Exploring the Odorant-Determined, Molecular Basis for Olfactory Perception Volume 52- Issue 2

Roshan Thilakarathne^1,4#, Neranjan Perera^2,5# and Chiquito Crasto³*

• ¹Office of Clinical Transformation, Texas Tech University Health Sciences Center, USA
• ²Institute for Systems Genomics, University of Connecticut, USA
• ³Center for Biotechnology and Genomics, Texas Tech University, USA
• ⁴Current Affiliation: UMC Health Systems, USA
• ⁵Current Affiliation: Connecticut Department of Public Health, USA
• ^#Both authors have contributed equally to the paper

Received: August 02, 2023;   Published: August 16, 2023

*Corresponding author: Chiquito Crasto, Center for Biotechnology and Genomics, Texas Tech University, Lubbock, Texas 79409, USA

DOI: 10.26717/BJSTR.2023.52.008224

Abstract PDF

The mechanistic underpinnings of olfactory receptor (OR)-odorant interactions have yet to be completely understood. This stymies our understanding of how we perceive specific odors, odorants or combinations of odorants. We used computational methodologies to address OR-odorant interactions from a novel perspective-that of the odorant. In this study, odorants that elicit a hedonistic olfactory response were identified. We used machine learning methodologies to cluster 85 odorant molecules that are perfumes, active ingredients in perfumes or fillers in perfumes. Our clustering process used neural networks, which leveraged structural (inter-atomic distances-bonded as well as non-bonded) and electronic (NMR chemical shifts) specific features from our odorant cohort. Chemical shifts were theoretically determined using ab initio Density Functional Theory (DFT)- methods. Intra-molecular distances were obtained following geometry optimization using the above DFT methods. The 85-odorant cohort was then clustered into five sub-groups, based on similar electronic and structural features. To provide this electronic-structural basis for olfactory “hedonistic” perception, we tested whether the perceptions of odors, based on super-smeller responses, were also similarly clustered. Odor- perceptions within the clusters matched odor types with very few overlaps. Outliers arose from odorants that have shown multiple perceptive responses.

This proof-of-concept experiment will go a long way towards elucidating the combinatorial and promiscuous nature of odorant-olfactory receptor interactions. The perspective of the odorant is a novel approach to identifying olfactory receptor-odorant interactions.

The discovery of olfactory receptors has been relatively recent [1]. The impact of this discovery is particularly critical for two reasons: the 2004 Nobel Prize in Medicine and Physiology [2]; and the identification of the olfactory receptor gene family being the largest in all the human and mammalian genomes [3-6]. Olfactory receptors (OR) reside in the cilia of the mucous membrane of the nostrils, each of which is associated with a single olfactory neuron [7]. An olfactory receptor’s interaction with an odorant marks the first step of olfaction. This OR-odorant interaction catalyzes a not well-understood signal transduction process that results in the perception of an odor [8]. The human genome has an olfactory receptor gene family consisting of approximately 400 receptor proteins [3,9,10]. Several hundred receptors are responsible for the discrimination of several thousand odorants and odors (combination of odorants-contributing to a particular odor). Elucidating the mechanism by which these odorants interact with and activate the olfactory receptors would provide molecular-level insights into how one can relate the odorant, its complementarily interacting olfactory receptor and the resulting odor-perception. Functional assessments that would result in the identification of specific odorants that will target one or more OR are limited. There exists no acceptable rational basis for testing a panel of odorants against an OR. Experimental efforts at functionally characterizing ORs have been made, with some success [11,12]. These studies have been few and far between.

Results of these efforts indicate that the nature of OR-odorant interactions is promiscuous in nature: one receptor will interact with several odorants, while several receptors will interact with the same odorant. An overarching mechanism at a molecular level to address these interactions from experimental, functional analyses, however, is stymied. Olfactory receptors are membrane-bound and-because of the nature of these proteins-are difficult to express and purify. The crystal structure of an olfactory receptor is currently not available. Several studies have used homology modeling methodologies to develop structures of ORs. Some of these studies also involve molecular dynamics simulations of the interactions between ORs and odorants [13-16]. Here, we used a novel approach towards elucidating OR-odorant interactions, from the perspective of the odor. Our aim was to identify aspects of an odorant that are likely to bind specific ORs–namely the odorant’s structural and electronic features. Our methodologies went beyond merely looking at functional groups or overall molecular size or structure of the odorant. We identified specific interatomic distances and electronic features associated with atoms that are common to specifically perceived odors. To achieve this, we rigorously optimized the molecular geometries and performed electronic charge distribution studies of perfume odorants using quantum chemistry calculations using density functional representations of the atom in molecules.

