Silage making of maize stover and banana pseudostem under South Ethiopian conditions: evolution of pH, dry matter and microbiological profile

Maize and Banana pseudostem by‐products can be considered as potential options in silage making under the environmental conditions of Ethiopia. Fermentation of banana pseudostem requires molasses to obtain a fast pH reduction.


Introduction
Tropical countries typically suffer from seasonal variations of available feed, both in quantity and quality (Salo, 2017). As a result, poor quality natural pastures and crop residues are the major components of today's livestock feed in Ethiopia (Kechero et al., 2014). A better use of locally available feed sources, which cannot be applied as human foods, is therefore getting more and more attention (Moselhy et al., 2015). In this regard, preservation of agricultural by-products, such as banana pseudostem and maize stover, may contribute to secure livestock feed. The banana plant is a herbaceous perennial species of the Musa family, grown widely as a fruit crop in both tropical and subtropical regions (Wang et al., 2016). About 69% (AE 37 thousand ha) of the Ethiopian land coverage for banana production is situated in the southern and south-western parts of the country, accounting for 77.53% (AE 370 thousand tonnes) of the Ethiopian annual fresh banana production by 22.38% (AE 1.5 million) of the banana producers (Alemu, 2017). The Gamo Gofa Zone is also known as a maize growing belt of the southern region covering about 51 442 ha of land and producing 189 655 tons of maize per annum (CSA, 2016).
Currently, smallholder farmers are underutilizing the byproducts of these two crops. The maize is delayed for drying and leaf shattering until the grain is harvested. The banana pseudostem is left aside in the field to decompose and to be used as organic fertilizer, but as there is an excess of banana biomass, the decomposition in the field can cause environmental pollution (Li et al., 2010;Yousef et al., 2019). Fermentation of forage is a lifelong tradition to conserve feed for periods of shortage caused by inadequate pasture and climate conditions (McEniry et al., 2006;Romero et al., 2015;Gharechahi et al., 2017). During silage, sugars are metabolized into organic acids by lactic acid bacteria (LAB) allowing storage for a long time while maintaining the nutritional quality and improving the palatability of the feed (Storm et al., 2010;Moselhy et al., 2015;Romero et al., 2015;Wang et al., 2016;Gharechahi et al., 2017;Ni et al., 2018). Several studies have been conducted on ensiling whole maize silage in different parts of the world (Israelsen et al., 2009;Storm et al., 2010), but not yet under tropical conditions as in South Ethiopia. Moreover, ensiling the whole maize plant including the fruits for feed purposes inhibits the use of the grain kernels for human use and is not appropriate for developing countries such as Ethiopia (Israelsen et al., 2009). In South Ethiopia, the spikes with grains are removed for human consumption and the crop is harvested relatively late. These two practices can be hypothesized to reduce the soluble carbohydrate content (WSC) available for microbial fermentation. Therefore, addition of molasses can substitute the WSC and enhance lactic acid fermentation (McDonald et al., 1991;Ni et al., 2018;Liu et al., 2020;Shiau et al., 2020). To the best of our knowledge, there is only one study in China (Wang et al., 2016) reporting on the feasibility of ensiling banana pseudostem for feed purposes, and bacterial profiles have not yet been determined by Illumina sequencing.
In this regard, the aim of this study was to evaluate the microbiological aspects of maize stover and banana pseudostem while ensiling under the environmental conditions prevailing in South Ethiopia, as a first step in judging the feasibility of the process to contribute to feed security. To this end, small-scale fermentations were performed with the two crops, while monitoring pH, dry matter changes and microbiological profile, using both culture-dependent (classic microbial counts) and cultureindependent (metagenetic analysis) methods.

