To describe, at high resolution, the bacterial population dynamics and chemical transformations during the ensiling of alfalfa and subsequent exposure to air.
To describe, at high resolution, the bacterial population dynamics and chemical transformations during the ensiling of alfalfa and subsequent exposure to air.
Samples of alfalfa, ensiled alfalfa and silage exposed to air were collected and their bacterial population structures compared using 16S rRNA gene libraries containing approximately 1900 sequences each. Cultural and chemical analyses were also performed to complement the 16S gene sequence data. Sequence analysis revealed significant differences (P < 0·05) in the bacterial populations at each time point. The alfalfa-derived library contained mostly sequences associated with the Gammaproteobacteria (including the genera: Enterobacter, Erwinia and Pantoea); the ensiled material contained mostly sequences associated with the lactic acid bacteria (LAB) (including the genera: Lactobacillus, Pediococcus and Lactococcus). Exposure to air resulted in even greater percentages of LAB, especially among the genus Lactobacillus, and a significant drop in bacterial diversity.
In-depth 16S rRNA gene sequence analysis revealed significant bacterial population structure changes during ensiling and again during exposure to air.
This in-depth description of the bacterial population dynamics that occurred during ensiling and simulated feed out expands our knowledge of these processes.
Ensiling is a method of preserving fresh forage crops, such as alfalfa, by fermentation for future use as a ruminant animal feed (McDonald et al. 1991; Muck 1991). The process can be divided into four stages: first, the plant material is chopped, wilted to 50–67% moisture and packed into airtight silos or sealed in plastic membranes (Pitt 1990). During this period, the plant material is exposed to atmospheric oxygen and the plant cells continue to respire and produce enzymes that can break down proteins, hemicellulose and sugars, reducing the nutritional value of the silage (Pitt et al. 1985). During this stage, aerobic fungi and spoilage bacteria are present and may produce undesirable by-products, and thus, it is essential to complete this stage as quickly as possible (Woolford 1990). In the second stage, the ensiled material becomes anaerobic due to plant and microbial respiration (Pitt et al. 1985; Woolford 1990). Once anaerobic conditions are obtained, the plant cell membranes lyse, releasing liquids and sugars that are fermented to lactic acid by the autochthonous lactic acid bacteria (LAB), lowering the pH of the ensiled material. Along with lactic acid, lesser amounts of acetic and formic acids, as well as alcohols, are produced (Zaunmuller et al. 2006). The acidic and anaerobic conditions generated during this stage prevent the growth of most fungi and spoilage bacteria, preserving the ensiled material during the third or storage stage (Muck and Pitt 1994). In the last stage, the silos or plastic membranes are opened, and portions of the silage are removed as needed to feed ruminant animals. During this stage, the silage is exposed to atmospheric oxygen, allowing fungi and aerobic spoilage bacteria that are present, but dormant to grow and produce undesirable products such as mycotoxins (Borreani et al. 2005; Garon et al. 2006), consume digestible dry matter (Bolsen et al. 1993) and degrade the nutritional value of the silage (Woolford 1990; Ashbell and Weinberg 1992).
Ensiling has been practised in Europe since the beginning of the 19th century and was described by the French agriculturist Auguste Goffart in 1877 and later translated into English by Brown (Goffart and Brown 1880). The microbial population dynamics that occur during ensiling have been described previously; however, the vast majority of these investigations utilized culture-based techniques (Kempton and San Clemente 1959; Langston and Bouma 1960; Lin et al. 1992), that while informative, are notorious for underestimating the level of bacterial diversity (Hugenholtz et al. 1998; Ercolini 2004). Recently, molecular techniques such as denaturing gradient gel electrophoresis (DGGE) (Parvin and Nishino 2009; Dolci et al. 2011), terminal restriction fragment length polymorphism (TRFLP) (McEniry et al. 2009) and random amplified polymorphic DNA (RAPD) (Rossi and Dellaglio 2007) have been used to study specific aspects of the ensiling process. These studies have enhanced our understanding of the microbial populations in silage. However, they only identified a few of the most abundant operational taxonomic units (OTU) present, because of their poor limits of detection (Dewettinck et al. 2001; Ogier et al. 2002; Fasoli et al. 2003; Temmerman et al. 2003). In this report, we describe the bacterial population dynamics during the ensiling of alfalfa by comparing 16S rRNA gene libraries generated from wilted alfalfa, ensiled alfalfa and ensiled alfalfa exposed to air containing ca. 1900 sequences each. We also provide data on the cultural, chemical and physical characteristics at each time point.
