across all possible shuffles. Data were binned at 2 ms for this analysis, and correlograms are smoothed using a 3-bin wide Gaussian filter. The x axes are in time (s) and the y axes are coincident probability. A: results from spike train records prior to series of odor pulses. Analysis was based on a total of available 18 prestimulation intervals, totaling 167 s of data. B: results from all 780-ms peristimulus spike train records. Analysis was based on a total of 180 peristimulation intervals totaling 140.4 s of data. For peristimulus data, shuffle corrections were first calculated for each odorant (20 shuffles) then results were averaged across all 9 odorants. In this case, shuffle correction will not be accurate because of pulse-dependent and nonstationary responses. Thus confidence limits were not included. C: results from spike train data that was recorded 2 s after the last stimulus of each train. There were a total of 15 intervals available totaling 80 s for this analysis.

To characterize the distinctiveness of the population response to odorant as a function of response time, a response trajectory was calculated using factors as separate dimensions. The Euclidean distance between response trajectories was a function of differences in both molecular features of odorants and response time (Fig. 9). For example, the average distance between ketones and alcohols (i.e., functional group differences) was greater than the average distance of comparisons within ketones or alcohols (i.e., carbon chain length differences; see Fig. 9A). The distance between ketones and alcohols maximized between 150 and 240 ms (between the 5th and 8th time bin). Beyond 240 ms the distance between alcohols and ketones decreased.

Distance between trajectories also increased as a function of differences in carbon chain length between comparison odorants within functional groups. This chain length effect was maximized between 180 and 240 ms as shown in Fig. 9B. This figure also shows that by 330 ms, the distance for all comparisons drops back to 1 SD. This effect is averaged across time for all four recordings, for both alcohols and ketones, in Fig. 9C. Indeed, Fig. 9C shows that as the difference in carbon chain length increased so too did the mean distance between trajectories. This systematic increase in distance as a function of difference in chain length can be characterized as a slope or gradient. The angles of the slopes in Fig. 9C were 16 and 14° for alcohols and ketones, respectively. In Fig. 9D, this slope angle has been calculated for each time bin and smoothed using a three-bin sliding average. Here, the systematic chain length effect creates a gradient that follows a time course. In this case, the angle is maximized at 17° for both ketones and alcohols by 240 and 270 ms, respectively. After this maximization, in both alcohols and ketones, there was a subsequent decrease in slope angle.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
The central objective of this study was to correlate spatiotemporal population responses in the AL to molecular features of odorants that have been behaviorally characterized in studies of odor generalization and discrimination (Daly and Smith 2000; Daly et al. 2001a,b). Furthermore, in characterizing population responses, we wanted to employ an analytical approach that would be sensitive to both the spatial and temporal structure of the recorded population. Using a series of analyses, we make the following set of observations. First, the excitatory aspect of the population response is odor dependent, temporally patterned, and distributed across many units. Second, there are subgroups of recorded units that act in a coordinated manner. In many cases, these coordinated subgroups tend to spike synchronously, even during spontaneous activity. Third, when the responses of all subgroups are considered collectively in a population analysis, they produce a sequence of bursting and silence that systematically becomes more unique as a function of odorant features; this was largely not the case for individual units or factors. Finally, the time course needed for population responses to become maximally unique was relatively brief.

Specifically, our results imply that the spatial pattern of a response can have potentially important temporal components, particularly during the bursting phase of an odor-driven response. If our rate-based measures of response trajectory distance are related to how the AL represents an odorant, then subtle changes in the sequence of bursting across the population may enhance discrimination. Sequenced bursting was brief, typically lasting 240 ms (e.g., see Fig. 6) even though odor-dependent responses were considerably longer. Therefore the neural patterns that represent odors quickly become distinct. As bursting ends, the distinctiveness of population responses begins to decline as indicated by a decrease in both trajectory distance and slope angle. It remains unclear whether the residual distinctiveness of odor-dependent responses that persist after stimulus offset and the completion of bursting is due to the intrinsic properties of the AL or to lingering odor stimulus in the boundary layer around the antenna.

