An Electronic Nose Analysis of the Headspace from Extra-Virgin Olive Oil–Saliva Interactions and Its Ability to Differentiate Between Individuals Based on Body Mass Index

Abstract

The interaction between fatty foods and saliva in individuals of different body weights may lead to differences in the release of volatile compounds in the mouth. This study investigates the ability of an electronic nose (E-nose) to discriminate between the headspace profiles of extra-virgin olive oil (EVOO) mixed with the saliva of 55 subjects of different body mass indices (BMI). The resulting data were analysed using linear discriminant analysis (LDA) and principal component analysis (PCA) to evaluate the E-nose’s ability to discriminate between groups. W5S, W1S, W2S, and W2W sensors exhibited the greatest variation in response intensity; in particular, they highlighted differences between obese and non-obese subjects. The LDA plot demonstrated a clear separation of samples corresponding to three BMI groups, with the first and second components accounting for 61.25% and 23.97% of the variance, respectively. Overall, the percentage of correct classification in the cross-validation results was 87.3%. These results highlight the potential of an electronic nose for use as a rapid and objective tool for screening olfactory profiles associated with food matrix–saliva interaction in different BMI groups, providing valuable insight for further research on food–saliva interactions.

1. Introduction

The orosensory perception of dietary fat involves a complex multisensory integration of olfactory, gustatory, and somatosensory signals [1]. A key factor in this process is human saliva, which plays a crucial role in fat perception [2] and has the potential to interact with all macronutrients due to its diverse enzymatic composition, including amylolytic, proteolytic, and lipolytic activities [1,3,4,5].

Although the limited contact time of food with saliva in the mouth may reduce the overall impact of these enzymes, enzymatic activity can still significantly influence the release of aroma compounds in a lipid matrix, such as virgin olive oil [6,7,8]. Olfactory perception is linked to the decomposition products of triacylglycerols and polyunsaturated fatty acids, with saliva’s antioxidant or pro-oxidant buffering capacity possibly aiding in the detection of oxidation-related notes [9]. Furthermore, variation in saliva composition contributes to different perceptions and enjoyment of the food itself [10].

It is well-established that lipase and α-amylase activity is higher in individuals with a higher body mass index (BMI). Specifically, individuals with a BMI between 25 and 29.9 (overweight) exhibit higher lipase activity compared to those with a BMI below 24.9 (healthy weight) [11]. Another study examined the intra- and inter-individual variability of salivary biochemicals, such as air flow, lipolysis, lipocalin, proteolysis, and total antioxidant status during food contact, linking these activities to fat perception and preference [3].

In individuals who are obese, saliva demonstrates elevated levels of total antioxidant activity, total proteins, albumin, transglutaminase E, and lactotransferrin. In contrast, proteins, such as cystatin-S and -SN, as well as cathepsin G, are more abundant in the saliva of individuals of a healthy weight [10]. The interaction of fat-rich foods like extra-virgin olive oil (EVOO) with human saliva in individuals of a healthy weight facilitate a greater release of C5 compounds, whereas saliva from individuals who are overweight and obese show increased levels of C6 compounds [12]. This variation in volatile compound profiles led us to hypothesize the potential for an electronic nose (E-nose) to rapidly discriminate between individuals in different BMI categories by analysing the headspace profiles generated when their saliva interacts with fat-based foods like EVOO.

An E-nose uses electronic chemical sensors to detect specific groups of volatile molecules, often employing multiple sensors to record various molecular groups [13]. The majority of medical applications have focused on using an E-nose to discriminate between the breath of individuals with certain medical conditions, including diabetes [14], lung cancer [15,16], oral and pulmonary cancer [17], liver cirrhosis [18], and that of healthy subjects. Food applications, on the other hand, have focused on using an E-nose to identify microbial contamination [19], chemical detergents and adulterations [20,21], oxidative rancidity [22], and phytopathologies [23] in food products to assess their freshness and quality.

To our knowledge, ours is the first study to evaluate the potential of an E-nose to screen fat–saliva matrices under simulated food consumption conditions. Its application in saliva/food interactions offers several advantages, including the elimination of the need for chemical analysis or sample pre-treatment. Additionally, it is a non-invasive and discreet method, requiring only saliva samples without direct intervention with the subjects themselves.

Since the sensory perception of lipid-rich foods can elicit different neurological responses related to satiety and satisfaction, an E-nose could be a valuable tool in the food industry in the development of functional foods tailored to specific consumer populations while ensuring comparable sensory perception [1,24].

Therefore, this study aims to evaluate the ability of an E-nose to detect differences in odour profiles from the interaction of extra-virgin olive oil (EVOO) as a fatty food matrix with human saliva from individuals with varying body mass indices (BMIs).