Variants of these methodologies have been used in drug design-- to identify features of a putative pharmaceutical product that is likely to bind and activate a receptor protein--an interaction that mirrors that of an odorant and olfactory receptor. Indeed, most protein receptors for drug-compounds are Class A GPCRs. This is the presumed structure of an olfactory receptor [17]. We used machine learning ideations that leveraged the intra-molecular distances and the NMRshifts of 85 odorant molecules from quantum chemistry studies to cluster odorants. Five clusters were generated. The odorant molecules in each cluster were then matched to their perceptions–which were determined by “smell experts” from companies that manufactured these odorant compounds.

Figure 1

Identification and Acquisition of Perfume Odorants

Eighty-five odorant molecules were identified from a search using the term “perfume” from the Good Scents resource http://www. thegoodscentscompany.com/. (Figure 1) shows a screen capture of a partial page for the odorant, Citronellol (Chemical Abstracts Service [CAS] number: 106-22-9; IUPAC name: 3,7-dimethyloct-6-en-1-ol). The Good Scents Company is a comprehensive resource that informs the users of the chemical nature of the compound. It also lists the perceived odor and flavor of the compound from all available sources, as determined by smell and taste experts. For Citronellol, the perceived odor information is from two different sources. Source 1: floral, leathery, waxy, rose, and citrus. Source 2: floral, rose, sweet, citrus, green, fatty, and terpenic. The flavor for this compound is listed as: floral, rose, sweet, green, fruity, and citrus. (http://www.thegoodscentscompany. com/data/rw1007032.html-- Figure 1). For every odorant identified using the search term “perfume” from The Good Scents resource, the perceived odors were cataloged. (Table 1) lists all the compounds and their perceived odors. (Table 2), the converse of (Table 1), lists the compounds for the different perceived odors. The three-dimensional structural coordinates for every compound were downloaded from the PubChem (https://pubchem.ncbi.nlm.nih.gov/) resource in the SDF (Structure Data Files) format.

Table 1: The odorants selected based on a search on keywords “perfume” in The Good Scents resource and the odor perceptions from every available resource for each odorant.

Table 2: The perceived odors of odorants selected based on a search on keywords “perfume” in The Good Scents resource. This is the converse of (Table 1). 162 unique perceived odors were identified for the 85 odorants.

Quantum Chemistry--NMR and Geometries

Coordinates of the SDF files were used as a starting point for the geometry optimization calculations. These SDF files were converted into x, y, z coordinates using the Open Babel resources (https://openbabel. org/) and then to Gaussian input files using customized bash scripts. Density functional theory calculations were carried out on the initial structures using the software Gaussian (version 16.A.03) [18] with the hybrid functional B3LYP and basis set 6-31G(d) for all the atoms, first to optimize the geometry and then to determine NMR chemical shifts for the carbon and nitrogen atoms. Shielding and isotropic coupling constants were computed using the gauge-including atomic orbitals (GIAOs) implemented in the Gaussian software. Isotropic shielding values were extracted, and the chemical shift values were calculated for each of the atoms. Optimized structures were then converted into SDF files and used for further clustering analysis [19].

Creating Feature Matrix

The data from the ab initio, quantum chemistry experiments were integrated into a feature matrix, which was the result of the elements of three matrices:

1) An adjacency matrix stored the bonded connectivities of the atoms from the odorants using Python libraries, Spektral and NetworkX;
2) A distance matrix comprised all intra-molecular distances (irrespective of whether the atoms were bonded or not) using the Python library, molmod; and
3) An NMR matrix. The NMR matrix elements were binned into blocks of chemical shifts of 20 ppm (parts per million). For Carbon- 13, the blocks ranged from 0 ppm to 220 ppm. 20 ppm blocks were also used for Nitrogen-15. To account for the entire range of chemical shifts for nitrogen, the NMR matrix included nitrogen bins ranging from 0 ppm to 400 ppm. All the matrices were combined into a single matrix. This features matrix comprised 1638 features.

