Early Detection of Monilinia laxa in Nectarine (Prunus persica var. nectarina) Using Electronic Nose Technology: A Non-Destructive Diagnostic Approach

Abstract

This study evaluates the application of an electronic nose (E-nose) system as a non-destructive tool for the early detection of Monilinia laxa infection in yellow nectarines (Prunus persica var. nectarine, cv. “Kinolea”) through the analysis of volatile organic compounds (VOCs). Two experimental groups were established: a control group of healthy fruit and a treatment group inoculated with the pathogen. The VOCs emitted by both groups were identified and quantified using gas chromatography-mass spectrometry (GC-MS). Simultaneously, the responses of the E-nose were recorded at three critical stages of fungal development: early, intermediate, and advanced. The electronic nose used consists of a set of 11 commercial metal oxide semiconductor (MOX) sensors. The signals from these sensors showed a strong correlation with the VOC profiles associated with fungal deterioration. Linear discriminant analysis (LDA) models based on E-nose data successfully distinguished between healthy and infected samples with 97% accuracy. Furthermore, the system accurately classified samples into three stages of contamination—control, early infection, and advanced infection—with 96% classification accuracy. These findings demonstrate that E-nose technology is an effective, rapid, and non-invasive method for the real-time monitoring of post-harvest fungal contamination in nectarines, offering significant potential for improving quality control during storage and distribution.

1. Introduction

Brown rot, caused by species of the genus Monilinia, is one of the most significant postharvest diseases affecting stone fruits. Among the primary causal agents are Monilinia laxa, Monilinia fructicola, and Monilinia fructigena. Historically, M. laxa has been the most prevalent in Europe, accounting for 85–90% of infections, followed by M. fructigena (10–15%) [1]. However, the introduction of M. fructicola in the early 21st century led to its rapid spread across several European countries, including Spain, Italy, Switzerland, and Slovakia [2,3,4]. M. fructicola was originally identified in North America in 1883 and subsequently reported in various countries, including California (1936), Canada (1976), Mexico (1999), Guatemala, and Panama (1976 and 1999, respectively), and several South American countries such as Argentina, Bolivia, Brazil, Peru, Venezuela (1976), and Ecuador, Paraguay, and Uruguay (1999) [5]. This rapid dissemination is attributed to its faster growth rate, higher spore production, and greater dispersal efficiency compared to M. laxa and M. fructigena. Moreover, M. fructicola exhibits a higher tendency to develop resistance to fungicides such as benzimidazoles, dicarboximides, and triazoles [6].

While control strategies for M. fructicola are well-documented, research on M. laxa remains limited, despite its continued impact. The growing demand for high-quality fruit and increasing restrictions on fungicide use have made M. laxa a major challenge for European stone fruit production [7]. Consequently, early detection methods are essential. Technologies such as near-infrared (NIR) spectroscopy, hyperspectral imaging (HSI), and electronic noses offer rapid, non-destructive solutions for assessing fruit quality and detecting spoilage. These techniques enable the quantification of quality parameters and classification of fruits based on attributes such as colour, shape, size, and texture [8].

The electronic nose (E-nose) is an emerging analytical technology characterized by its speed, low cost, and short response time. It typically comprises four main components: a gas sensor array, a sampling system, a data acquisition unit, and a data processing module. This technology has advanced rapidly and demonstrates broad applicability in the food industry. It has been successfully employed for tasks such as classification, identification, monitoring, quality control, and traceability [9]. Thus, Chen et al. [10] used an E-nose coupled with gas chromatography (GC) to classify ten different date varieties, employing Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) for data interpretation. Beyond varietal classification, E-nose systems have also proven effective in detecting sensory defects in foods—such as zapateria, butyric, putrid, and musty off-flavours in table olives—often resulting from abnormal fermentations [11].

Given its versatility, the E-nose is also valuable for the early detection of fungal contamination in fruits, which are highly susceptible to fungal infections during the pre-harvest, post-harvest, and storage stages [12]. Pan et al. [13] analyzed the volatiles emitted by control strawberries and strawberries inoculated with Botrytis spp., Penicillium spp. and Rhizopus spp. for 10 days by gas chromatography–mass spectrometry (GC–MS) and E-nose. The results showed that, on the second day, discrimination between control and infected strawberries was achieved with a classification accuracy of 96%. Similarly, Haghbin et al. [14] were able to detect early Botrytis spp. contamination in kiwifruit with a discrimination rate of over 90%.

The electronic nose system employed by Jia et al. [15] successfully distinguished between fresh and mould-infected apples inoculated with Penicillium expansum and Aspergillus niger. Similarly, Gu et al. [16] achieved a 92.86% accuracy rate in differentiating healthy rice batches from those contaminated with Aspergillus as early as the second day of incubation. Complementary gas chromatographic analysis revealed that the volatile profiles were closely associated with the specific fungal species present in the rice.

