Using culture ‘omics to explore the microbial structure and function in an equid in vitro digestion model

Abstract

The in vitro gas production system (GPS), developed to estimate degradation of ruminant feedstuffs, has been adapted for equine use. This study aimed to characterise the bacterial community profile and metabolome of donor faeces and faecal inoculum within the GPS when fermenting the same diet as faecal donors. Six Welsh ponies on identical diets were faecal donors with samples collected for microbiome profiling and system inoculation. Gas production (manual pressure transducer technique) was performed for 156 h with 2 replicate bottles from each donor harvested at 8,20,28 and 36 h. Faecal and inoculum samples were subject to PMAxx for viability PCR, 16S rRNA sequencing and 1^H NMR metabonomics. Time in the GPS effected bacterial community profile, metabolic phenotype and predicted metabolic pathways. Collectively a system dysbiosis was observed at 8 h. End point metabolic profile was similar to the donor faeces but GP fibre degrading microbiota better reflected previously reported literature on rumen microbiota, rather than those found in horses. The GPS estimates dry matter digestibility similar to in vivo digestibility, resulting in a similar metabolic profile to donor faeces. However, the GPS either favours rumen dwelling microbiota or demonstrates functional microbial redundancy compared to the donor faeces of equids.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-29936-w.

Introduction

Within livestock nutrition there is a requirement to consider the sustainability of animal diets in line with the United Nations (UN) Sustainable Development Goals (SDGs)^1,2. The effects of climate change on crop production and a requirement to become more selective in the feed materials used for feeding horses highlight the importance of effective ways to screen and measure digestibility of novel feedstuffs that may be suitable in horse nutrition. In vivo digestibility trials remain the gold standard approach to test the digestibility and suitability of feed materials, however these trials are labour intensive, costly to deliver and only provide end-point data which limits their value. In contrast in vitro methods are cost-effective and provide essential kinetic detail on feed degradation of single substrates and feed combinations that can enhance screening prior to undertaking an animal trial. At a recent European strategy meeting both technical specialists within the horse feed industry and academics defined the most important goal for the horse feed industry in order to navigate changes brought about by our changing climate, was a validated in vitro digestion system³.

The in vitro gas production system, originally developed to evaluate the degradability of ruminant feedstuffs^4,5 was soon adapted by Lowman et al.⁶ for equine use by the utilisation of equid faeces as the source of inoculum. Cumulative gas produced and substrate dry matter disappearance can be mathematically modelled to give fermentation kinetics on the rate of fermentation, the predicted flow rate through the animal and the overall potential digestibility⁷. Gas production is based upon three stages; the slow lag phase, the rapid exponential phase, and the slow phase of diminishing gas production⁸. Within the last 30 years the in vitro gas production system has been used to rank feed degradation, and more recently to model the horse’s hindgut ecosystem in vitro as well as how that ecosystem performs under differing conditions^9–13. There are however still unanswered questions around the use of this in vitro tool to model the horse’s hindgut microbiome e.g. it is currently unknown which microbiota from the donor faeces survive in the GP system and how this alters over the fermentation time, nor if the microbiome within the system evolves away from the faecal donor.

Leng et al.¹⁴ created a multi vessel in vitro equid hindgut model using faeces to inoculate the system. This system allows for the study of the microbiome and metabonome in a model environment, but does not enable measurable digestion of specific nutrients nor fermentation kinetics to be modelled; both being important parameters to measure the potential value of novel feedstuffs for equine use. Therefore, the gas production system might be advantageous over more complex models to screen the degradability of novel feedstuffs. However, if the gas production system is to be used to predict both degradation and microbiome function in horses in vitro then it is important to know how representative the bacterial community profile and metabonome of the faecal inoculum are compared with the original donor faecal samples, alongside the degradation profile. This information on the accuracy of the GP system is vital to the development of sustainable equine nutrition.

The aim of this study, was to characterise the viable bacterial community profile and their metabolic function over time in the manual in vitro gas production system when compared to the donor faeces.

Results

Gas production profiles were plotted for each donor animal and fitted to the France et al.⁷ fermentation kinetics model. The mean profile demonstrated a good fit to predict fermentation kinetics, R² 0.998. There was a mean lag time of 56 ± 54 min between inoculation of the system and the start of gas accumulation. The model predicted 506.75 (± 52.28) ml/g of degradable dry matter, similar to the dry matter disappearance of the same diet in vivo 591.35 (± 59.42) g/kg obtained by a separate, unpublished, digestibility experiment.

Bacterial community profile

A key objective was to identify the survival of microbiota from the faeces over time in the GP system. The number of Amplicon Sequence Variants (ASVs) representing all of the live microbiota across all of the sampling points was compared to the ASVs representing the total (live and dead) microbiota as a measure of survival within the GPS, see Fig. 1a. There was no difference in ASV counts between live (36785.2 ASVs) and total (40357.7 ASVs) microbiota (Mann Whitney n = 80, P = 0.61). Both alpha and beta diversity were explored for live and total microbiota in the system, Fig. 1b, and showed the survival of microbiota did not influence the alpha diversity, Shannon, (Mann Whitney n = 85, 179 vs. 172 P = 0.96) nor beta (Fig. 1c), Bray Curtis (Pairwise PERMANOVA n = 163, with 999 permutations, P = 0.62) diversity.

