1. Neuroscience
Download icon

The contribution of temporal coding to odor coding and odor perception in humans

  1. Ofer Perl
  2. Nahum Nahum
  3. Katya Belelovsky
  4. Rafi Haddad  Is a corresponding author
  1. Bar-Ilan University, Israel
Research Article
  • Cited 1
  • Views 672
  • Annotations
Cite this article as: eLife 2020;9:e49734 doi: 10.7554/eLife.49734

Abstract

Whether neurons encode information through their spike rates, their activity times or both is an ongoing debate in systems neuroscience. Here, we tested whether humans can discriminate between a pair of temporal odor mixtures (TOMs) composed of the same two components delivered in rapid succession in either one temporal order or its reverse. These TOMs presumably activate the same olfactory neurons but at different times and thus differ mainly in the time of neuron activation. We found that most participants could hardly discriminate between TOMs, although they easily discriminated between a TOM and one of its components. By contrast, participants succeeded in discriminating between the TOMs when they were notified of their successive nature in advance. We thus suggest that the time of glomerulus activation can be exploited to extract odor-related information, although it does not change the odor perception substantially, as should be expected from an odor code per se.

Introduction

How odors are encoded by the brain is an open fundamental research question in neuroscience, echoing a more general debate regarding how the brain encodes information (Uchida et al., 2014). Odor perception starts with odorant molecules binding to olfactory receptors. Different odorants bind to a unique set of possibly overlapping olfactory receptors. In rodents, each olfactory sensory neuron expresses one olfactory sensory receptor and projects to one glomerulus (Mombaerts, 2006). Each odorant activates a unique set of glomeruli, which in turn activate the olfactory bulb (OB) output neurons – the mitral and tufted (M/T) cells. These findings led to the hypothesis that odors are encoded by a combinatorial code composed of the set of glomeruli activated by each odor. This code was termed the spatial code or identity code (Friedrich and Korsching, 1997; Friedrich and Korsching, 1998; Galizia et al., 1999; Johnson and Leon, 2007; Malnic et al., 1999; Rubin and Katz, 1999; Uchida et al., 2000). Owing to the large number of different glomeruli, this coding scheme is theoretically capable of encoding any number of odors (Koulakov et al., 2007).

It has long been observed that odor stimulation evokes odor- and cell-specific temporal patterns of activity in the OB and antennal lobe, which are not directly related to the dynamics of the olfactory stimulus. These observations led to the hypothesis that odors are represented by spatially and temporally distributed ensembles of active neurons. As discussed in detail in Uchida et al. (2014), there are two main models that attempt to describe the role of temporal coding in olfaction: the Hopfield latency coding model and the Laurent slow-evolving decorrelation model.

According to the Hopfield model, odors are encoded by the identity of the odor-activated neurons and their time of activation in relation to some oscillatory cycle (Figure 1A) (Hopfield, 1995). This coding model provides a simple mechanism for concentration-invariant odor recognition: when the odor concentration changes, multiple neurons shift their spike timing together, such that the relative timing across neurons remains unchanged. According to this model, time is part of the odor code and therefore two odors that activate a similar set of glomeruli but at different times should evoke two distinguishable percepts. Time can be measured relative to some internal oscillatory cycle such as internal gamma oscillations or the respiration cycle (i.e., phase coding), or relative to inhalation onset regardless of the respiration duration (i.e., latency coding) or relative to the activation times of other neurons (i.e., relative time coding).

Figure 1 with 1 supplement see all
Construction of odors differing mainly in their odor-elicited temporal dynamics.

(A) A schematic diagram showing three glomeruli (colored circles) responding to two odors over two cycles of a putative oscillatory mechanism (black line). Odors A and B activate the same glomeruli but at different times. According to the Hopfield model, this difference in activation time should be sufficient to render these two odor codes as different odors. (B) Two odor trajectories plotted in phase space. The two trajectories represent two different odors as their trajectories are different in phase space. (C) Left panel: experimental setup. Participants were presented with two odors dispensed from two separate canisters using a computer-controlled air dilution olfactometer. Another canister was used to deliver clean air (marked as C). Nasal respiration was recorded using a nasal cannula attached to a pressure-sensitive spirometry sensor. Right panel: experimental design. Two odor pulses were presented consecutively spaced by a time delay (Δt) between them. Eight seconds later, we presented either the same order of odor presentations or a reversed one. Participants then reported whether the odors were the same or different. Twenty-three seconds later, this procedure was repeated. Each participant underwent 20 trials. Red, odor A; blue, odor B; ITI, inter-trial-interval. (D) A mockup model of glomeruli responses to two odors delivered one after the other with a short delay in between, both in forward and in reverse order. Red and blue circles represent glomeruli activated by odors A and B, respectively. Purple circles represent glomeruli activated by odors A+B. The order of odor presentation elicits different time sequences of glomeruli activation. (E) Possible trajectories of the two TOMs in phase space. The first trajectory (thick lines) is composed of odor A (red), odors A+B (purple) and then odor B (blue). The second trajectory (thin lines) represents odor B, odor B+A, and odor B. The two trajectories are relatively similar, although the directions are opposite. (F) A typical example of a 20-trial experimental session. Left panel: black traces depict several typical nasal respiration trials overlaid. Positive values represent inhalation. Red and blue triggers represent a single example of odor A and B pulses. Right panel: examples of trials in which the TOMs were presented entirely concurrent (top and bottom) or not concurrent (middle) with sniffing.

Experimental results have provided some support for the Hopfield model. In rodents, the latency of glomerular activation showed stimulus-specific temporal patterns (Spors, 2006; Spors and Grinvald, 2002). Moreover, increasing the odor concentration reduced the glomeruli onset times while preserving the temporal sequence of glomerular activation, suggesting that the temporal sequence of glomeruli activation encodes the odor identity, whereas the time of activation encodes concentration (Spors and Grinvald, 2002). In larvae of Xenopus laveis (a model organism that does not sniff), latency from stimulus onset across M/T cells reliably conveyed stimulus information about odor identity and concentration (Junek et al., 2010). Thus, these studies show that spike or glomeruli latency with respect to slow oscillations (e.g., respiration cycle), as well as latency from odor onset on the order of hundreds of milliseconds, can convey reliable odor information. Consistent with latency coding, a study reported that downstream neurons in mice are sensitive to the relative activation times of optogenetically activated output neurons in the olfactory bulb (Haddad et al., 2013). This finding suggested that cortical neurons can decode specific glomerular activation sequences. Finally, several behavioral studies have shown that mice are able to discriminate between optogenetic stimulations of olfactory neurons, which differ mainly in their activation times, either relative to the sniff cycle or to each other (Rebello et al., 2014; Smear et al., 2011; Smear et al., 2013). These results can be interpreted as behavioral evidence supporting relative- or phase-latency coding as part of the odor code.

A more recent variant of the latency model suggested that odors are encoded by the early components of the activated glomeruli (Schaefer and Margrie, 2007; Wilson et al., 2017), whereas the components activated later in the evolving spatiotemporal map could possibly be utilized to differentiate between highly similar odors (Schaefer and Margrie, 2007). Consistent with this model, a recent study showed that neurons in the piriform cortex tend to respond to the early components and ignore the later ones (Bolding and Franks, 2018). Regardless of the exact variant, all of these coding models posit that two odors that differ mainly in terms of their temporal code should elicit a different percept, as time is part of the odor code.

The Laurent model, by contrast, stresses that populations of neurons exhibit synchronized oscillatory activity, but that each neuron only transiently participates in this population activity (Laurent, 2002). The identities of neurons that participate in the oscillatory ensemble change over time and form the odor representation that is gradually decorrelated to improve odor encoding in neural space (Friedrich and Laurent, 2001; Friedrich and Laurent, 2004). Thus, the odor representation can be considered a trajectory in phase space (Laurent et al., 2001), where different odors generate different trajectories (Figure 1B). Whether the whole trajectory or only parts of it are used for decoding is unknown. According to this model, temporal dynamics are not a necessary part of the odor code because they only facilitate coding optimization (Laurent, 2002). According to one possible interpretation of this model, two odors that activate a similar set of glomeruli but at different times are not necessarily perceived as different, as long as some parts of their evolving trajectories are similar; see Laurent et al. (2001) for a full discussion of this model. Support for this model comes from observations of temporally evolving neural dynamics and subsequent analyses demonstrating that the odor representation becomes sparser and more decorrelated over time (Friedrich and Laurent, 2004; Friedrich and Laurent, 2001; Gschwend et al., 2015; Gupta and Stopfer, 2014; Laurent et al., 1996; Wehr and Laurent, 1996) and that disturbing gamma oscillations impairs odor discrimination (Stopfer et al., 1997).

Thus overall, the different variants of the latency model predict that two odors that differ substantially in glomerular activation time should be perceived as two different odors as such temporal dynamics are part of the odor code per se. According to the Laurent model, time may act as an orthogonal component to the odor percept, such that two odors that elicit neural responses that differ in their temporal dynamics may be perceived as the same if their neural trajectories resemble one another in some way.

There is still an ongoing debate as to which of these temporal coding models is used by the mammalian olfactory system. Here, we examined whether glomerular activation times affect odor perception in humans. Using temporal odor sequences, we manipulated temporal features that are involved in odor coding and measured their effect on odor perception during discrimination and odor-rating tasks.

Results

Construction of odors that differ in their odor-elicited neural temporal dynamics

To test how temporal coding affects odor perception, we manipulated relative glomerular activation times using temporal odor mixtures (TOMs). We presented two precise consecutive odor pulses (A and B) separated by a precise short delay (Δt), and then 8 s later, we presented the same odors again, either in the same order or the reverse order. That is, we first presented TOM AΔtB , and 8 s later, we presented either TOM BΔtA or TOM AΔtB  again (Figure 1C). Delivering two odors in a different order activates the glomeruli at a different relative time and forms a different latency code (Figure 1D), although the odor trajectories in phase space may still share some similar features (Figure 1E).

The participants were instructed to decide whether these two consecutively presented TOMs were the same or different. Each odorant was delivered for exactly 200 ms and the second odorant within each TOM was delivered precisely at a controllable, experimentally defined time (Δt) following the onset of the first odorant (Figure 1C). This design ensured that both TOMs had the exact same duration and thus safeguarded against the possible use of odor duration as a cue to facilitate discrimination. Participants were not informed that the TOMs were composed of two odors, to avoid discrimination based on identifying which odor was first (Laing et al., 1994).

We verified that odor delivery and clearance were precise across trials with a mini photo-ionizer detector (PID) (miniPID, Aurora Scientific). The average latency of the PID-reported odor signal onset was within the range of 20–40 ms, and the odor concentration returned to 10% of its peak value within 30–50 ms after odor offset (Figure 1—figure supplement 1A–B). The time of glomerulus activation depends on odorant-receptor dynamics, among other factors. That said, even assuming some delay in glomerulus activation due to these interactions, switching the order of the odors in the TOM will nevertheless create a substantial difference in the glomerular activation times, because the glomerular activation times of two odors in the first TOM were shifted by +Δt and by –Δt in the second TOM. Thus, our two TOMs were expected to activate a similar set of glomeruli, but the sequence of glomerular activation time of each TOM was substantially different.

We used an inter-stimulus-interval (ISI) of 8 s between the two TOMs rather than the more common 10–20 s interval employed in human olfaction experimentation, because we observed in pilot sessions that when ISI was longer, performance was significantly hampered even when the participants were asked to discriminate between the two constituent odors (i.e., odor A vs. odor B). This was probably due to the fact that the task essentially became an olfactory match-to-sample working memory task rather than a discrimination task, which is more difficult as the ISI gets longer (Arzi et al., 2014; Zelano et al., 2009). An inter-trial-interval (ITI) of at least 23 s separated discrimination trials, depending on the participants’ response times in the trial (Figure 1C). To minimize sensory fatigue resulting from repeated odor exposure, each participant conducted a maximum of 20 trials (i.e., 20 × 2 = 40 odor trials), which is within the range used in olfaction psychophysics experiments (Johnson et al., 2003; Weiss et al., 2012).

Thus, this experimental setup allowed us to disentangle the contribution of relative- and phase-latency coding to odor coding. If glomerular activation times are part of the odor code, as suggested by the Hopfield model, different TOMs should elicit different odor percepts. If different percepts are indeed evoked, we should be able to assess these differences by asking the participants to describe each TOM using a set of verbal odor descriptors.

