Chem. Senses 25: 429-443,
2000
© Oxford University Press 2000
REVIEW |
Quantification of Odor Quality
Chemosensory Perception Laboratory, Department of Surgery (Otolaryngology), University of California, San Diego, La Jolla, CA 92093-0957, USA and 1 Department of Psychology, Uppsala University, S-75142 Uppsala, Sweden
Correspondence to be sent to: William S. Cain, Chemosensory Perception Laboratory, Department of Surgery (Otolaryngology), Mail Code 0957, University of California, San Diego, La Jolla, CA 92093-0957, USA. e-mail: wcain{at}ucsd.edu
 |
Abstract |
|---|
 Top Abstract IntroductionCriteria for a measureHow have odors been...ConclusionsReferences |
|---|
The relationship between odor quality and molecular properties is arguably the most important issue in olfaction. Despite sophistication in the chemical characterization of molecules, accompanying perceptual characterization has had little quantitative usefulness, relying mostly on enumerative description. As a result of weak interest in the topic outside industry and little agreement regarding how to measure quality, the field of olfactory psychophysics has failed to develop a substantial database for odor quality and has offered little help to other researchers, e.g. neurobiologists, in choice of stimuli, interpretation of outcome or testable hypotheses. This review scrutinizes how psychophysicists and others have measured quality and offers criteria for useful techniques. Most measures have had a subjective component that makes them anachronistic with modern methodology in experimental behavioral science, indeterminate regarding the extent of individual differences, unusable with infrahumans and of unproved ability to discern small differences. Techniques based upon performance, rather than on the more common reporting of mental content, offer firmer possibilities for growth. These techniques inevitably tap the discriminative basis of perception. The nonsubjective techniques have high sensitivity, can have counterparts in infrahuman research, are suitable to examine individual differences and yield non-negotiable answers with potential archival value. Discriminative techniques have their limitations, too—principally excess sensitivity that abridges their use to comparisons between similar-smelling stimuli. Research has begun to extend that range and may overcome the limitation. Application of discriminative methods may have the side-effect of shifting focus in structure–activity research from searches for molecular least common denominators that underlie often vague similarity to the search for molecular properties of importance in discrimination of small differences.
|
Introduction |
|---|
|
TopAbstractIntroductionCriteria for a measureHow have odors been...ConclusionsReferences |
|---|
Can you measure the difference between one kind of smell and another? It is very obvious that we have very many different kinds of smells, all the way from the odour of violets and roses up to asafetida. But until you can measure their likeness and differences you can have no science of odour. (Alexander Graham Bell, 1914)
In the many decades since Bell made his observation, no such science of odor has materialized. Scientists have neither measured likenesses and differences very effectively nor deciphered what causes them. Various notions concerning the relationship between properties of molecules and their corresponding odors have appeared, but none has attained acceptance as a legitimate theory (Cain, 1988
; Rossiter, 1996, Chastrette, 1997). As Bell saw, it is axiomatic that any account of odor quality should develop around a corpus of measurements. Science makes incremental progress through cycles of data-collection followed by theorizing or model-building followed by more data-collection. In the case of odor, few trustworthy measurements of likenesses and differences have preceded theory and few have followed. Theory has accordingly had few data to explain and has stimulated little research.
To develop a corpus, the field needs trustworthy measuring techniques. This paper examines conditions for some choices.
|
Criteria for a measure |
|---|
|
TopAbstractIntroductionCriteria for a measureHow have odors been...ConclusionsReferences |
|---|
Techniques to measure quality could usefully meet three criteria: (i) they should resolve quite small differences since the clearest insights into molecular determinants of quality will undoubtedly come in the quest to account for small differences; (ii) they should produce indices with properties of interval-scale measurement (Stevens, 1950
, 1960) to allow metric comparisons with molecular properties; and (iii) they should avoid subjectivity. The third criterion places the most serious constraint on the techniques, for almost all used through the years have a subjective aspect, a matter not always acknowledged explicitly (Figure 1).
|
Before moving on, definitions of objective and subjective (as the terms will be used in this piece) might prove useful. An objective assessment is based on performance, whereas a subjective assessment is based on a report of mental content. Suppose, for example, one wished to assess the qualitative similarity of two odors. One might determine how well a subject can tell the two odors apart (a performance-based, or objective, assessment) or ask the subject how different the odors seem (a report of mental content, or subjective, assessment).
In the world of applied sensory measurement, subjective characterization can serve satisfactorily to communicate the aromas of materials, foods, beverages, personal products, etc., and in this role deserve no particular disparagement (Swan and Burtles, 1980; Stone and Sidel, 1985). In order to characterize an aroma absolutely, rather than comparatively, some such description may seem unavoidable, but when used to test the determinants of quality, description may limit progress by providing outcomes sometimes as vague as Rorschach inkblots (Figure 2). Some techniques entail greater subjectivity than others and it would be imprudent to tar all with the same brush. Nevertheless, reliance on subjective techniques at all has at least three important liabilities:
|
- Subjective answers are commonly circumstantial in that they depend on context, on experience, on attitude, on culture and on personal factors of unspecified origin. Technically, one may not dispute any person’s description, for that would violate a private domain. Yet, by the same token, every description is open to contradiction and negotiation as just one person’s opinion and hence not necessarily ‘true’. When different persons characterize the same odor differently, the reasons are generally left unexplained. Descriptions often achieve stability only when obtained from groups of subjects. Even so, a group of Americans may agree on one description and a group of French people on another. Which group has yielded the objective answer?
