Microbiota succession during aerobic stability of maize silage inoculated with Lentilactobacillus buchneri NCIMB 40788 and Lentilactobacillus hilgardii CNCM‐I‐4785

Abstract Aerobic deterioration of silage following feeding out is responsible for the deterioration of its quality. Inoculation of silage with lactic acid bacteria is one strategy to limit these effects. A trial was performed using whole‐plant corn ensiled in bag silo, and forage was inoculated with Lentilactobacillus buchneri NCIMB 40788 (Lactobacillus buchneri) and Lentilactobacillus hilgardii CNCM‐I‐4785 (Lactobacillus hilgardii) or not (Control silage). After 159 days of fermentation, the silos were opened and the silage was sampled at 24‐h intervals during a 10‐day aerobic stability assay to measure pH, the fermentation profile, mycotoxins, and microbial and fungal populations. In inoculated silage, lactic acid concentrations and pH remained stable during the aerobic phase and higher microorganism alpha‐diversity was observed. Treated silage was characterized by a high abundance of Saccharomycetes and maintenance of Lactobacillus throughout the aerobic stability assay. The high aerobic stability of the inoculated silage contrasted with the decrease in lactic acid contents and the increase in pH observed in the Control silage, concomitantly with an increase in lactate‐assimilating yeast (Pichia and Issatchenkia), and in Acetobacter and Paenibacillus OTUs. Remarkably, Penicillium and roquefortine C were detected in this silage by day 8 following exposure to air. Our study highlighted the fact that the use of L. buchneri with L. hilgardii modified the consequences of exposure to air by maintaining higher microbial diversity, avoiding the dominance of a few bacteria, and preventing fungi from having a detrimental effect on silage quality.

both result in material losses. Losses due to the diffusion of oxygen inside the silage have been estimated at 120 g kg −1 of fresh matter in bunker and clamp silos but can exceed 300 g kg −1 for the crop ensiled close to the surface (Wilkinson & Fenlon, 2013). It is estimated that losses of dry matter during aerobic deterioration are more important than the losses occurring in the initial aerobic phase and the main fermentation phase (Wilkinson & Davies, 2013). In addition to the important impact on fresh matter losses, aerobic deterioration can reduce the nutritional quality of the silage due to the catabolism of silage fermentation products, leading to lower feed intake, contamination by mycotoxins, and increased health-related issues for the animals (Gallo et al., 2015;Pitt & Muck, 1993).
Oxygen entering the silage allows microorganisms to use optimal biochemical pathways and dormant cells may be reactivated, leading to major biochemical changes (Wilkinson & Davies, 2013).
Spoiling will start with the degradation of organic acids by yeasts (Pahlow et al., 2003) and occasionally in the case of whole-crop silages, by acetic acid bacteria (AAB) (Dolci et al., 2011). Losses of organic acids due to respiration increase the pH, thereby allowing other microorganisms, including enterobacteria, AAB, and Bacillus to grow, with counts reported to be as high as 1 × 109 CFU × g −1 (Jonsson, 1991). Degradation of more complex compounds follows the oxidation of soluble substrates. The increased biochemical activity linked to the development of all these microorganisms triggers a rise in temperature inside the silage mass, which is the most noticeable sign of the process. Subsequently, the development of less challenging physicochemical conditions following the catabolism of organic acids can trigger the development of molds.
These organisms significantly reduce the nutritional quality of the feed by degrading carbohydrates, fiber, and protein. Synthesis of mycotoxins may occur simultaneously, as recently reported for aflatoxins (Cavallarin et al., 2011;Ferrero, Prencipe, et al., 2019).
Microbial diversity following exposure to air has recently been studied using next-generation sequencing, but several days after the opening of the silos. Testing aerobic deterioration of small grain silages (barley, oat, and triticale) the bacterial and fungal core microbiome composition was reported after 14 days of exposure (Dunière et al., 2017).

Developing strategies to limit aerobic deterioration is important
for optimizing feed storage, livestock performance, and farm profitability. Inoculation of the forage by lactic acid bacteria improves both the fermentation process and aerobic stability (Muck et al., 2018). The selection of highly competitive strains (Carvalho et al., 2020) can improve the production of lactic acid as well as that of other organic acids able to inhibit the proliferation of undesirable fungi (Guimarães et al., 2018). Driehuis et al. (1999) observed that strains of Lentilactobacillus buchneri (Lactobacillus buchneri (Zheng et al., 2020)) were able to degrade lactic acid into acetic acid and 1,2-propanediol (Oude Elferink et al., 2001), which could then be metabolized into propionic acid (Krooneman et al., 2002). Since both acetate and propionate are strong yeast inhibitors (Guaragnella et al., 2010;Lourenco et al., 2011), these modifications positively improve the aerobic stability of silage. More recently, co-inoculation with L. buchneri NCIMB 40788 and Lentilactobacillus hilgardii CNCM-I-4785 (Lactobacillus hilgardii) was reported to increase the stability of different silages Silva et al., 2020).
While microbial dynamics during fermentation was recently characterized in corn silage inoculated with these two microorganisms , little research has been performed to characterize microbial succession and mycotoxin production in inoculated vs uninoculated silages during the feed-out phase (Hu et al., 2018;Liu et al., 2019). Thus, the objective of this study was to investigate the positive impact of inoculation of whole-plant maize silage with L. buchneri NCIMB 40788 and L. hilgardii CNCM-I-4785 on microbial succession following aerobic exposure.