Clustering

In our customized Python program, we used three neural networks: a Graph Attention Neural Network (GANN) that mapped an odorant’s atomic connectivities from its positional coordinates [20]; a deep neural network (DNN) [21] that comprehensively extracted interatomic-bond distances; and a DNN that extracted electronic features of bonded and surrounding atoms within an odorant from the 13C and 15N NMR chemical shifts. We used K-means unsupervised learning algorithm to cluster the structural and electronic features from the feature matrix. The algorithm groups observations with similar attribute values together by measuring the Euclidian distance between points. As proof of concept, we tested the system on creating five clusters from the 85-odorant list.

(Table 3) illustrates the distribution of the odorant molecules among the five clusters. Cluster 1 had four molecules, cluster 2–by far the largest–had 60, cluster 3 had 16, cluster 4 had four, and cluster 5 had six odorants. For each cluster, the top ranked perceived odors are highlighted in (Table 4). In cluster 1–the smallest cluster–the representative odors are woody and powdery. In cluster 2, the top-ranking odors are sweet, green, camphereous, floral, herb and spicy. For cluster 3, the odors are floral, woody, powdery and amber. For cluster 4, the top-ranking are powdery, animal, fruity and musk. For cluster 5, the odors related to specific fruits predominate: apple, fruity, pineapple, and creamy.

Table 3: The distribution of the cohort of odorants in the five clusters following clustering analysis.

Table 4: The top ranked perceived odors for each cluster following clustering analysis.

To put the perceived odors into perspective of our methodology, it means that specific structure-electronic features of the molecules, which were the data fed through the feature matrix into the machine learning system are reproduced across the odorants that are in each cluster. For the electronic features, the NMR chemical shifts were considered for C-13 and N-15 atoms. NMR chemical shifts for H-1 were not determined. None of the cohorts of odorants had a phosphorus atom. The chemical shifts were binned into 20 ppm blocks and covered the entire range of the chemical shift spectrum in ppm–observed for every possible connectivity and electronic environment for carbon and nitrogen odorant atoms. Odorants that elicit a specific response, e.g., a fruity odor, possess one or more atom-pairs, each of which are separated by the same distance and have the same NMR chemical shift. The structure matrix consisted only of all intra-molecular distances and intra-molecular bond connectivities. No information related to functional groups, size of molecule, or structural features such as straight chains or rings were included. We explored the possibility that the reproducing “atom-pairs” as mentioned above, responsible for a perceived odor, go beyond bonded atom-pairs. The eventual goal is to identify specific atoms-pair(s) as being responsible for generating a specific olfactory response. (Table 4) only includes perceived odors that had high counts. This meant that a particular odor was elicited by most of the odorants in the cluster.

(Table 1) shows every possible odor identified by a smell-expert from The Good Scents resource, which was used in this work. Some odorants, e.g. Citronellol, have several odors (Figure 1), some of which are closely related and conceivably have perceptual overlaps, e.g., green vs. herb-like, or, sweet vs. floral. Some of the perceived odors also overlap across clusters (Table 4). Benzyl acetone (Cluster 2) has been identified as having the odors of: floral, aromatic, sweet, balsamic, jasmine, herbal, fruity, and chemical. Benzyl alcohol, from the same cluster has the odors: floral, phenolic, balsamic, and rose. In each case, the experts have been able to discern between floral and jasmine, and between floral and rose. The Good Scents resources identifies the odor perception based on the source of the expertise–and indeed, these differ, in some cases, depending on the expert. It, however, does not identify the specific concentration of the odorant used during the smell test. There is evidence that odor-perception changes with concentration [21,22]. An increasing body of evidence shows that odorant perception changes between demographic populations and often depends on the geographical region of an individual [23,24]. Epigenetic factors also likely influence odor perception. Expert smellers can presumably discriminate between can distinguish between green and herbal, oily and fatty, sweet and floral. But it is likely that the perceptions of highly similar odors are weighed by opinion.