- Subjective measures can lead to rapid generation of data, but the speed comes with a price, namely an inability to give careful consideration to individual differences (even if such differences did reflect true differences in reception). As discussed below, subjective techniques, unlike nonsubjective techniques that measure performance, entail the collection of just a small amount of data per odor comparison. Once a subject describes an odor, such as ethyl butyrate, as fruity–fragrant–aromatic–sweet, that subject’s job is typically done (Cain et al., 1998). This holds even for, say, numerical ratings of similarity between odors. After one or two ratings for a particular pair of odors, a subject will remember the ratings and on subsequent trials merely repeat them, showing spuriously high repeatability with himself, though adding no new information.
- Subjective techniques generally preclude isometric comparisons between the data of humans and that of other animals. Interspecies comparisons, combined with physiological data on the infrahuman, represent a historically important path to understanding. In light of the progress in the study of neural coding in olfaction (Mori and Shepherd, 1994; Yokoi et al., 1995; Buck, 1996), lack of commensurate progress in the psychophysics of smell seems a badly missed opportunity. Psychophysical research could provide key information about odors in the neurobiological experiment (Figure 3).
|
|
How have odors been measured? |
|---|
|
TopAbstractIntroductionCriteria for a measureHow have odors been...ConclusionsReferences |
|---|
Classification
From the time of the ancient Greeks, scholars have grouped odors by qualitative resemblance (Beare, 1906; Zwaardemaker, 1925; Boring, 1942; Harper et al., 1968; Cain, 1978). Such schemes can offer ordinal measurement. In the hierarchical categorization scheme that one can create as a dendrogram, similarity can be seen graphically to decrease ordinally the closer to the root one must progress to find a link between any two odors (Figure 4).
|
Categorization as a judgement may seem harmlessly nonsubjective, but a person’s background and personal theory of reality can actually influence categories as much as, or even more than, basic sensory similarity (Smith and Medin, 1981
; Murphy and Medin, 1985). Do people call various floral aromas by that name because they smell so much alike or because they seem to come from flowers? Do people call spicy aromas by that name because they smell categorically alike or because they come from seasonings (Chastrette et al., 1988)? The stimuli impose some constraints, but much of what subjects express depends upon their internally generated strategies.
Once knowledgeable about what the person ‘contributes’ to the classification, one cannot view a classification scheme as a rendition of truth, but of someone’s rendition of truth. As Hendrik Zwaardemaker (Zwaardemaker, 1925), the first major olfactory physiologist and the author of a well-known classification scheme, wrote: ‘If we choose to give the same name to several similar odors, it could often happen that the evidence of our designation eludes other people . . . and one may question the legitimacy of our identification’ (p. 179). In fact, others did question the legitimacy of Zwaardemaker’s scheme (Titchener, 1912; Henning, 1916; Hazzard, 1930), as one person will inevitably question another’s subjective organization of almost any complex matter. That, though, is the crux. As noted above, the measurements of interest should not be negotiable.
The notion that one scheme is right and another wrong is implicit in disputes over the ‘validity’ of one scheme versus another. History does not endorse any particular scheme as more valid than the rest, though some have proven quite useful for other reasons. For example, the sixfold scheme of Hans Henning (Henning, 1916) did not survive empirical scrutiny (Gamble, 1929; Boring, 1942; Harper et al., 1968; Cain, 1978), but it did spur progress in the measurement of quality. The techniques of Henning’s critics represented methodological advances, including odor profiling (Dimmick, 1922, 1927; Hazzard, 1930), triadic comparisons (Macdonald, 1922; Findley, 1924; Bentley, 1926) and direct ratings of similarity (Dimmick, 1927). John Amoore’s (Amoore, 1962a,b, 1964) scheme of seven odor ‘primaries’, and the accompanying models of receptors based largely on structural least-common denominators within categories, received modest support at best (Döving, 1965; Köster, 1965; Amoore and Venstrom, 1967; Johnson, 1967; Amoore, 1970) [see also (Buck and Axel, 1991; Wang et al., 1998)]. However, in the course of his work and that of others (Beets, 1978), it became evident that quality does relate to chemical structure, though specifics have proven elusive. Hypotheses regarding determinants of quality can certainly direct research, though there is something rather too facile and seductive about the search for least-common denominators with forgiving outcome variables (Rossiter, 1996).
Sorting
Some investigators have asked subjects to sort odors by qualitative resemblance (Lawless, 1989; MacRae et al., 1990, 1992; Stevens and O’Connell, 1996). When a subject places two odors in the same group it indicates similarity. The technique has naive subjects do in a controlled setting what classifiers such as Linnaeus and Zwaardemaker did by personal observation. Lay subjects, however, generally have no knowledge of the composition or origin of the substances they evaluate. Could these subjects in their ignorance of origin offer more objectivity than experts? Not likely. No matter who does it, sorting still qualifies as subjective. Average data show reasonable test–retest reliability (MacRae et al., 1990, 1992), but subjects differ in the number of categories they form. Sorting might in principle reflect true individual differences in reception, but even subjects within special subgroups differ in their categories (Stevens and O’Connell, 1996). Subjects accordingly seem to show diversity in their decisions about which odors belong in the same group. An attempt to force consistency by setting the number of categories simply imposes arbitrary structure on the data (Lawless, 1989).