Fig. 1.

(a) Amplicon sequence variants, (b) Shannon diversity index and (c) Bray Curtis distances for total (dead and live) microbiota compared to only live microbiota within faecal samples and faecal inoculum throughout the in vitro fermentation.

When examining the feed used in the system the mean ASV count for total microbiota (dead and live) was 21,943, whereas the live ASV count was 3497 suggesting many of the microbiota detected on the feed sample prior to inoculation were not viable in the system itself. With regards to microbiota diversity there was no difference in alpha nor beta diversity between the live microbiota and total microbiota feed samples.

With the exception of the feed sample, that displayed greater total ASVs than live ASVs, there were no differences in the bacterial community profile between the total and live ASVs thoughout the time points in the fermentation system, therefore the focus on the findings was placed on the live microbiota only.

Differences in ASVs at genera level are presented in Fig. 2 (LeFSe LDA > 2, FDR < 0.05). Overall there were 113 differing ASVs in total of the 422 ASVs observed at this classification in the biom table. There were 13 ASVs that differed in prevalance in the starting faecal sample compared to later time points. At 0 h in the inoculum for the in vitro system 10 ASVs differed to the other time points. By 8 h into fermentation one ASVs differed in prevalance to other time points, Streptococcus sp. with an LDA score of 4.25. After 20 h of fermentation three further ASVs differed in prevalance to other time points, Pseudobutyrivbrio, Veilonella and Agathobacter. The bacterial community profile continued to evolve over time and by 28 h a further 3 ASVs differed compared to the previous and future time points within the system. By 36 h into fermentation a further 11 ASVs differed to the other time points and finally by 156 h of fermentation there were 9 ASVs differing from all other time points within the system.

Fig. 2.

LEfSe plot of amplicon sequence variants at genus differing between time points within the fermentation system. Each ASV presented had a LDA > 2 and FDA < 0.05 reflecting differences in abundance of these ASVs over the differing time points within the GP system.

Three measures of alpha diversity were evaluated, Fig. 3. (1) observed ASVs; the number of different taxa, (2) Chao1 richness by allowing for rare taxa within samples and (3) shannon index which considers richness and evenness of microbiota. For observed ASVs there was no effect of differing taxa over the time points (Kruskall-Wallis P = 0.37) (Fig. 3a), for Chao1 there was no difference in richeness between any of the time point (Kruskall-Wallis P = 0.38) (Fig. 3b). However, for Shannon index of species richness there was a difference between time points (Kruskall-Wallis P < 0.001) (Fig. 3c). Differences were observed between the 0 h inoculum and 8 h, 8 to 20 h and 28to 36 h for species richness (each comparison; P > 0.0010, FDR > 0.001), but no differences observed in the species richness between 0 h and 156 h. For beta diversity there was a difference in diversity between time points, Fig. 4 (Bray-Curtis, Permanova P < 0.001). Differences were observed between the starting faecal sample and the 0 h faecal inoculum (P = 0.005, FDR 0.005), 0 h inoculum and 8 h (P = 0.001, FDR 0.001),8–20 h (P = 0.001, FDR 0.001) and 28–36 h (P = 0.005, FDR 0.005). There were no differences observed between the other sequential timepoints.

Fig. 3.

Alpha diversity within the in vitro system over time, Chao1 and Observed ASVs suggest limited difference in rare ASVs and richness of ASVs over time, whereas the Shannon diversity index shows a reduced diversity within the microbiome 8 h into fermentation.

Fig. 4.

Beta diversity, Bray curtis, within the fermentation system over time showing the faecal samples and 0 h inoculum samples were similar in diversity. By 8 h into fermentation beta diversity differed to the original samples, gradually over the time points beta diversity evolved to finish similar the the starting faecal sample.

The core microbiome, ASVs present in 75% of the samples, were collated for each time point. There were four core ASVs present in the faecal samples and the 0 h inoculum, two Streptococcus, Luteciea and Ruminococcacaeae. There were no unique ASVs in the core microbiome in the faecal samples and no unique ASVs between the starting faecal sample and the 0 h faecal innoculum created from the starting sample. A core microbiome venn diagram was constructed for 0–156 h time points, Fig. 5.

Fig. 5.

Venn diagram representing core ASVs in the fermentation inoculum from 0–156 h of fermentation, each sampling time is represented by a colour. The numbers in the diagram represent core ASVs at each sampling time, where numbers are within sections that overlap this represents common ASVs between those time points within the core microbiome of the GP system. The total number of ASVs present in the core at each time point is given with the time point outside of the Venn diagram.