Whole-plant alfalfa was harvested at the 10% bloom stage, at the third cut, from fields located at the University of California, Davis, over a 5-days period in July 2011. The alfalfa was cut into 3- to 6-cm lengths with a flail chopper and dried in windrows measuring approximately 3 m wide and 10 cm height. The windrows sat atop polyethylene sheets, outside in direct sunlight, at a temperature of approximately 30°C for 5–6 h, until it reached a dry matter content of approximately 35%. After this point, a sample was taken for chemical and microbiological analysis (BE). Seven kilograms of alfalfa was compacted at 390 kg m−3 wet basis (136 kg m−3 dry basis) in five identical mini-silos constructed from 20·8-l, high-density polyethylene buckets with airtight lids (Pleasant Hill Grains, Hampton, NE, USA) that contained pressure sensors (Ankom Gas Production System, Ankom Technology, Macedon, NY, USA) set to release gas at 1 KPa. The 20·8-l mini-silos were filled to approximately the 18-l mark, allowing a 2·8-l headspace measuring approximately 5 cm from the silage to the lid. The mini-silos were incubated at 25°C for 40 days, after which, the lids were opened and the silage was weighed and sampled for chemical and microbiological analysis (PE). The mini-silos were resealed, and for the next 20 days, 1 l of ultra-pure carrier grade air (Airgas Specialty Gases, Radnor, PA, USA) was injected into them via the injection port in the pressure sensors (Ankom Gas Production System, Ankom Technology) 3× daily. On day 60, the mini-silos were opened, the silage was weighed, and samples were taken for chemical and microbial analysis (PO).
Ten grams of wilted alfalfa, alfalfa silage or silage exposed to air was mixed with 90 ml of PBS + 0·01% Tween-80 and blended using a Waring commercial blender (Waring Products, Torrington, CT, USA) at high speed for one min. Ten-fold serial dilutions were performed in PBS and plated onto nutrient agar (NA) plates (Becton, Dickenson and Company, Sparks, MD, USA), incubated at 25°C and observed daily for 4 days. All chemical analyses were performed at Cumberland Valley Analytical Services (Maugansville, MD, USA) using the standard methods of the National Forage Testing Association available online at: http://www.foragelab.com/Lab-Services/Forage-and-Feed/Lab-Procedures/.
To extract DNA from the wilted alfalfa, alfalfa silage or silage exposed to air, 100 g of each sample was added to 300 ml of PBS + 0·01% Tween-80 in a sterile 1-l Erlenmeyer flask and shaken at 250 rev min−1 for 1 h at 25°C. The liquid portion was removed and centrifuged at 10 000 g for 10 min, and the resulting pellet was suspended in 1 ml of lysis buffer (500 mmol l−1 NaCl, 50 mmol l−1 Tris–HCl, pH 8·0, 50 mmol l−1 EDTA and 4% sodium dodecyl sulfate), transferred to a 2·5-ml microcentrifuge tube with 0·4 g of sterile 0·1 mm zirconia/silica beads, homogenized for 3 min on a Mini-Beadbeater (BioSpec Products, Bartlesville, OK, USA) at maximum speed, incubated at 70°C for 15 min and centrifuged at 16 000 g for 5 min. The supernatant was reserved and 300 μl of lysis buffer added to the pellet, the process repeated and the supernatants pooled. Two hundred sixty microlitres of 10 mol l−1 ammonium acetate was added to the supernatants, incubated at 4°C for 5 min and centrifuged at 4°C for 10 min at 16 000 g; the supernatant transferred to two 1·5-ml tubes with 1 volume of 4°C isopropanol, incubated at 4°C for 30 min and centrifuged at 16 000 g for 15 min at 4°C. The supernatant was discarded and the pellet washed with 70% ethanol, dried under vacuum and suspended in 100 μl of Tris–EDTA (pH 8·0). Two microlitres of DNase-free RNase (10 mg ml−1) was added, and the mixture was incubated at 37°C for 15 min, and 15 μl of proteinase K (600 mAU ml−1) and 200 μl of AL buffer from the QIAamp DNA Stool Mini kit (Qiagen, Valencia, CA, USA) were added and incubated at 70°C for 10 min. The mixture was cooled on ice, 200 μl of ethanol was added, and the DNA was purified using the QIAamp DNA Mini kit (Qiagen) as per the manufacturer's instructions.
