This was a multicenter observational study. The study population comprised 293 cows from 36 herds [10 in California, USA (CA); 12 in Canada (CAN); 10 in Australia (AU); and 4 in Wisconsin, USA (WI)], with a target of eight cows per herd (four primiparous and four of parity >1 and ≤4) between 10 and 100 DIM. The AU herds were in four geographical regions (the Western districts of Victoria, Finley in NSW, the South Coast of NSW, and the Macarthur region of NSW). The herds ranged in size from 57 to 6,294 cows and are summarized by Golder et al. (2023b). The diets of the North American herds were all total mixed rations, whereas those in Australia were primarily pasture-based with supplementary concentrates and/or silage.

Power calculations were carried out using the rdpower function in Stata (version 14.2; StataCorp LLC, College Station, TX) and were based on a difference of 5.5 mM in the VFA propionate, measured with a SD of 10 mM being sufficient to categorize non-acidotic cows (Bramley et al., 2008). Consequently, a difference of 29mM–34.5 mM in propionate would require 320 cows, given an intraclass correlation of 0.2. The study operated on two levels. The effect size (ES) was 0.55, the number of head per herd was eight (n), and the number of herds was 40 (m) to provide a power of 0.6.

Farms were selected based on their willingness to participate and if they had accurate details on cow identification, parentage, and calving history; predominately Holstein-Friesian lactating cattle; conducted herd recording; and were able to provide diet details and a sample of the diet for analysis. There were no restrictions based on herd size, production level, production system, or milking frequency per day.

Cows that met the following criteria were randomly selected: they were Holstein-Friesian cows, were fourth-parity cows at maximum (four cows of parity 1, and four cows of parity >1 and ≤4), were between 10 and 100 DIM, had no apparent current clinical illness, and were from different sires. This information was obtained using each herd’s herd management software.

Sample collections included rumen fluid for ruminal metabolomic analysis, bacterial taxonomy analysis, and determination of acidosis risk group; milk samples for production data; and samples of the diet for analysis of nutrient composition. Detailed descriptions of collection and analyses are provided by Golder et al. (2023b).

In the high-risk group, the abundances of Anaerocella_f and Acholeplasmataceae were increased, whereas those of AY244965_f, Anaerolineaceae, Christensenellaceae, Ruminococcaceae, AF050559_f, EF445272_f, and Planctomycetaceae were decreased compared with the medium- and low-risk groups, which had similar abundances ( Table 3 ). The abundance of Gammaproteobacteria_c was lowest, whereas that of Veillonellaceae was greatest in high-risk cows, followed by cows of low-risk and then medium-risk. Low-risk cows had an increased abundance of GU304534_f, EF436358_f, and Paracaedibacteraceae and a decreased abundance of Lachnospiraceae, compared with the other groups, which had similar abundances. The abundances of BS11_f and Erysipelotrichaceae were also greater in low-risk cows than in high-risk cows, but similar in high- and medium-risk cows, and also in medium- and low-risk cows. Medium-risk cows had greater abundances of RF16_f and Saccharimonadaceae.

Primiparous cows had greater abundances of Anaerocella_f, Eubacteriaceae, and Veillonellaceae than multiparous cows ( Tables 3 , 4 ). Region was significant for the abundance of all influential families except Veillonellaceae, EF436358_f, Gammaproteobacteria_c (tendency p= 0.079), and Paracaedibacteraceae. The cows from Canada had greater abundances of AY244965_f, GU304534_f, RF16_f, and Acetobacteraceae than cows from Australia and California. The cows from Canada also had greater abundances of BS11_f than cows from California, but Australian cows had similar abundances of BS11_f to cows from both California and Canada ( Tables 3 , 4 ).

The cows from Australia had greater abundances of Anaerocella_f and lower abundances of Christensenellaceae and Planctomycetaceae than cows from other regions. Cows from Australia had a lower abundance of EF445272_f and greater abundance of Eubacteriaceae than cows from California, but their abundances of these families were similar to those in cows from Canada, and cows from California and Canada also had similar abundances ( Tables 3 , 4 ). Cows from Australia had a greater abundance of Erysipelotrichaceae than cows from California, which was higher than that in cows from Canada.

