Electronic Nose and Head Space GC–IMS Provide Insights into the Dynamic Changes and Regularity of Volatile Compounds in Zangju (Citrus reticulata cv. Manau Gan) Peel at Different Maturation Stages

Zangju (Citrus reticulata cv. Manau Gan) is the main citrus cultivar in Derong County, China, with unique aroma and flavour characteristics, but the use of Zangju peel (CRZP) is limited due to a lack of research on its peel. In this study, electronic nose, headspace-gas chromatography–ion mobility spectrometry (HS-GC–IMS), and partial least squares-discriminant analysis (PLS-DA) methods were used to rapidly and comprehensively evaluate the volatile compounds of dried CRZP and to analyse the role of dynamic changes at different maturation stages. The results showed that seventy-eight volatile compounds, mainly aldehydes (25.27%) and monoterpenes (55.88%), were found in the samples at four maturity stages. The contents of alcohols and aldehydes that produce unripe fruit aromas are relatively high in the immature stage (October to November), while the contents of monoterpenoids, ketones and esters in ripe fruit aromas are relatively high in the full ripening stage (January to February). The PLS-DA model results showed that the samples collected at different maturity stages could be effectively discriminated. The VIP method identified 12 key volatile compounds that could be used as flavour markers for CRZP samples collected at different maturity stages. Specifically, the relative volatile organic compounds (VOCs) content of CRZP harvested in October is the highest. This study provides a basis for a comprehensive understanding of the flavour characteristics of CRZP in the ripening process, the application of CRZP as a byproduct in industrial production (food, cosmetics, flavour and fragrance), and a reference for similar research on other C. reticulata varieties.


Introduction
Citrus is grown in more than 140 countries and territories worldwide and mainly includes mandarin (Citrus reticulata), grapefruit (Citrus maxima), citron (Citrus medica), oranges (Citrus sinensis) and limes (Citrus aurantium) [1]. In particular, C. reticulata of local cultivars is very popular, such as sweet orange (C. sinensis), C. unshiu, C. reticulata 'Chazhi', C. reticulata 'Ponkan' and Zangju (C. reticulata cv. Manau Gan), among others. Zangju is known locally (in the Tibetan region) as 'Jiaxu', which means 'fruit of longevity', which may be related to the fact that the Zangju tree has been growing for hundreds of years (the common citrus tree lives between 40 and 50 years). Zanju is the local cultivar of C. reticulata cv. Manau Gan and is popular with people because of its unique aroma and flavour and easy peeling. It is mainly distributed in Derong, Muli and other counties [2] in the agricultural and pastoral areas of the middle and lower reaches of the Sanjiang River Valley (subtropical valley climate) in the southern section of the Hengduan Mountains, China. It basis for the development and utilization of the VOCs in CRZP. For other C. reticulata varieties, research on the volatile compounds of these varieties provides a reference.

Results of E-Nose
LDA can minimize intragroup variance and maximize intergroup variance; in other words, it can reduce intragroup differences and expand intergroup differences [22,23]. Therefore, LDA was used to distinguish CRZP in the five picking stages. As shown in Figure 1A, the variance contribution rates of LD1 and LD2 are 82.58% and 13.52%, respectively. Five groups of samples are far apart, especially the distance between ZJP and the other four groups, indicating that ZJP is significantly different from the other samples.
providing a theoretical basis for the development and utilization of the VOCs in CRZP. For other C. reticulata varieties, research on the volatile compounds of these varieties provides a reference.

Results of E-Nose
LDA can minimize intragroup variance and maximize intergroup variance; in other words, it can reduce intragroup differences and expand intergroup differences [22,23]. Therefore, LDA was used to distinguish CRZP in the five picking stages. As shown in Figure 1A, the variance contribution rates of LD1 and LD2 are 82.58% and 13.52%, respectively. Five groups of samples are far apart, especially the distance between ZJP and the other four groups, indicating that ZJP is significantly different from the other samples. Figure 1B shows that the W5C, W1S and W1W sensors have stronger induction of VOCs in the samples, indicating that CRZP may contain higher levels of alkane aromatics, terpenes and sulfur-containing organic compounds. However, E-nose cannot quantitatively or qualitatively identify the important aroma substances in citrus peel, such as alcohols, aldehydes and esters [9,18,24]. This indicates that the E-nose is not completely sensitive to CRZP aroma substances and can only represent part of the aroma characteristics. The results showed that there were significant differences in CRZP at different harvesting stages, which could be further identified and compared.