The effect of odor temporal dynamics on odor perception

We tested participants' ability to discriminate between TOMs. First, we used two odors that are highly familiar to the majority of the population and thus can be easily named and processed as consolidated olfactory objects (Frank et al., 2011; Olofsson and Gottfried, 2015). We used extracts of orange (ORG) and cinnamon (CIN), which are naturally occurring substances often encountered in the context of food products (see 'Materials and methods'). We first tested delays of 300 ms and 600 ms between the two constituting odors, as we assumed that these delays would be long enough to generate substantially distinguishable temporal dynamics that would result in the elicitation of distinct percepts by the two TOMs. We did not test delays longer than 600 ms to ensure that the two odors would both be delivered during a single inhalation phase. Furthermore, long delays increase the likelihood that participants may realize that the TOMs are composed of two odors arriving one after the other, and this the likelihood that they may use this information for discrimination rather than the hypothesized change in odor perception itself.

We observed that out of the 10 participants who were tested with a time delay of 300 ms, nine failed to discriminate between the TOMs in a significant manner; that is, their individual performance did not differ from chance (p>0.05, binomial test per participant). Group mean and median success rates were also close to chance levels (success rate: mean = 0.541, p=0.23, t-test; median = 0.55, p=0.51, two-sided sign test, N = 10, Figure 2A left violet panel). An analysis based on Bayesian statistics (see 'Materials and methods') further supported this conclusion (Bayesian one sample t-test: BF10 = 1.039, error = 8.6E-5%). When we used a longer interval of Δt = 600 ms between odor constituents, seven out of the nine participants (78% of the cohort) failed to discriminate significantly between the TOMs. Overall, this resulted in only a slight improvement in group success rate (success rate: mean = 0.60, p=0.16, t-test; median = 0.5, p=1, two-sided sign test, N = 9, Figure 2A, left violet panel; Bayesian one sample t-test: BF10 = 1.626, error = 2.3E–4%). Notably, one participant obtained a perfect score in the Δt = 600 ms condition. In a post-session debriefing, this participant reported perceiving the two odor constituents (CIN and ORG) arriving one after the other (had we discarded this participant from the analysis, the statistical output would have changed as follows: success rate: mean = 0.55 p=0.33, t-test; median = 0.5, p=1, two-sided sign test, N = 8; Bayesian one sample t-test: BF10 = 0.841, error = 1.9E–5%).

Figure 2 with 1 supplement see all
Participants could not discriminate between TOMs composed of two familiar odors.

(A) Violin graph of the success rates of individual participants in discriminating between TOMs composed of ORG and CIN with Δt = 300 ms and Δt = 600 ms. Gray data points represent the success rate of an individual participant. Participants who scored significantly higher than chance are circled in red (p<0.05 binomial test, uncorrected for multiple comparisons). Means and medians are marked by black and blue diamonds, respectively. Standard deviation around the mean is marked in black. The result of a two-sided sign test, comparing the median success rate to the expected chance success rate of 0.5, and the Bayesian statistics are denoted above each group. Overlapping data points are shifted sideways across the x axis for visualization purposes. The dashed line marks the chance success rate (0.5). The number of participants in each group is denoted by N. Left violet box: participants’ success rates for the TOM experiments. Right gray box: results of four control experiments validating that failure to discriminate between TOMs is not due to suboptimal choice of experimental parameters or odor delivery system. Note that in all control experiments, the majority of the cohort performed significantly above chance and the mean and median success rates were significantly above chance level. *, ** and *** represent p<0.05, p<0.01 and p<0.001, respectively (same applies for all figures in this manuscript). (B) Success rates of the TOM discrimination experiment split into trials in which the TOMs presented were the same or when they were different (Δt = 300 ms). No significant difference was detected between the two group means (two-tailed t-test, t(9) = −0.11 p=0.91). (C) Equally perceived intensity of the odors constituting the TOMs. The bar graph depicts the mean perceived intensity of the odors used (ORG and CIN). Error bars are the standard errors of the means. Gray dots denote the individual participant success rates. No significant difference was detected between the two group means (two-tailed t-test, t(10) = −0.45 p=0.66).

As it was evident that the mean was susceptible to outliers, we focused on the median statistic for the group success rate. That said, the results remained virtually the same when using the mean instead.

Participants made a similar number of wrong answers in trials in which the second TOM was the same as the first (‘Same’) or different (‘Diff.’). This suggests that neither condition was easier than the other and that there was no strong bias in the participants’ report towards one answer over the other (analysis for Δt = 300 ms: Same = 0.505 ± 0.028, Diff = 0.495 ± 0.011, paired t-test: t(9) = 0.11, p=0.92, Cohen’s d = 0.45, Figure 2B).

One explanation for this overall failure to discriminate between the TOMs could potentially stem from setting inappropriate parameters for the experimental paradigm (e.g., the relatively short ISI, the odor duration or the participants’ motivation). To rule out these alternatives, we ran an extensive battery of control experiments using a delay of 300 ms to minimize the probability that participants would notice that the TOMs were composed of two consecutive odors. First, we verified that the perceived intensity of each of the constituting odors was similar. There was no significant difference in intensity (perceived intensity: ORG = 3.89 ± 0.12, CIN = 4.14 ± 0.15, paired t-test: t(10) = –0.45, p = 0.66, Cohen’s d = –1.76, Figure 2C).

Then, we tested whether participants could discriminate between the two constituting odors when they were presented alone. We ran two experiments: 1) A vs. (A / B) and 2) B vs. (A / B). In this experiment, participants were presented with one odor and 8 s later they were presented with either the same odor or the second odor. They reported whether the two odors were the same or different. The logic behind this control was that if participants’ failure to discriminate between the TOMs was due to some inadequate experimental parameter or setup accuracy, this would persist in the control paradigm, and they should also fail to discriminate between these odor sets. We found that all participants except one could discriminate between these odors (median success rate = 0.9 for A vs. (A / B). two-sided sign test p = 0.0156, Bayesian one sample t-test BF10 = 425.2, error = undetectable, N = 7; median success rate = 0.84 for B vs. (A / B), two-sided sign test: p = 0.0156, Bayesian one sample t-test BF10 = 121.9, error = undetectable, N = 7, Figure 2A, gray panel).

Finally, we ran two additional control experiments to test whether participants could discriminate between the following two odor sets: 1) AB vs. (AB / AC), and 2) AB vs. (AB / BC) (AB denotes AΔtB, AC denotes AΔtC , BC denotes BΔtC , and C was a clean air stimulus of the same duration as odors A and B (200 ms)). These paradigms employed the same temporal dynamics as those in the TOMs experiment; however, here, the two odors also differed in odor content in that one TOM contained a clean air stimulus. Importantly, in these control experiments, the first odor was the same as that in the TOMs experiment (i.e., AB). In other words, this control experiment was used to validate that when the TOMs are the same, participants perceive them as same, and when they are in some way different, participants are able to notice this. The findings showed that most participants could indeed discriminate between these odors (AB vs (AB / AC) (median success rate = 0.75; two-sided sign test, p = 0.004; Bayesian one sample t-test, BF10 = 219.4; error = undetectable; N = 9) and AB vs. (AB / BC) (median success rate = 0.9; two-sided sign test, p = 0.031; Bayesian one sample t-test, BF10 = 1031; error = undetectable; N = 6; Figure 2A gray panel, right plots), although the task was slightly more difficult than discriminating between the constituting odors, probably because the two odors were more similar as they shared a component. Note that this control experiment also confirms that the constituent odors did not mask each other (at least for most of the cohort). Had this been the case, participants would have failed on the discrimination task for at least one of these odor pairs. The participants who took part in the control experiments did not participate in the TOMs experiment to prevent improvement through learning.

To compare performance on TOM discrimination and in the controls directly, we conducted an analysis of variance (ANOVA) and observed a strong main effect experiment type (F(5,42) = 10.061, partial eta-squared = 0.545, p<0.0001). Planned post-hoc comparisons (corrected with Tukey HSD) revealed that the discrimination accuracies for TOMs tested with a delay of 300 or 600 ms were similar (p=0.89) and, more importantly, that performance in TOM discrimination was significantly lower than that in all control experiments, with the exception of one comparison between Δt = 600 ms and the control AB vs. (AB/AC), which was marginally significant (p=0.067). All of the other controls maintained comparable high performance (all p-values for the TOM compared with the controls < 0.012, all p-values for cross-control comparisons > 0.68).

Note that there were fewer participants in the control experiments than in the main experiment. Nevertheless, the success rates of all four control experiments were high, whereas performance in the TOM experiments was close to chance. This indicates that the result was mostly independent of the number of participants. Moreover, the success rates in the TOM experiments were quite low (range: 50–60% for both 300 and 600 ms and for both the median and the mean, with Cohen’s d = 0.43 and 0.59, respectively, Figure 2A). Thus, it seems unlikely that increasing the number of participants would increase mean or median success rates, although it could have made them statistically significant. In other words, the main message conveyed by this set of experiments is that the success rates in the TOM experiments were low, whereas those in the controls were not.

Finally, the TOMs were delivered following an auditory tone, such that in each trial, the TOM could potentially be presented at a different phase of the inhalation. One potential concern is that the respiration phase in which the odor was encountered may have contributed to the odor code, so that participants may have failed to discriminate between TOMs when they were delivered at different respiration phases. That said, the fact that participants performed adequately in the control experiments suggests that phase differences were unlikely to be the reason for the failure in the TOMs experiment. We tested this directly by comparing respiratory phases across TOMs (first/second) and outcome (correct/incorrect) and observed a consistent phase distribution that was not associated with accuracy (see 'Materials and methods' and Figure 2—figure supplement 1).

Taken together, the results of these experiments and the ensuing analyses strongly indicate that the inability of participants to discriminate TOMs is unlikely to have been the result of an inadequate experimental parameter or setup limitation, but rather arose because the neural temporal dynamics generated in this experiment did not affect odor percept substantially.

The effect of rapid odor temporal dynamics and odor similarity on odor perception

The results so far imply that human participants perform poorly in discrimination between TOMs when the delay between the components is 300 or 600 ms and when the constituent odors are familiar. We next tested whether the failure to discriminate between these TOMs could stem from temporal dynamics within faster time scales than the ones we used. Another possible reason for the failure in discrimination is the fact that the constituent odors, although very familiar, were both pleasant, a core property in human olfactory perception, and could therefore have been perceived similarly (Haddad et al., 2008; Khan et al., 2007), which may have rendered the TOMs hard to discriminate. In fact, 3 out of 14 participants failed to discriminate between the TOM constituting odors and 4 out of 15 failed to discriminate between a TOM and a single odor constituent (Figure 2A, gray panel). With this in mind, we decided to introduce a second pair of odorants, namely citral (CTL) and dimethyl-trisulfide (DMTS). These odorants have markedly more distinct percepts in that CTL is perceived as lemony/citrusy, whereas DMTS is perceived as sulfurous or onion-like (see 'Materials and methods'). In this experiment, we also tested whether shorter Δt intervals of 150 and 75 ms along with intervals of 300 and 600 ms would have an effect, while maintaining all the other experimental parameters as in the previous experiment.

We first confirmed that this new set of odors reflected pronounced differences in percept (Figure 3—figure supplement 1). We then tested the participants’ ability to discriminate between TOMs composed of CTL and DMTS using four delay durations (Δt = 75, 150, 300 and 600 ms). In line with the previous experiment, the majority of the cohort (~88% or 38 out of 43 participants) failed to discriminate significantly between the two TOMs, regardless of the time delay (p>0.05, binomial test per each participant, uncorrected for multiple comparisons, Figure 3A left box). The median for the group success rates was also not significantly higher than chance for Δt = 75 and 150 ms and marginally significant for Δt = 300 and 600 ms (Figure 3A). These results suggest that when the delay is relatively long, and the two constituent odors are dissimilar, the two TOMs can be discriminated, although the mean and median success rates were only slightly above chance and the majority of the participants failed to discriminate between them. This observation further supports the finding that temporal dynamics as elicited here do not contribute substantially to odor percept.

Figure 3 with 1 supplement see all
The majority of the participants could not discriminate between TOMs that are composed of highly dissimilar odors.