The cows from California had greater abundances of Carnobacteriaceae, Lachnospiraceae, Lactobacillaceae, Leuconostocaceae, and Streptococcaceae and lower abundances of Prevotellaceae, S24-7_f, Acidaminococcaceae, AF050559_f, and Acholeplasmataceae, than those from other regions. The cows from California also had greater abundances of Coriobacteriaceae, Anaerolinaceae, Ruminococcaceae, and Saccharimonadaceae than cows from Australia, which was greater than abundances in cows from Canada ( Tables 3 , 4 ).

Group by region was significant for over one-third of the influential phyla ( Table 3 ). Group by parity was significant for Carnobacteriaceae and Ruminococcaceae ( Figures 2A, B ). Region by parity was significant for Anaerocella_f, Carnobacteriaceae, Christensenellaceae, Veillonellaceae, Saccharimonadaceae, and EF445272_f. There were three-way interactions for Anaerocella_f, Eubacteriaceae, Paracaedibacteraceae, and EF445272_f, despite only a significant main effect for acidosis group and no significant two-way interactions.

Figure 3 shows a correlation biplot of the redundancy analysis of bacterial families with respect to acidosis risk group, region, and dietary nutrients (namely, NDF, CP, sugar, starch, and crude fat). Only the 20 families with the best fit to this model are displayed and the x and y coordinates and Cfit2 values for each of these bacterial families are shown in Supplementary Table 1 . The variables explained 17.6% of the total variation. The high- and low-risk samples both differed to the rest of the samples and presented in inverse quadrants to each other; the high-risk acidotic category was associated with 4.1% of the variation (p= 0.022), the medium-risk category was associated with 0.7% of the variation (p= 0.154), and the low-risk category was associated with 3.3% of the variation (p= 0.022) and presented in the center of the biplot. Each region was statistically different and presented in different quadrants; California was associated with 6.3% of the variation (p= 0.022), Canada was associated with 5.1% of the variation (p= 0.022), and Australia was associated with 4.0% of the variation (p= 0.022). Sugar, CP, NDF, and starch were all statistically different (p= 0.022) and associated with 4.4%, 2.8%, 2.4%, and 1.0% of the variation, respectively. Crude fat was associated with 0.8% of the variation (p= 0.176). Sugar and CP content were associated with the high-risk group, whereas starch was associated with the low-risk group. The abundance of the family Erysipelotrichaceae was that most associated with the high-risk group; that of the family Gammaproteobacteria_c was most associated with the low-risk group; that of the family Prevotellaceae was most associated with NDF; and those of Coriobacteriaceae, Lachnospiraceae, and Streptococcaceae were most associated with sugar.

The relatively few phyla (only five in total and four overlapping between acidosis risk groups) and families associated with acidosis and ruminal conditions support the theory that ruminants have a core microbiota (Jami and Mizrahi, 2012). In a large cohort study where dairy cattle had the same diet and management, core genera accounted for 53.9% of correlations in a correlation network analysis among rumen bacteria, rumen short-chain fatty acids, and lactation performance (Xue et al., 2018). Large inter-animal variation in the relative abundance of core taxa was also observed (Xue et al., 2018). The large overlap in the number of influential families between acidosis group and region in the redundancy analyses adds further support for the existence of a core microbiota and emphasizes the dynamic nature of the non-core microbiota. The opposing point direction for coefficients for the phyla and families that predicted the high- and low-risk groups further supports a hypothesis that these are key core taxa that shift during ruminal disturbance. Furthermore, the consistency of these observations supports our use of the backward stepwise statistical approach to provide meaningful outcomes. Xue et al. (2018) suggest that enhanced knowledge of the core and non-core rumen microbiota enables their manipulation, and we concur with this view.

