State-dependent representations of mixtures by the olfactory bulb

  1. Aliya Mari Adefuin
  2. Sander Lindeman
  3. Janine Kristin Reinert
  4. Izumi Fukunaga  Is a corresponding author
  1. Sensory and Behavioural Neuroscience Unit, Okinawa Institute of Science and Technology Graduate University, Japan

Abstract

Sensory systems are often tasked to analyse complex signals from the environment, separating relevant from irrelevant parts. This process of decomposing signals is challenging when a mixture of signals does not equal the sum of its parts, leading to an unpredictable corruption of signal patterns. In olfaction, nonlinear summation is prevalent at various stages of sensory processing. Here, we investigate how the olfactory system deals with binary mixtures of odours under different brain states by two-photon imaging of olfactory bulb (OB) output neurons. Unlike previous studies using anaesthetised animals, we found that mixture summation is more linear in the early phase of evoked responses in awake, head-fixed mice performing an odour detection task, due to dampened responses. Despite smaller and more variable responses, decoding analyses indicated that the data from behaving mice was well discriminable. Curiously, the time course of decoding accuracy did not correlate strictly with the linearity of summation. Further, a comparison with naïve mice indicated that learning to accurately perform the mixture detection task is not accompanied by more linear mixture summation. Finally, using a simulation, we demonstrate that, while saturating sublinearity tends to degrade the discriminability, the extent of the impairment may depend on other factors, including pattern decorrelation. Altogether, our results demonstrate that the mixture representation in the primary olfactory area is state-dependent, but the analytical perception may not strictly correlate with linearity in summation.

Editor's evaluation

This study provides a strong evidence supporting that odour mixture interactions are more linear during awake animals compared to anaesthetised conditions, contrary to the previous notion that odour mixture interactions are sublinear – the conclusion obtained in anaesthetised animals. With their revisions, the authors have done new simulations to clarify how linear mixture interactions affect odour discriminability and interact with other factors (e.g. decorrelations). This new analysis provides a reasonable explanation as to when linearity helps improve discriminability and clarify the significance of sublinear interactions.

https://doi.org/10.7554/eLife.76882.sa0

Introduction

As animals in nature navigate through their environment in order to find food, mates, and to avoid dangers, their sensory systems are tasked to detect and recognise signals of interest despite a background of interfering signals. This figure-ground segregation is a ubiquitous task for many, if not all, sensory systems. In the visual system, for example, segmentation of spatial patterns of light allows animals to recognise objects despite some parts being obscured (Marr, 1982). In the auditory system, spectral combinations of sound waves are recognised and strung together over time to form a stream, allowing animals to recognise social calls from specific individuals among other noises (Bregman, 1990). In olfaction, too, animals face challenges in identifying an odour of interest in the presence of other molecules (Laing and Francis, 1989; Rokni et al., 2014).

Olfactory stimuli are first detected by a large family of olfactory receptors residing in the nasal epithelium (Buck and Axel, 1991). Due to a broad ligand-receptor binding (Araneda et al., 2000; Del Mármol et al., 2021; Malnic et al., 1999), each odour molecule may activate a number of olfactory receptor types, leading to a combinatorial representation (Malnic et al., 1999). As a result, when several compounds are present in a given mixture, they can activate overlapping sets of olfactory receptors, causing complex pharmacological interactions. For example, molecules may bind a common receptor, which, depending on the efficacy, can lead to antagonism (Cruz and Lowe, 2013; Kurahashi et al., 1994; Oka et al., 2004; Reddy et al., 2018; Singh et al., 2019) or enhancement (Xu et al., 2020). Recent large-scale studies demonstrate that this is a widespread phenomenon (Inagaki et al., 2020; Xu et al., 2020; Zak et al., 2020), which means that neural responses to mixtures often do not equal the sums of responses to the individual odours. In addition to the interactions at the periphery, there are many forms of nonlinear summation at multiple stages of olfactory processing, including the saturation of neural responses (Firestein et al., 1993; Wachowiak and Cohen, 2001) and inhibitory interactions within the olfactory bulb (OB) (Economo et al., 2016). Widespread suppressive interactions are also observed in downstream areas, including the piriform cortex (Penker et al., 2020; Stettler and Axel, 2009).

Nonlinear summation of signals in some brain areas may be desirable when specific combinations of signals carry special meanings (Agmon-Snir et al., 1998; Jacob et al., 2008). However, in primary sensory areas, it is sometimes considered information limiting (Laughlin, 1989). For olfaction, this is thought to limit the analytical ability – whether a mixture can be perceived in terms of the constituent qualities (Bell et al., 1987; Jinks and Laing, 1999; Laing and Glemarec, 1992). Nonlinear summations, or ‘interactions’, do not occur for all odour mixtures (Fletcher, 2011; Gupta et al., 2015; Tabor et al., 2004), but occur prevalently when the background and target activation patterns overlap. This happens when a mixture contains many components (Mathis et al., 2016), as well as when the background odours are structurally related to the target odour (Cruz and Lowe, 2013; Fletcher, 2011; Jinks and Laing, 1999; Kay et al., 2003; Mathis et al., 2016; Tabor et al., 2004). Nonlinear summation poses a difficulty because it may distort a pattern of interest brought by non-uniform addition of unpredictable background patterns, although more recent studies suggest beneficial effects of antagonism, for example, by reducing saturation-related loss of information (Reddy et al., 2018). Due to this difficulty, some studies have suggested that the olfactory system may not decompose mixture representations into component parts, but may solve the task by learning task-specific boundaries instead (Mathis et al., 2016; Wilson and Stevenson, 2003).

The question therefore remains: how does the mammalian olfactory system deal with nonlinear summation of responses? To investigate this, we used binary mixtures of odours to investigate mixture representations in mice performing an odour detection task. While temporal structures caused by turbulence may be used to segregate odours of interest from the background (Ackels et al., 2021; Hopfield, 1991), we tackle the case when temporal information from the environment is not available. We demonstrate, by comparing the mixture responses using various paradigms, that the property of mixture summation depends on the brain state.

Results

Olfactory figure-ground segregation is most difficult when the target and background odours evoke overlapping activity patterns (Rokni et al., 2014). To study this task using binary mixtures, we first characterised how our target odour relates to other odours in our panel. The target odour was ethyl butyrate (EB) and the rest of the odours in our set comprised a range of small esters structurally similar to EB, as well as non-esters (Figure 1A). According to Rokni et al., the masking index, which measures the amount of overlap in response patterns between the target odour and background odours, correlates well with behavioural performances (Rokni et al., 2014).

Masking indices of odours used with respect to the ethyl butyrate (EB) pattern.

(A) Odours in the panel with abbreviations used in the rest of the article. EB was the target odour for behavioural experiments. (B) Two-photon imaging of GCaMP6f signals from the apical dendrites of mitral and tufted (M/T) cells in Tbx21Cre::Ai95D mice under ketamine and xylazine anaesthesia. Scale bar = 0.1 mm. (C) Example GCaMP6f transients expressed as a change in fluorescence (ΔF/F). Scale bar = 1 ΔF/F. Grey, odour presentation (0.5 s). (D) Example evoked responses. Manually delineated regions of interest (ROIs) are shown with fluorescence changes evoked by odours, indicated with the corresponding colour map. The amplitude indicated is the average change during 1 s from the final valve opening. (E) Masking indices for all odours in the panel, with EB as the target. N = 4 fields of view, four mice. Mean and SEM of three trials or more shown.

To measure the masking indices from single-odour responses, we obtained activity patterns from the OB output neurons, the mitral and tufted (M/T) cells. Using a two-photon microscope, we imaged from the glomerular layer of the OB in Tbx21Cre::Ai95D mice, which express the calcium indicator GCaMP6f (Dana et al., 2019) in M/T cells (Haddad et al., 2013). To reproduce previous results, these initial imaging experiments took place in mice under anaesthesia (Figure 1B). We studied the degree of overlap between EB and other odour responses by presenting single odours in a randomised order. An analysis of glomerular activity patterns revealed that methyl butyrate (MB) responses overlap the most with the EB response patterns, followed by closely related esters (Figure 1C–E).

Previous studies from anaesthetised animals showed that neural responses to odour mixtures exhibit widespread nonlinear summation (Inagaki et al., 2020; Oka et al.; Reddy et al., 2018; Singh et al., 2019; Xu et al., 2020). In particular, suppressive interactions become dominant among large responses due to saturation (Mathis et al., 2016). This pattern is observed at many stages of olfactory processing, including in the OB (Economo et al., 2016; Fletcher, 2011) and anterior piriform cortex (Penker et al., 2020), but it depends on the complexity of mixtures, as well as odorant choices (Fletcher, 2011; Gupta et al., 2015; Rokni and Murthy, 2014; Tabor et al., 2004). We therefore characterised the property of binary mixture summation using our odour set.

We first confirmed that our olfactometer is capable of presenting stable concentrations of odours for mixtures. We used a photoionisation detector to ensure that, when two odours are mixed, ionisation levels sum linearly (Figure 2A–C). Then, to assess how the OB output represents binary mixtures, we imaged the individual somata of M/T cells in Tbx21Cre::Ai95D mice (Figure 2D). A typical session consisted of about 40 trials to minimise time-dependent effects. Single odours and their binary combinations were presented in a semi-random order. Since EB, MB, and the mixture of these two odours are of particular importance in this study, these three trial types appeared every 10 trials (Figure 2E; see Materials and methods). To assess how the mixture of EB and MB is represented, the amplitudes of the mixture responses were compared against those of linear sums of single-odour responses (Figure 2F and G). As reported before, a large proportion of M/T cells exhibited nonlinear summation (39.0% of M/T cells showed a deviation from linearity greater than 2; 71/182 regions of interest [ROIs], seven fields of view, four mice; see Materials and methods) with a large fraction showing sublinear summation (35.7%, 65/182 ROIs).

Mixture suppression dominates under anaesthesia.

(A–C) Validation of linear mixing by the olfactometer.(A) Binary mixtures are generated by mixing odorised air from two streams each equipped with a mass flow controller and presented as a stimulus when a five-way valve (‘final valve’) is actuated. To present single odours, odorised air from one line was mixed with air that passes through a blank canister in the other line. A photoionisation detector (PID) was used for calibration. (B) Example PID measurements for single odours (‘Odour A’ and ‘Odour B’), and their mixtures. Linear sum of the components (green trace) is shown superimposed with the observed PID signal (black). (C) Observed amplitudes for mixtures vs. linear sum of component amplitudes. Light blue line, the observed mixture amplitude equals the linear sum of components. (D–G) Investigation of mixture summation by mitral and tufted (M/T) cells under anaesthesia. (D) Top: schematic of two-photon GCaMP6f imaging from somata of M/T cells in naïve Tbx21Cre::Ai95D mice under ketamine/xylazine anaesthesia. Bottom: example field of view. Scale bar = 50 μm. (E) Example session structure. Single ethyl butyrate (EB) and methyl butyrate (MB) trials, as well as EB + MB mixture trials, are indicated with colour codes. (F) Transients from three example regions of interest (ROIs) as highlighted in (D). Linear sum (green trace) was constructed by linearly summing the averages of single EB and MB responses. Grey bar represents the time of odour presentation (0.5 s). (G) Scatter plot of observed mixture response amplitude against linear sum of components. N = 183 ROIs, seven fields of view, four mice.

Given that nonlinear summation is so widespread in the OB output, how well can mice analyse binary mixtures to accurately detect the presence of the target odour at the behavioural level? To assess this, we used a Go/No-Go paradigm for head-fixed mice. The rewarded stimulus (S+ odour) contained EB, either as a single odour or as a component of binary mixtures (Figure 3A). To train mice on this task efficiently, after habituation, the head-fixed mice were first trained to discriminate EB against other single odours (Figure 3A). The mice learned to perform this task well, reaching an accuracy of 80% within 200 trials on average (Figure 3B and C; number of trials to reach 80% accuracy = 194.3 ± 21.9; n = 7 mice). Subsequently, these mice were trained to detect the presence of EB in binary mixtures (Figure 3B and C). The mice performed the mixture task at a high accuracy from the beginning (90.2% ± 1.3% of trials with correct responses in the first session; n = 7 mice). However, the mistakes they made were odour-specific, in that they tended to lick more indiscriminately on MB-containing trials regardless of the presence of EB (Figure 3D and E; mean lick preference index for MB trials = 0.74 ± 0.04 vs. 0.93 for five other background odours, p = 0.0035, one-way ANOVA; n = 7 mice). However, with training, the performance on MB trials, too, became accurate, demonstrating that mice can learn to accurately detect the target odour in binary mixtures even when the background odour is similar.

Mice can learn to accurately analyse difficult binary mixtures.

(A) Behavioural paradigm: Go/No-Go task with head-fixed mice. Ethyl butyrate (EB)-containing olfactory stimulus was the rewarded stimulus. (B) After habituation, mice learned to discriminate EB against other single odours in the panel. Once proficient, mice learned to detect the presence of EB in binary mixtures. 30% of stimuli in the mixture stage were single odours. (C) Behavioural performance for all mice (n = 7 mice). Mean and SEM shown. (D) Odour-specific accuracy shown for the first (left) and last day of mixture training (right). Green shades indicate high accuracy. Top row corresponds to rewarded trials, and bottom six rows correspond to unrewarded trials. Average accuracy from all animals shown (n = 7 mice). Bl. = blank. (E) Lick preference index measures licks that occur preferentially on rewarded trials for a given background odour. A lick preference value of 1 occurs when all anticipatory licks were observed in rewarded trials only. Grey points, data from individual animals (n = 7 mice); mean and SEM shown in black.