Odors identified as watery (lilial, helional, beta-pinene, bourgeonal, cyclamen) or fresh (phenethyl alcohol, perillaldehyde, nerol, menthone, hydroxycitronellal, heptyl_acetate, helional, gamma-decalactone, benzyl_acetate, bourgeonal, calone, citral, cyclamen) are also vague, and likely vary among smell and taste experts. One possible future remedy is to only include as perceived odors for an odorant those that are common among experts. There is an overlap in some odors between clusters: woody (Cluster 1 and 3), powdery (Cluster 1, 3 and 4), and fruity (Cluster 4 and 5). Cluster 2 with many odorants is distinguishable because the top-ranking odor perceptions are generally sweet-smelling. Cluster 5 is distinguishable because of the predominance of specific fruit odors, in addition to the generic “fruity” odor. Our clustering software allows the programmer to determine the number of clusters in which to place the odorants. For this proofof- concept effort, we chose the creation of five clusters. It is likely that some of the overlap is because of the limited number of clusters which forced slightly different odorants into the same cluster. Future work would necessitate the need of re-clustering the odorants into 10, 15, 20 and 25 clusters to ascertain the ideal number of clusters for an odorant cohort of this size, while fine tuning the software’s ability to discriminate subtler electronic-structural differences. Another possible methodology would be to create 3-5 clusters of odorant molecules by iterating the clustering analysis over 25-50 random selected odorants out of the 85 tested.

This randomization process would remove any biases in determining the exact number of clusters needed. In depth analysis would also be required to pinpoint the exact electronic structural feature of the odorant specifically responsible for a olfactory perceptive response.

We present here a novel methodology of correlating electronic and structural features in odorants with odor perception. Our atom- pairs are not restricted to bonded atoms. This means that we can identify features that are not obvious from merely observing the overall structure of an odorant. These notions are like those advanced in the drug-design of putative pharmaceutical products domain. Drug design methods use Quality Structure Activity Relationships QSAR [25] and Spectral Data Activity Relationship (1-D, which used NMR, and 3-D, which used distances) to associate electronic structural features with drug-activity [26,27]. In the pharmaceutical industry, a drug’s activity is determined by quantification of pharmacophore or toxicophore effects of the compound on human subjects. In drawing a parallel between these methodologies and the ones described in this paper, activity in the case of odorants is the elicited perceived odor from the perspective of the expert-smeller. We have shown that our system can discriminate between perceived odors. One issue that needs to be further addressed is fine-tuning the methodology to determine the ideal number of clusters per odorant cohort and pinpointing specific structural features responsible for specific olfactory responses.