Golder et al. (2023b) found that the Firmicutes were the most abundant phylum for all groups and were most abundant in the high-risk acidosis group. The Firmicutes, Bacteroidetes, and Proteobacteria phyla are usually considered the dominant phyla in cattle (Hook et al., 2011). The Lachnospiraceae were the most abundant family in the high-risk group. Thus, it is not surprising that the Firmicutes phylum and the Lachnospiraceae family had the greatest increase in abundance in the models for the high-risk group, and that the inverse was observed for the low-risk group. The phylum with the second highest RA for all groups, Bacteroidetes (Golder et al., 2023b), was not associated with acidosis risk in our models. Bacteroidetes often have an inverse association with the RA of Firmicutes. Thus, taxa that had an increased likelihood of being increased in abundance in the low-risk group were phyla that were present in a low abundance. These findings align with the premise of Nagaraja and Titgemeyer (2007) that gram-positive Firmicutes may displace the gram-negative Bacteriodetes during ruminal acidosis. This also likely explains the absence of the abundant family Prevotellaceae, which is a member of the Bacteroidetes phylum, from the final models. The Prevotellaceae family contains a diverse population of species (Krieg, 2010), which may limit its value in prediction. Interestingly, Ruminococcaceae, a member of the Firmicutes phylum, was the third most abundant bacterial family overall on a RA basis and was the family with the greatest RA in the low-risk group that was also not associated with group inclusion, perhaps for a similar reason.

It is logical that the class Gammaproteobacteria from the Proteobacteria phylum was associated with groups as it contains a range of diverse medically important gram-negative rod bacteria, including Escherichia coli, Salmonella spp., and Klebsiella spp. from the Enterobacteriaceae family; Pasteurella spp. from the Pasteurellaceae family; Yersinia pestis (bubonic plague) from the Yersiniaceae family; and Vibrio spp. from the Vibrionaceae family.

This group was characterized by increased abundances of Anaerocella_f [which do not utilize carbohydrates (Abe et al., 2012)] and Veillonellaceae [which has various characteristics that include lactate utilization (Hungate, 1966; Bergey et al., 2011)], which suggests a higher risk of acidosis. Decreased abundances of several bacterial families with diverse characteristics (AF050559_f, Anaerolinaceae, AY244965_f, Christensenellaceae, EF445272_f, Gammaproteobacteria_c, Planctomycetaceae, and Ruminococcaceae) were associated with an increased risk of acidosis. These taxa belong to an array of different phyla that were not consistent in Gram stain and of these, only Gammaproteobacteria_c and Planctomycetaceae were also in the predictive model. The increased odds in the predictive model of the abundance of Lactobacillaceae, which is a family of lactic acid producers (Bergey et al., 2011), supports the classification of this group as high risk (Golder et al., 2023b). A high level of lactic acid accumulation in the rumen is most likely responsible for clinical acidosis only (Nagaraja and Titgemeyer, 2007). Clinical acidosis is less prevalent than milder forms of acidosis. Lactate concentrations concentrations did not differ among the acidosis risk groups in our study (Golder et al., 2023b); hence, our ruminal bacterial community composition should reflect this. This study was not an acidosis challenge study; thus, we were not likely to observe a high prevalence of clinical acidosis in the study population. Our data, being broader and from a range of herds with differing management and diets, is not directly comparable to evaluations of rumen bacterial community compositions from acidosis challenge studies that induce acidosis with particular substrates. Golder et al. (2012b) found that lactate concentrations peaked within 20 minutes of feed consumption but declined rapidly. Our rumen samples were not collected immediately after feeding, and it is possible that if lactate was generated then it had already been metabolized, with portions converted into propionate, valerate, and caproate, which are synthesized from three lactate molecules (Annison and Lewis, 1959), as these fermentation products were also highest in the high-risk acidosis group. Thus, perhaps there was a population of lactate utilizers present in this group. This suggestion is further supported by an increased abundance of the Veillonellaceae family in this group, which contains Megasphaera elsdenii, one of the producers of valerate from lactate (Hungate, 1966; Stewart et al., 1997).