To understand why mice were able to acquire the binary mixture task so easily, we analysed how M/T cells represent binary olfactory mixtures while the trained mice accurately performed the task. GCaMP6f fluorescence was measured from M/T cell somata. Data from the behaving mice was compared to the case when the same mice were later anaesthetised with ketamine and xylazine (Figure 4A–D). This revealed that the M/T cells in behaving mice tended to sum mixture responses more linearly than when the mice were anaesthetised (Figure 4D). To quantify this data, we expressed a deviation from the linear sum as a fraction of the predicted, linear sum of component responses (‘fractional deviation’; Figure 4E). Data from awake, behaving mice showed a clear shift in the distribution of this index compared to the anaesthetised case. In the anaesthetised case, a steep slope in the cumulative histogram occurs at a negative value (median fractional deviation = –0.39), indicating that the majority of the OB output neurons exhibit sublinear summation. On the other hand, M/T cells from behaving mice showed a broader distribution centred around 0 (median fractional deviation = –0.12; P = 3.83 × 10–10, two-sample Kolmogorov–Smirnov test comparing the behaving vs. anaesthetised data; n = 202 ROIs and 103 ROIs, respectively). The broader distribution for the awake, behaving condition likely reflects a greater variability in this condition. Consistent with this, adding a Gaussian noise to the data from anaesthetised mice made the distribution broader, but without affecting the median (Figure 4F). Using this index to compare the two conditions, we found that the difference was most striking in the first second after the odour onset (‘early phase’; Figure 4G). Further, we observed that the responses in the somata are more linear in the early phase compared to the apical dendrites (Figure 4—figure supplement 1). These observations together may indicate that dampened responses in the early phase of somatic responses may underlie the reduced mixture suppression, although some contribution may come from more decorrelated responses at the somata (Figure 4—figure supplement 1). Overall, the results indicate that the property of mixture summation is more linear in the awake, behaving mice, and that linearisation is not imprinted permanently as a result of learning.

Figure 4 with 1 supplement see all
Mixture summation is more linear in awake, behaving mice.

(A) Schematic of experimental setup. Imaging of GCaMP6f signals from the somata in Tbx21Cre::Ai95D mice performing the mixture detection task. On the last day, imaging took place under ketamine/xylazine anaesthesia (Anae). (B) Example field of view showing the same neuron from two imaging sessions. (C) Relative fluorescence change evoked by ethyl butyrate (EB), methyl butyrate (MB), and their mixture for the two conditions for the neuron indicated by arrow in (B). Grey bar, odour presentation (0.5 s). (D) Scatter plot of observed EB + MB mixture response amplitude against linear sum of component odour responses (average of 20 frames). Indicated time is relative to odour onset. (E) Cumulative histograms of fractional deviation from linear sum for data from anaesthetised mice (blue) and trained mice performing mixture detection task (black). N = 202 regions of interest (ROIs) from 13 fields of view, six mice for behaving case; 103 ROIs from eight imaging sessions, four mice for the anaesthetised case. (F) Effect of Gaussian noise on the fractional deviation distribution. Colours correspond to data with different amount of noise added. (G) Time course of median fractional deviation for the two datasets. Median deviation was obtained from each field of view; thick line shows the median for all fields of view, along with the 25th and 75th percentiles (shaded areas). See also Figure 4—figure supplement 1.

While the above result demonstrates that mixtures sum more linearly in awake, behaving mice, M/T cell responses are also significantly smaller and more variable in this condition. This is quantified in an analysis of trial-by-trial response similarity, which measures the Pearson correlation coefficient between population responses from individual trials (Figure 5A–C). This showed that, in the anaesthetised mice, the M/T cell response patterns are highly correlated, both within class (comparison between trials with EB) and across classes (Figure 5A and B). In the awake, behaving mice, as expected, responses were less consistent, indicated by the lower within-class correlation (Figure 5A and B). However, because across-class patterns were substantially less correlated (Figure 5A and B), S+ responses may be more discriminable in the behaving mice.

Figure 5 with 1 supplement see all
Encoding the presence of the target odour is better in behaving mice.

(A) Correlation matrices for data from anaesthetised mice (top) and awake, behaving mice (bottom), similarity of evoked responses between individual trials, for two time points, t1 and t2, as indicated in panel (B). S+ trials were trials with ethyl butyrate (EB) either as single odour or in binary mixtures. (B) Average correlation coefficient within S+ trials (magenta), within S- trials (blue), or across trials (green) for mitral and tufted cells (M/TC) responses from anaesthetised mice (top panel) and awake, behaving mice (bottom panel). (C) Discriminability index, which measures the proportion of trials in which the within-S+ correlation was higher than the across class, was compared at each time point for data from anaesthetised mice (blue trace) and awake, behaving mice (black). See also Figure 5—figure supplement 1. (D) Support vector machine (SVM) for discriminating EB vs. non-EB responses, trained on M/TC somatic responses from randomly selected 80% of trials, and tested on 20% of trials for anaesthetised (blue) and behaving (black) conditions. Both training data and test data contained random selection of single-odour responses (grey rectangles in scheme) and mixture responses (hollow rectangles). (E) SVM trained with single-odour responses, tested on mixture responses.

The ability to reliably encode the presence of EB was first quantified based on within-class vs. across-class correlation by counting the proportion of trials where within-class correlation was higher than across-class (Bridgeford et al., 2021). This indicated that M/T cells from anaesthetised mice showed better discriminability initially, but later, M/T cells in the behaving mice performed as well as the anaesthetised mice (Figure 5C). Note that a similar result is obtained when trials are sorted by the presence of MB (Figure 5—figure supplement 1). While useful, the Pearson correlation coefficient takes into account all ROIs, irrespective of how informative they are in discriminating patterns, and may not reflect the true ability of a population of neurons to encode odours. To address this, we trained support vector machines (SVMs) to discriminate responses evoked by odours containing EB (S+) vs. odours without EB (S-). We used 80% of randomly selected trials from each session for training, and the remaining 20% of the trials to test the performance (Figure 5D). This time, M/T cells from the two conditions performed similarly for the first 1 s, performing above chance soon after the odour onset (earliest time where accuracy is significantly above 0.5 = 0.67 s for the behaving case, and 1.12 s for anaesthetised data; t-test at a significance level of 0.05; n = 13 fields of view, six mice for behaving case; 8 fields of view, four mice for the anaesthetised case; Figure 5D). Further, the data from behaving mice outperformed the anaesthetised case in the later phase (median accuracy for 2–3 s after odour onset = 0.58 vs. 0.70 for anaesthetised vs. behaving mice, respectively; p = 0.027, Mann–Whitney U test for equal medians). We also wished to assess how mixture responses relate to single-odour patterns. To do so, we trained SVMs using single-odour responses (EB vs. other single odours) and tested the performance on mixture responses (Figure 5E). Again, the data from behaving mice outperformed the data from anaesthetised mice in the late phase, although the difference was not statistically significant (median accuracy for 2–3 s after odour onset = 0.55 vs. 0.64 for anaesthetised vs. behaving mice, respectively; p = 0.065, Mann–Whitney U test for equal medians). This may suggest that, over time, mixture responses come to resemble the target component odour pattern more in the awake, behaving mice than in the anaesthetised mice. Curiously, the time course of decoder accuracy did not strictly depend on the linearity of mixture summation (Figure 6—figure supplement 2). Overall, the result suggests that, despite the substantial differences in the amplitudes and variability of responses, M/T cells in behaving mice are able to encode the presence of the target odour well.

To what extent is the mixture representation in the OB affected by learning? Previous studies suggest that prior exposures and familiarity to odours affect the ability to analyse odour mixtures (Grabska-Barwińska et al., 2017; Poupon et al., 2018). To address this, we assessed how EB and MB mixtures are represented in naïve but awake, head-fixed mice. Since different levels of motivation, with accompanying changes in the sniff patterns, affect the olfactory system (Carey and Wachowiak, 2011; Jordan et al., 2018), we made one group naïve mice engaged and a second group of mice disengaged by changing the reward contingency (Figure 6A). That is, we presented the same sets of odour stimuli, but instead of associating specific odours with reward, the water reward was either delivered after odour onset but on randomly selected trials to engaged the mice, or about 15 s before the odour onset on all trials in order to disengage the mice (Figure 6A; n = 17 sessions, six mice, and 14 sessions, four mice, respectively). The level of engagement was confirmed by the speed of inhalations during odour presentations (Figure 6B) and the number of anticipatory licks generated (Figure 6—figure supplement 1).

Figure 6 with 3 supplements see all
Mixture representation depends on behavioural states.

(A) Reward contingencies used to achieve different behavioural states in awake mice. Top: mice were trained to discriminate between odours with ethyl butyrate (EB) vs. odours without EB. Middle: naïve mice received water reward 3 s after the onset of odour, on randomly selected trials. Bottom: naïve mice received water every trial, 15 s before odour onset. (B) Sniff patterns associated with trained and behaving mice (black), naïve and engaged mice (orange), and naïve and disengaged mice (light blue). Speed of inhalation during odour presentations shown. (C) Comparison of support vector machine (SVM) performance using data from behaving mice (black) vs. naïve engaged mice (left panel, orange), and vs. naïve disengaged mice (right panel, light blue). SVM was trained to discriminate responses to S+ vs. S- odours using randomly selected 80% of trials and tested on the remaining 20% of the trials. (D) Decoder performance for the results in (C) plotted using approximate z-values obtained from MATLAB implementation of Wilcoxon rank-sum test. (E) As in (C), but SVM was trained using responses to single odours, and tested on mixture responses. (F) As in (D) but for the results in (E). n = 13 fields of view, six mice for behaving, 17 fields of view, six mice for naïve, engaged, and 14 sessions, four mice for naïve, disengaged mice. See Figure 6—figure supplement 1 for linearity of summation.

In all awake conditions, we found the EB and MB responses to sum equally linearly (Figure 6—figure supplement 1), indicating that the linearity of summation does not depend on the behavioural states or learning. However, the performance of SVMs on these datasets showed a state-dependence. While the general ability to discriminate S+ vs. S- odours was comparable across the states, the ability to analyse mixtures in terms of the constituents – SVMs trained on single odours and tested on mixture patterns – was particularly poor for the disengaged mice (Figure 6C–F). Since the linearity of summation was comparable across the three groups (Figure 6—figure supplement 1), this, again, suggests that the ability to correctly decode the presence of an odour does not strictly correlate with the linearity of mixture summation. In all cases, the decoder performed most accurately in the late phase. We hypothesised that this extra time may reflect an involvement of recurrent interactions with the piriform cortex, which is thought to store learned representations and feed back to the OB (Chapuis and Wilson, 2011; Grabska-Barwińska et al., 2017). Our preliminary result from muscimol infusion in the ipsilateral piriform cortex, however, suggests that this may not be the case (Figure 6—figure supplement 3).

Under what circumstance does discriminability of odour mixtures decouple from linearity in mixture summation? To study the effect of saturating sublinearity in isolation, we made a simulation as follows. Linear sums of responses were constructed by adding component responses obtained from the imaging sessions. In one test, these sums, with added noise, were passed through SVMs that had been trained with single odour responses. In another test, the linearly summed responses with noise were transformed with a normalising function (Mathis et al., 2016; Penker et al., 2020), which adds sublinearity in an amplitude-dependent manner (Figure 7B; see Materials and methods). Then, these signals were passed through the same SVMs to assess how discriminable the activity patterns were. When the data from the anaesthetised mice was used, normalising sublinearity was particularly detrimental around 2 s after the odour onset (Figure 7C), qualitatively reproducing the transient decrease in the accuracy seen with the observed mixture responses (Figure 5E). In contrast, with the data from the behaving mice, normalising sublinearity did not have as significant an effect on the mixture discriminability, even in the late stage when sublinear summation becomes more widespread (Figure 7D and E). Thus, the functional consequence of nonlinear summation may depend on specific circumstances, for example, on how separable, or decorrelated, the activity patterns are. Overall, our result indicates that mixture responses in the OB are highly state-dependent and evolve over time (Figure 6F).

Simulation indicates data from behaving mice is less susceptible to normalising sublinearity.

(A) Simulation approach. For each field of view, component responses were averaged across trials and summed to obtain simulated mixture responses. Subsequently, uncorrelated Gaussian noise was added, and the summed responses were further transformed using a normalisation model (Mathis et al., 2016; Penker et al., 2020). These were then used as test inputs on support vector machines (SVMs) that had been trained with single-odour responses. (B) Parameter extraction. Scatter plot shows observed ethyl butyrate (EB) + methyl butyrate (MB) mixture responses against linear sums of component responses (black dots). Red line shows the model fit used in the simulation. Data was from behaving mice, at 2 s after the odour onset. (C) Comparison of SVM performance using data from anaesthetised mice. Black line corresponds to linear sum + noise as inputs; green line corresponds to the results with additional normalising sublinearity. Mean ± SEM shown. (D) Same as in (C) but for data from awake, behaving mice. (E) Comparison of SVM performance. SVM accuracy for simulated mixture responses with sublinearity expressed as a fraction of accuracy obtained with simulated linear sum. p = 0.03 for early phase (0–1.5 s) and 8.13 * 10–4 for late phase (1.5–3 s after odour onset); paired t-test. (F) Schematic of the finding: possible mechanisms of mixture representation in behaving mice may evolve over time. In awake, behaving mice, responses evoked in the olfactory bulb (OB) output neurons are dampened early on, remaining largely in the linear range of the normalisation curve. Over time, the responses become larger, making sublinearity more widespread, but a pattern decorrelation may make evoked mixture responses still discriminable.

