Research Article

Widespread receptor-driven modulation in peripheral olfactory coding

See allHide authors and affiliations

Science  10 Apr 2020:
Vol. 368, Issue 6487, eaaz5390
DOI: 10.1126/science.aaz5390

The perception of complex scents

It is generally assumed that olfactory receptors faithfully report information to the brain in the form of a linear, additive code. However, under realistic conditions, the olfactory system handles a far more complex input, usually mixtures of odors. Xu et al. found that when we smell scents, the nasal olfactory sensory neurons relay a more complex pattern of signals to the brain than previously thought. The responses of individual neurons within the peripheral olfactory epithelium were either amplified or attenuated by the presence of other odors, which could explain the common perception of one odor in a mixture dominating over others. This effect occurs within the peripheral sensory organ's receptors and not within the brain.

Science, this issue p. eaaz5390

Structured Abstract


The mammalian nose is arguably the best chemical sensor on the planet, able to detect and discriminate among a large and diverse repertoire of mostly small, organic molecules. It accomplishes this, at least in part, through a large family of G protein–coupled receptors (GPCRs) expressed in specialized olfactory sensory neurons (OSNs) arrayed over an epithelium deep within the nasal cavity. Each neuron expresses only one of the ~1000 receptor genes. It is thought that the specific activation of subsets of these receptors by a particular odor translates into a code that can be read by higher brain centers to create a perception. However, we rarely encounter pure odors. Our daily life is a stream of encounters with rich blends of odors, from garbage to cologne. Even a simple cup of coffee has >800 volatile components. To study how the olfactory system encodes this much more complex information, we explored how neurons within the peripheral olfactory epithelium of a mouse’s nose responded to a series of odor mixtures. Our analysis was enabled by a new high-speed three-dimensional (3D) imaging method called SCAPE (Swept Confocally Aligned Planar Excitation) microscopy, which allows the responses of thousands of single neurons within the intact olfactory epithelium to be monitored in parallel during delivery of repeated odor combinations.


Previous studies using single, monomolecular odors suggested that the diverse expression of receptors in OSNs could provide unambiguous representations of individual odors. However, it is unclear how the brain might be able to decode signals when multiple odor components within a blend generate overlapping patterns. Moreover, when smelling a mixture of odors, it is common to perceive one odor dominating another. Psychophysical tests have revealed both suppressive and enhancing effects of particular odors within a blend. However, it has long been assumed that such odor-coding interactions occur at higher processing levels within the brain. Here, we studied whether combinatorial effects of odors affect neural representations at the peripheral sensory level.


Using SCAPE microscopy to image calcium-sensitive fluorescent proteins in OSNs, we were able to simultaneously monitor the activity of cells within a large volume of the intact mouse olfactory epithelium with single-cell resolution. Analyzing the responses of thousands of single neurons to blends of up to three odors, we discovered a series of surprising interactions that distorted the representation of the odor mixture compared with a simple combinatorial sum of responses to individual odors. Among the eight chemically distinct odors tested, we observed that the presence of one odor could either enhance or suppress the response of a neuron to another odor, even if the modulating odor by itself did not elicit a response from the neuron. This means that an odorless molecule could alter the perception of another odor, and that a neuron’s response to an odor blend can be much larger or smaller than its response to components of the blend. Overall, we observed clear evidence of agonism, antagonism, partial agonism, and enhancement occurring at the receptor level, suggesting a richer repertoire of receptor modulation mechanisms than previously thought. Finally, we note that enhancement of responses may be evidence of an allosteric modulatory site, a rare finding in class A GPCRs that bind small molecules.


Although inhibition and enhancement are well established in sensory systems, they are only a feature of higher circuit processing. Here, we observed complex receptor modulation at the level of peripheral olfactory sensory receptors. We propose that these peripheral modulatory interactions are crucial for discriminating complex blends of odors with overlapping activation patterns because they prevent the saturation of receptors and allow each new component to alter the overall activation pattern, rendering it distinctive. This result suggests that higher brain regions may rely on pattern recognition rather than on reading an additive combinatorial code to build a perception. This work also demonstrates an exciting and versatile new paradigm for high-throughput characterization of single-cell responses in intact systems.

Large-scale single-cell recording of OSNs reveals receptor-driven modulation effects.

Volumetric imaging of GCaMP in intact olfactory epithelium using high-speed SCAPE microscopy enables analysis of responses to mixtures of different odor molecules. Single-neuron response time courses show that odor mixtures can enhance (red) or suppress (green) responses compared with individual odors. Heatmap shows assessment of >10,000 individual neurons across five mice.


Olfactory responses to single odors have been well characterized but in reality we are continually presented with complex mixtures of odors. We performed high-throughput analysis of single-cell responses to odor blends using Swept Confocally Aligned Planar Excitation (SCAPE) microscopy of intact mouse olfactory epithelium, imaging ~10,000 olfactory sensory neurons in parallel. In large numbers of responding cells, mixtures of odors did not elicit a simple sum of the responses to individual components of the blend. Instead, many neurons exhibited either antagonism or enhancement of their response in the presence of another odor. All eight odors tested acted as both agonists and antagonists at different receptors. We propose that this peripheral modulation of responses increases the capacity of the olfactory system to distinguish complex odor mixtures.