To interpret the complex datasets generated by chemical sensor arrays, advanced multivariate statistical and machine learning techniques are essential. Methods such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Partial Least Squares Regression (PLSR), along with algorithms like Support Vector Machines (SVMs) and Artificial Neural Networks (ANNs), have proven effective in extracting meaningful patterns and enhancing classification accuracy [17,18,19]. These analytical tools play a critical role in maximizing the diagnostic potential of E-nose systems in food quality and safety applications.

The primary objective of this study is to analyze the volatile organic compounds (VOCs) emitted by nectarines in order to identify potential biomarkers associated with fungal decay caused by Monilinia spp. These biomarkers must be characterized for each specific association between pathogen and fruit variety, as the VOC profile can vary significantly depending on the cultivar and the infecting species [18]. Special emphasis is placed on investigating the correlation between specific VOCs and the response patterns of metal oxide (MOX) sensors integrated into an electronic nose (E-nose) system. The most suitable MOX sensors must also be evaluated to ensure the appropriate design of the E-nose, optimizing its sensitivity and selectivity for detecting relevant VOCs [20,21,22,23]. Ultimately, the study aims to assess the feasibility and diagnostic performance of this portable, non-invasive technology for the early detection of Monilinia infections in nectarines under postharvest conditions. This contrasts with the more complex and laboratory-dependent setups used in previous studies [20,21,23], providing a technological framework closer to non-destructive applications that are potentially transferable to the post-harvest chain.

2. Materials and Methods

2.1. Experimental Design

The fungal strain Monilinia laxa CA1, previously characterized by Ruíz-Moyano et al. [24], was obtained from the Culture Collection of the CAMIALI research group at the University of Extremadura (UEx). For experimental use, the strain was cultured on potato dextrose agar (PDA) plates and incubated at 25 °C for 10 days to ensure optimal mycelial growth.

Yellow-fleshed nectarines (Prunus persica var. nectarina, cv. ‘Kinolea’) were sourced from a major commercial supermarket during August and September 2023, coinciding with their typical harvest period in early August. To preserve freshness, fruits were stored at 1 °C for no longer than 48 h prior to experimental use. Before inoculation, nectarines were surface-sanitated using 70% ethanol and air-dried under sterile conditions. Each fruit was then wounded at a single site using a sterile instrument to create a standardized lesion (3 mm wide × 3 mm deep). Inoculation was performed by placing a PDA cube containing actively growing M. laxa mycelium into the wound. Two experimental groups were established (inoculation factor): (i) inoculated fruits (ML) and (ii) non-inoculated, sound fruits serving as controls. Control nectarines were wounded in the same way as inoculated fruits but did not receive any fungal material.

Post-inoculation, fruits were individually placed in transparent polyethylene containers (25 × 25 × 11 cm; width × height × depth) and incubated at 25 °C. Volatile organic compound (VOC) analysis was conducted using gas chromatography coupled with mass spectrometry (GC-MS) and an electronic nose (E-nose). Three stages of infection were defined according to lesion diameter (stage factor): (i) early (18 mm, 2 days after inoculation), (ii) middle (37 mm, 3 days) and (iii) advanced (57 mm, 4 days). Each stage was considered as an independent batch, and each was associated with its corresponding control (healthy, non-inoculated nectarines stored under the same conditions). For each batch, three biological samples per treatment were used to ensure the reproducibility of the results.

2.2. Volatile Compound Analysis

2.2.1. Volatile Extraction

To capture volatile organic compounds (VOCs), samples were placed in glass jars with plastic lids, each with a capacity of 780 cm³. These jars were pre-conditioned before sealing. The headspace was purged with GC-grade air for 2 min to remove residual VOCs from previous samples. Volatile extractions were performed by directly inserting the HS-SPME fibre through the vial cap, extracting VOCs at 25 °C for 40 min. Gas chromatography/mass spectrometry (GC/MS) analysis was performed as detailed in Section 2.2.2. Each analysis was performed in triplicate, and the entire experiment was repeated three times to ensure reliability. Fibres were activated according to conditioning guidelines prior to initial use.

2.2.2. Gas Chromatography/Mass Spectrometry (GC/MS) Analysis

The identification and quantification of volatile organic compounds (VOCs) were performed using gas chromatography/mass spectrometry (GC/MS) following the methodology described by Serradilla et al. [25]. The analysis utilized an Agilent 6890 GC system coupled to a 5973 mass spectrometer (Agilent Technologies, Little Falls, DE, USA). Separation was carried out on a 50 m × 0.32 mm i.d. capillary column with a 1.05 μm film thickness, coated with 5% phenyl and 95% polydimethylsiloxane (HP-5MS equivalent, model 19091J-215, Agilent Technologies, Madrid, Spain).

Compound identification involved a two-pronged approach: first, Kovats retention indices, calculated using n-alkanes (R-8769, Sigma Chemical Co., St. Louis, MO, USA), were employed. Second, the obtained mass spectra were compared against the NIST/EPA/NIH library, with a minimum match quality of 90% required. Furthermore, the identity of specific compounds was verified by comparing their retention times and mass spectra to those in an in-house library generated from the analysis of pure standards under identical experimental conditions. Quantification was achieved by integrating peak areas in the total ion current chromatograms to determine relative abundance.