Discussion

Segmentation and extraction of relevant information from mixtures of signals are important and perpetual tasks for sensory systems. Nonlinear summation of signals is generally thought to pose difficulty in demixing component signals. Here, we demonstrate that the signal summation in the primary olfactory area of the mouse is significantly more linear in awake mice. This is notable because many previous studies reported that suppressive mixture interactions are widespread in the central nervous system, based on anaesthetised preparations (Bell et al., 1987; Cruz and Lowe, 2013; Penker et al., 2020; Stettler and Axel, 2009).

However, our analysis also indicates that the ability to discriminate binary mixture responses does not correlate strictly with linearity in summation. First, the behavioural improvement was not accompanied by more linear summation. Second, the time course of decoder accuracy does not follow the linearity of summation. Indeed, in the data from behaving mice, the decoder performance peaked when mixture responses were most sublinear. This raises a question: is sublinear summation in mixtures detrimental or beneficial to discriminating mixtures of odours? Our simple simulation indicates that, generally, a sublinear mixture summation due to saturating influences limits the discriminability of mixture responses. Thus, dampened responses in awake animals bring some advantage by maintaining the mixture responses in the linear range. However, in the behaving mice, even when the responses became larger at a later phase, the saturating influence was less detrimental to discrimination. Here, decorrelated response patterns may play a more crucial role as this is known to enhance classifier or behavioural performances (Bhattacharjee et al., 2019; Friedrich and Wiechert, 2014; Gschwend et al., 2015; Padmanabhan and Urban, 2010). While the behavioural task described in this study is simple and animals make accurate decisions within 1 s (771 ± 97 ms) of stimulus onset, the slower mechanisms described here may be important when larger and more reliable responses are required for accurate decisions. So, in addition to sampling time (Rinberg et al., 2006), this phenomenon may be a part of mechanisms needed to solve more difficult tasks accurately (Abraham et al., 2010; Wilson et al., 2017).

Mechanistically, the ability to accurately analyse, or discriminate between, mixtures based on the knowledge of component is hypothesised to involve the piriform cortex, through learned representations (Chapuis and Wilson, 2011; Grabska-Barwińska et al., 2017). For tufted cells, however, the relevant sources of feedback could be other brain regions, such as the anterior olfactory nucleus (Chae et al., 2020). It will be an intriguing future investigation to identify the source and mechanisms of state-dependent mixture representations.

Perception of olfactory mixtures has long fascinated investigators. Mixtures of odours often have qualities that are different from those of the individual components. Accurate recognition of components is particularly hard for human subjects. For untrained subjects, olfactory mixtures that contain about 30 components tend to smell alike (Weiss et al., 2012). Even highly trained people like perfumers can only accurately identify individual components if unfamiliar mixtures contained no more than five components (Poupon et al., 2018). These demonstrate an ultimate limit in the analytical perception of olfactory mixtures. In addition, in non-expert humans and rodents alike, when the task is to look for a particular aroma in the mixture, the tendency is to falsely report the presence of the target odour in target-odour detection tasks (Laing and Glemarec, 1992; Rokni et al., 2014). In all cases, an extensive training for specific odours can improve the ability to detect the target odour, as seen in the case for sommeliers who routinely analyse key components in wines, which can contain several hundred component mixtures (Ilc et al., 2016). Thus, while mixture perception is highly context specific (Rokni and Murthy, 2014), training in specific odours seems to be key to improving on mixture analysis. With more complex mixtures, mechanisms within the OB may reach a limit, and also is unlikely to be the only mechanism that is used to solve the task. Elucidating what roles the primary sensory areas plays is a crucial step towards a mechanistic understanding of complex sensory processing.

Materials and methods

Animals

All animal experiments have been approved by the OIST Graduate University’s Animal Care and Use Committee (protocol 2016-151 and 2020-310). Tbx21Cre (Haddad et al., 2013) and B6J.Cg-Gt(ROSA)26Sortm95.1(CAG-GCaMP6f)Hze/MwarJ, also known as Ai95D (Madisen et al., 2015), were originally obtained from the Jackson Laboratory (stock numbers 024507 and 028865, respectively). Tbx21Cre::Ai95D mice were generated by crossing homozygous Tbx21Cre and Ai95D mice, resulting in heterozygous animals used for imaging experiments. C57Bl6J mice were purchased from Japan CLEA (Shizuoka, Japan) and were acclimatised to the OIST facility for at least 1 week before they were used for experiments. All mice used in this study were adult males (8–11 weeks old at the time of surgery).

Olfactometry

Request a detailed protocol

A custom-made flow-dilution olfactometer was used to present odours. Briefly, custom Labview codes were used to control solenoid valves, and a flow controller (C1005-4S2-2L-N2, FCON, Japan) was used to regulate the rate of air flow. A pair of normally closed solenoid valves was assigned per odorant and used to odorise the air. These solenoid valves were attached to a manifold, such that a set of eight pairs had access to a common stream of air. To generate binary mixtures of odours while keeping concentrations stable, one odour from each manifold (lines A and B) was used. For single-odour presentations, odour was mixed with air that passed through an empty canister. The odorised air was directed towards the animal only when the solenoid valve closest to the animal (final valve) opened. Final valve was opened for a short time (0.1 s for experiments involving behavioural analysis only, and 0.5 s for all imaging experiments, to accommodate for slow respirations during anaesthesia), to avoid adaptation related to temporal filtering for high sniff frequencies associated with long odour pulses (Verhagen et al., 2007). Odours were presented at 1–5% of the saturated vapour. Total air flow, which is a sum of odorised air and the dilution air, was approximately 2 l/min, which was matched by the air that normally flows towards the animal. Inter-trial interval was approximately 40 s to purge airways with clean, pressurised air and to ensure that the flow controllers have stabilised before each odour presentation. All odorants were from Tokyo Chemical Industry (Tokyo, Japan), apart from EB (W242705), which was from Sigma-Aldrich. Product numbers were T0247 (ethyl tiglate), V0005 (methyl valerate), A0061 (acetophenone), A0500 (methyl anthranilate), B0757 (butyl butyrate), B0763 (MB), S0015 (methyl salicylate), S0004 (salicylaldehyde), T0248 (methyl tiglate), and A0232 (eugenol). Purity of all odorants was at least 98% at the time of purchase. Stock odorants were stored at room temperature in a cabinet filled with N2 and away from light.

Surgery

Head plate implantation

Request a detailed protocol

All recovery surgery was conducted in an aseptic condition. 8–11-week-old male C57Bl6/J mice were deeply anaesthetised with isoflurane. The body temperature was kept at 36.5°C using a heating blanket with a DC controller (FHC, Bowdoin, USA). To attach a custom head plate about 1 cm in width weighing a few grams, the skin over the parietal bones was excised and the soft tissue underneath was cleaned, exposing the skull. The exposed skull was gently scarred with a dental drill, cleaned, dried, and coated with cyanoacrylate (Histoacryl, B. Braun, Hessen, Germany) before placing the head plate and fixing with dental cement (Kulzer, Hanau, Germany). For optical window implantation, adult, male Tbx21Cre::Ai95D mice were deeply anaesthetised and underwent a window implantation procedure as previously described (Koldaeva et al., 2019). Briefly, after exposing the frontal bone, a craniotomy about 1 mm in diameter was made over the left OB and on the exposed dorsal surface, a cut piece of coverslip that snugly fit in the craniotomy, on the edge of the drilled bone, was gently pressed down, sealed with a cyanoacrylate and fixed with dental cement. Mice were allowed to recover in a warm chamber, returned to their cages, and given carprofen subcutaneously (5 mg/kg) for 3 consecutive days.

Habituation and behavioural measurements

Request a detailed protocol

Water restriction began 2 weeks after the surgery. Mice went through 3 days of habituation to head fixation, one session per day for approximately 30 min, until mice learned to lick vigorously for water reward. Respiration pattern was measured by measuring the air flow just outside the right nostril by placing a flow sensor (AWM3100V, Honeywell, NC), and the data was acquired at 1 kHz. Lick responses were measured using an IR beam sensor (PM-F25, Panasonic, Osaka, Japan) that was part of the water port. Nasal flow, an analog signal indicating the odours used, lick signal, a copy of the final valve, and water valve timing were acquired using a data acquisition interface (Power1401, CED, Cambridge, UK).

Discrimination training

Request a detailed protocol

After habituation, the head-fixed mice were trained to associate a water reward with a target odour (EB). The reward was two droplets of water (10 μl each), which arrived 3 s after the onset of the final valve opening. The mice underwent single-odour discrimination training first until they generated anticipatory licks in response to EB presentations and correctly refrained from licking in response to other single odours. Once the overall accuracy was above 80% in at least one behavioural session, the mice went through the mixture detection task. A typical training session comprised roughly 150 trials, lasting about 1 hr. Rewarded trials comprised a third of all trials. Two-photon imaging took place once the mice performed at 80% accuracy or above in MB trials.

Random association paradigm

Request a detailed protocol

After habituation, the head-fixed mice were presented with the same odour mixture stimuli as those that underwent the discrimination training. The water reward was delivered on randomly selected trials, 3 s after the onset of the final valve opening. One behavioural session was used to accustom the mice to the odours. Two-photon imaging commenced from the second behavioural session.

Disengagement paradigm

Request a detailed protocol

After habituation, the head-fixed mice were presented with the same odour mixture stimuli as above. The water reward was delivered every trial, 15 s before the onset of the final valve opening, arriving in the middle of the inter-trial interval, which was 40 s to ensure thorough purging to clear the lines, as well as to stabilise the flow controllers. Two-photon imaging commenced from the second behavioural session.

Odour mixture trial composition

Request a detailed protocol

Binary mixtures have been chosen due to the smaller number of possible odour combinations compared to more complex mixtures. However, even with 11 odours, there is a limit in the number of trials each head-fixed mouse can sample in a given session. We therefore decided to focus on the EB + MB mixtures. However, it was crucial that EB, MB, and EB + MB mixture is not presented too frequently to avoid possible adaptation, as well as mice becoming over familiar, especially in the EB-detection task, where the goal was not to train mice to remember specific odour combinations. Thus, we used a compromise paradigm where EB, MB, and EB + MB appeared every 10 trials.

Data analysis

Request a detailed protocol

The data was analysed offline using custom MATLAB codes. To calculate the accuracy, the number of licks during 3 s from the final valve onset was measured for each trial. Threshold for an anticipatory lick was set to 2, thus the correct response for rewarded trials was two or more beam breaks, and the correct response for unrewarded trials was one lick or less during the response time window. To calculate the learning curve, the accuracy was expressed as the proportion of correct trials in a given block of 50 trials. Time to inhalation peak: to calculate the speed of inhalation, onset of inhalation and peak of inhalation was detected using Spike2 (CED, Cambridge, UK), using the built-in event detection functions. Briefly, inhalation peaks were detected using the ‘rising peak’ function. These events were used to search backwards in time for the inhalation onset when the flow signal crossed a threshold value. Lick preference index: to measure how well mice discriminated rewarded vs. unrewarded mixtures, anticipatory licking patterns for the two types of trials were compared using the following formula:

Lick preference index=(LickrewardedLickunrewarded)/(Lickrewarded+Lickunrewarded).

where Lickrewarded corresponds to the average number of anticipatory licks on rewarded trials, and Lickunrewarded corresponds to that for unrewarded trials.

Imaging

Request a detailed protocol

A custom-made two-photon microscope (INSS, UK) with a resonant scanner was used to observe fluorescence from the OB of Tbx21Cre::Ai95D mice in vivo. 3D coordinates for imaged fields of view were recorded relative to the location of a reference, blood vessel pattern on the surface. Unless otherwise stated, imaging from somata was done relatively superficially, just below the glomeruli, thus mainly comprised TCs, which use firing rate modulation to represent odours (Fukunaga et al., 2012). Fields of view for glomerular and somatic levels were 512 μm × 512 μm, and 256 μm × 256 μm, respectively, and overlapped for awake and anaesthetised conditions, but fewer sessions took place under anaesthesia. In each trial, 400 image frames were acquired at 30 frames per second, with 200 frames before the final valve opening to obtain a steady baseline. Unless otherwise stated, the time window analysed for the odour-evoked responses was the first 1 s since the onset of final valve opening. For imaging under ketamine/xylazine anaesthesia (100 mg.kg–1/20 mg.kg–1 intraperitoneally), mice were kept on a warm blanket (FHC) to maintain the body temperature at 36°C.