2. Materials and Methods

2.1. Extra-Virgin Olive Oil Samples and Analyses

A sample of extra-virgin olive oil (EVOO) was obtained from a blend of 50% Ascolana Tenera variety, “Colli del Tifata”, from the “La Pianurella” farm (Sant’Angelo in Formis, Caserta, Italy) and the remaining 50% from “Olive 100% Italiane”, produced by Oleificio Frantoio F.A.M. S.A.S. (Venticano, Avellino, Italy).

Olive oil acidity (% oleic acid per 100 g oil), peroxide value (meq O₂ kg⁻¹ oil), UV determination (K₂₃₂, K₂₇₀, and ΔK), and sensory attributes were carried out according to the EC Reg. 2568/1991 and its later amendments and International Olive Council (IOC) standard methods [25]. Spectrophotometric analyses were carried out using a Shimadzu UV-1601 spectrophotometer (Shimadzu, Kyoto, Japan). ΔK was calculated from absorbances at 262, 268, and 274 nm using the following formula:

$$\Delta \mathrm{K}={\mathrm{K}}_{268}-{\displaystyle \frac{{\mathrm{K}}_{262}+{\mathrm{K}}_{274}}{2}}.$$

Analysis of fatty acids was carried out by preparing fatty acid methyl esters (FAMEs) following the method described by Sacchi, et al. [26]. EVOO was diluted in hexane to a 1% oil solution. Then, 0.4 mL of the diluted oil solution was mixed with 0.2 mL of a methanolic 2 N KOH solution. The mixture was vigorously shaken for 1 min to promote the formation of FAMEs. After shaking, the organic hexane phase was collected for gas chromatography (GC) analysis. The GC analysis was performed using a Shimadzu GC-17A gas chromatograph equipped with a flame ionization detector (FID) (Shimadzu Italia, Milan, Italy). Data acquisition and processing were managed using the Class-VP Chromatography data system, version 4.6 (Shimadzu Italia, Milano, Italy). Separation of FAMEs was achieved using a 60 m capillary column with an internal diameter of 0.25 mm and a film thickness of 0.25 μm coated with 50% cyanopropylphenyl-methylpolysiloxane (Quadrex Corporation, New Haven, CT, USA). The oven temperature was initially set to 170 °C and maintained for 20 min, then raised at a rate of 10 °C/min to 220 °C, where it was held for an additional 5 min. The injector and FID temperatures were both set to 250 °C. Helium was used as the carrier gas at a flow rate of 2 mL/min. The split ratio was 1:60 and the injected sample volume was 1 mL. Identification of fatty acids was performed by comparing their retention time with that of a reference standard mixture of the pure methyl esters of fatty acids (Larodan, Malmoe, Sweden), which were injected under identical conditions.

The chemical and sensory attributes of the EVOO used in the study are summarized in

Table 1. The chemical indices, sensory attributes, and fatty acids for the extra-virgin olive oil (EVOO) used in the study.

Chemical Index	Mean ± SD	EVOO Legal Limit *
Acidity (% oleic acid)	0.38 ± 0.02	≤0.8
Peroxide value (meq O2 kg−1 oil)	8.23 ± 0.08	<20
K232	1.96 ± 0.01	≤2.50
K270	0.16 ± 0.00	≤0.22
ΔK	−0.001 ± 0.002	≤0.01
Sensory Attribute	Median	EVOO Legal Limit *
Fruitiness	4.2	>0
Off-flavours	0	=0
Fatty Acid	Quantity (%)	Common Name
C 16:0	12.29 ± 0.12	Palmitic acid
C 18:0	2.03 ± 0.03	Stearic acid
C 18:1	75.65 ± 0.00	Oleic acid
C 18:2	7.10 ± 0.00	Linoleic acid
C 18:3	0.74 ± 0.00	α-Linolenic acid
SFA	14.94
MUFA	77.20
PUFA	7.86

* EEC Reg. 2568/91 and further modifications. Sensory attributes are expressed as a median on an unstructured 0–10 scale. Fatty acid composition is reported in % weight (g/100 g) of total methyl esters. SFA, saturated fatty acid; MUFA, monounsaturated fatty acid; PUFA, polyunsaturated fatty acid.

. Free acidity, peroxide value, K₂₃₂, K₂₇₀, ΔK, and sensory attributes of EVOO were within the legal limit of this market category (EC 2568/91). The main fatty acid was oleic acid, constituting 75.7% of the total fatty acid composition (Table 1).