HS-GC-IMS Visual Topographic Plot Comparison
In this study, HS-GC-IMS was used to analyse the volatile compounds in dried CRZP at different maturation stages to help identify the volatile compounds and their changes during the ripening process. To observe the differences in different periods more directly, FlavourSpec ® instrument software was used for visual analysis of the measured results, including three-dimensional spectra and corresponding two-dimensional spectra. The combination of the two spectra can more clearly reflect the differences and changes in volatile substances [24,25].  Figure 1B shows that the W5C, W1S and W1W sensors have stronger induction of VOCs in the samples, indicating that CRZP may contain higher levels of alkane aromatics, terpenes and sulfur-containing organic compounds. However, E-nose cannot quantitatively or qualitatively identify the important aroma substances in citrus peel, such as alcohols, aldehydes and esters [9,18,24]. This indicates that the E-nose is not completely sensitive to CRZP aroma substances and can only represent part of the aroma characteristics. The results showed that there were significant differences in CRZP at different harvesting stages, which could be further identified and compared.

HS-GC-IMS Visual Topographic Plot Comparison
In this study, HS-GC-IMS was used to analyse the volatile compounds in dried CRZP at different maturation stages to help identify the volatile compounds and their changes during the ripening process. To observe the differences in different periods more directly, FlavourSpec ® (Version 2.000) instrument software was used for visual analysis of the measured results, including three-dimensional spectra and corresponding two-dimensional spectra. The combination of the two spectra can more clearly reflect the differences and changes in volatile substances [24,25].
The x-, yand z-axes of the three-dimensional spectrum represent the migration time, retention time and peak intensity, respectively. The differences in VOCs in different samples can be visually seen in the three-dimensional diagram (Figure 2A). The results show that the peak signal distribution of volatile compounds in CRZP at different stages is basically similar, but the peak signal intensity is different. This shows the variation in volatile organic compound content in different samples. Compared with the three-dimensional spectrogram, the two-dimensional spectrogram has the same information except for the lack of a Z-axis, and the two-dimensional spectrogram can directly show the difference in compounds and concentrations among different samples. The background of the figure is blue ( Figure 2C), and the red vertical line at abscissa 1.0 is the RIP peak (reactive ion peak after normalization). Each dot in the diagram represents a volatile substance or dimer, and the colour represents the concentration of the substance, with white representing a lower concentration, red representing a higher concentration, and a darker colour representing a higher concentration.
representing a lower concentration, red representing a higher concentration, and a darker colour representing a higher concentration.
To be er compare the differences among samples, combined with the results of the E-nose, the spectrograph of samples at the commodity maturity stage in December was selected as a reference to obtain two-dimensional top views of samples at other maturity stages ( Figure 2D). White represents similar or consistent with the reference value; red represents higher than the reference value; and blue represents lower than the reference value. As shown in the figure, the composition and content of volatile compounds in tangerine peel changed significantly at different ripening stages. The results showed that the content of volatile compounds in immature tangerine peel changed significantly with increasing maturity, and the number of red dots was higher than that of blue dots in the full maturity period (January to February of the next year), indicating that the content of most volatile compounds increased after full maturity. Similar reports have also been reported in studies on fruits such as camellia tangerine, green plum and Passiflora edulis [24,26,27], which are related to the vigorous metabolism of ripe fruits.  To better compare the differences among samples, combined with the results of the E-nose, the spectrograph of samples at the commodity maturity stage in December was selected as a reference to obtain two-dimensional top views of samples at other maturity stages ( Figure 2D). White represents similar or consistent with the reference value; red represents higher than the reference value; and blue represents lower than the reference value. As shown in the figure, the composition and content of volatile compounds in tangerine peel changed significantly at different ripening stages. The results showed that the content of volatile compounds in immature tangerine peel changed significantly with increasing maturity, and the number of red dots was higher than that of blue dots in the full maturity period (January to February of the next year), indicating that the content of most volatile compounds increased after full maturity. Similar reports have also been reported in studies on fruits such as camellia tangerine, green plum and Passiflora edulis [24,26,27], which are related to the vigorous metabolism of ripe fruits.