Dry matter content and pH
The dry matter (DM) content at the start and during the four types of fermentation can be seen in Figure 1A. According to Romero et al. (2015), the silage evaluated in this study can be classified as low moisture silage having 54-64% DM content. The fresh, unsqueezed BPS exhibited a very low DM content about 12%, which is in agreement with the findings of Wang et al. (2016). Because a too high moisture content (or forage having less than 20% DM content) involves the risk of growth of spoilage microorganisms such as Clostridium species (McEniry et al., 2007;K€ onig et al., 2017), the chopped BPS was manually squeezed to increase the DM content. As can be seen in Figure 1A, there was statistically significant (P = 0.007) difference in the DM contents of the forages used at the start of ensiling. These variations can be due to differences between the two crops and also to the addition of molasses (Ni et al., 2017). There was also a statistically significant increase (P = 0.012, and P = 0.000) in DM content from day 0 to day 90 in all treatments. DM loss from volatile products that may have escaped during oven drying was not taken into account, however. The DM content increased much faster in the first 14 days compared to the rest of the fermentation period.
The pH of the biomass (Fig. 1B) recorded prior to ensiling showed significant differences (P = 0.000) between the two crop by-products. The variation in pH was likely due to differences in composition of the materials used in the experiment. At day 7, all treatments except BPSwo showed a significant decrease (P = 0.000) in pH. On day 90, there was a statistically significant difference (P = 0.000) in pH between maize, BPS with molasses and BPS without molasses. Moreover, no significant difference was found in maize with or without adding molasses. The maize stover fermentations displayed a rapid reduction in pH to values below 4 on day 7, regardless of molasses being present or not, after which it remained fairly constant. The reduction in pH was comparable with that observed by Gharechahi et al. (2017), who also studied the microbial dynamics of maize silage (under Iranian conditions). A significant decrease in pH was also observed for BPSw from day 0 to day 7 to a value of about 4.32 AE 0.07, and the decrease steadily continued to its lowest value reaching a value of 3.83 AE 0.04 on day 60 and tended to increase to 3.98 AE 0.11 on day 90. A comparable drop in pH was also reported by Wang et al. (2016) on the fermentation of banana pseudostem without squeezing (9.77 AE 0.16 dry matter content) and without adding molasses. Unlike the other three treatments, the pH of the BPSwo silages remained above 4.5 during the 90 days of ensiling. The significant difference (P = 0.000) between the two banana-based silages can only be related to the addition of the molasses that might have replaced the lost water soluble carbohydrates during squeezing (Ni et al., 2017;Bernardes et al., 2018). While some fungi and yeasts can grow at low pH values, a rapid decrease in pH to a value below 4.5 likely contributes to the inhibition of spoilage and pathogenic microorganisms (Hu et al., 2018;Liu et al., 2020).