Factors identify units with common responses and coincident spiking

We were able show that there were subgroups of units that acted as functional subassemblies using factor analysis. This analysis extracted from the raw spike-train data, subpopulations of units with similar temporal response patterns. Factors correlated well with (and hence were used to represent) these subpopulations of units (Figs. 6B and 7). Shuffle corrected cross-correlation analyses indicated that some factors identified groups of units that fire coincident spikes. Units demonstrating synchrony, defined in this manner, did so during spontaneous activity, both prior to and after odor presentation, and we provide evidence that they continue to do so during odor responses as well. However, nonstationary responses preclude our ability to accurately characterize synchrony during responses (Aertsen et al. 1989). In spite of this limitation, coincident spiking was still evident in the same pattern across units during odor-driven activity (Fig. 8). We are therefore confident that synchrony was still present during odor responses. Thus these results indicate that factor analysis can be used to prescreen for groups of units that appear to be effectively connected.

Spatial distribution of synchronous and nonsynchronous factors

When factors described units that were statistically synchronous (e.g., factors 1-3 in Fig. 6), those units were, with a single exception, recorded on the same tetrode. It is possible that tetrodes were positioned within and hence recorded multiple units from a single glomerulus or possibly from their output tracts. An ordinary glomerulus in M. sexta is 65-75 µm in diameter (L. P. Tolbert, personal communication), which is slightly larger than a tetrode. When factors described units that had the same response patterns to odor stimulation but showed no evidence of synchronous spiking (spontaneous or otherwise), those units were commonly recorded from different tetrodes.

These findings are generally consistent with studies that employed dual intracellular or patch recordings in the AL and OB (Lei et al. 2002; Schoppa and Westbrook 2001, 2002). In M. sexta, paired recordings of PNs from the same glomerulus can produce nearly indistinguishable, synchronized bursting responses (Lei et al. 2002). Previous studies in M. sexta, however, have not shown coincident spiking during spontaneous activity. This is probably because the relatively brief duration of dual impalement experiments does not provide an adequate volume of data (H. Lei, personal communication).

In either case, it has been demonstrated in M. sexta that GABAergic LN inhibition mediates a fast inward Cl^– conductance, which briefly hyperpolarizes PN cell membranes and is followed by membrane depolarization and bursting. This mechanism is thought to synchronize PN bursting responses within individual glomeruli (Christensen et al. 1993, 1998). LNs are furthermore spontaneously active (Christensen et al. 1993) and thus may impose a constant synchronizing force on membrane potentials among numerous PNs within a glomerulus. This would provide a mechanism by which coincident spiking might always be both evident and localized. Pairs of mitral cells that arborize the same glomerulus in the rat OB also produce synchronous spike trains due to a combination of dendro-dendritic gap junctions as well as reciprocal synaptic interactions (Schoppa and Westbrook 2001, 2002). Although these additional mechanisms have not been identified in M. sexta, they suggest additional means by which synchronous activity could be maintained within an olfactory system.

Factors produce odor-dependent response dimensions

Within most factors, the temporal aspect of a response varied in an odor-dependent manner. This was revealed in the numerous time-by-odor interactions (Table 3). Changes in response to different odors were in many cases categorical. That is, as carbon chain length increased, the temporal pattern of a factor might change abruptly. These types of categorical shifts were common throughout the data. Thus we observed little evidence, in single units or in single factors, consistent evidence of systematic and continuous changes that correlated to our previous behavioral results (Daly et al. 2001a). However, when all factors were considered simultaneously, they collectively produce a multidimensional "signature" or trajectory that did change in a systematic manner as a function of carbon chain length. As the difference in molecular features of odorants increased, the corresponding Euclidian distance between their trajectories also increased (Fig. 9). Thus several factors taken together were collectively better at resolving the structural relationship between odors than any single factor (or any single unit) was capable of resolving in isolation. This indicates that the spatiotemporal signatures we have described are consistent with, and have the capacity to represent, the odor dimensions we have demonstrated to be behaviorally relevant to M. sexta (Daly et al. 2001a).

There has not yet been a mechanism described in M. sexta that can produce these temporally patterned sequences of bursts across multiple subassemblies as we demonstrate here. Studies of honey bee olfactory coding, however, indicate that there are two distinct output pathways each with unique temporal response properties (Müller et al. 2002). These researchers have shown that PNs of the lateral antennocerebral tract respond to different odors with different spiking rates and may provide the initial information about odor identity at the level of the mushroom bodies. Projection neurons of the median antennocerebral tract respond to different odors with bursts of activity of different latencies. The information conveyed in these PNs lagged behind PNs of the lateral antennocerebral tract. These results are consistent with ours in two important ways; they support the hypotheses that there can be a sequence of outputs from the AL and that the latency of these outputs can change as a function of odor identity.

How much time is necessary for odor discrimination?