(A) Left box: success rates for discrimination between TOMs composed of CTL and DMTS as a function of Δt. Color code as in Figure 2. Gray right box: four control experiments demonstrating that the vast majority of participants can discriminate between the constituent odors or TOMs when one of the odors is clean air (labeled as C). '~' denotes marginally significant (p>0.05 and p<0.07). (B) Extensive training does not substantially improve discrimination between TOMs. Lower panel: discrimination score progression over five sessions of 40 trials each. Participants who reached and maintained a score benchmark of 0.7 (red horizontal line) in one session ended their participation and they did not return for additional sessions. Chance performance (0.5) is marked by a black horizontal line. Thin gray lines mark individual participants and the thick blue line is the average of all participants over a moving window of 20 trials. The upper panel reflects trial-by-trial performance (columns) per participant (rows), where the color of each tile denotes accuracy: correct (blue), incorrect (red), or removed because the odor presentation was not in sync with sniffing (black). (C) Mean success rate for each participant across the five sessions. Group mean per session is depicted by a thick gray line along with vertical bars representing the standard error.

We next applied the same battery of control experiments used previously and observed that group performance in discrimination between the odorants was markedly higher when the odorants were not presented as TOMs. Furthermore, performance remained high for discrimination between TOMs when the second odor stimulus was replaced with clean air of the same duration (Δt = 300 ms, Figure 3A right box).

A possible concern is that participants may have used the length of the first odor as a cue to facilitate discrimination in the control experiment, in which the second stimulus was replaced with air. To eliminate this eventuality, we conducted two additional control experiments. In these controls, we presented two identical pulses of the same odor constituent as the second TOM (e.g., AB vs. (AB/AA) and AB vs. (AB/BB). In these experiments, the two TOMs had the exact same odor durations and content. Furthermore, to verify that participants did not perform well on the AB vs. (AB/AC) and AB vs. (AB/BC) control experiments by exploiting possible changes in airflow (in these experiment air was coming from a third odor port), we conducted another control experiment in which we tested whether participants could discriminate between the TOMs: AB vs. (AB/C(B)A). In this experiment, we filled the air canister (the C channel) with odorant B (denoted as C(B)) so that the TOMs were the same as those in the main experiment, but they came from different ports, as in the control experiment. We found that participants could easily discriminate between AB vs. (AB/AA) and between AB vs. (AB/BB) but did not perform above chance in discriminating between AB vs. (AB/C(B)A) (Figure 4—figure supplement 1A–C). Notably, in these control experiments, we used a matched-pair experimental design in which we tested the same participants in both the main experimental paradigm and the control experiment (counterbalanced for order between subjects). This further shows that the same cohort succeeded in discriminating in the control experiment but performed poorly when discriminating between the TOMs used in the main paradigm (when merged, the results were as follows: for the main paradigm, median success rate = 0.81, mean success rate = 0.82 ± 0.095; for the control experiment, median success rate = 0.55, mean success rate = 0.547 ± 0.11; paired t-test: t(12)=8.68, p=1.617E–6). Notably, one participant who performed well in both the control (100% success rate) and main experiments (83% success rate) later reported realizing that the TOMs were composed of two odors coming one after the other.

Finally, we contrasted performance in TOM discrimination across different Δt values and controls by entering the data from all experiments into an omnibus ANOVA. There was a strong main effect for experiment type (F(7,66) = 23.13, partial eta-squared = 0.710, p<0.0001). We next carried out planned post-hoc comparisons (corrected with the Tukey HSD). These comparisons revealed that performance in TOM discrimination was significantly lower than that for all control experiments except one (AB vs. AB/BC, Δt = 600 ms, p=0.24). Critically, this persisted across Δt values (mean success rate: TOM Δt 75 ms = 0.514 ± 0.114, TOM Δt 150 ms = 0.558 ± 0.116, TOM Δt 300 ms = 0.588 ± 0.13, TOM Δt 600 ms = 0.666 ± 0.158, A vs. A/B = 0.978 ± 0.052, B vs. A/B = 0.948 ± 0.091, AB vs. AB/AC = 0.883 ± 0.092, AB vs. AB/BC = 0.802 ± 0.115.; all other p values of TOM compared with controls < 0.0042). To recapitulate, our results so far portrayed a picture in which participants did not perceive TOMs to be different enough for easy discrimination.

Extensive training does not substantially improve discrimination between TOMs

Previous experiments in rodent models reported that the exploitation of temporal features typically involved extensive training of the animals, sometimes for hundreds of trials, while maintaining high motivation (for example, through water deprivation). Given this experimental regimen, it is worth inquiring whether after extensive training, or when motivation is high enough, more olfactory information is extracted from odor-elicited temporal features that contributes to task performance. To probe this possibility, we replicated the same experimental paradigm in a new cohort of five participants, but this time, we asked them to take part in a total number of up to 200 trials (two sessions per day, 40 trials each, over three days). We used the same two highly dissimilar odors as before (CTL/DMTS) and, as before, there was a fixed delay of Δt = 300 ms.

Bearing in mind the obvious differences in reward circuitry between rodents and undergrad students, we attempted to increase participants' motivation by offering a monetary bonus (e.g., doubling the session payment) upon hitting a performance benchmark of clearly succeeding in discriminating between the two TOMs; that is, reaching and maintaining a success rate of 0.7 during a 40-trial session. Participants who reached this goal were fully rewarded and did not have to return for additional sessions. Mean success rates per day ranged from 0.51 to 0.64 and were not significantly different from chance. Critically, they did not improve as a function of session (two-sided sign test p>0.05 in 4 out of 5 sessions; note the marginal significance of cohort performance in the fourth session, p=0.0625, N = 4) (Figure 3B–C). This result suggests that the percepts evoked by the two TOMs were still mostly indistinguishable, even after extensive training.

The odor of TOMs is perceived as an intermediate of its components

Our behavioral results indicate that TOMs seem to be perceived as similar, even when the delay between the constituent odors is long (e.g., Δt = 300 or 600 ms) making it hard to discriminate between TOM AB and TOM BA. To assess odor perception elicited by the TOMs directly, we asked a new group of participants (N = 21) to rate the TOMs and their constituent odors on a list of eleven verbal odor descriptors that are commonly used in olfactory psychophysics (Dravnieks, 1982) (see 'Materials and methods'). TOMs were delivered using the same methodology as before, at a time delay of Δt = 300 ms (N = 12) or Δt = 600 ms (N = 9), with participants able to undergo more than a single smelling round before rating each TOM. To eliminate possible habituation, the time delay between smelling the two TOMs was set to at least 23 s.

We next projected the odor ratings for the two TOMs and their constituting odors onto a two-dimensional space using principal components analysis (PCA). We observed that the two TOMs were perceived as an intermediate percept, positioned in space between the constituting two odor perceptions (Figure 4A). Calculating the Euclidean distance between the average descriptor ratings of the four odor stimuli (e.g., A, B, AB, and BA for Δt = 300 and 600 ms) confirmed that while the two constituting odors were markedly different, the perceptual distance between the two TOMs was much smaller (Δt = 300 ms: distance (AU) A vs. B = 14.24 ± 4.09, AB vs. BA = 10.11 ± 3.48; two sample t-test, t(286) = −9.21; Cohen’s d = 1.08, p=7.35E–18; Figure 4B). An analysis of distance hierarchy between all stimuli confirmed the existence of a significant main effect for odor type (F(5,858) = 23.25, p<0.0001). Post-hoc comparisons confirmed that the perceptual distance between the mono-molecules was larger than for all other stimuli (all p<0.0001, Tukey HSD corrected) (Figure 4C). In other words, whenever a similarity rating included a TOM, it was not significantly different from other similarity ratings that included a TOM, and the perceptual distance between the mono-molecular constituents stood out in all of these comparisons. In addition, the perceptual distance between the two TOMs was smaller than all other compared distances except one (all p<0.05, Tukey HSD corrected, with the exception of AB vs. BA and A vs. AB, p=0.61). When the TOMs were delivered at a delay of 600 ms, a similar picture emerged in which the perceptual distance between the two TOMs remained distinctly smaller than the distance between their mono-molecular constituents (Figure 4D) (for Δt = 600 ms: Distance (AU) A vs. B = 18.86 ± 4.33, AB vs. BA = 10.98 ± 3.39; two sample t-test, t(160) = −7.98; Cohen’s d = 1.25, p=2.61E–13; Figure 4E). An analysis of distance hierarchy between all stimuli confirmed the existence of a significant main effect for odor type (F(5,480) = 18.31, p<0.0001). Post-hoc comparisons confirmed that even with a 600 ms interval between components, the perceptual distance between the mono-molecules was larger than that for all other stimuli (all p<0.0001, Tukey HSD corrected) (Figure 4F).

Figure 4 with 1 supplement see all
Perception of TOMs is an intermediate of that of their components.

(A) Projection of verbal descriptor ratings of the four odors onto the two main axes of principal component space for CTL (labeled 'A', red circles), DMTS (labeled 'B', blue circles) and two TOMs comprised of CTL and DMTS presented in two temporal sequences: AΔt=300B (brown) and BΔt=300A (purple). Each data point represents the ratings of a single participant for a given odor or TOM. The centroids of each cluster are marked by labeled circles following the same color scheme. (B) A matrix comprised of the Euclidean distances between eleven descriptor ratings provided by the participants for CTL (labeled ‘A’), DMTS (labeled ‘B’) and two TOMs comprised of CTL and DMTS. Hotter colors denote larger distances. Ratings are divided into four odorant subgroups by a black grid. Mosaic-like patterns within each compartment represent the between-participant variability of ratings for the same odor. (C) TOMs are perceived more similarly to each other than to their isolated constituents. Bar graph of mean perceptual distance between TOMs (AΔtB or BΔtA ) and their constituent odorants (A and B). Statistical significance is denoted for post-hoc comparisons of perceptual distance between the monomolecular constituents ('A vs. B') with all other stimuli in black and for the TOMs with all other stimuli in red. Error bars are mean ± SEM. (D–F) Same as panels (A–C) for Δt = 600 ms.

Odor-elicited temporal dynamics can be used to extract odor-related information

We next examined whether participants could use explicit information regarding the temporal structure of the stimuli to facilitate discrimination. We therefore informed the participants prior to the session that the two TOMs were composed of two odors delivered one after the other in a different order. We examined two variations: in the experiment composed of the similar, more nameable odors, participants were told that the TOMs were composed of cinnamon (CIN) and orange (ORG). In the experiment presenting dissimilar odors (CTL and DMTS), participants were told that the one odor was pleasant and the other less so.

To test whether awareness of the TOM temporal composition affected performance as a function of Δt, we conducted an ANOVA separately for each odorant pair (i.e., ORG/CIN or CTL/DMTS) with the variable ‘Δt’ (75/150/300/600 ms). For ORG/CIN, we observed a significant main effect for ‘Δt’, implying that the success rate varied considerably as a function of the TOM Δt (F(3,44) = 6.05, partial eta-squared = 0.292, p=0.0015, N = 48, Figure 5A). A parallel analysis conducted on the CTL/DMTS odorant pair indicated a similar significant main effect for ‘Δt’ (F(3,28) = 19.94, partial eta-squared = 0.681, p<0.00001, N = 36, Figure 5B).

Temporal odor dynamics can be used to extract odor-related information.

(A) Success rate of discrimination between the TOMs constituting the ORG and CIN odors as a function of Δt when participants were aware of the constituent odors and their temporal features. Color and legend code as in Figure 2. (B) Same as in panel (A), but for the TOMs composed of CTL and DMTS.

In other words, when participants were explicitly instructed that the incoming odor stimuli were composed of two consecutive odorants, we observed an incremental improvement in both individual participants’ success rates and the group’s median performance. This trend was similar for both odor pairs despite minor variations (in the Δt = 150 ms delay, performance was better in the ORG/CIN odor set than in the CTL/DMTS, yet the Δt = 600 ms exhibited the opposite trend; Figure 5A–B). This ‘unlocking’ of discrimination ability suggests that temporal dynamics can be utilized to extract odor information.

Thus, overall, glomeruli activation times do not elicit a prominent difference in odor percept; however, they can be used to extract temporal odor information such as detecting the order of odor delivery in a sequence.

Discussion

Effects of latency coding on odor perception in humans are inconsequential

Odor-elicited temporal dynamics that are not directly related to stimulus dynamics have long been observed in the olfactory system. These observations gave rise to the hypothesis that odors are encoded by a spatiotemporal code. What a temporal code means in the context of the olfactory system has been interpreted in several ways; one prominent model suggests that the activated glomerulus time relative to some internal or external event is part of the odor code. In the current study, we tested whether the time of glomeruli activation affected odor perception as predicted by this model. We used odor stimuli composed of two odors sequentially presented in different orders (TOMs) and observed that most of the participants had a poor ability to discriminate between TOMs (Figures 2A, 3A and B).