Within the system there were 23 ASVs in the 0 h inoculum that were unique to that time period, 4 core ASVs that were shared with all other time periods (two differing Stroptococcus, Luteciae and Weissella), 1 ASV shared with 36 h and 1 ASV with 156 h. After 8 h of fermentation the same 4 core ASVs remained, 2 ASVs shared with 28 h, 1 shared with 36 h and one ASV (Aggregatibacter) unique to 8 h. By 20 h of fermentation there were 2 core ASVs shared with 8 h, 4 ASVs shared with 0 h, 2 ASVs shared with 28 h, 1 ASV shared with 28, 36 and 156 h, 1 ASV shared with 36 and 156 h, 5 ASVs shared with 28 and 36 h, 1 ASV shared only with 36 h, 1 ASV shared only with 156 h and 4 ASVs unique to 20 h. By 28 h there were no ASVs shared with the 0–8 h fermentation inoculum, 2 ASVs were shared with 20 h and 7 with 36 h. There was one ASV shared between 28, 36 and 156 h, 2 ASVs shared between 8, 20, 28, 36 and 156 h. After 36 h of fermentation there were 11 unique ASVs in the core microbiome, 1 ASV shared with 8 h, 1 ASV shared with 28 and 36 h. There was one ASV shared with 8, 20 and 36 h, 1 ASV shared with 20 h, 1 ASV shared with 36 and 156 h, 1 ASV shared with 20, 28, 26 and 156 h. There were 7 ASVs shared with 28 h and 36 h only. By 156 h there were 4 ASVs unique to that time point and 9 shared between the other time points. A full table of the core microbiome ASVs can be found in supplementary materials S1.

The functions of the microbiota based upon the marker gene sequences suggest a similar functional profile between the faecal samples and the faecal inoculum. However, by 8 h into fermentation the functional profile was different to the faecal sample used to inoculate the system and later time points. As the incubation time progressed the functions within the microbiome became similar to those in the starting inoculum and the faecal sample, after 28 h there was no functional clustering by time point, see Fig. 6. Figure 6 (E–H) shows differences in specific KEGG orthologs over time considering general carbohydrate metabolism, as well as more specifically butyrate, propionate and lactate metabolism.

Fig. 6.

PCA plot of predicted microbiome function by time point, (A) showing the differences in predicted function from the faeces through to 156 h of fermentation. (B) Functional differences between 0 h and 8 h of fermentation, (C) between 8 and 20 h clustering of functional pathways was reduced. (D) By 28 h there was no clustering of functional pathways compared to 20 h and no difference between the sequential time points for the rest of the fermentation period. (E) Carbohydrate metabolism differences from K00001 between time points, (F) Butyrate metabolism, differences between time points, (G) Propionate differences between time points and (H) lactate dehydrogenase metabolism differences between time points.

Predicted microbiome function was generated using PiCruSt2, and the functional profile over the time points in the score plots, Fig. 6. Within the inoculum at 0 h there were five pathways that were different from other time points. By 8 h into fermentation there were 73 differing pathways to the other time points in the system. Only one pathway was different from the other time points at 20 h into fermentation and a further four pathways by 28 h. Seven pathways differed from other time points by 36 h and only one final pathway differed from other time points by 156 h of fermentation.

Metabolic profiling

Initially a principal component analysis (PCA) was constructed for all of the time points within the study for the metabolic profiles (Fig. 7a). These data suggest that the metabolic phenotype of the faecal samples and the starting inoculum (0 h) were very similar. By 8 h into fermentation the metabolic phenotype appeared different from all other time points. From 20 h on to 156 h the metabolic phenotype of these time points started to cluster together.

Fig. 7.

(a) PCA scores plot of the differing time points sampled from the in vitro system, (b) S-line plot of the differences in the metabolic phenotype between 0 and 8 h of fermentation. (c) S-line plot of the differences in the metabolic phenotype between 8–20 h of fermentation, (d) S-Line plot of the differences between 36 and 156 h of fermentation and (e) S-line plots showing the differences between the starting faecal samples and 156 h of fermentation.