The large number of individual olfactory receptors present within the mammalian nose is critical to our ability to detect and distinguish among a vast number and diversity of odors (15). It has generally been assumed that these receptors faithfully report information to the brain in the form of a combinatorial, additive code. For monomolecular odor stimuli, it is indeed possible to identify subsets of receptors that respond to particular odors, generating a code that could serve to unambiguously discriminate one odor from another (611). However, under realistic conditions, the olfactory system must handle a far more complex input consisting of diverse mixtures of odors. It is common for our perception of a mixture of odors to be distorted by, for example, heightening the effect of one or more odor components in favor of another (1216), an effect that is contrary to the idea of simple combinatorial summation. In other sensory systems, inhibition and enhancement of signals is commonly performed by higher-level circuits. However, the neural locus of these perceptible suppression or enhancement effects in the olfactory system remains largely unknown.

In the current study, we characterized the peripheral olfactory system’s response to odor blends using Swept Confocally Aligned Planar Excitation (SCAPE) microscopy, a high-speed, single-objective, light-sheet microscopy technique. SCAPE microscopy enabled high-throughput dynamic volumetric imaging of thousands of single-cell calcium responses to sequences of odor blends in olfactory sensory neurons (OSNs) within an intact mouse olfactory epithelium preparation (1719).


Figure 1A illustrates our hemihead mouse brain preparation and the inverted geometry of the SCAPE microscope used to image the intact olfactory epithelium (for details, see the materials and methods). The mice expressed the genetically encoded Ca2+ indicator GCaMP6f under the mature OSN-specific OMP promoter (OMP-GCaMP6f). This preparation enabled a variety of odor stimuli to be perfused over the intact tissue while monitoring the simultaneous responses of thousands of olfactory neurons layered within the epithelium of the curved turbinates using SCAPE microscopy (Fig. 1B). 3D SCAPE images were acquired at between 2 and 5 volumes per second during a >1-h perfusion with >20 applications of different odor stimuli. Movie S1 shows an example of dynamic neural responses to an odor stimulus. Postprocessing of the resulting four-dimensional (4D) datasets extracted the response profiles of each individual cell to the sequence of odors delivered (for full details, see the materials and methods).

Fig. 1 Imaging of intact olfactory epithelium using SCAPE microscopy.

(A) Schematic of an intact olfactory epithelium imaging platform for SCAPE microscopy. A custom-designed glass-bottomed perfusion chamber was placed above the inverted objective with water immersion. The right half of the mouse head was mounted in the perfusion chamber with the olfactory turbinates exposed. The perfusion chamber was designed to control the perfusion flow through the nasal cavity with the inlet at the nostril and the outlet at the throat (blue arrows). The imaging area typically covered the ventral half of either turbinate IIb or III and some of the neighboring turbinates (yellow rectangle). (B) Three-dimensional volumetric rendering of SCAPE data acquired from the olfactory epithelium at the resting level without odor stimulus showing a 1600 × 1200 × 350 μm field of view. A zoom-in side view (Y-Z, orange box) and a top-down view (X-Y, yellow box) are shown on the right. Both views are the maximum-intensity projections of 10-μm substacks. Scale bar, 50 μm.

Because mature OSNs express only one allele of an odorant receptor gene (20), the response of a single neuron to a sequence of presented odor blends represents the specific properties of a particular receptor. Thus, our method permits characterization of the properties of a large population of receptors without requiring knowledge of the genetic identity of each individual receptor.

Application of forskolin (50 μM) at the end of each imaging session was used to uniformly excite all OSNs. We could thus test how many responsive neurons were present in the field of view and ensure that every odor-specific neuronal response came from a viable OSN. Image segmentation of OSNs coexpressing nuclear-localized tdTomato confirmed that we could simultaneously observe a range of 6000 to 12,000 forskolin-responsive, healthy OSNs from the imaging session of each mouse.

Response to odor set 1

The first odor set (Fig. 2A) contained acetophenone, benzyl acetate, and citral, all at 100 μM. These odors have different nuances: Acetophenone is often described as having an almond or mimosa scent, benzyl acetate as a floral and/or jasmine scent, and citral as a citrus scent. These components were chosen because they are chemically varied and perceptually distinct (at least to humans) and because, when combined, they should activate a large number of cells for analyses. These odor stimuli were pseudorandomly presented as the total mixture, each component singly and in the three possible binary pairs.

Fig. 2 GCaMP time courses of individual OSNs extracted from the raw SCAPE image.

(A) Chemical structures of the three odors used as odor stimuli (odor set 1). (B) OSN responses to acetophenone (ACE, 100 μM) alone and to the three-odor mixture (MIX, each at 100 μM). Each image was cropped from a 1000 × 500 × 200 μm volume taken at peak responses. Two OSNs are highlighted. Scale bar, 20 μm. (C and C′) Time courses of the highlighted OSNs [yellow and orange boxes in (B)]. Data were extracted directly from the raw volumetric time series and calculated as ΔF/F. A 30-s-long odor stimulus (MIX or each odor alone) was delivered in each trial, with a 2.5-min interval between stimulus applications.

Single-cell analysis revealed a diversity of complex responses to the odors and the mixture. Figure 2, B and C, shows the response profiles of two cells to presentation of odor set 1. The cell highlighted in Fig. 2C (Fig. 2B, yellow box) exhibited a preferential response to acetophenone, with much smaller responses to benzyl acetate and citral. This cell’s response to a mixture of all three of these odors at the same concentration was slightly higher than the acetophenone response, as might be expected for a linear summation. However, this linear relationship does not hold for the cell shown in Fig. 2C′ (Fig. 2B, orange box). Here, the cell exhibited a lower-amplitude response to a mixture containing acetophenone than it did to acetophenone alone at the same concentration.

