Whole-crop corn (Zea mays L., ZhongBei 410, Wo Da Feng Ltd., Shan Xi, Tai Yuan, China) silage was collected in a standard silage pile on a dairy farm in Hohhot City, Inner Mongolia Autonomous Region, China (111.697° E, 40.485° N). The main parameters of the corn silage were as follows: the harvest date was August 2021; the growth stage of the feedstock was the half-milk line; the cut length was 10-20 mm; the compression density was 210±4 kg dry DM/m³; and the storage period was 90 days. To obtain high-quality silage, each 4000±50 g of silage were manually collected from nine locations along the vertical profile of the pile (3 × 2 m), according to the nine-point sampling method (GB/T 14699.1-2005, 2005). Subsequently, all the silage was thoroughly mixed on a plastic film. Then, 18 high-quality silage samples were equally divided from the above-mixed pile. Of these, each sample weighed 2000±50 g for the subsequent aerobic exposure treatment.

To perform the aerobic exposure treatment, each sample was placed in a polystyrene box (30 × 40 × 30 cm) covered with a sheet of aluminum foil with small holes to avoid moisture loss and ensure sufficient exposure to air (Da Silva et al., 2018). Specifically, each 2-kg sample was placed in the box without compaction (8 cm thick) to ensure that the thickness was sufficient to simulate aerobic exposure. For the aerobic exposure of the silage samples, a total of 18 high-quality silages were divided into nine treatments (two replicates per treatment) and aerobically exposed for different durations. The duration of aerobic exposure for each treatment was 0 days (D0 feed), 0.25 days (D0.25 feed), 0.5 days (D0.5 feed), one day (D1 feed), two days (D2 feed), three days (D3 feed), four days (D4 feed), eight days (D8 feed), and 12 days (D12 feed). After aerobic exposure, each sample was reduced to 60 g based on a quadratic method, of which 50 g was used for sensory quality evaluation and 10 g for measurement of volatile compounds measurement. The quadratic method was carried out as follows: each sample was placed on a clean plastic film and thoroughly mixed into a flat square; the square sample was divided into four sub-samples in diagonal directions, and any pair of orthogonal sub-samples were retained to reduce the weight of the samples; the first and second execution steps were repeated in sequence until 60 g of sub-sample remained. Finally, all samples were immediately frozen at −18 °C.

Prior to testing, the fiber (50/30 μm divinylbenzene/carboxen/polydimethylsiloxane [DVB/CAR/PDMS], Supelco Inc., Bellefonte, PA, USA) was pre-aged at 300 °C for 30 min. Samples were ground into powder in liquid nitrogen. The silage powder (10±0.001 g) was transferred into a 20-mL headspace sample vial, which was sealed with a Teflon/silicone septum. Then, the vials were stabilized at 65 °C for 10 min. After that, the aged fiber was exposed to the headspace of the vial at 65 °C for 10 min. Finally, the fiber was withdrawn from the vial and desorbed at 280 °C for 5 min in splitless mode.

An HP 7890B-7000C GC-MS instrument analyzed the volatile compounds in the fiber with an HP-5MS capillary column (60 m × 0.25 mm, 0.25 μm; Agilent Technologies, California, USA). Helium (He, purity > 99.999%) was used as the carrier gas flowing at 1.2 mL/min. The oven temperature programming was as follows: 40 °C initial temperature for 5 min, increased to 250 °C at 10 °C/min, and then to 280 °C at 5 °C/min for 5 min. The MS conditions were as follows: 230 °C for ion source temperature, EI ionization source, 70 eV for ionization energy, 250 °C for transmission line temperature, 250 °C for interface temperature, and 150 °C for quadrupole temperature. The mass scan range was m/z 35-550.

The compounds were identified by comparing the mass spectral information of samples with the NIST 05 data and matching of the retention indices with the corresponding standards (positive and negative matches > 800; RI). The RI (Eq. 1) of compounds were calculated by n-alkanes (C7-C40; Sigma-Aldrich Co., St. Louis, MO, USA).

wherein n is the number of carbon atoms and tRn < tRy < tR (n + 1); tRy is the adjusted retention time of the test compound y in minutes; tRn is the n-alkane composed of n carbon number, and the retention time is in minutes; and tR (n + 1) is the retention time of a normal structure with a carbon number of n + 1 in minutes.

Moreover, the relative content of volatile compounds was calculated by the percentage of each substance peak area to the total peaks area in GC-MS total ion chromatograms (TIC), which was the peak area normalization method.