Comparing the starting faecal samples and the 0 h inoculum an OPLS-DA model with good predictive ability (Q²Y 0.948, P < 0.001) was constructed, however, there were no differing metabolites between these two time points. By 8 h of fermentation (Q²Y 0.978, P < 0.001) there was a significant decrease in formate (n = 24, 1.15 to 0.49 µM P < 0.001) and ethanol (n = 24 8.88 to 5.91 µM P = 0.005), Propionate (n = 24 19.08 to 9.68µM, P < 0.001), Acetate (n = 24 16.67 to 9.79µM, P = < 0.001) and Butyrate (n = 24 6.67 to 2.53µM, P < 0.001). Both dihydroxyacetone and ethlyglycol dimethlineither were detected in the model as differing between groups, however, there was no difference in the concentration of dihydroxyacetone (n = 24 8.81 to 8.86µM P = 0.84) and ethlyglycol dimethlineither (n = 24, 2.80 to 2.33µM P = 0.11) between these two time points. Lactate increased in the system (n = 24 7.72 to 10.59µM P < 0.001) compared to 0 h. By 20 h of fermentation (Q²Y 0.937. P < 0.001) the system had greater levels of the SCFAs butyrate (9.27 to 14.45µM, P = 0.01), and sacchropine (3.6 to 10µM P < 0.001). For propionate (7.45 to 6.33 µM, P = 0.19), gamma hydroxybutyrate (2.8 to 2.91µM, P = 0.8), N-Acetylalanine (3.26 to 4.29µM, P = 0.1) and acetate (8.9 to 10.9 µM, P = 0.14) there was no difference in metabolite concentration between 8 and 20 h however the overall metabolic profile differed between these two time points in the model. Lactate decreased at 20 h (15.24 to 9.15µM, P = < 0.001), as did succinate (5.41 to 3.62µM, P = 0.002) and propionaylcarnatine (1.28 to 0.62µM, P < 0.001) (Fig. 7c). There was a good predictive model between 20 h and 28 h (Q²Y 0.61, P < 0.001) but no differing metabolites were detected within the model. For 28 to 36 h there was a poor predictive model (Q²Y0.37) and therefore this model was not explored further. For 36 h and 156 h (Fig. 7d) there was a good predictive model (Q²Y 0.65, P = 0.005). Several metabolites were detected as differing in the overall metabolic profile but there were no concentration differences between 36 and 156 h, Acetate (n = 24 12.07 to 16.98µM, P = 0.24), Ethanol (24.72 to 17.61µM, P = 0.22) and ethyleneglycol dimethlenether (1.12 to 1.6µM, P = 0.37). A final set of pairwise comparisons was undertaken between the starting faecal sample and the 156 h fermentation media (Q²Y 0.987, P < 0.001), which showed that the fermentation media had a greater level of acetate (2.43 to 14.61, P = < 0.001), butyrate (0.55 to 9.51µM, P < 0.001), propionate (1.9 to 8.02µM, P < 0.001), adipate (2.83 to 9.5µM, P < 0.001) and sacchropine (0.56 to 7.98µM, P < 0.001) (Fig. 7e).

ASVs identified as differing from other time points in the study were correlated with metabolites that differed at the same corresponding time point. In the starting faecal sample there was a negative correlation between the concentration of formate and the prevalence of Fibrobacters (R² −0.72). In the 0 h inoculum there was a positive correlation between Oscillospiraceae UCG-005 and propionate (R² 0.5). There were no correlated ASVs and metabolites at 8 h, but after 20 h there was a negative relationship between formate concentration and Psudobuytrivbrio (R² −0.65). By 36 h into fermentation there was a positive correlation between Ethyleneglycol-dimethylether concentration and Lachnospiraceae.FD2005 abundance (R² 0.6) and no relationships between metabolites and ASVs at 156 h.

Discussion

A main objective of this study was to identify how the viable bacterial community profile in the GP system compared to the donor faeces when the same diet was used in both. Faeces are accepted as a non-invasive proxy for a range of experiments that aim to characterise either hindgut bacterial community profile, metabolic function or in vitro gas production^6,15,16. The data in this present study show no difference in alpha nor beta diversity in live or total ASVs within the GPS. These data suggest that microbiota from faeces can survive and are broadly represented within the GPS as a simple model of the horses hindgut, these general findings are in broad agreement with the work of Dougal et al.¹⁷ who showed no differences in alpha diversity between the differing hindgut regions. More recently Leng et al.¹⁴ successfully used faeces to inoculate a multi stage hindgut model which reflected a similar bacterial community profile to faecal samples collected directly from regions of the hindgut. The findings from our study support this whereby we observed a similar bacterial community profile in the GP system at the end of fermentation compared to the donor faeces. The present study also considered metabolic function, both predicted and the metabolic phenotype from the metabonome. Despite some individual pony variability and the GP system variation, the functional findings of the predicted metabolic pathways and metabolic outputs supported the bacterial community profile we observed.

The structure and function of the bacterial community profile within the fermentation system altered over time within this study. The concept itself of bacterial turnover in the GPS has previously been considered from a functional perspective to consider the effect on fermentation¹⁸. To the authors knowledge this is the first study to consider the effects of GP on the microbiome structure and function within the GPS system over the fermentation time period, when compared to the original donor faeces profile. These data allow the evaluation of both the accuracy and application of the GPS as a tool to screen feed substrates alongside fermentation kinetics. Initially, there was a reduction in species richness in the first 8 h of fermentation, compared to the faeces used to inoculate the system. At this early stage of fermentation there was an increase in the lactate producing streptococcus sp. and Ligilactobacillus sp. as well as the lactate concentration which may explain the differences in the metabolic profile at this time. There were 73 differing predicted metabolic pathways at 8 h into fermentation compared to the other time points in this study. This number of differing metabolic pathways, much greater than all other comparisons observed, suggests this early stage of fermentation demonstrated the most significant change within the system. When looking at the core microbiome in the system at 8 h into fermentation there was 1 ASV unique to that time point, 1 ASV was shared with 36 h into incubation and 1 was shared with both 28 h and 36 h into incubation. This core ASV prediction supports the species richness observed in the reduced alpha diversity seen at this time point. When observing the top 10 ASVs in this experiment streptococcus was greatest at 8–20 h and Ligilactobacillus greatest at 8–36 h, whereas both genera were lowest in abundance in the faeces and 0 h into fermentation. These data suggest a dysbiosis occurred within the GP system, as these lactate producing bacteria proliferated in the system during the early part of fermentation, whereas these microbiotas were in lower prevalence within the donor faecal samples.