Profiling
Profiling of odors relies on ratings of the applicability of odor-related words or reference odorants for their relevance to a test odorant. The words usually comprise fixed descriptors (Pilgrim and Schutz, 1957; Yoshida, 1964b; Harper et al., 1968; Moskowitz and Barbe, 1977; Coxon et al., 1978; Dravnieks et al., 1978; Dravnieks, 1982; Laing and Willcox, 1983). The use of a list imposes some constraint on outcome and therefore plays it own role (see Figure 1). In the widely known scheme of Dravnieks (Dravnieks, 1985), subjects rate the applicability of each of 146 descriptors (e.g. cardboard-like, fishy, leather-like, sauerkraut-like) to a given test-odor on a scale from 0 to 5. In the other type of profiling, subjects evaluate odors against chemical references (Crocker, 1945; Schutz, 1964; Wright and Michels, 1964; Amoore and Venstrom, 1967; Yoshida, 1975; Polak et al., 1978). Wright and Michel (Wright and Michel, 1964), for example, gave their subjects a scale of 1 to 7 to indicate the qualitative similarity between a test-odor and nine references. Amoore (Amoore, 1969) gave his subjects a scale of 1 to 9 to indicate similarity between 107 test odors and seven references (Figure 5).
|
Dravnieks’s list of 146 descriptors covers a broad range of qualities, but Dravnieks himself anticipated that investigators would need to add more depending on the stimuli under scrutiny. Specialists have tailored systems to the specifications of the substances they evaluate. Noble’s ‘wine aroma wheel’, for example, includes 87 notes relevant to wine (Noble et al., 1984
, 1987). With the need for so many verbal descriptors to obtain resolution of quality, one wonders how many chemical references it would take to characterize a given odor. Hundreds, as with words? Certainly more than seven or nine. Without a comprehensive list of descriptors (or references), one risks obtaining only a parochial picture of similarity. To illustrate, suppose one wished to profile a group of esters present in various fruits. If the list failed to include an adequate number of fruit-relevant descriptors (e.g. pineapple, banana, orange), one might underestimate differences among the esters. On the other hand, if one tailored the scheme to this group of chemicals by including only fruity descriptors, one might overestimate their differences compared with other stimuli. To a considerable extent, therefore, the experimenter’s choice of descriptors determines outcome.
Almost needless to say, the use of verbal descriptors assumes subjects to have a common base of olfactory experience and to use the same words in the same way to describe sensations. That subjects differ both in the kind and number of descriptors they apply to a given odor makes this assumption questionable (Dravnieks et al., 1978; Dravnieks, 1982, 1985; Cain et al., 1998). Individual differences in reception accordingly become confounded with individual differences in cognition, culture and experience (Wysocki et al., 1991). Dravnieks chose to average across 15 or more subjects and to purge aggregate profiles of infrequently used descriptors. Average profiles created in this fashion exhibited impressive reliability, but of course provided no information about individual differences (Dravnieks et al., 1978; Dravnieks, 1982, 1985; Jeltema and Southwick, 1986); nor do they have a counterpart in animal research or offer a simple distillation of the profile into an index.
Direct ratings of similarity
In various investigations, subjects have been instructed to give numerical ratings of similarity to all pairwise combinations among sets of odors (Engen, 1964; Yoshida, 1964a,b; Döving and Lange, 1967; Woskow, 1968; Berglund et al., 1973; Dravnieks, 1974; Moskowitz and Gerbers, 1974; Davis, 1979; Schiffman, 1981, 1984; Seeman et al., 1989). Unlike profiles, direct ratings do not require the experimenter to derive a measure of similarity indirectly from either words or other odors. The numerical judgement stands for the similarity.
Direct scaling of similarity has focused in some measure on data for individuals to address whether subjects rate similarity similarly. Most studies showed poor agreement among individuals (Dimmick, 1927; Yoshida, 1964a,b; Gregson, 1972; Berglund et al., 1973). Intersubject variation is not necessarily random, as Dimmick (Dimmick, 1927) first noted when his four subjects separated themselves into two pairs, with substantial agreement within pairs but poor agreement between them. Much later, Gregson (Gregson, 1972) factor-analyzed a matrix of inter-correlations between subjects and found several factors. Different subjects showed strong loadings on different factors (Davis, 1979). It appears that different subjects may judge similarity by different criteria.
Average ratings, which essentially treat differences between subjects as noise, seem to offer somewhat degenerate information regarding similarity. Multidimensional analyses of average ratings typically yield a dominant hedonic dimension/factor, with one or two others that defy clear interpretation (Yoshida, 1964a,b, 1972; Woskow, 1968; Moskowitz and Gerbers, 1974; Schiffman, 1981, 1984). Faced with an unfamiliar task and lacking clear instructions regarding criteria, subjects may resort to judgements of pleasantness as a way to convey some information. Beyond the salient aspect of pleasantness, the task may be analogous to asking someone to rate the similarity of a pencil and a baseball bat. One person may judge on the basis of size and find the objects dissimilar, whereas another may judge on the basis of shape and find them similar. Even if individuals use personal criteria consistently, the differences in non-hedonic criteria may tend to cancel out in the averaging. The hedonic dimension may survive because it alone figures in the ratings of most subjects.
Some investigators have worked with related chemicals that differ from one another in regular and specific ways (Engen, 1964; Southwick and Schiffman, 1980; Schiffman and Leffingwell, 1981). In such cases, subjects show better agreement, perhaps because related odors do not differ along so many dimensions and because relevant criteria may define themselves.