2.2. Human Saliva Sampling

Saliva was collected from 23 individuals of healthy weight (HW), 18 overweight (OW), and 14 obese (O) individuals (45.45% female and 54.55% male). All participants were volunteers aged between 18 and 64 years (

Table 2. The descriptive characteristics of the study’s participants, grouped by Body Mass Index (BMI), into healthy weight (HW), overweight (OW), and obese (O).

	Healthy Weight (HW)	Overweight (OW)	Obese (O)
Number (n)	23	18	14
Sex (M/F)	8/15	12/6	8/6
Age, mean (range) (years)	28.4 (23–50)	34.6 (24–63)	34.7 (18–64)
BMI, mean, (range, SD) (kg/m2)	22.1 a (18.6–24.8; 1.7)	27.6 b (25.92–29.4; 1.2)	35.5 c (30.14–38.09; 4.7)
HW (18.5 ≤ BMI ≤ 24.9)	23
OW (25 ≤ BMI ≤ 29.9)		18
O class I (30 ≤ BMI ≤ 34.9)			5
O class II (35 ≤ BMI ≤ 39.9)			7
O class III (BMI ≥ 40)			2
Smoking	2	5	3

Values are expressed as a mean and/or range. BMI is reported as a mean (range and standard deviation, SD). Different letters (p < 0.05) indicate significant differences in BMI values.

). Donors were recruited amongst students, researchers, professors, and staff from the University of Naples Federico II, all of whom had average taste and olfactory function and maintained proper oral hygiene.

All procedures were conducted in accordance with the Helsinki Declaration of 1964 and its later amendments, and equivalent ethical standards. Protocols to protect the rights and privacy of participants were strictly followed to ensure no data were released without their consent. Data were collected anonymously. Participation was voluntary and without coercion. Prior to the experiments, assessors signed a written informed consent form that fully explained the voluntary nature of participation.

Saliva collection occurred approximately two hours after breakfast to minimize potential biases from diet and food intake. Each volunteer performed a thorough cleaning of their teeth and oral cavity using a non-aromatic antimicrobial toothpaste and mouthwash to eliminate any confounding factors that could influence saliva composition [7]. Prior to saliva sampling, each volunteer rinsed their mouth with water for 30 s to further reduce contamination risk. Approximately 5 mL of saliva was collected from each individual using a standardized passive drool technique. After collection, the samples were immediately vortexed to ensure homogenization, split into several aliquots, and stored at –20 °C until subsequent analyses could be performed. Before use, the saliva samples were placed in a thermal bath at 37 °C and shaken to dissolve any suspension.

2.3. Electronic Nose Analysis

The portable electronic nose, PEN2 (WMA Airsense Analytics GmbH, Schwerin, Germany), equipped with ten metal oxide semiconductor (MOS) thin-film sensors, was used. The method used was adapted from Buratti, et al. [27]. For the analysis, 1.5 g of oil and 0.3 mL of saliva were placed in 20 mL Pirex^® vials sealed with a pierceable Teflon/silicon septum cap. Water was also used in place of saliva as a reference sample for comparison. The samples were incubated in a temperature-controlled bath at 37 °C for 10 min prior to inserting the E-nose sampling needle. The headspace odour-sampling system included an air injection needle and a sample aspiration needle that extracted and transferred the volatile fraction to the E-nose’s MOS sensors at a constant velocity of 400 mL/min, with the headspace being pumped over the sensors’ surfaces for 100 s while recording the sensors’ signals. Sensor signals were recorded every second by a computer connected to the E-nose system (Winmuster v.1.6 software) and the recovery time for flushing the sensors with reference air was 240 s. The E-nose was used at 25 °C ± 1 in all the experiments. The mean G/G₀ values for each sensor response, used in data pattern construction, were calculated based on measurements taken during the last part of the experiment (96–100 s), when the MOS sensors were stable [28], using Winmuster v.1.6 software (Airsense Analytics GmbH, Schwerin, Germany).

2.4. Statistical Analyses

The E-nose responses gathered from the 10 MOS sensors for the 63 samples in the 96–100 s time interval were selected because this phase ensured signal stabilization. The responses were used to construct a data pattern with Winmuster v.1.6 software (Airsense Analytics GmbH, Schwerin, Germany). Data analysis was performed by two pattern recognition methods: supervised linear discriminant analysis (LDA) and principal component analysis (PCA). PCA was applied to reduce the dimensionality of the data and identify trends, while supervised LDA was used to classify the samples based on BMI group. Analysis of variance (ANOVA), with Tukey’s HSD test and a significance level set at p < 0.05, was used to evaluate statistical differences in sensor responses between the three BMI groups. Analyses were conducted using Winmuster v.1.6 software (Airsense Analytics GmbH, Schwerin, Germany) and XLStat (Version 2019 v.2.2), an add-on for Microsoft Excel (Addinsoft Corp., Paris, France).