The OPLS algorithm was used to confirm the relationship between key compounds and silage odor. First, the relative contents of 63 compounds and the odors of 18 silages were used separately as inputs to construct the OPLS model. Subsequently, a biplot was plotted based on the odor information of the compounds and silage. In the biplot, compounds located near silage odor points illustrated a strong correlation with the corresponding odor.

Origin 2022 (Northampton, MA, USA) was used to plot the temperature of silage and the relative content of volatile compound groups. GraphPad Prism 8 (GraphPad Software Inc., San Diego, CA, USA) was used to plot the changes in the relative content of volatile compounds in silage. Median normalization, log transformation, and autoscaling were applied to preprocess of the generated data before the OPLS-DA and OPLS analysis, which was established using SIMCA 14.1 (Umetrics, Sweden).

There was a significant difference between the relative contents of various functional groups of compounds in silages (Figure 2), with the main families of compounds being acids (54.67-70.87%), esters (6.17-29.58%), and phenols (3.19-9.26%), while terpenes (0.99-6.35%), alkanes (1.21-4.85%), and heterocyclics (0.34-0.88%) had lower contents. In addition, a wide variety was found in the relative contents of chemical families between the fresh- and deteriorated-grade silages. The relative content of esters ranged from 20.99 to 28.86% in fresh-grade silages, while their content ranged from 6.08 to 20.99% in deteriorated-grade silages. Deteriorated-grade silages had higher relative contents of acids (64.42 to 70.17%), phenols (3.41 to 9.60), terpenoids (1.54 to 6.35%), and alkanes (1.80 to 4.85%) than fresh-grade silages (54.35 to 60.73% for acids, 2.57 to 3.47% for phenols, 0.99 to 1.43% for terpenoids, and 1.21 to 1.65% for alkanes). In particular, the contents of n-hexylferulic acid, palmitic acid, linoleic acid, (Z)-oleic acid, linolenic acid, ethyl palmitate, ethyl linoleate, (Z)-ethyl oleate, and ethyl linolenate were higher (above 3% in individual silages).

A total of 63 volatile compounds were detected in all silages (Figure 3), including 19 acids, 17 esters, 11 phenols, four terpenes, three alkanes, three heterocycles, three aldehydes, two alcohols, and one amide. A total of 61 and 52 compounds were present in fresh-grade silages and deteriorated-grade silages, respectively. There were 31 mutual compounds in all samples, which might define the odor tone of the silages. The two different grade silages contained unique components. For instance, fresh-grade silages had 11 unique components, including butyric acid, valeric acid, (E)-3-hexenoic acid, heptanoic acid, phenyl acetic acid, 3-phenyl propionic acid, para-vanillic acid, ethyl lactate, ethyl valerate, ethyl acid succinate, and phenethyl alcohol. Ortho-guaiacol and 2,6-dimethoxyphenol were detected only in deteriorated-grade silages. These unique compounds might have increased the odor variance between the two different grade silages.

The OPLS was used to investigate the relationship between key compounds and silage odor (Figure 7). In terms of the performance of the OPLS model, R2X, R2Y, and Q2 were 0.945, 0.958, and 0.724, respectively, which demonstrated acceptable performance. Combined with the results of sensory evaluation, in the OPLS, we defined fruity odor corresponding to D0 and D0.25 feeds, strong sour and spicy odor corresponding to D0.5 and D1 feeds, roasted odor corresponding to D2 feeds, roasted, musty, and ammonia odor corresponding to D3 feeds, strong roasted and musty odor corresponding to D4 feeds, and strong roasted, malodorous, and musty odor corresponding to D8 and D12 feeds (Figure 7). Moreover, we defined a group of odor points that have a tendency to cluster: strong sour and spicy point and fruity point were a group, which were located in the positive direction of the X-axis and negative direction of the Y-axis; strong roasted and musty point, roasted point and roasted, musty and ammonia point were a group, which were located in the positive direction of the X-axis; strong roasted, malodorous, and musty points are grouped in the negative direction of the X- and Y-axes. It was noted that eugenol and ethyl linoleate were positively correlated with fruity and sour and spicy odors, respectively. In addition, guaiacol, 4-ethyl guaiacol, and 2-methoxy-4-vinyl phenol contributed to the roasted, musty, and putrid odor groups. 4-Ethylphenol contributed to both the upper and left groups of odors on the plot.