Ipsilateral muscimol infusion in the anterior piriform cortex

Animals were unilaterally implanted with a 26G cannula (10 mm length; C315GS-4/SPC, Plastics One). The cannula was inserted through a small craniotomy at a stereotactic coordinate (AP, 2.2 mm; ML, 2.4 mm relative to bregma) and advanced over 6.1 mm at 45° to target the anterior piriform cortex, and fixed with dental cement (Kulzer, Hanau, Germany). For infusion, 500 nl of 2 mM muscimol (M1523, Sigma-Aldrich, MO) was injected using a Hamilton microsyringe, at a rate of 100 nl/min approximately 10 min before the imaging started.

Post-hoc verification of infusion cannula placement

Request a detailed protocol

Animals were infused with 500 nl DiI (Invitrogen, V22885), immediately transcardially perfused with phosphate buffer (225.7 mM NaH2PO4, 774.0 mM Na2HPO4. pH 7.4) and PFA (4%), dissolved in phosphate buffer. The heads (including the implants) were postfixed in PFA at 4°C for at 48–96 hr before extraction of the brains. Coronal sections of 100 µm thickness were cut on a vibratome (5100 mz-Plus, Campden Instruments, Leicestershire, UK) and counterstained using DAPI (D9542, Sigma-Aldrich). Images were acquired using a Leica SP8 confocal microscope using a ×10 (NA 0.40 Plan-Apochromat, Leica) objective. Images were taken at a resolution of 1024 × 1024 pixels per field of view (1163.64 × 1163.64 μm).

Image analysis

Transients were extracted as follows. ROIs were manually delineated using an average frame from each imaging session. For glomeruli, ROIs were delineated by tracing the outer rim of congregated labelled processes surrounded by a dark region. Pixel values within each ROI were averaged to obtain a time series. Imaging sessions with motion artefacts and drifts were removed from analyses. For each transient, the baseline period was defined as 2 s preceding the final valve opening. Relative fluorescence change (ΔF/F) was calculated with respect to this baseline. Odour response period was 1 s (30 frames) starting at the onset of the final valve opening, unless otherwise stated. Awake mice tended to adjust sniff patterns, so the final valve opening was not triggered by nasal flow. Masking index for the glomerular level was calculated as previously described (Rokni et al., 2014). Briefly, responses to single odours were obtained from glomeruli of anaesthetised mice. In each field of view, only the glomeruli responsive to the target odour (EB) were analysed. To obtain glomeruli evoked by the target odour, the evoked response amplitudes were converted into z-scores. Glomeruli that responded to EB with z-scores higher than 2 were considered. The masking index was the average overlap in the evoked response, where the maximum value for each glomerulus was 1. For each odour, responses were averaged over three or more trials. Mean and standard error of the mean are shown in figures, unless otherwise stated. Boxplots were constructed using the MATLAB function boxplot and show the median, 25th and 75th percentiles. Outliers are shown with red crosses. The experiments were not done blindly since the stimulus-reward contingency was visible to the experimenter. However, the olfactometer performance, age and sex of the mice, and analysis codes used were the same for all conditions.

Fractional deviation from linearity was calculated as,

(RobservedRlinearsum)/Rlinearsum,

where Robserved is the mean response amplitude for observed mixture (e.g. response to EB + MB mixture), Rlinear sum is the trial average linear sum of component responses (e.g. EB response + MB response). Median fractional deviation was obtained from each field of view. For testing the effect of trial-by-trial variability, Gaussian noise was generated using the MATLAB function normrnd with the mean set to 0, and added to responses from each trial, before the fractional deviation was calculated.

Deviation from linearity was the difference between an observed mixture response amplitude and a linear sum of components normalised by the joint standard deviation:

(RobservedRlinearsum)/(s.e.m.observed+s.e.m.linearsum)

Discriminability index based on correlation coefficient was calculated in the same way as Discr, described in Bridgeford et al., 2021, except that the distance measure used was 1 – Pearson’s correlation coefficient, instead of the Euclidean distance.

Decoding analysis

Request a detailed protocol

SVMs were trained using the MATLAB function fitclinear with linear model intercept (‘FitBias’) set to 0. Test datasets were used to predict the trial types using the MATLAB function predict. Signals were averaged over 30 frames (~1 s) for each time point to obtain the time course.

Simulating the effect of normalising sublinearity on discriminability

Request a detailed protocol

SVMs were trained as above using the single-odour responses only. Then simulated mixture responses were constructed as follows. Single-odour responses were averaged over trials. Combinations of two averaged responses were added linearly, and uncorrelated Gaussian noise (mean = 0) was added. The noise term comprised two components: fixed, baseline standard deviation (0.2 for anaesthetised data, and 2 for awake, behaving case) and standard deviation that scaled with response amplitude by taking the slope of a linear regression of the observed data. In one test, linear sum with added noise was used as inputs to the above SVMs. In a second set of tests, these simulated responses were further passed through a normalising sublinearity and tested on the same SVMs. The equation of normalisation (Penker et al., 2020) was

Rj*= Rmax21+e-s.Rj-1

where Rj represents the linear-sum amplitude of the jth neuron, and Rj* is the normalised response amplitude of jth neuron. Parameters s and Rmax were obtained by fitting the data to observed mixture responses (on data from behaving animals at t = 2 s after odour onset). Rmax was 6 and s was 0.2.

Data availability

The data generated in the study is available in Dryad Digital Repository at 10.5061/dryad.p2ngf1vrh. The files consist of individual data to compare linear sum vs. observed mixture responses (300 - 1000 ms after odour onset).

The following data sets were generated

References

  1. Book
    1. Marr D.
    (1982)
    Vision: A Computational Investigation into the Human Representation and Processing of Visual Information
    W.H.Freeman and Company.

Decision letter

  1. Naoshige Uchida
    Reviewing Editor; Harvard University, United States
  2. Gary L Westbrook
    Senior Editor; Oregon Health and Science University, United States
  3. Naoshige Uchida
    Reviewer; Harvard University, United States
  4. Florin Albeanu
    Reviewer

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

Decision letter after peer review:

[Editors’ note: the authors resubmitted a revised version of the paper for consideration. What follows is the authors’ response to the first round of review.]

Thank you for submitting the paper "State-dependent linearisation of mixture representations by the olfactory bulb" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including Naoshige Uchida as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. We are sorry to say that, after consultation with the reviewers, we have decided that this work will not be considered further for publication by eLife.

Summary:

The reviewers agreed that this study addresses an important question: how complex odors are processed in the olfactory bulb, more specifically, whether mixture components are represented as a linear sum of components or sublinear interactions occur as suggested by previous studies. The results are potentially interesting, but the reviewers have raised substantive concerns.

1. The reviewers were not convinced by the analysis based on the fractions of sublinear versus linear interactions. The authors report the fraction of non-linear interactions as defined by the activity showing a deviation from linearity using a statistical criterion (> 2 standard deviation). The reviewers thought the authors need to perform more analysis by considering trial-to-trial variability, magnitude difference, and response saturation etc.. Each reviewer made useful suggestions (see below).

2. Odor responses in awake conditions are generally weaker than those under anesthesia. It is, therefore, unclear whether the overall discrimination is better during awake condition. It is important to perform decoding analysis using a classifier trained using odor responses in each condition (not a classifier built using predicted responses).

3. The reviewers had somewhat different opinions on overall significance of the work. One reviewer thought that the difference in awake and anesthetized conditions, if shown convincingly by addressing the above points, is important because previous studies have predominantly reported sublinear interactions. One reviewer, however, thought that the difference between awake and anesthetized conditions is not sufficiently novel, because many other differences have already been reported. One reviewer thought that further mechanistic insights would be important.

Overall, strong demonstrations of the above point 1 and 2, and/or further mechanistic insights would need to be considered for eLife. If the authors can address the above issues, we would be willing to reconsider a resubmitted version but without guarantees concerning publication.

Reviewer #1:

Adefuin and colleagues examined the interaction between components of binary odor mixtures in odor responses in mice. The authors used two-photon calcium imaging from the soma and apical dendrites of mitral/tufted cells in the olfactory bulb. Odor responses were measured in various conditions: under anesthesia (ketamine/xylazine), while well-trained mice were engaged in an odor discrimination task, or disengaged. The authors first show that mixture components interacted sublinearly in a large fraction of mitral/tufted cells (46%; Figure 6D) consistent with previous studies. However, when odor responses were measured in awake animals, very few mitral/tufted cells showed sublinear responses at soma (8-9%; Figure 6D). Interestingly, sublinear interaction was evident in apical dendrites of mitral/tufted cells (45%). Whether mixture components are represented linearly or not in the olfactory system is an important question, related to the animal's ability to identify or segment mixture components. Somewhat contrary to previous studies, this study demonstrates largely linear interactions. Furthermore, this study compares various behavioral conditions. These results are important and of interest to those who study sensory systems. I have a few concerns regarding data analysis.

1. Non-linear interactions are detected by the activity showing a deviation from linearity greater than 2 standard deviations. Using this criterion, non-linear interactions might decrease if the trial-by-trial activity becomes more variable. This is concerning because the activity might be less variable in the anesthetized condition, and the reduction in sublinear interactions in awake conditions may be due to a general increase in response variability during awake. Can the authors exclude the possibility that the decrease in sublinear interactions is merely due to an increase in response variability in the awake conditions. This issue also applies to the comparison between apical dendrites versus soma; are the signals in apical dendrite less variable (maybe due to some averaging across dendrites from multiple cells; see the following point 5)?

2. Related to the above issue, it would be useful to analyze the difference between conditions using different metrics to fully understand what really are different between conditions. The scatter plots shown in various figures do not show drastic differences between awake and anesthetized conditions, as might be indicated by the percent of sublinear responses. It would be useful to characterize the magnitude of sublinear/supralinear effects. For example, one can calculate a fractional change in the mean response. Does this measure show consistent difference between awake and anesthetized conditions?

Reviewer #2:

This study addresses how complex stimuli are represented in neural responses. This is particularly relevant to olfaction because the vast majority of stimuli are complex mixtures that perceptually, are not easy to decompose into parts. Nonetheless, the ability to discern a relevant odor from background odors is essential. This process is easier when neural responses to mixtures reflect the linear sum of the responses to the individual components. The main conclusion of this study is that the linearity of olfactory bulb responses to two-component mixtures increases awake versus anesthetized states. The authors provide some evidence to support this claim. However, this could be better quantified and there is a temporal aspect of linearization that is not addressed. Perhaps the most interesting aspect of the study is the difference in linearity between the dendrites and the somata of the mitral/tufted cells. But a statistical analysis of this finding was not evident. Overall, a mechanistic or functional approach to understanding these findings is lacking. The differences linearity between the anesthetized and awake are simply explained by response saturation anesthetized animals. There are hints at mechanism by which linearity is supported in the OB with comparisons between soma and dendrite, but these are not well developed. There is a model that addresses the functional significance of linearity, but this is only supplemental and not well described.

1) The quantification of linearity is incomplete. The deviation from linearity is a fine metric but just reporting the change in fraction of ROIs that show deviation from linearity in the sublinear direction is insufficient. There are many ROIs that show supralinear summation as well. Moreover, it is not clear where the cut-off is between linear and non-linear responses. There are statistical analyses that can be done on proportional data to determine if the proportions of non-linear ROIs are statistically different in the anesthetized and awake states. These should be done.

2) Another approach is to compare the slope of the lines fit to data in the Observed vs. Linear sum plots and to determine how much these deviate from 1, and if this is statistically significant. These slopes could also be compared between states.

3) It is also necessary to be clear about which timepoint is being analyzed. It is apparent from figure 4 that the linearity of the population of ROIs evolves with time, becoming most linear 3 sec after odor onset. This temporal evolution of linearity is not addressed at all. But could provide some insight into mechanism.

4) There is no statistical analysis of the dendrite versus soma data. The figure is not very convincing of differences. Why wasn't the temporal progression of responses shown for the dendrite versus soma data?

5) Linearity seems very correlated with the strength of the response. Sublinearity increases with strength suggesting saturation is a major contributing factor. Conversely, supralinear responses are more apparent with weaker component responses. It is well known that OB responses are much stronger in anesthetized animals. The deltaF/F values are greater in the anesthetized state compared to the awake state. So, the differences between the states may be trivially explained by saturation (as shown in Supplemental Figure 6). This doesn't really say anything about how the OB linearizes responses. However, the temporal evolution of linearization could speak to other mechanisms such as adaptation, recruitment of inhibition, descending control etc. Analyses of the dendrite versus soma signals over time could give insight as to where/when these mechanisms may be occurring. I think this was attempted in the supplement to figure 7, but the figure does not have sufficient explanation to figure out what is going on.

6) There was a model of how linearity might function in odor detection, but this is not well described or motivated in the text.

7) Overall, all the supplemental figures and data are essential to the study. Many of my concerns are somewhat addressed by these additional analyses. But they are not well explained. These should be integrated into the main figures of the paper and adequately described in the Results section.

Reviewer #3:

Adefuin et al. use multiphoton imaging of M/T cell responses to investigate whether neuronal representations of binary mixtures can be explained as a sum of the components. The current view in the field (built largely from studies in anesthetized animals), is that mixture summation is non-linear and increases with the degree in glomerular response overlap elicited by the components. The authors reproduce these results and ask whether the same phenomenon is observed in the awake state, in particular when the animals are engaged in an odor discrimination task. Unlike in the anesthetized state, the authors find that mixture representations are linear in the awake brain. They use a series of systematic behavioral paradigms to show that the observed linearity in the awake state (compared to anesthetized) is not dependent on task engagement (reward is given randomly, post-odor) or stimulus relevance (reward is given before odor). While the experiments are well done and the data is presented clearly, I have several major concerns about the interpretation of their results.