2.3. E-Nose Analysis

2.3.1. E-Nose System

A custom-built electronic nose (E-nose) system, developed by the Sensory Systems research team at the University of Extremadura [26], was used in this study (Figure 1). This E-nose features an array of four commercially available metal oxide semiconductor (MOX) sensors providing a total of eleven signals, enabling a broad spectrum of detection. These sensors were sourced from several manufacturers: (i) Bosch BME680, which measures environmental parameters such as temperature (°C), pressure (hPa), humidity (%RH), and gas resistance (Ω); (ii) Sensirion SGP30, which detects equivalent CO₂ (eCO₂) concentration (CO₂SGP30), total volatile organic compound concentration (TVOCSGP30), and raw resistive values for hydrogen (H₂SGP30) and ethanol (EtanolSGP30); and (iii) ScioSense CCS811 and (iv) iAQ-Core, both of which measure eCO₂, TVOC, and sensor resistance (CO₂CCS811, TVOCCCS811, ResohmCCS811, CO₂iAQ, TVOCiAQ, RiAQCore). Detailed specifications of the sensors are provided in Appendix A (

Table A1. Specifications of the gas sensors used in the study.

Sensor	Manufacturer	Type	Measured Parameters (Signals)	Unit	Variable Name	Measurement Range
BME680	Bosch Sensortec GmbH (Reutlingen, Germany)	MOX	Temperature, Pressure, Relative humidity, Gas resistance	°C, hPa, %RH, Ω	RBME680	Temp: −40–85 °C; Pressure: 300–1100 hPa; Humidity: 0–100%RH
SGP30	Sensirion AG (Stäfa, Switzerland)	MOX	Equivalent CO2 concentration, Total VOC, Hydrogen (resistive), Ethanol (resistive)	ppm, ppb, -,-	CO2SGP30, TVOCSGP30, H2SGP30, EtanolSGP30	eCO2: 400–60,000 ppm; eTVOC: 0–60,000 ppb
CCS811	ScioSense B.V. (Eindhoven, The Netherlands)	MOX	Equivalent CO2 concentration (eCO2), Equivalent total VOC (eTVOC), Sensor resistance	ppm, ppb, Ω	CO2CCS811, TVOCCCS811, ResohmCCS811	eCO2: 400–29,206 ppm; eTVOC: 0–32,768 ppb
iAQ-Core	ScioSense B.V. (Eindhoven, The Netherlands)	MOX	Equivalent CO2 concentration, Total VOC, Sensor resistance	ppm, ppb, Ω	CO2iAQ, TVOCiAQ, RiAQCore	eCO2: 450–2000 ppm; eTVOC: 125–600 ppb

).

Each sensor is integrated within a compact module that includes analogue conditioning circuits, analogue-to-digital converters, a microcontroller, communication capabilities, and a heated microplate. The system is powered by a 3.7 V lithium battery and transmits data wirelessly via Bluetooth to a dedicated mobile phone application.

2.3.2. Measurement Process and Data Analysis

For this study, a total of 98 measurements were taken with the E-nose, distributed among the different batches established (Table S1). Three samples of nectarines were analyzed individually for each batch established: each piece was placed in an airtight container to allow volatile compounds to accumulate in its headspace.

The E-nose captured these volatiles for 120 s, in what is called the Adsorption Phase, in which the sensors were directly exposed to the headspace of the sample to allow the odour molecules to be adsorbed onto their surface. This was followed by a Desorption Phase, also lasting 120 s, during which the sensors were exposed to clean air to establish a baseline for subsequent odour differentiation. The instrument recorded sensor responses every 2 s, providing around 60 data points per phase (approximately 120 points per cycle). This measurement cycle (adsorption + desorption) was repeated five times per sample, and the complete response curves were stored.

To obtain a representative value for each measurement cycle, a baseline manipulation algorithm was implemented. This characteristic value was calculated as the proportional difference between a reference value (V_ref) and the value recorded during exposure to volatile compounds (V_odor). The reference value (V_ref) was defined as the average of the last five readings taken during the sensor’s exposure to the air phase. On the other hand, the value of the volatiles (V_odor) was defined as the average of the last five readings recorded during the sensor’s exposure to the headspace of the sample. The final value (V_f) was calculated using the following expression:

$$V_f=\left(V_{ref}-V_{odor}\right)\times 100-1$$

Although only these steady-state values were used in the present analysis, recording the full adsorption and desorption curves ensured sensor stability and recovery, avoided carry-over effects, and also provides the possibility of applying more complex feature extraction methods in future studies.

2.4. Statistical Analysis

All statistical analyses were performed using SPSS Statistics 21.0 (IBM Corp., Armonk, NY, USA). A descriptive analysis was first conducted to summarize the concentration and distribution of volatile organic compounds (VOCs) identified in both control and inoculated nectarine samples.

