All animal procedures were carried out according to the Animals (Scientific Procedures) Act 1986 (PLL 40/3653; PIL 40/9798) in accordance with the guidelines of the European Directive 2010/63/EU and after approval by the Aberystwyth University’s Internal Ethical Review Panel.

Rumen fluid was collected separately from four dairy cows fed a diet composed of perennial ryegrass hay and concentrate at 67:33 on DM basis via permanently established rumen cannulae, and strained through a double layer of muslin and stored under CO₂ at 39°C (see Table S1 in Supplementary Material for chemical composition of experimental diets). Rumen fluid was diluted 2:1 with incubation solution (Theodorou et al., 1994) under CO₂ and 50 mL of buffered rumen fluid was incubated with 500 mg of a mixed diet (30:70 ground barley/alfalfa mix ground to pass through a 1 mm² sieve). General batch incubation procedures were all performed following (Belanche et al., 2016).

Fresh fecal material was collected separately from four horses the morning of the experiment, kept warm and mixed with buffer solution (Theodorou et al., 1994) under CO₂ at 39°C (1:2 w/v). Then, the mix was strained through a double layer of muslin and 50 mL was incubated with 500 mg of a pre-digested substrate as described below.

Unflavored kefir was produced by The Chuckling Goat LTD (Llandysul, Wales, SA44 6DS) from raw goat milk containing 2.9% fat. Samples were kept on ice until use. Samples (500 mL) of kefir to be pasteurized were aseptically transferred to a glass Erlenmeyer flask (2 L), heated for 30 min at 62.5°C and then cooled rapidly in water and ice. Samples (500 mL) of kefir to be autoclaved were aseptically transferred to a glass Erlenmeyer flask (2 L), heated for 20 min at 121°C and then cooled rapidly in water and ice.

All incubations were run in duplicate for each animal. Kefir was added at a concentration of 10 mL per L of incubation (0.5 mL per bottle) prior to the addition of buffered rumen fluid or equine digesta, according to the following treatments: CTR (no added kefir), KEF (unaltered kefir), AUT (autoclaved kefir), and PAS (pasteurized kefir). All incubations were in 120 mL Wheaton bottles under CO₂ sealed with an airtight seal, and incubated at 39°C. Blanks (without substrate) of the four treatments (CTR, KEF, AUT, and PAS) were also included in the incubation (a summary of the treatments in shown in Table 1). Samples of buffered rumen/horse fluid at inoculation and kefir used as inoculate were retained to determine initial LAB, AAB, and yeast levels (see Figure 1) as explained below.

Gas production was measured at 2, 4, 6, 8, 12, 24, 48, and 72 h of incubation using a TP704 Manometer (DELTA OHM, Italy). To convert pressure to volume we used a linear regression between pressure and known air volumes at standard incubation temperatures. After 24 h of incubation a sample (10% v/v) of liquid incubation media was retained for determining volatile fatty acid (VFA) (1.6 mL of sample in 0.4 mL of 20% orthophosphoric containing 20 mM 2-ethylbutric acid as internal standard), ammonia (0.8 mL of sample in 0.2 mL of 25% TCA) by gas liquid chromatography (Jouany, 1982) and a phenol hypochlorite method (Weatherburn, 1967), respectively; pH was also recorded after 24 h incubation. Concentrations of total lactate were measured using the EnzytecTM D/L-Lactic Acid kit (r-biopharm Rhone, Ltd, Glasgow).

Lactic acid bacteria counts from samples after 24 h of incubation were performed on MRS medium (pH 6.5) from Thermo Fisher (De Man et al., 1960) and incubated at 30°C for 3 days. Plates were overlaid to create a microaerophilic environment. To inhibit yeast growth pimaricin (100 mg/l) was added to the media. Yeasts and molds were grown on OGYE medium (Oxytetracycline-Glucose-Yeast Extract Agar, pH 7.6; Thermo Fisher), supplemented with Oxytetracycline GYE Selective supplement and incubated at 25°C for 7 days. Acetic acid bacterial were counted on the medium described by Guillamón (2000) containing 5% glucose, 1% yeast extract, and 2% agar. To inhibit yeast and LAB growth pimaricin (100 mg/l) and penicillin (3 μg/mL) were added to the media. Plates were overlaid to create a microaerophilic environment and incubated at 25°C for 2 days.

