A large-scale analysis of odor detection in the olfactory epithelium
To obtain a more comprehensive understanding of odor coding in the OE, we sought to analyze the responses of thousands of individual mouse OSNs to a large number and variety of odorants with diverse structures and perceived odors in humans. Since each OSN expresses only one OR gene and each OR gene is expressed, on average, in ~1/1000 OSNs, we reasoned that such an analysis could provide a broad view of odorant recognition not only by the OSN repertoire but also by the mouse OR family.
We first selected 125 odorants with diverse structures and perceived odors (in humans) and grouped them into 13 odorant mixtures according to structural features (). In some cases, these structural features correlate, at least to some extent, with perceived odors in humans: (1) amines (fishy, ammonia); (2) thiols (sulfurous); (3) alcohols (floral, fruity); (4) esters (fruity, floral); (5) ethers (floral); (6) aldehydes (aldehydic, citrusy); (7) cyclic alkanes (woody); (8) terpenes (green, minty); (9) vanillin-like (sweet); (10) camphors (camphor); (11) azines (pungent, animalic); (12) musks (musky); and (13) ketones/others (varied). Also included in the mixtures were a fox predator odor (13-12) (Day et al., 2004
) and five mouse pheromones (Leinders-Zufall et al., 2000
), one present in mixture 11 and the remainder in mixture 13.
Odorants. A total of 125 odorants with varied structures and perceived odors in humans were grouped into 13 mixtures on the basis of structural features.
To analyze the responses of OSNs to the odorants, we used calcium imaging (Malnic et al., 1999
). Mouse OE cells were dissociated, loaded with the calcium indicator, fura-2, and then plated on glass coverslips. Individual OSNs were monitored for increases in intracellular calcium during sequential perfusion with the 13 odorant mixtures (containing 50 μm
of each odorant) and then, in most cases, with single odorants (at 50 μm
) from mixtures that had elicited a response. Many OSNs were subsequently tested with lower concentrations of stimulatory odorants (5 and/or 0.5 μm
). Finally, cells were assessed for viability by exposure to 87.4 mm
KCl, which induces calcium influx in living OSNs. Because of their limited survival time after isolation, OSNs that had responded to multiple mixtures were usually tested with single odorants from only some mixtures. Only OSNs that had responded to KCl (“KCl+ OSNs”) were included in data analyses.
We tested 3000 KCl+ OSNs with the 13 odorant mixtures, a total of 39,000 potential OSN–mixture pairings and 375,000 potential OSN–odorant pairings. Of OSNs tested with elevated KCl, 308 responded to at least one mixture, but 60 were KCl− and another 31 OSNs were excluded from further analysis for other reasons (see Materials and Methods). Of the 3000 KCl+ OSNs, 217 (7.2%) responded to one or more mixtures and were suitable for analysis (). Of the 217 OSNs analyzed, 197 were subsequently tested with single odorants from activating mixtures and 169 of those responded to at least one odorant. In some cases, but not others, an odorant also stimulated an OSN at a lower concentration (5 and/or 0.5 μm
), consistent with previous studies (Sato et al., 1994
; Malnic et al., 1999
; Bozza et al., 2002
). Some OSNs failed to respond to single odorants from one or more stimulating mixtures. One possible explanation for this is that some mixture responses resulted from a summation of responses to multiple mixture components that alone did not stimulate a response at the concentration tested.
Figure 2. Responses of individual OSNs to odorant mixtures. This diagram shows the responses of 217 OSNs (rows) to the 13 odorant mixtures (columns). The red boxes indicate mixtures to which the neurons responded with an increase in intracellular calcium, as measured (more ...)
When 263 additional KCl+ OSNs were later tested with 18 mixtures containing 176 odorants, the percentage of mixture responsive OSNs increased to 13.2% (data not shown). This increase in the percentage of responsive OSNs suggested that a large proportion of the KCl+ OSNs analyzed by these methods were capable of responding if tested with an appropriate odorant, although the exact proportion cannot be ascertained. Since each OR gene is expressed in roughly 1/1000 OSNs, it is likely that these studies queried a large proportion of the OR repertoire.