To evaluate the effects of fungal inoculation and infection stage on VOC profiles, a two-way analysis of variance (ANOVA) was applied. This method allowed for the assessment of main effects and interactions between the two factors. Post hoc comparisons were conducted using Tukey’s HSD test to identify significant differences between groups (p < 0.05).

Principal Component Analysis (PCA) was employed to reduce the dimensionality of the VOC dataset and to visualize the clustering of samples based on their volatile profiles. PCA was conducted using the correlation matrix, and the first few principal components explaining the highest variance were retained for interpretation.

To explore the relationship between VOCs and E-nose sensor responses, a Hierarchical Cluster Analysis (HCA) was performed. Pearson correlation coefficients were calculated between the integrated peak areas of VOCs and the response values of individual metal oxide (MOX) sensors. Clustering was based on squared Euclidean distances and the agglomerative intergroup linkage method, enabling the identification of sensor response patterns associated with different infection stages.

Linear Discriminant Analysis (LDA) was used to classify samples according to infection stage (control, early, and advanced). A stepwise selection method based on Wilks’ lambda was applied to identify the most discriminative variables. The performance of the LDA model was evaluated using classification accuracy, and the results were visualized by projecting the samples onto the space defined by the first two discriminant functions. In addition, variable loadings derived from the sensor signals were projected onto the discriminant function space (DF1 and DF2) to visualize the contribution of each sensor to the classification model.

3. Results and Discussion

3.1. Volatile Organic Compounds

A total of 54 volatile organic compounds (VOCs) were identified in yellow-fleshed nectarines. These compounds were categorized into the following chemical families: esters (31), alcohols (5), hydrocarbons (5), other compound as lactone, aromatic compound or furan (5), aldehydes (4), terpenoid (2), ketones (1), carboxylic acid (1) (

Table 1. Volatile compounds identified in nectarine samples.

					Mean 5			p Values 7
RT 1	CD 2	Volatile Compounds	ID 3	KI 4	AAU	%	RSD 6	Ps	Pi
	Hydrocarbons				3872	2.00
6.6	v16	2,4-dimethylheptane	A	834	156	0.08	160		--
8.4	v18	Ethylbenzene	C	897	336	0.18	128	---	---
14.3	v25	2,2,4,6,6-pentamethylheptane	C	992	866	0.47	67	---	---
32.7	v53	Pentadecane	A	1500	1473	0.79	71	-
36.2	v54	Heptadecane	A	1700	886	0.48	63	+
	Alcohol				14,729	7.93
1.9	v2	Propan-1-ol	A	595	74	0.04	244	--	++
2.9	v6	Pent-1-en-3-ol	B	689	882	0.47	131		+++
3.8	v10	2-methylbutan-1-ol	B	731	4510	2.43	142		+++
4.6	v11	Pentan-1-ol	B	762	9059	4.88	130		+++
21.1	v35	(Z)-non-3-en-1-ol	B	1156	204	0.11	127		+++
	Aldehyde				2461	1.33
2.6	v5	3-methylbutanal	C	658	86	0.05	201	+++	+++
5.7	v14	Hexanal	B	803	834	0.45	310
19.1	v32	Nonanal	B	1105	974	0.52	64	--
23	v39	Decanal	B	1240	567	0.31	30	--	+++
	Ketone				4241	2.28
3.1	v7	Pentan-3-one	B	704	4241	2.28	89	---	+++
	Carboxilic acid				76	0.04
27.7	v45	(4E)-3-methyl-4-decenoic acid	C	1430	76	0.04	203	--	+++
	Ester				117,912	63.5
1.7	v1	Methyl acetate	A	522	1369	0.74	187	--	++
2.1	v3	Ethyl acetate	A	605	21,945	11.82	66	+++	---
3.4	v8	Propyl acetate	B	715	996	0.54	105
3.6	v9	Methyl butanoate	B	723	595	0.32	226	+++	+++
4.8	v12	2-methylpropyl acetate	B	769	4334	2.33	120
5	v13	Methyl 3-methylbutanoate	B	777	412	0.22	166
5.8	v15	Ethyl butanoate	B	807	3642	1.96	141	+++	+++
8.2	v17	Ethyl 3-methylbutanoate	B	890	4481	2.41	76	--	+++
9.3	v20	2-methylbutyl acetate	B	913	2415	1.30	73	---	++
10.4	v21	Ethyl pentanoate	B	930	982	0.53	195	+++	+++
11.1	v23	Pentyl acetate	B	941	9012	4.85	119	-	+++
11.5	v24	Methyl hexanoate	B	948	762	0.41	112		+++
14.9	v27	Ethyl hexanoate	B	1002	5514	2.97	66	---	+++
15.2	v28	[(E)-hex-3-enyl] acetate	C	1010	4876	2.63	99
15.5	v29	Hexyl acetate	B	1017	2532	1.36	184	--
18.8	v30	Pentyl butanoate	B	1098	896	0.48	131	---	+++
20	v33	Methyl octanoate	B	1128	8440	4.54	100	---	+++
20.7	v34	Pentyl 3-methylbutanoate	C	1146	2767	1.49	133	+	+++
22.4	v37	Ethyl oct-7-enoate	C	1190	1066	0.57	65	--	++
22.7	v38	Ethyl octanoate	B	1197	36,175	19.48	59	---	+++
24.1	v40	[(Z)-hex-3-enyl] 3-methylbutanoate	C	1335	568	0.31	76	---
24.3	v41	Hexyl 3-methylbutanoate	C	1343	142	0.08	100	-
25.9	v42	Pentyl hexanoate	C	1404	512	0.28	139	---	+++
26.1	v43	Propyl octanoate	C	1407	84	0.05	177		++
26.7	v44	Methyl dec-4-enoate	C	1415	555	0.30	123		+++
27.9	v46	2-methylpropyl octanoate	C	1432	215	0.12	136	---	+++
29	v47	Ethyl (E)-dec-4-enoate	C	1448	1214	0.65	78	---	+++
29.2	v48	Pentyl heptanoate	C	1451	81	0.04	196	--	+++
31.1	v49	3-methylbutyl octanoate	C	1477	68	0.04	191		++
31.2	v50	2-methylbutyl octanoate	C	1479	50	0.03	203		++
32.3	v52	Pentyl octanoate	C	1494	1211	0.65	135	---	+++
	Terpeniods				25,617	13.79
19	v31	Linalool	B	1103	25,469	13.71	98	--
22	v36	2-methylisoborneol	C	1179	148	0.08	254	+++	+++
	Other compunds				16,800	9.13
2.2	v4	Unidentified compound	D	616	7469	4.02	254	---
8.6	v19	4-methylactone	B	902	155	0.08	171		--
10.9	v22	Methyl (Z)-N-hydroxybenzenecarboximidate	C	938	253	0.14	152		-
14.4	v26	Pentylfuran	B	994	7106	3.83	70		+++
31.8	v51	2(3H)-furanone, 5-hexyldihydro	C	1487	1972	1.06	70