To mitigate any concerns that the rather poor performance in the task was a result of some inadequate experimental condition, we employed a comprehensive battery of control experiments (Figures 2A and 3A gray panels, and Figure 4—figure supplement 1). These experiments showed that participants performed well when the two odors differed not just in terms of their temporal order. Note that the success rates in the control experiments were high, despite the small cohort, compared to the main experiment, suggesting that the number of participants was adequate to allow this conclusion to be drawn. In addition, performance rates improved dramatically when the participants became aware of the constituent odors (Figure 5), further indicating that the failure to discriminate TOMs was not due to technical shortcomings.

We chose to conduct these experiments with human participants to take advantage of a robust and widely accepted supposition that humans can tell what they smell. In fact, an analysis of the percept elicited by each TOM and their constituent odors confirmed that temporal mixtures evoked odor percepts that were in between those of the two constituent odors, regardless of their order of presentation (Figure 4). This type of experiment is significantly more challenging to interpret in animal models, because even when performance is adequate, one cannot attribute this to the different percepts evoked by the TOMs or to an ability of the animals to identify the TOM temporal structure, that is, to identify that two odors have been presented consequently. Importantly, even if under some conditions participants could have discriminated between TOMs, the fact that the TOMs were perceived as similar rather than as generating a new odor percept is strong evidence that temporal coding (of the type examined here) is not a substantial part of the odor code per se.

Temporal coding models

We view these results as more consistent with one possible interpretation of the Laurent model — that time is not necessarily part of the code — and less so with the Hopfield model that time is part of the code. Moreover, the results do not align with a temporal coding model in which the first activated glomeruli form the odor code (Schaefer and Margrie, 2007; Wilson et al., 2017). We argue that if that were the case, participants should have easily discriminated between the TOMs, as the first-activated glomeruli in the TOM AΔtB were most probably very different from those activated in the TOM BΔtA.

One possible interpretation of the Hopfield model that can be reconciled with our results is that although the time of glomeruli activation is not part of the odor code, the time of glomeruli activation in relation to the internal gamma oscillation or the exact spike times of each neuron are indeed part of the code. The odor manipulations in this study could not have affected spike time relative to the gamma oscillation cycle, so this remains a viable possibility.

Temporal dynamics can be used to exploit odor-related features

When participants became aware of the temporal structure of the TOMs and their constituent odors (i.e., rapid onset of two consecutive odors) before the session, we observed higher performance in discrimination when the Δt was set to 300 and 600 ms, but the participants still performed around chance when this delay was shortened to 75 ms to allow for a more rapid succession (Figure 5). This may suggest that although the latency code only weakly affects the odor percept of the TOM, it could be used to disentangle the constituent odors, provided that the sequential nature of the stimuli is disclosed, and the temporal dynamics do not evolve too rapidly. One possible interpretation of this result is that when a delay was introduced, there was a substantial duration of time in which there was no odor at all. This pause between the two odors might have been used to detect the existence of two odors and might therefore have contributed to perceiving them as two odors delivered one after the other. Another possible explanation is that participants performed well because they employed a pattern-matching algorithm. A few participants reported that they were actively searching for a specific odor to occur at the beginning or end of the stimulus, suggesting that they employed a matching algorithm for one of the constituents. When the delay was set to be shorter than 150 ms, this matching failed because the first odor was presented in a partial temporal overlap with the second one. This strategy is also in line with a previous study in which participants were able to name which odor out of two known odors was presented first when the delay was 200–400 ms (Laing et al., 1994) or when it was presented in the presence of a background stimulus (Smith et al., 2010). The mechanism governing this is currently unknown, but we speculate that top-down mechanisms that either shut down background activity or improve sensitivity for the target odor are likely to be involved. These results thus suggest that glomeruli activation time can be utilized to extract odor-related information such as when the odor was delivered relative to other odors or relative to respiration phase.

Human versus non-human organisms

Our experiments were conducted in humans who are generally (and some say wrongfully) regarded as a microsmatic species (Rouquier et al., 2000). One can speculate that in mascrosmatic animals who rely more heavily on their sense of smell, such as rodents and insects, temporal coding is part of the odor code. Several studies have recently probed the ability of mice to exploit temporal coding to extract odor information. Owing to the considerable challenge posed by delivering two odors within the short inhalation time of mice (typically <100 ms in actively smelling mice), the authors conducted these experiments using optogenetic stimulation in mice expressing Channelrhodopsin2 in the olfactory sensory neurons (OSN) or the M/T cells. Mice learned to discriminate between two light stimulations of the OSNs, M/Ts or even a single glomerulus that were a few milliseconds apart, either relative to, or irrespective of respiration (Rebello et al., 2014; Smear et al., 2011; Smear et al., 2013). These experiments clearly demonstrated the use of timing in the olfactory system of mice. However, they did not demonstrate that mice perceived the optogenetic stimulations as two different odor percepts as mice cannot report their odor perception directly. One possible interpretation is that the trained mice perceived the two optogenetic stimuli as the same odor presented at different respiration phases. Alternatively, the mice may have perceived the optogenetic stimuli as two different ‘odor’ durations or two odors delivered one after the other. Furthermore, the behavioral results from non-human model systems frequently rely on extensive training (e.g., thousands of trials performed by highly motivated animals), thus making it hard to distinguish between what the olfactory system does and what the olfactory system can do. It is thus possible that although latency coding does not substantially affect odor perception, temporal dynamics can be utilized under extreme conditions. This may be achieved through neurons that are sensitive to differences in M/T-relative activation times (Haddad et al., 2013).

Limitations

Our study is not without limitations. First, for practical reasons, we focused on a limited set of odor mixtures that were always composed of only two odorants. Second, by using TOMs, we could not change the spike timing of individual olfactory neurons. Several studies have shown that a substantial amount of odor information can be extracted from ‘sub-sniff’ neuron spike timings (Bathellier et al., 2008; Bolding and Franks, 2017; Cury and Uchida, 2010; Sirotin et al., 2015). It thus remains a clear possibility that the odor-elicited spike timing of each neuron is a carrier of odor identity. This coding scheme suggests that changing the spike timing of a particular neuron or neurons will result in a different odor percept. This opens up exciting new possibilities, but it remains a daunting technical challenge. Finally, it is still possible that the participants failed to discriminate between TOMs as a result of some technical shortcoming in our odor-delivery system or some other caveat that we were unaware of. This is always a possibility in experimental research, but we find it unlikely as participants discriminated between the constituent odors with ease (Figures 2B and 3), and were able to discriminate between TOMs when they were made aware of their temporal composition (Figure 5) or when we replaced one odor with the presentation of clean air or the same odor again (Figures 2A and 3A and Figure 4—figure supplement 1).

Thus, overall, whether the temporal dynamics observed in odor-elicited neural responses in rodents are part of the odor code remains an ongoing debate. Our experiments in humans indicate that latency coding could be utilized to extract odor-related information without substantially affecting odor perception.

Materials and methods

Participants

This manuscript describes a series of experiments conducted on an overall cohort of 284 participants (191 female, 93 male, age range 18–41 years). All participants were healthy, self-reported to be normosmic and in good health at the time of the experiment. None of the female participants were pregnant at the time of the experiment (self-report). Participants were all university students, recruited via advertisement on campus grounds. Written informed consent and consent to publish was obtained from the participants in accordance with the ethical standards of the Declaration of Helsinki (1964). The experiment was approved by the institutional ethics committee of Bar Ilan University (reference number ISU20140804001). Participants were paid for their participation. To avoid cross-learning, each participant was tested using only one condition with exception of the experiments reported in Figure 4—figure supplement 1.

Experimental design

Request a detailed protocol

The temporal odor mixtures (TOMs) were comprised of fragrant oil mixtures of orange (ORG) and cinnamon (CIN), or of the monomolecular odorants Citral (CTL) (CAS: 5392-40-5, Sigma-Aldrich) and Dimethyl Trisulfide (DMTS) (CAS: 3658-80-8, Sigma Aldrich). In both cases, odors were diluted to equally perceived intensity (CTL: 1:100, ORG: 1:1, DMTS, 1:10000, CIN: 1:1). The compositions of the fragrant oil mixtures were analyzed using gas-chromatography mass-spectrometry (GCMS). Tenax tubes were used for trapping volatiles, and these were subjected to GCMS combined with dynamic headspace sampling. The main constituents of each oil according to their relative contribution to the mixture were as follows. Cinnamon oil: E-cinnamaldehyde (CAS 14371-10-9), linalool (CAS 78-70-6), o-cymene (CAS 527-84-4), terpineol (CAS 98-55-5) and cinnamyl acetate (CAS 103-54-8). Orange oil: limonene (CAS 138-86-3), gardenol (CAS 93-92-5) and linalool (CAS 78-70-9).

We used a computer-controlled air-dilution custom-built olfactometer to deliver odors in varying orders and sequence lengths. Odor stimuli lasted 200 ms. The inter-stimulus interval (i.e., the time between two TOMs) was set to 8 s. The next trial (i.e., the next pair of TOMs) was presented 23 s after the participant submitted his or her answer (which usually lasted ~2–5 s). Three short tones (1 s) informed the participant about the incoming odor presentation. Participants underwent several training trials to practice the synchronization of a nasal inhalation longer than 1 s, just before the odors were delivered. To simplify the task, participants were instructed that the first TOM was fixed. Participants were allowed three trials to familiarize themselves with the two TOMs and the experimental setup, and then conducted 20 trials. Nasal respiration was recorded using a nasal cannula attached to a pressure-sensitive transducer, which translated these changes in pressure into an electrical signal via a USB interface (Plotkin et al., 2010). The signal was acquired and digitized at a sampling rate of 1 KHz. The recording of respiration and odor triggers and the user interface were designed and carried out in a LabView environment (National Instruments).

Odor delivery

Request a detailed protocol

Odor and air stimuli were delivered via a three-port custom-built odor-delivery system. To prevent cross effects between the two odor ports (e.g., air pressure changes or odor contamination), we used three completely independent channels (Figure 1A). To deliver an odor, purified air (medical grade 99.999%, Maxima, Israel) was streamed through a glass vial containing a liquid odor. Odor selection was done using a digitally controlled solenoid valve (07V113, AIGNEP) on each of the dedicated delivery channels. One-way check valves were used to prevent backflow of the odorized air stream. The air flow was set to 1.8 LPM using a mass flow controller (Alicat Scientific). Odors and the odor-delivery plastic tubing were replaced every day.

To prevent contamination, participants were seated in front of three dedicated odor ports, one for each of the odorants (the distance from the odor ports was ~5 cm, Figure 1A). We used a relatively short delay (8 s) between the first and the second TOM because pretesting showed that longer delays between the two TOM presentations diminished the success rates, even when the participants were asked to discriminate between two very different odors. We verified that this short ISI was adequate by testing participants’ ability to discriminate between the two pairs of odors (Figure 2). In the first set of experiments (ORG and CIN), the first TOM was set to be AB and the second TOM was either AB or BA. In part of the second set of experiments (CTL and DMTS), we changed the first TOM to BA. We modified this design to make sure there that was no bias related to the identity of the first TOM. As the results were the same, we pooled the data.

To ensure that odor presentation matched the inhalation phase of the respiratory cycle, we instructed participants to inhale for a duration of at least 1 s just before the odor was presented (using an auditory cue prior to each odor arrival). Trials in which the odor presentation extended into the exhalation period or started prematurely before inhalation were excluded from the analysis (Figure 1F, methods). The delay between the two odors, Δt, was set to ≤600 ms. We did not test longer delays because the overall duration of a TOM had to be shorter than a typical inhalation period (about 1–1.5 s) to be perceived in full and to minimize the possibility of discriminating between the two TOMs by realizing that they were composed of two consecutive odors. In fact, at long delays (e.g., 300 and 600 ms) a few participants realized that the TOMs were composed of two consecutive odors. We did not exclude these participants from our analysis to prevent bias.

PID measurements

Request a detailed protocol

To estimate odor concentration, we used a mini photo-ionizer detector (miniPID, Aurora Scientific). PID response amplitude depends on the odor identity, its concentration and the distance from the inlet. Each odor elicits a different response that is related to how effectively the measured odor can be ionized. We measured the door PID responses to the odors and their temporal dynamics at different delays. All TOMs elicited similar response dynamics. PID responses varied between these delays and the order of odor presentation, reflecting the between-odor interactions and inter-pipe flow interactions. However, the odor concentration of the odor constituents and the TOMs in three tested delays was highly similar across trials of the same condition. Thus, each TOM was expected to elicit the same odor percept across all trials. Measurements of odor concentration when placing the PID inlet ~5 cm from the odor ports (i.e., the location of the participants’ nose) revealed that the order of odor presentation was preserved, although the odor concentration was more variable across trials when measured at a 5 cm distance (Figure 1—figure supplement 1C).