1) Given the data the authors present, it is unclear if one can conclude that the olfactory system is more or less linear in the awake state compared to the anaesthetised one. What seems to change most across the awake vs. anesthetized state is the response amplitude. Responses appear to be ~3x smaller in the awake mice. In the anesthetized state, non-linearity seems most apparent for large response amplitudes (>5 dF/F) with mixture responses being sub-linear, most likely due to saturation effects. The authors themselves do an analysis in Figure 6 – supplement 1 to show that most of the observed non-linearity in the anesthetized animals can be explained away after accounting for amplitude normalisation. The authors use this analysis to comment that the level of linearity is the same across all the three awake states, but the same figure shows that it is in fact the same even for the anaesthetized state.

To put it differently, it is indeed true from the authors data that the OB response gain is significantly lower in the awake state, but it is unclear if the summation is more linear if measured at similar response amplitude regimes in both awake and anaesthetised mice.

2) The authors argue that keeping response amplitudes small in the awake brain prevents sub-linear summation and therefore may lead to better mixture decomposition. They do a decoding analysis in anaesthetised mice to show that linear mixture representations (instead of using observed sub-linear representations) make odor classification easier. However, I find this analysis uninformative and misleading. It is no surprise that the decoders trained on single odor representations should perform better (or equivalent) when using linear sums as input instead of observed sub-linear representations. The authors use this observation to suggest that this mechanism aids discrimination ability in the awake state. However, given that even the single odor responses are much weaker and noisier in the awake state, it is likely that even the single odor discrimination ability is poorer in the awake state. By the same logic, mixture decomposition might be also much poorer in the awake brain than the anesthetized brain, even though summation is more linear, just because responses are weaker and noisier. In my opinion, the authors should compare decoding accuracy across awake vs. anesthetized responses if they want to assert that linearisation of responses in the awake brain leads to easier decomposition. Because otherwise, while linearisation in principle can aid decomposition, at least in the form that the authors observe here, it may come at a high cost on signal-to-noise ratio which would undo the gain that linearity provides, in principle, for discrimination.

3) At a more philosophical level, to this Reviewer, it is unclear if anesthesia vs. awake state difference in response should constitute the main focus of the manuscript. The authors explore summation properties under four different brain states, one of which is anaesthesia (also least behaviorally relevant). In three out of four states, they observe that summation is linear. In the fourth (anaesthesia), they observe that summation is sub-linear, but this happens at much larger response amplitude regimes compared to the three awake states sampled, presumably due to saturation. To me, it seems that the Authors here show that mixture summation in the OB, is largely independent of brain state since it is unaffected by whether the animal is task engaged or motivated etc.

4) It is unclear how to interpret the dendritic imaging comparison. First, the dendritic signal is pooled across many cells. If any of the cells that are being pooled shows sub-linearity, the pooled population response will look sub-linear, albeit less so than at the single cell level. Second, again like for the anesthetized vs. awake comparison, there is a discrepancy in response amplitudes – dendritic responses are ~2x stronger than the somatic responses and sub-linear summation would be more apparent as one approaches the saturation regime. Third, dendritic responses pool both mitral and tufted, while the somatic data the authors present is predominantly from tufted cells.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "State-dependent representations of mixtures by the olfactory bulb" for further consideration by eLife. Your revised article has been reviewed by three peer reviewers, including Naoshige Uchida as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Gary Westbrook (Senior Editor). The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

The authors have performed new data analysis and revised the manuscript. Overall, all reviewers thought that these changes have greatly improved the manuscript. In particular, the new metric to characterize linearity of mixture interactions -- median fractional change -- provides a more quantitative information, and the main conclusion has been strongly supported by this new method. The decoding analysis now shows that the odor information remains at least as informative in awake conditions as in anesthetized conditions, despite overall reduced magnitude of response in awake conditions. The new analysis also shows that discriminability does not necessarily correlate with the extent of linearity in mixture summation.

Although the manuscript has greatly improved, there are some remaining issues. Specifically, Reviewer 2 and 3 have raised concerns that need to be clarified. In particular, Reviewer 3 has raised the issue regarding the dissociation between discriminability and linear mixture summation. The functional significance of linear mixture summation, or the limitation thereof, needs to be discussed more carefully, and it would be useful to include some discussion on potential causes of increased discriminability in awake conditions. In addition, Reviewer 2, raised several points about clarifying and adjusting statements in the abstract and the text of the manuscript, and analyzing the temporal dependence of decoding accuracy to the instantaneous level of non-linearity in response when comparing across different brain states. We would like you to respond to these remaining issues.

Reviewer #1:

The authors have addressed my previous concerns by reanalyzing the data and revising the manuscript. Importantly, the authors now provide a more convincing metric for linearity of mixture interaction using the median fractional change over the population. I am more convinced by this metric which supports that mixture interaction is more linear in awake conditions. This new analysis addresses my most important concern. A new analysis also finds that decoding or behavioral ability to detect a mixture component did not correlate with the linearity of mixture interactions in odor responses. This finding, in a sense, makes the significance of the finding of linearity in odor responses a little unclear. On the other hand, I think that the demonstration that odor decoding does not benefit from linear mixture interaction has an important message to the field as one might assume that it would be the case. It would be more useful if the authors can demonstrate this point (odor decoding does not benefit from linear mixture interaction) in a more abstract form without relying on the data -- can we take this conclusion as a general result, or does it rely on some particular features in the data. A simple modeling might help address this issue. Although such demonstration is not required, if the authors can address this in a short period of time, it would enhance the manuscript significantly, although this point is a little secondary compared to the main point (linearity of odor mixture interaction).

Overall, I am satisfied by the more convincing demonstration of linearity of mixture interaction. Because sublinearity has been emphasized in the previous literature, the results presented in this manuscript provide an important result.

Reviewer #2:

I commend the authors for re-structuring the manuscript and performing additional analyses to address the concerns raised. These changes have indeed improved the manuscript substantially.

However, I still have concerns about the disparity in response amplitude and dynamics across the different states and how that may affect the interpretation of the results. I think these can be addressed by further clarifying and editing the statements in the abstract and rest of the manuscript before publication.

1. The authors use the new metric – Fractional deviation – to compare differences in the degree of linearity between the anesthesia and awake states. While this metric normalizes deviations from linearity with the response amplitude, it does not correct for the overall difference between the means of the two response amplitude distributions – in anesthesia versus awake states. If majority of the responses during the awake state are smaller than during anesthesia, and the amount of non-linearity scales with response amplitude – irrespective of the metric chosen, there will be overall more non-linear instances in anesthesia than in the awake state. This does not imply that large amplitude responses will sum 'more' linearly in the awake state. To this reviewer, an adequate comparison to test this possibility is sub-sampling the response distribution under anesthesia and comparing differences in non-linearity between responses of similar amplitudes. Do similar amplitude responses tend to sum more, or less linearly in awake versus anaesthetized states?

I think that the observation that, in response to the same stimuli, at the level of mitral and tufted cells, are kept in the linear regime of the system is insightful and would be important to document. It would enable the field to move forward. In my opinion, one way for the authors to clarify this matter is to show both the normalized and un-normalized data in the manuscript and simply state that in the awake state, responses to the same stimuli tend to be smaller in amplitude than under anesthesia, and, therefore, remain in the linear summation regime (i.e. do not get to reach high amplitudes that would sum up in a more non-linear fashion). This could be due to, for example, to stronger gain control mechanisms in the awake state, etc.

2. The authors added analysis to compare the discriminability between the anesthesia and awake state. While it is reassuring that discriminability is decent even with the smaller amplitude responses in the awake state. I find that the overall interpretation of the results could potentially lead to confusion in the readers' minds. Therefore, I propose to clarify by adding a few explanatory statements, and adjusting some of the statements in the abstract.

If I understand correctly, the authors use a 0.5s pulse of odor and find that discriminability during the first second or so, is comparable (for training an testing on random selection of single odor responses and mixture responses) or better (for training on single odors, testing on mixtures) during anesthesia. In the later phase (>1.5 s), awake responses appear to do better at discriminability. I am not sure how to interpret this finding. Could this simply be related to differences in response amplitudes and temporal dynamics? If so, stating this, would be very useful to bring clarity to the message. From the examples in Figures 1,2,4 it appears that responses during anesthesia peak early and decay back to baseline within 1 second from odor OFF (so, mostly back to baseline at 1.5 s from odor ON), whereas responses during the awake state have longer latencies and peak later. Therefore, the time-course differences in responses might explain the time-course difference in discriminability. It would be important to state that these differences in response dynamics across the two states may be the basis for the time differences in discriminability.

From a behavioral point of view, it is unclear how to interpret these time-courses, discriminability peaks much later than behaviorally relevant time-scales in the awake state (<1.5s for the odor concentration used, 1-5%), and while odor identities can be decoded in anesthesia – animals don't make decisions under KX. When comparing the discriminability between naïve engaged, it would be important to clarify whether these differences arise early, at behaviorally relevant time-scales, or only appear in the later phase of the responses (>1.5 s) from odor onset and adjust the statements in the abstract accordingly. A side by side comparison would clarify this point.

3. Discriminability as a function of linearity of summation. It is unclear what the authors mean by 'the time course of decoding accuracy did not correlate with the linearity of summation'. The authors do not quantify linearity as a function of time, or explicitly analyze the temporal dependence of decoding accuracy with the instantaneous level of non-linearity. To support the above-mentioned sentence, such analysis would be necessary. Additionally, while this is a nice result (that decoding accuracy is not a direct correlate of linearity), it dampens the significance of assessing differences in linearity between various brain states.

Reviewer #3:

The authors have made a good attempt to address concerns raised in the first review. They have reanalyzed the data and used an alternate metric- mean fractional deviation, to evaluate response linearity. Importantly their original result still holds, M/T responses are more linear in the awake state than anesthetized. There is still some issue with stating the importance of this finding. It still reads like a non-finding from a functional perspective. Linearity only seems to be helpful with respect to discrimination in a time frame (>2s) that is after the animal has already started licking. But it is also important to note that despite a reduction in discrimination in early odor responses compared to anesthetized this does present a problem at the behavioral level. Particularly for a simple task such as the one used here. The longer time frame maybe important for more difficult tasks. Also, a potentially underestimated finding is the overall reduction in correlation between M/T cells in the awake state (Figure 5B). Such a reduction in correlation decreases redundancy which may enhance coding (Padmanbhan and Urban, 2010). These points should be emphasized more.

https://doi.org/10.7554/eLife.76882.sa1

Author response

Summary:

The reviewers agreed that this study addresses an important question: how complex odors are processed in the olfactory bulb, more specifically, whether mixture components are represented as a linear sum of components or sublinear interactions occur as suggested by previous studies. The results are potentially interesting, but the reviewers have raised substantive concerns.

1. The reviewers were not convinced by the analysis based on the fractions of sublinear versus linear interactions. The authors report the fraction of non-linear interactions as defined by the activity showing a deviation from linearity using a statistical criterion (> 2 standard deviation). The reviewers thought the authors need to perform more analysis by considering trial-to-trial variability, magnitude difference, and response saturation etc.. Each reviewer made useful suggestions (see below).

2. Odor responses in awake conditions are generally weaker than those under anesthesia. It is, therefore, unclear whether the overall discrimination is better during awake condition. It is important to perform decoding analysis using a classifier trained using odor responses in each condition (not a classifier built using predicted responses).

3. The reviewers had somewhat different opinions on overall significance of the work. One reviewer thought that the difference in awake and anesthetized conditions, if shown convincingly by addressing the above points, is important because previous studies have predominantly reported sublinear interactions. One reviewer, however, thought that the difference between awake and anesthetized conditions is not sufficiently novel, because many other differences have already been reported. One reviewer thought that further mechanistic insights would be important.

Overall, strong demonstrations of the above point 1 and 2, and/or further mechanistic insights would need to be considered for eLife. If the authors can address the above issues, we would be willing to reconsider a resubmitted version but without guarantees concerning publication.

We thank you for your thoughtful and helpful comments on the earlier version of our manuscript. We are pleased to re-submit what we believe to be a substantially improved version, with more robust analyses, and more insights, manifesting in new main figures, and 3 new supplemental figures.

First, we re-analysed the data, taking into account the trial-by-trial variability in the awake data (Distilled Pointed #1 above). We revised how we describe the mixture summation, which is now more robust to differences in noise and amplitudes across different conditions. Specifically, we decided to use median fractional deviation. As the name suggests, it is the median of fractional deviation, which we demonstrate in the revised Figure 4 to be robust to noise. In addition, we no longer use an arbitrary cut-off to classify the types of summation. We still find the mixture summation to be more linear in awake animals in the early phase of the response.

Following the second suggestion (Distilled Point #2), we also carried out (1) an analysis of trial-bytrial variability and (2) analyses using discriminant decoders.

We found, as the reviewers noted, that the trial-by-trial variability is greater in the awake mice (now included in the revised Figure 5). This is reflected in the lower correlation between S+ odour responses (i.e., within-class correlation) in this condition, compared to the anaesthetised case. However, we also observed that the across-class correlation (S+ vs. S- responses) is lower in the awake case i.e., odour patterns across classes are more decorrelated.