We also found that the amount of time needed for population response trajectories to maximally diverge was relatively brief (Fig. 9D). The distance between response trajectories, relative to functional group and chain length, was maximized by 240 ms or 120 ms after the initiation of the bursting phase of the response. Distance was maintained until 270-310 ms after stimulus onset. This time course also appeared to be independent of odor similarity. This period of optimal divergence was consistent with the bursting phase of the response. This suggests that even the longer and more complex bursting response patterns we describe may add to the power of the population to statistically discriminate these odorants. After the bursting phase of the responses, however, statistical power to discriminate odorants began to drop as indicated by both decreasing trajectory distance measures as well as decreasing measures of slope angle.

Friedrich and Laurent (2001) have reported comparable results in the zebrafish. They show, in a series of 200-ms sequential steps in response time, that there is a net decorrelation of odor representations in a multidimensional space. In fact, throughout our data there remains consistent odor-driven activity, even beyond the sampling 780-ms window we used here (e.g., Fig. 1). One issue that remains unresolved is whether and how these prolonged odor-dependent responses (i.e., beyond 370 ms) contribute information about odor identity in the behaving animal. Using 2.4-s odor presentations, Friedrich and Laurent (2001) show for the zebrafish that this decorrelation takes on the order of 800 ms. Our data, which assess statistical discrimination at a finer temporal resolution and with a briefer odor pulse length, indicate that decorrelation, as evidenced by divergence of the population response trajectories, is optimized faster. Our previous behavioral results were based on training and testing stimulations that were 4 s in duration. Whether M. sexta can demonstrate odor discrimination using these brief stimulus durations is currently unknown.

Although this difference between studies may be species specific, we cannot rule out the possibility that the amount of time required to maximize the trajectories may be related to differences either in stimulus duration or in the analytical methods employed in these two studies. For example, with a longer-duration odor pulse, our response trajectory measures could continue to diverge. However, recent results by Stopfer et al. (2003), using related analytical techniques and 1-s odor pulses, show that odor discrimination in the locust may occur within a time frame consistent with ours. These results suggest that longer stimulation would not enhance our discrimination measures. Furthermore, behavioral data suggest that discrimination in the rat can occur in as little as 200 ms (Goldberg and Moulton 1987; Uchida and Mainen 2003), although it may take much longer in humans (Wise and Cain 2000).

Temporal patterning and stimulus dynamics

Our results also support the long held hypothesis that odor identification, and hence discrimination, while not instantaneous, should occur on a relatively rapid time scale. The dynamics of natural stimuli may dictate that discrimination occurs on such a rapid time scale. Behavioral evidence in moths indicates that both brief and intermittent odor stimulation are necessary to achieve successful pheromone-tracking behaviors (e.g., Willis and Baker 1984), although this has not yet been tested for nonpheromonal odorants. These behavioral results are consistent with matched physiological results indicating that PNs track brief odor pulses (Christensen et al. 1998, 2003; Lei et al. 2002; Vickers et al. 2001). This relatively fast time scale therefore allows olfactory systems to respond to the dynamics of natural stimuli (Koehl et al. 2001; Vickers et al. 2001).

What remains to be determined is the extent to which the types of temporal patterning we describe here are further dependent on, or interact with, stimulus dynamics (Christensen and Hildebrand 1997; Christensen et al. 1998, 2000; Vickers et al. 2001). Odors are carried in a turbulent medium that produces a plume structure that is complex (Koehl et al. 2001; Vickers et al. 2001). Our results are consistent with prior conclusions that the AL simultaneously processes information about odor identity as well as the context in which odor is experienced (Christensen et al. 1998). For example, a series of repeated pulses modulated responses to odor. The onset of an excitatory response was delayed in later pulses, perhaps due to sensory adaptation, decreasing odor concentration from the odor cartridge or other effects intrinsic to the AL circuitry. The complexity of odor plumes therefore makes it difficult to pick a single stimulus presentation method and expect that it would adequately represent the natural situation. Thus a more in-depth exploration of the effect of different variables related to stimulus dynamics is needed.

Odor-independent responses

All of the factors we described produced temporally patterned responses but not all factors produced significant odor-dependent responses (see Table 3). For example, we found factors that became silent and did not recover for the duration of the peristimulus time sample. We also found factors that were initially excited with little or no evidence of a prior decrease in firing rate. In both cases, the units represented by these factors tended to respond in the same manner to all odorants. Units correlated to these factors were typically recorded from two or more tetrodes spanning 200 µm² and showed little evidence of synchrony. Coordinated early responses that occurred in a non odor-dependent and nonsynchronous manner lead us to speculate that these could be subpopulations of LNs, which received independent sensory input from different glomeruli. Sensory input into glomeruli is likely to be roughly simultaneous, thus distributed LN activity should follow and thereby create nonsynchronous coactivation.

