Abstract
Background The red macroalgae Asparagopsis taxiformis is a potent natural supplement for reducing methane production from cattle. A. taxiformis contains several anti-methanogenic compounds including bromoform that inhibits directly methanogenesis. The positive and adverse effects of A. taxiformis on the rumen microbiota are dose-dependent and operate in a dynamic fashion. It is therefore key to characterize the dynamic response of the rumen microbial fermentation for identifying optimal conditions on the use of A. taxiformis as a dietary supplement for methane mitigation. Accordingly, the objective of this work was to model the effect of A. taxiformis supplementation on the rumen microbial fermentation under in vitro conditions. We adapted a published mathematical model of rumen microbial fermentation to account for A. taxiformis supplementation. We modelled the impact of A. taxiformis on the fermentation and methane production by two mechanisms, namely (i) direct inhibition of the growth rate of methanogenesis by bromoform and (ii) hydrogen control on sugars utilization and on the flux distribution towards volatile fatty acids production. We calibrated our model using a multi-experiment estimation approach that integrated experimental data with six macroalgae supplementation levels from a published in vitro study assessing the dose-response impact of A. taxiformis on rumen fermentation.
Results our model captured satisfactorily the effect of A. taxiformis on the dynamic profile of rumen microbial fermentation for the six supplementation levels of A. taxiformis with an average determination coefficient of 0.88 and an average coefficient of variation of the root mean squared error of 15.2% for acetate, butyrate, propionate, ammonia and methane.
Conclusions our results indicated the potential of our model as prediction tool for assessing the impact of additives such as seaweeds on the rumen microbial fermentation and methane production in vitro. Additional dynamic data on hydrogen and bromoform are required to validate our model structure and look for model structure improvements. We are working on model extensions to account for in vivo conditions. We expect this model development can be useful to help the design of sustainable nutritional strategies promoting healthy rumen function and low environmental footprint.
1. Background
Some macroalgae (seaweeds) have the potential to be used as natural supplement for reducing methane (CH4) production from cattle (Wang et al., 2008; Dubois et al., 2013; Maia et al., 2016). This anti-methanogenic activity adds value to the nutritional and healthy promoting properties of macroalgae in livestock diets (Evans and Critchley, 2014; Makkar et al., 2016). The species of the red macroalgae Asparagopsis have proven a strong anti-methanogenic effect both in vitro (Machado et al., 2014) and in vivo (Roque et al., 2019). In particular, Asparagopsis taxiformis appears as the most potent species for methane mitigation with studies reporting a reduction in enteric methane up to 80% in sheep (Li et al., 2016) and up to 80% and 98% in beef cattle (Kinley et al., 2020; Roque et al., 2020). The anti-methanogenic power of A. taxiformis results from the action of its multiple secondary metabolites with antimicrobial activities, being bromoform the most abundant anti-methanogenic compound (Machado et al., 2016b). It should be said, however, that despite the promising anti-methanogenic capacity of bromoform, the feasibility of supplying bromoform-containing macroalgae requires a global assessment to insure safety of feeding and low environmental footprint from the algae processing (Beauchemin et al., 2020).