Post-processing and analysis

Request a detailed protocol

The respiration signal was zeroed and baselined such that positive values denoted inhalation and negative values denoted exhalation. A 0.05–100 Hz band pass filter was applied to the respiratory trace to remove high-frequency artifacts and drift. Next, following normalization by a z-score, the entire recorded trace underwent segmentation into epochs, each consisting of a single trial.

To be sure that the odor pulses arrived in synchrony with the inhalation period, we meticulously verified all odor pulses occurring within the inhalation period. We scored the proportional duration of the odor pulses within each TOM, or in other words, how much of the odor pulse was presented concurrently with nasal inhalation. An automatic algorithm assigned a score to each trial, according to the combined score of all odor triggers in a given trial. We applied a strict criterion that set the cut-off to 1.00 (all odor pulses were presented fully within inhalations) to reject any trials that were questionable. Finally, this automated process was backed by a trial-by-trial visual inspection. The results remained the same when we used a less stringent cut-off (e.g., 0.9) or when we assumed that odor offset was longer than the estimate obtained from the PID measurements (e.g., assuming a very long odor offset of 200 ms).

In the main experimental paradigm, each session consisted of 20 trials. Any participant with a total number of fewer than 14 valid trials (70% of the session) was excluded from further analysis. This led to the exclusion of 73 participants following analysis of their trial-by-trial data, leaving a total of 284 participants. The number of eliminated trials within the pool of remaining participants who took part in the TOM paradigms totaled 162 out of 1940 and 223 out of 3972 trials, or 8.3% and 5.6% of all events in the CIN/ORG and CTL/DMTS experiments, respectively. The results were virtually the same when we used a less stringent criterion for excluding trials.

As detailed above, two TOMs were presented in each trial. Each was comprised of two rapid odor pulses spaced apart by a pre-defined delay. Odor stimuli were presented subsequent to an auditory cue instructing the participants to inhale. The stimuli were not, however, triggered to lock with a certain phase of nasal inhalation, and as a result, were encountered at different respiration phases. Respiratory phase was calculated with MATLAB’s ‘angle’ function applied to the Hilbert transform of the respiratory trace. The product of this calculation is the phase that gradually increases from –π/2 to π/2 over the course of the inhalatory phase of the respiratory cycle (see Figure 2D for visualization).

Statistical analysis

Request a detailed protocol

Given the two-alternative nature of the task, the outcome of each trial could either be ‘correct’ or ‘incorrect’, and the success rate at chance level was 0.50. A success rate significantly higher than chance in a 20-trial experiment was therefore calculated to be 0.70 (binomial cumulative distribution function). In cases where the number of valid trials was lower than 20, this threshold was adjusted accordingly, such that it significantly exceeded chance at a significance level of p<0.05. Similarly, the median group performance was compared to chance (0.50) using a two-sided sign-test. A comparison of group performance across conditions was carried out using an analysis of variance (ANOVA). To reduce the effect of outliers on the result of this study, we focused our analysis on the group medians. However, the results remained the same even when we used the group mean (i.e., performed a one-sample two-sided t-test).

In addition to standard testing of the data against a null hypothesis, we also subjected the data in each analysis to Bayesian one-sample t-tests with success rate as the dependent variable, compared to chance performance (0.50) with a Cauchy prior of 0.707 (Good, 1962) The added insight gained from this approach stems from its ability to quantify the evidence in favor of two different models. Bayesian statistics are advantageous in assessing the relative probability of the null hypothesis over the experimental hypothesis. This advantage becomes a necessity when one does not reject H0 (i.e., ‘non-significant results’) and needs to quantify the evidence to support this claim (Leech and Morgan, 2002). We therefore detailed our Bayesian statistics alongside each regular sign-test. The output Bayesian statistic used was the BF10, which depicts an odds ratio; namely, the probability, or simply how likely the data are under both hypotheses. In our interpretation, we used the standard recommendation that a BF10 between 1 and 3 implies anecdotal evidence, 3–10 substantial, and 10–30 strong evidence, where BF10 quantifies evidence for the alternative hypothesis relative to the null hypothesis. All the Bayesian statistical analyses were conducted in JASP (2019) version 0.9.2. Statistical analyses concerning the values of the respiratory phase were carried out using functions implemented in CircStat MATLAB, a toolbox for circular statistics that are analogous to the regular t-test or ANOVA (Berens, 2009).

Odor similarity analysis

Request a detailed protocol

To estimate olfactory perceptual distance between all odors, we asked a separate cohort (N = 12 for ΔT = 300 ms and N = 9 for ΔT = 600 ms) to rate all four odors used in this study (ORG, CIN, CTL and DMTS) on the basis of eleven verbal descriptors curated from a list of commonly used descriptors (Dravnieks, 1985). To calculate odor similarity, we then projected participants’ ratings into a two-dimensional space using principal component analysis (PCA), a dimensionality reduction method common in olfactory research (Haddad et al., 2010; Khan et al., 2007). As expected, ORG and CIN odors encompassed an overlapping area in this perceptual space, whereas CTL and DMTS were markedly divergent, with CTL more similar to ORG/CIN (Figure 2—figure supplement 1A). We quantified similarity by calculating the mean Euclidean distance between the ratings for each odor. The perceptual distance between CTL and DMTS was significantly higher than the distance between ORG and CIN (CTL/DMTS = 5.52 ± 1.12; CIN/ORG = 4.18 ± 1.25, paired t test: t(110) = 5.95, p=3.1E–8, Cohen’s d = 1.12, Figure 2—figure supplement 1C). Furthermore, this perceptual distance was supported by a predictive algorithm allowing for the estimation of perceptual similarity from molecular structure (Snitz et al., 2013). The distance between ORG and CIN was 0.0189 radians, but the distance between CIT and DMTS was 1.0846 radians. In other words, discriminating between TOMs composed of citral and DMTS was expected to be an order of magnitude easier than discriminating between ORG and CIN. Last, as with the previous TOMs, we verified that these two odors had similar intensities (CTL = 7.37 ± 1.19, DMTS = 7.12 ± 0.99, paired t-test, t(7) = −0.84, Cohen’s d = 0.23, p=0.43, N = 8).

TOM phase analysis

Request a detailed protocol

We computed the respiration phases of the two TOMs for correct and incorrect trials. The respiration phases for the first and second TOMs were correlated, both for correct and incorrect trials (correct: circular correlation r = 0.67, p=1.33E-8, N = 100; incorrect: r = 0.549, p=1.02E-5, N = 84). The regression line slopes were close to one (correct: a = 0.942, incorrect: a = 0.859), indicating that participants tended to encounter the two TOMs at similar respiration phases (Figure 2—figure supplement 1A–B). Moreover, comparing the differences between the TOMs’ respiration phases (e.g., TOM1 – TOM2) in correct and incorrect trials further showed that the phase differences in the correct and incorrect trials were not significantly different (non-parametric multi-sample test for equal medians, Kruskal-Wallis test for circular data: shared population median = −0.478 rad, KW(P)=1.402, p=0.236; Figure 2—figure supplement 1C–D). Finally, an adaptation of the Kolmogorov-Smirnoff test for circular data (Kuiper test) conducted iteratively 10,000 times on the phase data suggested an average p-value of 0.985, with an average k statistic of 480.15 ± 126.4; in other words, the phase distributions were highly similar.

Olfactory perception of TOMs

Request a detailed protocol

Odorant and TOMs were rated using a set of eleven verbal descriptors (pleasant, fruity, edible, hot, chemical, medicinal, smoky, alcoholic, attractive, earthy, and sulfurous) on a scale of ranging from 0 to 10 (where 0 corresponded to ‘not at all’ and 10 corresponded to ‘very much’). It should be noted that in rating sessions involving TOMs, these were always rated before the monomolecular odors in order to prevent bias in the perception of the mixtures, given that their isolated components had not yet been presented separately. Participants could undergo several smelling rounds of the same odor before rating, and an inter-stimulus interval of 23 s was imposed between any two odor presentations. Data for two descriptors of a single participant were corrupted and were replaced by the group average for that score. Prior to projection into the principal component space, descriptor ratings were normalized using the Z-score.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
    Atlas of Odor Character Profiles
    1. A Dravnieks
    (1985)
    PA: ASTM Press.
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
    Geophysical Journal International
    1. IJ Good
    (1962)
    555–558, Theory of Probability Harold Jeffreys (Third edition, 447 + ix pp., Oxford Univ. Press, 84s.), Geophysical Journal International, 6, Oxford Press, 10.1093/gji.6.4.555.
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53

Decision letter

  1. Upinder Singh Bhalla
    Reviewing Editor; Tata Institute of Fundamental Research, India
  2. Catherine Dulac
    Senior Editor; Harvard University, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

This paper studies whether humans use time-differences in odor arrival as part of the precept of odor discrimination. The study makes the surprising case that odor arrival time does not contribute strongly to odor identification, and this finding runs counter to considerable prior work albeit mostly in rodents. Because this claim is surprising, and the readouts are behavioural and lack access to physiology in human subjects, the report has been supported with a comprehensive set of controls. These reveal the suggestive nuance that when humans are primed to look for sequential odor delivery, their performance on the exact same task improves. It will be interesting to see how this set of findings is reconciled with the rodent literature.

Decision letter after peer review:

Thank you for submitting your article "The contribution of temporal coding to odor coding and odor perception in humans" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Catherine Dulac as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

The reviewers all agreed that the results were surprising and potentially interesting. They brought up a number of technical and conceptual points that they felt could be cleared up by some rather straightforward experiments of the same design.

1) The authors should perform calibration experiments on humans to clear the concern that the control and Temporal Odor Mixture (TOM) discrimination experiments were different. These differences are both by way of total odor and in the way the delivery apparatus was designed.

2) The authors should extend their longest odor interval from 300 ms to around 600 ms. This would address the concern that their longest tested interval, expressed as a fraction of respiration cycle duration, is actually rather shorter than most experiments in rodents where temporal discrimination has been reported.

In addition, the reviewers have made several other suggestions that the authors may wish to address to improve the manuscript.

Reviewer #1:

In this behavioural study Perl et al. ask whether humans can discriminate between odour pairs delivered in different temporal orders. They find that humans can discriminate if they know that two odour will be given in sequence, but not otherwise. This study adds to the substantial literature on the importance of timing in odour perception.

The authors take some pains to deliver odour precisely with a fixed duration of 200 ms and varying intervals between odour. They also explore a variety of relevant control stimuli in Figure 2. Overall the experiments seem carefully done.

The intriguing finding of this study is that odour order discrimination is possible at 300 ms interval, but only if the subjects know that the stimuli are sequential. The authors argue that if timing were an important component of odour coding then the percepts for different odour should be quite different.

1) I wonder if the authors have narrowed their interval proportionately to the discrimination time in model organisms. The authors begin to see discriminability at 300ms, of a respiration respiration cycle time of about 6 seconds (e.g., Figure 1F). Here a 300 ms interval is 1/20 of the cycle. In various animal studies the separation of stimuli is from 13 ms (Rebello et al) to 50 ms (Cury and Uchida and others). The mouse/rat sniff cycle can be well under 100 ms, though note that the Cury and Uchida study points out timing the initial respiration phase is quite similar even for slower respiration. Thus the phase separation is from 1/8 to 1/2 of a cycle. Thus in this rather rough calculation, one might expect that if human discrimination scales similarly to their respiration rates, we should expect discrimination to start to improve at about 600 ms, which is twice as long as what the authors try. Thus one explanation for their findings may simply be that they used too short an interval.

2) I feel that the study is a little uni-dimensional: here is a human perceptual observation, without mechanistic underpinnings, and only using two pairs of odour. While it does have interesting interpretations in terms of coding as suggested by the authors, it would have been good to have these interpretations backed by further observations within the constraints of human experimentation.

Reviewer #2:

In this manuscript, Perl et al. aim to assess the contribution of glomerular activation timing to the problem of odor identification. To this end, they use temporal odor mixtures (TOMs: A->B vs. B->A, each presented within one sniff) to presumably activate the same set of glomeruli in two sniffs but in different temporal sequences. They ask human participants to report whether the odor percepts across the two sniffs are the 'same' or 'different'.