Conclusions

There is still debate as to the relative importance of both spatial and temporal information in olfactory coding (e.g., Bazhenov et al. 2001; Christensen and White 2000; Christensen et al. 2000; Hildebrand 1995; Hildebrand and Shepherd 1997; Kashiwadani et al. 1999; Laurent 1996; Laurent and Davidowitz 1994; Laurent et al. 2001; Lei et al. 2002; Malnic et al. 1999; Mori 1995; Slotnick and Bodyak 2002; Slotnick et al. 1997; Uchida et al. 2000), although it is clear that both may be important. Here we show that the slow temporal sequence of bursts among subassemblies of coordinated units is correlated to odorant molecular features in a manner consistent with behavioral data (Clelend et al. 2002; Daly et al. 2001a; Uchida and Mainen 2003). Future work must move beyond the current correlational analysis, however, by applying ensemble recording and analysis techniques, such as those described here, in conjunction with pharmacology. This will allow us to assess the relative contribution of modulatory and inhibitory pathways on our measures of ensemble discrimination. Furthermore, such studies must be matched with behavior-pharmacological studies of olfactory learning and odor discrimination to assess changes in brain function in conjunction with changes in the abilities of these animals.

	GRANTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
This work was supported by an award from National Institutes of Health Grants 9 R01 RR-14166 to B. H. Smith and 1 R03 DC-05535 to K. C. Daly. Initial support was provided by a contract from Defense Advanced Research Projects Agency/Controlled Biological Systems to J. G. Hildebrand.

	ACKNOWLEDGMENTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
The authors thank the University of Michigan Center for Neural Communication Technology (Grant P41 RR-09754) for electrode supply. We also thank T. A. Christensen, H. Lei, L. P. Tolbert, A., Dacks, and G. L. Gerstein for comments on this work.

	FOOTNOTES

 
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

¹ The Supplementary Material for this article (a figure and a table) is available online at http://jn.physiology.org/cgi/content/full/01132.2003/DC1.

Address for reprint requests and other correspondence: K. C. Daly, Dept of Entomology, The Ohio State University, 400 Aronoff Laboratory, 318 W. 12th Ave., Columbus, OH 43210 (E-mail: daly.40{at}osu.edu).

	REFERENCES

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
Abeles M. Quantification, smoothing, and confidence limits for single-units' histograms. J Neurosci Methods 5: 317–325, 1982.[CrossRef][Web of Science][Medline]

Aertsen AM, Gerstein GL, Habib MK, and Palm G. Dynamics of neuronal firing correlation: modulation of "effective connectivity." J Neurophysiol 61: 900–917, 1989.[Abstract/Free Full Text]

Baker SN and Gerstein GL. Determination of response latency and its application to normalization of cross-correlation measures. Neural Comput 13: 1351–1377, 2001.[CrossRef][Web of Science][Medline]

Bazhenov M, Stopfer M, Rabinovich M, Huerta R, Abarbanel HD, Sejnowski TJ, and Laurent G. Model of transient oscillatory synchronization in the locust antennal lobe. Neuron 30: 553–567, 2001.[CrossRef][Web of Science][Medline]

Bell RA and Joachim EG. Techniques for rearing laboratory colonies of tobacco hornworms and pink bollworms. Ann Ent Soc Am 69: 365–372, 1976.

Bhalerao S, Sen A, Stocker R, and Rodrigues V. Olfactory neurons expressing identified receptor genes project to subsets of glomeruli within the antennal lobe of Drosophila melanogaster. J Neurobiol 54: 577–592, 2003.[CrossRef][Web of Science][Medline]

Brody CD. Slow covariations in neuronal resting potentials can lead to artefactually fast cross-correlations in their spike trains. J Neurophysiol 80: 3345–3351, 1998.[Abstract/Free Full Text]

Brody CD. Correlations without synchrony Neural Comput 11: 1537–1551, 1999.[CrossRef][Web of Science][Medline]

Buck L and Axel R. A novel multigene family may encode odorant receptors: a molecular basis for odor recognition. Cell 65: 175–187, 1991.[CrossRef][Web of Science][Medline]

Chapin JK and Nicolelis MA. Principal component analysis of neuronal ensemble activity reveals multidimensional somatosensory representations. J Neurosci Methods 94: 121–140, 1999.[CrossRef][Web of Science][Medline]