An exponential model was used (Ørskov and McDonald, 1979) to describe the kinetics of the accumulated gas production profiles:

Where “Y” is the accumulated gas produced over time in mL; “t” is time in hours; “a” and “b” parameters are used to explain the potential of fermentation (“a + b” reflects the maximum potential of fermentation), and “c” parameter is the gas production rate constant, used to explain the speed of fermentation.

Kefir extracts effects were analyzed with JMP Pro 11 (SAS, 2013) following a one-way ANOVA, considering number of bottles as replicates and “treatment” as main factor. Moreover, to evaluate the indirect influence of kefir on the microbial population in a second analysis, values observed in the blanks were subtracted to each treatment and analyzed again following a one-way ANOVA. Means were compared with a Tukey test at P < 0.05. P between 0.1 and 0.05 were considered trends.

Multivariate analyses, including Permutational Analysis of variance (Adonis) and Canonical Correspondence Analysis, were performed using the package “vegan” from the R statistical program (R Core Team, 2014); cluster analysis and analysis of the dispersion of the multivariate data were conducted calculating the similarity distances by the “Bray–Curtis” method, also from the package “vegan” in R. Red circumferences were displayed assuming a confidence interval of 95%.

Results obtained after subtracting the specific blank to the respective incubation are summarized in Tables 4, 5, 7, Supplementary Tables S4, S5, and Figures 4, 5B. A complete definition of terms CTRB, PASB, AUTB, and KEFB is explained in Table 1. The most obvious differences observed in the foregut model were the alteration in the acetate and propionate proportions, and thus the C2:C3 ratio (Table 4). Molar proportion of propionate increased more in CTRB than in kefir-based treatments. Moreover, the increase of molar proportion of acetate was higher in PASB than in AUTB and KEFB (P < 0.05). Neither ammonia nor lactate concentrations were modified by the kefir-based treatments compared to the control (P > 0.05).

Permutational ANOVA applied to the dataset indicated that samples did not group separately (P = 0.513), most probably due to the higher variability observed within samples from the same treatment (Figure 4A). The same effect was then confirmed in the pairwise comparison (Table 7), in which no treatment showed statistical differences between them.

Results obtained in the hindgut model also followed a similar trend to those from the foregut model, in which no major differences were observed between treatment, apart from a decrease in pH (P < 0.05) in CTRB samples and a decrease in the speed of fermentation (coefficient “c”) in kefir-based treatments.

Permutational ANOVA indicated no differences due to treatment (P = 0.572), confirming the pairwise comparison among treatments (Table 7).

Comparative analysis between both fermentative models (Foregut vs. hindgut) is shown in Figures 5, 6. Multivariate analysis showed a clear fermentation model effect (P < 0.001) in both direct (Figure 5A) and differential (Figure 5B) analysis. Moreover, to study the variance in response from the two models (i.e., dispersion), a simple ANOVA was performed on the data generated by the distance to the centroids in the two fermentative models (see Figure 3B). As can be seen in Figure 5, both models grouped separately for both situations (direct effect and “differential” effect), and the dispersion (measured as the distance to the centroid) in the hindgut model was higher in both situations, becoming significant when the “differential” effect was studied (Figure 6B, P = 0.009).

The effect of kefir on digestive systems has been described by several authors in terms of anti-inflammatory, anti-allergic (Lee et al., 2007) and probiotic effect (Farnworth, 2006). It is known that kefir can interact with gut microbiota (Carasi et al., 2015) and control pathogenic infections such as Clostridium difficile (Bolla et al., 2013). In this study we investigated the action of kefir using two gut models: a foregut model (RUM) that represents ruminant fermentation, and hindgut model (HOR), that represents the non-ruminant herbivore species. To achieve this objective, we used rumen fluid as an inoculate for the foregut model and horse fecal material for the hindgut model. The microbial community present in both systems is known to behave differently in terms of the nutritional ecology of the host (Van Soest, 1994). The rumen has a major role in stabilizing host nutrition (Newbold et al., 2001) and its capability as detoxifier has been reviewed (Upadhaya et al., 2010). On the other hand, horse hindgut microbiota is highly sensitive to diet changes and certain diet related pathologies, like lameness have been associated with a lower core microbial population in the hindgut (Steelman et al., 2012). This study aimed to use the kefir as an agent to challenge these two distinctive models (one robust and other fragile) and induce an alteration in the fermentation pattern. The dose used in this study was chosen based on preliminary results obtained in similar incubation conditions, tested in both fermentation models and following a dose-response analysis (data not shown). No predigestion of kefir was performed in the hindgut model, because the objective of this study was to observe the effect of the same type of disturbance in both fermentation models, and hence, the predigestion of kefir would indubitably have modified the original properties of this additive. In order to apply this knowledge in practical situations, it would be needed to include a dose-response study at in vivo conditions, since the feeding of kefir would imply its digestion prior to reaching the hindgut.