The brief lag phase of GP may also be linked to the rapidly degradable fraction of the diet available for fermentation. The diet contained a fibre content of approximately 85%, the remaining 15% was made up of; protein (7.4%), NSC (11.4%) and minerals suggesting limited non-structural carbohydrate for rapid fermentation, this may be attributed to the metabolic changes in the GP system at this time. In the first 8 h of fermentation fibre degrading microbiota were reduced in abundance and replaced with saccharolytic degrading microbiota. Cellulolytic bacteria, Fibrobacters, are typically obligate anaerobes whereas amylolytic bacteria, e.g., streptococcus, are facultative anaerobes. It is plausible that during the process of GP setup, even when keeping the process as anaerobic as possible the presence of even small amounts of oxygen may have influenced the systems microbiome, or alternatively a lack of available rapidly degradable substrate. Sileshi et al.¹⁹ identified during GP inoculation the presence of oxygen would likely reduce cellulolytic microbiota (obligate anaerobes) as opposed to amylolytic (facultative anaerobes), thereby reducing fibre digestion and increasing lag time. The France et al.⁷ model predicted a lag time of 56 min in this present study, however at 8 h into fermentation the ml/gas/hr was the lowest of all of the early time points (0–50 h) measured in the experiment suggesting fermentation was affected by the bacterial community profile at that time point. As the lactate concentration in the fermentation system was highest at 8 h into fermentation this may also be linked to the lowest abundance of key fibre degraders at this time point, as lactic acid will have lowered the pH in the inoculum and Fibrobacters are particularly pH sensitive. The metabolic pathway predictions also highlighted an increase in the lactate dehydrogenase pathway at 8 h into fermentation. Previous studies^12,20 have shown that the pH in the gas production system reaches the lowest within 24 h of fermentation where NSC from starch based complementary feed is present, after this time the system appears to buffer the pH which may also be linked to increased fibre fermentation at this time. Alongside the increase in lactate concentration, acetate, butyrate and propionate concentrations all decreased when comparing 0 to 8 h of fermentation. Thus, the lag phase of in vitro gas production could be partly attributed to dysbiosis in the newly inoculated system, linked to microbial turnover and oxygen sensitivity of the fibre degrading population as well as the amylolytic population making use of the NSC in the substrate. The species richness noted between 8 and 20 h, the change in lactate and succinate concentrations and metabolic pathway activity may all suggest an element of metabolic stress²¹ which has been associated with dysbiosis in human and animal models²².

There were no differences in metabolite concentrations between 36 and 156 h where gas production was going into the diminishing phase. When comparing the final 156 h inoculum to the starting faeces the metabolic profile showed that for each of the short chain fatty acids and amino acids present there was a greater concentration at 156 h than in the faeces and this is likely to be due to the nature of a closed batch culture system where the concentration of these metabolites will build up in the system over time as they cannot be removed.

Considering the fibre degraders in the bacterial community profile over time, Ruminobacter typically found in the rumen were present in the system between 20 and 36 h but had a very low prevalence in the faecal samples and a low prevalence by 156 h of fermentation. It is plausible that the media for the GPS which was designed to be optimal for ruminants supported the proliferation of these microbiota within the GPS. Similarly Bacteroides, typically fibre degraders were represented in higher prevalence in the batch culture fermentation system than in the faecal samples. The viability of major gut colonizers, Bacteriodetes and Fibrobacters, over the incubation period suggest that the fibre substrate was degraded in the fermentation system to an equivalent dry matter disappearance to feeding the same diet in vivo. However, the fibre degrading microbiota profile in vitro may not be the same as those degrading the substrate in the horse’s hindgut. The in vitro GP system derived to test ruminant feedstuffs may favour microbiota common to the rumen and thus be less favourable to those specific to the hindgut of the horse. This could be linked to the type of media or that these microbiota are more robust in a batch culture system. However, many of the fibre degrading ASVs in the faeces used to inoculate the system did increase in proportion by 156 h of fermentation, suggesting the GPS microbiome took time to restabilise after the early dysbiosis. In the multi vessel model of Leng et al.¹⁴ Spirochaetes were a phyla that were not supported within the system, this was not a finding in our present study as Spirochaetes were observed increasing in abundance in the core microbiome from 20 h into fermentation. The media used in the multi vessel model of Leng et al.¹⁴ differs to that used in the gas production system, by being more focused on an equid model, whereas gas production is based upon a ruminant model. It is plausible that either the multi vessel construction or the media used in the model did not support this phyla. However, in our present study spirochaetes reappeared gradually from 20 to 156 h of fermentation suggesting that these phyla of acetate producers can be supported within the gas production system.