The traditional model of aetiology of ruminal acidosis, which is largely based on culture-based techniques, suggests that Streptococcus bovis and lactic acid utilizers are primarily responsible for acidosis (Nocek, 1997). Culture-based techniques are widely recognized to have limitations in the evaluation of microbial populations, namely their vast underestimation of microbial diversity and richness (McSweeney et al., 2007; Fernando et al., 2010). There is an increasing body of evidence showing that acidosis is complex and that this model of the aetiology of ruminal acidosis needs refinement (Britton and Stock, 1989; Plaizier et al., 2018; Plaizier et al., 2022).

Our work shows that several bacterial taxa are likely to be involved in the etiology of ruminal acidosis. This is consistent with the knowledge that the rumen microbiome is dynamic with diversified adaptive functions (Gharechahi et al., 2021). The microbiome has high metabolic redundancy (Taxis et al., 2015), which probably explains why multiple microbial taxa may be associated with acidosis. Rumen microbes have symbiotic relationships with each other (Hungate, 1966); however, a survey by Henderson et al. (2015) of several different ruminants from different parts of the globe that are likely to have differed vastly in diet and management found that rumen bacteria rarely had similar co-occurrence patterns. It is critical to consider the difference between abundance compared to co-dependency compared to function. Metabolically active bacteria within and without biofilms may facilitate the growth and multiplication of other bacteria through substrate influence in the exometabolome (Morris et al., 2012; Douglas, 2020). Our study suggests there is enough commonality in the microbiome between acidosis cases in early-lactation Holsteins to use the abundance of rumen bacteria to predict acidosis risk. Currently, collection, storage, processing, interpretation of results, costs, and turn-around time of results prevent this from being a viable method of diagnosis for producers. Development of a rapid cow side test, used in combination with other diagnostic methods and clinical history, may be viable in the future.

Overall, the medium-risk group had very few associations with the microbiome, only being characterized by increased abundances of RF16_f and Saccharimonadaceae. These Gram-negative and Gram-positive bacterial families, respectively, although knowledge on them and their substrates is limited, are frequently found in cattle (Abbas et al., 2020; Daghio et al., 2021; Ouyang et al., 2021). The medium-risk group did not have any predictive phyla and had the smallest number of predictive families. This is consistent with the lack of differences in milk production (Golder et al., 2023b). However, Bainbridge et al. (2016) suggested that correlations between milk production and rumen microbial populations do not imply causation, as both may be strongly influenced by multiple factors. This group did have distinctive ruminal metabolomic measures, having the greatest rumen pH and least acetate, propionate, butyrate, valerate, iso-butyrate, iso-valerate, and caproate concentrations, and numerically lowest total VFA (Golder et al., 2023b). It is evident from these measures and the PCA plots that cattle in this group differ from those of the other two groups, but we may not have captured all the sources of heterogeneity. Golder et al. (2023b) suggested that this group could contain underfed cows, cows that are inappetent or that were not recently feeding, or cows in recovery from acidosis or an illness, as the mean total VFA was low. The concentration of ruminal VFA is directly related to bacterial community composition, amount of fermentation, and rumen epithelial absorption (Brockman, 2005). The greater abundance of Streptococcoceae in this group is interesting. This family contains Streptococcus bovis, which has traditionally been associated with the etiology of acidosis by precipitating a decline in rumen pH and consequent increase in lactic acid (Nocek, 1997). Interestingly, in grain-based subacute ruminal acidosis studies, responses have differed with both increases (Fernando et al., 2010) and decreases (Petri et al., 2013; Plaizier et al., 2017) observed in the populations of S. bovis. Neither lactic acid nor the abundance of Lactobacillaceae, the family that includes of the Lactobacillus genus, were increased in this group. Three out of the five predictive taxa were not included in the 29 influential taxa and were present in only minor abundances, but the three with the highest odds were all Gram-positive members of the Firmicutes phylum, supporting the hypothesis that this group could be recovering from acidosis. An increase in Firmicutes numbers occurs with an increase in rapidly fermentable carbohydrates (Hungate, 1966; Mao et al., 2013; Plaizier et al., 2022). Given the prevalence of cattle categorized as high risk for acidosis in the study population, and the time that is likely needed for the rumen to recover from dysbiosis, a population of cattle should exist that is in recovery from acidosis, and perhaps the medium-risk group represents these. Brede et al. (2020) showed that the rumen bacterial abundance of most bacterial groups returned to initial abundance 5 days after a subacute ruminal acidosis challenge using a rumen simulation technique (RUSITEC).