Data from different laboratories agree, though imperfectly [see (Callegari et al., 1997) for a review of studies with some common test-odors]. Some inconsistencies could have come from methodological differences, but since different studies employed different sets of odors, some differences could have come from effects of context. Kurtz and collaborators (Kurtz et al., 2000) found ratings of similarity among a group of odorants changed with substitution of one odor in the group. Thus, subjects do change the criteria by which they judge similarity.
Numerical ratings, such as those used in direct scaling of similarity, occupy a very important place in psychophysical research. They have generated much useful data, mostly on perceived intensity. Some investigators have maintained that ratings have high validity and some that they have virtually none. Some have argued for the validity of just one class of rating, e.g. magnitude estimation and related techniques, and some just for another kind, e.g. category rating.
Anderson (Anderson, 1982) has set out criteria to establish the linearity of numerical ratings with the internal representations of stimuli. The criteria entail collection of data on additivity—a matter not always possible to do— but the spirit of such tests has considerable appeal. If, for example, one could establish lawfulness of similarity estimates for the perceived quality of mixtures where an experimenter could exert some control of quality, then standard means of testing for the additivity of estimates of similarity might qualify them in a way that no one has done for any other of the methodologies. This would not make their outcome isometric with animal data, nor would it necessarily prove that such estimations would have the requisite sensitivity, but the possibility that methodology to measure similarity could earn its stripes should remain open.
Triadic comparisons
In triadic comparisons, subjects pick the most similar and the least similar pairs of odors within triads (MacRae et al., 1990, 1992). In this technique, subjects need not quantify similarity, but need make only ordinal judgements. As with ratings of similarity, different subjects might use different criteria. To our knowledge, no recent studies have examined individual differences. However, researchers who used similar techniques to evaluate Henning’s prism (Macdonald, 1922; Findley, 1924; Bentley, 1926) reported that subjects differed widely in their judgements and suggested that subjects do in fact employ different criteria. The matter remains open.
Measures of performance
For more than a century, experimental psychologists have grappled with the issue of whether to build psychological science on studies of the content of the mind or on studies of human abilities (Cattell, 1893). To a large degree, one’s position on the matter dictated permissible methodologies. If in the former camp, a report of mental content, such as a direct rating of the brightness of a scene, was permissible. If in the latter camp, quantification of perceived brightness would need to be approached less mentalistically. One could ask a subject to match one brightness to another (and by enumerating errors over trials could measure accuracy), or to state whether one brightness exceeded another, or otherwise to discriminate brightnesses, but not to quantify them directly. Why not? Because differences in ratings could have origins outside perceived brightnesses. The same issue exists here.
The rise of cognitive psychology and cognitive science has seen increasing interest in mental content, though cognitive scientists would find the term ‘mental operations’ more agreeable, but with refinement of experimental techniques to deduce operations via tests of ability, e.g. how fast subjects might notice features of complex stimuli. In this way, they avoid the pitfalls of ‘mentalism’. The same could in principle occur in the investigation of odor quality. In olfactory psychophysics, investigators have examined how fast and how accurately subjects can sort odor mixtures on quality in order to determine if subjects can selectively attend to one qualitative attribute and ignore another. Whether or not subjects can do this for all odors, some odors or none has implications for olfactory coding (Schwartz et al., 1987). Whether subjects can ‘find’ the correct number of constituents in a mixture may also have such relevance (Livermore and Laing, 1998). As unlikely as it may seem on the surface, these tests of ability, and various others (e.g. capacity to identify odors), could serve to examine aspects of structure–activity relationships. There are no a priori limits on the number or variety of tests of ability that one might use to address aspects of coding or structure–activity relationships.
Cross-adaptation
Exposure to an adapting-odor tends to raise the threshold for, lower the perceived intensity of and increase simple reaction time to a test-odor (Köster and de Wijk, 1991; Cometto-Muñiz and Cain, 1995). In general, the largest effects occur when adapting- and test-odors nominally match (self-adaptation), but adaptation can also occur when they differ (cross-adaptation). Scholars have proposed that cross-adaptation occurs when two odors stimulate overlapping groups of sensory channels or physiological mechanisms (Zwaademaker, 1925; Cheesman and Mayne, 1953; Moncrief, 1956; Cain and Polak, 1992). Could this functional similarity lead to an index of perceived similarity? Cross-adaptation, at least as defined as changes in threshold and reaction time, could qualify as reasonably objective.
Cross-adaptation could yield the desired index if degree of cross-adaptation reflects perceived similarity. Some studies have produced evidence consistent with this notion (e.g. Cheesman and Mayne, 1953; Moncrief, 1956; Todrank et al., 1991; Cain and Polak, 1992). However, phenomenologically similar odors do not always cross-adapt and dissimilar ones sometimes do (O’Connell et al., 1994; Pierce et al., 1995). Further, asymmetric or non-reciprocal cross-adaptation often occurs (Cain and Engen, 1969; Cain, 1970; Köster, 1971; deWijk, 1989; Todrank et al., 1991; Stevens and O’Connell, 1996). Both results suggest that degree of cross-adaptation entails more than sensory similarity. Finally, exposure to one odorant can sometimes enhance sensitivity to another odorant (Engen and Bosack, 1969; Berglund et al., 1971; de Wijk, 1989; Stevens and O’Connell, 1996).
A valid scale of similarity may help investigators understand the complexities of cross-adaptation. In this context, it might prove instructive to compare degree of cross-adaptation with both perceived similarity and similarity of molecular properties (Pierce et al., 1995). This approach, combined with physiological studies of receptors and central neural mechanisms, might shed light on the mechanisms of adaptation and the coding of odor quality. However, investigators have yet to demonstrate a clear relationship between degree of cross-adaptation and perceived similarity.