To assess the prevalence of this effect, Fig. 3A shows a summary of 11,936 OSN responses in a heatmap format, corresponding to all cells found to respond to at least one of the presented odors or combinations in odor set 1 across five mice. For this heatmap visualization, peak responses were calculated for each odor, and all responses for a given cell were normalized to the cell’s maximal response across all odors. Each row represents the response of a single cell and each column represents an odor-stimulus condition. Neurons were sorted into subgroups on the basis of their response patterns across odor combinations using k-means clustering, and then combined into eight major subgroups on the basis of their response patterns to the three individual odors (see the materials and methods). From top to bottom, these eight subgroups correspond to: cells activated by the three-odor mixture but only minimally activated by an individual odor (subgroup I), cells dominantly activated by one of the three individual odors (subgroups II to IV), cells dominantly activated by a binary pair of two of the three individual odors (subgroups V to VII), and cells activated by all three individual odors (subgroup VIII). Response time courses of individual neurons in each subgroup are shown in fig. S1.

Fig. 3 Response profile of odor set 1.

(A) Heatmap of normalized peak responses (N = 11,936, 5 mice) to 100 μM concentrations of ACE, BEA, CIT, or their mixtures. Odor stimuli (columns) were given in a pseudorandom manner for each mouse and realigned for this presentation as denoted by the colored squares at the top. OSNs (rows) were clustered into eight subgroups (I to VIII) on the basis of k-means clustering. (B) Time courses of individual OSNs in subgroups I and II showing inhibition and suppression (cells I and ii; primarily suppressed by citral), no effect (cell iii), and enhancement (cells iv and v). The final peak shows the response to 50 μM forskolin (FORK) as a control for all viable OSNs. (C) Quantification of the modulation effects. For suppression, Imod was calculated as (d0d1)/d1, where d0 is the corrected response to the mixture and d1 is the dominant single-odor response. When calculating enhancement, the formula was modified to Imod = max{[(d0d2d3) – d1]/d1, 0}, where d2 and d3 are the responses to nondominant odors. When [(d0d2d3) – d1]/d1 resulted in a negative value, the Imod value was set to zero to avoid misinterpretation. Note that this method may underestimate the effect of enhancement. (D) Representations of the modulation effects on ACE-, BEA-, and CIT-dominant neurons (subgroups II, III, and IV, respectively). Red indicates that cell responses to the individual odors were inhibited by the mixture and blue indicates enhancement.

Grouping of cells that are dominantly activated by one of the three individual odors (Fig. 3A, subgroups II to IV) permits straightforward comparison between responses to that individual odor and responses to the mixture containing that odor. Many of the cells were unaffected by the other components of the mixture, and their response to the mixture was a nearly linear summation of the three individual responses. However, as indicated in Fig. 2 and seen clearly in Fig. 3, there were numerous and unexpected exceptions. In subgroup II, there was a group of cells that dominantly responded to acetophenone, but whose responses to the three-odor mixture were suppressed or completely inhibited (see example time courses in Fig. 3B, cells i and ii). Conversely, many responses were seen to be enhanced by the presence of one of the other odors; for example, in cell iv, in which benzyl acetate paired with acetophenone strongly enhanced the response to acetophenone despite the cell having only a minimal response to benzyl acetate alone. Cell v from subgroup I exhibited responses to the three-odor mixture yet did not respond to any of the three odors individually. Suppression and enhancement were also observed in the benzyl acetate– and citral-dominant subgroups (III and IV). To quantify these effects, a modulation index (Imod; Fig. 3C) was calculated to represent the relative response of each cell to the three-odor mixture compared with its dominant single-odor response. An Imod value of –1 represents complete suppression, i.e., that a single-odor responding neuron has no response to the three-odor mixture. An Imod > 1 represents more than twofold enhancement of the response to the three-odor mixture compared with the dominant single-odor response (for full details, see the materials and methods).

The distributions of these modulation effects for each of the single-odor dominant sets of neurons (subgroups II to IV) are plotted in Fig. 3D. Most of the cells showed no modulation by the other odors in the mixture. However, for acetophenone-, benzyl acetate–, and citral-dominant cells, we found that 19, 3, and 3% of OSNs, respectively, had an Imod value < –0.3 (>30% suppression), and 20, 23, and 27% of OSNs had an Imod value > 0.3 (>30% enhancement), respectively. Suppression effects were considerably stronger in acetophenone-dominant cells, with >6 times the proportion (19 versus 3%) of acetophenone-dominant neurons being suppressed than those of benzyl acetate or citral. Conversely, benzyl acetate– and citral-dominant cells were more likely to undergo enhancement.

Dose-response analysis of inhibition

To gain a deeper understanding of the mechanism of the observed modulation effects, we undertook a series of dose-response experiments with acetophenone and citral. We found 11,774 OSNs across five mice that were activated by either component in this binary odor pair (fig. S3A). To explore suppression effects, we focused our analysis on a subpopulation of 2620 cells that were activated by acetophenone or the mixture of acetophenone and citral but showed little or no response to citral alone. Thus, acetophenone was used as the agonist and citral as the modulator. After k-means clustering and data sorting (see the materials and methods), we found 410 OSNs (16%) that exhibited a suppressed (Imod < –0.3) or completely inhibited response to acetophenone in the presence of 100 μM citral (Fig. 4, A and B).

Fig. 4 Dose-dependent suppression or enhancement of acetophenone by 100 μM citral.