3. Results and Discussion

The E-nose generates responses by a conductance ratio (G/G0), where G indicates the sensor’s conductivity in response to a sample gas and G0 denotes the conductivity for a baseline gas. The signal responses from the ten MOS sensors were employed to construct a data pattern for the samples under investigation.

Table 3. The effect of body mass index on the mean G/G0 values in the 96–100 s data range of MOS sensors from electronic nose analysis, including 23 samples from healthy weight, 18 from overweight, and 14 from obese subjects (see Table 1 for details).

	(G/G0)
MOS Sensor	HW	OW	O	REF
W1C	0.56 ± 0.03 b	0.55 ± 0.02 b	0.58 ± 0.03 a	0.55 ± 0.03 b
W5S	17.28 ± 1.34 a	17.74 ± 1.23 a	15.23 ± 1.03 b	16.83 ± 1.21 a
W3C	0.67 ± 0.02 ab	0.65 ± 0.02 b	0.69 ± 0.02 a	0.67 ± 0.02 ab
W6S	1.05 ± 0.02 b	1.09 ± 0.02 a	1.04 ± 0.05 bc	1.01 ± 0.01 c
W5C	0.45 ± 0.03 ab	0.44 ± 0.02 b	0.46 ± 0.01 a	0.46 ± 0.01 a
W1S	2.77 ± 0.27 ab	2.81 ± 0.20 a	2.55 ± 0.14 c	2.55 ± 0.08 bc
W1W	1.45 ± 0.05 a	1.42 ± 0.03 a	1.40 ± 0.09 a	1.42 ± 0.06 a
W2S	2.16 ± 0.22 ab	2.27 ± 0.24 a	2.03 ± 0.27 bc	1.81 ± 0.08 c
W2W	4.36 ± 0.24 a	4.33 ± 0.20 a	3.92 ± 0.19 b	4.21 ± 0.24 a
W3S	1.10 ± 0.04 a	1.12 ± 0.04 a	1.08 ± 0.10 ab	1.03 ± 0.03 b

Healthy weight (HW), obese (O), overweight (OW), and reference (REF). REF refers to the reference sample in which water replaced saliva. Data were reported as a mean ± standard deviation. Different letters indicate statistically significant differences (p < 0.05).

shows the average signal intensity response of the sensors for the three subject groups and the reference samples (using water instead of saliva). The sensors that showed the greatest difference in response intensity between the samples were W5S, W1S, W1W, W2S, and W2W, with W5S giving the highest signal intensity. In particular, higher differences were detected between the obese group and the healthy weight and overweight groups, where the MOS sensors showed a lower response intensity in the O group (Table 3).

W1S, W2S, and W5S are broad-range sensors with a higher sensitivity to aliphatic compounds (contain hydrogen atoms and CH3 groups), which are volatile compounds released during EVOO consumption. W2W is sensitive to sulfuric compounds present in the oral cavity and produced by the oral microbiota, and are subsequently found in the breath [29,30].

Therefore, in order to investigate the E-nose’s ability to distinguish between EVOO–saliva samples from the three BMI groups, HW, OW and O, we conducted a data pattern analysis using two pattern recognition techniques: linear discriminant analysis (LDA) and principal component analysis (PCA).

The data from the 63 samples were analysed using linear discriminant analysis (LDA) to classify them based on sensor response. The sensor responses for each sample served as input features in the LDA model, which was trained to distinguish between the BMI groups. To address class imbalances, the analysis assumed equal within-class covariance matrices and incorporated prior probabilities. As shown in Figure 1, the LDA plot reveals a clear separation of samples across the three BMI groups. The first component (F1) accounts for 61.25% of the variance while the second component (F2) accounts for 23.97%. Overall, the three BMI groups were positioned in three distinct areas of the plot. The distribution of the samples indicates the EVOO–water samples cluster close to the healthy weight group. In contrast, the OW samples exhibit a clearer separation from the HW samples, with a tendency to be more similar to the O group.

To evaluate the model’s performance, we applied leave-one-out cross-validation, where each sample served as both training and testing data. Classification accuracy was determined as the proportion of correctly classified samples. A confusion matrix, assuming equal within-class covariance matrices and prior probabilities, was used to assess performance, with the significance level set to 5%.

As shown in

Table 4. The confusion matrix obtained from the discriminant analysis of the E-nose data pattern, showing the classification results for the training samples and cross-validation results and the number of correct and incorrect classifications across the different BMI categories.