| Corn silage
Whole-plant corn (hybrid Mycogen TMF2Q419) was harvested 95 days after sowing (Chazy, NY, USA: N44° 53.421′ W73° 28.103′) at a mean dry matter content of 383 g kg −1 The forage was harvested with a John Deere 8800 series harvester and chopped to a theoretical fragment length of 12.7 mm using a kernel processor unit (John Deere KernelStar). The corn was harvested in the field following a randomized five-block design to minimize field-based variations.
The forage was transported to the ensiling site for inoculation using a Dohrman DE-25 applicator mounted on the bagging system (Ag-Bag Systems G6170, Wisconsin, USA). The applicator nozzles were mounted on the filling ramp of the bagger and adjusted to deliver 800 ml min −1 of either water (Control) or a mix of

Lentilactobacillus buchneri NCIMB 40788 (Lactobacillus buchneri)
and Lentilactobacillus hilgardii CNCM-I-4785 (Lactobacillus hilgardii) (LB + LH), each set at 2 × 10 5 CFU g −1 of fresh forage. Individual replication (n = 5/treatment) corresponded to two metric tons of forage separated by one ton of uninoculated forage as a spacer. The treated region was marked with spray paint on the outside of the bag. Each replication corresponded to one of the five randomized field-based blocks of the crop, which was used for both treatments. To reduce contamination, LB + LH was inoculated and bagged after the control by turning the applicator on and off in turn to manage the spacers.
Both treatments were completed within 2 h. The two bags were placed next to one another to guarantee similar storage conditions. The bag silos were 3.5 m in width.
The bagged silage was opened after 159 days of fermentation by removing the plastic cover along the complete length of the bags.
Silage from the outer core was removed and silage samples were taken from the central core (1.5 m from the side) of each silage replication using a loader. One set of (250 g) samples was frozen and used for the analysis of the fermentation parameters. The second set of (250 g) samples was used for microbial counts and the determination of the dry matter in a forced-air oven (at 55°C for 48-72 h).
Lastly, the third set of (20 g) samples was frozen at −80°C for DNA extraction and high-throughput sequencing.

| Aerobic stability assay
The aerobic stability assay was performed according to the Honig method (1990) by collecting silage from the central core of the bag silage for all five replications. For each experimental unit, two one-kg layers of silage were loosely packed in a 7-L container and covered with a layer of cheese-cloth. Air-permeable lids were used to cover the container and reduce moisture loss. A temperature probe (TMC6-HD, Onset Computer, Massachusetts, USA) was positioned between each one-kg layer. The temperature probes were connected to data loggers (UX120-006M, Onset Computer, Massachusetts, USA). Ambient temperature was recorded using the same probes inserted in a bucket filled with perlite to buffer rapid temperature changes. All 10 buckets (5 replications × 2 treatments) were incubated in an incubation chamber set at 20°C.

| Experimental design of the aerobic stability succession trial
Additional samples were collected from the central core to perform an aerobic stability succession trial at the same time as the AS assay. For this purpose, 60 experimental units were collected from the central core for replications 1, 2, and 3 of the bag silos in each treatment, according to a design based on 10 experimental units per replication. These additional AS experimental units were prepared using the same technique as described above and were incubated in the same temperature-controlled room. The experimental units were randomly distributed as one per period of AS, equal to day 1,2,3,4,5,6,7,8,9, and 10 of exposure to air, totaling three replications per treatment per day of aerobic exposure (2 treatments × 3 replications × 10 sampling periods = 60). At 24-h intervals during the AS succession trial, about 1 kg of the top layer of silage was removed and two 250 g samples were collected per unit from the middle of the six containers (2 treatments × 3 replications = 6 per 24 h period) for further analysis and the remaining silage was discarded. One set of samples was used for chemical analysis and the second for DNA isolation and high-throughput sequencing.