(A) Normalized response heatmap of acetophenone-activated neurons suppressed by citral (N = 410). Neurons were stimulated by an increasing concentration of ACE (10 to 300 μM) in the presence or absence of 100 μM CIT, as denoted by colored squares at the top. (B) Time course of an individual OSN suppressed by 100 μM citral. (C) Effect of 100 μM citral as an antagonist. Dose-dependent responses (mean ± SEM) were plotted and fitted with the Hill equation. Responses to acetophenone alone are plotted in black (Hill coefficient = 1.34, EC50 = 23.0 μM); responses to acetophenone + citral are plotted in red (Hill coefficient = 1.99, EC50 = 101.2 μM). (D) Normalized response heatmap of acetophenone-activated neurons enhanced by citral (N = 301). (E) Time course of an individual OSN showing enhancement. (F) Effect of 100 μM citral as an enhancer. Only the 119 OSNs with no baseline response to 100 μM citral were used to plot the dose-dependent curves (mean ± SEM). Responses to acetophenone alone are plotted in black and responses to acetophenone + citral are plotted in red. Acetophenone alone: Hill coefficient = 1.95, EC50 = 125.1 μM; acetophenone + citral: Hill coefficient = 0.87, EC50 = 44.1 μM.

The averaged responses of all 410 OSNs with and without citral are plotted against acetophenone concentration in Fig. 4C. This dose-response curve shows that the suppression effect of citral could be overcome at higher concentrations of acetophenone, consistent with a competitive antagonism effect. In a few cases (28/410), citral alone activated an OSN but nevertheless suppressed the neuron’s response to acetophenone (fig. S3B). This response pattern fits the standard model of partial agonism.

Dose dependence of enhancement

Supra-additive enhanced responses are more difficult to explain mechanistically than antagonism. An enhanced response to an agonist is generally ascribed to an allosteric mechanism (21). However, allosteric modulation is rarely seen in classic small-molecule class A G protein–coupled receptors (GPCRs). Within our acetophenone versus citral dose-response experiments, a subpopulation of 2620 cells that showed little or no response to citral alone included 301 OSNs with responses to acetophenone that were enhanced by 100 μM citral (Imod > 0.3; Fig. 4D). In most of these cases, a low concentration of acetophenone was insufficient to activate these OSNs unless it was mixed with 100 μM citral, whereas high concentrations of acetophenone (>100 μM) alone were sufficient to activate these OSNs (Fig. 4, D and E).

Among these neurons, 182 showed a small response to 100 μM citral; therefore, the enhancement could be a synergistic effect of acetophenone and citral together, or acetophenone may enhance citral. However, the remaining 119 OSNs did not show a response to 100 μM citral and were used to plot the dose-response curves in Fig. 4F. This plot shows a clear shift to the left (enhancement) in the presence of citral. Because these cells were not activated by citral alone acting at the orthosteric site, the mechanism of enhancement is most likely an allosteric effect of citral on this subset of receptors.

Some neurons showed no response to acetophenone alone, even at 300 μM; however, the mixture of 100 μM citral with lower concentrations of acetophenone could elicit activity (one such example is shown in fig. S3C). This argues against the enhancement resulting just from increased ligand concentration at the orthosteric binding site.

This effect may also explain the subset of cells seen in Fig. 3A, subgroup I. In this subgroup, cells had no or minimal responses to any of the individual odors (all at 100 μM concentrations) but did generate a response to the three-odor mixture, further indicating an allosteric function for one or more of the odors in the blend.

Although small-molecule allosteric modulation of class A GPCRs has been observed only rarely (2227), the wide diversity of the OSN receptor family appears to have extended the occurrence of this mechanism in GPCRs of this type. We are unable to determine the allosteric site from these data. However, the odor molecules are relatively hydrophobic and, from a chemical perspective, could easily access the lipid membrane and bind to sites within the transmembrane regions of the receptor or alter the lipid membrane environment.

Responses to odor set 2

To gain an appreciation for the importance of receptor modulation in natural or composed blends of odors, we performed a similar analysis using a three-odor blend known as “woody accord,” which was designed by perfumers to impart a pleasant and harmonious woody scent to human perceivers. This specific formula contains 148 μM dorisyl, 48 μM dartanol, and 127 μM isoraldeine (fig. S4A).

Applying the same analytic strategy that we used in odor set 1, we found 1303 odor-activated OSNs over three mice (fig. S4B). Within the dorisyl-, dartanol-, and isoraldeine-dominant subgroups (I to III), we found 15, 4, and 33% of neurons showing suppression (Imod < –0.3), and 9, 5, and 3% of neurons showing enhancement (Imod > 0.3), respectively. A large proportion of dorisyl-dominant neurons exhibited suppression (fig. S4B, black box). OSN responses to the binary mixtures of the odor set further indicated that isoraldeine was the major inhibitor rather than dartanol. The Imod distributions for these subsets are summarized in fig. S4D.

Repeating dose-response measurements with the dorisyl-isoraldeine odor pair yielded 3396 responding neurons over three mice, among which 1692 cells were activated by dorisyl or the mixture of dorisyl and isoraldeine and showed little or no response to isoraldeine alone. We observed 306 of these cells exhibiting dose-dependent inhibition (Imod < –0.3), in which dorisyl was the agonist and isoraldeine the antagonist (fig. S5, A to C). We also observed 158 cells exhibiting enhancement (Imod > 0.3) (fig. S5, D to F). These enhancement effects are consistent with the interactions between acetophenone and citral observed above (Fig. 4, D to F), reinforcing the possibility that allosteric modulations could occur between different odor pairs.