	HW	O	OW	Total	Correct Response (%)
Training sample
HW	22	1	0	23	95.7%
O	1	13	0	14	92.9%
OW	0	2	16	18	88.9%
Total	23	16	16	55	92.7%
Cross-validation results
HW	20	3	0	23	87.0%
O	2	12	0	14	85.7%
OW	0	2	16	18	88.9%
Total	22	17	16	55	87.3%

Healthy weight (HW), obese (O), and overweight (OW). The percentages represent the number of correct classifications in each group.

, the classification results indicate a clear separation of the samples across the three BMI groups, with correct classification rates of 92.7% for the training samples and 87.3% for the cross-validation results. Regarding the cross-validation results, the obese (O) group was misclassified as healthy weight (HW) on two occasions. The overweight (OW) group was misclassified as O twice. Meanwhile, the HW group was misclassified three times as O (Table 4).

PCA was employed to analyse the E-nose data pattern, offering a partial visualization of the dataset in a reduced dimension space. Unlike the supervised approach of LDA, PCA is a non-parametric, linear, and unsupervised method that reduces dataset dimensionality, facilitates the examination of primary sources of variability, and identifies differences and similarities amongst samples [31]. An observation plot showing the two principal components is presented in Figure 2, where these two axes account for 72.1% of the cumulative variance. This analysis demonstrates a clear separation of the samples into three distinct groups, supporting the findings from the LDA. In particular, the O samples were more distinctly separated from the other groups.

Our results indicate that the E-nose effectively discriminates between oil samples in contact with water and those in contact with saliva (Figure 1 and Figure 2). This outcome is primarily attributed to compositional differences between saliva and water. Although saliva is a hypotonic fluid containing approximately 98% water [32], the remaining 2% of solutes are responsible for the difference in the volatile compounds released from EVOO. Genovese, Rispoli and Sacchi [12] reported that the interaction between EVOO and water resulted in a lower release of volatile compounds compared to the interaction between saliva and EVOO. In particular, when artificial saliva is added to water/oil emulsions, hydrophilic compounds are preferentially diluted in saliva, which is primarily water-based, thereby reducing their release to the air phase. Conversely, hydrophobic compounds remain largely solubilized in the oil phase, minimizing their partition to the air phase [33]. This behavior was also observed in studies on ice cream, where the release of hydrophobic volatile compounds was found to be higher in low-fat ice cream compared to ice cream with a higher fat content [34,35]. These findings suggest that fat serves as a reservoir for hydrophobic compounds, retaining them within the matrix, and thereby reducing their availability for release to the air phase.

In addition, the salivary composition of subjects with different body mass indices (BMIs) varies significantly [36]. For instance, proteins, such as albumins, lactotransferrin, and transglutaminase E, are found to be more prevalent in individuals who are obese [10,37]. It is also known that the lipase and α-amylase activity in saliva is higher in subjects with a larger body mass index [11]. Along with mucin and other proteins, α-amylase plays a crucial role in flocculation and coalescence phenomema, which lead to emulsion destabilization and changes in viscosity [38]. These changes affect the aroma perception of fatty matrices by altering the release of volatile compounds (VOCs) [39]. Therefore, variance in the composition of saliva based on body weight may influence the release of VOCs through both chemical factors, such as enzymatic activity, and physical factors, such as changes in matrix viscosity.

Lipases can hydrolyse triglycerides, leading to the formation of free fatty acids [40]. These enzymes can release free fatty acids from triglycerides within 1-5 s [41], potentially giving rise to certain volatile compounds in the mouth through the action of other enzymes involved in the oxidation of linoleic and linolenic acids, as suggested by Feron and Poette [1] and studied by Genovese, Rispoli and Sacchi [12]. These authors also reported that individuals who are overweight (OW) or obese (O) release a higher amount of C6 volatile compounds compared to individuals of a healthy weight (HW), who release more C5 compounds [12]. Therefore, the release of specific volatile compounds can vary with salivary composition and between individuals of different BMIs.

In conclusion, the MOS sensors in the E-nose demonstrate good sensitivity to detect changes in volatile compound release in interactions between EVOO and saliva from donors of different BMIs. The E-nose can assist the food industry in developing tailored food products based on individual sensory and nutritional responses, offering a valuable non-human-based alternative to sensory analysis.

While this study provides valuable insight as a preliminary investigation, we acknowledge that the sample size is limited. Therefore, further research is needed to validate these findings in a larger and more diverse population. Moreover, investigating the composition of salivary volatile compounds, as influenced by body weight, using both E-nose and GC/MS techniques, would offer valuable insight into the contribution of BMI to the salivary volatilome, a topic that remains largely unexplored in the current literature.