¹ RT: Retention time (min). ² CD: Code of volatile compound used in Figure 1 and Figure 2. ³ ID: Reliability of identification: A, identified by a comparison to standard compounds; B, tentatively identified by the NIST/EPA/NIH mass spectrum library (comparison quality > 90%) and Kovats index; C, tentatively identified by the NIST/EPA/NIH mass spectrum library (comparison quality < 90%); D, Unidentified compound. ⁴ KI: Kovats retention index. ⁵ AAUd: Arbitrary Area Units; (%): Relative percentage. ⁶ RSD: Relative Standard Deviation. ⁷ Ps: p values of stage factor; Pi: p values of inoculation factor. The significance of the effects is indicated by + (positive effect) or - (negative effect). One, two, and three symbols correspond to p-values less than 0.1, 0.05, and 0.01, respectively.

). The aroma profile of stone fruits is notably diverse, as evidenced by the extensive body of literature dedicated to their characterization [27,28,29]. Among the various classes of volatile compounds, esters represent the predominant group, playing a central role in imparting the characteristic fruity and floral aromas of peaches and nectarines [30].

In the present study, esters emerged as the most abundant class, both in terms of diversity and relative concentration. The major ester compounds detected were ethyl octanoate (v38) (19.48%), ethyl acetate (v3) (11.82%), pentyl acetate (v23) (4.85%), and methyl octanoate (v33) (4.54%), underscoring their significant contribution to the overall aroma profile. In general, ester concentrations increased in nectarine samples infected with M. laxa, with the notable exception of ethyl acetate—a compound typically abundant in stone fruits such as peaches and nectarines [30]. Increase in esters in stone fruit inoculated with Monilinia spp. has also been described by Fanesi et al. [31], agreeing on compounds such as propyl acetate, methyl 3-methylbutanoate, pentyl acetate, ethyl hexanoate, ethyl oct-7-enoate, ethyl octanoate, 2-methylpropyl octanoate and ethyl (E)-dec-4-enoate. As storage time progressed, a general decline in ester concentrations was observed, consistent with the findings reported by Visai et al. [32]. However, several compounds deviated from this overall trend. Notably, ethyl acetate (v3), methyl butanoate (v9), ethyl butanoate (v15), ethyl pentanoate (v21), and pentyl 3-methylbutanoate (v34) exhibited increasing concentrations over time (Table 1 and Table S2). These results indicate a compound-specific response to both fungal infection and postharvest storage conditions.

Among the alcohols identified, pentan-1-ol (v11) emerged as the most abundant compound in terms of relative percentage (4.88%), followed by 2-methylbutan-1-ol (v10) and pent-1-en-3-ol (v6) (Table 1). This pattern aligns with findings by Fanesi et al. [31], who also reported pentan-1-ol as the predominant alcohol in M. fructicola-infected peaches. In the present study, all quantified alcohols demonstrated statistically significant associations with the inoculation factor, with p-values indicating strong positive effects. Conversely, the influence of the storage stage was limited, with only propan-1-ol (v2) exhibiting a moderately significant negative response.

