Effect of Storage Time and Packing Method on the Freshness of Dried Lycium Fruit Using Electronic Nose and Chemometrics

China-Malaysia National Joint Lab, Biomedical Research Center, Northwest Minzu University, Lanzhou 730030, Gansu, China College of Life Science and Engineering, Northwest Minzu University, Lanzhou 730030, Gansu, China Department of Biosystems Engineering, Zhejiang University, 886 Yuhangtang Road, Hangzhou 310058, Zhejiang, China Gannan Research Institute of Yak Milk, 32 Huanxi Road, Hezuo 747000, Gansu, China


Introduction
Lycium barbarum L., a Solanaceae defoliated shrubbery that grows in China, Tibet, and other parts of Asia, reportedly have many health-enhancing effects, such as immune modulation, antioxidant effects, and eye health benefits [1,2]. e dried fruits have been used as traditional medicine in China and in many countries for more than 2500 years [3].
However, the content of the bioactive substances in dried Lycium fruits declines with storage, such as the content of Lycium barbarum polysaccharides [4], chlorogenic acid, [5,6], and total flavonoids [7], indicating that stale Lycium fruits weaken in terms of pharmacodynamic effects. It is very difficult for customs to recognize the freshness of dried Lycium fruits. Guaranteeing the quality of commercialized products of L. barbarum is important.
For the quality detection of dried Lycium fruits, sensory evaluation [8], physical and chemical component analysis and fingerprint technology, and spectroscopy technology are used. For sensory evaluation, it is the traditional method used for quality differentiation. However, it relies on the specialists and unable to give the content of internal components, which limits the application. For methods based on physical and chemical component analysis and fingerprint technology [9], they were effective in regional distribution [10][11][12], variety identification, and differentiation [13,14]. e complex sample pretreatment, long detection cycle, and effects by external conditions could not be ignored. For spectroscopy technology, it is widely used for tracing the origin of L. barbarum [15,16] varieties, and species differentiation of eight Lycium species with two-dimensional infrared spectroscopic fingerprinting have been studied [17]. However, samples were usually crushed to powder [16] to guarantee the sample homogeneity and spectral data stability. us, it was not a nondestructive test. e methods used to detect the quality of dried Lycium fruits could not meet the requirement of market supervision department for fast and nondestructive detection methods. It is important to develop a fast, nondestructive method to distinguish and predict the freshness of dried Lycium fruits. e difference in the composition and content of inner components of dried Lycium fruits results in varied odors of volatiles on the surface, thereby providing a basis for the nondestructive internal quality assessment of dried Lycium fruits by using the information of volatile compounds. However, the composition of volatiles is complex and difficult to detect and analyze. Works conducted mainly focused on the composition and content of volatile compounds for Goji berry of different origins [18] and processing methods [19]. Few studies were conducted on the judgment of storage years and other factors on the freshness by using flavor compounds.
Gas sensor arrays are sensitive to the volatiles of samples and give a comprehensive odor fingerprint, which could be used for qualitative and quantitative analysis of the nondestructive freshness judgment of dried Lycium fruits. Detection of the volatile odors released from the whole-dried Lycium fruits allows the development of a fast, objective, and nondestructive method for rapid evaluation of the freshness of dried Lycium fruits in the market. Electronic nose (Enose) is a device developed for sensing of odor information by using a gas sensor array that is sensitive to volatile compounds released from the whole sample. With the advantages of simple sample preparation, high sensitivity, reliability, and repeatability, E-nose has been widely used in food and agricultural products for quality control, environmental monitoring, medical diagnosis, explosive detection, and other fields.
Preliminary studies on E-nose have been conducted on agricultural products used for medicine and food. e E-nose was used for variety differentiation and identification. It was used for determination of the difference between American and Asian ginseng [20], as well as variety differentiation of three cultivars of Perilla frutescens [21], 10 different varieties of Chinese jujubes [22], and 13 species of Chinese medicine ( [23]. For origin traceability, Radix Angelica sinensis (Danggui) from 4 different producing areas were distinguished by the E-nose [24]. In addition, raw samples of Licorice roots were classified according to their origin with the help of the E-nose [25]. For the identification of processing methods for products used for medicine and food, differences between fresh and steamed Arctium lappa L. [26] and stir-fried processed Crataegus in different degrees [27] were effectively determined by the E-nose. For quantification prediction, the correlation between the Eurycoma longifolia extract content and E-nose response was studied [28], and an effective predictive model for rapidly identifying total glucosinolate content in Maca was built [29]. Moreover, coupled with back propagation-artificial neural networks, the E-nose showed high ability to predict ginseng age [30].
Recently, the E-nose was used to identify production areas of Zhongning goji berries [31]. However, the smell of dried Lycium fruits was seldom employed for freshness detection, and few studies were conducted on the correlation between E-nose sensor responses and the freshness of dried Lycium fruits.
Dried Lycium fruits that differed in freshness were detected using the E-nose to do the following: (1) to evaluate the changes in odor of Lycium fruits due to differences in freshness; (2) to analyze the correlation between odor and freshness of Lycium fruits; and (3) to provide references and establish a new method for market surveillance of dried Lycium fruits.