PCR amplification of 16S rRNA genes was carried out using the primers 27f (5′ AGAGTTTGATCCTGGCTCAG 3′) and 1392r (5′ GACGGGCGGTGTGTAC 3′) (Lane 1991). PCR were performed as recommended by Polz and Cavanaugh (1998) to reduce bias in amplification. Briefly, 50 μl reaction volumes contained 25 μl High-Fidelity PCR Master Mix (Roche, Nutley, NJ, USA), 50 ng DNA and 1 μmol l−1 of each primer. PCR were performed in a Tetrad Thermocycler (Bio-Rad, Hercules, CA, USA) under the following conditions: one cycle of 95°C for 5 min, 20 cycles of 94°C for 30 s, 55°C for 30 s, 72°C for 1·5 min, and one cycle of 10 min at 72°C. PCR products were purified using the Zymo DNA Clean and Concentrator kit (Zymo Research, Orange, CA, USA), cloned using the TOPO TA Cloning kit (Invitrogen, Carlsbad, CA, USA) as per the manufacturer's instructions and transformed into E. coli TOP10 competent cells (Invitrogen). Clones were grown on LB agar (Fisher Scientific, Fair Lawn, NJ, USA) plates containing kanamycin (Km) (50 μg ml−1) and 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside (X-gal) (80 μg ml−1) at 37°C for 18 h. White colonies were selected and grown in 96-well plates in LB broth (Fisher Scientific) supplemented with Km. For each sample type, four 96-well plates were picked from each of 5 replicates for a total of 1920 clones.
DNA templates were prepared from 0·2 μl overnight cultures using the TempliPhi HT Amplification kit (GE Healthcare, Piscataway, NJ, USA) as per the manufacturer's instructions. Sequencing reactions were performed using the primer 1392r and the BigDye Terminator v3·1 Cycle Sequencing kit (Applied Biosystems, Foster City, CA, USA). Sequencing reactions were purified using the BigDye XTerminator Purification kit (Applied Biosystems); electrophoresis and readout were performed using an Applied Biosystems 3730XL Genetic Analyzer (Applied Biosystems). For each sample type, 4 96-well plates of clones were sequenced from each of 5 replicates for a total of 1920 sequences.
DNA sequences were edited manually to correct falsely called bases and trimmed at both the 5′ and 3′ ends using the SeqMan software (DNASTAR Inc., Madison, WI, USA) and analysed by Chimera Check to ensure they were not artefacts of amplification. Only sequences with unambiguous reads of 650 bp starting at position 1367 and terminating at 717 (E. coli numbering) covering hypervariable regions V5-V8 were used. The DNA sequences were grouped into OTU defined as sequences with >97% sequence identity using the FastGroup program (Seguritan and Rohwer 2001). Representatives of each OTU were analysed using the megaBLAST program and the NCBI nr/nt database (available at http://blast.ncbi.nlm.nih.gov/). DNA sequences representative of the 10 most prevalent OTU from each library were deposited in GenBank under the accession numbers JQ970489-JQ970518.