In the same way, the analysis of carbonyl and carboxylic acid profiles revealed that inoculation exerted a consistently strong positive effect, markedly enhancing the concentrations of several compounds, including 3-methylbutanal (v5), decanal (v39), pentan-3-one (v7), and (4E)-3-methyl-4-decenoic acid (v45) (Table 1). Notably, decanal has been previously identified as a potential agent for brown rot suppression. However, its dual origin—from both healthy nectarine tissue and M. laxa cultured in vitro on peach juice-based medium—raises questions regarding its specificity and functional role [33]. In contrast, the stage factor was associated with a significant reduction in the levels of nonanal (v32), decanal (v39), pentan-3-one (v7), and the carboxylic acid, suggesting a stage-dependent modulation of these volatile compounds.

Terpenoids represent a significant portion of the volatile profile in nectarines, accounting for 13.79% of the total signal intensity. Among them, linalool (v31) was the predominant compound, contributing 13.71% of the total volatiles (Table 1). This monoterpene alcohol is well-documented as a key aroma compound in peaches and nectarines, particularly in mature fruit [34,35]. Its concentration is known to be cultivar-dependent [30] and sensitive to postharvest conditions. In our study, linalool levels remained high but showed no statistically significant variation in response to either cold storage or inoculation. In contrast, 2-methylisoborneol (v36), although present at a much lower concentration (0.08%), exhibited a highly significant increase in response to both storage and inoculation. This compound, often associated with earthy or musty odours, is indicative of fungal activity, and its presence could serve as a potential marker for fruit deterioration or infection [36].

Hydrocarbons are generally considered minor contributors to the aroma profile of nectarines, with previous studies reporting their presence in low concentrations and limited sensory relevance [30]. Consistent with these findings, our analysis revealed that hydrocarbons accounted for only 2.00% of the total volatile profile (Table 1). Among the identified hydrocarbons, pentadecane (v53) and heptadecane (v54) were the most abundant, with relative abundance of 0.79 and 0.48%, respectively. Most other hydrocarbons showed no significant or negative response to either inoculation or storage stage, and their relative abundances remained below 0.5%.

Within the group categorized as “Other compounds,” pentylfuran (v26) (3.83%) and 2(3H)-furanone, 5-hexyldihydro (v51) (1.06%) emerged as notable contributors. Pentylfuran, in particular, exhibited a highly significant positive response to inoculation (Pi: +++), suggesting that its production may be stimulated by M. laxa infection or associated stress responses of fruits.

3.2. Relationship Between VOCs and Signals from E-Nose Sensors

To further elucidate the multivariate relationships among the volatile compounds correlated with the MOXs signals, a principal component analysis (PCA) was conducted (Figure 2). This dimensionality reduction technique enables the visualization of complex patterns and interactions between storage and inoculation factors, facilitating the identification of key volatiles that contribute most to the differentiation of nectarine samples.

The first principal component (PC1), accounting for 58.43% of the total variance, effectively discriminated control samples across storage stages (C_1S, C_2S, and C_3S) from those inoculated with Monilinia, with the former and latter distributed along the negative and positive axes of PC1, respectively (Figure 2B). Volatile compounds such as ethyl acetate (v3), methyl (Z)-N-hydroxybenzenecarboximidate (v22), 4-methylactane (v19), ethylbenzene (v18) and 2,4-dimethylheptane (v16) were predominantly associated with control batches, suggesting their potential as markers of unaltered fruit. In fact, lactones have been associated with the intense aroma of various peach cultivars [30,34]. In addition, peaches infected with M. fructicola exhibit significantly lower concentrations of these compounds, as well as reduced levels of hydrocarbons [31].

The second principal component (PC2), explaining an additional 16.02% of the variance, enabled further differentiation among inoculated samples, particularly separating early and mid-stage infections (M_1S and M_2S) from late-stage infections (M_3S). Volatiles such as decanal (v39), ethyl (E)-dec-4-enoate (v47), pentyl octanoate (v52), and 2-methylpropyl octanoate (v46) were linked to early infection stages, while compounds like 2-methylisoborneol (v36), methyl butanoate (v9), and ethyl pentanoate (v21) were indicative of advanced fungal colonization [36]. These findings highlight the distinct profiles of volatile compounds associated with varying degrees of fruit spoilage, underscoring their potential utility in the development of an E-nose system for rapid, non-invasive detection of postharvest spoilage in nectarines.

The electronic nose is an emerging analytical tool with broad applications across various food matrices, including beverages such as wine and tea, as well as grains, oils, meats, fruits, and vegetables [10,37]. To further classify or group samples based on similarities in their volatile profiles, Hierarchical Cluster Analysis (HCA) is often employed [38]. The integration of complementary data sources, such as volatile compound analysis, can enhance the accuracy and robustness of these classification models [39,40].