Since, as the reviewers noted, it was unclear how these factors affect the ability to detect ethyl butyrate in mixtures, we carried out a decoder analysis. We found that discriminant decoders (SVMs) using data from behaving mice performed equally well, or even better than decoders using the anaesthetised data during the late phase (revised Figure 5).

Interestingly, we also observed that the time course of decoder accuracy does not match the linearity in summation, making it unlikely that the linearity is critically needed for mixture analysis. This is noted in the following place.

Abstract (lines 19-21): “…decoding analyses indicated that the data from behaving mice was able to encode mixture responses well, though the time course of decoding accuracy did not correlate with the linearity of summation”. Result (lines 200 – 203): “Curiously, the time course of decoder accuracy did not correlate with the linearity of mixture summation. Overall, the result suggests that, despite the substantial differences in the amplitudes and variability of responses, M/T cells in behaving mice are able to encode the presence of the target odour well.”

Regarding the data from awake mice with different training and engagement levels, a decoder analysis revealed a state-dependent difference, where data from disengaged mice performed particularly poorly, suggesting that mixture responses may be subject to modulation by behavioural contexts (revised Figure 6). Again, the decoder accuracy did not depend on the linearity of summation.

In terms of significance (Distilled Point #3), we understand the difficulty the reviewers may have faced with the previous version. However, as one reviewer recognises, prior to our investigation, study after study demonstrated widespread mixture suppressions, using anaesthetised animals. It is important for the field to know that this is not the case in awake animals. That is, while our observation indeed falls into the general category of “awake vs. anaesthesia differences”, it is still important for the field to understand how mixtures are represented in the primary olfactory area. Further, currently, whether or not nonlinear summation (“interaction between mixture components”) is detrimental to the analytical perception of odour mixtures is still under debate, with a number of classical studies suggesting a detrimental relationship (Bell et al., 1987; Laing, 1994), while more recent studies suggest otherwise, i.e., antagonism could be beneficial to the analytical perception (Reddy et al., 2018). Our observation that a behavioural improvement is not accompanied by a greater linearisation, and our (new) observation that the decoder performance does not correlate with linearity, should be relevant.

In summary, we believe that the revised manuscript addresses the major concerns and brings insights on how mixtures are represented in the primary olfactory area. We are grateful for your time and for your helpful comments, which have greatly improved our manuscript.

Please find our point-by-point responses below.

Reviewer #1:

Adefuin and colleagues examined the interaction between components of binary odor mixtures in odor responses in mice. The authors used two-photon calcium imaging from the soma and apical dendrites of mitral/tufted cells in the olfactory bulb. Odor responses were measured in various conditions: under anesthesia (ketamine/xylazine), while well-trained mice were engaged in an odor discrimination task, or disengaged. The authors first show that mixture components interacted sublinearly in a large fraction of mitral/tufted cells (46%; Figure 6D) consistent with previous studies. However, when odor responses were measured in awake animals, very few mitral/tufted cells showed sublinear responses at soma (8-9%; Figure 6D). Interestingly, sublinear interaction was evident in apical dendrites of mitral/tufted cells (45%). Whether mixture components are represented linearly or not in the olfactory system is an important question, related to the animal's ability to identify or segment mixture components. Somewhat contrary to previous studies, this study demonstrates largely linear interactions. Furthermore, this study compares various behavioral conditions. These results are important and of interest to those who study sensory systems. I have a few concerns regarding data analysis.

Thank you for your helpful review, and for recognising the relevance our work. We hope that the reviewer finds the our point-by-point responses satisfactory.

1. Non-linear interactions are detected by the activity showing a deviation from linearity greater than 2 standard deviations. Using this criterion, non-linear interactions might decrease if the trial-by-trial activity becomes more variable. This is concerning because the activity might be less variable in the anesthetized condition, and the reduction in sublinear interactions in awake conditions may be due to a general increase in response variability during awake. Can the authors exclude the possibility that the decrease in sublinear interactions is merely due to an increase in response variability in the awake conditions. This issue also applies to the comparison between apical dendrites versus soma; are the signals in apical dendrite less variable (maybe due to some averaging across dendrites from multiple cells; see the following point 5)?

Thank you for raising this valid point and for suggesting alternative analyses. We agree that the index we used previously is susceptible to noise, and not appropriate for comparing two datasets with different trial-by-trial variability. To quantify the deviation from linear sum more robustly, we now use the “Median fractional deviation”, which expresses a deviation from the linear sum as a fraction of predicted, linear sum – not normalised by the standard deviation – and take the median of the distribution from each field of view. As we describe in the revised Figure 4Figure , this measure is more robust to noise. Notably, our finding that mixture summation is generally less sublinear in awake mice still stands for the early phase.

In the revised manuscript, we use the median fractional deviation whenever we compare linearity of summation across different conditions, which includes the comparison of anaesthetised vs. awake, behaving conditions (revised Figure 4), comparison of dendrites vs. somata (revised Figure 4—figure supplement 1), and comparisons of awake states (revised Figure 6). This has given us, too, more confidence about our interpretation, so we are grateful for the reviewer’s suggestions.

2. Related to the above issue, it would be useful to analyze the difference between conditions using different metrics to fully understand what really are different between conditions. The scatter plots shown in various figures do not show drastic differences between awake and anesthetized conditions, as might be indicated by the percent of sublinear responses. It would be useful to characterize the magnitude of sublinear/supralinear effects. For example, one can calculate a fractional change in the mean response. Does this measure show consistent difference between awake and anesthetized conditions?

Thank you for suggesting this analysis. As described in Figure 4 , we now use the fractional deviation to quantify how mixture summations differ from linear sums, which turned out to be a very useful way to express the property of summation (N.B.: noise is amplified for small responses when fractional deviation is used, which is another reason we use the median now). We thank the reviewer for suggesting this analysis.

Reviewer #2:

This study addresses how complex stimuli are represented in neural responses. This is particularly relevant to olfaction because the vast majority of stimuli are complex mixtures that perceptually, are not easy to decompose into parts. Nonetheless, the ability to discern a relevant odor from background odors is essential. This process is easier when neural responses to mixtures reflect the linear sum of the responses to the individual components. The main conclusion of this study is that the linearity of olfactory bulb responses to two-component mixtures increases awake versus anesthetized states. The authors provide some evidence to support this claim. However, this could be better quantified and there is a temporal aspect of linearization that is not addressed. Perhaps the most interesting aspect of the study is the difference in linearity between the dendrites and the somata of the mitral/tufted cells. But a statistical analysis of this finding was not evident. Overall, a mechanistic or functional approach to understanding these findings is lacking. The differences linearity between the anesthetized and awake are simply explained by response saturation anesthetized animals. There are hints at mechanism by which linearity is supported in the OB with comparisons between soma and dendrite, but these are not well developed. There is a model that addresses the functional significance of linearity, but this is only supplemental and not well described.

Thank you for appreciating the significance of our work, and for your constructive comments. Please see below for our point-by-point responses to your concerns.

1) The quantification of linearity is incomplete. The deviation from linearity is a fine metric but just reporting the change in fraction of ROIs that show deviation from linearity in the sublinear direction is insufficient. There are many ROIs that show supralinear summation as well. Moreover, it is not clear where the cut-off is between linear and non-linear responses. There are statistical analyses that can be done on proportional data to determine if the proportions of non-linear ROIs are statistically different in the anesthetized and awake states. These should be done.

Thank you for raising this point. The problem with the way we quantified our data previously was raised by other reviewers, too, and we agree. In the revised manuscript, first, we refrain from using arbitrary cut-offs. Using the median fractional deviation (please see responses to Distilled Point #1 far above for details), we plot cumulative histograms to compare groups, which avoids forcing arbitrary categories (i.e., sublinear, linear, and supralinear). Please see below for the comparison of cumulative histograms (Figure 4).

2) Another approach is to compare the slope of the lines fit to data in the Observed vs. Linear sum plots and to determine how much these deviate from 1, and if this is statistically significant. These slopes could also be compared between states.

Thank you for suggesting the alternative analysis. We hope that our new measure, the fractional deviation, addresses the concern.

3) It is also necessary to be clear about which timepoint is being analyzed. It is apparent from figure 4 that the linearity of the population of ROIs evolves with time, becoming most linear 3 sec after odor onset. This temporal evolution of linearity is not addressed at all. But could provide some insight into mechanism.

Thank you for your suggestion to clarify our presentation. Whenever a single time point is presented, we annotate in the figure which time point the analysis relates to (Figures 4E,F; Figure 5A, Figure 4—figure supplement 1 D).

4) There is no statistical analysis of the dendrite versus soma data. The figure is not very convincing of differences. Why wasn't the temporal progression of responses shown for the dendrite versus soma data?

We apologise for not describing the dendrite data as fully. We now include the time course information for dendrites and somata in a way that is easier to relate the amplitudes to the linearity of summation (Figure 4 – supplement 1). The difference between the dendrites and somata seems to relate to the time course of activation somewhat – in general, Ca2+ rises earlier for the apical dendrites, which seems to manifest in the earlier appearance of sublinearity in this structure. We now also include data from MC somata to further understand the dendritic signal.

5) Linearity seems very correlated with the strength of the response. Sublinearity increases with strength suggesting saturation is a major contributing factor. Conversely, supralinear responses are more apparent with weaker component responses. It is well known that OB responses are much stronger in anesthetized animals. The deltaF/F values are greater in the anesthetized state compared to the awake state. So, the differences between the states may be trivially explained by saturation (as shown in Supplemental Figure 6). This doesn't really say anything about how the OB linearizes responses. However, the temporal evolution of linearization could speak to other mechanisms such as adaptation, recruitment of inhibition, descending control etc. Analyses of the dendrite versus soma signals over time could give insight as to where/when these mechanisms may be occurring. I think this was attempted in the supplement to figure 7, but the figure does not have sufficient explanation to figure out what is going on.

Thank you for the valuable comments. We agree that a deeper mechanistic insight would be useful. Indeed, as the reviewer suggests, the time course can give useful hints. To address this, we include the time course of response amplitudes next to the linearity of summation in our new supplemental figure (Figure 4, supplement 1). There is a rough correspondence between these two factors – as the reviewer points out, in general, the larger the response, the more sublinear the summation. However, the correspondence is not perfect and suggests something else at play. We additionally considered if correlation between component responses (i.e., EB vs. MB response correlation) could explain how linear the summations may be (Figure 4, supplement 1 panel E). This analysis suggests that pattern similarity may contribute, but again, not completely. We suspect that a comprehensive model of olfactory bulb circuits, with perhaps piriform feedback, may be needed to fully understand how the olfactory bulb processes mixture information. While this is somewhat beyond the scope of the current study, we hope to address it in our future investigations. Importantly, however, we hope that the additional data and analyses help.

6) There was a model of how linearity might function in odor detection, but this is not well described or motivated in the text.

Thank you for this helpful comment. We were not sure which model the reviewer was referring to, but we now discuss a recent work by (Reddy et al., 2018) where antagonism can, in principle, bring benefits to odour detection tasks, in addition to discussing the difficulty in mixture analysis due to nonlinear interactions. Thus, our introduction is more balanced, but the fact remains that we need to understand how mixtures are represented in behaviourally relevant context. We hope that the reviewer finds our revised description better. The revised introductory text is as follows (lines 68-70): “Nonlinear summation poses a difficulty because it may distort a pattern of interest brought by non-uniform addition of unpredictable background patterns, although more recent studies suggest beneficial effects of antagonism, for example by reducing saturation-related loss of information (Reddy et al., 2018).”, followed by (lines 75 – 76): “The question therefore remains: how does the mammalian olfactory system deal with nonlinear summation of responses?”

7) Overall, all the supplemental figures and data are essential to the study. Many of my concerns are somewhat addressed by these additional analyses. But they are not well explained. These should be integrated into the main figures of the paper and adequately described in the Results section.

Thank you very much for valuing our analyses that we had presented in the supplement. By addressing many excellent comments from the reviewers, our revised manuscript has been heavily edited. With the changes, we believe the original supplementary figures are either modified substantially, incorporated better, or are no longer needed.

Reviewer #3:

Adefuin et al. use multiphoton imaging of M/T cell responses to investigate whether neuronal representations of binary mixtures can be explained as a sum of the components. The current view in the field (built largely from studies in anesthetized animals), is that mixture summation is non-linear and increases with the degree in glomerular response overlap elicited by the components. The authors reproduce these results and ask whether the same phenomenon is observed in the awake state, in particular when the animals are engaged in an odor discrimination task. Unlike in the anesthetized state, the authors find that mixture representations are linear in the awake brain. They use a series of systematic behavioral paradigms to show that the observed linearity in the awake state (compared to anesthetized) is not dependent on task engagement (reward is given randomly, post-odor) or stimulus relevance (reward is given before odor). While the experiments are well done and the data is presented clearly, I have several major concerns about the interpretation of their results.

1) Given the data the authors present, it is unclear if one can conclude that the olfactory system is more or less linear in the awake state compared to the anaesthetised one. What seems to change most across the awake vs. anesthetized state is the response amplitude. Responses appear to be ~3x smaller in the awake mice. In the anesthetized state, non-linearity seems most apparent for large response amplitudes (>5 dF/F) with mixture responses being sub-linear, most likely due to saturation effects. The authors themselves do an analysis in Figure 6 – supplement 1 to show that most of the observed non-linearity in the anesthetized animals can be explained away after accounting for amplitude normalisation. The authors use this analysis to comment that the level of linearity is the same across all the three awake states, but the same figure shows that it is in fact the same even for the anaesthetized state.