| Microbial counts
Upon sampling (day 0 to day 10), for the microbial counts, 20 g samples of silage were refrigerated and transported as rapidly as possible to the laboratory for analysis. The samples were suspended with 180 ml of NaCl buffer (0.85%) for two 60-s periods in a Stomacher 400 paddle blender mixer (Seaward, UK). Serial dilutions were then performed using the same NaCl buffer. Lactic acid bacteria were enumerated using De Man-Rogosa-Sharpe agar plates (Oxoid-Thermo Scientific, Hampshire, UK) containing 100 µg L −1 cycloheximide (Sigma-Aldrich, USA). Yeast and molds were enumerated on malt extract agar plates (Oxoid-Thermo Scientific, Hampshire, UK) containing 2 g L −1 of Rose Bengal (Fisher Scientific, USA), streptomycin at 100 µg L −1 , and neomycin at 50 µg L −1 (Sigma-Aldrich, USA).

| Chemical analysis
The nutritional characteristics (ADF, NDF, starch, crude protein, soluble protein, and ethanol soluble carbohydrates) of samples of fresh forage were analyzed at Cumberland Valley Analytical Services (Waynesboro, PA, USA) using wet chemistry methods (https://www. forag elab.com/Resou rces/Lab-Proce dures). Volatile fatty acids (VFA), lactic acid, ethanol, and pH were also analyzed at Cumberland Valley Analytical Services. Briefly, each fermented silage sample was mixed, and a 25 g wet sample was taken and diluted with 200 ml deionized water. After incubation for 2 h in the refrigerator, the sample mixture was then blended for 2 min and filtered through a coarse (20-25 µm particle retention) filter paper. For VFA quantification and ethanol, 3 ml of extract was filtered through a 0.2 µm filter membrane, and a 1.0 µl subsample was injected into a Perkin Elmer AutoSystem gas chromatograph (Model 710, Perkin Elmer, USA) equipped with a Restek column packed with Stabilwax-DA (Restek, USA). For lactic acid quantification, a 1:1 ratio of extract to deionized water was placed in a YSI 2700 Select Biochemistry Analyzer to determine lactic acid. pH was measured with a Mettler DL12 Titrator (Mettler-Toledo, USA) using 0.1 N NaOH to a pH of 6.5.

| DNA isolation
The DNA isolation and purification protocols were adapted from the methodology proposed by Romero et al. (2018) and Zhou et al. The concentration of DNA was measured on a spectrophotometer (Nanodrop Technology, Cambridge, UK) and quality was measured by agarose gel electrophoresis (1% agarose). The concentration of DNA was standardized at 2 ng μL −1 for all samples.

| High-throughput sequencing and bioinformatics analysis
For amplicon sequencing, the libraries were prepared according to the Illumina 16S Metagenomic Sequencing Library Preparation guide (Part # 15044223 Rev. B), except that a Qiagen HotStar MasterMix (Toronto, Ontario, Canada) was used for the first PCR (amplicon PCR) and half the volume of reagents was used for the second PCR (index PCR). This protocol includes a PCR cleanup step that uses AMPure XP beads to purify amplicons from free primers and primer dimers. The template-specific primers were as follows (without the overhang adapter sequence): 515F (5′-GTG CCA GCM GCC GCG GTA A-3′) and 806R (5′-GGA CTA CHV GGG TWT CTA AT-3′) for 16S amplification from the V4 hypervariable region , and with ITS region 1 specific primers ITS1F (5′-CTT GGT CAT TTA GAG GAA GTA A-3′) and 58A2R (5′-CTG CGT TCT TCA TCG AT-3′) for the amplification of fungi (Yergeau et al., 2017). The amplicon PCR reaction was carried out for 30 cycles with annealing temperatures of 55°C for 16S and 45°C for ITS. Diluted Sequencing data were analyzed using AmpliconTagger (Tremblay & Yergeau, 2019). Briefly, raw reads were scanned for sequencing adapters and PhiX spike-in sequences, and the remaining reads were merged using their common overlapping part with FLASH (Magoč & Salzberg, 2011). Primer sequences were removed from merged sequences and the remaining sequences were filtered for quality such that sequences with an average quality (Phred) score of less than 27 or one or more undefined bases, (N) or more than 10 bases with a quality score of less than 15 were discarded. The remaining sequences were clustered at 100% identity and then clustered/denoized at 99% identity (DNACLUST v3, (Ghodsi et al., 2011)). Clusters with abundances lower than three reads were discarded. The remaining clusters were scanned for chimeras with the VSEARCH version of UCHIME de novo and UCHIME reference (Edgar et al., 2011;Rognes et al., 2016) and clustered at 97% (DNACLUST) to form the final clusters/OTUs. A global read count summary is provided in Table A1. Bacterial and fungal OTUs were then assigned a taxonomic lineage with the RDP classifier (Wang et al., 2007) using the AmpliconTagger 16S and ITS training sets, respectively (Tremblay, 2019; https://doi.org/10.5281/zenodo.3560149). The RDP classifier attributes a score (0 to 1) to each taxonomic depth of each OTU.
Each taxonomic depth with a score of >= 0.5 was retained to reconstruct the final lineage. Taxonomic lineages were combined with the cluster abundance matrix obtained above to generate a raw OTU Alpha (observed species) diversity metrics and taxonomic summaries were then computed using the QIIME v1.9.1 software suite (Caporaso et al., 2010;Kuczynski et al., 2011) using the consensus rarefied OTU table and phylogenetic tree (i.e., to build a UniFrac distance matrix). Analysis of the ITS amplicons was performed similarly, but the alignment was performed using the Silva core reference database (Quast et al., 2012).
The 16S and ITS rDNA raw reads from the microbiota analyses have been deposited at the NCBI BioProject repository under study accession number PRJNA595554.
The solutions were first vortexed for 30 s, sonicated at 35°C for 30 min, and finally shaken on a thermomixer (35°C, 1400 rpm) for 30 min. The samples were then centrifuged and 400 μl extracts were removed. The extracts were diluted 1:1 with 20/79/1 (v/v/v) acetonitrile/water/acetic acid prior to LC/MS analysis. The samples were analyzed using a non-targeted, data-independent acquisitionscreening method as described in Renaud et al.  to allow the identification of a multitude of mycotoxins and fungal secondary metabolites by accurate mass, retention time as well as MS/MS spectra. Briefly, the samples were analyzed in a non-quantitative way that allowed the identifiable features of all the ionizable compounds present to be recorded. The resulting data contain high-resolution accurate mass information of all precursors and product ions as well as the retention times. Analytes that were putatively first detected by accurate mass to charge ratio (m/z) (±3 ppm) and retention time were confirmed by matching the product ions m/z (±3 ppm) with standard spectra deposited in the MassBank Europe spectral library (https://massb ank.eu/MassBank).
Semi-quantification was then performed by running a calibration curve containing the detected mycotoxins with a matrix-matched calibration solution.