The repertoire is biased
These studies revealed several striking features of the mouse OSN repertoire. First, the repertoire is biased. Although each odorant mixture stimulated a subset of OSNs, some mixtures activated far more OSNs than others (, ). Remarkably, 59% of the 217 OSNs activated by mixtures responded to the aldehyde mixture, whereas, in contrast, only 15% of OSNs responded to the amine mixture. Taking into account the number of odorants per mixture, the aldehyde mixture activated an average of 9.9 OSNs/odorant, whereas, at the other extreme, the musk mixture activated only one-fifth as many (2.0 OSNs/odorant) and the azine and amine mixtures only stimulated 2.3 and 2.5 OSNs/odorant, respectively (). Alcohols and terpenes activated intermediate numbers of OSNs (6.2 and 5.1 OSNs/odorant, respectively).
Figure 3. Quantitation of OSN responses to odorant mixtures. a, Individual OSNs responded to 1–12 mixtures, but most responded to only 1 or a few mixtures. The number of neurons that responded to the indicated number of mixtures is shown above each bar. (more ...)
Bias at the level of single odorants was also evident but was most striking among the 81 OSNs tested with single aldehydes (). Echoing a report that octanal stimulates a high percentage of rat OSNs (Araneda et al., 2004
), 33 of the 81 OSNs (40.7%) responded to octanal (6-3) and 30 (37%) responded to decanal (6-4), whereas other aldehydes stimulated 2–21 OSNs each. Bias was also seen among odorants from some other mixtures, although it was less extreme than that observed for aldehydes. For example, individual alcohols stimulated 0–11 of 34 OSNs tested and different esters activated 0–7 of 26 tested OSNs ().
Figure 4. Responses of OSNs to single odorants. These diagrams show the responses of individual OSNs (rows) to single odorants (50 μm) (columns) from different odorant mixtures, as indicated. The blue boxes indicate odorants to which OSNs responded with (more ...)
The observed biases in OSN responses to different odorants could reflect bias in either the number of ORs that recognize different odorants or bias in the proportion of OSNs that express different ORs. To explore the source of the observed bias to octanal and decanal, we compared the response profiles of OSNs activated by these odorants in terms of their responses to, first, single aldehydes and, second, different mixtures. The 33 OSNs activated by octanal showed 28 different mixture/aldehyde response profiles and the 30 OSNs activated by decanal showed 26 different profiles (). Possibly because of variations in the level of expression of a given OR among OSNs, some OSNs expressing a particular OR may respond to two odorants at different thresholds, whereas others respond only to the lower threshold odorant (Bozza et al., 2002
). Thus, OSNs with the same OR can show related, but different response profiles. For this reason, the actual number of ORs involved in the responses to octanal or decanal is unknown.
Nevertheless, the larger number of different response profiles seen for these two odorants than for other aldehydes indicates that there are likely to be more ORs that recognize octanal or decanal than other odorants, a conclusion similar to that reached for octanal in rat (Araneda et al., 2004
). These results support the idea that there is bias not only in the OSN repertoire, but also the OR repertoire, and that there are likely to be many more ORs that recognize some odorants than others.
The repertoire exhibits extreme diversity
One of the most striking features of the OSN responses seen in these experiments was their extreme diversity. This diversity was apparent, first, in the responses of OSNs to the 13 odorant mixtures. The 217 mixture-responsive OSNs showed 93 different response profiles composed of responses to single mixtures or combinations of mixtures ().
OSN responses to individual odorants were also extremely diverse. For example, 81 OSNs tested with single aldehydes showed 36 different response profiles composed of responses to single aldehydes or combinations of different aldehydes (). Similarly, 34 OSNs tested with single alcohols showed 15 response profiles and 6 OSNs tested with different musk odorants showed 5 different profiles ().
The diversity of OSN responses seen in these experiments is consistent with the large size of the mouse OR repertoire and the extensive diversity seen in OR protein sequences (Zhang and Firestein, 2002
; Godfrey et al., 2004
). However, as already noted, OSNs with the same OR can show slightly different response profiles, possibly because of variations in the level of OR gene expression among neurons. Thus, while much of the response diversity seen in these experiments is likely to reflect diversity in the recognition properties of different ORs, some is likely to be due to variations in the response properties of OSNs expressing the same OR.
Most mouse OSNs are narrowly tuned
The large number of OSNs and odorants tested in these studies permitted analysis of the extent to which individual OSNs are narrowly tuned to recognize a relatively small number of related odorants versus broadly tuned to recognize a comparatively large number and variety of odorants.
These experiments indicate that the majority of mouse OSNs are narrowly tuned. Narrow tuning was apparent, first, from OSN responses to the odorant mixtures. Of 217 OSNs activated by mixtures, 44.7% (97 of 217) responded to only one mixture containing structurally related odorants ().
