Subscriber access provided by HACETTEPE UNIVERSITESI KUTUPHANESI
Perspective
Encoding odorant images Madeleine Marie Rochelle, Géraldine Julie Prévost, and Terry Edward Acree J. Agric. Food Chem., Just Accepted Manuscript • DOI: 10.1021/acs.jafc.6b05573 • Publication Date (Web): 12 Mar 2017 Downloaded from http://pubs.acs.org on March 14, 2017
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Journal of Agricultural and Food Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.
Page 1 of 30
Journal of Agricultural and Food Chemistry
Computing Odor Images Madeleine M. Rochelle Cornell University Food Science Department 114 Tower Road Ithaca , NY 14853
[email protected] Géraldine Julie Prévost 1 Cornell University Food Science Department 114 Tower Road Ithaca , NY 14853
[email protected] Terry E Acree * Cornell University Food Science Department, 347 114 Tower Road Ithaca , NY 14853
[email protected] 1
present address Procter & Gamble Services Company Temselaan 100 1853 Strombeek-Bever Belgium
1 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 2 of 30
1
Abstract
2
This perspective examines psychophysical methods that may reveal the algorithms that
3
encode odor images by integrating current data from sensory measurement into a
4
computational model of odor perception. There is evidence that algorithms used by the
5
nervous system to process odor sensations require input from only a few odorants,
6
between 3 and 8.1-7 Furthermore, the number of recognizable odors in foods that
7
contribute anything to the aroma of all foods is approximately 250.8 This may imply that
8
it is the ratio of a small number of key odorants (KOs) that create a multitude of food
9
odor. Studies with large mixtures of odorants (formulated to be of equal potency) show
10
that a subject’s ability to detect individual odorants in these mixtures was vanishingly
11
small. These large mixtures had weak and nondescript but similar odor character. If only
12
a few stimulants are used to represent complex images, it is direct evidence of the
13
simplicity and therefore the tractability of the computational process.9
14
Keywords: sniff olfactometry, odor image, odorant mixtures, Laing limit, olfactory white
2 ACS Paragon Plus Environment
Page 3 of 30
Journal of Agricultural and Food Chemistry
15
Introduction
16
Multitudes of odorants surround us every day. From fresh cut grass to a pot of brewing
17
coffee, the odors we encounter in our daily routines capture our attention, modify our
18
memories, and shape our experiences. It is a truth universally acknowledged that there
19
are hundreds of odorants working behind the scenes, subliminal, while still activating
20
our olfactory receptors. Nonetheless, a number of issues remain unresolved in this
21
context, in particular the relation between the olfactory stimulus and percept.
22 23
In a seminal study from 1989, David Laing tested the human ability to identify individual
24
components in complex odor mixtures. For this, 127 human subjects were trained to
25
associate labels with 7 odorants until they could do so accurately. The subjects were
26
asked to identify all individual elements in the mixtures, containing up to 5 of the 7
27
odorants. The frequency at which subjects could correctly identify both odorants in
28
binary mixtures was less than 35 %, all three in tertiary mixture was less than 14%, and
29
identifying all 4 in a quaternary mixtures was an insignificant 4%.1 However, Laing’s
30
subsequent comparative studies with 10 more rigorously trained subjects yielded
31
different results. Here, subjects were directed to perform a different task, namely to
32
identify a single odorant in a complex mixture, and they could perform this task
33
accurately for mixtures containing up to 8 odorants. 5, 10-12
34 35
This represents an example of one of the three levels of analysis described for the study
36
of visual perception, as described by the visual scientist David Marr. These three levels
37
involve a “computational” element, referring to what the system does, an “algorithmic” or
3 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
Page 4 of 30
38
rule-based one that apply to the input during computation to yield an output, and the
39
“implementation,” meaning the biology that does the work (the “wetware”).9 If we apply
40
this model to olfaction, the few odor limits to the analysis of mixtures appears as a
41
computational limit that is imposed by some algorithm and that allows us to ignore most
42
odorants in a mixture, even though these odorants are above their threshold and
43
activating receptors. In addition to such filtering effects, the perception of a mixture of
44
odorants in small numbers is modulated by associative learning. For example, the
45
enhanced ability to discriminate binary mixtures after association of one of the pairs with
46
aversive shocks and the shift in odor quality when two of the odorants are paired
47
together in a binary mixture indicates the profound and perhaps unavoidable effects of
48
associative learning on odor mixture perception.7, 13-16 Somewhere, in the
49
neuroanatomy are cellular implementations and neuronal connections housing the
50
algorithms that produced these computational results.9, 17 Noticeably, the reproducibility
51
of psychophysical experiments indicate that the olfactory system operates on a robust
52
encoding-decoding process.18, 19 The psychophysical experiments described here are
53
intended to provide insight into the computational behavior of a sensing organism and to
54
define the algorithms that underlie them. Central to these experiments is the
55
standardization of experimental techniques and parameters to assure a sufficient
56
comparability of results obtained from different models.20
57 58
Recognizable odorants in mixtures.