Adaptation could potentially play a role whenever subjects must make repeated judgements within a session (this caution also applies to the discriminative techniques discussed below). Repeated exposure tends to attenuate intensity, but might also decrease subjects’ ability to discern differences in quality independent of changes in intensity. No empirical evidence exists to support this notion. Subjectively, however, the phenomenon can prove quite compelling. For example, those who work in retail sales of fragrance often caution against sampling too many perfumes in rapid succession, lest they begin to smell alike. Future research could examine the effects of repeated exposure, both to gain a better understanding of olfactory processing and to determine an inter-trial interval that would ensure a reasonable degree of independence between judgements.
Discrimination
At the most fundamental level, a sensory system does two things: it responds to a certain type of energy (radiative, mechanical, chemical) and within that type it resolves diffeences in kind (wavelength, molecular properties). Discrimination measures success of resolution. Success at resolution across many stimuli holds a key to what aspects of the stimulus are important. Beneath the surface, almost all tests of olfactory performance prove to be limited in some measure by the property of discrimination.
In the most basic discriminative paradigm, subjects receive some pairs of odors that do and some that do not differ in quality (Doty, 1991). In other paradigms, subjects receive three odors during a trial (Jones and Elliot, 1975; Eskenazi et al., 1983, 1986; Hormann and Cowart, 1993). In an oddity paradigm, two of the three odors match and subjects must identify the ‘odd’ one. In an ABX paradigm, subjects must chose which of two different ‘comparison-odors’ matches a standard. In all cases, the frequency of incorrect responses represents qualitative similarity provided the odors do not differ in perceived intensity or trigeminal impact (such could serve as cues for discrimination).
Techniques of discrimination minimize or eliminate subjectivity. Accordingly, discriminative results are essentially non-negotiable. Indeed, there would be little disagreement about most results based on discrimination. We are not naive enough to think that data obtained by discrimination can invariably be deposited in some vault, but this is probably more true of such data than of estimates of similarity that purport to indicate how similar rose odor is to peanut odor. Hence, if data on discriminability exist, they could in principle be dovetailed to other discrimination-based data without harm. As tests of performance, rather than tests of mental content, they can have exact counterparts in animal studies. Recent similarities in outcome of discriminations between squirrel monkeys and humans reinforce the virtue of such comparability (Laska and Hudson, 1993; Laska and Freyer, 1997; Laska et al., 1999).
Techniques of discrimination can resolve small differences in quality, even for individuals (Laska and Freyer, 1997; Laska and Teubner, 1999a; Laska et al., 1999). There is no a priori limitation on the number of judgements an individual can contribute to a discriminative result. The data from an individual could in principle be collected with enough precision to distinguish heretofore unexplored subtleties, e.g. those present within family members versus those outside the family. Indeed, if one wanted to know whether another technique allowed for good resolution, one would use a direct discriminative technique as a standard.
The sensitivity of discrimination, an asset for resolution of small differences, becomes a liability if odors differ markedly. Performance in simple discrimination may then lie at asymptotic level. Only a few investigators have used discriminability to assess similarity outside studies of enantiomers, where the issue was whether the odors differed more than how much they differed (Theimer and McDaniel, 1970; Friedman and Miller, 1971; Jones and Elliot, 1975; Hummel et al., 1992; Hormann and Cowart, 1993).
On general grounds, if forced to choose between adequate measurement of small differences versus large differences, one should choose the former. Various investigators have compared a wide variety of odorants in hope of capturing the major dimensions of ‘odor space’ (Schutz, 1964; Woskow, 1968; Yoshida, 1975; Schiffman, 1981). Accordingly, the researchers compared molecules that differed considerably in molecular parameters. This approach assumes that the many candidates for appropriate molecular parameters appear in a breadth necessary to see significant relationships. One might need to study hundreds or thousands of stimuli to uncover the relevant molecular parameters. Mapping changes that occur with gradual, systematic changes in molecular parameters would seem a shorter path to the goal of understanding. Since discrimination resolves small differences in quality, it seems well-suited to this task.
Laska and colleagues have begun to explore the discriminability of substances that differ by methylene groups in aliphatic series (Laska and Freyer, 1997; Laska and Teubner, 1999b; Laska et al., 1999). Within a series, discriminability increased with difference in carbon chain-length, though did reach asymptotic levels at a difference beyond about three carbon atoms. It would be desirable to map trends in quality over a wider range, an entirely achievable goal, via: (i) addition of a new dependent variable to the task of discrimination, without any change in the stimuli themselves; or (ii) increasing the difficulty of the discrimination with a change in the stimuli
A large literature suggests that the time needed to compare two physical stimuli varies inversely with the perceptual distance between them, even beyond the point of perfect discriminability (Cattell, 1902; Henmon, 1906; Kellogg, 1931; Woodworth and Schlosberg, 1954; Crossman, 1955; Welford, 1960; Vickers, 1980). Wise and Cain (Wise and Cain, 2000) found that measures of discriminability based on latency to discriminate correlated strongly with measures based on errors of discrimination, but latency-based measures provided better resolution among odor-pairs. Application of latency-to-discriminate to mixtures endorsed the conclusion that the odor qualities of binary combinations lie approximately midway between the qualities of their components. More importantly, the discriminabilities between pairs of mixtures not previously measured proved predictable from the discriminabilities of their components. Latency to discriminate shows considerable promise as a dependent variable. It has begun to earn its stripes.