| Statistical analysis
The effect of the day of exposure to air on the chemical parameters, mycotoxins, and the differences between treatments was analyzed by a two-way ANOVA using R version 3.6.1 (R Core Team, 2020).
The treatments and the day of aerobic exposure were considered as fixed factors. Interactions between treatments and days were tested. The following parameters were individually tested by the model: pH, lactic acid, acetic acid, ethanol, and the ratio of lactic acid to acetic acid. If the interactions were significant, a pairwise comparison was performed using one Kruskal-Wallis rank test per individual sampling period. For the mycotoxins, values below the threshold of detection were replaced by one half of the detection limit. Aerobic stability was tested by performing a Kaplan-Meier analysis using the time needed to reach the 2°C and 3°C thresholds above the ambient temperature. The maximum temperature was also recorded.
Similarly, differences in the abundance of specific OTUs were tested using a Kruskal-Wallis rank test within an individual day. Significant differences were p < 0.05.
According to our experimental design, differentially abundant OTUs (DAO) were assessed with edgeR (v3.19.7) using its GLM approach (Chen et al., 2016) with OTU table raw count matrix as input.
OTU having a logFC (log foldchange) ratio equal to or higher than 1.5 and false discovery rate (FDR) < 0.05 were considered as differentially abundant between each aerobic stability period. The log transformation of the FC (foldchange) and CPM (counts per million) values were used to generate histograms using in-house scripts, including ggplot functions. showed that the parameters tested in the Control silages (pH, lactic acid) did not remain constant with time, whereas in the LB + LH silages, they did not change significantly (Table 2). For acetic acid, the concentration upon opening was 1.01 g kg −1 DM for Control silage and 1.67 g kg −1 DM for LB + LH. During aerobic exposure, they were of 1.67 (p = 0.107 over time) and 0.72 g kg −1 DM (p = 0.873 over time)

| Fermentation parameters
for Control and LB + LH, respectively.
Indeed, the pH was below 4.0 in the two silages upon opening, and significantly lower in Control than in LB + LH silages (−0.1 unit, Table 2). After exposure to air, it followed two trends depending on the treatment. In the control silage, the pH increased from 3.9 to 5.5 (p < 0.001-Kruskal-Wallis over time), while it remained relatively stable, between 3.97 and 4.10, in the LB + LH silage. In  (Table 2).
NDF, ADF, and the concentration of starch were not influenced by aerobic deterioration in either treatment, with p of 0.394, 0.396, and 0.805, respectively, for the length of aerobic exposure. Temperature profiles during AS are presented in Figure 1. Fresh matter losses during AS averaged 127.7 ± 23.2 g, representing 7.9%

| Aerobic stability
of the original weight of the silage in the Control silage while averaging 72.7 ± 5.6 g in LB + LH silage, that is 4.6% of the original weight of the silage (p < 0.001).

| Microbial counts
Upon opening, the counts of LAB, yeast, and molds in the silos were 7.8, 2.7, and 2.3 log 10 g −1 fresh forage (ff), respectively, in the Control silage and 7.4, 3.1, and below the threshold of detection level of 2 log 10 g −1 ff in the LB + LH silage. The difference between treatments was only significant for the count of LAB (p < 0.001).