59
According to Laing (Figure 1), after the limiting number of 3, there is a sharp drop in
60
the ability to correctly identify all components of a mixture with significance. Moreover,
4 ACS Paragon Plus Environment
Page 5 of 30
Journal of Agricultural and Food Chemistry
61
the ability to recognize a single odorant in presence of 9 to 60 odorants is
62
insignificant.1, 21 Training and expertise showed little to no effect on the number of
63
recognizable odorants.4, 22 While the type of odorants used did not change the number
64
of odorants detected, familiarity with the odorants increased identification accuracy,
65
indicating that memory, as well as odorant concentration, contribute to the rapid
66
identification of odorants in mixtures.1, 4 This remains true even when the odor is
67
misidentified as something familiar. In multiple studies, reviewed in “Learning to Smell,”
68
the relationship between familiarity and perceived intensity of an odor were directly
69
correlated.23 This restricted capacity of the human olfactory system to recognize more
70
than a few odorants in a mixture may seem like a weakness at first. However, it was
71
suggested that, "Kthe apparent inability may in fact reflect a highly efficient neural
72
encoding mechanism which facilitates the rapid discrimination and identification of
73
multicomponent object odors in the environment." 4, 24 From this perspective, further
74
analysis of such a limit to the number of odorants may account for particular analytical
75
components in odor perception that could prove integral in the future of the study of
76
olfaction and of enormous practical importance to the flavor industry.25
77 78
A limit of recognizable odorants implies that any real complex odor mixture can be
79
recreated with just a few odorants, something that perfumers and flavorists are
80
convinced they do every day. There is considerable evidence that a few odorants in a
81
complex mixture can be detected analytically and/or they can also be configured into a
82
single perception from this pattern of individual responses to each of the component
83
odorants. In philosophical terms, the whole may be more than the sum of its parts. For
5 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
84
example, a study of tertiary mixtures of odorants that smelled like pineapple to
85
humans, while none of the individual components had a pineapple smell on their own,
86
is evidence that mixtures of odorants can be perceived as a single configuration of
87
elements (pineapple) instead of a pattern of elemental odors (violet, caramel, and
88
strawberry).26, 27 Similar results were obtained with newborn rabbits using the same
89
three compounds. 28-30 The ongoing question is how processing such few odorants
90
can explain the discriminatory skill of a sommelier to identify wine, the ability of many
91
animals to use odorant mixtures to understand their environment, and the models of
92
olfactory processing emerging from neurobiology9. In other words, how do organisms
93
decode odor mixtures? How do they use salient and subliminal information extracted
94
from a whiff to stimulate perception and behavior? An answer to this question must be
95
able to explain the 3 KO phenomena. Also, it must account for the related issue of the
96
simultaneous suppression of odorants in mixtures and their adaptation in sequential
97
presentations.
Page 6 of 30
98 99
Suppression and Adaptation
100
It is likely that the limit on the number of recognizable odorants in mixtures is due to
101
the interaction between odorant signals in the network of neurons that process
102
odorants into odors. Suppression and adaptation are among the chief modulating
103
effects observed many times in the study of mixture perception, and these effects
104
indicate that odorants interact with each other in different ways. When this
105
phenomenon was examined in binary mixtures both components remained detectable,
106
but some features of each component were suppressed with the addition of others.31
6 ACS Paragon Plus Environment
Page 7 of 30
Journal of Agricultural and Food Chemistry
107
Notably, the suppression and adaptation in odor mixtures can be somewhat predicted
108
from the similarities and differences in odorant qualities. For example, the structurally
109
similar “citrus” smelling odorants C8, C10, C11 n - aldehydes, cross-adapted each
110
other but did not adapt to the green smelling and structurally similar n - C6 aldehyde,
111
hexanal.32 Furthermore, the 3 similar smelling “citrusy” aldehydes do not suppress
112
each other in mixtures while they do suppress hexanal, and vice versa.33 It is not
113
surprising then that the rI7 receptor, first de-orphanized in 1998, has all the citrus
114
smelling odorants in its receptive field while hexanal is not.34 Examining the interaction
115
of C6, C8, and C10 aldehydes in binary mixtures, the similar smelling C8 and C10
116
aldehydes cross-adapt but the dissimilar smelling aldehydes, C6 and C8, suppressed
117
each other when mixed. It seems that odorants cross-adapt when they smell alike and
118
suppress each other when they do not.33 At the very least, this implies that the
119
processes that regulate suppression are somewhat different from those that govern
120
adaptation. In other studies of binary odorant mixtures the overall intensity was less
121
than the sum of the intensities of the odorants individually, indicating suppression, but
122
always stronger than the mean intensities of the individual odorants and follow a vector
123
model of addition.35, 36
124 125
Non-human models also show non-additive effects of odorants in mixtures. For
126
example, rats are instinctively attracted to, or repelled by, specific odors for safety and
127
reproductive purposes. However, when these attractive and aversive odorants are
128
combined in mixtures, the rats respond more to the attractive odorant, indicating that
129
the attractive odorant is suppressing the aversive odorant.37 The effect of suppression
7 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
