Adaptation invariant concentration discrimination in an insect olfactory system
- Department of Biomedical Engineering, Washington University in St. Louis, United States
- Department of Electrical and Systems Engineering, Washington University in St. Louis, United States
Figures
Figure 1 with 4 supplements
Behavioral response facilitation upon odor repetition.
(a) Schematic of the innate palp-opening response (POR) experimental paradigm. Opening of the locust’s maxillary palps within 15 s of the odor onset was considered a POR. (b) Schematic of the block odor stimulation protocol. Each block consisted of ten trials, and a 4 s odor pulse was presented in each trial. A 15 min no-stimulation period separated the blocks. (c) Response matrices are shown summarizing individual locust PORs (rows) for blocks of ten trials (columns). White boxes indicate PORs in a specific trial, while black boxes indicate the absence of PORs in that trial. Locusts were sorted such that the least responsive locusts are shown at the top and the most responsive ones are near the bottom. PORs varied between locusts across trials. The left matrix depicts responses of 36 locusts to isoamyl acetate. The right matrix shows the responses of 64 locusts to benzaldehyde. (d) The probability of PORs across locusts (p(POR)) is shown as a function of trial number for two odorants: iaa and bza at high intensities (1% v/v). (e) p(POR)s as a function of trial number is shown for four odorants (oct, hex, iaa, bza) at two different intensities (0.1% v/v – low and 1% v/v – high). The p(POR) of both odorant intensities increased over trials. Further, note that the p(POR) values for high-intensity odor exposures were notably higher than p(PORs) for low-intensity odor presentations. For this analysis, we computed the p(POR) for each locust at low and high concentrations of an odorant. This resulted in 36 paired values for isoamyl acetate, 33 paired values for hexanol, 47 paired values for octanol, and 64 paired values for benzaldehyde. We used a left-tailed t-test to check whether the p(POR) across all ten trials of lower intensity odor exposures was significantly smaller than the p(POR) during higher intensity presentations of the same odorant. (f) Bar plots compare the average P(POR) for the low- and high-intensity presentations of the same odor. Error bars indicate the standard error of the mean. p(POR) values observed at high and low intensities of an odorant were significantly different for all four odorants tested (p<0.05). n=36 locusts for iaa, n=33 locusts for hex, n=47 locusts for oct, and n=64 for bza.
Figure 1—figure supplement 1
POR responses of locusts across trials and odor intensities.
(Left panel) POR responses of each locust (rows) across trials (columns) are shown for both low and high intensity exposures for all four odorants used in the study. For each odorant, locusts are shown such that each row indicates the response of the same locust to low and high-intensity odor presentations. Black boxes indicate no POR responses, and white boxes indicate POR responses in that trial. (Right panel) The probability of palp-opening responses across locusts is shown as a function of trial number. Replotted from Figure 1.
Figure 1—figure supplement 2
POR responses of locusts to hexanol vs. paraffin oil control.
The probability of palp opening responses (averaged across locusts; 10 trials per locust) between an odorant (hexanol) and the solvent used for diluting odorants (paraffin oil) is shown. Only odorants evoked significant palp-opening responses (p<0.05; one-sided t-test; n=10 locusts). Error bars indicate the standard error of the mean.
Figure 1—figure supplement 3
Palp-opening responses in early vs. late trials.
The average POR in the first five (left panels) and last five trials (right) is compared between high and low-intensity odor exposures (p≤0.05, one-tailed t-test). The difference in p(POR) was significant for all four odorants during the later trials. Error bars indicate the standard error of the mean.
Figure 1—video 1
A representative video showing locust palp-opening response.
Figure 2
Ensemble neural activity systematically changes across repeated encounters with an odorant.