The low-risk group was characterized by increased abundances of Gammaproteobacteria_c, EF436358_f, GU304534_f, and Paracaedibacteraceae, and decreased abundances of Lachnospiraceae and Veillonellaceae. Many members of the Gammaproteobacteria class have deleterious effects in cattle but are also natural inhabitants of the rumen. The release of LPS from the lysis of Gram-negative bacteria such as these may contribute to the pathogenesis of ruminal acidosis (Plaizier et al., 2018). The fact that these were more abundant in the low-risk group than in the high-risk group suggests that cattle in the low-risk group were at risk of disorders other than acidosis, that these bacteria may not pose a concern, or that these cattle had more robust and resilient microbial ecosystems that resist the potential negative impacts of these bacteria. It could be expected that some cattle would be in a state of flux between acidosis risk groups; each cow has a different distance to the centroid of each acidosis risk group, which reflects the strength of association at the time of sampling (Golder et al., 2023b). Ruminal acidosis is a multifactorial disorder and has substantial sequelae (Enemark, 2008; Plaizier et al., 2008). There may be additional sequelae that are yet to be identified as clinically or metabolically associated with acidosis.

The Planctomycetaceae family, which had the largest increase in odds ratio in the models, is from the Planctomycetes phylum, which is Gram-negative and plays a role in global carbon and nitrogen cycles (Wiegand et al., 2018). Many species of the Planctomycetes phylum are involved with anaerobic ammonium oxidation, a function consistent with the low-risk group having the highest ammonia concentrations (Golder et al., 2023b); however, these are usually found in the Brocadiaceae family (Wiegand et al., 2018). It is interesting that a decrease in the abundance of Fibrobacteraceae, a group likely to contain fiber digesters, was associated with increased odds of being in the low-risk group when we had hypothesized that the opposite would occur. Krajcarski-Hunt et al. (2002) showed that the rumen digestion of NDF from grass hay, legume hay, and corn silage was reduced by induced subacute ruminal acidosis.

Golder et al. (2023b) and Bramley et al. (2008) proposed that cattle in the low-risk group may have had a greater synergy between ruminal energy and protein metabolism than those in the high-risk group. This leads to slower fermentation, favoring the production of acetate and butyrate, which provide “safer” electron sinks and overall a more stable rumen. A lower abundance of Firmicutes at the phylum level, in addition to the family members of this phylum, the Carnobacteriaceae, Peptostreptococcaceae, and Lachnospiraceae, having increased odds of being in the low-risk group supports these cattle being at a lower risk of acidosis. The Carnobacteriaceae produce lactic acid as their major end product (Bergey et al., 2011), which is considered an “unsafe” electron sink and has a pKa of 3.0, compared with that of 4.8 for ruminal VFA (Oetzel, 2003). A decreased abundance of Veillonellaceae in the mixed-model analysis provides further support to this hypothesis. The Tenericutes phylum, which contains the Mycoplasma genus, had the greatest association with odds of inclusion in the low-risk group at the phylum level, but none of the predictive families for the low-risk group is a member of this phylum. The pattern of significance for the abundance of families was more similar between the medium- and low-risk groups than between the high- and low-risk groups.