130
and adaptation can also be seen in olfactory studies related to time. In both humans
131
and non-human models, it has been shown that when faced with mixtures of similar
132
smelling odorants, subjects tend to take longer to identify the individual odorants
133
present in the mixture. Perhaps suppression is occurring in mixtures of similar
134
odorants, making the analytical process of identifying odorants more difficult.38
Page 8 of 30
135 136
Mixture perception
137
Contrary to the phenomena that limits the number of detectable odorants in a mixture,
138
humans seem to be unable to detect a single component in mixture of 8 different
139
odorants when they are at the same odor intensity.5 This phenomenon was more
140
dramatically demonstrated when a mixture of 60 different odorants was prepared at
141
concentrations of similar odor intensities. None of the component odorants could be
142
recognized in the mixture, but there was a faint nondescript smell. Furthermore, when
143
the mix of 60 was subdivided into 2 random mixtures of 30 components they both
144
smelled the same.21 It may seem that the computational mechanisms of the human
145
olfactory system process the signals from complex mixtures through a small subset of
146
the individual components that are present at an odor potency somewhat larger than
147
the remainder of the constituents. Gas chromatography-olfactometry (GCO) data
148
published in the last 30 years shows the same pattern in natural products. A few Key
149
Odorants (KO) dominate natural product GCO data, as supported by the publications
150
(~ 900) that make up the Flavornet39. Furthermore, an analysis of 119 publications that
151
fit rigorous criteria for odor activity involving 220 foods, only 230 unique odorants were
152
found.8 There are over 113,000,000 possible distinguishable odorant patterns from a
8 ACS Paragon Plus Environment
Page 9 of 30
Journal of Agricultural and Food Chemistry
153
combination of any 4 of these 230 food odors – more than enough patterns to encode
154
ecologically important features of any organism’s olfactory space. This does not even
155
include the number of patterns that can be created with different ratios of the 4
156
odorants. For this reason, a small number of stimulants may be all that is needed to
157
encode complex images, and the limit on recognition may be evidence of the
158
simplifying of algorithms involved in computational processing.
159 160
To the extent that the encoding process is similar in rats and humans, maps of the
161
neural projections from the glomerulus to the anterior piriform cortex indicate that
162
connections between these two bodies are unique but not ordered in any simple way,
163
e.g. the map in the main olfactory bulb is not a major feature of the “cortical response
164
mosaic”. However, reproducible psychophysical behavior indicates that a robust
165
encoding – decoding process must be in operation.18, 19 Although this perspective
166
examines odor perception in terms of top-down processing, recent experiments with
167
heterologous expression systems using recombinate odorant mixtures for “butter”
168
containing the three KOs, diacetyl, butanoic acid, and racemic δ-decalactone showed
169
that 1.) the receptor activity pattern of the recombinate did not just appear as the sum
170
of patterns of the single compounds (at least in class-I ORs), and that 2.) at the level
171
of a single receptor response, there was a synergism of the single compounds in the
172
binary mixtures and in the recombinate.40 It is likely that evidence for the
173
computational processes will also be visible at the receptor level, in the peripheral
174
nervous system, or even in the circulatory system where these cultured butter KOs are
175
expressed in leukocytes.40 The challenge will be to determine how humans process
9 ACS Paragon Plus Environment
Journal of Agricultural and Food Chemistry
176
relatively simple mixtures of odorants first into a simple input code and then into more
177
complex output: a multitude of unique odor objects representing our olfactory space.
Page 10 of 30
178 179
Sniff Olfactometry
180
Most research on odor mixture perception in humans has relied on the correlation of
181
psychophysical and sensory measurements of intensity with measures of odorant
182
concentration. These methods tend to generate data on individuals very slowly. As
183
pointed out by Wilson & Stevenson “[There is a] ... need to develop new approaches to
184
testing olfactory discrimination that enhance both the sensitivity and speedK [Most
185
sensory tests] ...are very time consuming and yield relatively little data per participant.
186
Techniques such as those pioneered by Rabin and Cain41, which involve the
187
identification of a target odor in a mixture are the sort of thing we have in mind.” 23
188 189
Sniff Olfactometry (SO) was developed to address these questions. To improve the
190
study of odor images in humans, stimulated by simple mixtures of key odorants, an
191
olfactometer was used to deliver defined compositions with minimal stimulus exposure
192
to 98%), and 2-ethyl-3,5-dimethylpyrazine CAS number 27043-
204
05-6 (>95%) (2E3,5DP) were from Sigma Aldrich (St Louis, USA). Solutions were
205
made in distilled water containing 10% v/v ethanol (food grade). Test solutions ranged
206
from 1 ppm to 200 ppm for MAL, 10ppm to 1000ppm MOL and 50ppb to 300ppb
207
2E3,5DP.
208 209
Subjects – Two female subjects both members of the lab were tested
210 211
Sniff Olfactometer - The design criteria for the SO was based on 250 ml PFA
212
squeeze bottles.46 The PFA exhibits low odorant absorption, can contain a model
213
bolus to represent retronasal smell, a headspace designed to deliver any odorant
214
concentration in air released from model mixtures, or material representing many
215
different ecological odorant sources: foods, beverages, ingredients, etc. The PFA
216
bottles are easily managed when pre-installed in a 3-bottle assembly that maximizes
217
sample exchanges (