Kefir changed the fermentation pattern in our foregut model (Table 2) in terms of metabolite concentration, C2:C3 ratio and the potential extent of fermentation (“a+b” parameter). These results can be explained by the metabolite composition of kefir and its probiotic effect, which could have provided extra sources of energy and protein to the microbiota (Farnworth, 2006). The effects are probably due to the combination of both microflora in the kefir and its ingredients/nutrients that boost the fermentation. In general, a greater increase compared to the control in most of the analyzed metabolites was observed when pasteurized kefir was used (Table 2), suggesting that most of this effect may have been caused by the nutrients provided by the kefir and not by the probiotic microflora. Most of metabolites concentrations did not differ between pasteurized and untreated kefir. Autoclaved kefir resulted in a lower concentration of metabolites compared to the other two sources of kefir. Autoclaving is an aggressive method of sterilization that can lead to protein denaturalization and formation of poorly fermentable material (Parsons et al., 1992). It seems that part of the extra substrate provided by the kefir became unavailable after its sterilization by autoclaving.

A similar pattern was observed in the hindgut model, in which most of the major metabolites increased when kefir was included in the incubation. Moreover, a shift toward propionate fermentation occurred, resulting in higher proportions of propionate and lower of acetate and C2:C3 ratio when compared to the control. A higher concentration of LAB in those incubations could potentially have promoted an increase in the production of propionate, and hence a shift in the fermentation pattern.

It’s noticeable than both lactic and AAB concentrations in PAS treatment were numerically higher than in KEF, suggesting that after 24 h bacteria belonged to the kefir inoculum did not survive or at least reached an equilibrium with the microbiota present at the incubations. Moreover, a higher concentration of fermentable metabolites and in particular lactic acid could have promoted a lower pH and a change in the fermentation pattern. Pearson’s negative correlations between lactate concentration and either pH or C2:C3 ratio confirmed this process (r = -0.594, P = 0.01 for pH and r = -0.662, P = 0.005 for C2:C3 ratio). Although it appears that positive effects are promoted by Kefir in rumen and hindgut models, long term studies are needed to evaluate the possible transience of kefir effects over time on gut microbial communities.

In this study we present two types of information in each of the two gut models. First (Tables 2, 3) we tried to assess the overall effect of the inclusion of kefir, either unaltered or after autoclaving or pasteurization on the fermentation pattern in both foregut and hindgut models. Secondly, we aimed to assess the indirect effect of kefir on the microbial population, by studying the changes in the fermentation of the substrate when incubated with kefir. This “differential” effect was calculated by subtracting the values obtained in the specific blanks from the correspondent treatments (Table 1). Thus, differences among the treatments in the values obtained after subtracting their respective blanks could indicate any effect not directly related to fermentation of the kefir but indirectly via effects on microbial activity (Tables 4, 5). This is a novel approach that could potentially identify changes in the microbial structure. However, some of the results must be interpreted with care, since efficiency of utilization of a compound can change in presence of additional substrates (Hall and Weimer, 2007; Rezaii et al., 2010), and with this approach we are assuming a similar efficiency of utilization between samples.

In general no major differences were observed after including this correction, apart from changes in the molar proportions of acetic and propionic (P < 0.01). The production of total VFA was also numerically higher in kefir based treatments and in the potential of fermentation (“a+b”), suggesting a more efficient microbiota in using fermentable material. However, the high variability made impossible to find statistical differences between treatments.

The differential approach performed in the hindgut model showed even less significant differences than those observed in the foregut model (Table 5). Total VFA production was affected in the opposite way to that observed in the foregut model (P < 0.1), suggesting that inclusion of pasteurized kefir can indirectly reduce the efficiency of substrate utilization. The pH decrease due to the substrate fermentation was more acute in bottles with either live or pasteurized kefir; moreover, the speed of fermentation was lower in those cases, confirming previous studies showing an inverse relation between pH and rate of fermentation (Grant and Mertens, 1992).