For dendrogram construction, 30 partial 16S rDNA sequences representing the 10 most prevalent OTU from each sample type were aligned using MUSCLE (Edgar 2004). Also included in this alignment were the most similar 16S rRNA gene sequences to each OTU from the NCBI nr/nt database. These 16S rRNA gene sequences were reverse complemented and trimmed to the approximate start point and length of the OTU sequences. Phylogenetic and molecular evolutionary analyses were performed using MEGA5 (Tamura et al. 2011). The dendrogram was constructed using the neighbour-joining algorithm (Saitou and Nei 1987) and the maximum composite likelihood distance estimation method (Felsenstein 1985).
Rarefaction analysis was performed using the approximation algorithm of Hurlbert (1971) with 95% confidence intervals estimated as described by Heck et al. (1975) using the freeware program aRarefactWin by Holland (1998). The percentage of the total OTU identified in each sample was calculated using the equation C = 1−(n/N) × 100, where C is the percentage coverage, n is the number of singleton OTU and N is the number of clones examined. Shannon entropy and evenness were calculated using the FastGroupII software (Yu et al. 2006).
Each OTU was assigned to a phylum using the Classifier software (Wang et al. 2007), which phylogenetically classifies an OTU sequence using a naïve Bayesian rRNA classifier trained on the known type strain 16S sequences. Once the OTU were classified, pairwise comparisons of the OTU within the libraries were performed using the Library Compare software (Wang et al. 2007), which estimates the likelihood that the frequency of membership in a given OTU is the same for the two libraries using the equation:
where N1 and N2 are the total number of sequences for library 1 and 2, respectively, and x and y are the number of sequences assigned to an OTU from library 1 and 2, respectively. Statistical significance was set at P < 0·05. The aerobic plate count data were statistically analysed using the Student's t-test, and statistical significance was set at P < 0·05. The chemical and physical data were statistically analysed using the Proc GLM procedures (SAS version 9.3; SAS Institute, Cary, NC, USA) using repeated measures with variance components on day 0, 40 and 60 with replication treated as a random variable. Statistical significance for the Tukey's test was also set at P < 0·05.
Cultural analysis revealed that the number of bacteria capable of growing on nutrient agar plates increased significantly (P < 0·05), from 1·0 × 107 to 2·1 × 108 CFU g−1 after 40 days of ensiling (Table 1). Exposure to air yielded a slight, but insignificant increase in these populations. Chemical and physical analysis revealed that during ensiling, significant reductions (P < 0·05) in weight, dry matter, sugar, starch, nitrate and pH occurred (Table 1). Concomitantly, significant increases (P < 0·05) in crude protein, soluble protein, crude fat, ammonia, total acids, lactate, acetate and the ratio of lactate to acetate were observed. After exposure to air, significant reductions (P < 0·05) in total weight, crude protein, sugar, starch, crude fat, pH and lactate occurred. Exposure to air also resulted in significant increases (P < 0·05) in dry matter and soluble protein.