Microbial counts
Microbial counts measured during ensiling are shown in Table 1. On day 0, the TVC did not differ significantly (P = 0.815) among treatments. On day 7, only the Mwo, Mw and BPSw treatments had reached a maximum in the TVC when considering the whole silage period, while the BPSwo reached a maximum on day 14. Between day 0 and day 90, all TVC decreased with 2.17, 2.75, 1.92 and 1.34 log cfu g À1 for Mwo, Mw, BPSwo and BPSw, respectively, which can be explained by a depletion of nutrients for growth and/or a decreasing pH.
Ensiling is based on the available nutrients, such as soluble carbohydrates, and the naturally occurring LAB on the leaves and stems of forages (Wang et al., 2016).
LAB are responsible for the preservation of fermented forages through production of organic acids, mainly lactic acid and acetic acid (Brusetti et al., 2006;Storm et al., 2010). Even though statistical differences were noted between the treatments at day 0, from a microbiological perspective, the biomass contained about the same density of LAB in all four types of silages ranging between 6.38 and 6.93 log cfu g À1 . A maximum in the LAB count was obtained at day 7 for all treatments, and also for all treatments, the counts decreased from then on to result in the lowest values (and for all treatments significantly lower than the value at day 0) at day 90. Interestingly, BPSwo attained the lowest TVC of all four treatments at days 7, 14 and 30, albeit not always statistically. It is plausible that the growth of LAB being the least  pronounced for BPSwo is caused by the expected loss of nutrients (not analysed in the current study) due to the squeezing. This could reduce the lactic acid production and explain the diminished pH reduction among all treatments (Duni ere et al., 2013). A further consequence of this is likely to be seen in the dynamics of the Enterobacteriaceae. A first observation for this bacterial subgroup is that on day 0, the two banana-based treatments contained a statistically (P = 0.000) higher Enterobacteriaceae count than the two maize-based treatments. For all treatments, except for BPSwo, the number of Enterobacteriaceae was below the detection limit of 1.00 log cfu g À1 as from day 7. In the case of BPSwo, Enterobacteriaceae were only below the detection limit as from day 30, probably due to the fact that pH was not so low and LAB counts not so high as in the other three treatments. Therefore, the squeezing of banana pseudostem without addition of molasses may hinder the initiation and dominance of beneficial microorganisms within the recommended 7-10 days of ensiling (Borreani et al., 2018).
On day 0, the yeast and mould counts showed no statistically significant differences among treatments. On day 7, there was statistically significant (P = 0.000) reduction in the yeast and mould counts in the Mwo and Mw treatments. A reduction was also observed in BPSwo and BPSw treatments, but it appeared later, i.e. at day 14. At the end of ensiling, the yeast and mould counts did not show a significant difference (P = 0.117) between the treatments and the average counts per treatment were between 4.24 and 4.67 log cfu g À1 . Yeasts and moulds in general can grow at a lower pH than bacteria, but the reduction in the pH occurring in the fermentations in our study, possibly combined with a depletion of available nutrients, caused a general decline in the counts for all treatments. Although no clear relation exists between fungal growth and mycotoxin production, high fungal counts are to be avoided in order to limit the risk of contamination with mycotoxins and other deleterious effects, such as nutrient use and taste changes in the ensiled crop (Wambacq et al., 2013).
Finally, all treatments contained Clostridium spores at the start of silage were above the limit of detection 2 log cfu g À1 (K€ onig et al., 2017). The higher count seen on maize compared to banana pseudostem might relate to the soil contamination present on the former, whereas for banana pseudostem, the outer layers were discarded (Duni ere et al., 2013). While during the first weeks of silage, all treatments exhibited a reduction in the number of Clostridium spores, they all revealed a slight increase during the last two months of the experiment, yielding values that did not statistically differ among treatments (P = 0.699). Attention should be paid to the slight increase in Clostridium spores, because when pH remains above 4.5, as in the case of BPSwo, there is a risk of secondary fermentation by Clostridium of lactic acid into for instance acetic and unpalatable butyric acid and of degradation of proteins and amino acids (Wilkinson, 2005). Growth of Clostridium is unlikely in forages having a DM content of greater than 40% (K€ onig et al., 2017). Such forages used for silage making are characterized by a high osmotic pressure and a reduced dissociation of organic acids produced during fermentation to keep the pH low (Silva et al., 2015;K€ onig et al., 2017). Therefore, the squeezing activity performed on the chopped BPS resulted in an increase in DM content from 12% to about 56%, thereby reducing Clostridium growth and sporulation, even at the higher pH in BPSwo (K€ onig et al., 2017). Generally, silages containing Clostridium spore counts of more than 2 log cfu g À1 , a high pH (i.e. above 4.5), a low DM content (i.e. below 20%), substantial Enterobacteriaceae counts and an unpleasant smell can be considered as bad silages that may negatively affect the performance and health of livestock (McEniry et al., 2006;Wambacq et al., 2013).