Composition of LAB, AAB, and yeast in both foregut and hindgut models differed (see Figure 1), presenting the latter a higher (numerical) concentration of LAB and a lower of AAB and yeast. Viable microorganisms in kefir inoculates included a major contribution of LAB and yeast. Pasteurization seemed to affect mostly LAB but not yeast numbers, and autoclaving eliminated all the viable bacteria. In this scenario, it was hypothesized that kefir in their different presentations would affect in a different way the foregut and the hindgut systems. However, after 24 h of incubation no differences were observed in microorganisms concentrations (P > 0.1 in both models, Tables 2, 3), indicating the capability of both models to react to the perturbation induced (inclusion of kefir in different formats) and to get to a new equilibrium. Resilience is generally higher in species rich environments, but the disturbance plays out in two different organizational levels: changing biodiversity can have a strong influence on the community level, but the effect on the ecosystem not necessarily is evident (Liebergesell et al., 2016).

Inclusion of kefir seemed to exert a more pronounced effect in the foregut model, increasing the concentration of most of the measured metabolites. Only with total VFA did kefir addition induce a higher change in the hindgut model. Microbial cell synthesis depends on the availability of precursors (such as simple sugars, nucleic acids, ammonia, or minerals). If bacteria are limited through energy source, total energy and growth utilization can be explained by growth and maintenance functions of the microorganisms (Russell and Cook, 1995; Russell, 2007). However, if growth is not limited by energy but by other nutrients (e.g., nitrogen), energetic uncoupling occurs, and bacteria can spill ATP in reactions that cannot be categorized as maintenance per se (Russell and Cook, 1995; Russell, 2007). In the hindgut, the substrate that reaches the hindgut highly depends on pre-cecal digestibility of feeds, which means that cytoplasmic protein (nitrogen) and soluble sugars reach the hindgut in low amounts. However, the majority of cell wall carbohydrates and linked nitrogen will reach the hindgut since limited hydrolysis of these constituents occurs either in the stomach or in the pre-cecal environment. If we consider the type of substrate that gets to the hindgut, we can expect it to be nitrogen limiting, and a situation of energetic uncoupling seems feasible. The hindgut environment would aim to favor VFAs production by increasing microbial metabolism without net microbial growth. In this study, it was observed an increase of 15.5% in the production of VFAs, compared with that observed in the foregut (5.2%). However, the increase of ammonia was only of 8.9%, indicating a lower activity related with the nitrogen catabolism and cell growth. As the substrate that reaches the hindgut will be nitrogen limiting, the environment should promote microbial activity but without net growth thus favoring VFAs production (Russell, 2007; Santos et al., 2011).

Both host diet (Tajima et al., 2001) and the relationships among groups of organisms can influence gut bacterial diversity (Ley et al., 2008), based on fecal microbial composition, found that foregut and hindgut herbivores fermenters grouped separately. Differences between foregut and hindgut environments include different viscosity due to the water content (Hecker and Grovum, 1971), pH (bile and bicarbonate buffering) (Mackie and Wilkins, 1988), particle size and VFA concentration fluctuations (Sato and Shiogama, 2010). Microbial communities altered by disturbances do not return to their original composition for after some time (Allison and Martiny, 2008). However even if the microbial composition may change, the new community that appears after the disturbance might be similar to the original in terms of metabolic activity (Weimer, 2015). Considering this, expression of changes (for instance, by shifts in the fermentation pattern) can indicate a more permanent disturbance in the microbial community structure. In this experiment, we aimed to study the effect that kefir could cause in both fermentation models and whether this effect was due directly to the fermentation of kefir inoculate or indirectly by a change in the resilience of the microbiota; permutational analysis of the gathered data indicated a distinctive response in both fermentative models to both direct or indirect effects (Figure 5), and a significant effect of the addition of kefir on the fermentation pattern. However, presence of kefir did not indirectly change in the activity of microbial population over the fermented substrate (Figure 5B, P_treatment = 0.389), but induced a more variable response in the hindgut rather than the foregut model (Figure 6). Although both models effectively reacted to the new conditions by adapting to the disturbance induced by kefir, the variability of that response, measured by the dispersion of that response throughout the different samples was not the same, suggesting that foregut model is more robust than the hindgut one, which was more variable in such response. Thus, these results suggest that hindgut model is less stable than the foregut one, and hence more prone to experiment disturbances that could be maintained over a longer period of time. Moreover, the dispersion of the samples with respect with their centroid (Figure 6B), indicated that the increase in the variability became more significant when the indirect influence of kefir was studied.