Christensen TA and Hildebrand JG. Frequency coding by central olfactory neurons in the sphinx moth Manduca sexta. Chem Senses 13: 123–130, 1988.[Abstract/Free Full Text]

Christensen TA and Hildebrand JG. Coincident stimulation with pheromone components improves temporal pattern resolution in central olfactory neurons. J Neurophysiol 77: 775–781, 1997.[Abstract/Free Full Text]

Christensen TA, Lei H, and Hildebrand JG. Coordination of central odor representations through transient, non-oscillatory synchronization of glomerular output neurons. Proc Natl Acad Sci USA. 100: 11076–11081, 2003.[Abstract/Free Full Text]

Christensen TA, Pawowski VM, Lei H, and Hildebrand JG. Multi-unit recordings reveal context-dependent modulation of synchrony in odor-dependent neural ensembles. Nat Neurosci 3: 927–931, 2000.[CrossRef][Web of Science][Medline]

Christensen TA, Waldrop BR, Harrow ID, and Hildebrand JG. Local interneurons and information processing in the olfactory glomeruli of the moth Manduca sexta. J Comp Physiol [A] 173: 385–399, 1993.

Christensen TA, Waldrop BR, and Hildebrand JG. Multitasking on the olfactory system: context-dependent responses to odours reveal dual GABA-regulated coding mechanisms in single olfactory projection neurons. J Neurosci 18: 5999–6008, 1998.[Abstract/Free Full Text]

Christensen TA and White J. Representation of olfactory information in the brain. In: The Neurobiology of Taste and Smell; Second Edition, edited by Finger T E, Silver WL, and Restrepo D. NY: Wiley-Liss, 2000, p. 201-232.

Cleland TA, Morse A, Yue EL, and Linster C. Behavioral models of odor similarity. Behav Neurosci 116: 222–231, 2002.[CrossRef][Web of Science][Medline]

Cohen J and Cohen P. Applied Regression/Correlation Analysis for the Behavioral Sciences (2nd ed.). Hillsdale, NJ: Erlbaum, 1983.

Daly KC, Chandra S, Durtschi ML, and Smith BH. Generalization of olfactory-based conditioned response reveals unique but overlapping odor representations in the moth, Manduca sexta. J Exp Biol 204: 3085–3095, 2001a.[Abstract/Free Full Text]

Daly KC, Durtschi ML, and Smith BH. Olfactory based discrimination learning in Manduca sexta. J Insect Physiol 47: 375–384, 2001b.[CrossRef][Web of Science][Medline]

Daly KC and Smith BH. Associative olfactory learning in the moth Manduca sexta. J Exp Biol 203: 2025–2038, 2000.[Abstract]

Eaton JL. Morphology of the head and thorax of the adult tobacco hornworm Manduca sexta (Lepidoptera: Sphingidae). I. Skeleton muscles. Ann Ent Soc Am 64: 437–445, 1971.

Friedrich RW and Laurent G. Dynamic optimization of odor representations by slow temporal patterning of mitral cell activity. Science 291: 889–94, 2001.[Abstract/Free Full Text]

Fine-Levy JB, Girardot MN, Derby CD, and Daniel PC. Differential associative conditioning and olfactory discrimination in the spiny lobster Panulirus argus. Behav Neural Biol 49: 315–331, 1988.[CrossRef][Web of Science][Medline]

Firestein S. How the olfactory system makes sense of scents. Nature 413: 211–218, 2001.[CrossRef][Medline]

Galizia CG, Sachse S, Rappert A, and Menzel R. The glomerular code for odor representation is species specific in the honeybee Apis mellifera. Nat Neurosci 2: 473–487, 1999.[CrossRef][Web of Science][Medline]

Goldberg SJ and Moulton DG. Olfactory bulb responses telemetered during an odor discrimination task in rats. Exp Neurol 96: 430–442, 1987.[CrossRef][Web of Science][Medline]

Gorsuch RL. Factor Analysis (2nd ed.). Hillsdale, NJ: Erlbaum, 1983.