Further consideration of the structure and function of the gas production system for equine faecal inoculum needs to be considered if the system is to accurately represent in vivo digestion and microbiome function in the horse. Previous studies have used the in vitro gas production system as a model to predict the effects of diet or other interventions on fermentation and metabolic profile as a model for the horses’ hindgut^11–13,20. It seems likely that functional redundancy²³ occurs within the system, as in the present study the overall dry matter disappearance was equivalent to the in vivo value and therefore functionally it seems to be representative. Some of the proportional differences in the metabolic profile could be attributed to the use of a closed system compared to an ‘open’ system of metabolites excreted in faeces. Mechanistically dry matter disappearance and VFA production occurs proportionately to in vivo digestibility when using in vitro gas production. However, some caution should be applied when extrapolating the effects of diet on the structure and function of the microbiome of the in vitro system and predicting the same will occur in the equine gut microbiome, as the specific microbiota pathways involved may differ. Our findings support the use of in vitro gas production as a method of estimating digestibility in equids and support the work that has been undertaken in this area to date. However there remains a gap in the knowledge of how accurately in vitro models for equids reflect in vivo digestibility of diets, further studies are required to validate such models going forwards.

There were some limitations to this study that should be considered when extrapolating the findings. This in vitro study used a combination of amplicon sequencing, predicted metabolic pathways and metabonomics to elucidate the structure and function of the microbiome in this batch culture system. Repeating this using shotgun sequencing would confirm the findings. The diet digested in vitro in this study had not undergone an in vitro pre-caecal digestion. Most gas production studies incubate raw substrate and this is more typical of ruminant models, whereas the horse is a hindgut fermenter and enzymatically digests feed substrate prior to fermentation. Future work needs to consider if undertaking a pre-caecal digestion in vitro alters the bacterial community structure and function in the gas production system. Notwithstanding these limitations we believe that this study provides new insights into the bacterial community structure and function within the in vitro gas production system using equine faeces as a source of inoculum.

As the equine sector faces the need for changes to feed ingredients aligned to improving sustainability and in response to the effect of climatic change on feedstuff availability, the in vitro gas production system provides an economic tool to evaluate the fermentation characteristics of feedstuffs and their suitability as horse feeds. In this study the metabolic profile of both the faeces and the GP system suggest both produce the same metabolites, suggesting the gas production system can provide an acceptable in vitro model to study equine hindgut fermentation. Therefore, GP enables both the modelling of fermentation profile and the metabolic profiling of feedstuffs, prior to needing to trial potentially valuable substrates in animal trials. Using the in vitro GP system allows a pre-screening for digestibility, reducing the number of in vivo animal trials required when testing new diets. In vivo trials would still be required to check palatability of feedstuffs and to be able to accurately measure rate of digesta passage of those substrates that were favourable after in vitro screening. The differences in the fibre degraders surviving in the system may be due to the setup, the media, or a functional redundancy. Future studies should consider if an equid specific media is required for in vitro gas production to replicate a microbiome environment more similar to that of the horse’s hindgut rather than that of the rumen for in vitro gas production studies.

Method

Ethics declaration

This project was approved by the Royal Agricultural University research ethics committee, 20204703-Daniels for both the in vitro and in vivo elements. The in vivo element of this study has been reported in line with the ARRIVE guidelines, all methods were conduced in line with the Royal Agricultural University Non-ASPA policy.

Animals

Six Welsh section A geldings, aged 9 years fed a diet of 90% hay and 10% proprietary complementary feed (PMix), Table 1, were recruited as faecal donors. In vivo digestibility data was available for this diet from these ponies from the same time period from a separate study which included a five-day, total faecal collection digestibility trial. At the time of this study ponies had been fed this diet for 12 weeks and therefore fully adapted to these feed substrates. This diet provided 100% of the ponies feed ration during this time. They were stabled on shavings and spent social/exercise time together daily in a drylot.

Table 1.

Diet composition of the diet fed to the donor ponies and used in the gas production system.

	DM (g/kg)	ADF (g/kg)	NDF (g/kg)	Cellulose (g/kg)	Hemi-cellulose (g/kg)	ADL (g/kg)	CP (g/kg)	NSC (g/kg)	OM (g/Kg)	Ash (g/kg)
Hay	957.9	359.62	564.59	250.06	204.96	53.38	69.97	121.06	943.76	56.24
PMix	898.5	294.71	418.25	254.67	123.54	40.04	114.60	51.48	930.43	69.57
Total Diet	952.0	353.13	549.96	250.52	196.82	52.05	74.43	114.10	942.43	57.57

PMix – proprietary complementary feed mix, DM – dry matter, ADF – acid detergent fibre, NDF- neutral detergent fibre, ADL – acid detergent lignin, CP – crude protein, NSC – non-structural carbohydrate (starch + water soluble carbohydrate), OM – organic matter.

Faecal sampling

Faecal samples were collected into cryo-vials in sextuple on the day of the experiment. Fresh spontaneously voided faecal samples were collected between 06:00–08:00 h into pre warmed thermos flasks for GP, samples were collected in sextuple into cryo-vials from the same faecal pile transported to the laboratory on ice for microbiome and metabonome analysis. All samples in cryovials were placed immediately on ice and were frozen at −80 °C within four hours of collection.