In the perceptual ‘space’ between any two odors, one can conceive of a continuum over which one odor becomes gradually transformed into the other by progressive dilution and concentration. Hence, as clove odor becomes increasing diluted with lemon odor its discriminability from clove grows progressively and its discriminability from lemon diminishes. In principle, discriminations between a binary mixture and an unmixed component should prove more difficult when the two components are similar, and less difficult when the two components are not similar. Hence, beginning with two similar odors, progressive dilution of one by the other should reflect itself in poor differential sensitivity (Figure 6) (Duncan et al., 1992; Olsson and Cain, 2000). Variations such as this may need to earn their stripes from collection of data convergent with that of other techniques.
|
One might also increase the difficulty of discrimination by lowering the perceived intensity of the stimuli. Discriminations of hue and pitch become more difficult at lower intensities (Shower and Biddulph, 1931
; Brown, 1951; Harris, 1952; Walraven and Bouman, 1966; Reitner et al., 1992). This undoubtedly holds true for olfactory discriminations as well (Jones and Elliot, 1975; de Wijk and Cain, 1994). Odor quality may change with changes in concentration for some molecules (Moncrief, 1967; Gross-Isseroff and Lancet, 1988). Such effects could complicate matters to a certain extent from the standpoint of measurement, but might also shed light on olfactory coding. Future investigations could determine how the discriminative odor-space, i.e. both absolute and relative differences among odors, changes with alterations of perceived intensity. As mentioned above, future investigations could also determine how the discriminative odor-spaces of individuals differ. To what degree do normal subjects vary? How many subjects will we need to study for a general picture of similarity? Effective measurement of differences in quality might require answers to these questions.
Of course, individual differences provide opportunities as well as challenges. For example, the reduced discriminative capacity of patients with damage to certain areas of the brain might shed light on the processing of odor quality in the normal brain (Eskenazi et al., 1986; Martinez et al., 1993). The discrimination of persons with a specific anosmia [poor sensitivity to a certain chemical despite otherwise normal olfactory acuity (Amoore, 1969, 1970; O’Connell et al., 1989; Stevens and O’Connell, 1991, 1996)] might also prove enlightening, though investigators have yet to define specific anosmia precisely [see (O’Connell et al., 1994) for a discussion], much less explain it.
When considered across the sense-modalities, no area of sensory measurement has received more attention than that of discrimination. The auditory psychophysics of simple and moderately complex stimuli, developed by techniques of discrimination, of matching and of numerical scaling, has succeeded so well as to have reached a point of diminishing returns. The theory of signal detectability (TSD) played a significant role in this accomplishment. TSD offers many opportunities to turn ordinal data of the sort generated in studies of performance into metric indices (MacMillan and Creelman, 1991; Gescheider, 1997). Some such indices have already seen use in olfaction (Cain, 1977; Rabin and Cain, 1989; Swets, 1986; Cain and Potts, 1996). Whereas research could conceivably lead to a metric measure for a descriptive technique, it already offers choices for techniques based upon performance.
|
Conclusions |
|---|
|
TopAbstractIntroductionCriteria for a measureHow have odors been...ConclusionsReferences |
|---|
Understanding the relationship between molecular properties and odor quality arguably poses the single most significant problem in olfaction. It seems curious that in recent times no psychophysical laboratory has dedicated itself to systematic pursuit of the problem. Essential understanding will require more than just psychophysical correlations between structure and activity, whether in humans or other animals. It will require knowledge of interactions between molecules and receptors. Neither pursuit alone, though, will solve the problem, but this need not inhibit any level of approach. Any given molecule will undoubtedly stimulate more than one type of receptor, perhaps even hundreds, with probabilities reminiscent of those of resonance structures. How such stimulation reflects itself in a neural code will require modeling of mechanisms of neural computation at virtually every stage of the olfactory nervous system. How the neural code for this dual discriminative–regulatory modality varies with hunger, thirst, sexual needs, arousal, mood, metabolism, state of health, adaptation, and the presence of other odoriferous and nonodoriferous stimuli will remain issues of importance. Curiously, ancillary issues such as these have received more attention in general than the primary issue of how to define the stimulus in the first order.
Lack of psychophysical direction plays its role in the languor of olfactory structure–activity relationships. Psychophysical studies have rarely built on one another. Data from one study rarely seem trustworthy enough for another investigator to use them. Hence, every study starts anew, as if the first of what might become a long string of studies, but in fact never does (Cain, 1978). How many areas of sensory science can say: ‘We have no archival data. If you bring us a new item [in this case an odorant], we will likely have no idea what its threshold will be and no idea what its perceived quality will be’? Oh, yes, there are trivial cases, a simple mercaptan or ester or amine, where any chemist can guess that the stimulus will smell skunky, fruity or fishy–uriny, something chemists could do well over a century ago, but such trivia aside, the field languishes as much for a way to accumulate data as anything. This is not an essential problem, for it has a solution. It is an ‘accidental’ problem in the Aristotelian sense. Unscrutinized psychophysical methodology has contributed to it and more circumspect measures based upon performance should help solve it. In the pursuit of an answer, we feel responsive to the eminent chemosensory scientist Lloyd Beidler (Beidler, 1976), who noted: ‘Rigor in defining the quantity and quality of the [sensory] response measured must be as good as that used in determining the physicochemical properties of the stimulus molecule if reliable correlations [between structure and activity] are to be expected’ (p. 295). Beidler shared Alexander Graham Bell’s priority.