To put it differently, it is indeed true from the authors data that the OB response gain is significantly lower in the awake state, but it is unclear if the summation is more linear if measured at similar response amplitude regimes in both awake and anaesthetised mice.

Thank you for the valuable comments. We agree that many differences between the anaesthetised vs. awake states should have been taken into account when comparing the linearity of summation. We address the reviewer’s concern now by expressing the deviation as a fraction of the predicted, linear sum of component responses. Further, we also considered another factor that could influence the anaesthetised vs. awake comparison, namely, the trial-by-trial variability.

2) The authors argue that keeping response amplitudes small in the awake brain prevents sub-linear summation and therefore may lead to better mixture decomposition. They do a decoding analysis in anaesthetised mice to show that linear mixture representations (instead of using observed sub-linear representations) make odor classification easier. However, I find this analysis uninformative and misleading. It is no surprise that the decoders trained on single odor representations should perform better (or equivalent) when using linear sums as input instead of observed sub-linear representations. The authors use this observation to suggest that this mechanism aids discrimination ability in the awake state. However, given that even the single odor responses are much weaker and noisier in the awake state, it is likely that even the single odor discrimination ability is poorer in the awake state. By the same logic, mixture decomposition might be also much poorer in the awake brain than the anesthetized brain, even though summation is more linear, just because responses are weaker and noisier. In my opinion, the authors should compare decoding accuracy across awake vs. anesthetized responses if they want to assert that linearisation of responses in the awake brain leads to easier decomposition. Because otherwise, while linearisation in principle can aid decomposition, at least in the form that the authors observe here, it may come at a high cost on signal-to-noise ratio which would undo the gain that linearity provides, in principle, for discrimination.

Thank you very much for the insight and for the excellent suggestion to consider the discriminability of stimuli. In particular, we now include an analysis where a decoder trained on single responses is tested on observed mixture responses. Surprisingly, despite the substantial differences in the amplitudes of response and trial-by-trial variability, decoders using data from awake mice performed well, even better than anaesthetised data for the late phase of responses. This is now described in the revised figures (revised Figure 5). We thank the reviewer for the excellent suggestion.

Interestingly, though, the time course of the decoder performance does not correlate well with the linearity of summation. This observation is now described in the abstract (lines 19-21): “…decoding analyses indicated that the data from behaving mice was able to encode mixture responses well, though the time course of decoding accuracy did not correlate with the linearity of summation“.

3) At a more philosophical level, to this Reviewer, it is unclear if anesthesia vs. awake state difference in response should constitute the main focus of the manuscript. The authors explore summation properties under four different brain states, one of which is anaesthesia (also least behaviorally relevant). In three out of four states, they observe that summation is linear. In the fourth (anaesthesia), they observe that summation is sub-linear, but this happens at much larger response amplitude regimes compared to the three awake states sampled, presumably due to saturation. To me, it seems that the Authors here show that mixture summation in the OB, is largely independent of brain state since it is unaffected by whether the animal is task engaged or motivated etc.

Thank you for this thoughtful comment. This has made us reflect on the essence of our study. We believe we make three main observations. First, the anaesthesia vs. awake difference in the property of summation differ, and should be reported, because of the large volume of prior works reporting sublinear summations. However, as the reviewer recommends and as mentioned next, this is no longer the sole focus of our study. Our second observation is that the linearity of summation does not necessarily correlate with the ability to analyse mixtures, based on the decoder performance. We believe it is important to share this observation, since a number of previous studies speculated that nonlinear summation contributes to perceptual difficulty (Bell et al., 1987; Laing, 1994). Third, the decoder performance – especially one that is trained on single odour responses and tested on mixtures – shows differences depending on the awake states, where data from disengaged mice performed particularly poorly. This result is shown in the revised Figure 6Figure . Further, we have edited the abstract and results to ensure that these are clearly communicated. We hope that this is more balanced and reflects the data better.

4) It is unclear how to interpret the dendritic imaging comparison. First, the dendritic signal is pooled across many cells. If any of the cells that are being pooled shows sub-linearity, the pooled population response will look sub-linear, albeit less so than at the single cell level. Second, again like for the anesthetized vs. awake comparison, there is a discrepancy in response amplitudes – dendritic responses are ~2x stronger than the somatic responses and sub-linear summation would be more apparent as one approaches the saturation regime. Third, dendritic responses pool both mitral and tufted, while the somatic data the authors present is predominantly from tufted cells.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Although the manuscript has greatly improved, there are some remaining issues. Specifically, Reviewer 2 and 3 have raised concerns that need to be clarified. In particular, Reviewer 3 has raised the issue regarding the dissociation between discriminability and linear mixture summation. The functional significance of linear mixture summation, or the limitation thereof, needs to be discussed more carefully, and it would be useful to include some discussion on potential causes of increased discriminability in awake conditions. In addition, Reviewer 2, raised several points about clarifying and adjusting statements in the abstract and the text of the manuscript, and analyzing the temporal dependence of decoding accuracy to the instantaneous level of non-linearity in response when comparing across different brain states. We would like you to respond to these remaining issues.

Thank you for recognising the significant improvement with the previous revision.

In this revision, we addressed the remaining concerns by providing additional analyses and text changes. Key to this was to follow Reviewer 1's helpful suggestion, namely, to model the functional consequence of nonlinear summation. Specifically, we simulated the normalising sublinearity on simulated linear sums of mixture responses and assessed how the SVM performance is affected by this. This provided a missing link that now allows us to interpret the discriminability of mixture responses and isolate the functional consequence of non-linear summation more clearly.

Generally, our simulation indicates that the sublinear mixture summation due to saturating influence limits the discriminability of mixture responses. Thus, dampened responses in awake animals bring some advantage by maintaining the mixture responses in the linear range, especially in the early phase. However, when evoked response patterns are decorrelated in the late phase, the saturating sublinearity is less detrimental to discrimination.

The simulation, as well as a graphical summary of our findings, are shown in a new figure (Figure 7).

Further, to explain our findings, particularly in light of our new analyses, we edited the abstract, relevant sections of the result and discussion are also revised concordantly as detailed below.

Lines 17- 27 (abstract): ".… we found that mixture summation is more linear in the early phase of evoked responses in awake, head-fixed mice performing an odour detection task, due to dampened responses. Despite this, and responses being more variable, decoding analyses indicated that the data from behaving mice was well discriminable. Curiously, the time course of decoding accuracy did not correlate strictly with the linearity of summation. Further, a comparison with naïve mice indicated that learning to accurately perform the mixture detection task is not accompanied by more linear mixture summation. Finally, using a simulation, we demonstrate that, while saturating sublinearity tends to degrade the discriminability, the extent of the impairment may depend on other factors, including pattern decorrelation. Altogether, our results demonstrate that the mixture representation in the primary olfactory area is state-dependent, but the analytical perception may not strictly correlate with linearity in summation."

Lines 240- 256 (results): Under what circumstance does discriminability of odour mixtures decouple from linearity in mixture summation? To study the effect of saturating sublinearity in isolation, we made a simulation as follows. Linear sums of responses were constructed by adding component responses obtained from the imaging sessions. In one test, these sums, with added noise, were passed through SVMs that had been trained with component responses. In another test, the linearly summed responses with noise were transformed with a normalising function (Mathis et al., 2016; Penker et al., 2020), which adds sublinearity in an amplitude-dependent manner (Figure 7B; see methods). Then, these signals were passed through the same SVMs to assess how discriminable the activity patterns were. When the data from the anaesthetised mice was used, normalising sublinearity was particularly detrimental around 2 seconds after the odour onset (Figure 7C), qualitatively reproducing the transient decrease in the accuracy seen with the observed mixture responses (Figure 5E). In contrast, with the data from the behaving mice, normalising sublinearity did not have as significant an effect on the mixture discriminability, even at the later stage when sublinear summation becomes more widespread (Figure 7D,E). Thus, the functional consequence of nonlinear summation may depend on specific circumstances, for example on how separable, or decorrelated, the activity patterns are. Overall, our result indicates that mixture responses in the olfactory bulb are highly state-dependent and evolve over time (Figure 6F)".

Lines 272-285 (discussion): " This raises a question: is sublinear summation in mixtures detrimental, or beneficial to discriminating mixtures of odours? Our simple simulation indicates that, generally, a sublinear mixture summation due to saturating influences limits the discriminability of mixture responses. Thus, dampened responses in awake animals bring some advantage by maintaining the mixture responses in the linear range. However, in the behaving mice, even when the responses became larger at a later phase, the saturating influence was less detrimental to discrimination. Here, decorrelated response patterns may play a more crucial role, as it is known to enhance classifier performances (Bhattacharjee et al., 2019; Friedrich and Wiechert, 2014; Gschwend et al., 2015; Padmanabhan and Urban, 2010). While the behavioural task described in this study is simple and animals make accurate decisions within 1 s (771 ± 97 ms) of stimulus onset, the slower mechanisms described here may be important when larger and more reliable responses are required for accurate decisions. So, in addition to sampling time (Rinberg et al., 2006), this phenomenon may be a part of mechanisms needed to solve more difficult tasks accurately (Abraham et al., 2010; Wilson et al., 2017). "

We hope that these additional analysis, simulation, and text changes address the remaining concerns and that the findings are explained more clearly.

Reviewer #1:

The authors have addressed my previous concerns by reanalyzing the data and revising the manuscript. Importantly, the authors now provide a more convincing metric for linearity of mixture interaction using the median fractional change over the population. I am more convinced by this metric which supports that mixture interaction is more linear in awake conditions. This new analysis addresses my most important concern. A new analysis also finds that decoding or behavioral ability to detect a mixture component did not correlate with the linearity of mixture interactions in odor responses. This finding, in a sense, makes the significance of the finding of linearity in odor responses a little unclear. On the other hand, I think that the demonstration that odor decoding does not benefit from linear mixture interaction has an important message to the field as one might assume that it would be the case. It would be more useful if the authors can demonstrate this point (odor decoding does not benefit from linear mixture interaction) in a more abstract form without relying on the data -- can we take this conclusion as a general result, or does it rely on some particular features in the data. A simple modeling might help address this issue. Although such demonstration is not required, if the authors can address this in a short period of time, it would enhance the manuscript significantly, although this point is a little secondary compared to the main point (linearity of odor mixture interaction).

Overall, I am satisfied by the more convincing demonstration of linearity of mixture interaction. Because sublinearity has been emphasized in the previous literature, the results presented in this manuscript provide an important result.

Thank you for noting that our manuscript had improved, and for your valuable inputs. In particular, the suggestion to model the nonlinearity to understand the functional consequence (on discriminability) proved to be key to interpreting the result. Generally, our simulation indicates that the sublinear mixture summation due to saturating influence limits the discriminability of mixture responses, but the extent of impairment may depend on other factors such as decorrelation.

Reviewer #2:

I commend the authors for re-structuring the manuscript and performing additional analyses to address the concerns raised. These changes have indeed improved the manuscript substantially.

However, I still have concerns about the disparity in response amplitude and dynamics across the different states and how that may affect the interpretation of the results. I think these can be addressed by further clarifying and editing the statements in the abstract and rest of the manuscript before publication.

1. The authors use the new metric – Fractional deviation – to compare differences in the degree of linearity between the anesthesia and awake states. While this metric normalizes deviations from linearity with the response amplitude, it does not correct for the overall difference between the means of the two response amplitude distributions – in anesthesia versus awake states. If majority of the responses during the awake state are smaller than during anesthesia, and the amount of non-linearity scales with response amplitude – irrespective of the metric chosen, there will be overall more non-linear instances in anesthesia than in the awake state. This does not imply that large amplitude responses will sum 'more' linearly in the awake state. To this reviewer, an adequate comparison to test this possibility is sub-sampling the response distribution under anesthesia and comparing differences in non-linearity between responses of similar amplitudes. Do similar amplitude responses tend to sum more, or less linearly in awake versus anaesthetized states?

I think that the observation that, in response to the same stimuli, at the level of mitral and tufted cells, are kept in the linear regime of the system is insightful and would be important to document. It would enable the field to move forward. In my opinion, one way for the authors to clarify this matter is to show both the normalized and un-normalized data in the manuscript and simply state that in the awake state, responses to the same stimuli tend to be smaller in amplitude than under anesthesia, and, therefore, remain in the linear summation regime (i.e. do not get to reach high amplitudes that would sum up in a more non-linear fashion). This could be due to, for example, to stronger gain control mechanisms in the awake state, etc.

Thank you for your valuable comments. We apologise that our explanations were not clear enough. Essentially, we believe that dampened responses make the mixture responses in the awake state to remain in the linear range, which, if we understood correctly, agrees with your intuition. We now explain this as a graphical summary (Figure 7E). We hope that this presentation is clearer. Thank you for raising this point.

2. The authors added analysis to compare the discriminability between the anesthesia and awake state. While it is reassuring that discriminability is decent even with the smaller amplitude responses in the awake state. I find that the overall interpretation of the results could potentially lead to confusion in the readers' minds. Therefore, I propose to clarify by adding a few explanatory statements, and adjusting some of the statements in the abstract.