Bromoform is released from specialised gland cells of the macroalage (Paul et al., 2006) in to the culture medium. The mode of action of the anti-methanogenic activity of bromoform is similar to that described for bromochloromethane (Denman et al., 2007), following the mechanism suggested for halogenated hydrocarbons (Wood et al., 1968; Czerkawski and Breckenridge, 1975). Accordingly, bromoform inhibits the cobamid dependent methyl-transfer reactions that lead to methane formation. In addition to the direct effect on the methanogenesis, the antimicrobial activity of A. taxiformis impacts the fermentation profile and the structure of the rumen microbiota (Machado et al., 2018; Roque et al., 2019). Fermentation changes may have detrimental effects on animal health and productivity (Chalupa, 1977; Li et al., 2016). The positive and adverse effects of A. taxiformis on the rumen microbiota are dose-dependent (Machado et al., 2016a) and operate in a dynamic fashion. It is therefore key to characterize the dynamic response of the rumen microbial fermentation for identifying optimal conditions on the use of the A. taxiformis as a dietary supplement for methane mitigation. The development of dynamic mathematical models provides valuable tools for the assessment of feeding and mitigation strategies (Ellis et al., 2012) including developments in the manipulation of the flows of metabolic hydrogen to control rumen fermentation (Ungerfeld, 2020). Accordingly, the objective of this work was to model the effect of A. taxiformis supplementation on the dynamics of rumen microbial fermentation under in vitro conditions. We adapted a published rumen fermentation model (Muñoz-Tamayo et al., 2016) to account for the impact of A. taxiformis on rumen fermentation and methane production evaluated in vitro at six supplementation levels (Chagas et al., 2019).
2. Methods
2.1. Experimental data
Model calibration was performed using experimental data from an in vitro study assessing the dose-response impact of A. taxiformis on fermentation and methane production (Chagas et al., 2019). In such a study, in vitro fermentation was carried out with rumen inoculum from two lactating Swedish Red cows. A basal of timothy grass (Phleum pratense), rolled barley (Hordeum vulgare), and rapeseed (Brassica napus) meal in a ratio of 545:363:92 g/kg diet dry matter (DM), and composed of 94.4% organic matter (OM), 16% crude protein (CP) and 38.7% neutral detergent fiber (NDF) were weighted (1000 mg of DM) into serum bottles prior to the incubations. A. taxiformis was supplemented at six treatment levels (0, 0.06, 0.13, 0.25, 0.5, and 1.0 % of diet OM). The content of bromoform in A. taxiformis was 6.84 mg/g DM. Diet samples were incubated for 48 h in 60 ml of buffered rumen fluid (rumen fluid:buffer ratio of 1:4 by volume). The in vitro fermentation was run in a fully automated system that allow continuous recording of gas production (Ramin and Huhtanen, 2012).
Methane production, acetate, butyrate, propionate, and ammonia were measured along the fermentation. Methane was measured at 0, 2, 4, 8, 24, 36 and 48 h according to (Ramin and Huhtanen, 2012). The volatile fatty acids (VFAs) were measured at 0, 8, 24 and 48 h. Ammonia was measured at 0 and 24h. For model calibration, we only considered data until 24h, since microbial fermentation stopped around this time.
2.2 Mathematical modelling
We adapted the mathematical model of in vitro rumen fermentation developed by (Muñoz-Tamayo et al., 2016) to account for the effect of A. taxiformis on the fermentation. This model represents the rumen microbiota by four microbial functional groups (sugar utilisers, amino acid utilisers and methanogens). Hexose monomers are represented by glucose and amino acids are represented by an average amino acid. The model is an aggregated representation of the anaerobic digestion process that comprises the hydrolysis of cell wall carbohydrates (NDF - Neutral Detergent Fiber), non-fiber carbohydrates (NSC – Non Structural Carbohydrates) and proteins, the fermentation of soluble monomers producing the VFAs acetate, butyrate, propionate, and the hydrogenotrophic methanonogenesis. Figure 1 displays a schematic representation of the rumen fermentation model indicating the effect of A. taxiformis on the fermentation.
The model is derived from mass balance equations. It is described in compact way as follows
Where ξ is the vector of state variables (metabolites), ρ(·) is a vector function with the kinetic rates of hydrolysis and substrate (sugars, amino acids, hydrogen) utilization. Hydrolysis rates are described by first-order kinetics. Substrate utilization rates are described by the Monod kinetics. S is the stoichiometry matrix containing the yield factors (Yi,j) of each metabolite (i) for each reaction (j), g(·) is a vector function with the equations representing transport phenomena (liquid-gas transfer), and p is the vector of the model parameters. The original model has 18 state variables (compartments in Figure. 1) and was implemented in Matlab (the code is accessible at https://doi.org/10.5281/zenodo.4047640). An implementation in R software is also available (Kettle et al., 2018). In the present work, we incorporated an additional state variable to represent the dynamics of bromoform concentration. The original model was extended to account for the impact of A. taxiformis on the rumen fermentation. While the original model predicts the pH, we set the pH value to 6.6.