(a) Schematic of the olfactory stimulation protocol. Each block consisted of 25 trials with a 4-s odor pulse delivered in each trial. The inter-trial interval was 60 s. Two datasets were collected. Each dataset consisted of five randomized blocks of four odorants (dataset 1: hex (H), hex (L), oct (H), oct (L), hex (H)-repeat; dataset 2: iaa (H), iaa (L), bza (H), bza (L), iaa (H)-repeat). A 15-min no-odor stimulation period separated blocks of trials. (b) Raster plots of eight representative projection neurons (PNs) in the locust antennal lobe are shown. Spiking activities are shown for 25 trials (rows) with earlier trials shown at the top and later trials at the bottom. The shaded region indicates the 4-s odor stimulation period, and the identity of the stimulus is indicated in each plot. (c) Raster plots are shown for four representative PNs during two blocks of trials. The same odorant was presented in both blocks. Note that spiking activity changes are repeatable across the block of trials. (d) Schematic showing how vesicular depletion and lateral inhibition facilitation models would change spiking activity in individual neurons. The y-axis represents the change in response over trials (25th trial response – 1st trial response). Positive numbers indicate that the last trial had a stronger response, hence the response increase. Negative numbers indicate that the first trial had a stronger response, hence response reduction. Along the x-axis, the response in the first trial is shown. Vesicular depletion should impact the strongly activated neurons more, whereas lateral inhibition facilitation should progressively suppress weak responders. (e) Following the schematic in panel d, response changes observed in PNs are shown. All PNs were included in this plot. Different colors and symbols are used to denote the identity of the odorant and its intensity (low-intensity trials – triangles and high-intensity trials – circles). As can be noted, the R2 values are ~0.48–0.56 indicating that the vesicle depletion facilitation model captures the adaptation trends in our datasets to a certain extent. (f) Changes in PN spiking activity over trials during high-intensity odor exposures (y-axis) are plotted against response changes observed for the same PN during low-intensity exposures of the same odorant. The poor correlation values (hex, R2=0.17; oct, R2=0.12; iaa R2=0.21; bza R2=0.08) indicate that reductions in neural response amplitude for one odor intensity do not model reductions in response amplitude for another odor intensity.
Figure 3
Stimulus repetition and intensity decrements reduce spiking responses.
(a) Peristimulus time histograms (PSTHs) across all PNs (hex and oct, n=80 PNs; iaa and bza, n=81 PNs) for each odorant. PSTHs of trials 1 and 25 are shown. (b) PSTHs across all PNs are compared between high and low-intensity odor exposures. Response during the first trial is shown. (c) Summed spike counts (across PNs and four-second odor presentation) are calculated and shown as a function of the trial number. The dotted line indicates the spike count of the 25th trial of the high-intensity odor exposure.
Figure 4
Odor identity and intensity information are maintained across trials.
(a) A schematic showing how the ensemble neural activity might change between two odorants and across multiple trials or repetitions. The combination of neural activation should differ between different odorants, and therefore, the odor identity should be represented by population response vectors that differ in their direction. Repetitions should reduce response strength without altering the combination of neurons activated. If this were the case, the later repetitions of the same odorant would evoke a response that can be represented by vectors that maintain directions while progressively becoming shorter in length or magnitude. (b) A schematic coding scheme for achieving adaptation-invariant intensity coding. The combination of neurons activated changes markedly with odor identity and subtly with odor intensity. These responses become less intense without altering the combination of neurons activated. If this were the case, then the ensemble vector direction would change with both odor identity and intensity, and the vector direction would be robustly maintained even though the vector length continues to change with repetition. (c) Trial-by-trial odor-evoked ensemble PN response trajectories after dimensionality reduction are shown for hex and oct. Only high-concentration exposures of both odorants are included in this plot. For each odor, the temporal response trajectories for the 1st, 5th, 10th, 15th, 20th, and 25th trials are shown (color gradient from light to dark as trial number increases). Note that the odor-response trajectories become increasingly smaller during later trials, but the direction of the trajectories is robustly maintained. (d) Trial-by-trial response trajectories for low-concentration exposures of the hex and oct are shown. Similar convention as in panel c. (e) Similar plot as in panel c, but now comparing the odor-evoked response trajectories elicited by hex at high and low intensities. (f) Similar plot as in panel d, but comparing the response trajectories elicited by oct at high and low intensities.
Figure 5 with 2 supplements
High and low stimulus intensities activate distinct ensembles.
(a) Correlations between neural responses observed in different trials are shown. Each pixel/matrix element represents a similarity between mean neural responses in one trial versus those in another trial. Diagonal blocks reveal the correlation between trials when the same odorant at a specific intensity was repeatedly presented. (b) A dendrogram was generated using a correlation distance metric comparing trial-by-trial ensemble spiking activities evoked by two different stimuli at two different intensities (see Methods). Two major response clusters that correspond to stimulus identity and intensity were identified. The number at the leaf node represents the trial number.