Additional modulators of acetophenone responses

We do not expect that the modulation effects observed here are only found in the specific molecules chosen. All odors could feasibly act as an agonist, antagonist, or enhancer at different receptors depending on what other molecules may be present. To test this hypothesis, we selected four different odors (dartanol, isoraldeine, γ-terpinene, and isoamyl acetate; see Fig. 5A) and paired them with acetophenone. We identified 6178 cells across three mice that were activated by at least one of these odors or their binary mixtures (fig. S6). Among these OSNs, 1309 were activated by acetophenone. After k-means clustering and cell sorting on the basis of different response patterns to different odor stimuli, 80 of these cells (6%) were identified as showing response suppression (Imod < –0.3) by one or more of the other compounds. These neuron responses are plotted as a heatmap in Fig. 5B, with suppression effects highlighted in boxes. Time courses of individual OSNs showing suppression are shown in Fig. 5C. These OSNs showed diversified response patterns to the four individual compounds, indicating that a given odor likely modulated the activity of different receptors. For example, cells i and ii were both inhibited by isoraldeine but showed differential responses to dorisyl. In rarer cases, cells were inhibited by more than one odor (Fig. 5C, cell iii) or even suppressed by all four odors equivalently (Fig. 5C, cell iv).

Fig. 5 Acetophenone responses can be modulated by multiple odors.

(A) Chemical structures of acetophenone and odors tested as modulators. (B) Normalized response heatmap of suppressed OSNs (N = 80). OSNs were first stimulated with 100 μM dorisyl, isoraldeine, γ-terpinene, or isoamyl acetate individually. Each of the odors was then tested against 30 μM ACE. Odor stimuli are denoted by colored squares at the top [color coding as in (A)]. The columns of the heatmap were reordered for easier visualization of the suppression effects. Neurons showing suppression effects are boxed in different colors corresponding to the different modulators. (C) Time courses of four OSNs showing suppression. Responses to 30 μM acetophenone alone are highlighted by pink rectangles; arrows with different colors indicate suppression and/or inhibition by the corresponding odors. (D) Normalized response heatmap of enhanced OSNs (N = 73). The columns of the heatmap were reordered for easier visualization of the enhancement effects. OSNs showing enhancement effects were boxed in different colors corresponding to the different modulators. (E) Time courses of four OSNs showing enhancement. Arrows with different colors indicate enhancement by the corresponding odors.

Among the same 6178 cells, we also observed enhancement (Imod > 0.3) in 73 cells, as shown in a heatmap format in Fig. 5D, with enhancement effects highlighted in boxes. Time courses of individual OSNs exhibiting enhancement are shown in Fig. 5E. Cell vii is of particular interest in that it did not respond to isoraldeine, γ-terpinene, or acetophenone, but both binary odor pairs (isoraldeine/acetophenone and γ-terpinene/acetophenone) did activate the neuron. Thus, a wide variety of molecules, even those without an odor, could potentially act as modulators at different receptors.


Our first insight from our results is the occurrence of likely allosteric enhancement at classic small-molecule class A GPCRs, an effect long searched for but only rarely observed among this group of receptors (2227). Our observations should reinvigorate the search for potential drug candidates in other members of this important class of receptors.

The second, and main, result is the widespread modulation of peripheral sensory responses in the detection and discrimination of blends of odors. Suppression of the perception of a particular odor within a blend has been well documented in psychophysical tests (1216). For example, isoraldeine (also known as γ-methyl ionone) has been known to perfumers as a masking agent since at least 2001 (28). However, even the locus of this effect—peripheral, central, or both—has remained undetermined. Antagonism has previously been observed at a few individual receptors, and theoretical models have been proposed to describe the possible consequences of odor mixtures on coding (8, 2937). However, studies to date have been limited by the technical difficulty of performing comprehensive investigations of odor interactions in mixtures. Although monomolecular odors as stimuli can reveal ligand-receptor relations and categories of odor sensitivity, such studies cannot resolve the mechanisms at play in the more realistic case of smelling blends or mixtures of odors. By leveraging SCAPE microscopy to screen widespread cell-specific responses to more odor blend stimuli, we demonstrate here that the receptors themselves are engaged in a variety of modulatory effects, including antagonism, partial agonism, and enhancement, before any further synaptic-mediated processing of the stimulus at higher system levels.

Our results demonstrate that odors can act as agonists at one receptor and antagonists or partial agonists at others, and that odors can also function as enhancers; this paints a much more complex picture of how odor sensing leads to perception of mixed odor blends. Modulation of responses by suppression and enhancement was conspicuous, even in simple three-component mixtures. Although this complexity in the peripheral sensory system demands more than a simple combinatorial coding strategy, it is consistent with data showing no discernable patterning or topographic arrangement of inputs from olfactory bulb to piriform cortex (3840).

A simple combinatorial code of olfactory receptor responses is insufficient to account for the modulatory effects that we saw here with blends of odors. One possible value of this effect is that, by modifying the strength of the signals detected within the periphery to enhance or suppress different components of a mixture, the sensory system could increase its dynamic range. This model, illustrated in Fig. 6, shows that if responses to individual odors generate an expected identifiable odor-specific code, then a simple summation or maximum projection of the codes of multiple odors, as in a blend, could quickly fill up the representation of the summed receptors (model 1). This effect could make numerous mixtures of different odors indistinguishable from one another. The small changes in the combined code resulting from the effects of suppression and enhancement bestows a specificity on the combined code for the mixture (model 2). In some cases, this effect may interfere with the ability to identify every odor component within a blend but would instead provide the mixture with a recognizable identity. To relate this model to our data, we note that measured cell-response profiles shown in Fig. 3, cells i and v, correspond to the presumptive response patterns R2 and R8 in the proposed model.