The impact of A. taxiformis on the fermentation and methane production was ascribed to two mechanisms, namely the (i) direct inhibition of the growth rate of methanogenesis by bromoform and (ii) hydrogen control on sugars utilization and on the flux distribution towards volatile fatty acids production. These aspects are detailed below.
For the methanogenesis, the reaction rate of hydrogen utilization ρH2, (mol/(L h)) is given by where sH2(mol/L) is the hydrogen concentration in liquid phase, xH2, (mol/L) is the concentration of hydrogen-utilizing microbes (methanogens), km,su (mol/(mol h)) is the maximum specific utilization rate constant of hydrogen and Ks,H2 (mol/L) is the Monod affinity constant of hydrogen utilization, and IIN is a nitrogen limitation factor. The kinetic rate is inhibited by the anti-methanogenic compounds of A. taxiformis. The factor Ibr represents this inhibition as function of the bromoform concentration. We used the following sigmoid function to describe Ibr where sbr is the bromoform concentration (g/L) and p1, p2 are the parameters of the sigmoid function. We included in our model the dynamics of bromoform using a first-order kinetics to take into account that the inhibition of A. taxiformis declines on time as a result of the consumption of anti-methanogenic compounds (Kinley et al., 2016). The dynamics of sbr is where kbr (1/h) is the kinetic rate constant of bromoform utilization.
With regard to sugars utilization, we assumed that the effect of A. taxiformis is ascribed to hydrogen control due to accumulation of hydrogen resulting from the methanogenesis inhibition. Hydrogen level influences the fermentation pattern (Janssen, 2010). We used the structure proposed by (Mosey, 1983) to account for hydrogen control on sugar utilization and flux distribution. However, we used different parametric functions to those proposed by (Mosey, 1983). The functions proposed by (Mosey, 1983) did not provide satisfactory results.
In our model, the kinetic rate of sugar utilization is described by where ssu (mol/L) is the concentration of sugars, xsu (mol/L), is the concentration of sugar utilizers microbes, (km,su (mol/(mol h)) is the maximum specific utilization rate constant of sugars and Ks,su (mol/L) is the Monod affinity constant of sugars utilization. The factor IH2 describes the hydrogen inhibition: with zH2 the hydrogen partial pressure (pH2).
In our model, the rumen fermentation is represented by the macroscopic reactions in Table 1.
Table 1 shows that VFA production from glucose utilization occurs via reactions R1-R3. The pattern of the fermentation is determined by the flux distribution of glucose utilization through these three reactions. We denote λk as the molar fraction of the sugars utilized via reaction k. It follows that λ1 + λ2 + λ3 = 1.
The fermentation pattern (represented in our model by the flux distribution parameters λk) is controlled by thermodynamic conditions and by electron-mediating cofactors such as nicotinamide adenine dinucleotide (NAD) that drive anaerobic metabolism via the transfer of electrons in metabolic redox reactions (Mosey, 1983; Hoelzle et al., 2014; van Lingen et al., 2019). In the present study, the regulation of flux distribution was set to be dependent on the hydrogen partial pressure following the work of (Mosey, 1983; Costello et al., 1991). The flux distribution parameters were represented by the following sigmoid functions:
2.3 Parameter estimation
We used the maximum likelihood estimator that minimizes the following objective function
Where p is the vector of parameters to be estimated, ny is the number of measured variables, ntk is the number of observation times of the variable k, tik is the ith measurement time for the variable yk, and ymk is the value predicted by the model. The measured variables are the concentrations of acetate, butyrate, propionate, NH3, and the moles of methane produced.
