There is interest in understanding the links between the mammalian genotype and the closely associated enteric biome. Along with the influence of these associations on disease. Ruminal acidosis is a complex nutritional disorder of worldwide concern. Estimates of the prevalence in dairy cattle are between 10 and 26.7% (Bramley et al., 2008; Tajik et al., 2009) and feedlot cattle and sheep are also at high risk of ruminal acidosis. The condition is associated with an accumulation of organic acids within the rumen initiated by the combination of consumption of large amounts of readily fermentable carbohydrates and insufficient intake of physically effective fiber (Nagaraja and Titgemeyer, 2007; Bramley et al., 2008). It can be difficult to achieve a balance between feeding readily fermentable carbohydrates to meet the high energy demands required for productivity and to minimize adverse health risks that result from feeding starches and sugars. Several studies have noted considerable variation in responses among cattle fed a common diet designed to induce ruminal acidosis (Brown et al., 2000; Bevans et al., 2005; Penner et al., 2009; Golder et al., 2014a). Current control strategies are unlikely to manage ruminal acidosis in the entire herd due to this considerable variation (Golder et al., 2014a).

The review by Firkins and Yu (2015) notes that differences in the microbiome structure among animals within the same diet group often exceeds those among different diet groups. These differences have been explained by different host genetics and interactions with the rumen microbiome (Weimer et al., 2010; King et al., 2011; Hernandez-Sanabria et al., 2013). As suggested in humans and mice (Benson et al., 2010), bovines appear to have a core microbiome that differs among individuals (Li et al., 2009; Welkie et al., 2010; Jami and Mizrahi, 2012; Henderson et al., 2015). It has not been established if this core microbiome is the result of genetics, environment, or their interaction, or how this may influence susceptibility to ruminal acidosis or other disorders. A mechanistic understanding of gut function and gut microbiota is relevant to our understanding of acidosis in ruminants as well as other disease conditions and production outcomes. Currently, no data are available in the Animal quantitative trait loci (QTL) database for rumen parameters or microbial composition. Genetic selection for desirable traits has improved production and welfare in several animal industries including cattle; therefore, identification of markers linked to favorable rumen conditions could lead to genetic selection toward more acidosis resistant and productive cattle.

The aim of this work was to examine associations between the bovine genome, metabolome, and microbiome, hence to investigate the large among animal variation observed in cattle fed feed additive interventions and subjected to a non-life-threatening, but substantial, starch and fructose challenge (Golder et al., 2014a). The objective is to use this exploratory work to provide a rationale for a larger field study where markers for the susceptibility of ruminal acidosis may be identified for possible use in breeding decisions. We hypothesized that associations would occur between the bovine genome, metabolome, and microbiome and QTL would be identified.

This study was carried out in accordance with the recommendations of The Australian Code for Care and Use of Animals for Scientific Purposes, Scibus Animal Ethics Committee. The protocol was approved by the Scibus Animal Ethics Committee (Scibus 0512-0513 and 0414-0714).

Phenotypic and genotypic data were obtained from 36 pregnant and 4 non-pregnant Holstein heifers. These heifers were sourced from a commercial Holstein herd that primarily used an artificial insemination breeding program. These heifers were enrolled in a study to investigate the efficacy of feed additives during a starch and fructose challenge. The experimental design and diets are described in detail by Golder et al. (2014a). Briefly, all heifers were between 15 and 21 months of age and a mean bodyweight (BW) of 383 ± 49 kg on arrival at the study site located at Cobbitty, New South Wales (NSW), Australia. Each heifer was enrolled in the study for a period of 29-days consisting of 5 experimental periods: (1) preadaptation (days −2 to 0), (2) adaptation I (days 1–10), (3) adaptation II (days 11–20), (4) challenge (day 21), and (5) post-challenge (days 22–26; Figure 1). The preadaptation, adaptation I, and adaptation II periods are often referred to collectively as the pre-challenge periods. The preadaptation period was designed to allow collection of baseline samples and acclimatize the heifers to the study pens and a total mixed ration. The adaptation I period was designed to adapt the heifers' rumen microbiota to the feed additives, while the adaptation II period was designed to examine the influence of fructose on the microbiota and the efficacy of the feed additives. For the duration of the study all heifers, when not being fed or sampled, were kept as one herd in a paddock with little or no available feed and with ad libitum water access. Each heifer was allocated to one of five feed additive groups (n = 8 heifers/feed additive group) that are detailed in Table 1. The efficacy of feed additives to reduce acidosis risk during a non-life-threatening, but substantial, starch and fructose challenge was assessed to allow producers, nutritionists, and veterinarians to make informed management decisions when considering use of feed additives and assist in the development of the most prudent use strategies for antimicrobial and other agents that modify rumen function (Golder et al., 2014a). Heifers were fed twice daily at approximately 07:00 and 14:00 h a 62% forage:38% concentrate total mixed ration at 1.25% of BW dry matter/day for a 20-day adaptation period with their additive(s) in individual concrete floor feeding pens. Pens were cleaned between heifers and feeding sessions to avoid cross-contamination of feed additives. Fructose (0.1% of BW/day) was added to the ration for the last 10 days of pre-challenge (adaptation II). On day 21 heifers were challenged with a ration consisting of 1.0% of BW dry matter milled wheat and 0.2% of BW fructose plus their additive(s). The challenge ration had an estimated non-fiber carbohydrate content of 76.3% of dry matter (CPM Dairy ration Analyzer, Cornell University, Ithaca, NY, USA). Hence, was designed to provide a substantial but non-life-threatening acidosis challenge to test the efficacy of the feed additives.