Dried Lycium Fruit Samples.
e fruits of cultivated L. barbarum were dried naturally immediately after picking. e dried Lycium fruits were collected from June to September in 2015, 2016, and 2017 in Zhongning, Ningxia Hui Autonomous Region, China. As listed in Table 1, sample 1 was produced in the year 2015, sample 2 was produced in the year 2016, and samples 3-6 were produced in the year 2017 with different grades. e samples were stored for 3, 2, and 1 years, respectively. Each kind of dried Lycium fruits were sealed in two zipper bags, and then stored at 4°C in the refrigerator to prevent moisture absorption and softening of the dried fruits.
All the dried Lycium fruit samples were brought to room temperature before E-nose detection. In addition, 31 duplications were prepared for each kind of samples. 1∼2 responses were eliminated as the E-nose needs time to adapt to a new sample.

Electronic Nose (E-Nose).
To obtain the odor of volatiles emitted from dried Lycium fruits, an E-nose of PEN2 (Airsense Corporation, Germany) was used. e E-nose system consists of three parts, namely, a sensor array, a detector unit, and a pattern recognition system. e nomenclature and characteristics of the 10 metal oxide sensors are listed in Table 2, showing that each sensor has a certain degree of affinity toward specific chemical or volatile compounds. e software Win Muster v.1.6 was used for data recording.
In our former study, the experimental conditions for the detection of dried Lycium fruits by the E-nose were optimized [32] and given as follows. Before detected by the E-nose, the dried Lycium fruit samples were brought to room temperature. e headspace used for the detection of E-nose was generated by the following processes. Without any preparation, 20.0 g of the dried Lycium fruits was placed in a beaker (500 mL). e beaker was sealed by a plastic and kept for 30 min at room temperature of 25°C ± 3°C for the headspace to stabilize. During sampling, the stabilized headspace was transferred into the sensor chamber at a flow rate of 300 mL·min − 1 . e detection procedure was as follows. For each sample, the measurement time was set to 70 s at an interval of 1 s, and the sensors were rinsed for 80 s with clean air before the detection of the next sample. All the samples were detected at room temperature with 31 duplications.

Granularity and Dry Weight of 100 Seeds.
e granularity and dry weight of 100 seeds, indicating the seed size, maturity, and the quality, were detected according to the national standards of [33], and the results are listed in Table 1.