Identification of Substances
HS-GC-IMS was used to qualitatively analyse the VOCs from CRZP at different maturation stages. Ninety signal peaks were determined by comparing the drift time and retention index with the control, and 78 volatile compounds were preliminarily identified, as shown in Table 1. It mainly includes ten alcohols, twenty-three aldehydes, nineteen monoterpenoids, nine ketones, eight esters, and nine other volatile substances, with relative contents of 2.13%, 25.27%, 55.88%, 4.11%, 3.87% and 8.74%, respectively ( Figure 2B). In particular, 18 substances, including α-terpineol, citronellal, linalool, decanal, cyclohexanone and ethyl acetate, formed dimers, which were mainly related to the concentration of volatile compounds and proton affinity. The protons of volatile substances with higher proton affinity than water and higher concentrations can be transferred to substances with high proton affinity, thus contributing to the formation of dimers [24,28].

Alcohols
A total of ten volatile alcohol compounds were identified in five different harvesting stages of CRZP, the number of which was third only to aldehydes and monoterpenoids. Examples include 1-penten-3-ol, (z)-2-penten-1-ol, and tetrahydrolinalool. Studies have shown that linalool and tetrahydrolinalool are important volatile compounds in citrus peel and have the aroma of citrus fruits, which are usually produced through the MEP pathway [24,29]. Other alcohols, such as α-terpineol and benzyl alcohol, were also considered to have the aroma characteristics of jasmine and lilac, and benzyl alcohol was a unique component of CRZP compared with C. reticulata 'Chachi', C. reticulata 'ponkan' and navel orange. In addition, α-terpineol, linalool and tetrahydrohydrolinalool have antibacterial, antioxidant and sedative effects [30]. Benzyl alcohol has anti-louse, anti-insect, analgesic and antiseptic properties. These include being used for Benzyl Alcohol Lotion 5% [31], antilice treatments, and injection preparation solvents, such as the solvent of spectinomycin hydrochloride that is used for injections [16].

Aldehydes
A total of 23 volatile compounds of aldehydes were identified in the CRZP, most of which were small molecules of C4-C9. Previous studies have shown that these smallmolecule aldehydes, which are widely found in the fruits and peels of citrus plants such as Shatian pomelos, oranges, lemons and tangerines, are highly volatile, have a low threshold, and are a major component of fresh and unripe fruit aromas. They are usually produced by the degradation of fatty acids [4,27,32,33]. Dodecanal, Decanal, etc. are C10-C12 molecules with high volatility and a citrus aroma. Among them, dodecanal and citronellal also have the aromas of pine leaf oil and cherry, respectively. These three compounds exist in the essential oils of lemon, citrus and sweet orange peel and mainly serve as spices and edible flavours [5,34,35]. Other volatile compounds, such as hexanal, heptanal, octanal, etc., have a fruity aroma, and benzaldehyde has a bitter almond taste, which is often used in the production of spices. Heptanal is also the raw material for pharmaceutical and rubber products. In addition, isovaleraldehyde, butanal and other aldehydes are intermediates that produce fragrances, spices, and other organic compounds [5,9,18]. Interestingly, the citronellal identified from CRZP in this study was not found in previous studies of other cultivated species of C. reticulata (C. reticulata 'Chachi', C. unshiu and C. reticulata 'Ponkan', etc.) [21,27,36].
The eight esters identified in this study included crotonic acid, hexyl ester, ethyl salicylate, 2-nonynoic acid methyl ester, ethyl acetate, butyl acetate, styralyl acetate, and methyl salicylate. Esters have fruit and wood aromas, which are an important component of citrus aroma and an important evaluation index of citrus aroma. It is mainly produced through the esterification of alcohols and acids, which has poor stability and is easy to decompose in the ripening process [9,41]. Crotonic acid, hexyl ester, ethyl salicylate, ethyl acetate, methyl salicylate, ethyl acetate, methyl salicylate, etc. [27,37]. Ethyl salicylate and methyl salicylate were reported for the first time in this study, which may be the reason for the distinctive aroma of Zangju.