|Total weight (kg)||7·00 (0·00)||6·92 (0·14)a||6·24 (0·03)a|
|Dry matter (g kg−1)||349·6 (79·36)||323·7 (37·27)a||363·2 (62·97)a|
|Crude protein (g kgDM−1)||210·6 (15·20)||221·6 (9·15)a||214·0 (9·13)a|
|Soluble protein (g kgCP−1)||402·6 (18·74)||686·6 (39·96)a||692·5 (28·23)a|
|Ammonia (g kgDM−1)||5·60 (0·56)||28·6 (4·78)a||27·8 (7·23)|
|Glucose (g kgDM−1)||70·6 (6·26)||37·2 (7·72)a||36·4 (6·06)a|
|Starch (g kgDM−1)||29·2 (2·77)||27·6 (7·22)a||24·0 (2·79)a|
|Crude fat (g kgDM−1)||22·8 (1·92)||34·0 (3·33)a||32·0 (2·00)a|
|Nitrate (g kgDM−1)||1·7 (0·40)||0·6 (0·50)a||0·8 (0·32)|
|pH||5·8 (0·08)||5·1 (0·13)a||5·0 (0·09)a|
|Total acids (g kgDM−1)||6·5 (4·07)||83·4 (5·54)a||71·0 (15·78)|
|Lactate (g kgDM−1)||1·8 (2·16)||51·0 (14·33)a||41·4 (12·01)a|
|Acetate (g kgDM−1)||4·7 (2·05)||24·2 (5·83)a||26·6 (11·99)|
|Propionate (g kgDM−1)||0·1 (0·09)||1·1 (2·54)||0·9 (1·92)|
|Butyrate (g kgDM−1)||0·0 (0·00)||6·3 (14·04)||0·0 (0·00)|
|Lactate/acetate||0·38 (0·49)||2·1 (0·78)a||1·6 (0·83)|
|APC (CFU ml−1)||1·0 × 107 (1·1 × 107)||2·1 × 108 (1·3 × 108)a||6·5 × 108 (3·7 × 108)|
Clone libraries containing approximately 1900 16S rRNA gene sequences each were constructed to evaluate the bacterial population dynamics associated with the ensiling of wilted alfalfa and subsequent exposure to air. Before ensiling, 89·6% of the bacterial 16S rRNA gene sequences were associated with the phylum Proteobacteria, and 8·1% were associated with the phylum Firmicutes; other phyla identified were the Actinobacteria (0·2%), the Bacteroidetes (0·2%) and the candidate division TM7 (0·1%) (Table 2). Of the Proteobacteria-associated sequences, 99% were members of the class Gammaproteobacteria, of which 86·3% were related to the order Enterobacteriales and 10·3% were associated with the order Pseudomonadales. The sequences associated with the Firmicutes were exclusively members of the class Bacilli and were split into the orders Lactobacillales (85·1%) and Bacillales (14·9%). The Lactobacillales were primarily composed of members of the family Streptococcaceae (86·2%) and lesser amounts of the Leuconostocaceae (7·4%), the Lactobacillaceae (3·2%) and the Enterococcaceae (3·2%). The most commonly occurring OTU are shown in Table 3 and their phylogenetic relationships are shown in Fig. 1.
|Phyla||% BE||% PE||% PO|
|OTU||% Total||Classifiera (% confidence)||Best matchb (% similarity)||GenBankc||Previously isolated from|
|BE1||25·1||Enterobacteriaceae (100)||Unknown Enterobacteriaceae (99)||JQ726780||Arthropod|
|BE2||21·3||Erwinia (99)||Erwinia amylovora (99)||AB680317||Plum tree|
|BE3||16·7||Enterobacteriaceae (100)||Enterobacter sp. J33 (99)||JN969595||Peanut Rhizosphere|
|BE4||8·8||Pseudomonas (96)||Pseudomonas oryzihabitans (99)||JQ661238||Soil|
|BE5||4·8||Lactococcus (80)||Lactococcus garvieae (99)||JQ446487||Fermented food|
|BE6||2·4||Enterobacteriaceae (100)||Erwinia persicina (98)||EU681952||Pea plant|
|BE7||2·2||Erwinia (98)||Erw. amylovora (96)||FN434113||Crataegus sp.|
|BE8||1·1||Pantoea (100)||Pantoea brenneri (100)||FJ611808||Human infection|
|BE9||0·8||Pseudomonas (88)||Pseudomonas oleovorans (99)||GU357740||Textile effluent|
|BE10||0·8||Enterobacteriaceae||Enterobacter aerogenes (99)||JF430156||Chickpea|
|PE1||21·3||Lactobacillus (100)||Lactobacillus buchneri (99)||HQ711363||Vinegar|
|PE2||12·3||Enterobacteriaceae (100)||Enterobacter sp. MPR16 (99)||JN680700||Fermented food|