In our study,

Table 2. Cluster groups of volatile compounds according to Pearson correlation values between their area and the responses values of the different MOX used.

			MOX 1,2
Cluster	CD	VOCs	M_1	M_4	M_5	M_8	M_2	M_3	M_6	M_7	M_9	M_10	M_11
1	V3	Ethyl acetate	--	--	--	--	++	++	++	++	++	++	-
3	V16	2,4-dimethylheptane	-	-	-	-			+				-
3	V18	Ethylbenzene							+	+
3	V19	4-methylactone	-	-	-	-			+				--
3	V22	Methyl (Z)-N-hydroxybenzenecarboximidate	--	--	--	--			+		+	+	--
2	V5	3-methylbutanal	+	+	+	+	--	--	-	--	--	--
2	V6	Pent-1-en-3-ol	++	++	++	++	--	--	--	--	--	--	+
2	V7	Pentan-3-one	++	++	++	++	--	--	--	--	--	--	++
2	V9	Methyl butanoate	++	+	+	++	--	--	--	--	--	--
2	V10	2-methylbutan-1-ol	++	++	++	++	--	--	--	--	--	--
2	V11	Pentan-1-ol	+	+	++	++	-	-	--	-	--	--
2	V12	2-methylpropyl acetate	+			+	-	-	-	-
2	V15	Ethyl butanoate	++	++	+	++	--	--	--	--	--	--
2	V21	Ethyl pentanoate	++	+	+	++	--	--	--	--	--	--
2	V23	Pentyl acetate	+	+	++	++			--		-	-	+
2	V24	Methyl hexanoate	++	++	++	++	--	--	--	--	--	--
2	V26	Pentylfuran	+	+	+	+	-	-	--	-	-	-
2	V30	Pentyl butanoate	+	+	++	++	-	-	--	-	--	--
2	V33	Methyl octanoate	+	+	++	++	-	-	--	-	-	-	+
2	V34	Pentyl 3-methylbutanoate			+	+			-
2	V35	(Z)-non-3-en-1-ol	++	++	++	++	--	--	--	--	--	--	+
2	V36	2-methylisoborneol	++	+		+	-	--	-	--	--	--
2	V42	Pentyl hexanoate		+	+	+			--
2	V44	Methyl dec-4-enoate	++	++	++	++	--	--	--	--	--	--	+
2	V46	2-methylpropyl octanoate	+	+	++	++			--		-	-
2	V47	Ethyl (E)-dec-4-enoate	+	++	++	++			--				++
2	V52	Pentyl octanoate	+	+	++	++			--		-	-
4	V27	Ethyl hexanoate		+									+
4	V39	Decanal							-

¹ MOX: RBME680 (1); CO₂SGP30 (2); TVOCSGP30 (3); H₂SGP30 (4); EtanolSGP30 (5); CO₂CCS811 (6); TVOCCCS811 (7); ResohmCCS811 (8); CO₂iAQ (9); TVOCiAQ (10); RiAQCore (11). ² Positive (+) or negative (-) correlation (p < 0.01, ***; p < 0.05, **).

presents the clustering of volatile organic compounds (VOCs) based on their Pearson correlation with the response values of 11 different metal oxide (MOX) sensors.

The analysis of sensor responses across the four VOC clusters reveals two main behavioural patterns. Clusters 1 and 3, although composed of different compounds, share a similar sensor response profile: both show positive correlations primarily with CO₂CCS811 (M_6) and TVOCCCS811 (M_7), while exhibiting negative or weak correlations with RBME680 (M_1), H₂SGP30 (M_4), and EtanolSGP30 (M_5). Cluster 1 includes only ethyl acetate (v3), whereas cluster 3 comprises hydrocarbons and ketones such as 2,4-dimethylheptane (v16) and 4-methylactone (v19), yet both elicit comparable sensor behaviour. In contrast, clusters 2 and 4 display an opposite pattern. Cluster 2, which includes a wide range of esters and alcohols associated with nectarine aroma and spoilage [30], shows strong positive correlations with RBME680 (M_1), H2SGP30 (M_4), EtanolSGP30 (M_5), and ResohmCCS811 (M_8), and strong negative correlations with CO₂SGP30 (M_2), TVOCSGP30 (M_3), CO₂CCS811 (M_6), TVOCCCS811 (M_7), CO₂iAQ (M_9), and TVOCiAQ (M_10). Cluster 4, although limited to ethyl hexanoate and decanal, mirrors this trend with weak positive responses from H₂SGP30 (M_4) and RiAQCore (M_11) and a negative response from CO₂CCS811 (M_6).

The consistent pattern of cluster 2 suggests that these sensors are particularly responsive to the chemical characteristics of this VOC group, which may be indicative of early to inter-mediate stages of fruit spoilage. The relationship of these sensors with volatile organic compounds produced by moulds has already been described by Martínez et al. [18] in apples inoculated with Penicillium species. In general, the esters identified in that study showed a positive correlation with sensors RBME680, H₂SGP30, EtanolSGP30, ResohmCCS811 and RiAQCore.