Analysis of VOC Fingerprints
In order to more intuitively show the variation rule and relative content comparison of volatile substances in the CRZP at different maturity stages, we used the Gallery Plot plug-in to draw the fingerprint of volatile substances. Compared with the topographic map showing the trend of volatile compounds, fingerprints can intuitively and quantitatively compare the changes in volatile compounds (increase, decrease, or fluctuation) between different samples through the strength of different spots and reveal the dynamic changes in each substance [27,44]. In the fingerprint, each row represents a sample (three parallel at each stage), and each column represents a volatile component (the darker the colour, the higher the content). The VOC fingerprint of the CRZP at five harvesting stages is shown in Figure 3. The fingerprint showed the diversity of volatile compounds in CRZP at different harvesting periods. A total of 90 peaks (78 VOCs were defined) were preliminarily identified. The seventy-eight different VOCs included ten alcohols, twenty-three aldehydes, nineteen monoterpenoids, nine ketones, eight esters and eight other substances.

Analysis of Dynamic Changes and Formation Regularity of VOCs
Further analysis of the fingerprint ( Figure 3) and Table 2 showed that with increasing maturity, ketones first increased, reached the highest value in the commodity maturity stage (December), and then tended to be flat. Specifically, cyclohexanone M/D, 2-hexanone, 3-hydroxy-2-butanone M/D, and 2-butanone have a fruity aroma. During the ripening process, fa y acids accumulate in Zangju, providing sufficient substrate for the generation of ketones [46,47]. Alcohols, esters and other compounds showed a precipitous decline, then increased, and finally tended to fla en out. The lowest value was reached in the immature maturity stage (November), and the highest value was reached in the full maturity stage (January). It is manifested as α-terpineol M/D, tetrahydrohydrolinalool, ethanol, crotonic acid, hexyl ester, ethyl salicylate, 2-nonynoic acid methyl ester, ethyl acetate M/D, and styralyl acetate, with a citrus fruit aroma and wood aroma. This may be related to the influence of climatic conditions (temperature, precipitation, etc.) on enzyme activities related to the synthesis of such substances [27,47]. The aldehydes and terpenes showed a slow change trend of first decreasing and then increasing, reaching their lowest value in the full maturity stage (January). The specific pa ern was dodecanal, octanal, benzaldehyde, (E)-2-pentenal M/D, pentanal, alpha-terpinene M/D, alpha-thujene, nor-