Perl et al. find that human participants largely fail to discriminate between TOMs (chance performance) while showing significantly higher discrimination performance on simpler discrimination tasks (A vs. B, A->B vs. A or A->B vs. B). However, TOM discrimination does improve when participants are pre-informed about the composition of TOMs as sequences of two different odors. The authors propose that latency differences in timing of glomerular activation may not be automatically perceived as differences in odor identity. However, information pertaining to latency differences can be extracted to enhance odorant discriminability when required.

The degree to which glomerular timing contributes to odor identification is a longstanding question. As the authors themselves note, substantial body of work in non-human systems has shown that small differences in glomerular activation times elicit changes in both neuronal and behavioral responses. However, behavioral results from non-human model systems frequently rely on extensive training, thus making it hard to distinguish between what the olfactory system does versus what the olfactory system can do?

In this context, the results of Perl et al. are extremely relevant: they provide the clearer view of what features of the glomerular activity patterns olfactory system natively tunes to in order to identify odors. Overall, the study is well executed and the results are presented clearly.

However, my enthusiasm to support publication at this stage is dampened by several concerns regarding the interpretation of their results. These should be addressable by one additional set of behavioral experiments (point 1).

Concerns:

The authors' primary conclusion (latency differences in glomerular timing do not contribute to odor identification) rests on one key observation: human subjects show poorer discrimination performance for TOM discrimination versus control conditions.

While the authors do perform a battery of controls, the stimulus properties in the control conditions differ significantly from those in the TOM discrimination experiments. The total amount of odor delivered per sniff and spatial distribution of airflow across the final tubes (see points 1 and 2) differ in the TOM discrimination and in the controls. This makes the results hard to interpret: the deterioration in discrimination performance for TOMs may result from factors other than just differences in glomerular activation times. It is essential to rule out these alternative factors with additional control experiments (see point 1) before publication.

1) The authors' odor discrimination paradigm requires subjects to compare percepts across two sniffs (~8 s apart). In the TOM experiments, the subjects experience odors for a total of 400 ms (200ms of A + 200ms of B) in each sniff. However, in control conditions the subjects experience either overall less amount of odor (odor A vs. B; 200ms in each sniff) or different amounts of odor across the two sniffs (task: A->B vs. A/B->air; 400 ms in the first sniff and 200 ms in the second sniff).

It is unclear how discrimination performance changes with total odor content per sniff. Do the subjects perform poorer in the TOM task simply because of the higher odor content in both the sniffs being compared? Likewise, can the subjects simply rely on the difference in total odor content between the two sniffs to assess stimulus similarity in the second set of controls (A->B vs. A/B->air)?

In my opinion, a more interpretable control is to test whether subjects can discriminate A->B vs. A->A or A->B vs. B->B better than the TOM condition. This would allow the authors to maintain the same odor content across each sniff across both experimental and control conditions.

2) Flow related cues might allow subjects to achieve a higher performance in control conditions compared to the TOM task. The authors use a non-typical final port design for odor delivery, where each stimulus (A, B and air) is delivered through a separate tube. From the Materials and methods section, it is not clear whether the airflow is ON through all three tubes at all times. When odor A is OFF, is the tube dedicated to odor A delivering clean air instead, such as to maintain the same net air flow?

If not, another axis along which subjects can discriminate across stimuli is by comparing spatial distribution of airflow patterns across tubes. The TOM discrimination experiment is the only experiment where the airflow is ON through same set of tubes (tubes A and B) across the two sniffs. For all the control conditions, airflow switches from one set of tubes in the first sniff to another set of tubes in the second sniff. For the first set of controls (A vs. B), the flow switches from tube A to tube B. For the second set of controls (A->B vs. A->air), the flow switches from tube A to the air tube. Given these limitations, it is impossible to rule out that poorer discrimination in the TOM task (compare to controls) simply results from the smaller differences in flow cues across the two sniffs.

3) Discrimination performance varies across control conditions (Figure 2A and 3A) in ways that are not obvious. Naively, it appears, that discriminating A vs. B should be easiest. While this is true for CTL-DMTS odor pair, it does not seem to be the case for the CIN-ORG pair. There are asymmetries in performance across controls that are expected to show similar outcomes. Why is A->B vs. A->air harder to discriminate than A->B vs. B->air?

Since the deterioration in discrimination performance is the metric that the authors base their claims on, the authors should comment on the variations in this metric and the factors that underlie these variations. Are these differences across control conditions significant, especially given that the control conditions consistently have fewer subjects than the TOM experiments? If significant, the authors should comment on possible explanations of these differences in a manner that is consistent with the results observed in the TOM discrimination task.

Reviewer #3:

In this paper, Perl and colleagues attempt to analyze if glomeruli activation times affect odor perception by performing olfactory discrimination experiments with human subjects. Authors designed temporal odor mixtures (TOMs) composed of two components and challenged the subjects to discriminate between the stimuli where the sequences of presentation of these components were different. Results show that the subjects were unable to discriminate these TOMs when they were not informed about the sequential presentation of components. This is an important topic that needs to be discussed in the context of temporal coding in olfaction.

As the authors mention, temporal coding in olfaction has been discussed extensively and given many interpretations. Earlier works have provided strong experimental evidences for relative time-based code in rodent olfactory system (Haddad et al., 2013; Smear et al., 2013). Here authors rely on the behavioral readouts to study how the time-based code contributes to olfactory information processing. While I agree that this is an important topic, lack of clarity with the interpretations and the flaws with the experimental design question the aptness of this article for a publication in eLife in the present format. The study needs to be revised.

Here are my major concerns:

1) This study tried to address how temporal coding affects odor perception. To probe this, authors designed the TOMs, made of odors A and B that varied in sequence of presentations. When the sequence of presentation changes from A→B to B→A, the relative time of glomeruli activation changes. I agree with authors' claim of differences in the relative time of glomerular activation for odors A and B separately ("Delivering two odors at different order activates the glomeruli at a different relative time and form a different latency code"). But, are authors neglecting the information processed by odor pulse B, when it is presented first in the sequence in a discrimination context? Is this the optimal stimuli combination for the question they are addressing?

2) In the section of "temporal dynamics can be used to exploit odor related features", authors discuss that glomeruli activation time can be utilized to extract odor-related information such as when the odor was delivered relative to other odors or relative to respiration phase. This can be tested by challenging the subjects with a discrimination task using the same target odor pulse at different timings in the background of another odor. For me, this is more relevant as we have to detect and discriminate specific odors in olfactory enriched environments.

3) What do authors try to address by reducing the time delay between the onset of pulses to 75 ms and 150 ms for CTL/DMTS discrimination? This would allow the mixing of two odors and the subjects could perceive this as a single entity (?).

4) I would like to see a more detailed discussion and comparison between this study and other studies (Haddad et al., 2013, Rebello et al., 2014 and Smear et al., 2013) to draw more robust conclusions about the neural mechanisms involved.

Other concerns:

1) Technical issues: authors have used Orange and Cinnamon extracts. These extracts are mixtures of different monomolecular odors that vary in their physico-chemical properties. Does this affect authors' conclusion?

2) Authors have cited Friedrich and Laurent, 2001 and Gschwend et al., 2015 for the decorrelation model. While I respect authors' freedom to interpret the results differently, these two studies are reporting totally different time scales for the decorrelation (slow vs. fast).

3) Interpretations given for Rebello et al., 2014 and Smear et al., 2013 are misleading.

4) In the last experiment, when participants were informed about the nature of the task, authors observed incremental improvement in the performance. Authors say that it helped in 'unlocking' the discriminability. What is the neural mechanism underlying this "unlocking"?

5) Few p values reported are the same (Subsection “The effect of odor temporal dynamics on odor perception” paragraph three and subsection “The effect of rapid odor temporal dynamics and odor similarity on odor perception” paragraph two).

https://doi.org/10.7554/eLife.49734.sa1

Author response

The reviewers all agreed that the results were surprising and potentially interesting. They brought up a number of technical and conceptual points that they felt could be cleared up by some rather straightforward experiments of the same design.

1) The authors should perform calibration experiments on humans to clear the concern that the control and Temporal Odor Mixture (TOM) discrimination experiments were different. These differences are both by way of total odor and in the way the delivery apparatus was designed.

2) The authors should extend their longest odor interval from 300 ms to around 600 ms. This would address the concern that their longest tested interval, expressed as a fraction of respiration cycle duration, is actually rather shorter than most experiments in rodents where temporal discrimination has been reported.

In addition, the reviewers have made several other suggestions that the authors may wish to address to improve the manuscript.

We thank all reviewers and the reviewing editor for their helpful and constructive comments.

All reviewers agree that this finding is interesting as it sheds new light on the contribution of latency coding in odor perception in humans. Two technical concerns were raised (as summarized above by the reviewing editor). We have addressed these concerns by conducting the suggested three experiments and conducting new analyses. We have recruited overall additional 59 participants for these three new experiments (reported in Figures 2A, 3A, 4D-F and Figure 4—figure supplement 1).

To summarize, we ran two additional control experiments, specifically designed to mitigate the reviewers’ concerns, namely to address the possibility that the control and the TOMs are different in terms of odor duration and airflows (These are now detailed in Figure 4—figure supplement 1).

We have also expanded the range of our testing by introducing a new, longer, delay of 600 ms between odor constituents within the TOMs, asking whether participants can discriminate between TOMs in terms of accuracy and perception. Notably, this effort is reflected horizontally in essentially all main experiments in the manuscript ms (Discrimination – Figures 2A, 3A, and 5; Perception – Figure 4D-F).

The results of all these three experiments further strengthen the study and we thank the reviewers for the opportunity to clarify these possible concerns.

Finally, we have implemented your comments regarding minor issues of clarity throughout the manuscript and have further elaborated where was required.

Reviewer #1:

In this behavioural study Perl et al. ask whether humans can discriminate between odour pairs delivered in different temporal orders. They find that humans can discriminate if they know that two odour will be given in sequence, but not otherwise. This study adds to the substantial literature on the importance of timing in odour perception.

The authors take some pains to deliver odour precisely with a fixed duration of 200 ms and varying intervals between odour. They also explore a variety of relevant control stimuli in Figure 2. Overall the experiments seem carefully done.

The intriguing finding of this study is that odour order discrimination is possible at 300 ms interval, but only if the subjects know that the stimuli are sequential. The authors argue that if timing were an important component of odour coding then the percepts for different odour should be quite different.

1) I wonder if the authors have narrowed their interval proportionately to the discrimination time in model organisms. The authors begin to see discriminability at 300ms, of a respiration respiration cycle time of about 6 seconds (e.g., Figure 1F). Here a 300 ms interval is 1/20 of the cycle. In various animal studies the separation of stimuli is from 13 ms (Rebello et al) to 50 ms (Cury and Uchida and others). The mouse/rat sniff cycle can be well under 100 ms, though note that the Cury and Uchida study points out timing the initial respiration phase is quite similar even for slower respiration. Thus the phase separation is from 1/8 to 1/2 of a cycle. Thus in this rather rough calculation, one might expect that if human discrimination scales similarly to their respiration rates, we should expect discrimination to start to improve at about 600 ms, which is twice as long as what the authors try. Thus one explanation for their findings may simply be that they used too short an interval.

We thank the reviewer for this interesting suggestion. The reviewer is suggesting that temporal coding might be proportional to the sniff length. That is, in mice temporal coding is in the range of 10-50 ms (in Smear et al. they report a temporal resolution of 10ms), while in model organisms possessing longer sniffs this could scale up to be hundreds of milliseconds. We think this is quite an interesting claim. We are not aware of any study suggesting that the scale of temporal coding is relative to the sniff length. Notably, in insects, there is no respiratory cycle per se onto which sniffs can map. Furthermore, if this suggestion is correct then it has deep implications on the general notion of how neurons work: Downstream neurons in humans should possess a very long integration window so that they could be sensitive only to very long temporal differences but not short ones. That said, we extended our study to include a delay of 600 ms and the results are very similar to those observed with 300 ms (now reported in Figure 2A and 3A and Figure 4D-E and Figure 5).

Originally, we did not test a 600 ms delay since it was technically challenging for two reasons: First, with a 600 ms delay, the overall duration of each TOM is 800 ms (200 ms+400 ms+200 ms) which is sometimes at the limit of the inhalation length (~1-1.5 seconds). This resulted in many trials that had to be excluded since our predefined inclusion criterion demanded that the entire odor stimulus has to occur within the inhalatory phase. Second, and more importantly, when the delay is as long as 600 ms it often happens that participants perceived the stimulus as two odors arriving one after the other. This may result in a scenario where we are not testing if participants perceive the TOM as a new odor but rather if they can tell the order of the odor sequences (which is similar to the second set of experiments we conducted, in which we notified the participants in advance each trial is constructed of two odors arriving one after the other).