For the four types of fermentations, it can be observed that there was a distinct shift in the bacterial profiles between day 0 and day 7, and from then on there was a more gradual further evolution of the profile towards the end of the silage period. For Mwo ( Fig. 2A) (Gharechahi et al., 2017). In Mw (Fig. 2B), a relatively similar evolution can be observed, with the difference that Buttiauxella (OTU 1) was present at a higher relative abundance on day 90 than in Mwo. The facultative anaerobic Buttiauxella species belong to the family of the Enterobacteriaceae. In the colony counts, no Enterobacteriaceae were detected as from day 7, but the DNA of Buttiauxella species was observed in the metagenetic analysis throughout the fermentation period. Either the detected DNA belonged to dead cells, or they were viable but not cultivable anymore (Andeta et al., 2018). The strictly aerobic bacteria that were present on day 0, i.e. Methylobacterium (OTU 34; 3.88%) and Sphingomonas (OTU 37; 3.98%) disappeared completely after day 0. For BPSwo (Fig. 3A) and BPSw (Fig. 3B), the dynamics are somewhat less evident, but for both types of silages, the abundances Leuconostoc (OTU 3) and Buttiauxella (OTU 1) faded away towards the end of the fermentation, while Bifidobacterium (OTU 4) clearly established a relative abundance of 28.5% and 24.5% at day 90 for BPSwo and BPSw respectively. In a study conducted by Li et al. (2019), in the evaluation of tropical forage grasses, Bifidobacterium was seen dominant with prolonged fermentation and increased temperature conditions. Bifidobacterium species are characterized by the conversion of free conjugate linoleic acid into conjugated linoleic acid, and they are known for their health benefits for both humans and animals (Coakley et al., 2003;Li et al., 2019). The dominance of Bifidobacterium, as seen in this study, may benefit ruminant animals (Li et al., 2019). The impact on the use of molasses on the bacterial profiles is more clear for banana pseudostem than for maize stover. ). This finding largely coincided with the LAB counts on day 90. Therefore, in the silage of banana pseudostem squeezed to reduce its moisture content, there is a need for molasses addition. Remarkably, OTUs related to Clostridium were only detected on day 30 and 60 in BPSwo. On these days, the Clostridium spore counts were also (slightly) higher for this treatment than for the other three treatments. It has to be noted, however, that as seen in previous research (Andeta et al., 2018(Andeta et al., , 2019, endospores are resilient to many traditional methods of DNA isolation and can therefore remain underdetected in metagenomics, even when a tailored DNA extraction method is used (Filippidou et al., 2015). As shown in Table 2, the coverage, a parameter that indicates the percentage of individuals sampled in a microbial community, ranged from 60.4% to 80.5%. The observed richness indicates the number of OTUs observed in a sample, but the Chao1 index, estimates the true species diversity of a sample by also taking into account rare species (singletons and doubletons). The Chao1 index was higher than the observed richness. Based on these observations, it can be concluded that the bacterial community was not completely covered during sequencing. Nevertheless, it can be deduced that for maize stover, there was no clear trend when comparing Mwo and Mw, but for banana pseudostem, the observed richness of BPSwo was higher at any time in the fermentation than for BPSw. The higher diversity was likely due to the relatively higher pH values of BPSwo silages, which could not suppress the microbial growth to a substantial extent and subsequently reduce the bacterial community diversity (Ni et al., 2017). Therefore, the addition of molasses clearly affected the bacterial community into fewer but more dominant species in the silages of BPS. The Shannon-Wiener index describes diversity in terms of the observed richness combined with the relative abundance of the OTUs. The index increases as both richness and evenness of the community increase. For both crops, when molasses was used, this diversity index slightly decreased during the first 14 days and from then on it increased steadily. This phenomenon was not evident for any of the crops when no molasses was added.