Golder et al. (2023b) suggested that region may act to confound acidosis group determination for some measures in this study and hypothesized that this would result from differences in diet and management among regions. Although the multi-center, multi-country study design provides strong external validity of our findings, the regional effects and their influence on group associations may be a limitation. Bacteria utilize different substrates (Hungate, 1966), thus the presence and abundance of bacteria should be predicted from dietary substrates. The dietary nutrient components that were most consistently associated with our key bacterial taxa were sugars and CP.

Cattle can be exposed to a range of sugar contents from different sources in their diet. Sugars have a greater k_d than starch and other carbohydrate fractions (Sniffen et al., 1992) and are therefore more likely to overwhelm the buffering capacity of the rumen than less readily fermentable carbohydrates. Sugars are more likely to be fermented into lactic acid than starches (Harmon et al., 1985; Heldt et al., 1999; Golder et al., 2012b) and create an increased risk of ruminal acidosis (Nagaraja and Titgemeyer, 2007; Golder et al., 2012b; Golder et al., 2014a).

The potential involvement of dietary protein content in the pathogenesis of ruminal acidosis is largely unexplored. The supply of rumen-degradable protein has influenced organic acid and lactate pool sizes (Hall, 2013) and increases in levels of rumen-available protein have increased organic acid concentrations, regardless of the levels of rumen-available carbohydrate (Herrera-Saldana and Huber, 1989). The rumen-undegradable portion of protein from the diet bypasses the rumen and is absorbed as amino acids in the small intestine, thereby supplying approximately 30% more energy than starch (Klopfenstein, 2001) without the generation of hydrogen in the rumen. A total of 8% of the dry weight of bacteria is hydrogen (Todar, 2012), which provides a considerable sink for the hydrogen generated during ruminal catabolism of carbohydrates. Consequently, it appears possible that protein can be either beneficial or detrimental to the risk of ruminal acidosis, depending on substrate availability, concentration, and dietary management. The synchrony of rumen-available protein and carbohydrate has been proposed as a method to increase the efficiency of microbial nitrogen production and animal productivity (Johnson, 1976) and could influence the risk of ruminal acidosis. Ammonia may “buffer” changes in rumen pH by neutralizing between 10% and 15% of the rumen VFA concentrations produced (Owens et al., 1998), a function which could reduce the risk of ruminal acidosis. Golder et al. (2014c)demonstrated that feeding canola meal to increase crude protein content produced higher ruminal ammonia concentrations and estimated metabolizable protein yield, and was associated with reduced risk of acidosis. Golder et al. (2014c) postulated that these cows would have an increase in rumen-undegradable protein availability and microbial growth. Fermenten^®and Biochlor^®increased in vitro rumen ammonia nitrogen concentrations and stimulated microbial protein nitrogen production by approximately 24.6% and 13.5%, respectively (Lean et al., 2005). Cows fed Fermenten^® had reduced ruminal acidosis risk in vivo (Golder, 2014). These findings in totoindicate the potential for ammonia and peptides to reduce the risk of acidosis through neutralizing VFA and increasing microbial protein production, thereby reducing the futile cycles leading to increased VFA production and energy spilling (Russell, 2007).

Overall, our findings support the concept that the rumen ecosystem is host specific and comprises a core rumen microbiota that has a unique ability to adapt to different substrates and may contribute to a host’s individual susceptibility to disorders such as ruminal acidosis (Golder et al., 2014b). Microbial interactions are extremely complex; it is a challenge to evaluate the entire microbiome and the influence of other factors, including protozoa, fungi, biofilms, and inflammatory agents, and markers requires exploration. Only a small percentage of rumen bacteria species have been characterized (McSweeney et al., 2007), highlighting the infancy of this field. Flux through metabolic pathways requires characterization, as bacteria prevalence may not reflect metabolic function. Furthermore, recognized metabolic fermentation pathways fail to explain all generated fermentation products (Hackmann et al., 2017), emphasizing a need for a continuation of microbial culturing to accompany omic technologies (Gruninger et al., 2019).