Gray CM, Maldonado PE, Wilson M, and McNaughton B. Tetrodes markedly improve the reliability and yield of multiple single-unit isolation from multi-unit recordings in cat striate cortex. J Neurosci Methods 63: 43–54, 1995.[CrossRef][Web of Science][Medline]

Harris KD, Henze DA, Csicsvari J, Hirase H, and Buzsaki G. Accuracy of tetrode spike separation as determined by simultaneous intracellular and extracellular measurements. J Neurophysiol. 84: 401–414, 2000.[Abstract/Free Full Text]

Henze DA, Borhegyi Z, Csicsvari J, Mamiya A, Harris KD, and Buzsaki G. Intracellular features predicted by extracellular recordings in the hippocampus in vivo. J Neurophysiol 84: 390–400, 2000.[Abstract/Free Full Text]

Hildebrand JG. Analysis of chemical signals by the nervous system. Proc Natl Acad Sci USA 92: 67–74, 1995.[Abstract/Free Full Text]

Hildebrand JG. Olfactory control of behavior in moths: central processing of odor information and the functional significance of olfactory glomeruli. J Comp Physiol [A] 178: 5–19, 1996.

Hildebrand JG and Shepherd GM. Mechanisms of olfactory discrimination: converging evidence for common principles across phyla. Annu Rev Neurosci 20: 595–631, 1997.[CrossRef][Web of Science][Medline]

Homberg U, Christensen TA, and Hildebrand JG. Structure and function of the deutocerebrum in insects. Annu Rev Ent 34: 477–501, 1989.[CrossRef][Web of Science][Medline]

Johnson BA, Woo CC, and Leon M. Spatial coding of odorant features in the glomerular layer of the rat olfactory bulb. J Comp Neurol 393: 457–471, 1998.[CrossRef][Web of Science][Medline]

Kashiwadani H, Sasaki YF, Uchida N, and Mori K. Synchronized oscillatory discharges of mitral/tufted cells with different molecular receptive ranges in the rabbit olfactory bulb. J NeuroPhysiol 82: 1786–1792, 1999.[Abstract/Free Full Text]

Kjaer TW, Hertz JA, and Richmond BJ. Decoding cortical neuronal signals: network models, information estimation and spatial tuning. J Comput Neurosci 1: 109–139, 1994.[CrossRef][Medline]

Koehl MA, Koseff JR, Crimaldi JP, McCay MG, Cooper T, Wiley MB, and Moore PA. Lobster sniffing: antennule design and hydrodynamic filtering of information in an odor plume. Science 294: 1948–1951, 2001.[Abstract/Free Full Text]

Laska M, Ayabe-Kanamura S, Hubener F, and Saito S. Olfactory discrimination ability for aliphatic odorants as a function of oxygen moiety. Chem Senses 25: 189–197, 2000.[Abstract/Free Full Text]

Laubach M, Shuler M, and Nicolelis MA. Independent component analyses for quantifying neuronal ensemble interactions. J Neurosci Methods 94: 141–154, 1999.[CrossRef][Web of Science][Medline]

Laurent G. Dynamical representaion of odors by oscillating and evolving neural assemblies. Trends Neurosci 19: 489–496, 1996.[CrossRef][Web of Science][Medline]

Laurent G. Olfactory network dynamics and the coding of multidimensional signals. Nat Rev Neurosci 3: 884–895, 2002.[CrossRef][Web of Science][Medline]

Laurent G and Davidowitz H. Encoding of olfactory information with oscillating neural assemblies. Science 265: 1872–1820, 1994.[Abstract/Free Full Text]

Laurent G and Naraghi M. Odorant-induced oscillations in the mushroom bodies of the locust. J Neurosci 14: 2993–3004, 1994.[Abstract]

Laurent G, Stopfer M, Friedrich RW, Rabinovich M, Volkovskii A, and Di Abarbanel H. Odor encoding as an active dynamical process: experiments, computation, and theory. Annu Rev Neurosci 24: 263–297, 2001.[CrossRef][Web of Science][Medline]

Lei H, Christensen TA, and Hildebrand JG. Local inhibition modulates odor-evoked synchronization of glomerulus-specific output neurons. Nat Neurosci 5: 557–65, 2002.[CrossRef][Web of Science][Medline]

Linster C, Johnson BA, Yue E, Morse A, Xu Z, Hingco EE, Choi Y, Choi M, Messiha A, and Leon M. Perceptual correlates of neural representations evoked by odorant enantiomers. J Neurosci 21: 9837–9843, 2001.[Abstract/Free Full Text]

MacLeod K and Laurent G. Distinct mechanisms for synchronization and temporal patterning of odor-encoding neural assemblies. Science 274: 976–979, 1996.[Abstract/Free Full Text]

Malnic B, Hirono J, Sato T, and Buck LB. Combinatorial receptor codes for odors. Cell 96: 713–723, 1999.[CrossRef][Web of Science][Medline]