We have focused principally on choice of a methodology, but we recognize that a methodology based on performance comes at the price of time. Automated presentation of stimuli could reduce the burden and sorely needs development. Devices that test human beings automatically could, with appropriate modification, also test animals. Development of an animal psychophysics of odor quality could greatly speed up accumulation of data, which might conceivably have as much relevance to human perception as human responses themselves. Finally, in our focus on the measurement of activity in structure–activity relationships, we may have seemed to underestimate complexities of the characterization of molecules. Molecules of odorant can be characterized in considerable detail, but the chemical challenge in structure– activity relationships is to decide both the relevant type of characterization and the details. As Chastrette (Chastrette, 1997) indicated, choices for type include: (i) global properties, including boiling point, molar volume and calculated shape; (ii) geometric variables, commonly distances within a molecule; (iii) electrostatic variables; and (iv) fragments and patterns as epitomized in hypothesized osomophoric groups and odortopes (Mori and Shepherd, 1994). To sort out the rules will require testing of hypotheses with both adequate measures of structure and adequate measures of activity. This will undoubtedly require collaboration of scientists on both sides of the structure–activity equation.
|
Acknowledgments |
|---|
Supported by research grant 5 RO1 DC 00284 from the National Institute on Deafness and Other Communication Disorders, National Institutes of Health. We would like to thank three anonymous reviewers and Dr Craig Warren for useful comments.
|
References |
|---|
|
TopAbstractIntroductionCriteria for a measureHow have odors been...ConclusionsReferences |
|---|
Amoore, J.E. (1962a) The stereochemical theory of olfaction 1: Identification of the seven primary odours. Proceedings of the Scientific Section, Toilet Goods Association, No. 37 suppl., 1–12.
Amoore, J.E. (1962b) The stereochemical theory of olfaction 2: Elucidation of the stereochemical properties of the olfactory receptor sites. Proceedings of the Scientific Section, Toilet Goods Association, No. 37 suppl., 13–23.
Amoore, J.E. (1964) Current status of the steric theory of odor. Ann. N.Y. Acad. Sci., 116, 457–476.
Amoore, J.E. (1969) A plan to identify most of the primary odors. In Pfaffmann, C. (ed.), Olfaction and Taste III. Rockefeller University Press, New York, pp. 158–171.
Amoore, J.E. (1970) Molecular Basis of Odor. Charles C. Thomas, Springfield, IL.
Amoore, J.E. and Venstrom, D. (1967) Correlations between stereochemical assessments and organoleptic analysis of odorous compounds. In Hayashi, T. (ed.), Olfaction and Taste II. Pergamon Press, Oxford, pp. 3–17.
Anderson, N.H. (1982) Methods of Information Integration Theory. Academic Press, New York.
Beare, J.I. (1906) Greek Theories of Elementary Cognition from Alcmaeon to Aristotle. Clarendon Press, Oxford.
Beets, M.G.J. (1978) Structure–Activity Relationships in Human Chemoreception. Applied Science Publisher.
Beidler, L.M. (1976) Taste and smell: review of G. Benz (Ed.), Structure– Activity Relationships in Chemoreception. Trends Biochem. Sci., 1, 295–296.
Bentley, M. (1926) Qualitative resemblance among odors. Psychol. Monogr., 35, 144–151.
Berglund, B., Berglund, U., Engen, T. and Ekman, G. (1973) Multidimensional analysis of twenty-one odors. Scand. J. Psychol., 14, 131–137.[ISI][Medline]
Boring, E.G. (1942) Sensation and Perception in the History of Experimental Psychology. Irvington, New York.
Brown, W.R.J. (1951) The influence of luminance level on visual sensitivity to colour differences. J. Opt. Soc. Am., 41, 684–688.
Buck, L. (1996) Information coding in the vertebtrate olfactory system. Annu. Rev. Neurosci., 19, 517–544.[ISI][Medline]
Buck, L. and Axel, R. (1991) A novel multigene family may encode odorant receptors: a molecular basis for odor recognition. Cell, 65, 175–187.[ISI][Medline]
Cain, W.S. (1970) Odor intensity after self-adaptation and cross-adaptation. Percept. Psychophys., 7, 271–275.
Cain, W.S. (1977) Differential sensitivity for smell: ‘noise’ at the nose. Science, 195, 796–798.
Cain, W.S. (1978) History of research on smell. In Carterette, E.C. and Friedman, M.P. (eds), Handbook of Perception: Tasting and Smelling. Academic Press, New York, Vol. VIA, pp. 197–229.
Cain, W.S. (1988) Olfaction. In Atkinson, R.C., Herrnstein, R.J., Lindzey, G. and Luce, R.D. (eds), Stevens’ Handbook of Experimental Psychology: Perception and Motivation. Wiley, New York, Vol. 1, pp. 409–459.
Cain, W.S. and Engen, T. (1969) Olfactory adaptation and the scaling of odor intensity. In Pfaffmann, C. (ed.), Olfaction and Taste. Rockefeller University Press, New York.
Cain, W.S. and Polak, E.H. (1992) Olfactory adaptation as an aspect of odor similarity. Chem. Senses, 17, 481–491.
Cain, W.S. and Potts, B.C. (1996) Switch and bait: probing the discriminative basis of odor identification via recognition memory. Chem. Senses, 21, 35–44.