|PE3||11·7||Lactococcus (91)||L. garvieae (100)||JQ446487||Fermented food|
|PE4||8·2||Pediococcus (100)||Pediococcus pentosaceus (100)||JQ446479||Fermented food|
|PE5||7·2||Lactobacillus (100)||Lactobacillus lindneri (99)||AB512778||Stinky Tofu|
|PE6||4·5||Lactobacillus (100)||Lactobacillus plantarum (99)||JQ446567||Fermented food|
|PE7||4·4||Enterobacter (91)||Enterobacter ludwigii (99)||JQ308612||Soil|
|PE8||3·2||Lactobacillales (100)||L. garvieae (95)||JQ446487||Fermented food|
|PE9||2·9||Enterobacteriaceae (100)||Unknown Enterobacteriaceae (99)||JQ726780||Arthropod hindgut|
|PE10||2·1||Enterococcus (100)||Enterococcus durans (99)||HM218738||Fermented dairy product|
|PO1||32·9||Lactobacillus (100)||Lact. lindneri (99)||AB512778||Stinky Tofu|
|PO2||32·7||Lactobacillus (100)||Lact. plantarum (99)||JQ446567||Fermented food|
|PO3||9·5||Pediococcus (100)||Ped. pentosaceus (99)||JQ446479||Fermented food|
|PO4||6·5||Lactobacillus (100)||Lact. buchneri (97)||HQ711363||Vinegar|
|PO5||6·0||Lactococcus (84)||L. garvieae (99)||JQ446487||Fermented food|
|PO6||1·4||Enterobacteriaceae (100)||Enterobacter hormaechei (98)||HQ220153||Citrus roots|
|PO7||0·9||Lactobacillales (89)||Lactobacillus sp. TS4 (98)||AB284947||Fermented tea|
|PO8||0·9||Enterobacter (96)||Ent. ludwigii (99)||JQ308612||Soil|
|PO9||0·4||Weissella (100)||Weissella kandleri (99)||AB022922||Kimchi|
|PO10||0·2||Enterobacteriaceae (100)||Enterobacter sp. FMB-1 (99)||DQ855282||Soil|
After 40 days of ensiling, the number of sequences associated with the Proteobacteria dropped significantly (P < 0·05), from 89·6 to 26·9% with a concomitant significant increase (P < 0·05) in the phylum Firmicutes, from 8·1 to 70·6% (Table 2). Of the remaining proteobacterial sequences, 96·9% were associated with the class Gammaproteobacteria and the remaining 3·1% were associated with the Alphaproteobacteria. Of the Gammaproteobacteria-associated sequences, 97·6% were related to the family Enterobacteriaceae, and the majority of those were associated with the genus Enterobacter (~70%). Of the sequences associated with the phylum Firmicutes, 99·5% were associated with the class Bacilli, which were almost exclusively related to the order Lactobacillales (99·6%). The Lactobacillales-associated sequences were comprised of the families Lactobacillaceae (72·2%), Streptococcaceae (17·1%), Leuconostocaceae (2·2%) and the Enterococcaceae (2·1%). The most commonly occurring OTU are shown in Table 3 and their phylogenetic relationships are shown in Fig. 1.
After 20 days of oxygen exposure, the percentage of sequences associated with the Proteobacteria again dropped significantly (P < 0·05), from 26·9 to 3·5%; and those associated with the Firmicutes increased significantly (P < 0·05), from 70·6 to 95·0% (Table 2). The Proteobacteria that persisted were almost exclusively associated with the family Enterobacteriaceae (85·8%), with the genus Enterobacter making up the majority (>60%). Of the sequences associated with the phylum Firmicutes, the vast majority were associated with the class Bacilli (99·7%), of which 99·9% were associated with the order Lactobacillales, and the families Lactobacillaceae (92·8%) and the Streptococcaceae (3·9%). The most commonly occurring OTU are shown in Table 3 and their phylogenetic relationships are shown in Fig. 1.