Narrow tuning was further evident in the responses of OSNs to individual odorants. Of the 97 OSNs that responded to only one odorant mixture, 76 subsequently responded to at least one odorant from that mixture. More than one-half of those OSNs [43 of 76 (56.6%)] responded to only one odorant and another 35.5% (27 of 76) responded to two to three odorants, whereas only 7.9% (6 of 76) responded to four to five odorants, and none responded to all odorants in the mixture.
In most cases, the odorants recognized by the narrowly tuned OSNs had related structures. Two examples shown in are OSN223, which selectively responded to four structurally related odorants from the vanillin-like mixture, and OSN366, which responded only to indole and skatole, two structurally related odorants from the azine mixture that share an animalic-fecal odor (Yokoyama and Carlson, 1979
; Garner et al., 2007
). Narrow tuning to structurally related odorants was also seen among OSNs that recognized more than one mixture. For example, one OSN (OSN166) responded to structurally related odorants in mixtures 7 and 10 (odorants 7-13, 10-2, 10-3, and 10-5) (data not shown). Two other examples are OSN175 and OSN319, each of which recognized aliphatic odorants with long carbon chains present in different mixtures (data not shown). The first responded to a undecylenic alcohol (3-8) and decanal (6-4), while the latter responded to heptane thiol (2-2), heptanol (3-2), and octanal (6-3) (data not shown).
These results suggest that the majority of mouse OSNs are narrowly tuned to recognize a relatively small assortment of odorants that share a particular structural motif. Narrow tuning clearly extends beyond the recognition of a single obvious structural motif, however, since individual OSNs responded to varied subsets of odorants with the same motif and, in some cases, the odorants recognized by an OSN did not share any obvious structural feature.
Some OSNs are specific for animal-associated chemicals
The odorants tested in these studies included a small number that are associated with animals, at least some of which can elicit an innate response in mice. These included several mouse pheromones (11-6, 13-5, 13-10, 13-11, 13-13) (Leinders-Zufall et al., 2000
), a predator odor present in fox feces, TMT (2,5-dihydro-2,4,5-trimethylthiazoline) (13-12) (Morrow et al., 2000
; Kobayakawa et al., 2007
), two odorants with a pronounced fecal odor [indole (11-1) and skatole (11-2)], one odorant that has the odor of decaying flesh [cadaverine (1-12)], and the odorants in the musk mixture, which are characterized by an animalic musky odor. One question was whether the OSNs that recognize these odorants also detect other odorants or whether they are instead highly specific.
Of the five mouse pheromones, only three elicited a response in OSNs and those three stimulated a total of five OSNs. Four of the OSNs responded not only to pheromones, but also other odorants. However, one OSN (OSN20) responded to only one mixture and to only one compound in that mixture, the pheromone α-farnesene (13-11) (). It is conceivable that OSNs that respond to pheromones as well as other odorants are involved in the perception of the pheromone as an odorant, whereas an OSN such as OSN20, which may be specific for a pheromone, is involved in the generation of instinctive responses to pheromones detected in the OE. In this regard, it may be relevant that α-farnesene, a pheromone in male urine that accelerates female puberty onset, can activate some neurons in the olfactory cortex that communicate with hypothalamic gonadotropin-releasing hormone (GnRH) neurons that regulate reproductive hormones (Boehm et al., 2005
Another case in which we observed extreme odorant specificity that might be relevant to animal behavior was in responses to the odorants skatole and indole. These two closely related odorants are present in feces, both have fecal odors, and skatole is believed to give feces its characteristic odor (Yokoyama and Carlson, 1979
; Garner et al., 2007
). Among the neurons responsive to these odorants, we found one (OSN366) that detected only one mixture and only two odorants in that mixture, indole and skatole ().
We also identified one OSN (OSN293) that was highly specific for cadaverine (1-12), the odorant with the odor of decaying flesh (data not shown) (). Interestingly, this was the only OSN that responded to cadaverine among all those examined. Another OSN highly specific for cadaverine (OSN446) was identified when additional OSNs were tested with 176 different odorants (). One question is whether indole, skatole, and cadaverine, all of which are repulsive to humans, elicit innate responses, such as avoidance, in mice.