Statistical Analysis.
e ability of the E-nose in the evaluation of dried Lycium fruits freshness was judged by the chemometrics. Principle component analysis (PCA), canonical discriminate analysis (CDA), and cluster analysis (CA) were employed to visualize the difference of dried Lycium fruits with different freshness. To study the correlation between the storage years of the dried Lycium fruits and E-nose sensor responses, multiple linear regression (MLR) with the step regression method, partial least square (PLS) regression with cross-validation, and back propagation neural network (BPNN) were performed. And the predictive capacity of the E-nose for the storage years of dried Lycium fruits was compared. Both the qualitative and quantitive methods were compared to find the better one.
For data analysis, the SAS version 8 (SAS Institute Inc., Gary, USA) was used. e figures were plotted with Origin Pro 8.

Sensor Array Responses to Dried Lycium Fruits.
e typical responses of E-nose sensors to dried Lycium fruits produced in the years 2015, 2016, and 2017 are shown in Figure 1. For one sample, each curve represents a sensor transient. For the E-nose, the output signal of each sensor is given as G/G0, where G0 represents the electronic conductivity (resistance) of sensor to the zero gas, which is ambient air after filtered by active carbon, and G represents the electronic conductivity (resistance) of the sensor to the sample gas.
After an initial period of low and stable conductivity, the conductivities for sensor S2 increased sharply within the first 30 s and then stabilized after 50 s of collection time. e conductivities of sensor S6, S7, S8, and S9 increased slowly during detection. For sensors S1, S3, and S5, the conductivities dropped slightly and stabilized after 40 s collection time. e responses of S4 and S10 to dried Lycium fruits were  weak. e sensor signal stabilized generally and was considered for further analysis.
As shown in Figure 1, the response tendency of the sensors was quite similar for dried Lycium fruits produced in different years, whereas the response varied greatly in intensity for S2 and S9. e stabilized sensor responses at the 70 th second were extracted and analyzed by ANOVA, and the results are shown in Table 3. As shown in Table 3, except for S4, the responses of sensors S1-S3 and S5-S10 were significantly affected by the producing year of dried Lycium fruits, thereby laying the foundation for the differentiation of dried Lycium fruits produced in different years.

Discrimination Storage Years and Packing Method of Dried Lycium Fruits with PCA and CDA.
For the volatile components of dried Lycium fruit, it is reported that the nitrogenous substances, such as 6-methyl-6-nitroheptan-2one, made the responses of S2 changed greatly. For the high content of aldehydes, such as hexanal and 2,6,6-trimethyl-1, 3-cyclohexadiene-1-carboxaldehyde, alcohols of 1-octen-3ol, and ketones of 2,3-butanedione, makes the responses of S8 and S6 differed with each other. For the major aromatic compounds of alkanes, furan, pyrazines, and aromatic ether, the sensors S1, S3, S5, S9, and S10 respond to these flavors. For terpenes, 10 compounds were reported, which made the response of S7 changes. Furthermore, hexanal and 1octen-3-ol, the oxidation products of fatty acids, and 2, 6, 6trimethy-1, 3-cyclohexadiene-1-carboxaldehyde, the oxidation products of carotene components, were the prime candidates of undesirable odor of dried Lycium fruits and its products. During the storage of dried Lycium fruits, the smell changed due to the oxidation, hydrolysis of carotene components, ω5 group of the C18 fatty acids, and other components [9]. All these changes of volatile compounds were detected by the E-nose. e extracted sensor responses of the E-nose to dried Lycium fruits at the 70 th second were analyzed by PCA and CDA, and the results are shown in Figure 2. In Figure 2 With PCA, the stale fruits and different grades of Lycium fruits can be discriminated; the same results were obtained by the taste information obtained by the taste sensing system with artificial lipid-based membrane sensors in our former studies [34]. e ability of the E-nose to discriminate staledried Lycium fruits from fresh ones provided a new nondestructive method for the detection of adulteration of dried Lycium fruits with stale ones.
e CDA results were shown in Figure 2(b). Can1, Can2, and Can3 explained 96.02% of the total variance with the value of 100%. us, the first three CANs can give most information of the E-nose data set. e samples were grouped into three clusters according to the producing year. is observation could be used in the discrimination of adulteration of dried Lycium fruits with stale ones.