Analysis of Dynamic Changes and Formation Regularity of VOCs
Further analysis of the fingerprint (Figure 3) and Table 2 showed that with increasing maturity, ketones first increased, reached the highest value in the commodity maturity stage (December), and then tended to be flat. Specifically, cyclohexanone M/D, 2-hexanone, 3-hydroxy-2-butanone M/D, and 2-butanone have a fruity aroma. During the ripening process, fatty acids accumulate in Zangju, providing sufficient substrate for the generation of ketones [45,46]. Alcohols, esters and other compounds showed a precipitous decline, then increased, and finally tended to flatten out. The lowest value was reached in the immature maturity stage (November), and the highest value was reached in the full maturity stage (January). It is manifested as α-terpineol M/D, tetrahydrohydrolinalool, ethanol, crotonic acid, hexyl ester, ethyl salicylate, 2-nonynoic acid methyl ester, ethyl acetate M/D, and styralyl acetate, with a citrus fruit aroma and wood aroma. This may be related to the influence of climatic conditions (temperature, precipitation, etc.) on enzyme activities related to the synthesis of such substances [27,47]. The aldehydes and terpenes showed a slow change trend of first decreasing and then increasing, reaching their lowest value in the full maturity stage (January). The specific pattern was dodecanal, octanal, benzaldehyde, (E)-2-pentenal M/D, pentanal, alpha-terpinene M/D, alpha-thujene, norbornane, 7,7-Dimethyl-2-meth, 3-Carene, α-Phellandrene, with citrus aromas of sweet orange, lemon, pine leaf oil and cherry. This may be due to the activation of the lip oxygenase (LOX) pathway with increasing maturity. Active hydroperoxide lyase (HPL) and alcohol dehydrogenase (ADH) react hydroperoxide with alcohol, thus reducing the alcohol content and producing other substances [24,26], which reaches the lowest value in the full maturity stage (January) and increases slightly in February. This may be related to the influence of the growth environment (temperature) on related enzyme activities [22,27,47]. The lower temperature inhibits enzyme activity, leading to a decrease in the biosynthesis of VOCs, resulting in a decrease in the relative content of VOCs in December and January. In general, with increasing maturity, the relative content of VOCs in tangerine peel showed a downward trend and then increased slightly in the CRZP harvested in February, indicating that the relative abundance of VOCs in the CRZP harvested in October was the highest compared with the other four periods. To better analyse the dynamic changes in VOCs in the CRZP at different picking stages, we drew a clustering heatmap based on the fingerprint results ( Figure 4). The darker the red or blue, the higher or lower the peak intensity, respectively. The samples are clustered into five different categories [24,48], and the samples with similar accumulation patterns have a higher correlation. The first category includes alcohols, aldehydes, ketones and monoterpenes, such as 1-butanol, citronellal M, 2-butanone, 3-methyl-1-butanol, 2methylbutyraldehyde M/D, cyclohexanone D, 2-methyl-1-pentanol, α-thujene, β-myrcene, γ-terpinene, and octanal. These compounds produce citrus, cherry, and other fruit aromas whose intensity increases with maturity, reaching a maximum in January and then decreasing in February. Compounds 3-Methyl-1-butanol, α-thujene, and citronellal M are unique VOCs for samples taken in January. The second category contains alcohols, aldehydes, ketones, terpenoids, esters, 1-penten-3-one, (z)-2-penten-1-ol, α-terpinene M/D, crotonic acid, hexyl ester, 2-nonynoic acid methyl ester, ethyl salicylate, and dodecanal. The intensity of these C6-C10-producing VOCs with complex unripe aromas decreased with increasing maturity, which was related to the activation of the LOX pathway, resulting in a reduction in the biosynthesis ability of unripe VOCs, indicating that the samples collected in October had more complex unripe aroma characteristics. The third group includes ketones, monoterpenes, alcohols, and aldehydes, such as (E)-ocimene, β-pinene, α-terpinene, 2-methylpropanoic acid, 2-hexanone, 3-hydroxy-2-butanone, decanal, and linalool. The intensity of these fragrance-producing substances showed a trend of first increasing and then decreasing, reaching a maximum in December. The fourth kind includes most alcohols and aldehydes as well as some esters and terpenes, such as α-terpineol    To clarify the characteristic differences among samples and the similarities and differences in VOC composition, PCA and PLS-DA models were adopted to further process the chromatographic peak data of CRZP at different maturity stages [24,27]. The VIP method was used to search for characteristic aromas. The total sample size of the model was fifteen (five periods × three). The Y variable in the model refers to the dried fruit skins collected in five harvesting periods (October 2022-February 2023). The X variable represents the 78 VOCs identified in this study. Figure 5A shows the PCA scoring maps of VOCs of the CRZP at five different harvesting stages measured by GC-IMS. It can be seen from Figure 5A that after dimension reduction, the contribution rates of the first two principal compounds were 66.52%, and the first principal component (PC1) and the second principal component (PC2) were 40.73% and 25.79%, respectively. This shows that after feature compression, relatively complete information is still retained, and the feature difference of the original variable can be better represented. Samples ZGP, ZIP, ZJP, ZKP and ZLP were obviously separated for four different maturation stages, indicating that there were obvious characteristic differences among samples.
The PLS-DA results are shown in Figure 5B. The first component and the second component accounted for 47.6% and 27.6%, respectively, in the PLS-DA scoring chart. R2 and Q2 have high accuracy, both exceeding 0.91 and 0.6 ( Figure 5D), indicating that the model has good accuracy and reliability [49]. The score map is used to directly visualize the similarities and differences between samples, and the remote clustering in the score map represents the differences between samples. Samples from five different harvesting periods can be separated well, indicating that the HS-GC-IMS method is very accurate in classifying the CRZP at different stages [24,26,47]. To further determine the key aroma compounds of CRZP in different harvesting periods, the VIP value was used to evaluate the degree of influence and explanatory power of each variable on sample classification. We took a VIP value greater than one as a significant criterion ( Figure 5C) [24,27]. A total of twelve VIP > one volatile compound were identified, among which hexanal D and cisianethol were the key discriminant compounds in the CRZP collected in October. (E)-2-Hexenal D, norbornane, and 7,7-dimethyl-2-meth were the key discriminant compounds in the sample collected in November, while α-Pinene, Decanal M, Linalool M, and α-Thujene were identified in the sample collected in December. Octanal, 4-allylanisole, and βmyrcene were identified in the samples collected in January and February of the following year, and decanal D in the samples collected in February was the key volatile compound to distinguish the samples collected in January and February. These compounds can be used as potential markers to distinguish CRZP samples collected during different harvesting periods.