1. Buck L, Axel R (1991) A novel multigene family may encode odorant receptors: a molecular basis for odor recognition. Cell 65(1): 175-187.
2. Burbach JP (2004) Nobel Prize for Physiology or Medicine for research into smell receptors and the organization of the olfactory system]. Ned Tijdschr Geneeskd 148(52): 2576-2579.
3. Niimura Y, Nei M (2003) Evolution of olfactory receptor genes in the human genome. Proc Natl Acad Sci U S A 100(21): 12235-12240.
4. Malnic B, Godfrey PA, Buck LB (2004) The human olfactory receptor gene family. Proc Natl Acad Sci U S A 101(8): 2584-2589.
5. Olender T, Lancet D, Nebert DW (2008) Update on the olfactory receptor (OR) gene superfamily. Hum Genomics 3(1): 87-97.
6. Ache BW (2020) Position Review: Functional Selectivity in Mammalian Olfactory Receptors. Chem Senses 45(7): 503-508.
7. McClintock TS, Khan N, Xie C, Martens JR (2020) Maturation of the Olfactory Sensory Neuron and Its Cilia. Chem Senses 45(9): 805-822.
8. Glezer I, Malnic B (2019) Olfactory receptor function. Handb Clin Neurol 164: 67-78.
9. Zozulya S, Echeverri F, Nguyen T (2001) The human olfactory receptor repertoire. Genome Biol 2(6): RESEARCH0018.
10. Glusman G, Yanai I, Rubin I, Lancet D (2001) The complete human olfactory subgenome. Genome Res 11(5): 685-702.
11. Malnic B, Hirono J, Sato T, Buck LB (1999) Combinatorial receptor codes for odors. Cell 96(5): 713-723.
12. Mainland JD, Li YR, Zhou T, Liu WL, Matsunami H, et al. (2015) Human olfactory receptor responses to odorants. Sci Data 2: 150002.
13. Lai PC, Crasto CJ (2012) Beyond modeling: all-atom olfactory receptor model simulations. Front Genet 3: 61.
14. Charlier L, Topin J, de March CA, Lai PC, Crasto CJ, et al. (2013) Molecular modelling of odorant/olfactory receptor complexes. Methods Mol Biol 1003: 53-65.
15. Vaidehi N, Floriano WB, Trabanino R, Hall SE, Freddolino P, et al. (2002) Prediction of structure and function of G protein-coupled receptors. Proc Natl Acad Sci U S A 99(20): 12622-12627.
16. Cong X, Topin J, Golebiowski J (2017) Class A GPCRs: Structure, Function, Modeling and Structure-based Ligand Design. Curr Pharm Des 23(29): 4390-4409.
17. Gaussian 16, Revision C.01, Frisch, M. J , Trucks, G. W , Schlegel, H. B , Scuseria, G. E , Robb, MA , Cheeseman, JR , Scalmani, G, Barone, V , Petersson, GA , Nakatsuji, H, Li, X , Caricato, M, Marenich, AV, Bloino, J, Janesko, BG, Gomperts, R, Mennucci, B, Hratchian, HP, Ortiz, JV , Izmaylov, A. F , Sonnenberg, J. L , Williams-Young, D , Ding, F , Lipparini, F , Egidi, F, Goings, J, Peng, B , Petrone, A, Henderson, T, Ranasinghe, D, Zakrzewski, VG , Gao, J, Rega, N, Zheng, G, Liang, W, Hada, M, Ehara, M, Toyota, K, Fukuda, R, Hasegawa, J, Ishida, M, Nakajima, T, Honda, Y, Kitao, O, Nakai, H, Vreven, T, Throssell, K, Montgomery, JA, Jr , Peralta, JE, Ogliaro, F, Bearpark, MJ, Heyd, JJ, Brothers, EN, Kudin, KN, Staroverov, VN, Keith, TA, Kobayashi, R, Normand, J, Raghavachari, K, Rendell, AP, Burant, JC, Iyengar, S.S, Tomasi, J, Cossi, M, Millam, J. M, Klene, M, Adamo, C, Cammi, R, Ochterski, JW, Martin, RL, Morokuma, K, Farkas, O , Foresman, JB, Fox, DJ, Gaussian, Inc., Wallingford CT, 2016.
18. Mehta R, Lavingia K, Khanpara P, Dulera V (2020) Graph Neural Network based Recommender System. In: 2022 7^th International Conference on Communication and Electronics Systems (ICCES) Pp. 1377-1381.
19. LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521(7553): 436-444.
20. Mainland JD, Lundström JN, Reisert J, Lowe G (2014) From molecule to mind: An integrative perspective on odor intensity. Trends Neurosci 37(8): 443-454.
21. Gross Isseroff R, Lancet D (1988) Concentration-dependent changes of perceived odor quality. Chem Senses 13(2): 191-204.
22. Demattè ML, Endrizzi I, Biasioli F, Corollaro ML, Zampini M, et al. (2011) Individual variability in the awareness of odors: Demographic parameters and odor identification ability. Chemosensory Perception 4: 175-185.
23. Trimmer C, Keller A, Murphy NR, Snyder LL, Willer JR, et al. (2019) Genetic variation across the human olfactory receptor repertoire alters odor perception. Proceedings of the National Academy of Sciences 116(19): 9475-9480.
24. Tropsha A (2010) Best practices for QSAR model development, validation, and exploitation. Molecular informatics 29(6‐7): 476-488.
25. Slavov SH, Wilkes JG, Buzatu DA, Kruhlak NL, Willard JM, et al. (2014) Computational identification of a phospholipidosis toxicophore using (13)C and (15)N NMR-distance based fingerprints. Bioorg Med Chem 22(23): 6706-6714.
26. McPhail B, Tie Y, Hong H, Pearce BA, Schnackenberg LK, et al. (2012) Modeling chemical interaction profiles: I. Spectral data-activity relationship and structure-activity relationship models for inhibitors and non-inhibitors of cytochrome P450 CYP3A4 and CYP2D6 isozymes. Molecules 17(3): 3383-3406.
27. Stoyanova Slavova IB, Slavov SH, Buzatu DA, Beger RD, Wilkes JG, et al. (2017) 3D-SDAR modeling of hERG potassium channel affinity: A case study in model design and toxicophore identification. J Mol Graph Model 72: 246-255.