Classification of Storage Years and Packing Method of
Dried Lycium Fruit with CA. For dried Lycium fruits, CA was conducted based on odor information. e dendrogram obtained by CA is shown in Figure 3.
In Figure 3, the results obtained by CA showed that, at the distance D � 1.3711, the samples of dried Lycium fruits were clustered into two groups, namely, cluster 1 With sensor responses, the E-nose system could recognize the fresh and stale fruits of dried Lycium fruits by PCA, CDA, and CA methods.

Rapid Characterization of Storage Years of Dried Lycium
Fruits. To establish the relationship between the responses of E-nose and storage years of dried Lycium fruits, analytical methods of MLR, PLS, and BPNN were used, and the results were compared to find the best predictive model. Data set containing dried Lycium fruit samples (156 for calibration and 22 for validation) was used to build the predictive model for the storage years of dried Lycium fruits. e correlation coefficient (R 2 ) and root mean square error (RMSE) between predicted and experimental values (RMSE) were used to evaluate the model performance. Large R 2 and low RMSE lead to good calibration model. e results are shown in Figure 4 and Table 4.
With the stepwise method, the MLR algorithm established a model that describes the relationship between the E-nose sensor signals and the storage years of dried Lycium fruits. e predictive models for storage years of dried Lycium fruits were given as follows.
Large R 2 and low RMSE indicate adequate fits. As shown in Table 4, the predictive ability of the E-nose was listed. A linear correlation was observed between the responses of E-nose sensors and the storage years of dried Lycium fruits. When the model was used to predict the storage years of dried Lycium fruits in the test data set, R 2 was 0.9239, thereby showing good ability to predict the storage years of dried Lycium fruits.
For models built by PLS, the coefficient R 2 � 0.9316 was found between the E-nose sensor responses and storage years of dried Lycium fruit samples. When the model was used to predict samples in the test set, R 2 of 0.9480 was obtained. For the prediction of storage years of dried Lycium fruits, the PLS methods are slightly better than that of MLR.
For BPNN, the experimental design was completely randomized with each sample as an experimental unit. e architecture of the chosen artificial neural network was N × (2N + 1) × M three-layer back propagation. e input layer, i.e., one hidden layer, was designed as 10 neurons according to the sensor array of 10 sensors. e output layer      had three neurons for different storage years of Lycium fruits. e results obtained by the training model of BPNN showed that all the training and testing samples were correctly classified according to the storage years of Lycium fruits. Moreover, the correlations between the observed and predicted storage years of dried Lycium fruits were higher than 0.98 (Table 4) for the training and testing sets. e low errors of prediction and the high correlation of the BPNN model suggested that the E-nose can be successfully applied to the determination of the storage years of dried Lycium fruits.
In conclusion, the best results were obtained by the BPNN model, with high correlation (higher than 0.98) for the training and testing subsets.
For the E-nose, it has shown it ability in the discrimination of dried Lycium fruits for the freshness and the packing methods and also in the prediction of storage years. Former studies were conducted on the Lycium ruthenicum Murr., for producing area, species, and drying conditions [38]. In the future, more applications of the E-nose on quality detection should be studied to give more objective results.

Conclusions
Multivariate techniques (PCA, CDA, and CA) were used to prove the E-nose's usefulness for discriminating dried Lycium fruits according to freshness. Among the three methods, CDA was suitable for the freshness differentiation of dried Lycium fruits according to the producing year. Strong correlation was observed between the E-nose response and freshness of dried Lycium fruits, and BPNN was the most suitable for the prediction of storage years. e results obtained by this study are promising in terms of further development of rapid and nondestructive detection methods based on E-nose systems for discrimination between dried Lycium fruits and stale ones.

Data Availability
e research data about this manuscript used to support the findings of this study are included within the article.

Disclosure
is study has used the E-nose for nondestructive detection of dried Lycium fruit freshness, which provides a method to evaluate the quality of Lycium fruit.

Conflicts of Interest
e authors declare that they have no conflicts of interest.