We used the IDEAS Matlab® (Muñoz-Tamayo et al., 2009) (freely available at http://genome.jouy.inra.fr/logiciels/IDEAS) to generate the function files for solving the optimization problem locally. Then, we used the generated files by IDEAS to look for global optimal solutions using the Matlab optimization toolbox MEIGO (Egea et al., 2014) that implements the enhanced scatter search method developed by (Egea et al., 2010) for global optimization.
We reduced substantially the number of parameters to be estimated by setting most of the model parameters to the values reported in the original model implementation and using the information obtained from the in vitro study (Chagas et al., 2019). For example, the hydrolysis rate constant for NDF was obtained from (Chagas et al., 2019) whereas the hydrolysis rate constants of NSC (khydr,nsc) and proteins (khydr,pro) were included in the parameter estimation problem. The kinetic rate constant for hydrogen utilization km,H2, was set 16 mol/(mol h) using an average value of the values we obtained for the predominant archaea Methanobrevibacter ruminantium and Methanobrevibacter smithii (Muñoz-Tamayo et al., 2019) using a microbial yield factor of 0.006 mol biomass/mol H2 (Pavlostathis et al., 1990). With this strategy, we penalize the goodness-of-fit of the model. But, on the other hand, we reduce practical identifiability problems typically found when calibrating biological kinetic-based models (Vanrolleghem et al., 1995). The parameter vector for the estimation is then p: {khydr,nsc, khydr,pro, kbr, p1, p2, p9}. The optimization was set in a multi-experiment fitting context that integrates the data of all treatments. To evaluate the model performance, we computed the determination coefficient (R2), the Lin’s concordance correlation coefficient (CCC) (Lin, 1989), the Root mean squared error (RMSE) and the coefficient of variation of the RMSE (CVRMSE). We also performed residual analysis for bias assessment according to (St-Pierre, 2003).
3. Results
3.1. Dynamic prediction of rumen fermentations
The extended model developed in the present work to account for the impact of A. taxiformis on the rumen fermentation is freely available at https://doi.org/10.5281/zenodo.4090332 with all the detailed information of the model and the experimental data used for model calibration. An open source version in the Scilab software (https://www.scilab.org/) was made available to facilitate reproductibility since Scilab files can be opened with a text editor.
Figure 2 shows the dynamic data of fermentation variables for the levels of A. taxiformis at 0.06% and 0.25% compared against the model predicted variables. Figure 3 displays the comparison of all observations against model predictions. Figure 4 shows the residuals for all variables against centred predicted values. Table 2 shows the statistical indicators of model performance. For methane, butyrate and NH3 the mean and linear biases were not significant at the 5% significance level. Acetate and propionate exhibited significant linear bias. The liquid compounds have an average coefficient of variation of the RMSE (CV(RMSE)) of 11.25%. Methane had the higher CV(RMSE) (31%). The concordance correlation coefficients were higher than 0.93. Propionate had the lowest determination coefficient (R2=0.82) while methane and the other compounds had a R2 close to 0.9.
3.2. Prediction of the factors representing the impact of A. taxiformis on rumen fermentation
Figure 5 plots the factors that represent the effect of A. taxiformis on rumen fermentation. Direct inhibition of the methanogenesis due to the anti-methanogenic action of bromoform is represented by the factor Ibr. Methanogenesis inhibition results in hydrogen accumulation impacting the flux distribution of sugars utilization.
Figure 6 displays the simulated dynamics of hydrogen for all the supplementation levels of A. taxiformis. For supplementation levels higher than 0.25%, the methanogenesis inhibition resulted in a substantial hydrogen accumulation.