Cain, W.S., de Wijk, R.A., Lulejian, C., Schiet, F. and See, L.C. (1998) Odor identification: perceptual and semantic dimensions. Chem. Senses, 23, 309–326.[Abstract]
Callegari, P., Rouault, J. and Laffort, P. (1997) Olfactory quality: from descriptor profiles to similarities. Chem. Senses, 22, 1–8.
Cattell, J.McK. (1893) On errors of observation. Am. J. Psychol., 5, 285–293.
Cattell, J.McK. (1902) The time of perception as a measure of differences in intensity. Philos. Stud., 19, 63–68.
Chastrette, M. (1997) Trends in structure–odor relationships. SAR QSAR Environ. Res., 6, 215–254.[Medline]
Chastrette, M., Elmouaffek, A. and Sauvegrain, P. (1988) A multidimensional statistical study of similarities between 74 notes used in perfumery. Chem. Senses, 13, 295–305.
Cheesman, G.H. and Mayne, S. (1953) The influence of adaptation on absolute threshold measurements of olfactory stimuli. Quart. J. Exp. Psychol., 5, 22–30.
Cometto-Muñiz, E.C. and Cain, W.S. (1995) Olfactory adaptation. In Doty, R.L. (ed.), Handbook of Olfaction and Gustation. New York, Marcel Dekker, pp. 257–282.
Coxon, J.M., Gregson, A.M. and Paddick, R.G. (1978) Multidimensional scaling of perceived odour of bicyclo [2.2.1] heptane, 1,7,7-Trimethyl-bicyclo [2.2.1] heptane and Cyclohexane derivatives. Chem. Senses Flav., 3, 431–441.
Crocker, E.C. (1945) Flavor. McGraw-Hill, New York.
Crossman, E.R.F.W. (1955) The measurement of discriminability. Quart. J. Exp. Psychol., 7, 176–195.
Davis, R.G. (1979) Olfactory perceptual space models compared by quantitative methods. Chem. Senses Flav., 4, 21–33.
de Wijk, R.A. (1989) Temporal factors in human olfactory perception. Unpublished doctoral dissertation, University of Utrecht.
de Wijk, R.A. and Cain, W.S. (1994) Odor quality: discrimination versus free and cued identification. Percept. Psychophys., 56, 12–18.[ISI][Medline]
Dimmick, F.L. (1922) A note on Henning’s smell series. Am. J. Psychol., 33, 423–25.
Dimmick, F.L. (1927) The investigation of the olfactory qualities. Psychol. Rev., 34, 321–335.
Doty, R.L. (1991) Psychophysical measurement of odor perception in humans. In Laing, D.G., Doty, R.L. and Breipohl, W. (eds), The Human Sense of Smell. Springer-Verlag, Berlin, pp. 95–151.
Döving, K.B. (1965) Studies on the responses of bulbar neurons of frog to different odour stimuli. Rev. Laryngol. (Suppl.), 845–854.
Döving, K.B. and Lange, A.L. (1967) Comparative studies of sensory relatedness of odors. Scand. J. Psychol., 8, 47–51.[ISI][Medline]
Dravnieks, A. (1974) A building-block model for the characterization of odorant molecules and their odors. Ann. N.Y. Acad. Sci., 237, 144–163.[ISI][Medline]
Dravnieks, A. (1982) Odor quality: semantically generated multidimensional profiles are stable. Science, 218, 799–801.
Dravnieks, A. (1985) Atlas of Odor Character Profiles, Vol. 61. American Society for Testing and Materials, Philadelphia, PA.
Dravnieks, A., Bock, F.C., Tibbets, M. and Ford, M. (1978) Comparison of odors directly and through odor profiling. Chem. Senses, 3, 191–220.
Duncan, H.J., Beauchamp, G.K. and Yamazaki, K. (1992) Assessing odor generalization in the rat: a sensitive technique. Physiol. Behav., 52, 617–620.[Medline]
Engen, T. (1964) Psychophysical scaling of intensity and quality. Ann. N.Y. Acad. Sci., 116, 504–516.
Eskenazi, B., Cain, W.S., Novelly, R.A. and Friend, K. (1983) Olfactory functioning in temporal lobectomy patients. Neuropsychologia, 21, 365–374.[ISI][Medline]
Eskenazi, B., Cain, W.S. and Friend, K. (1986) Olfactory functioning in temporal lobectomy patients. Neuropsychologia, 24, 553–562.[ISI][Medline]
Findley, A.E. (1924) Further studies of Henning’s system of olfactory qualities. Am. J. Psychol., 35, 436–445.
Fráter, G. and Lamparsky, D. (1991) Synthetic products. In Müller, P.M. and Lamparsky, D. (eds), Perfumes: Art, Science, Technology. Kluwer Academic Publishers, Dordrecht, pp. 533–628.
Friedman, L. and Miller, J.G. (1971) Odor incongruity and chirality. Science, 172, 1044–1046.
Gamble, E.M. (1916) Taste and smell. Psychol. Bull., 13, 134–137.
Gamble, E.M. (1929) Psychology of taste and smell: status as of 1929. Psychol. Bull., 16, 566–569.
Gescheider, G.A. (1997) Psychophysics: The Fundamentals, 3rd edn. Lawrence Erlbaum Associates, Mahwah, NJ.
Gregson, R.M. (1972) Odour similarities and their multidimensional metric representation. Multivar. Behav. Res., 4, 165–175.
Gross-Isseroff, R. and Lancet, D. (1988) Concentration-dependent changes of percieved odor quality. Chem. Senses, 13, 191–204.