The rest of the volatile compounds did not present significant correlations with any of the sensors. Among them, ethyl octanoate (v38) stands out. This compound, despite being the most abundant volatile compound, exhibits a relatively low relative standard deviation (RSD = 59%) across samples. This lower variability likely contributes to its limited correlation with sensor responses, as MOX sensors tend to be more responsive to compounds with higher dynamic ranges. In contrast, other volatiles with greater variability may elicit stronger and more distinguishable sensor signals, enhancing their detectability and correlation with sensor outputs [41].

Overall, the clustering analysis underscores the potential of MOX sensor arrays to discriminate between VOC profiles associated with different stages or types of fruit spoilage. The distinct sensor response patterns across clusters provide a foundation for the development of targeted E-nose algorithms capable of rapid, non-invasive quality assessment in postharvest fruit monitoring.

3.3. Determination of Incipient Fungal Decay of Nectarines by E-Nose During Postharvest Storage

Linear Discriminant Analysis (LDA) models were developed using E-nose data to investigate the spoilage behaviour of nectarines during storage, comparing control lots with those inoculated with M. laxa. To rigorously assess the predictive power of these LDA models, a leave-more-out cross-validation technique was employed. The LDA models were constructed using group assignments derived from the Principal Component Analysis (PCA) of the volatile compounds, which revealed distinct clustering patterns corresponding to control and diseased fruit samples.

Table 3. Performance in calibration and prediction of Linear Discriminant Analysis (LDA) classification applied to the nectarine samples grouped in two batches.

Correctly Classified Nectarine Counts			Total
Batches	Control	M. laxa
Total	52	46	98
Computed classes	51	45	96
Predicted classes	50	45	95
Selected variable	CO2CCS811
	ResohmCCS811
	RBME680
	EtanolSGP30

summarizes the outcomes of the LDA, demonstrating its efficacy in differentiating healthy nectarines from those exhibiting fungal decay. The table prioritizes the most influential E-nose sensors, ordered by their significance as determined by the step-by-step selection algorithm utilized in this study. It also provides the number and percentage of correct classifications achieved during both calibration and prediction phases, offering a comprehensive evaluation of the model’s performance. Specifically, the CO₂CCS811, ResohmCCS811, RBME680, and EtanolSGP30 sensors collectively achieved a 97% correct classification rate in discriminating between healthy and M. laxa-infected nectarines (Table 3; Figure 3).

Furthermore, a separate analysis utilizing the CO₂CCS811, TVOCCCS811, TVOCSGP30, TVOCiAQ, RBME680, and CO₂SGP30 sensors successfully discriminated among healthy nectarines, those with incipient contamination (early and middle stages; M_1S + M_2S), and those in their advanced contamination stage (M_3S). This model achieved a 96% correct classification rate (

Table 4. Performance in calibration and prediction of Linear Discriminant Analysis (LDA) classification applied to the nectarine samples grouped in control and spoiled samples (early stage (M_1S + M_2S) and advanced stage (M_3S)).

Correctly Classified Nectarine Counts				Total
Batches	Control	M_1S + M_2S	M_3S
Total	44	34	12	90
Computed classes	43	33	12	88
Predicted classes	43	31	12	86
	CO2CCS811
	TVOCCCS811
Selected variable	TVOCSGP30
	TVOCiAQ
	RBME680
	CO2SGP30

; Figure 4).

The robust efficacy of LDA models, when applied to E-nose sensor data for accurate sample group differentiation, is well-established in the literature. Martínez et al. [18] demonstrated this by achieving over 97% accuracy in discriminating between healthy ‘Golden Delicious’ apples and those exhibiting incipient Penicillium expansum contamination. Rezaee et al. [42] successfully employed LDA models to discriminate pistachios contaminated with Aspergillus flavus across various storage durations, reporting an impressive 90% accuracy. Notably, their models achieved 100% accuracy in differentiating between varying contamination concentrations. These studies show how effective LDA models are at achieving high classification rates across a variety of sample groups using E-nose sensor data, confirming their usefulness in for the early detection of fungal spoilage in fruits.

4. Conclusions

This study analyses the efficacy of the E-nose in the quality control of “Kinolea” nectarines during storage. The results show a strong correlation between the signals of some sensors and changes in the aromatic profile of the fruit, mainly those associated with spoilage caused by M. laxa. The LDA model made it possible to classify healthy nectarines and those contaminated with M. laxa with an accuracy of 97% and 96% for healthy nectarines, nectarines with incipient contamination (M_1S + M_2S), and nectarines in their last stage of storage (M_3S).

Furthermore, it could be implemented at the industrial level, which would require external validation to obtain clear and easy-to-interpret results for all stakeholders. In this study, Linear Discriminant Analysis (LDA) models have been established to process the data, but looking to the future, artificial neural networks (ANNs) could be implemented to optimize data processing.