If I understand correctly, the authors use a 0.5s pulse of odor and find that discriminability during the first second or so, is comparable (for training an testing on random selection of single odor responses and mixture responses) or better (for training on single odors, testing on mixtures) during anesthesia. In the later phase (>1.5 s), awake responses appear to do better at discriminability. I am not sure how to interpret this finding. Could this simply be related to differences in response amplitudes and temporal dynamics? If so, stating this, would be very useful to bring clarity to the message. From the examples in Figures 1,2,4 it appears that responses during anesthesia peak early and decay back to baseline within 1 second from odor OFF (so, mostly back to baseline at 1.5 s from odor ON), whereas responses during the awake state have longer latencies and peak later. Therefore, the time-course differences in responses might explain the time-course difference in discriminability. It would be important to state that these differences in response dynamics across the two states may be the basis for the time differences in discriminability.

Thank you for highlighting the potential source of confusion. To address this point, we have edited the abstract in the following manner.

Lines 16-21: " Unlike previous studies using anaesthetised animals, we found that mixture summation is more linear in the early phase of evoked responses in awake, head-fixed mice performing an odour detection task, due to dampened responses. Despite this, and responses being more variable, decoding analyses indicated that the data from behaving mice was well discriminable. Curiously, the time course of decoding accuracy did not correlate strictly with the linearity of summation.".

In addition, we added the following text in the discussion as below.

Lines 275-280: " Thus, dampened responses in awake animals bring some advantage by maintaining the mixture responses in the linear range. However, in the behaving mice, even when the responses became larger at a later phase, the saturating influence was less detrimental to discrimination. Here, decorrelated response patterns may play a more crucial role, as it is known to enhance classifier or behavioural performances (Bhattacharjee et al., 2019; Friedrich and Wiechert, 2014; Gschwend et al., 2015; Padmanabhan and Urban, 2010).".

We hope that the temporal evolution of the responses, their effects on the properties of mixture summation, as well as the functional consequence, are now better explained.

From a behavioral point of view, it is unclear how to interpret these time-courses, discriminability peaks much later than behaviorally relevant time-scales in the awake state (<1.5s for the odor concentration used, 1-5%), and while odor identities can be decoded in anesthesia – animals don't make decisions under KX. When comparing the discriminability between naïve engaged, it would be important to clarify whether these differences arise early, at behaviorally relevant time-scales, or only appear in the later phase of the responses (>1.5 s) from odor onset and adjust the statements in the abstract accordingly. A side by side comparison would clarify this point.

Thank you for this comment. While, indeed, the discriminability peaks later, the SVMs start to perform above chance at an earlier time point. It is well possible that this is enough for the downstream areas to lead to good behavioural performance level. It should be noted that, in the imaging sessions, we sample from a very limited number of the output neurons, whereas the brain has access to all OB outputs. That is, the discriminability could reach a good level even earlier if more informative neurons were included. To clarify this timing point, we added the following text in the result section:

Lines 192-195: ".…M/T cells from the two conditions performed similarly for the first 1 second, performing above chance soon after the odour onset (earliest time where accuracy is significantly above 0.5 = 0.67 s for behaving case, and 1.12 s for anaesthetised data; t-test at significance level of 0.05; n = 13 fields of view, 6 mice for behaving case; 8 fields of view, 4 mice for the anaesthetised case). ".

3. Discriminability as a function of linearity of summation. It is unclear what the authors mean by 'the time course of decoding accuracy did not correlate with the linearity of summation'. The authors do not quantify linearity as a function of time, or explicitly analyze the temporal dependence of decoding accuracy with the instantaneous level of non-linearity. To support the above-mentioned sentence, such analysis would be necessary. Additionally, while this is a nice result (that decoding accuracy is not a direct correlate of linearity), it dampens the significance of assessing differences in linearity between various brain states.

Thank you for this suggestion. We have added a supplementary figure (Figure 6—figure supplement 2) to address this. Specifically, we plotted the instantaneous relationship between the SVM performance and the linearity of summation to direct comparison .

Reviewer #3:

The authors have made a good attempt to address concerns raised in the first review. They have reanalyzed the data and used an alternate metric- mean fractional deviation, to evaluate response linearity. Importantly their original result still holds, M/T responses are more linear in the awake state than anesthetized. There is still some issue with stating the importance of this finding. It still reads like a non-finding from a functional perspective. Linearity only seems to be helpful with respect to discrimination in a time frame (>2s) that is after the animal has already started licking. But it is also important to note that despite a reduction in discrimination in early odor responses compared to anesthetized this does present a problem at the behavioral level. Particularly for a simple task such as the one used here. The longer time frame maybe important for more difficult tasks. Also, a potentially underestimated finding is the overall reduction in correlation between M/T cells in the awake state (Figure 5B). Such a reduction in correlation decreases redundancy which may enhance coding (Padmanbhan and Urban, 2010). These points should be emphasized more.

Thank you for noting the improvement. Regarding the main message, we agree that it needs clarification. We were able to clarify, with simulation, that saturating sublinearity is generally detrimental, but at a later phase, when sublinear summation occurs in conjunction with other phenomena like decorrelation, it may not affect the discriminability as much. Due to the larger responses at late time points, the benefit of linear summation seems to be most relevant in the early phase (t < 1.5s), which is when patterns are less decorrelated. This is now described directly in the new Figure 7.

We are grateful for your acute observation that pattern decorrelation may be a contributing factor. We have edited the abstract and the discussion to reflect this, adding the reference you cited above. In addition, we have also edited the discussion to incorporate the potential relevance of late onset mechanisms (i.e., decorrelation and peak in the discriminability).

Text additions/changes:

Lines 21- 25 (abstract): "Finally, using a simulation, we demonstrate that, while saturating sublinearity tends to degrade the discriminability, the extent of the impairment may depend on other factors, including pattern decorrelation."

Lines 240- 156 (results): Under what circumstance does discriminability of odour mixtures decouple from linearity in mixture summation? To study the effect of saturating sublinearity in isolation, we made a simulation as follows. Linear sums of responses were constructed by adding component responses obtained from the imaging sessions. In one test, these sums, with added noise, were passed through SVMs that had been trained with component responses. In another test, the linearly summed responses with noise were transformed with a normalising function (Mathis et al., 2016; Penker et al., 2020), which adds sublinearity in an amplitude-dependent manner (Figure 7B; see methods). Then, these signals were passed through the same SVMs to assess how discriminable the activity patterns were. When the data from the anaesthetised mice was used, normalising sublinearity was particularly detrimental around 2 seconds after the odour onset (Figure 7C), qualitatively reproducing the transient decrease in the accuracy seen with the observed mixture responses (Figure 5E). In contrast, with the data from the behaving mice, normalising sublinearity did not have as significant an effect on the mixture discriminability, even at the later stage when sublinear summation becomes more widespread (Figure 7D,E). Thus, the functional consequence of nonlinear summation may depend on specific circumstances, for example on how separable, or decorrelated, the activity patterns are. Overall, our result indicates that mixture responses in the olfactory bulb are highly state-dependent and evolve over time (Figure 6F). ".

Lines 272-285 (discussion): " This raises a question: is sublinear summation in mixtures detrimental, or beneficial? Our simple simulation indicates that, generally, a sublinear mixture summation due to normalising influences limits the discriminability of mixture responses. Thus, dampened responses in awake animals bring some advantage by maintaining the mixture responses in the linear range. However, in the behaving mice, even when the responses became larger at a later phase, the normalising effect was less detrimental to discrimination. Here, decorrelated response patterns may play a more crucial role, as it is known to enhance classifier performances (Friedrich and Wiechert, 2014; Gschwend et al., 2015; Padmanabhan and Urban, 2010). While the behavioural task described in this study is simple and animals make accurate decisions within 1 s (771 ± 97 ms) of stimulus onset, the slower mechanisms described here may be important when larger and more reliable responses are required for accurate decisions. So, in addition to sampling time (Rinberg et al., 2006), this phenomenon may be a part of mechanisms needed to solve more difficult tasks accurately (Abraham et al., 2010; Wilson et al., 2017). "

We hope that the main message is now clear.

https://doi.org/10.7554/eLife.76882.sa2

Article and author information

Author details

  1. Aliya Mari Adefuin

    Sensory and Behavioural Neuroscience Unit, Okinawa Institute of Science and Technology Graduate University, Okinawa, Japan
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Writing – review and editing
    Contributed equally with
    Sander Lindeman and Janine Kristin Reinert
    Competing interests
    No competing interests declared
  2. Sander Lindeman

    Sensory and Behavioural Neuroscience Unit, Okinawa Institute of Science and Technology Graduate University, Okinawa, Japan
    Contribution
    Investigation, Supervision, Writing – review and editing
    Contributed equally with
    Aliya Mari Adefuin and Janine Kristin Reinert
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2965-744X
  3. Janine Kristin Reinert

    Sensory and Behavioural Neuroscience Unit, Okinawa Institute of Science and Technology Graduate University, Okinawa, Japan
    Contribution
    Investigation, Supervision, Writing – review and editing
    Contributed equally with
    Aliya Mari Adefuin and Sander Lindeman
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1495-7431
  4. Izumi Fukunaga

    Sensory and Behavioural Neuroscience Unit, Okinawa Institute of Science and Technology Graduate University, Okinawa, Japan
    Contribution
    Conceptualization, Formal analysis, Funding acquisition, Investigation, Supervision, Writing - original draft
    For correspondence
    izumi.fukunaga@oist.jp
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1860-5377

Funding

Okinawa Institute of Science and Technology Graduate University

  • Aliya Mari Adefuin
  • Sander Lindeman
  • Janine K Reinert
  • Izumi Fukunaga

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

Acknowledgements

We thank Yu-Pei Huang and OIST’s Animal Resource Service staff for their dedicated assistance, the members of Andreas Schaefer’s group and Kevin Franks’ group for comments on earlier versions of the work, and Adam Mago, Xiaochen Fu, and Josefine Reuschenbach for comments on the manuscript. This work was supported by the OIST Graduate University.

Ethics

All procedures described in this study have been approved by the OIST Graduate University's Animal Care and Use Committee (Protocol 2016-151 and 2020-310).

Senior Editor

  1. Gary L Westbrook, Oregon Health and Science University, United States

Reviewing Editor

  1. Naoshige Uchida, Harvard University, United States

Reviewers

  1. Naoshige Uchida, Harvard University, United States
  2. Florin Albeanu

Publication history

  1. Preprint posted: September 24, 2021 (view preprint)
  2. Received: January 7, 2022
  3. Accepted: March 5, 2022
  4. Accepted Manuscript published: March 7, 2022 (version 1)
  5. Version of Record published: March 21, 2022 (version 2)

Copyright

© 2022, Adefuin 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

  • 1,218
    Page views
  • 172
    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)

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)

  1. Aliya Mari Adefuin
  2. Sander Lindeman
  3. Janine Kristin Reinert
  4. Izumi Fukunaga
(2022)
State-dependent representations of mixtures by the olfactory bulb
eLife 11:e76882.
https://doi.org/10.7554/eLife.76882

Further reading

    1. Neuroscience
    Movitz Lenninger, Mikael Skoglund ... Arvind Kumar
    Research Article Updated

    According to the efficient coding hypothesis, sensory neurons are adapted to provide maximal information about the environment, given some biophysical constraints. In early visual areas, stimulus-induced modulations of neural activity (or tunings) are predominantly single-peaked. However, periodic tuning, as exhibited by grid cells, has been linked to a significant increase in decoding performance. Does this imply that the tuning curves in early visual areas are sub-optimal? We argue that the time scale at which neurons encode information is imperative to understand the advantages of single-peaked and periodic tuning curves, respectively. Here, we show that the possibility of catastrophic (large) errors creates a trade-off between decoding time and decoding ability. We investigate how decoding time and stimulus dimensionality affect the optimal shape of tuning curves for removing catastrophic errors. In particular, we focus on the spatial periods of the tuning curves for a class of circular tuning curves. We show an overall trend for minimal decoding time to increase with increasing Fisher information, implying a trade-off between accuracy and speed. This trade-off is reinforced whenever the stimulus dimensionality is high, or there is ongoing activity. Thus, given constraints on processing speed, we present normative arguments for the existence of the single-peaked tuning organization observed in early visual areas.

    1. Genetics and Genomics
    2. Neuroscience
    Timothy Wu, Jennifer M Deger ... Joshua M Shulman
    Research Article Updated

    Aging is a major risk factor for Alzheimer’s disease (AD), and cell-type vulnerability underlies its characteristic clinical manifestations. We have performed longitudinal, single-cell RNA-sequencing in Drosophila with pan-neuronal expression of human tau, which forms AD neurofibrillary tangle pathology. Whereas tau- and aging-induced gene expression strongly overlap (93%), they differ in the affected cell types. In contrast to the broad impact of aging, tau-triggered changes are strongly polarized to excitatory neurons and glia. Further, tau can either activate or suppress innate immune gene expression signatures in a cell-type-specific manner. Integration of cellular abundance and gene expression pinpoints nuclear factor kappa B signaling in neurons as a marker for cellular vulnerability. We also highlight the conservation of cell-type-specific transcriptional patterns between Drosophila and human postmortem brain tissue. Overall, our results create a resource for dissection of dynamic, age-dependent gene expression changes at cellular resolution in a genetically tractable model of tauopathy.