Plant Materials and Preparation
To exclude the influence of the growing environment and genetic factors on the experimental results, this study used the location and specific plants to study the peel of an orange at different maturity stages. All samples were taken from Bentu Township, Derong  Figure 5D), indicating that the model has good accuracy and reliability [49]. The score map is used to directly visualize the similarities and differences between samples, and the remote clustering in the score map represents the differences between samples. Samples from five different harvesting periods can be separated well, indicating that the HS-GC-IMS method is very accurate in classifying the CRZP at different stages [24,26,47]. To further determine the key aroma compounds of CRZP in different harvesting periods, the VIP value was used to evaluate the degree of influence and explanatory power of each variable on sample classification. We took a VIP value greater than one as a significant criterion ( Figure 5C) [24,27]. A total of twelve VIP > one volatile compound were identified, among which hexanal D and cisi-anethol were the key discriminant compounds in the CRZP collected in October. (E)-2-Hexenal D, norbornane, and 7,7-dimethyl-2-meth were the key discriminant compounds in the sample collected in November, while α-Pinene, Decanal M, Linalool M, and α-Thujene were identified in the sample collected in December. Octanal, 4-allylanisole, and β-myrcene were identified in the samples collected in January and February of the following year, and decanal D in the samples collected in February was the key volatile compound to distinguish the samples collected in January and February. These compounds can be used as potential markers to distinguish CRZP samples collected during different harvesting periods.

Plant Materials and Preparation
To exclude the influence of the growing environment and genetic factors on the experimental results, this study used the location and specific plants to study the peel of an orange at different maturity stages. All samples were taken from Bentu Township, Derong County, Garze Prefecture, Sichuan Province, China (99 • 16 37 E; 28 • 32 32 N; altitude: 2225.15 m); 4 adjacent Zangju trees with high yield and stable quality were randomly selected, numbered and listed. A total of 6 fresh fruits of similar size and free of pests and diseases were randomly picked from each tree, sampled once a month (October 2022 to February 2023), and numbered ZGP, ZIP, ZJP, ZKP, and ZLP in turn. According to the appearance and colour characteristics, the five harvesting periods were divided into four ripening stages: the immature stage (October) showed dark green peel, the incomplete maturity stage (November) showed chartreuse peel, the commercial ripening stage (December) showed aurantiacus pericarp and the fully matured stage showed orangered pericarp (January and February). The collected samples were washed, peeled manually, and dried by hot air at a constant temperature of 50 • C for 18 h [20]. After preparation, all samples were stored in a dry place at room temperature. The CRZP samples for each stage are shown in Figure 6.  [20]. After preparation, all samples were stored in a dry place at room temperature. The CRZP samples for each stage are shown in Figure 6.

E-Nose Analysis.
E-nose analysis was performed using an 8 MEMS-MOS E-nose (Isensortalk Co., Ltd., Beijing, China). First, 1.5 g of the sample was placed into a 20-millilitre headspace vial. Two probes were inserted into the headspace vial, one connected to the sensor chamber and the other to a charcoal filter. Each sample was run in triplicate, and data were obtained from the sensor array 60 s after injection to determine the stability of the sensor signals. Data from 55 s to 58 s were taken for subsequent statistical analysis. After sample analysis, the system was purified with filtered, clean air for 120 s to reestablish the instrument baseline. All samples were tested at room temperature.

HS-GC-IMS Analysis
In this research, VOCs were determined using a GC-IMS instrument (FlavourSpec ® , G.A.S., Dortmund, Germany) equipped with an automatic sampler. First, 0.2 g of CRZP was accurately weighed, placed in a 20-millilitre headspace injection bo le and incubated for 15 min at 80 °C and 500 rpm. The volume of automatic headspace injection was 200 µL, and the temperature of the injection needle was 85 °C. Then, the extracted volatile compounds were pre-separated on a gas chromatographic column (MXT-5, 15 m, ID: 0.53 mm) and transported to IMS with carrier gas for a total of 40 min of analysis. Each sample

E-Nose Analysis
E-nose analysis was performed using an 8 MEMS-MOS E-nose (Isensortalk Co., Ltd., Beijing, China). First, 1.5 g of the sample was placed into a 20-millilitre headspace vial. Two probes were inserted into the headspace vial, one connected to the sensor chamber and the other to a charcoal filter. Each sample was run in triplicate, and data were obtained from the sensor array 60 s after injection to determine the stability of the sensor signals. Data from 55 s to 58 s were taken for subsequent statistical analysis. After sample analysis, the system was purified with filtered, clean air for 120 s to reestablish the instrument baseline. All samples were tested at room temperature.