Bovine DNA was successfully extracted from 25 whole blood samples and 9 serum samples. A heifer from the monensin + tylosin group who had a successful DNA extraction from whole blood was removed as she consumed only 12% of her challenge ration, leaving a total of 33 heifers in the analysis. Thirty one of these 33 heifers were pregnant.

Associations between the bovine genome, metabolome, and microbiome and identification of QTL, as hypothesized, can be related to physiological challenge and suggest that some of the variation in responses in microbial data (Table 2) and metabolomic data that were observed previously (Golder et al., 2014a) may be attributed to the host genome. The study design used in the initial animal experiment (Golder et al., 2014a) from which the data for this current work is derived was not initially intended for use to explore genomic relationships; hence, the study population in the current work is not ideal for genomic analysis. Typically, a larger population size, evaluated at the same physiological status, and fed an identical diet would be used. However, with the limited data currently available in this field and the conservative nature of the statistical approach used, this work provides indications for future directions. These findings provide a strong rationale for a subsequent study with a larger number of animals.

Benson et al. (2010) used a similar approach to our study utilizing a mouse model, in which, they examined factors that influence microbial composition in an intercross line (n = 645). They determined that family/litter (10%) and cohort (26%) had significant impacts on gut microbiota composition which left host genetics to play a major role in explaining the remaining variation. Benson et al. (2010) used a set of 530 markers, while the present study utilized a final filtered set of approximately 95,000 markers, but in a smaller number of animals (33). As in the present study, Benson et al. (2010) reported that microbial phyla behaved like polygenic traits with multiple QTL reported for each phylum and additional QTL identified at the level of family and genus. As this current work is only exploratory, family and genera were not investigated but are intended to be explored in a larger field study. Similar to the current study, Benson et al. (2010) also observed pleiotropy with several QTL that appeared to be relevant on multiple taxa. The relationship among microbial phyla, host genome, and metabolite, specifically for the region on chromosome 4 that had overlapping putative QTL for concentration of total lactate and L-lactate and for relative abundance of Actinobacteria, which have been previously associated with subacute ruminal acidosis (Mao et al., 2013), provides support for the approach taken in this study. Further, the identification of interesting positional candidate genes such as MEPE in regions associated with the relative abundance of Actinobacteria, Euryarchaeota, and Fibrobacteres may provide novel directions for future research.

Benson et al. (2010) demonstrated in murine models, that the gut microbiota can now be viewed as an environmental factor that itself is controlled in part by host genetic factors and potentially by interactions between host and microbial genomes. This view implies that genetic predisposition to complex diseases may be manifested in part by a predisposition to aberrant patterns of microbial colonization, which in turn contribute to disease processes (Benson et al., 2010).

We recognize that a larger study with different diets and expression of acidosis will provide more insights to the associations studied. Factors that may have influenced our responses include; the exposure of cattle to feed additives; a limited opportunity for the microbiomes to adapt to the three ration changes; inherent variability among individuals in the time that the microbiome adapts to diets; and pregnancy status and stage of pregnancy. The differences in number of associations identified for each of the phenotypes between the pre-challenge and challenge periods highlights the importance of exploring genomic relationships at each of the 8-time points and suggests an influence of diet. Interpretations of these differences are limited by the low population size, but it is important to note that the associations and QTL identified for D-, L-, and total lactate concentration occurred pre-challenge, coinciding with lower among animal variation; whereas, associations for A:P and valerate concentration occurred from samples collected on the Challenge day.

Our findings that suggest responses to rumen disturbances could be host specific, as indicated by the associations between the host genome, metabolome, and microbiome support those of Weimer et al. (2010). These authors showed the ability of the rumen to revert to pre-exchange VFA concentration and rumen pH and nearly return to pre-exchange bacterial community composition within 24-h of a 95% exchange of ruminal content with a cow on a similar diet. A second cow took a longer period to revert indicating the potential for variability in this response (Weimer et al., 2010). While Chen et al. (2012) discussed possible links between acidosis and immune function, there has been limited work in this field to examine the basis for diversity in responses of cattle to different dietary challenges.