| The bacterial population
Upon opening, the Shannon diversity index and the number of observed OTUs were not affected by inoculation: 0.97 ± 0.40 and 17.2 ± 5.9 in the Control silage and 1.14 ± 0.23 and 21.0 ± 6.1 in the LB + LH silage (Figure 2c)

| Diversity during the aerobic stability trial
Over the 10 days, both the Shannon index and the number of observed OTUs in LB + LH silage were significantly higher, with a mean of 17.8 compared to 10.1 in the Control silage (p < 0.001). There was no significant effect of the length of aerobic exposure, with p of 0.956 and 0.060 for the Shannon index and the observed OTUs, the interaction was not significant (p above 0.7) in either parameter. In the Control silage, the Shannon index remained between 0.60 and 1.00 throughout the study except on day 10 when it reached 1.39.
In the LB + LH silage, the Shannon index ranged between 1.0 and 1.5 throughout the study, with higher values measured on days 6, 7, and 10.
The bacterial taxonomic profiles in the two treatments evolved differently during the incubation period of the AS assay ( Figures   2 and 3 The increase in the abundance of Acetobacter OTUs was followed by an increase in OTUs related to Paenibacillus, reaching 11.5% by day 10. However, like Acetobacter, the marked variation in abundance between replications did not result in a significant difference between the two treatments. OTUs related to Weissella were observed in the LB + LH silage throughout the incubation period of the AS assay ranging from 3.2% to 6.3%. Although the relative abundance of Weissella and Pediococcus did not differ significantly, it dropped in the control silage while remaining more stable in the LB + LH silage. Contrast analysis revealed no differences between silage treatments up to day 6, with only two Lactobacillus OTUs being significantly more abundant in the Control than in the LB + LH silage ( Figure 3). From day 6 on, a time that corresponded to changes in the microbial composition ( Figure 2) and the first heating period (Figure 1)

| Fungal population
Upon opening, the Shannon diversity index and the number of ob-

| Kinetics during the aerobic stability trial
The Shannon diversity index and the number of observed OTUs did not differ significantly between sampling periods (p = 0.620) but on average was significantly higher in the LB + LH silage over the complete 10-day period (p < 0.001). There was no interaction between time and treatment ( Figure 2c).
During aerobic exposure, OTUs belonging to Saccharomyces, Issatchenkia, Kazachstania were dominant in the Control silage until day 6. From then on, fewer Kazachstania were detected, whereas the relative abundance of OTUs related to Penicillium increased ( Figure 2b).
Interestingly, in the LB + LH silage, the relative abundance of Saccharomyces increased with time to reach more than 80% on day 10 and with less variability between replicates, while the abundance of OTUs related to Issatchenkia was high from 48 h to 72 h, with 31.5% and 20.0% of the relative abundance, but from then on was never higher than 2.1% ( Figure 2b). Also, it is interesting to note that several taxa were present at low relative abundances on days 1, 2, 3, notably Pleosporales, Eurotiales, Tremellales-related OTUs.

Contrast analysis revealed that yeasts belonging to Malassezia and
Filobasidium genera, the ubiquitous mold Cladosporium, and fungi belonging to the orders Pleosporales, Eurotiales, Cystobasidiales, and Tremellales were significantly more abundant at days 1, 2, and 3 after F I G U R E 3 Contrast analysis of the 16S amplicon sequencing data of the samples collected at 24-h intervals during incubation in the aerobic stability assay analyzed using the EdgeR package. Only OTUs whose abundance differed significantly between the Control and the LB + LH inoculation treatment are shown. No significant difference was observed at days 3, 4, and 5, so these days are not included in On day 5, the abundance of Penicillium OTUs was only 0.3% but increased to 7.6% in the following 24 h (p = 0.164). Abundance reached 39.2% by day 7 and then decreased to around 27%-28%. The mean daily abundance of the Penicillium OTUs was never higher than 0.9% in the LB + LH silage.
The abundance of the most abundant yeast-related OTUs followed different trends depending on the treatment ( Figure 5).