On each return to the laboratory for all samples collected, three of the replicate samples of faeces from each pony were divided into two, one triplicate was frozen straight away at −80 °C and the second triplicate were treated with propidium monoazide (PMAxx) prior to freezing at −80 °C.

Propidium monoazide treatment

For viability analysis of microbiota in the faeces, feed and inoculum PMAxx was used. In a dark room 1 µl of PMAxx was added to 400 µl of inoculum. For faeces and feed 0.3 g was added to 1 ml ddH₂O vortexed for 10 min then centrifuged for 15 min and 10 000 X g and 400 µl of supernatant was treated with PMAxx. Once PMAxx was added to vials these were incubated in the dark on a rocker for 10 min. Microfuge tubes were then fluoresced on a Glo-Plate (Biotium) for 20 min. After light treatment microfuge tubes were stored at −80 °C prior to analysis.

In vitro gas production

The manual in vitro gas production method of Theodorou et al.⁵ was used in this study. Prior to the main study commencing a pilot gas production experiment was run using faeces from two ponies to inoculate the system. The pilot was used to inform sampling time points in the main study to estimate time to reach the gas production asymptote and identify when background fermentation from the donor faeces ceased. Both hay and complementary feed were dried in a force drawn oven at 50 °C until reaching a constant dry weight. Feeds were milled to 1 mm sized particles (Fritsch, Pulversette 19, Germany) and 1 g of substrate, 0.9 g hay + 0.1 g of Pmix, were weighed into 125 ml serum bottles. For each pony donor, three replicate bottles with substrate were prepared to incubate throughout the gas production experiment plus three blank bottles with no substrate. For each pony donor a further eight bottles with substrate were also prepared to stop and sample at 8, 20, 28, 36 h into incubation. In total 78 bottles were inoculated as described by Theodorou et al.⁵. with media and faecal inoculum from the donor ponies. Inoculum were made up using a 1:3 ratio of faeces to van Soest media. Bottles were inoculated under a stream of CO₂ and sealed with crimp top rubber seals. Once all of the bottles were inoculated ambient pressure in each bottle was adjusted to zero on the pressure transducer and bottles were incubated at 39 °C as described by Lowman et al.⁶.

Gas accumulation was recorded at 0, 4, 8, 12, 20, 28, 36, 45, 53, 70, 95, 107, 131 and 156 h following incubation by recording the headspace pressure and the ml of gas produced within the headspace. On reading the gas was removed and discarded, reducing the ambient bottle pressure to zero at each reading.

After 156 h all bottles were stopped, remaining contents of each bottle were filtered using pre-weighed Whatman No.1 filter papers and dried in a force drawn oven at 60 °C until reaching a constant weight. Dry matter loss was calculated and used in mathematical modelling for the gas production kinetics. Initially volumes were corrected against pressure by linear regression as described Theodorou et al.⁶. Curves were fitted to the cumulative gas production profiles, maximum likelihood (MLP) using the France et al.⁷ quadratic equation was used to generate fermentation kinetics.

In vitro system sampling

The inocula in the system from each donor pony was sampled throughout the experiment from the harvested replicate bottles to allow omics analysis at set time points during incubation. Samples were taken from the inoculum representing each donor pony at the time the bottles were inoculated to represent time zero. Harvested replicate bottles were stopped and sampled at 8, 20, 28, 36 h and the final bottles at 156 h. At each of these time points samples of filtered inoculum were taken from the replicate bottles in triplicate, six per pony, for 16 S rRNA sequencing and 1^H NMR metabolomics. For bacterial community profiling samples were divided into two, half of the replicates were frozen immediately at −80 °C and half treated with PMAxx prior to freezing at −80 °C.

Bacterial community profiling

Bacterial DNA were extracted using the DNeasy Power soil Pro^® kit, following the manufacturer’s instructions. DNA presence was confirmed in all samples by gel electrophoresis. Pre-sequencing, aliquots of extracted DNA were amplified with universal primers for the V4 and V5 regions of the 16S rRNA gene using primers U515F (5’-GTGYCAGCMGCCGCGGTA) and U927R (5’-CCCGYCAATTCMTTTRAGT)²⁴. Amplicons were purified using 0.8 volumes of Ampure XP magnetic beads (Beckman Coulter). Each sample was then tagged with a unique pair of indices and the sequencing primer, using Nextera XT v2 Index kits, and 2x KAPA HiFi HotStart ReadyMix using the following cycling conditions: 95 °C for 3 min; 10 cycles of 95 °C for 30 s, 55 °C for 30 s, 72 °C for 30 s; followed by 72 °C for 5 min. Index-tagged amplicons were purified using 0.8 volumes of Ampure XP magnetic beads (Beckman Coulter). The concentration of each sample was measured using the fluorescence-based Quantifluor assay (Promega). Concentrations were normalized before pooling all samples, each of which would be subsequently identified by its unique index combination. Sequencing was performed on an Illumina MiSeq with 2 × 300 base reads according to the manufacturer’s instructions (Illumina Cambridge UK) by the Animal and Plant Health Agency. Bioinformatics were conducted in Qiime2²⁵ version 2022.08, following the Qiime2 workflow (qiime2.org). Raw reads were paired and quality checked and denoised using DADA2²⁶. The Qiime2 pipeline was used for downstream processing, the Microbiome Analyst pipeline was used for alpha and beta diversity analysis and to identify core features. Qiime2 was used to determine the core microbiome at genus based upon prevalence where ASVs were present in 75% of samples at each sampling time point. For taxonomy assignment the SILVA database (version 138.1) was referenced using RESCRIPt²⁷ using the primers to create an amplicon specific classifier. From the Qiime2 feature table PICRUSt2²⁸ was used to predict metabolic pathways. Differences between sampling points and metabolic pathways were analysed using LEfSe²⁹ within the Microbiome Analyst platform³⁰. For bioinformatics the biom tables were not rarefied³¹.