4. Discussion
The goal of this work was to model the impact of A. taxiformis supplementation on the rumen microbial fermentation and methane production under in vitro conditions using experimental data from (Chagas et al., 2019). Overall, our model was able to capture the dynamics of VFA, ammonia and methane production for different levels of A. taxiformis indicating the potential of the model structure towards the development of predictive models for assessing methane mitigation strategies in ruminants. With the exception of propionate, the slope of observed vs predicted variables is very close to one. Model limitations will be discussed further. We modelled the effect of A. taxiformis on rumen fermentation by two mechanisms. The first mechanism is associated to the direct inhibition of the methanogenesis growth rate by the anti-methanogenic compounds of A. taxiformis documented in different studies (Kinley et al., 2016; Machado et al., 2016a; Roque et al., 2019). In our model, we ascribed the inhibitory effect of A. taxiformis only to the concentration of bromoform. The first-order kinetic rate for bromoform consumption and the inhibition factor (Ibr) (Fig. 5) allowed our model to account for the observed dynamic decline in methanogenesis inhibition (Kinley et al., 2016). It should be noted that although bromoform is the most abundant anti-methanogenic compound in A. taxiformis, the anti-methanogenic capacity of A. taxiformis is the result of the synergetic action of all halogenated products present in the macroalgae (Machado et al., 2016b). Accordingly, it will be useful to include further in our model other secondary compounds such as dibromochloromethane. To enhance our model, it will be central to perform novel experiments to characterize the dynamics of anti-methanogenic compounds. This aspect is of great relevance to allow the model to be adapted to different applications of seaweed supplementation since it is known that the composition of halogenic compounds can vary with respect to the season, harvesting and drying methods.
The second mechanism that accounts for the impact of A. taxiformis on the fermentation is hydrogen control, which it is discussed below.
4.1. Hydrogen control
The anti-methanogenic capacity of A. taxiformis leads to hydrogen accumulation (Kinley et al., 2020; Roque et al., 2020) as predicted by our model in Fig. 6. The level of hydrogen increases as the dose of A. taxiformis increases. The predicted values of hydrogen levels for low doses of A. taxiformis are in agreement with in vitro reported values (Serment et al., 2016). The level of hydrogen can impact electron-mediating cofactors such as nicotinamide adenine dinucleotide (NAD) which are important drivers of anaerobic metabolism via the transfer of electrons in metabolic redox reactions (Hoelzle et al., 2014). van Lingen et al., 2019 extended the rumen model developed by (Dijkstra et al., 1992) to incorporate the regulation of NADH/NAD+ on the fermentation. In our model, the regulation of NADH/NAD+ was incorporated via the control of hydrogen partial pressure assuming a linearity between the couple NADH/NAD+ and the pH2 and following the model structure proposed by (Mosey, 1983) with a different parameterisation for the functions describing the effect of pH2 on the rate of glucose utilization and on the flux distribution. The linearity assumption between NADH/NAD+ and the pH2 might not be fulfilled for all values of pH2 (De Kok et al., 2013).
In the experimental conditions used in the experiment here analysed (Chagas et al., 2019) and under rumen physiological conditions, the linearity between NADH/NAD+ might be valid.
With regard to the hydrogen control on glucose utilization, our model predicts that the control is effective at pH2 higher than 0.2 bars (factor IH2 in Fig. 5). In vivo, the rumen physiological is lower than 0.02 bars indicating that the regulation of glucose utilization by pH2 might not take place under rumen physiological conditions in agreement with theoretical thermodynamic calculations (van Lingen et al., 2016). These results might explain the insignificant changes in total VFA concentration between a control diet and diets with A. taxiformis supplementation in vivo (Kinley et al., 2020). With regard to the fermentation pattern, when the hydrogen level increases the hydrogen control operates by increasing the flux of carbon towards propionate (λ2) while the flux towards the reaction that produces only acetate (λ2) decreases (Fig. 5). Incorporating hydrogen control on the fermentation pattern in our model enabled us to predict the decrease of the acetate to propionate ratio observed at levels of A. taxiformis supplementation leading to substantial methane reduction both in vitro (Machado et al., 2016a; Chagas et al., 2019) and in vivo (Kinley et al., 2020).