The control silage was mainly characterized by the emergence of
Issatchenkia in the first few days following exposure to air and a gradual decrease in Kazachstania was also observed. Also in the control silage, OTUs related to Saccharomyces increased slightly on the first day of exposure to air and remained stable thereafter. Conversely, a F I G U R E 4 Contrast analysis of ITS amplicon sequencing data of the samples collected at 24-h intervals during incubation in the aerobic stability assay analyzed using the EdgeR package. Only OTUs whose abundance differed significantly between the Control and the LB + LH inoculation treatments are shown. LogFC (fold changes) and logCPM (count per million) are used for the representation. Only OTUs with a logFC ≤ 1.5 (red line in each panel) and false discovery rate (FDR) value <0.05 were considered as being differentially abundant. Color coding is based on the taxonomic classification of individual OTUs.

| Mycotoxins
The samples were screened for 84 mycotoxins and related fungal metabolites (Table A2) among which 13 were identified, including metabolites produced by Penicillium, Fusarium, and Alternaria (Table   A3).

| DISCUSS ION
Silage exposed to air is prone to deterioration as bacteria, yeasts, and molds can be activated and oxidize different substrates, thereby affecting silage quality. The aerobic stability of silage is thus a key factor to ensure that silage provides well-preserved nutrients for animals containing minimum levels of spores and toxins. Further development of strategies to limit problems related to aerobic deterioration is thus needed. Adding microbial inoculants during ensiling and opening periods is one possible strategy. Microbial silage additives are usually classified into two groups, homofermentative and heterofermentative. Homofermentative LAB produces lactic acid (McDonald et al., 1991), whereas heterofermentative LAB not only produce lactic acid, but also complex organic acids such as acetic acid and propionic acid (McDonald et al., 1991;Oude Elferink et al., 2001) that are strong inhibitors of yeast growth. To give an example, the heterofermentative LAB L. buchneri in corn silage improved AS (Dolci et al., 2011;, probably through a synthesis of 1,2-propanediol and subsequent conversion to propionic acid (Oude Elferink et al., 2001). These specific properties of L. buchneri can be improved by co-inoculation of a second heterofermentative LAB, L. hilgardii (Reis et al., 2018). The positive impact of this co-inoculation was tested in the present study as inoculation increased the time required to reach the threshold of 2°C over ambient temperature by 77.6 h. Knowing that we analyzed the impact of the inoculation of the combination on microbial and biochemical parameters for 10 days of exposure to air.
Upon opening after 159 days of ensiling, the pH of the two types of silage was still below 4 and Lactobacillus related OTUs dominated, in agreement with results in the literature (Gharechahi et al., 2017;Guan et al., 2018), and reflecting the conservation process was appropriate in both cases. A recent study (with a 64-day incubation period) using the same two lactic acid bacteria as inoculants showed F I G U R E 6 Histograms of the changes in concentrations of the mycotoxins Roquefortine C, Fumonisin B1, Deoxynivalenol, and Beauvericin every 24 hours after the start of air exposure and differences between Coinntrol and LB + LH inoculation treatments.
that the microbiota of the inoculated corn silage was characterized by a higher abundance of Lactobacillus than the non-inoculated corn silage and hosted a distinctive yeast population with a lower alpha-diversity of the fungal related OTUs . This was not confirmed in the current study, where we observed a slightly lower relative abundance of Lactobacillus and a higher abundance of Weissella in the LB + LH silage. However, the length of the fermentation period and incubation temperature differed in the two studies, and in the former study, the bag silos were stored outside.
Considering the lower average temperatures encountered during the incubation of the bag silos, we would expect microbial succession to be slower than in mini-silos stored at room temperature (Zhou et al., 2019). The presence of Leuconostocaceae at this late stage of ensiling is rare in corn silage, considering that upon the opening of the bag silos, pH was below 4.0, a pH to which species belonging to this family are sensitive. This level of abundance of Weissella sp. has sometimes been reported but in non-inoculated silage (Keshri et al., 2018). The fungal population was mainly composed of the yeasts

