Validation of honeybee smelling profile by using a commercial electronic nose

Honey is a natural sweetener and its quality labels are associated to its botanical or geographical origin, which is being established by palynological and sensorial analysis. The use of fast and non-invasive techniques such as an electronic nose can become an alternative for honey classification. In this study, the operational parameters of a commercial electronic nose were validated to determine the honey odor profile. A central composite design with five factors, three levels and 28 assays was used, varying sample amounts (1, 2 and 3 g), incubation temperature (30, 40 and 50 °C), incubation time 30 min), gas flow (50, 150 and 250 mL/min) and injection time (100, 200 and 300 s). The commercial nose had ten sensors. Repeatability was evaluated with a coefficient of variation of 10 %. The response surface methodology was used and the optimal operating conditions were: 3 g of sample, incubation at 50 °C for 17 min, gas flow of 100 mL/min and sampling time of 150 s. Finally, these parameters were used to analyze 19 samples of honey, which were classified according to their odor profiles, showing that it can be a useful tool to classify honey.


Introduction
Food volatile compounds analysis is very important and, normally, it is related to smell, which is one of the most important sensory parameters.Generally, volatile compounds analysis is performed by using gas-chromatographic methods, which are robust and powerful (Agila & Barringer, 2012;Castro-Vázquez, Díaz-Maroto, González-Viñas, & Pérez-Coello, 2009; 1 Food Engineer.Universidad de la Amazonia, Colombia.Master Food Science and Technology (C), Universidad Nacional de Colombia, Colombia.Affiliation: Young researcher, Group Quality assurance food and development of new products, Food Science and Technology Institute -ICTA, Universidad Nacional de Colombia.E-mail: arcorream@unal.edu.co. 2 Chemical Engineer.Universidad Nacional de Colombia, Colombia.Master Chemical Engineering and Ph.D. Engineering -Chemical Engineering, Universidad Nacional de Colombia, Colombia.Affiliation: Associated Professor at Escuela de Ingeniería Química, Pontificia Universidad Católica de Valparaíso.E-mail: mmcuencaq@unal.edu.co.Papotti, Bertelli, & Plessi, 2012), but it is necessary to preprocess sample which is time-consuming and it is difficult to determine it.
Electronic Noses (e-Noses) are an alternative, and generally they have an electrochemical sensors array that provides a fingerprint of a given sample headspace (Romano et al., 2016).Typically, an e-Nose, trained using samples of known origin, can be employed to recognize and predict sample identity on the basis of a specific fingerprint (Gliszczyńska-Świgło & Chmielewski, 2016).The e-Nose provides little information as to the actual composition of the sample headspace but they are generally easy to use, they provide a high analytical throughput and they are relatively inexpensive.
In order to determining a smelling profile with an electronic nose, its operating conditions have to be considered: sample temperature (depending on amount and volatility of compounds presents), sampling time, gas flow, incubation time, and cleaning time of the sensors (Quicazán et al., 2014).This study aimed to validate the operating conditions for a commercial electronic nose PEN 3 (Airsense, Germany) to obtain smelling profiles for honey-bee samples, demonstrating that it can be a portable and low-cost technique even if it does not provide quantitative information about sample headspace composition.

Honey samples
For validation, it was used an Acacia honey (Robinia pseudoancia) from the local market of Bolzano (Italy).
For performing the final test with different types of honey, there were used 19 different honeys from different places, presented in Table 3.