Fig. 6 Diversified coding capacity through modulation.

Two conceptual models are shown to contrast their robustness in odor mixture coding. (Left) No Modulation model: Suppose odors X, Y, and Z (all monomolecular compounds) can each activate a subset of odorant receptors. In this model, mixing odor Y with odor X would recruit two more receptors, but adding Z will not produce a different perception, although its response profile only partially overlaps with X and Y. (Right) Modulation model: In this model, all receptors are subject to modulation in addition to their activation profiles. Under one possible circumstance (one similar to what we observed), mixing odors X and Y results in the inhibition of receptor 2 and the enhancement of receptor 8. Adding odor Z into the mixture inhibits receptor 5 and enhances receptor 7. As a result, the sparsity is increased because of inhibition and the spectrum of odor coding is expanded through enhancement. Together, these modulation effects serve to increase the robustness of pattern detection as a mechanism of perception. This model also implies that “silent” receptors (R7 and R8 in this case) might be as important as the activated ones in pattern recognition of an olfactory object.

Although humans have an estimated 400 different olfactory receptors, there are several orders of magnitude higher numbers of potential ligands (i.e., odors). Making conservative estimates that any given odor molecule can activate three to five different receptor types at a medium level of concentration, then a blend of just 10 odors could occupy as many as 50 receptors, more than 10% of the family of human receptors (4, 41). The situation is worsened if some of these receptors have overlapping sensitivities, which will result in fewer differences between two blends of 10 similar compounds. The number of available unoccupied receptors is further reduced with the addition of each new component, eventually saturating the system and making it impossible to discriminate between complex blends or to identify any components within the blends. The modulatory actions that we report here could ameliorate this problem, as detailed above and in Fig. 6, by providing additional receptor patterns for complex blends to occupy. This model suggests that higher brain regions may use pattern recognition as an alternative to either combinatorial or analytical coding strategies.

In other nonchemosensory systems (vision, audition, somatosensory, etc.), there are no demonstrated cases of a stimulus activating one receptor or detector and inhibiting another. In these systems, complex interactions among stimuli occur at higher levels of processing, e.g., red-green opponency in retinal ganglion cells. A nearly similar instance of peripheral modulation occurs in the auditory system, where the delivery of two tones of particular frequencies can set up interference waves on the basilar membrane causing auditory illusions (42). However, these are entirely mechanical processes and do not involve physiological responses of the primary sensory neurons. Olfaction thus appears unusual in using stimulus-induced complex activity starting at the level of primary sensory receptors. This unusual complexity at such an early level of sensory discrimination raises crucial questions about the similarity or dissimilarity of higher olfactory processing to that found in other sensory systems. Therefore, there is strong motivation to consider alternative coding strategies for olfaction that are distinct from those identified in other sensory systems.

Materials and Methods


Mice were housed and handled in accordance with protocols approved by the Columbia University Institutional Animal Care and Use Committee (IACUC). The OMP-Cre–driven GCaMP6f strain was generated by crossing the OMP-Cre strain (JAX006668) with Ai95D (CAG-GCaMP6f, JAX024105). Six- to 8-week-old male mice with a genotype of OMP-Cre+/− GCaMP6f−/− were used for SCAPE imaging. A total of 23 mice were used in this study.

To estimate the number of viable OSNs that could be imaged with SCAPE, OMP-Cre−/− GCaMP6f−/− mice were crossed further with Ai75D mice (RCL-nT, Jax025106) so that nuclear-localized tdTomato and GCaMP6f were coexpressed in mature OSNs. One 8-week-old male mouse was then imaged in both green and red fluorescence emission channels with SCAPE microscopy, and 3D segmentation was applied to estimate the total number of OMP-positive neurons within the imaging volume.

Tissue preparation

Mice were overdosed with anesthetics (ketamine 90 mg/kg, xylazine 10 mg/kg, i.p.) and decapitated in accordance with IACUC-approved procedures. The head was cut open sagittally and the septum was removed to expose the surface of the olfactory turbinates. Only the right half of the head was used for experiments. The tissue was placed in cold modified Ringer’s solution containing the following (in mM): 113 NaCl, 25 NaHCO3, 5 KCl, 2 CaCl2, 3 MgCl2, 20 HEPES, and 20 glucose, pH 7.4, for 40 min before imaging.

For SCAPE imaging, the right half of a mouse head with the olfactory turbinates exposed was mounted in a custom-designed, 3D-printed, glass-bottomed perfusion chamber. The perfusion chamber was designed to control the perfusion flow in the nasal cavity with the inlet at the nostril and the outlet at the throat (Fig. 1A, blue trace and arrows). A small amount of light-cured dental composite (Tetric EvoFlow, Ivoclar Vivadent) was applied to adhere the tissue to the chamber. During experiments, the tissue was continuously perfused with carboxygenated (95% O2, 5% CO2) modified Ringer’s solution at room temperature at 0.75 ml/min. Depending on individual differences and the particulars of the tissue mounting, there was some variation in the precise region that was imaged. Typically, the field of view covered the ventral half of either turbinate IIb or III (43) and some portion of the neighboring turbinates. Every experiment covered a large and similar region of the epithelium (Fig. 1A, yellow rectangle). Combined data from all regions recorded were analyzed because we saw no apparent differences in regional responses.