Saccharomyces and Kazachstania, but with a higher abundance of
Bionectriaceae in some samples of LB + LH silages. Bionectriaceae has rarely been reported in corn silage but has been observed in sorghum before ensiling (Gonda et al., 2019). Interestingly, in this family, the Clonostachys related-OTU is an epiphytic mycoparasitic fungus that can be used for biocontrol of Fusarium graminearum (Gimeno et al., 2019), and also has the enzymatic potential to degrade zearalenone (Ogunade et al., 2018). The heterogeneity of the fungal population in the LB + LH samples may be due to the number of fungal counts below the threshold of detection (<log 10 of 2.0) for the molds. Overall, data upon opening indicated that both silages were well stored and presented a typical core microbiome.
During exposure to air, the temperature pattern of the Control silage was typical, with a two-phase profile in the temperature curve, as also reported by Merry and Davies (1999). These authors described the first peak corresponding to an increase in yeast and AAB, and the second peak corresponding to the growth of molds. In our study, the change in metabolites and microbiota corresponded to a sharp drop in lactic acid, dominance of Saccharomycetales and more specifically an increase in OTUs linked to Issatchenkia in the first days after opening. The growth of Issatchankia was followed by an increase in the relative abundance of Pichia, Acetobacter, Bacillus, and Paenibacillus. It should be noted that the phenotypic classification of yeast species is often difficult due to switching between the teleomorph and anamorphic states (Kurtzman & Robnett, 2003).
This may have introduced classification errors in the ITS database as already reported for Candida krusei, Issatchenkia orientalis, Candida glycerinogenes, and Pichia kudriavzevii that were shown to belong to the same species (Douglass et al., 2018), potentially explaining why the number of observed OTUs related to the Issatchenkia genus was high in our study. In all cases, all the microorganisms we detected-bacterial and fungal-can use lactic acid, as supported by the decrease in lactic acid concentrations along with an increase in pH. More specifically, the growth of Acetobacter could occur when yeasts are outnumbered or impaired by the accumulation of propionic acid (Dolci et al., 2011). This genus can convert ethanol into acetic acid in the presence of oxygen (Mamlouk & Gullo, 2013).
It can also oxidize lactate and acetate into carbon dioxide and water (Pahlow et al., 2003). The presence of ethanol was observed since its concentration decreased more than 20 times from the highest value measured on day 0 to its lowest value after 10 days in the control, while it was reduced only by half in LB + LH. More than half the ethanol measured on day 0 was detected around day 5 and 6 when the peak temperature was reached. These modifications were followed by the growth of Penicillium-related OTUs. The observed changes in pH, lactic acid, acetic acid, and ethanol were thus fully explained by the succession of different bacterial and fungal OTUs (Figure 1).
Few studies testing the microbial dynamic following aerobic exposure of silage have been published. When testing wholeplant corn silage, Hu et al. (2018) showed that the abundance of Sporolactobacillus increased sharply after 72 h of exposure to air, but fungal diversity was not tested in their study. In an assay performed on small grain silage, Dunière et al. (2017) observed that Bacillales and Lactobacillales were part of the core microbiome after 14 days of exposure to air. Also, Saccharomycetales constituted 70% of the fungal population, and most OTUs were assigned to Kazhachstania and Pichia. These observations confirm the importance of lactate-assimilating yeast during aerobic deterioration. Still, the correlation between yeast counts after fermentation and aerobic stability of different silages is not yet clear. The specific genera of yeast could be involved, but changes in the bacterial population could also act as a precursor of shifts in the yeast population and numbers. Among bacteria, Acetobacter seems to be of interest. Indeed, Dolci et al. (2011) demonstrated that Acetobacter was present after five days of exposure in corn silage under polyethylene treatment, and Kazachstania (day 7) was observed before Pichia (day 9). Similarly, high abundances of Acetobacter and Lactobacillus have been reported in whole-plant maize silage exposed to oxygen for seven days (Li & Nishino, 2013).
The same authors noticed that the presence of Acetobacter was specific to corn silage and that the decrease in Acetobacter was concomitant with an increase in the abundance of Pediococcus.
Interestingly, we also observed this relationship between the presence of Pediococcus and a lower abundance of Acetobacter in treated silage. Overall, our results on microbiota are in agreement with those obtained in previous studies in which non-inoculated silage was exposed to oxygen. However, inoculated and non-inoculated silages were not compared in any of the above-mentioned studies.
Inoculation with L. buchneri and L. hilgardii significantly improved the AS of bag silage, a result consistent with published data Nair et al., 2019). The temperature started to increase only on day 6 after opening without the two successive peaks observed in the Control silage. Interestingly, the lactic acid concentration remained stable with time, resulting in higher stability after silage opening, as suggested by the pH ranged between 3.97 and 4.10 during exposure to air. Similarly, the concentrations of acetic acid (4.4-9.9 g kg −1 DM) and ethanol (3.4-3.8 g kg −1 DM) remained in the same range during the 10 days of exposure to air. As a result, the lactic:total VFA ratio did not decrease in the LB + LH silage, whereas it decreased dramatically to reach 0.3 in the Control silage. In addition to these observations on the fermentation parameters, differences in microbiota can be summarized as a higher diversity of OTUs for both the bacterial and fungal microbiota, a more dynamic population of LAB, and a non-lactate consuming yeast population (Figures 2-5).
Indeed, the abundance of Lactobacillus was more than 78% in the first eight days and was still 57% on day 10. This higher abundance of Lactobacillus might have contributed to the biocontrol of undesirable microorganisms (Castellano et al., 2017;Li et al., 2015).
Remarkably, the presence of Acetobacter was observed in only one of the three replications using the LB + LH silage and was accom-  (Kasmaei et al., 2017). This trend enabled a higher proportion of Saccharomyces-related OTUs during the AS assay (Dunière et al., 2015). Overall, silage inoculation allowed a switch from Pichia-Issatchenkia to Saccharomyces-related yeast OTUs.
All observed changes in the fermentation parameters were fully explained by the succession of the different bacterial and fungal OTUs. This study is the first to highlight how silage inoculation with L. buchneri and L. hilgardii improves aerobic stability thanks to maintaining higher microbial diversity and LAB fitness upon opening, avoiding the domination of a few yeasts and opportunistic facultative anaerobe bacteria that can use lactic acid (Figure 1).
Mycotoxins are secondary metabolites secreted by fungal organisms mostly belonging to the genera Aspergillus, Fusarium, Alternaria, and Penicillium. They have toxic effects on animals and may reduce feed intake and hence performance while inducing inflammatory status. Mycotoxins are not widely studied in silage, even though silage represents a major part of the diet of ruminants. The total amount of mycotoxins ingested by cows may be greater than the maximum concentrations allowed or recommended in ruminant diets by the US FDA or the EU. Different strategies to limit the production of mycotoxins in silage during the field, pre-harvest, harvest, and ensiling phases exist and were recently reviewed (Ogunade et al., 2018). These strategies included the use of microbial inoculants to insure good ensiling conditions and to improve aerobic stability. One objective of our study was to investigate if the change in bacterial and fungal populations thanks to silage inoculation could influence mycotoxin content during the opening phase. Upon opening, three mycotoxins were detected, namely, fumonisin, beauvericin, and zearalenone (ZEA). Their presence at low levels confirmed good ensiling conditions.
During the opening period, concentrations of FB1 fluctuated with time but no particular trend was identified, while FB 2 /B 3 concentrations remained low in all samples. The concentration of ZEA, mainly produced by F. graminearum in corn remained constant over the 10-day AS period and below the European Commission guidance level for ruminant feed, which is 0.5 mg kg −1 (2006/576/EC). The levels of beauvericin, a toxin produced by Fusarium and considered as an emerging mycotoxin, were similar to or higher than the reported maximum concentration (Reisinger et al., 2019). All these Fusarium metabolites are considered to be "pre-harvest" metabolites and are not produced during the ensilage process or the feed-out stage.
DON mycotoxin was not detected upon opening but appeared during the AS trial, particularly in the Control silage. DON is also a "pre-harvest" mycotoxin that is produced mainly by F. graminearum in corn, which also biosynthesizes ZEA. An interesting observation was the increase in DON in contrast to the relatively constant concentration of ZEA, and to the low abundance of Fusarium related OTUs. Contamination by Fusarium may increase later in the aerobic deterioration process, as observed previously (Borreani et al., 2005).
It is not clear whether the observed increase in the concentration of DON is the result of fungal activity occurring within the ensiled material or is rather the result of the conversion of DON-conjugates and glycosides into DON in plantae (Castellano et al., 2017). The presence of DON glycosides in the samples suggests that the in plantae conversion metabolism is active. Unlike Fusarium metabolites that are generally considered to be "pre-harvest" in origin, roquefortine C is produced by Penicillium and is considered a mostly "post-harvest" mycotoxin (Sumarah et al., 2005). We suggest that LB + LH favors a microorganism with mycotoxin-degrading ability. In the same way, DON and roquefortine C were almost not detected in the LB + LH silages during the current AS trial.
Only a few studies have tested the effects of microbial inoculants on mycotoxin degradation, especially during exposure to air. Even if the conclusions differ from one study to another one, it appears that the use of microbial inoculants can reduce concentrations of mycotoxins or prevent the growth of toxigenic fungi (Kung et al., 2003). This study is the first to suggest that inoculating silage with LB + LH could improve the safety of silage following exposure to air. Further work is now required to understand these mechanisms.