HS-GC-IMS Analysis
In this research, VOCs were determined using a GC-IMS instrument (FlavourSpec ® , G.A.S., Dortmund, Germany) equipped with an automatic sampler. First, 0.2 g of CRZP was accurately weighed, placed in a 20-millilitre headspace injection bottle and incubated for 15 min at 80 • C and 500 rpm. The volume of automatic headspace injection was 200 µL, and the temperature of the injection needle was 85 • C. Then, the extracted volatile compounds were pre-separated on a gas chromatographic column (MXT-5, 15 m, ID: 0.53 mm) and transported to IMS with carrier gas for a total of 40 min of analysis. Each sample was analysed in parallel 3 times. The column temperature and IMS temperature were 60 • C and 45 • C, respectively. The carrier gas/drift gas was high-purity nitrogen (99.99%). The flow rate of the drift gas was set at 150 mL/min. The flow rate of the carrier gas was 2 mL/min in the initial 2 min, increased to 10 mL/min in 2-10 min, increased to 100 mL/min in 10-20 min, and then maintained for 20 min. The whole process took 40 min. Volatile compounds were identified based on RI and the draft time of standard substances in the GC-IMS library (G.A.S.).

Statistical Analysis
The results listed in the table and figure are the average values of three experiments. VOCal software (v 2.2) views the qualitative and quantitative analysis of spectra and data and uses the built-in NIST and IMS databases of VOCal software to conduct qualitative analysis of substances. Each point obtained in the spectra represents a volatile component. The Reporter plug-in compares the spectral differences between samples (three-dimensional and differential spectrograms); the Gallery Plot plugin performs fingerprint comparison to visually and quantitatively compare the differences in volatile compounds between different samples. WinMester software (1.6.2.18, Airsense Analytics GmbH, Schwerin, Mecklenburg, Germany) was used to draw LDA diagrams and radar charts. PLS-DA analyses were performed using MetaboAnalyst 5.0. R software was used to visualize the PCA, cluster analysis and heatmap data.

Conclusions
In this study, E-nose and HS-GC-IMS were used to detect VOCs in five different harvesting periods of the CRZP. The VOC fingerprints were established based on the HS-GC-IMS and PLS-DA models. Seventy-eight volatile compounds were identified, including ten alcohols, twenty-five aldehydes, seventeen terpenoids, nine ketones, eight esters and nine other volatile substances. The E-nose is not sensitive to important aroma compounds such as alcohols and aldehydes in CRZP and can only represent part of the aroma, which includes alkane aromatics, terpenes and sulfur-containing organic compounds. The results of HS-GC-IMS showed that the contents of alcohols, aldehydes and monoterpenoids were relatively high. The VOCs in the CRZP increased, decreased and fluctuated during different harvesting periods. The alcohols and aldehydes that produce unripe fruit aromas, such as (z)-2-penten-1-ol, α-terpinene M/D and dodecanal, were relatively high in the immature stage (October to November). In contrast, monoterpenoids, ketones and esters with ripe fruit aromas, such as 2-butanone, 2-methylbutyraldehyde M/D, cyclohexanone D, α-thujene, β-myrcene, γ-terpinene, crotonic acid, hexyl ester, etc., are relatively high in the full maturity stage (January to February). The results of PCA and PLS-DA showed that the volatile organic compounds in the samples at different harvesting stages were well separated and easy to distinguish. A total of twelve key volatile compounds were identified by the VIP method to distinguish samples picked at different stages. In other words, the relative VOC content of CRZPs harvested in October was the highest. The results of this study can help us comprehensively understand the flavour characteristics of CRZP during the ripening process, provide a basis for the application of CRZP, a byproduct of Zangju, in industrial production (food, cosmetics, flavour and fragrance), and provide a reference for similar research on other C. reticulata varieties.