Matsumoto SG and Hildebrand JG. Olfactory mechanisms in the moth Manduca sexta: response characteristics and morphology of central neurons in the antennal lobes. Proc R Soc Lond B Biol Sci 213: 249–277, 1981.[Abstract/Free Full Text]

McClurkin JW, Optican LM, Richmond BJ, and Gawne TJ. Concurrent processing and complexity of temporally encoded neuronal messages in visual perception. Science 253: 675–677, 1991.[Abstract/Free Full Text]

McNaughton BL, O'Keefe J, and Barnes CA. The stereotrode: a new technique for simultaneous isolation of several single units in the central nervous system from multiple unit records. J Neurosci Methods 8: 391–397, 1983.[CrossRef][Web of Science][Medline]

Mombaerts P. Seven-transmembrane proteins as odorant and chemosensory receptors. Science 286: 707–711, 1999.[Abstract/Free Full Text]

Mori K. Relation of chemical structure to specificity of response in olfactory glomeruli. Curr Opin Neurobiol 5: 467–474, 1995.[CrossRef][Web of Science][Medline]

Müller D, Abel R, Brandt R, Zockler M, and Menzel R. Differential parallel processing of olfactory information in the honeybee, Apis mellifera L. J Comp Physiol [A] 188: 359–370, 2002.[CrossRef]

Musial PG, Baker SN, Gerstein GL, King EA, and Keating JG. Signal-to-noise ratio improvement in multiple electrode recording. J Neurosci Methods 115: 29–43, 2002.[CrossRef][Web of Science][Medline]

Nicolelis MA and Chapin JK. Spatiotemporal structure of somatosensory responses of many-neuron ensembles in the rat ventral posterior medial nucleus of the thalamus. J Neurosci 14: 3511–3532, 1994.[Abstract]

Ressler KJ, Sullivan SL, and Buck LB. Information coding in the olfactory system: evidence for a stereotyped and highly organized epitope map in the olfactory bulb. Cell 79: 1245–1255, 1994.[CrossRef][Web of Science][Medline]

Rubin BD and Katz LC. Optical imaging of odorant representations in the mammalian olfactory bulb. Neuron 23: 499–511, 1999.[CrossRef][Web of Science][Medline]

SAS Institute Inc. SAS/STAT User's Guide Version 6 (4th ed.). Cary, NC: SAS Institute, 2001.

Sachse S, Rappert A, and Galizia GC. The spatial representation of chemical structures in the antennal lobe of honeybees: steps towards the olfactory code. Eu J Neurosci 11: 3970–3982, 1999.

Sachse S and Galizia CG. Role of inhibition for temporal and spatial odor representation in olfactory output neurons: a calcium imaging study. J Neurophysiol 87: 1106–1117, 2002.[Abstract/Free Full Text]

Schmidt M and Ache BW. Processing of antennular input in the brain of the spiny lobster, Panulirus argus. I. Non-olfactory chemosensory and mechanosensory pathway of the lateral and median antennular neuropils. J Comp Physiol [A] 178: 579–604, 1996.[CrossRef]

Schoppa NE and Westbrook GL. Glomerulus-specific synchronization of mitral cells in the olfactory bulb. Neuron 31: 639–651, 2001.[CrossRef][Web of Science][Medline]

Schoppa NE and Westbrook GL. AMPA autoreceptors drive correlated spiking in olfactory bulb glomeruli. Nat Neurosci 5: 1194–1202, 2002.[CrossRef][Web of Science][Medline]

Shepherd GM. A molecular vocabulary for olfaction. Ann NY Acad Sci 510: 98–103, 1987.[Web of Science][Medline]

Shields VDC and Hildebrand JG. Responses of a population of antennal olfactory receptor cells in the female moth Manduca sexta to plant-associated volatile organic compounds. J Comp Physiol [A] 186: 1135–1151, 2001.[CrossRef]

Shipley MT and Ennis M. Functional organization of the olfactory system. J Neurobiol 30: 123–176, 1996.[CrossRef][Web of Science][Medline]

Slotnick BM, Bell GA, Panhuber H, and Laing DG. Detection and discrimination of propionic acid after removal of its 2-DG identified major focus in the olfactory bulb: a psychophysical analysis. Brain Res 762: 89–96, 1997.[CrossRef][Web of Science][Medline]

Slotnick B and Bodyak N. Odor discrimination and odor quality perception in rats with disruption of connections between the olfactory epithelium and olfactory bulbs. J Neurosci 22: 4205–4216, 2002.[Abstract/Free Full Text]

Smith BH and Menzel R. An analysis of variability in the feeding motor program of the honey bee: the role of learning in releasing a modal action pattern. Ethol 82: 68–81, 1989.