Metabolic profiling

Faecal samples were defrosted and 0.3 g of the samples were placed into a microfuge tube with 1.2 ml of phosphate buffer. 1.7 mm zirconia beads were also inserted in the tubes before placing them into a bead breaker for 10 min to ensure lysis. Tubes were then placed in the centrifuge for 10 min at 16 000 x g. Centrifugation was repeated twice more, with supernatants being transferred into new tubes each time. Faecal inoculum samples were removed from the freezer and defrosted, then vortexed. An aliquot of 400 µl was deposited into microfuge tubes with 200 µl of the premixed phosphate buffer. Samples were once again vortexed and then placed into a centrifuge at 10 000 x g for 10 min. 550 µl of supernatant was then placed into 5 mm internal diameter NMR tubes using a pipette. The samples were analysed on a Bruker Avance NEO 600 MHz equipped with a TCI Cryoprobe Prodigy (Bruker). Spectra were acquired at 298 K and consisted for each sample of a ¹H Carr-Purcell-Meiboom-Gill (CPMG) spectrum, with 64 scans, an acquisition time of 2.62s, a relaxation time of 4 s, a total spin echo delay of 30 ms and a spectral width of 20.8 ppm. After acquisition, the spectra were then phase corrected, baseline corrected, and the chemical shifts were referenced to the TrimethylSilylPropionate (TSP) peak at 0.0 ppm. For metabolite identification, 2D spectra were acquired, consisting of a Total Correlation SpectroscopY (TOCSY), using the Bruker pulse sequence “dipsi2gpphzs”, slightly modified to include presaturation, which 8 scans, 256 t₁ increments, a spectral width of 13.7 ppm in both dimensions and a relaxation delay of 2 s, along with a Heteronuclear Single-Quantum Correlation (HSQC) spectrum, using the Bruker pulse sequence “hsqcetgpsisp2.2”, with 16 scans, 256 t₁ increments, a spectral width of 210 ppm in the ¹³C dimension and 20.8 ppm in the ¹H dimension and a relaxation time of 1.75 s. Alignment and normalisation (total quotient method) of spectra was carried out in Mathworks Matlab R2014a. Samples were analysed using multivariate statistics, initially unsupervised principal component analysis (PCA) models were constructed. Where there was spatial clustering between groups, supervised models were constructed using orthogonal projections to latent-structures-discriminate analysis (OPLS-DA). OPLS-DA models were used to identify group-specific metabolic changes. Only OPLS-DA models with Q²Y values > 0.5 and metabolite abundances differences R² > 0.5 are reported. It is possible with these complex models that over-fitting occurs, where too many latent variables are added reducing the predictive ability and creating bias within the model³². To ensure models were not over-fitting OPLS-DA models underwent permutation testing, to 1000 permutations, to ascertain the significance by comparing the permutated model to the original OPLS-DA model. Only models with good prediction ability and no over-fitting (p < 0.01) were kept. Where models were kept and differing metabolites were observed concentrations were predicted by measuring the peak area and number of protons against that and the concentration of the TSP singlet. Differences between metabolites in pairwise comparisons were initially checked for normal distribution by Shapiro wilk and where normally distributed analysed by T-test, any comparisons not meeting the assumptions of T-test were analysed by Mann-Whitney U test in Genstat 22nd Ed.

Relationships between ASVs and metabolites

To explore relationships ASVs differing over the time points in the in vitro system and metabolites differing at the same time points were correlated using spearman’s rank correlation.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

S.D, M.MC and P.H were all involved with the conceptualization and study design. S.D, S.M and M.MC undertook the laboratory work and collected the data. S.D undertook all of the bioinformatics, metabolic profiling, mathematical modelling and analysed the data. S.D, M.MC and P.H drafted the manuscript, all authors were involved in editing and finalising the manuscript.

The authors would like to acknowledge the European Society of Veterinary and Comparative Nutrition for supporting this work.

Data availability

Amplicon sequencing data have been uploaded to the European Nucleotide Archive under accession PRJEB94521. Bioinformatics scripts for 16 S rRNA sequences and metaboomics can be found here: https://github.com/simond83/Bio-Scripts/pull/1/commits/0bdd321d8db24e4b6b3f59740348b3cb39c9547a. Other data are available from the corresponding author on request.

Declarations

Competing interests

The authors declare no competing interests.