As already noted, only a small proportion of OSNs responded to the musk mixture. Of the six OSNs that subsequently responded to individual musk odorants, three responded not only to musks but also to other types of odorants. However, the other three (OSNs 216, 339, and 355) all responded to only the musk mixture and each of those responded to a single musk odorant (12-1, 12-3, or 12-7) (data not shown) (). Another OSN from the set tested with 176 odorants (OSN454) responded to only the musk mixture and then to several different musk compounds (). As with indole, skatole, and cadaverine, it remains to be seen whether or not these animalic odorants stimulate specific responses in mice.
Although it cannot be excluded that the OSNs that responded to these animal-associated odorants also recognize other odorants that were not tested, these results raise the possibility that there might be some pheromones or other animalic odorants that are recognized by highly specific OSNs and ORs that provide signals to the brain that stimulate innate responses.
The repertoire contains broadly tuned components
Surprisingly, these studies revealed that a small proportion of mouse OSNs are broadly tuned. In contrast to the majority of OSNs examined, these OSNs responded to a relatively large number and variety of odorants.
The existence of broadly tuned OSNs was suggested, first, by the responses of some OSNs to odorant mixtures. Of 217 OSNs responsive to mixtures, 29 OSNs (13.4%) responded to 5–9 of the 13 mixtures and, remarkably, 4 OSNs (1.8%) responded to 10–12 mixtures (, ).
Although it was not possible to test those OSNs with single odorants from all active mixtures, those tested with odorants from several mixtures were informative. For many of those OSNs, the stimulatory odorants shared a particular structural feature, such an extended carbon chain or an aldehyde or ester group (data not shown). However, other OSNs, such as OSN226 and OSN273, were activated by some odorants that shared a structural motif and others that did not (), suggesting the possible involvement of less obvious physicochemical characteristics.
Further analysis of one broadly tuned OSN, OSN226, demonstrated that the broad tuning of this OSN derived from broad tuning of the OR it expressed. OSN226 responded to 10 odorant mixtures and to 12 single odorants with which it was tested (). Using single-cell RT-PCR, we identified the OR gene expressed in this neuron as Olfr42. When we cloned Olfr42 and expressed it in HEK293T cells, we found that the OR responded to 9 of 11 of odorants that had stimulated OSN226 as well as to 4 additional odorants (). This is consistent with a previous study showing that a different mouse OR is broadly tuned (Grosmaitre et al., 2009
Figure 6. Olfr42 is a broadly tuned OR. After identifying Olfr42 as the OR expressed in OSN226, a broadly tuned OSN (), HEK293T cells were cotransfected with expression vectors encoding Olfr42 (or vector alone), RTP1s, and Ric8b together with a vector containing (more ...)
Most odor codes are unique and combinatorial
Previous studies have indicated that different odorants are detected, and thereby encoded, by different combinations of ORs. The present studies allowed analysis of the extent to which this combinatorial scheme extends to a larger number and variety of odorants than were previously tested.
In these studies, we tested 125 odorants, 102 of which activated one or more OSNs. Comparison of OSNs activated by single odorants from the same mixture showed that the vast majority of odorants [96 of 102 (94.1%)] stimulated a unique set of OSNs (). Moreover, while some odorants were recognized by only one OSN, the majority [78 of 102 (76.5%)] were recognized by a combination of different OSNs (). For example, the 13 different aldehydes stimulated 13 different combinations of OSNs. Similarly, each of the 15 esters that activated OSNs stimulated a different set of OSNs, with 14 of 15 stimulating more than one OSN (). These findings indicate that the principle of combinatorial coding extends to a wide variety of odorants with different types of structures and perceived odors. It also shows how this principle, in combination with the extreme diversity of OSN odorant recognition, can generate a multitude of unique codes that permit a vast number of odorants to be discriminated.
Most odorants are recognized by a unique set of OSNs
Analysis of n
-aliphatic odorants with six or seven carbon atoms and different functional groups (amino, thiol, hydroxyl, or aldehyde) showed that, despite their similarity, each odorant was recognized by a unique combination of OSNs (). As in a previous study of n
-aliphatic odorants with other functional groups (Malnic et al., 1999
), a change in either carbon chain length or functional group changed the combination of OSNs recognizing an odorant (its “combinatorial code”). Given the relatedness of human and mouse OR families (Zhang and Firestein, 2002
; Godfrey et al., 2004
; Malnic et al., 2004
), human ORs are presumably used in a similar fashion, providing an explanation for the ability of these odorants to elicit different odor perceptions in humans (Malnic et al., 1999
Figure 7. Structurally related odorants are recognized by different combinations of OSNs. n-Aliphatic odorants with six or seven carbon atoms and different functional groups (amino, thiol, hydroxyl, or aldehyde) (rows) each elicited responses in a different combination (more ...)