Odors and odor stimuli

All odor chemicals in this study were from Sigma-Aldrich except benzyl acetate, dorisyl, dartanol, and isoraldeine, which were gifts from Firmenich SA. In odor set 1, acetophenone, benzyl acetate, and citral were first diluted in dimethyl sulfoxide (DMSO) to make stock solutions and then diluted in modified Ringer’s solution to 100 μM. For two- and three-component mixtures, acetophenone, benzyl acetate, and citral were mixed and then diluted so that each component had a final concentration of 100 μM. In odor set 2, dorisyl, dartanol, and isoraldeine were mixed at a volume ratio of 45%: 15%: 40% to reproduce the woody accord blend, which has been widely used in the perfume industry. Final concentrations of the three odors were 148, 48, and 127 μM, respectively, both in single-odor solutions and in mixtures. All odor solutions had a final DMSO concentration of 1 to 2.5‰ depending on the solubility of the odors; the DMSO concentration was balanced across all odor stimuli within the same experiment.

Odors were applied for 30 s using a 1260 Infinity HPLC system (Agilent Technologies, Santa Clara, CA, USA) with 2.5-min time intervals between stimuli. For the three-odor mixture experiments, the sequence of individual and binary combinations of odors between the three-odor mixture stimuli was randomized among mice, although the three-odor mixture was always delivered second, sixth, and tenth to ensure repeatability and for correction of rundown when calculating the Imod. Odor delivery in dose-response experiments was performed in the fixed order shown in Fig. 4 and fig. S5. The adenylate cyclase activator forskolin (50 μM, Sigma-Aldrich) was applied at the end of each experiment to assess the viability of OSNs.

SCAPE imaging

High-speed volumetric imaging of intact epithelium was performed on a custom SCAPE microscope (1719). SCAPE is a form of light-sheet microscopy that provides low phototoxicity combined with very high-speed 3D imaging of intact samples through a single, stationary objective lens. Briefly, SCAPE’s high-speed 3D imaging is achieved by illuminating the sample with an oblique light sheet through a 1.0 numerical aperture (NA) primary objective lens. Fluorescence signal excited by this sheet (extending in the y-z′ direction) is collected by the same objective lens (in this case, an Olympus XLUMPLFLN 20XW 1.0 NA water-immersion objective with a 2-mm working distance). A galvanometer mirror in the system is positioned to both cause the oblique light sheet to scan from side to side across the sample (in the x direction) but also to descan returning fluorescence light. This optical path results in an intermediate, descanned oblique image plane that is stationary yet always coaligned with the plane in the sample that is being illuminated by the scanning light sheet. Image rotation optics and a fast sCMOS camera (Andor Zyla 4.2+) were then focused to capture these y-z′ images at >800 frames per second as the sheet was repeatedly scanned across the sample in the x direction. All other system parts, including the objective and sample stage, were stationary during high-speed 3D image acquisition. Data were reformed into a 3D volume by stacking successive y-z′ planes according to the scanning mirror’s x position and deskewing to correct for the oblique sheet angle.

In this study, the stationary objective in SCAPE was configured in an inverted arrangement to image under the perfusion chamber. The overall magnification of the system was configured to be 4.66×. A 488-nm laser was used for excitation (<1.4 mW at the sample) with a 500-nm long-pass filter in the emission path. The system’s sCMOS camera was used at various frame rates for different specific regions of interest (800 to 1300 fps). The maximum field of view of the SCAPE system can be as large as 1600 × 1200 × 350 μm (x-y-z; Fig. 1B). To achieve optimal spatiotemporal resolution, the sample was typically imaged with an x-direction scanning step of 2 μm over an 800 × 1000 × 240 μm field of view (x-y-z, 2.0 × 1.39 × 1.1 μm per pixel) at 2 volumes per second (VPS). For 5 VPS imaging, the x range was 600 μm with a 3-μm step. Because of the variation of the field of view and the tissue structure, the number of cells acquired ranged from 6000 to 12,000 per mouse. Each trial was acquired for 75 s and each mouse was imaged for >20 trials with a 2.5-min intertrial interval.

Volumetric image data processing

Sample drifts were corrected using custom MATLAB code based on NoRMCorre (44). Because tissue drift and motion artifact within each 75-s trial were negligible, a single volume from each trial was taken and registered to a template with manual correction when necessary. The same transform matrix was then applied to all the volumes in each trial. After registration. After registration, a constrained non-negative matrix factorization algorithm (CNMF) was used to extract the locations and time courses of responding neurons (45). To minimize the computational load for large-scale data analysis, CNMF was performed on successive 2D x-y planes extracted from the 3D datasets as mean intensity projections of 7.7-μm-thick layers, concatenated over all different odor trials. These 2D planes were spaced 3.3 μm apart throughout the depth of the epithelium to minimize overlapping neurons. Forskolin trials were omitted from the initial CNMF analysis to extract only the neurons responding to the delivered odors and odor combinations. The identified responding neuron locations were then used to update the time course of each neuron to include the forskolin response.

CNMF-extracted time courses and spatial loci were initially screened with a convolutional neural network to exclude components with spontaneous activity, motion-induced baseline fluctuation, or inconsistent responses to repetitive odor stimuli. All cells were later validated manually to ensure data fidelity.