Estimations of evenness, richness, coverage and the Shannon index revealed that although the libraries constructed from wilted alfalfa and ensiled alfalfa were dissimilar in composition; they contained similar levels of diversity (Table 4). The BE and the PE libraries contained similar levels of richness with 178 and 167 OTU, respectively. The percentage coverage for these libraries was approximately 31 and 35%, and the Shannon indexes were 2·63 and 3·11 for the BE and the PE libraries, respectively. The PO library contained less diversity than the other two, with less than half the richness of the BE and PE libraries (74 OTU), and a Shannon index of 1·93 (Table 4). The level of coverage in the PO library was greater than the other libraries at 43·2%, which is consistent with it having lower diversity. Rarefaction analysis is in agreement with the estimations of diversity, with the plots of the BE and the PE libraries almost superimposed, while the average slope of the PO curve is significantly lower (Fig. 2).
|Library||No. Clones||% Coverage||Richness (S)||Evenness||Shannon entropy (H)|
The preservation of herbage as silage is dependent on two critical factors: (i) the attainment of anaerobic conditions that stop plant and microbial respiration and encourage the growth of lactic acid bacteria and (ii) the establishment of a low pH environment that inhibits proteases and unwanted micro-organisms (Muck and Pitt 1994). Unlike cereal crops, such as corn, alfalfa is low in fermentable carbohydrates and high in buffering protein, making it one of the more difficult crops to ensile (McAlliser et al. 1998). However, all of the physical and chemical parameters we measured indicate that the alfalfa in our mini-silos produced high-quality silage. For example, the postensiling drop in pH to approximately 5 is similar to what others have reported for alfalfa silage (McAlliser et al. 1998; Denoncourt et al. 2006; Rossi and Dellaglio 2007). The pH reduction was due to the fermentation of sugars to mostly lactic and acetic acids, at concentrations similar to those reported previously (Denoncourt et al. 2006; Stevenson et al. 2006). We observed low amounts of butyrate and ammonia, indicating minimal clostridial fermentation and amino acid deamination (Driehuis and Oude Elferink 2000). In addition, the total weight loss of approximately 2·2% is well below the suggested maximum loss level of 4% for good lactic acid fermentation (McDonald and Whittenbury 1967; McDonald et al. 1973). The silage also had significant bacterial growth, mostly by lactic acid bacteria and minimal contamination by fungi and yeasts (data not shown). All of these factors indicate that the silage was well fermented and preserved.
Before ensiling, the majority of the 16S rRNA sequences obtained from the wilted alfalfa were associated with the phylum Proteobacteria, with the genera Erwinia, Escherichia, Pseudomonas, Pantoea and Enterobacter being the most abundant. To the best of our knowledge, the bacterial population structure of the alfalfa phyllosphere has not been reported previously; however, Redford and Fierer (2009) reported similar bacterial populations on the leaves of cottonwood trees. In addition, Delmotte et al. (2009) reported a predominance of the Proteobacteria on the surfaces of several plants including soybean, clover and arabidopsis; however, they observed substantially more Alphaproteobacteria. Lastly, Yang et al. (2001) observed a predominance of the Gammaproteobacteria on citrus tree leaves using culture-based techniques; however, their culture independent analysis gave inconsistent results. There was also a substantial amount of sequences representative of the firmicute Lactococcus garvieae (4·8%), which has previously been isolated from environmental samples including grass, soil and cattle wastewater (Klihn et al. 1995). It is likely that the colonization of plant surfaces by bacteria is a complex process that is dependent on many factors including plant species, climate, geographical location and type of fertilizer used.
The bacterial population structure of the alfalfa shifted significantly during the ensiling process, from predominantly Gammaproteobacteria to mostly Firmicutes. Of the sequences associated with the Firmicutes, >95% were associated with the lactic acid bacteria (LAB), most of which were homofermentative. Homofermentative LAB are desirable during ensiling because they produce two moles of lactic acid from each mole of glucose fermented, while heterofermentative LAB produce one mole of lactic acid, one mole of carbon dioxide and either a mole of ethanol or acetic acid (Muck 2010). Thus, homofermentative LAB have the ability to lower the pH of the ensiled material to a greater extent than heterofermentative LAB. However, the most abundant sequence we observed was associated with the heterofermentative LAB Lactobacillus buchneri. Sequences associated with the homofermentative LAB Pediococcus pentosaceus, L. garvieae and Lactobacillus plantarum, all of which have previously been associated with alfalfa silage (Stevenson et al. 2006), were also found in high percentages. Although we observed a preponderance of the LAB after ensiling, three of the ten most prevalent OTU were associated with the Enterobacteriaceae (PE2, PE7 and PE9). PE9, which is synonymous to BE1, the predominant OTU on the alfalfa before ensiling, is likely metabolically dormant, as its percentage decreased by over eightfold after ensiling. However, PE2, which has previously been associated with fermented foods, appears to be able to thrive in the anaerobic and acidic environment of the mini-silos, as the percentage of its sequences increased by over 20-fold after ensiling.