Figure 5—figure supplement 1
Classification analysis confirms robust adaptation-invariant odor recognition using ensemble neural activity patterns.
Results from classification analysis are shown for the two datasets: hexanol–octanol at different concentrations (dataset 1; 80 PNs), and isoamyl acetate and benzaldehyde (dataset 2; 81 PNs). A one-nearest neighbor approach with correlation distance was used for predicting odor labels. We did a leave-one-trial-out validation. The true odor label is shown along the x-axis, and the predicted odor label is shown along the y-axis. As can be noted, the class labels for every single trial were correctly predicted in both datasets. The result after class labels were shuffled is also shown for comparison. These results strongly support our conclusion that odor intensity information is preserved, and odor concentration can be recognized independent of adaptation.
Figure 5—figure supplement 2
Combinatorial response patterns across two intensities of the same odorant.
Comparison of odor-evoked responses between high and low intensity exposures of all four odorants used in the study. Spiking activities of individual PNs were summed over the entire odor presentation window, sorted based on their response to the high-intensity presentations, and plotted to reveal the combinatorial response. The mean ± S.D. across the 25 trials is shown.
Figure 6
Neural adaptation inversely correlates with behavioral facilitation.
(a) The probability of odor-evoked POR (P(POR)) for a given trial is plotted against the total spike counts elicited during the trial or repetition number. Therefore, each point represents a single trial. Symbols with light and dark colors are used for differential trials of high and low-concentration odor exposures. The line represents the regression fit between the behavioral and neural responses (hex (L), R2=0.65; hex (H), R2=0.79). The negative slope of the regression line indicates that while the neural responses diminish over trials, the behavioral responses increase over trials. (b–d) Similar plots as in panel a but for oct (L/H), iaa (L/H), and bza (L/H).
Figure 7
Adaptation results in odor-specific neural response reductions and behavioral output facilitation.
(a) Two blocks of trials were used. First, a block of 30 trials where one odorant (hex) was presented in all trials except the 26th trial (the catch trial). During the catch trial, a deviant stimulus (iaa or app) was presented. After a 15 min no-odor reset period, a second block of 10 trials of the deviant stimulus was presented (trials 31–40). This was done to determine the unadapted (first trial) and adapted (later trials) responses of the same set of PNs to the stimulus used in the catch trial. (b) Summed spike counts (during 4 s odor presentation period) across all PNs were calculated and shown as a function of trials (hex repetitions). Repeated presentations of odor A resulted in a substantial reduction in odor-evoked spike counts. However, the presentation of the deviant odor (iaa) during the catch trial (trial # 26) resulted in a marked increase in spike counts. The response strength was similar to the non-adapted responses (Trial # 31) for iaa (H). (right) Similar results are shown when hex was repeatedly presented and the app was used as the deviant stimulus. Note that the response to the app during the catch trial did not recover as was noted for iaa (left). (c) Correlations between odor-evoked responses with hex response in the first trial are shown as a function of trial number. (d) A similar catch-trial paradigm was followed for behavioral experiments. Each locust was presented with a repeating odorant for eight trials. A deviant stimulus was presented in the ninth trial. This was followed by six more repetitions of the recurring stimulus. (e) The response matrix summarizing the POR responses of individual locust PORs is shown. Same convention as in Figure 1. (f) The P(POR) across locusts was calculated and plotted as a function of the trial number. Notably, when the deviant stimulus was presented, there was a marked decrease in P(POR). The dotted line indicates the unadapted P(POR) value for the deviant stimulus.
Author response image 1
POR response dynamics in a conditioned locust.
The palps were painted in this case (left panel), and the distance between the palps was tracked as a function of time (right panel).
Author response image 2
Correlation as a function of trial number.
All correlations were made with respect to the odor-evoked responses in the last odor trial of hex(H) and bza(H).
Additional files
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)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Adaptation invariant concentration discrimination in an insect olfactory system
eLife 12:RP89330.
https://doi.org/10.7554/eLife.89330.3