Time courses of each neuron’s GCaMP activity are shown as ΔF/F, where ΔF is the real-time fluorescent intensity change relative to F and F is the baseline fluorescent signal. Peak response amplitudes were calculated as the average of a 3-s window around the maximal value during odor delivery minus the median of a 6-s window baseline extracted 15 s before each odor was delivered. Cells were considered to be exhibiting responses if their peak response amplitude exceeded 5 standard deviations above the baseline noise.

Clustering cells and plotting normalized response heatmaps

Peak responses were calculated as detailed above and normalized for each cell to its maximum odor response across all odors. Because odors were administered in a pseudorandom manner, data from different animals were reordered before combination into the heatmap.

Neurons were clustered into subgroups using the k-means clustering function in MATLAB with squared Euclidean distance. The number of clusters was determined for each dataset using inspection and similar clusters were combined. For heatmap visualization, in some cases, cells within each subgroup were reordered on the basis of the primary odor response amplitude or their Imod.

A more stringent criterion was applied when defining subgroups II to IV shown in Fig. 3 and subgroups I to III shown in fig. S4: that nondominant odor responses must be ≤30% of the dominant odor response, as detailed in the legend of Fig. 3. Individual neuron responses within each of these clusters were visually inspected to ensure correct classification.

Calculating the Imod

As shown in Fig. 3C, when calculating the Imod, responses to the mixture were linearly corrected on the basis of responses to the two closest neighboring three-odor mixture responses. To start, Imod was calculated as (d0d1)/d1, where d0 is the linearly corrected response to the mixture and d1 is the response to the dominant odor. Neurons for which this Imod was <0 were classified as showing suppression.

Neurons with Imod > 0 were then tested for enhancement effects using the modified equation Imod = max{[(d0d2d3) – d1]/d1, 0}, where d0 is the linearly corrected response to the mixture, d1 is the response to the dominant odor, and d2 and d3 are the responses to the nondominant odors (Fig. 3C). This extended version of the equation is intended to account for the amplitude of neuron responses to nondominant odors, effectively removing those responses from the three-mixture response. This calculation may result in a lower enhancement Imod by estimating a lower magnitude (d0d2d3) than the actual three-mixture response without modulation, which might thus underestimate the degree of enhancement while overestimating the number of neurons showing no modulation effect. We note that the responses to d2 and d3 are not incorporated into the calculation of Imod for suppression effects because they could cause overestimation of suppression effects. When calculating the enhancement effect for binary odor pairs (e.g., acetophenone paired with citral or other odors), the equation is modified to Imod = max{[(d0d2) – d1]/d1, 0}, where d0 is the magnitude of the response to the binary pair, d1 is the linearly corrected magnitude of the response to acetophenone alone, and d2 is the response to the modulator alone (Figs. 4 and 5; see also figs. S3 and S6).

In the intact epithelial preparation used here, rundown of cell responses over time was comparable to or less than that commonly experienced in dissociated cell preparations (9). Figure S2 shows a control experiment for the linear rundown estimation with repeated delivery of the three-odor mixture. Here, a mock Imod was calculated as (d0d1)/d1, where d0 is the linearly corrected response to the second mixture on the basis of the neighboring mixture responses and d1 is its real response to the mixture. In the ideal case, the calculated Imod would equal 0. Here, the mean of Imod was 0.002 with a standard deviation of 0.09 (N = 1302, 3 mice). A sample time course and the histogram of Imod are shown in fig. S2.

Supplementary Materials

References and Notes

Acknowledgments: We thank Z. Peterlin (Firmenich SA), C. Laudamiel (DreamAir LLC), and S. Lomvardas (Columbia University) for discussions and advice on experiment design; K. B. Patel and C. Perez Campos from the Hillman laboratory for maintenance of the SCAPE imaging system; C. Zhang from the Firestein laboratory for technical support in mouse breeding and genotyping; L. Paninski (Columbia University) and E. A. Pnevmatikakis (Flatiron) for advice on data analysis; and other members of the Firestein and Hillman laboratories for their support of this work. Funding: Funding for this work was provided by NIH 2 R01 DC013553, Firmenich contract 3000615937 (to S.F.), NIH BRAIN initiative grants U01NS09429 and UF1NS108213, NCI grant U01CA236554, Department of Defense grant MURI W911NF-12-24 1-0594, the Simons Foundation Collaboration on the Global Brain, the Kavli Institute for Brain Science (to E.M.C.H.), and the National Science Foundation (IGERT funding to V.V. and CAREER CBET-0954796 to E.M.C.H.). Author contributions: L.X., S.F., D.Z., W.L., and E.M.C.H. designed the experiments. V.V., W.L., and E.M.C.H. designed, constructed, and maintained the imaging system. L.X. and W.L. designed, constructed the experimental setup and performed the experiments. L.X., W.L., and E.M.C.H. analyzed the data. L.X., W.L., D.Z., E.M.C.H., and S.F. prepared the manuscript. All authors discussed and contributed to writing the manuscript. Competing interests: E.M.C.H., W.L., and V.V. declare a potential financial conflict of interest relating to the licensing of SCAPE microscopy intellectual property to Leica Microsystems for commercial development. S.F., E.M.C.H., V.V., and W.L. receive funds for advising Firmenich SA in work related to that presented here. Data and materials availability: The data that support the findings of this study are stored on multiple servers at Columbia University and, because of their size, will only be made available from the corresponding authors upon reasonable request.

Stay Connected to Science

Navigate This Article