Conclusions
It can be concluded from this study that both maize stover and banana pseudostem, when ensiled under conditions prevailing in South Ethiopia, show a microbial profile during as generally expected for ensiled crops, but in the case of banana pseudostem, squeezing or pressing as well as the addition of molasses are necessary before the fermentation. Except for banana pseudostem without molasses, the pH and the microbiota during fermentation showed dynamics that are known to characterize proper silage processes. The culture-dependent counts and the metagenetic analyses revealed corresponding findings, but at the same time provided complementary information. The dominance of Bifidobacterium observed in the last days of the silages of banana origin may have additional health benefits for ruminants. Further research should focus on the chemical and nutritional quality of the silages as well as on the opportunity to ferment mixtures of the two crop by-products.

Forage harvest and silage preparation
Prior to collecting biomass for fermentation, 60 microsilos were constructed with a capacity of 4 l (10 cm diameter and 50 cm height) using unplasticized polyvinyl chloride (uPVC) tubes. The microsilos were sealed at the bottom using gas tight glue (PVC glue, China), checked for water leakage and disinfected using 98% ethanol.
Maize (Zea Mays, BH-140) was grown at the Arba Minch University farm, which is located at 437 km south of Addis Ababa in the southern hot humid rift valley region of Ethiopia. The spikes of the maize at milk-line kernel maturity stages were harvested for human use (Israelsen et al., 2009;Gharechahi et al., 2017). The maize stover was immediately harvested manually at about 10 cm height and chopped using a forage chopper (Electro Mecce, Engineering Service, Ethiopia) to an average particle length of 2 to 4 cm. In this way, a total of 200 kg of maize stover were obtained, homogenized and equally divided into two 100 kg aliquots. One aliquot was divided over a series of 15 microsilos, and as no molasses was added, this treatment is further referred to as maize without molasses (Mwo). The other half (100 kg) was mixed thoroughly with 5% (w/w) molasses and divided over another series of 15 microsilos, referred to as maize with molasses (Mw). In the current study, molasses with 16.2% DM content was added to compensate for the loss of WSC by removing the spikes (grain) from the maize or by squeezing the banana pseudostem to increase its DM content. The filling of the microsilos was performed using hand tools, and the biomass was compressed to 80% of the microsilo volume. Compression forced out the trapped oxygen to reduce further plant respiration and to obtain anaerobic conditions (McEniry et al., 2008;Gharechahi et al., 2017). In each microsilo, a cement stone designed to tightly fit within the microsilo and weighing 1.5 kg was placed on the top of the biomass to maintain pressure before the microsilo was sealed with a PVC cap and scotch tape (Wang et al., 2016;Andeta et al., 2018). Subsequently, the top caps were equipped with a CO 2 valve.
Banana pseudostem (Musa acuminata, BPS) was randomly collected from a local farmer's field in the neighbourhood of the Arba Minch University after the banana fruits were harvested. The BPS was cut into short pieces using big knife and chopped to an average length of 2-4 cm by using the same chopping machine mentioned above. Then, the chopped banana pseudostem products were squeezed manually to reduce the moisture content following the procedure stated by Andeta et al. (2018), in enset (Ensete ventricosum) fermentation. A total amount of 200 kg chopped and squeezed BPS was obtained and divided into two aliquots of 100 kg. One aliquot was divided over a series of 15 microsilos, and as no molasses was added, this treatment is further referred to as BPS without molasses (BPSwo). The other half was mixed thoroughly with 5% molasses (w/w) and divided over another series of 15 microsilos, referred to as BPS with molasses (BPSw). The microsilos were compacted Different superscripts within the same column and within the same treatment for the different sampling days indicate significant differences (P < 0.05).
A,B,C Different superscripts within different treatments, but same sampling day, indicate significant differences (P < 0.05). 1. Data are the mean values of two sequenced extracts AE standard deviations. 2. Chao1 index: the total number of OTUs estimated for infinite sampling (Chao, 1984). 3. Coverage was determined by using the equation (observed richness/Chao1 estimate) 9 100. 4. Shannon-Wiener index: index to characterize species diversity based on species richness as well as their relative abundance. A higher value means more diversity (Shannon, 1948). Data among sampling days and sample types were analysed using one-way analysis of variance (ANOVA). The statistical differences among means having significant difference were determined using Turkey's multiple comparisons. Differences were considering significance at P < 0.05. and closed in the same manner as described for maize. All microsilos were stored in the same laboratory room until their respective sampling dates.