The size of the code varies among odorants
One question raised by previous studies, but unanswerable because of their smaller scale, was how large the “code” is for various odorants. What proportion of OSNs and ORs are used to encode the identities of individual odorants?
The data collected in the present studies indicate that the size of the code can vary extensively among odorants. The number of OSNs activated by different odorants from the same mixture (excluding odorants that stimulated no OSNs) was 1-11 for alcohols, 1-7 for esters, 2-33 for aldehydes, 1-6 for cyclic alkanes, and 1-7 for vanillin-like odorants (). These results suggest that, even among structurally related odorants, some odorants may be encoded by 10-30 times as many OSNs, and likely ORs, as others.
Is there a functional logic to these differences, such as larger codes for food odors? Using “odor type” classifications from online resources, the most stimulatory odorants among tested alcohols, esters, aldehydes, cyclic alkanes, and vanillin-like compounds were classified as green/citrus, fruity, aldehydic (bitter, fatty, waxy), herbal, and anisic (sweet), respectively, while the least stimulatory were classified as camphor/alcoholic/fermented, fruity, spicy, amber/woody, and spicy/minty. These results do not suggest any functional logic to differences in the number of OSNs that recognized different odorants from the same mixture, at least not in reference to perceived odors in humans.
As already discussed, the aldehyde, ester, and alcohol mixtures stimulated many more OSNs on a per odorant basis than did the amine, musk, and azine mixtures. In addition to being classified as belonging to particular odor types, individual odorants can be assigned one or more “odor descriptors” (odor qualities or subqualities). Although the tested aldehydes, esters, and alcohols have multiple odor descriptors, many of the tested aldehydes and esters are described as “citrus” or “fruity,” descriptors also given to some of the alcohols. In contrast, most amines have “fishy/ammonia” odors, musks have musky odors considered to be animalic, and the tested azines are described as animalic, fishy, or green. This suggests that there may be a slight bias toward structural classes of odorants that include those with citrus or fruity odors. The relatively small proportion of OSNs that recognize odorants with animalic odorants could be of greater significance, however, since some odorants of that class could conceivably serve as social cues that elicit specific physiological responses or behaviors.
How do combinatorial codes convey odor qualities?
One question raised by these and previous studies is how an odorant's combinatorial code conveys its odor quality. Is it possible that some ORs can convey a particular odor quality, such as minty, or different subqualities of the same odorant? If so, one might expect to find some OSNs that recognize only a single odorant or odorants that share an odor quality. Although it is impossible to determine whether this is the case without testing every possible odorant with human ORs, the present studies did uncover some interestingly relationships between odorants and mouse OSNs, which express ORs related to those found in humans.
First, as already discussed, some OSNs recognized certain animal-associated odorants, such as cadaverine or individual musk odorants, but no other tested odorants. Second, the odorants recognized by some OSNs shared not only a structural motif but also an odor quality or odor descriptor in humans. Among 92 OSNs that were tested with single odorants from every mixture to which they had responded and were activated by at least one odorant from each of those mixtures, 49 responded to two or more odorants. Of those, 39 of 49 (79.6%) recognized odorants that all shared an odor descriptor (, ). These findings raise the intriguing possibility that, at least in some cases, a particular OR may convey a specific odor quality or subquality, such as minty or fishy.
Odorants recognized by the same OSN often share an odor quality
Figure 8. Individual OSNs can recognize related odorants with similar or dissimilar odors. Shown here are the odorants recognized by a series of different OSNs that recognized odorants with related structures. In many cases, the odorants detected by an OSN shared (more ...)
However, many of the odorants shown in were recognized not only by such seemingly “odor-specific” OSNs but also by OSNs that responded to other odorants with unrelated odors. Moreover, as already discussed, some odorants with related structures but very different odors were recognized by partially overlapping sets of OSNs ().
Figure 9. Individual odorants can be recognized by a combination of highly specific and broadly tuned OSNs. Some OSNs (columns) responded exclusively to one or few odorants (rows) of a particular odor type (woody/camphor, minty/mentholic, fishy, or fruity), as (more ...)
Studies using human ORs and larger panels of odorants will ultimately be required to assess how ORs give rise to human odor perceptions. However, like other proteins, ORs found in human and mouse are related, suggesting that they are likely to have related ligand specificities. The above findings raise the possibility that, while there might be ORs that convey a particular odor quality, there may be many more ORs that do not do so.