Stopfer M, Bhagavan S, Smith BH, and Laurent G. Imparied odour discrimination on desynchronization of odor-encoding neural assemblies. Nature 390: 70–74, 1997.[CrossRef][Medline]

Stopfer M, Jayaraman V, and Laurent G. Intensity versus identity coding in an olfactory system. Neuron 39: 991–1004, 2003.[CrossRef][Web of Science][Medline]

Stopfer M and Laurent G. Short-term memory in olfactory network dynamics. Nature 402: 664–668, 1999.[CrossRef][Medline]

Uchida N and Mainen ZF. Speed and accuracy of olfactory discrimination in the rat. Nat Neurosci 6: 1224–1229, 2003.[CrossRef][Web of Science][Medline]

Uchida N, Takahashi YK, Tanifuji M, and Mori K. Odor maps in the mammalian olfactory bulb: domain organization and odorant structural features. Nat Neurosci 3: 1035–1043, 2000.[CrossRef][Web of Science][Medline]

Vassar R, Chao SK, Sitcheran R, Nunez JM, Vosshall LB, and Axel R. Topographic organization of sensory projections to the olfactory bulb. Cell 79: 981–991, 1994.[CrossRef][Web of Science][Medline]

Vickers NJ, Christensen TA, Baker TC, and Hildebrand JG. Odour-plume dynamics influence the brain's olfactory code. Nature 410: 466–470, 2001.[CrossRef][Medline]

Vosshall LB, Wong AM, and Axel R. An olfactory sensory map in the fly brain. Cell 102: 147–159, 2000.[CrossRef][Web of Science][Medline]

Wehr M and Laurent G. Odor encoding by temporal sequences of firing in oscillating neural assemblies. Nature 384: 162–166, 1996.[CrossRef][Medline]

Willis MA and Baker TC. Effects of intermittent and continuous pheromone stimulation on the flight behavior of the oriental fruit moth, Grapholita molesta. Physiol Ent 9: 341–358, 1984.[CrossRef]

Wise PM and Cain WS. Latency and accuracy of discriminations of odor quality between binary mixtures and their components. Chem Senses 25: 247–265, 2000.[Abstract/Free Full Text]

This article has been cited by other articles:

				A. M. Dacks, T. A. Christensen, and J. G. Hildebrand Modulation of Olfactory Information Processing in the Antennal Lobe of Manduca sexta by Serotonin J Neurophysiol, May 1, 2008; 99(5): 2077 - 2085. [Abstract] [Full Text] [PDF]

				E. K. Mwilaria, C. Ghatak, and K. C. Daly Disruption of GABAA in the Insect Antennal Lobe Generally Increases Odor Detection and Discrimination Thresholds Chem Senses, March 1, 2008; 33(3): 267 - 281. [Abstract] [Full Text] [PDF]

				K. C. Daly, L. A. Carrell, and E. Mwilaria Characterizing Psychophysical Measures of Discrimination Thresholds and the Effects of Concentration on Discrimination Learning in the Moth Manduca sexta Chem Senses, January 1, 2008; 33(1): 95 - 106. [Abstract] [Full Text] [PDF]

				S. Decker, S. McConnaughey, and T. L. Page Circadian regulation of insect olfactory learning PNAS, October 2, 2007; 104(40): 15905 - 15910. [Abstract] [Full Text] [PDF]

				T. A. Christensen Making Scents Out of Spatial and Temporal Codes in Specialist and Generalist Olfactory Networks Chem Senses, January 1, 2005; 30(suppl_1): i283 - i284. [Full Text] [PDF]

				H. Lei, T. A. Christensen, and J. G. Hildebrand Spatial and Temporal Organization of Ensemble Representations for Different Odor Classes in the Moth Antennal Lobe J. Neurosci., December 8, 2004; 24(49): 11108 - 11119. [Abstract] [Full Text] [PDF]

				K. C. Daly, T. A. Christensen, H. Lei, B. H. Smith, and J. G. Hildebrand Learning modulates the ensemble representations for odors in primary olfactory networks PNAS, July 13, 2004; 101(28): 10476 - 10481. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) Supplemental Data All Versions of this Article: 92/1/236    most recent 01132.2003v1 Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (16) Citing Articles via Google Scholar Google Scholar Articles by Daly, K. C. Articles by Smith, B. H. Search for Related Content PubMed PubMed Citation Articles by Daly, K. C. Articles by Smith, B. H.