| CON CLUS ION
To conclude, our study confirmed that exposure to air increases results in a switch in both the composition and in the activity of fungal and bacterial communities. It also highlighted the fact that the use of L. buchneri in combination with L. hilgardii modified the consequences of exposure to air by maintaining higher microbial diversity after the first 24 h of opening, avoiding the dominance of a few bacteria and fungi that could be detrimental to the silage quality. The combination of LB and LH could thus be used not only to ensure good conditions during ensiling but also to reduce microbial spoilage during exposure to air. We expect that performing a similar trial at an earlier stage of fermentation would result in even greater differences in microbial diversity in comparison with untreated silage. It will be interesting to go further in our research to integrate all data from harvest to feed-out, and, ultimately, to test the impact of the inoculants on the nutritional quality of silage and the zootechnical performance of cattle. It would also be interesting to study the relationships between epiphytic microbiota and the production of mycotoxins in the field and at harvest, the kinetic evolution of these parameters during ensiling, and their impact during the exposure to air, with or without inoculants.

ACK N OWLED G M ENTS
We thank the staff of the William H. Miner Institute for their assistance in harvesting and preparation of the bag silage, and the use of some laboratory equipment. We also express our thanks to Éric Chevaux and Mathieu Castex for their critical review of the manuscript. We thank Compute Canada for access to the Graham High-Performance Computing cluster.

E TH I C S S TATEM ENT
None required.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data are provided in full in the paper except for the 16S and ITS rDNA raw reads from the microbiota analyses that have been de-