Nevertheless, to overrule this possible concern, we run another set of experiments in which we tested TOM discrimination in the presence of a 600 ms delay. As expected, we had to exclude quite a few participants in which the TOM wasn’t fully presented during the inhalation period. The results are presented in Figures 2A and 3A. The vast majority of the participants failed to significantly discriminate between the TOMs in an accuracy surpassing chance level. (Figures 2A and 3A in the 600 ms condition). The success rates of the group in the 600 ms delay is similar to the one in 300 ms. In both experiments the success rates are still far lower than the rates obtained in the control experiments or when the participants are aware of the nature of the TOMs (reported in Figures 5A and 5B). As expected, few participants reported realizing the TOM are composed of two consecutive odors. We did not exclude these participants although their success rates were above chance level.

To further examine the effect of a 600 ms delay on odor perception, we asked 9 new participants to rate the TOMs with a 600 ms delay and their components using the same verbal descriptors we used in Figure 4A. These results are reported in text and are now added to Figure 4D-F. This experiment clearly shows that even with a 600 ms delay humans perceive the TOMs as similar and not as two distinct odors as is expected from a temporal code per-se.

Finally, we examined performance in the presence of a 600 ms delay when the participants were aware of the temporal dynamics within the stimulus and observed that in this case they were able to easily discriminate between the TOMs (and even better than all shorter delays previously tested). These results are now reported in text and added to Figures 5A and 5B.

Taken together, these experiments demonstrate that latency coding as defined here does not contribute substantially to odor percept even when the delay is as long as 600 ms.

2) I feel that the study is a little uni-dimensional: here is a human perceptual observation, without mechanistic underpinnings, and only using two pairs of odour. While it does have interesting interpretations in terms of coding as suggested by the authors, it would have been good to have these interpretations backed by further observations within the constraints of human experimentation.

We do understand the reviewer’s concern. However, this is already quite a large study compared to other human psychophysics experiments. Overall, we report results from a cohort of 284 participants. This is a within the upper bounds of cohorts reported in a single study in the field of olfactory psychophysics and perception. Furthermore, in the main paradigms presented here each subject’s performance was derived from 20 trials in most cases (and up to 200 trials per subject in the paradigm which unfolded over days, see Figures 3B-C). This culminates into overall approximately 6000 trials which were analyzed in this study. This is somewhat comparable with high-throughput automated studies conducted in rodents (1).

We think using two odor pairs (two familiar and two that are highly dissimilar) is reasonable to convey the main point. The argument is that if temporal coding is as indispensable to perception then any two odor pairs which activate the glomeruli at different orders should elicit a different percept. In the discussion we note that it is possible that under some other circumstances (different odors, timing etc.) latency coding could contribute more to perception. Furthermore, unlike studies in rodents, we took advantage of the ability of humans to report what they smell to directly estimate how TOMs are perceived. We think this approach is an important contribution to the study of temporal coding which gives an additional and unique insight.

Finally, we do provide an interesting observation that once participants are told that the TOMs are composed of two consecutive odors, they do seem to dramatically improve in discriminating between them. We did examine how participants were able to succeed in this experiment, and it turns out they were applying an odor matching procedure of either the first odor or the second. This is similar to the known phenomena that people are much better when they actively look for something they know (e.g. a specific pen on a cluttered table) than when they need to find an unknown pen. We wish we could probe the underlying neural mechanism further as it is very interesting (which also is now on our minds as we plan future projects using this setup) but unfortunately it is not something that we could address in humans with a respectable depth.

Reviewer #2:

In this manuscript, Perl et al. aim to assess the contribution of glomerular activation timing to the problem of odor identification. To this end, they use temporal odor mixtures (TOMs: A->B vs. B->A, each presented within one sniff) to presumably activate the same set of glomeruli in two sniffs but in different temporal sequences. They ask human participants to report whether the odor percepts across the two sniffs are the 'same' or 'different'.

Perl et al. find that human participants largely fail to discriminate between TOMs (chance performance) while showing significantly higher discrimination performance on simpler discrimination tasks (A vs. B, A->B vs. A or A->B vs. B). However, TOM discrimination does improve when participants are pre-informed about the composition of TOMs as sequences of two different odors. The authors propose that latency differences in timing of glomerular activation may not be automatically perceived as differences in odor identity. However, information pertaining to latency differences can be extracted to enhance odorant discriminability when required.

The degree to which glomerular timing contributes to odor identification is a longstanding question. As the authors themselves note, substantial body of work in non-human systems has shown that small differences in glomerular activation times elicit changes in both neuronal and behavioral responses. However, behavioral results from non-human model systems frequently rely on extensive training, thus making it hard to distinguish between what the olfactory system does versus what the olfactory system can do?

We are thankful to the reviewer and their comprehensive comments. The last sentence is an insightful and well phrased way to describe the finding. We added the following sentence to the Discussion:

“Furthermore, behavioral results from non-human model systems frequently rely on extensive training (e.g. thousands of trials of highly motivated animals), thus making it hard to distinguish between what the olfactory system does versus what the olfactory system can do. It is thus possible that although latency coding does not substantially affect odor perception temporal dynamics can be utilized under extreme conditions.”

In this context, the results of Perl et al. are extremely relevant: they provide the clearer view of what features of the glomerular activity patterns olfactory system natively tunes to in order to identify odors. Overall, the study is well executed and the results are presented clearly.

However, my enthusiasm to support publication at this stage is dampened by several concerns regarding the interpretation of their results. These should be addressable by one additional set of behavioral experiments (point 1).

Concerns:

The authors' primary conclusion (latency differences in glomerular timing do not contribute to odor identification) rests on one key observation: human subjects show poorer discrimination performance for TOM discrimination versus control conditions.

While the authors do perform a battery of controls, the stimulus properties in the control conditions differ significantly from those in the TOM discrimination experiments. The total amount of odor delivered per sniff and spatial distribution of airflow across the final tubes (see points 1 and 2) differ in the TOM discrimination and in the controls. This makes the results hard to interpret: the deterioration in discrimination performance for TOMs may result from factors other than just differences in glomerular activation times. It is essential to rule out these alternative factors with additional control experiments (see point 1) before publication.

1) The authors' odor discrimination paradigm requires subjects to compare percepts across two sniffs (~8 s apart). In the TOM experiments, the subjects experience odors for a total of 400 ms (200ms of A + 200ms of B) in each sniff. However, in control conditions the subjects experience either overall less amount of odor (odor A vs. B; 200ms in each sniff) or different amounts of odor across the two sniffs (task: A->B vs. A/B->air; 400 ms in the first sniff and 200 ms in the second sniff).

It is unclear how discrimination performance changes with total odor content per sniff. Do the subjects perform poorer in the TOM task simply because of the higher odor content in both the sniffs being compared? Likewise, can the subjects simply rely on the difference in total odor content between the two sniffs to assess stimulus similarity in the second set of controls (A->B vs. A/B->air)?

In my opinion, a more interpretable control is to test whether subjects can discriminate A->B vs. A->A or A->B vs. B->B better than the TOM condition. This would allow the authors to maintain the same odor content across each sniff across both experimental and control conditions.

We thank the reviewer for raising these concerns. This is a valid point which also reflects a comprehensive thinking effort on the reviewer’s part and we are thankful for this observation. With your advice in mind we carried out the exact controls you suggested (A->B vs. A->A and A->B Vs B->B). Performance in these controls, centered around 80% accuracy, indicating that participants can easily discriminate these two set of TOMs (now presented in Figure 4—figure supplement 1). Furthermore, to assert this finding is not due to serendipitous recruiting of high-performing participants, we verified that these same participants cannot discriminate between A->B Vs. B->A, thus forming a matched pair test which is statistically stronger and further strengthen the results. This contrasts between high performance in control and a 50-60% accuracy in main paradigm conveys a clear message – even when equating total odor content and duration, participants cannot discriminate between the TOMs tested here.

2) Flow related cues might allow subjects to achieve a higher performance in control conditions compared to the TOM task. The authors use a non-typical final port design for odor delivery, where each stimulus (A, B and air) is delivered through a separate tube. From the Materials and methods section, it is not clear whether the airflow is ON through all three tubes at all times. When odor A is OFF, is the tube dedicated to odor A delivering clean air instead, such as to maintain the same net air flow?

If not, another axis along which subjects can discriminate across stimuli is by comparing spatial distribution of airflow patterns across tubes. The TOM discrimination experiment is the only experiment where the airflow is ON through same set of tubes (tubes A and B) across the two sniffs. For all the control conditions, airflow switches from one set of tubes in the first sniff to another set of tubes in the second sniff. For the first set of controls (A vs. B), the flow switches from tube A to tube B. For the second set of controls (A->B vs. A->air), the flow switches from tube A to the air tube. Given these limitations, it is impossible to rule out that poorer discrimination in the TOM task (compare to controls) simply results from the smaller differences in flow cues across the two sniffs.

The reviewer raises the possible concern that perhaps participants could more easily discriminate between A→B and A→C because in this experiment, different tubes are involved in each TOM and therefore participants could have exploited some changes in airflow (or spatial cues).

First, we think it is unlikely that these differences in success rates can be explained solely by the changes in airflow. The setup is built such that all odor ports and air ports are exactly the same (same bottle types, same connectors, same connecting tubes and valves etc.) so the flow should be the same across ports and was verified to be the same using a flow meter (Alicat Scientific Flow meter).

That said, in order to address this concern directly, we ran an additional control experiment in which we repeated the A→B Vs B→A experiment, but this time we used the ports previously designated as C for the B odor when we delivered the B→A TOM (we call it C(B)→A). C(B) delivers odor B through the C stream and therefore this experiment resembles the A→B Vs. A→C experiment in terms of changes in airflow while still delivering the same odors: A→B and B→A. If the airflow (or any other difference in the C port) provides information to help participants distinguish between AB and AC then it should also allow distinguishing between AB and B(C)A. We found that all participants but one failed to discriminate between A→B and C(B)→A indicating that changes in airflow don’t provide sufficient cues to help in discriminating between the two TOMs (Figure 4—figure supplement 1). Interestingly, the one participant who did succeed reported perceiving that the TOM is composed of two odors arriving one after the other which, for this individual, renders this experiment similar to the ones we report in Figure 5 in which participants easily discriminated between the TOMs when we have notified them in advance that the TOMs are composed of two consecutive odors.

3) Discrimination performance varies across control conditions (Figure 2A and 3A) in ways that are not obvious. Naively, it appears, that discriminating A vs. B should be easiest. While this is true for CTL-DMTS odor pair, it does not seem to be the case for the CIN-ORG pair. There are asymmetries in performance across controls that are expected to show similar outcomes. Why is A->B vs. A->air harder to discriminate than A->B vs. B->air?

Since the deterioration in discrimination performance is the metric that the authors base their claims on, the authors should comment on the variations in this metric and the factors that underlie these variations. Are these differences across control conditions significant, especially given that the control conditions consistently have fewer subjects than the TOM experiments? If significant, the authors should comment on possible explanations of these differences in a manner that is consistent with the results observed in the TOM discrimination task.

We thank the reviewer for this important comment and we have now explicitly addressed it in the manuscript. We think the reason for the higher success rate in discriminating between the CTL-DMTS is because these two odors are far more dissimilar than the CIN-ORG group. We now added an analysis in which we directly assess the perceptual distances between these odor pairs. CIN and ORG are two pleasant smells and as such it is not as easy to discriminate as DMTS (an unpleasant odor reminiscent of rotten egg) and CTL (a generally perceived as pleasant odor reminiscent of an orange).

It is possible that AB/AC is slightly harder to discriminate than AB/BC in the CIN-ORG experiment, because maybe A, as the first odor in the sequence, is slightly more dominant at least for some of the participants causing AB to be more similar to AC but not to BC. However, it is important to note that the success rates of all control experiments are not significantly different as stated: “…, all p’s of cross-control comparisons > 0.68)”, and therefore, it is possible that these differences are just statistical fluctuations caused by the selection of some less competent participants and because the group size may not be large enough to mitigate such fluctuations. Importantly, this difference does not occur in the CTL-DMTS pair.

The difference in success rates between AB/AC compared to A/B is expected as AB/AC share an odor component and therefore by definition are more similar.