4.2. Model limitations and perspectives
In our model, the quantification of the impact of A. taxiformis was ascribed by the action of bromoform on the methanogenesis growth rate and by the action of pH2 on the fermentation pattern. However, in the experimental study (Chagas et al., 2019), nor bromoform nor PH were measured. From our bibliography search, we did not find studies reporting dynamic measurements of bromoform. The lack of bromoform and hydrogen data in our work might result in structural identifiability (Muñoz-Tamayo et al., 2018) and model distinguishability problems (Walter and Pronzato, 1996). We will then require external data to validate our model. Experiments to be done within the MASTER project (https://www.master-h2020.eu/contact.html) will fill this gap and provide data for challenging and improving our model.
Our model aligns with the efforts of enhancing the dynamic prediction of ruminal metabolism via the incorporation of thermodynamics and regulation factors (Offner and Sauvant, 2006; Ghimire et al., 2014; van Lingen et al., 2019). While our work focused only on hydrogen control on sugars metabolism, future work is needed to incorporate the impact of hydrogen on amino acid fermentation (Janssen, 2010). We modelled the regulation of sugars metabolism by hydrogen control following a grey-box modelling approach where the regulation factors were assigned to sigmoid functions without an explicit mechanistic interpretation. However, to enhance the understanding of rumen fermentation, it will be useful to pursue an approach incorporating the role of internal electron mediating cofactors on the direction of electrons towards hydrogen or VFA (Hoelzle et al., 2014; Ungerfeld, 2020). Recent progress in this area (van Lingen et al., 2019) opens a direction for improving the prediction of rumen models.
The ultimate goal of this work is to pursue a model extension to account for in vivo conditions. In this endeavour, experimental data in semi-continuous devices such as the Rusitec (Roque et al., 2019a) will be instrumental for model improvement. In vivo, in addition to the impact on fermentation, A. taxiformis can induce changes in rumen mucosa (Li et al., 2016). These mucosa changes might translate in changes on the rate of absorption of ruminal VFA. This effect on the rate of VFA absorption should be quantified and incorporated into an extended model.
Finally, although our model developments focused on the impact of A. taxiformis on rumen fermentation and methane production, we think our model structure has the potential to be applied to other additives such as 3-nitrooxypropanol (Hristov et al., 2015; Duin et al., 2016) whose action is specifically directed to inhibit methanogenic archaea, as the halogenated compounds of A. taxiformis.
5. Conclusions
We have developed a rumen fermentation model that accounts for the impact of A. taxiformis supply on in vitro rumen fermentation and methane production. Our model was effective in representing the dynamics of VFA, ammonia and methane for six supplementation levels of A. taxiformis, providing a promising prediction tool for assessing the impact of additives such as seaweeds on rumen microbial fermentation and methane production in vitro. Additional data is required to improve our model structure. We are working on model extensions to account for in vivo conditions. We expect these model developments can be useful to help the design of sustainable nutritional strategies promoting healthy rumen function and low environmental footprint.
6. Declarations
Ethics approval and consent to participates
Not applicable
Consent for publication
Not applicable
Availability of data and material
The datasets and codes used in this study are available at https://doi.org/10.5281/zenodo.4090332
Funding
Authors acknowledge funding from the RumenPredict project funded by the Horizon2020 Research & Innovation Programme under grant agreement No 696356. Rafael Muñoz-Tamayo acknowledges funding from the MASTER project, an Innovation Action funded by the European Union’s Horizon 2020 research and innovation programme under grant agreement No 818368.
Authors’ contributions
JCC, MH and SJC produced the experimental data of the study. RMT developed the mathematical model and drafted the article. All authors contributed to the analysis and interpretation of the results. All authors read and approved the final manuscript.
Competing interests
The authors declare that they have no competing interests.