Sampling
On day 0 prior to ensiling, fresh samples were taken and stored under refrigerated conditions for maximum 24 h until analyses of the microbial population of the raw material was performed. After 7, 14, 30, 60 and 90 days silage, triplicate microsilos were destructively sampled per treatments as described by McEniry et al. (2008) and Andeta et al. (2018). Samples of about 80 g were taken at a depth of about 20-30 cm of each microsilo in a sterile way and immediately subjected for the DM, pH and microbial count analyses explained in the next sections. 15 g of each of these samples was freeze-dried for 24 h at À50°C (Alpha 1-4 LD plus freeze-drier, Martin Christ, Osterode am Harz, Germany) and stored at À18°C for a period of 4 months before metagenetic analyses.

Dry matter content and pH
The dry matter content and pH were measured according to the methods used by Moselhy et al. (2015) and Andeta et al. (2018) using the oven-drying method and a digital pH meter (PH 1100H, VWR International, Darmstadt, Germany) respectively. The average and standard deviation were calculated for the three microsilos per treatment.

Microbial counts
Classic colony counts were determined as described by Andeta et al. (2018) and McEniry et al. (2008). Five grams of sample was transferred aseptically into a sterile stomacher bag, and 45 ml of peptone physiological salt solution (0.85% NaCl, 0.1% peptone, Biokar Diagnostics, Beauvais, France) was added. The mixture was homogenized for 60 s (StarBlender TM LB 400, VWR International, Fontenay Sous Bois Cedex, France). A ten-fold serial dilution was plated on different media using the pour-plate method. Total viable aerobic counts were determined on Plate Count Agar incubated at 30°C for 3 days, Enterobacteriaceae on Violet Red Bile Glucose Agar incubated at 37°C for 24 h, LAB on de Man Rogosa Sharpe Agar incubated at 30°C for 3 days, yeasts and moulds on Dichloran Glycerol (DG-18) agar supplemented with 0.1 g l À1 chloramphenicol, incubated at 25°C for 5 days. Clostridium endospores were counted by giving the ten-fold serial dilution a heat shock treatment (10 min at 80°C), followed by serial dilution, plating onto Reinforced Clostridium Agar (RCA) and anaerobic incubation at 37°C for 48 h using anaerobic jars, gas generating kits and indicator strips (IVD, Microbiology Anaerotest â , Merck KGaA, Darmstadt, Germany). As a heat shock treatment was included in the counting protocol for Clostridium, only Clostridium spores were counted rather than a total Clostridium count including also the vegetative cells. This approach was used because the medium available, RCA, also supports the growth of lactobacilli and other anaerobes. By combining it with a heat treatment and anaerobic incubation, Clostridium could be determined specifically. All media were obtained from Biokar Diagnostics (Beauvais, France). Microbial counts were determined for each replicate and expressed as log cfu g À1 .

Bacterial community composition by metagenetics
Five grams of freeze-dried silage samples was aseptically crushed and homogenized in 45 ml of sterile peptone water (peptone, 1 g l À1 ) for 2 min at 260 rpm using a Bagmixer â Stomacher (Interscience, Saint Nom, France). The homogenate was centrifuged at 16 000 g for 5 min. The total bacterial DNA was extracted from 0.25 g of pellet using the Power soil DNA Isolation Kit (Mo Bio Laboratories, Carlsbad, California, USA) following the manufacturer's instruction. Afterwards, replicate DNA extracts were pooled and diluted 10 times. Subsequently, PCR amplification, library preparation, sequencing, sequence processing and diversity analyses were performed as described by Andeta et al. (2019). Observed richness, Chao1, Coverage and Shannon-Wiener diversity indices were calculated using the R package Phyloseq v1.16.2 (R-package, R Development Core Team 2013).

Statistical analysis
Differences in the dry matter content, pH, microbial counts and diversity indices (richness, Chao1, Coverage and Shannon-Wiener diversity index) throughout the different sampling days and between the four sample types were analysed using one-way analysis of variance (ANOVA). All tests were performed using SPSS (IBM©SPSS Statistics v. 23, New York, USA). For treatment means showing significant differences, multiple comparisons were performed using Turkey's HSD, while considering a significance level of 0.05.