Determination of volatile compounds profile for electronic nose
It was used a portable commercial electronic nose Airsense PEN 3 (Airsense, Germany), with an array of 10 semiconductor sensors (Table 1).Honey samples were served and weighed into glass vials of 10 mL.The vials were hermetically sealed with lids containing septa silicone.Operating parameters were changed manually for each test.The obtained responses were recorded by the sensors through Win Muster software (Airsense, Germany), and quantitatively expressed as a conductance value.
It was obtained a data matrix of "m" columns "n" rows, where "m" columns represent the number of sensors of the electronic nose and "n" the number of times the analysis was performed.From the matrix for each sensor, it was obtained the medium coefficient differential value nuance response curve of each sensor corresponding to the value of the differential coefficient (mcdv) calculated by using Equation (1) (Yin & Tian, 2007).
Where mdcv is the result of the characteristic value for each sensor profile of each sample, N is the number of time intervals analyzed, Xi and Xi+1 result of conductance in times i and i+1, respectively; Δt is the time interval between conductance data, which by default is 1s.The values obtained reflect the average speed of sensors responses and represent their principal characteristics (Quicazán et al., 2014).

Operating parameters evaluation
A central composite design with five factors with three levels (Table 3) and 28 trials were used.Responses were conductance values for each sensor (10) of the E-nose.Response surface methodology was used and optimal operating conditions were found by a responses optimization design.Source: Authors

Repeatability evaluation
Smelling profile of 10 samples of acacia honey (Robinia pseudoancia L.) from the same batch were determined by using optimal operating conditions.It was used 10% maximal variation coefficient (VC) criteria to evaluate repeatability, which measures a dispersion that correlates the average (X) and the standard deviation (s) according to Equation (2): VC S = × χ 100% (2)

Honey classification with optimized parameters
Smelling profile of 19 different honey samples were performed by using optimized e-nose parameters.10 replicates were performed.With average mcdv values, a Principal Components Analysis was performed.

Smelling profiles
The mcdv for all sensors in each of 28 trials (Table 4) were calculated from the data matrix obtained by using Equation (1).All sensors recorded conductance values different for each of the tests performed, demonstrating all conditions reflect different responses.

Operating parameters evaluation
The highest statistically significant changes in relation to each factor and interactions between factors were evaluated.With a responses optimization design, it was found that the best operating conditions were 3 g sample, incubation temperature 50 °C, incubation time 1020 s, gas flow of 100 mL/min and 150 s sampling time, result that confirm the importance not only of the sample but also of operating conditions.

Repeatability
mcdv was determined for each of 10 measurements of smelling profile by using Equation (1).From the data matrix, statistical parameters like average, range, standard deviation and coefficient of variation were determined by using Equation ( 2), as it is shown on Table 6.Variation coefficients of the responses were within the maximal limit (10%).Operating conditions allow obtaining repeatable honey smelling profiles, demonstrating that validation of a smelling profile depends on the sample but also on sampling conditions.

Honey classification
In Figure 1 is presented the biplot corresponding to a PCA analysis which explains the 80,72% of total variance for 19 honey samples.It is noticeable all samples showed a different smelling profile, especially samples 1, 2, 7 and 12, due to its botanical and geographical origin.Samples 3, 16 and 19 are mixed floral honey and present similar smelling characteristics.Even small smelling characteristics make a difference among honey samples, which is observed by performing their e-nose analysis.

Conclusions
It is concluded that optimized operating conditions found for acacia honey smelling profile were standardized: 3 g sample, incubation temperature 50°C, 1020 s incubation time, gas flow of 100 mL/min and 150 s sampling time, giving repeatable responses for all sensors.Optimized parameters smelling evaluation for 19 different honey samples shows all e-nose sensors give information related to its smelling profile, which is different for all samples, confirming that a validated methodology allows to use this technique as a quick and easy alternative for honey differentiation and classification according to its botanical and geographical origin.

Figure 1 .
Figure 1.PCA for different kinds of honey using optimized parameters.Source: Authors

Table 1 .
Symbols and groups of compounds detected by each E-nose sensor Source: Authors

Table 2 .
Factors and levels of the central composite design

Table 3 .
Different honey samples used for classification

Table 4 .
MCDV for each sensor response at each trial

Table 5 .
p values for Surface Response Analysis

Table 6 .
Statistical parameters for repeatability evaluation Source: Authors