In this study, the changing pattern of relative contents of some compounds with aerobic exposure was analyzed by previous studies. The relative content of butyric, valeric, and hexanoic acids gradually decreased with the increasing duration of aerobic exposure, which may be due to the depletion of yeast (Zhang et al., 2018). The relative content of long-chain fatty acids (number of carbon atoms > 6), such as palmitic, linoleic, (Z)-oleic, and linolenic, increased with aerobic exposure. This could be a consequence of their weaker volatility (Tong et al., 2007). Esters were important for silage odor characterization (Worku et al., 2021). Ethyl lactate, ethyl valerate, ethyl palmitate, and ethyl linoleate were the ethyl esters that have been identified in silage (Figueiredo et al., 2007; Zhang et al., 2021). The relative content of DL-alpha-tocopherol was increased by aerobic exposure time. This phenomenon was caused by the proliferation of microorganisms during the aerobic exposure period (Liu et al., 2019). Zong et al. (2022) found higher squalene content in spoiled alfalfa silages than in quality forages. Similarly, the relative squalene content of the D8 and D12 feeds was several times higher than that of the D0 feed.

In this study, 4-ethyl phenol, eugenol, and ethyl linoleate were the key compounds that characterized whether the silages deteriorated. We considered the effects of these compounds on 2-grade silage odors as follows: in fresh-grade silage, the odor of 4-ethyl phenol might have been masked, eugenol contributed a spicy odor, and ethyl linoleate contributed a fruity odor; in deteriorated-grade silage, 4-ethyl phenol contributed a roasted odor and eugenol and ethyl linoleate odors might have been masked. Specific reasons were given as follows:

The 4-ethyl phenol had a smoky odor (Figure 3), which was shown to be associated with the roasted odor of silage (Figure 7). The aROAV of 4-ethyl phenol in deteriorated-grade silage was 2.5, and the aROAV of guaiacol (smoke flavor) was 32.9 (Table 3), both having similar odors. Therefore, these two compounds might work together to provide a roasted odor to deteriorated-grade silage. In addition, odors of low odor intensity may be masked by odors of higher odor intensity (Ma et al., 2021). The 4-ethyl phenol had an aROAV of 1.6 in fresh-grade silage at a relatively low level, suggesting that its odor may be masked by compounds with higher aROAV.

Eugenol was spicy (Figure 3) and was shown to be associated with the spicy odor of silage (Figure 7). Eugenol had a higher aROAV in fresh grade silage than most compounds, except for ethyl valerate and linoleic acid (Table 3). This indicated that the odor of eugenol may be revealed. In contrast, the aROAV of 1.3 for eugenol in deteriorated-grade silage was lower, suggesting that its odor may be masked by compounds with high aROAV.

Ethyl linoleate had a fruity odor (Figure 3) and was shown to be associated with the fruity odor of silage (Figure 7). The aROAV for ethyl linoleate in fresh-grade silage was 2.0 and the similar odor of ethyl valerate (fruity) had an aROAV of 5.8 (Table 3). This suggested that ethyl linoleate may synergistically contribute to the fruity odor of fresh-grade silage with ethyl valerate. The relatively low aROAV of 1.1 for ethyl linoleate in deteriorated-grade feeds suggests that its odor may be masked by compounds with higher aROAV.

In similar studies, principal component analysis (PCA) and ROAV were used to analyze key compounds in chicken breast (Bi et al., 2021); the ROAV, PCA, and sensory evaluation were integrated to analyze key compounds in safflower seed oil (Wang et al., 2020); the OPLS-DA combined with ROAV mined 11 key compounds characterizing black tea grades (Su et al., 2022). Although the above studies provided satisfactory results, methods to quantify the intrinsic link between key compounds and the odor of the measured object were scarce. Therefore, this study innovatively utilized OPLS to analyze the relationship between key compounds and silage odor. The results showed that OPLS could effectively mine the intrinsic link between silage odor and key compounds.

However, the limitations of this study were as follows: silage volatiles were not quantified during aerobic exposure; only volatiles associated with silage fruity, spicy, and roasted odors were screened, and additional key compounds were not found to fully characterize silage odors. Therefore, future work could aim to quantitatively determine the volatile compounds in silage with reference to similar studies, and introduce more intelligent algorithms (e.g., the partial least square (PLS)) for in-depth screening of key compounds that characterize silage odor (Su et al., 2022).