Although knowledge of associations between the host, the microbiome, and the metabolome is lacking in ruminants, the effects of different diets on the bacterial microbiome and links to the metabolome have been made (Petri et al., 2013). Further, associations to production outcomes have also been made, for example in dairy cattle Jami et al. (2014) demonstrated that the Firmicutes to Bacteroidetes ratio was strongly positively correlated to milk fat (Pearson R = 0.72). These authors also describe correlations between other production and efficiency measures and genus abundance. Lima et al. (2015) associated bacterial taxa with milk production and milk components. Butyrivibrio spp. and Prevotellaceae 2 were some of the bacteria that had the highest positive correlation with milk production (Lima et al., 2015). In beef cattle, Hernandez-Sanabria et al. (2010) linked PCR-DGGE patterns to ruminal fermentation patterns, rumen metabolites, and feed efficiency traits. Roehe et al. (2016) demonstrated host genetic links with microbial methane production in a factorial experiment with crossbred breed types and diet.

The overlapping genetic region between the relative abundance of Actinobacteria, Euryarchaeota, and Fibrobacteres, and A:P is of interest. Diets high in starch induce propionate production which lowers the amount of H₂ gas produced and also methane production, in comparison to diets with more fiber which tend to increase methane production relative to dry matter intake (Annison and Lewis, 1959; Hungate, 1966). Therefore, we could expect that a high starch diet would lower the abundance of Fibrobacteres, and Euryarchaeota relative to the total microbial population and have a lower A:P. Many members of the phylum Actinobacteria are lactate producers, while others are lactate consumers. Therefore, it is difficult to hypothesize the relationship between this phylum and the A:P as it is likely to depend on the abundance of the different genera that comprise this phylum.

The overlapping regions between Chloroflexi, Tenericutes, and A:P need further investigation, as it is not known whether members of the Chloroflexi phylum are resident bacteria and have any significant role in the rumen (Kim et al., 2011). They are only typically found at low abundances in the rumen and, as sequence datasets are yet to contain any culturable isolates from this phylum, it is difficult to ascertain their metabolism (Kim et al., 2011). The Tenericutes are small prokaryotes that do not contain a cell wall and are therefore resistant to penicillin (Brown, 2011). This may suggest they could also be unaffected by the ionophore and antibiotics (monensin, tylosin, or virginiamycin) given in this study. Mao et al. (2013) showed no change in relative abundance of Chloroflexi and Tenericutes between dairy cattle fed with a subacute acidosis induction diet and controls, but the A:P was decreased in the cattle fed with the subacute acidosis diet. Golder et al. (2014b) showed a tendency toward an increase in Tenericutes in grain-fed heifers, compared to those fed no grain, but there was no effect on relative abundance of Chloroflexi. The A:P was numerically lower in these grain-fed animals, but A:P was not statistically analyzed in the study by Golder et al. (2014b).

Although the Bacteroidetes and Firmicutes phyla are the dominant bacterial phyla in ruminants, these had few marker associations. This could reflect both the diversity of bacterial taxa within these phyla, or perhaps suggests that these phyla comprise a large portion of the core microbiome and are less inherently variable or are preserved as essential to rumen function. The A:P is commonly reported in studies of rumen function and can give an indication of rumen health, when combined with other measures. Hence, it is not surprising that this ratio had more associations than the acetate or propionate measures alone.

It is not surprising that rumen pH had few markers associated as although ruminal pH is commonly used as a diagnostic indicator for acidosis, on its own, point estimates of rumen pH are a poor indicator of acidosis (Golder et al., 2012). Receiver operator curves produced from the dataset of Bramley et al. (2008) that assessed ruminal status of 800 cows from 100 dairies using ruminal, performance, feed, and fecal characteristics showed that ruminal pH measured from ruminal fluid obtained using a stomach tube and rumenocentesis were only moderately sensitive (0.68 and 0.74, respectively) and specific (0.84 and 0.79, respectively) indicators of ruminal acidosis (Golder et al., 2012).

This research, although preliminary, suggests there is a link between host genetics, the microbiome, and physiological responses and may lead to selection toward acidosis resistant cattle. Such selection is likely to be occurring in the field in herds that are consistently feeding diets that have a high risk of ruminal acidosis. In these herds, there is likely to be a forced selection toward acidosis resistant dams' due to voluntary and involuntary culling as a direct or indirect response to ruminal acidosis.

In summary, our hypothesis that the bovine genome, metabolome, and microbiome are associated in cattle subjected to a non-life threatening, but substantial starch and fructose challenge, with large among animal variation observed is supported by these exploratory observations. These findings assist in understanding the occurrence of inter-animal variation in responses to physiological challenge both in the field and experimentally. They provide a strong rationale for a large field study of natural variation in ruminal acidosis. More research is needed to determine the heritability of microbial constituents and correlated responses to physiological changes. In the future rumen function markers may aid in identification and, therefore, management of animals predisposed to rumen issues.

HG and IL: conception, design, animal experiment, and metabolome analysis; HG, JT, and IL: host genome analysis and drafting manuscript; HG, SD, and CM: microbial analysis. All authors: interpretation, critical revision, and final approval and accountability.