Although there were significant changes in the bacterial population structure after ensiling, the overall levels of bacterial diversity changed very little. Our results showed that the estimates of evenness, richness, Shannon entropy, coverage and rarefaction before and after ensiling were very similar. Hartmann and Windmer (2006) observed a similar phenomenon, that is, significantly different bacterial population structures with similar levels of diversity, when comparing agricultural soils managed under organic or conventional methodologies. They concluded that significant changes in bacterial populations do not necessarily lead to altered diversity indexes, if the changes in the level of some taxonomic groups are offset by opposite changes in other groups.
Exposure of silage to air is unavoidable during the feed-out period. Previous studies have shown that as little as 100 mg O2 per kg dry matter per day is adequate to deteriorate alfalfa silage (Woolford 1990). We exposed our silage to low levels of oxygen for 20 days and evaluated its effects on the silage. We found no significant effect on the pH of the silage, likely due to the alfalfa's high buffering capacity. However, there were significant losses in total weight, lactate, crude protein, sugar, starch and crude fat. There was also a >threefold increase in the aerobic bacterial counts, similar to those observed previously (Cai et al. 1999). From these data, we conclude that low levels of aerobic deterioration occurred during the 20 days of simulated feed out.
The bacterial population structure of the silage changed significantly during oxygen exposure. We observed significant increases in the percentages of 16S rRNA sequences associated with the phyla Firmicutes, Actinobacteria and Fusobacteria and a significant decrease in the percentage of the phylum Proteobacteria. Among the Firmicutes, which accounted for 95% of the sequences, the greatest increase was among the genus Lactobacillus. Although the lactobacilli produce lactic acid under anaerobic conditions, they can also convert lactic acid to acetic acid under aerobic conditions, and some species such as Lact. buchneri can do this anaerobically as well (Driehuis et al. 1999; Oude Elferink et al. 2001). While this conversion reduces the acidity of the silage, it also produces greater levels of acetic acid and 1-propanol, which have been shown to inhibit fungal growth (Driehuis et al. 1999). In addition, these products are of lesser metabolic value for microbial growth, thus improving the aerobic stability of the silage during feed out (Wilkinson and Davies 2013). The percentage of sequences associated with Lact. plantarum also increased significantly during oxygen exposure. Some strains of Lact. plantarum have been shown to enhance the aerobic stability of silage because they produce antifungal compounds such as phenyllactic and 4-hydroxy-phenyllactic acids (Lavermicocca et al. 2000). It is possible that the production of these antifungal compounds is the reason we did not observe a significant increase in the fungi or yeasts counts in the silage during oxygen exposure (data not shown).
The ensiling of herbage is a dynamic process in which the bacterial populations shift from mostly proteobacterial populations associated with the plant phyllosphere to predominantly lactic acid bacteria that rapidly multiply under the anaerobic conditions of the silo. Many producers choose to add commercial inoculants that contain freeze-dried cultures of homo-fermentative LAB to improve the efficiency of fermentation (i.e. less proteolysis and better DM/energy recovery) and hetero-fermentative LAB to enhance the aerobic stability of the silage. In future research, we will examine the effects of the addition of such inoculants on the bacterial population structure of silage.
The authors would like to thank Mr. Jason Kish for assistance with library construction and sequencing and Ms. Teresa O'Keeffe for quantification of fungi and yeasts. This work was partially funded by a grant from the California Air Resources Board (CARB).