Reviewer #3:

In this paper, Perl and colleagues attempt to analyze if glomeruli activation times affect odor perception by performing olfactory discrimination experiments with human subjects. Authors designed temporal odor mixtures (TOMs) composed of two components and challenged the subjects to discriminate between the stimuli where the sequences of presentation of these components were different. Results show that the subjects were unable to discriminate these TOMs when they were not informed about the sequential presentation of components. This is an important topic that needs to be discussed in the context of temporal coding in olfaction.

As the authors mention, temporal coding in olfaction has been discussed extensively and given many interpretations. Earlier works have provided strong experimental evidences for relative time-based code in rodent olfactory system (Haddad et al., 2013; Smear et al., 2013). Here authors rely on the behavioral readouts to study how the time-based code contributes to olfactory information processing. While I agree that this is an important topic, lack of clarity with the interpretations and the flaws with the experimental design question the aptness of this article for a publication in eLife in the present format. The study needs to be revised.

Here are my major concerns:

1) This study tried to address how temporal coding affects odor perception. To probe this, authors designed the TOMs, made of odors A and B that varied in sequence of presentations. When the sequence of presentation changes from A→B to B→A, the relative time of glomeruli activation changes. I agree with authors' claim of differences in the relative time of glomerular activation for odors A and B separately ("Delivering two odors at different order activates the glomeruli at a different relative time and form a different latency code"). But, are authors neglecting the information processed by odor pulse B, when it is presented first in the sequence in a discrimination context? Is this the optimal stimuli combination for the question they are addressing?

We thank the reviewer for this comment which reflects a need for clarification on our side. Our initial finding with the odors of orange and cinnamon was indeed tested only with the TOMs of AB Vs. AB / BA. Following coming across the main finding of this study (namely the poor discrimination ability) and the application of adequate controls (Figure 2) we have next tested both A->B and B->A as the first stimuli within the TOM in ensuing experiments (see Figure 3 and its legend). Thus, both options are covered. We hope this addresses the reviewer concern.

2) In the section of "temporal dynamics can be used to exploit odor related features", authors discuss that glomeruli activation time can be utilized to extract odor-related information such as when the odor was delivered relative to other odors or relative to respiration phase. This can be tested by challenging the subjects with a discrimination task using the same target odor pulse at different timings in the background of another odor. For me, this is more relevant as we have to detect and discriminate specific odors in olfactory enriched environments.

We thank the reviewer for this comment. The concept of presenting an odor pulse in the background is indeed an elegant way to probe discrimination through target-background stimulus extraction. This exact method was employed before to study adaptation in human olfaction by David Smith (see Smith et al., 2010).

We found this design to be very different from the one used in our study and therefore we did not refer to it, let alone attempted to replicate its methodology.

We now acknowledge that some may find this study to be of relevance in the context of interpretation of our results and therefore we added the following text to the Discussion:

“Few participants reported they were actively searching for a specific odor to occur at the beginning or end of the stimulus, suggesting they employ a matching algorithm for one of the constituents. When the delay was set to be shorter than 150 ms, this matching failed because the first odor was presented in partial temporal overlap with the second one. This strategy is also in line with a previous study in which participants were able to name which odor out of two known odors was presented first when the delay was 200-400 ms (Laing et al., 1994) or when it is presented in the presence of a background stimulus (Smith et al., 2010).”

3) What do authors try to address by reducing the time delay between the onset of pulses to 75 ms and 150 ms for CTL/DMTS discrimination? This would allow the mixing of two odors and the subjects could perceive this as a single entity (?).

We thank the reviewer for this question. We did not want to base all our finding on a single parameter for delay between TOMs. We therefore explored other delays as well. Initially we did not want to go above 300 ms for reasons detailed in the manuscript as well as within this response letter. Previous experiments found that mice can discriminate optogenetic stimulations when the delay is in the order of few tens of milliseconds. We therefore wanted to examine if temporal coding might be more effective in shorter delays. Furthermore, reducing the delay offered some insights as to a (somewhat crude) performance curve which offered a richer understanding of the phenomenon we set out to study.

Please see our reply to your first comment in the “Other concerns” section just below, regarding the question of a potential “mixing of two odors into a single entity”.

Finally, as also suggested by the first reviewer, we have now extended the study to include long delays of 600 ms. The results further support the main claim of this study.

4) I would like to see a more detailed discussion and comparison between this study and other studies (Haddad et al., 2013, Rebello et al., 2014 and Smear et al., 2013) to draw more robust conclusions about the neural mechanisms involved.

We thank the reviewer for this suggestion. We have now completely revised the first paragraph in the Discussion regarding possible neural mechanisms and comparisons with these studies.

"Odor-elicited temporal dynamics not directly related to stimulus dynamics have been long observed in the olfactory system. These observations gave rise to the hypothesis that odors are encoded by a spatiotemporal code. What does a temporal code mean in the context of the olfactory system has been given several interpretations; one prominent model suggests that the time of activated glomeruli relative to some internal or external event is part of the odor code. In the current study, we set out to test whether the time of glomeruli activation affects odor perception as expected from this model. We used odor stimuli composed of two odors sequentially presented at different orders (TOMs) and observed that the greater majority of participants could poorly discriminate between TOMs (Figures 2A, 3A and 3B)."

And:

“Due to the considerable challenge posed by delivering two odors within the short inhalation time of mice (typically <100 ms in actively smelling mice), the authors conducted these experiments using optogenetic stimulations in mice expressing Channelrhodopsin2 in the olfactory sensory neurons (OSN) or the M/T cells. Mice learned to discriminate between two light stimulations of the OSNs, M/Ts or even a single glomerulus that are few milliseconds apart, either relative to-, or irrespective of respiration (Rebello et al., 2014; Smear et al., 2011, 2013). These experiments clearly demonstrated the use of timing in the olfactory system of mice. However, they did not demonstrate that mice perceived the optogenetic stimulations as two different odor percepts as mice cannot report their odor perception directly. One possible interpretation is that the trained mice perceived the two optogenetic stimuli as the same odor presented at different respiration phases. Alternatively, mice may have perceived the optogenetic stimuli as two different “odor” durations or two odors arriving one after the other. Furthermore, behavioral results from non-human model systems frequently rely on extensive training (e.g., thousands of trials of highly motivated animals), thus making it hard to distinguish between what the olfactory system does versus what the olfactory system can do. It is thus possible that although latency coding does not substantially affect odor perception, temporal dynamics can be utilized under extreme conditions. This may be achieved with the help of neurons that are sensitive to differences in M/T relative activation times (Haddad et al., 2013).”

Other concerns:

1) Technical issues: authors have used Orange and Cinnamon extracts. These extracts are mixtures of different monomolecular odors that vary in their physico-chemical properties. Does this affect authors' conclusion?

We thank the reviewer for this comment which reflects a comprehensive grasp of our methodology. The way humans perceive (and probably non-human as well) olfactory stimuli (often referred to as “olfactory objects”) is far from clear. It is largely held that people, especially non-professional, naïve participants, perceive odors synthetically. That is, they cannot tell if an odor is a mono-molecule or a mixture of odorants. See a host of studies by D. G. Laing from the late 90’s which provided mounting evidence to back up this claim as well as a recent review by Benjamin D. Young (Mind and Language, 2019) who summarized those to claim that “The inability to perceptually identify the constituents within a complex smell is best explained in light of the aforementioned evidence that the nature of sensory and cortical encoding of olfactory stimuli does not always encode complex odors as the concatenation of their constituents”.

We therefore believe our approach presenting TOMs of mixtures of Orange and Cinnamon smells is valid.

Furthermore, we verified that both mixtures are perceived as equally intense and outsourced them to a gas-chromatography mass-spectrometry analysis for a breakdown of their components which is listed in the Materials and methods section of the manuscript under “experimental design”.

Finally, as noted in the manuscript, we had several reasons to conduct the next experiments with a newly introduced odor pair of CTL and DMTS, both of which are mono-molecules. One of the reasons was to move into “cleaner” perception. As we detail in the manuscript, both types of odor pairs resulted in poor discrimination ability between temporal sequences of odors.

2) Authors have cited Friedrich and Laurent, 2001 and Gschwend et al., 2015 for the decorrelation model. While I respect authors' freedom to interpret the results differently, these two studies are reporting totally different time scales for the decorrelation (slow vs. fast).

We concur. We removed the Gschwend et al. citation from this section.

3) Interpretations given for Rebello et al., 2014 and Smear et al., 2013 are misleading.

We absolutely agree with the reviewer. We removed this paragraph.

4) In the last experiment, when participants were informed about the nature of the task, authors observed incremental improvement in the performance. Authors say that it helped in 'unlocking' the discriminability. What is the neural mechanism underlying this "unlocking"?

We thank the reviewer for this question. Unlocking is a very fitting word to describe a sudden improvement in perceptual abilities, yet it conveys very little as to the underlying Mechanism. In the discussion we speculated it might be related to pattern matching mechanisms similar to the way one can easily find a pen on a cluttered table when he/she knows how it looks like and it takes much more time to find it when you are just looking for a general pen. To reveal any neural mechanism underlying this phenomenon would necessitate recording of neurons of olfactory circuits which is not feasible with this human cohort and is out of the scope of this study. We do agree this is a highly intriguing question which worth pursuing with the right tools. For human experimentation, probing modulation of attention to odor patterns was studied before using fMRI (see studies by Zelano et al., 2011 Neuron and Plailly et al. 2008, JNS) however to the best of our knowledge, never with TOMs.

We added to the discussion the following sentences:

“This may suggest that although latency code only weakly affects the odor percept of the TOM, it could be used to untangle the constituting odors, provided that the sequential nature of the stimuli was revealed and the temporal dynamics do not evolve too rapidly. One possible interpretation of this result is that when a delay was introduced, there is a substantial duration of time in which there is no odor at all. This break between the two odors might be used to detect the existence of two odors and therefore to aid in perceiving them as two odors arriving one after the other. Another possible explanation is that participants performed well because they employed a pattern matching algorithm. Few participants reported they were actively searching for a specific odor to occur at the beginning or end of the stimulus, suggesting they employ a matching algorithm for one of the constituents. When the delay was set to be shorter than 150 ms, this matching failed because the first odor was presented in partial temporal overlap with the second one.”

5) Few p values reported are the same (Subsection “The effect of odor temporal dynamics on odor perception” paragraph three and subsection “The effect of rapid odor temporal dynamics and odor similarity on odor perception” paragraph two).

This is common in a ranked signed test as it compares ranks and the specific value of each data point does not affect the computation (sign-test considers their ranks and not their nominal values and therefore the same p value will be obtained if the number of points and their ranks is the same).

https://doi.org/10.7554/eLife.49734.sa2

Article and author information

Author details

  1. Ofer Perl

    The Gonda Multidisciplinary Brain Research Center, Bar-Ilan University, Ramat-Gan, Israel
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Investigation, Visualization, Methodology
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-3560-4344
  2. Nahum Nahum

    Department of Psychology, Bar-Ilan University, Ramat-Gan, Israel
    Contribution
    Conceptualization, Data curation, Software, Formal analysis
    Competing interests
    No competing interests declared
  3. Katya Belelovsky

    The Gonda Multidisciplinary Brain Research Center, Bar-Ilan University, Ramat-Gan, Israel
    Contribution
    Data curation, Project administration
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9960-7336
  4. Rafi Haddad

    The Gonda Multidisciplinary Brain Research Center, Bar-Ilan University, Ramat-Gan, Israel
    Contribution
    Conceptualization, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology
    For correspondence
    rafihaddad@gmail.com
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8285-5210

Funding

Israel Science Foundation (204/17)

  • Rafi Haddad

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We would like to extend our thanks to Shani Agron for help with GCMS analyses. We are indebted to Noam Sobel for helpful insights provided during the preparation of this manuscript. This study was supported by the ISF- 204/17 grant.

Ethics

Human subjects: Participants were all university students, recruited via an advertisement on campus grounds. Written informed consent and consent to publish were obtained from participants in accordance with the ethical standards of the Declaration of Helsinki (1964). The experiment was approved by the institutional ethics committee of Bar Ilan University (reference number: ISU20140804001). All experimental sessions were conducted after obtaining informed consent and the participants were paid for their participation.

Senior Editor

  1. Catherine Dulac, Harvard University, United States

Reviewing Editor

  1. Upinder Singh Bhalla, Tata Institute of Fundamental Research, India

Publication history

  1. Received: June 27, 2019
  2. Accepted: January 15, 2020
  3. Version of Record published: February 7, 2020 (version 1)

Copyright

© 2020, Perl et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 672
    Page views
  • 80
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)