All procedures described were conducted under experimental license from the Irish Department of Health in accordance with the Cruelty to Animals Act 1876 and the European Communities (Amendments of the Cruelty to Animals Act 1876) Regulations, 1994. Twenty triplet-bearing ewes from the sheep herd at University College Dublin, Lyons Research Farm, Newcastle, Co., Dublin, Ireland were used in the current study. All ewes were bred using artificial insemination and lambed within 1 day of each other.

At 24 h post-partum, 20 triplet bearing and rearing ewes were enrolled in the study. Ewes were blocked based on combined litter weight (11.6 ± 1.51kg) and treatments were balanced for maternal live weight at parturition. The experimental design was a 2 × 2 factorial design. From birth to 6 weeks post-partum ewes (and lambs) were offered a grass silage based diet plus one of two concentrate types (with linseed oil (L) or without linseed oil (C) inclusion). After weaning lambs were offered one of the two concentrate types described above to meet 100% of their nutrient requirements. This lead to four post-weaning dietary treatments: C:concentrate without linseed oil fed both before and after weaning, L:concentrate containing linseed oil fed both before and after weaning, L-P:concentrate with linseed oil fed before weaning and concentrate without linseed oil fed post-weaning, L-POST: concentrate without linseed oil fed before weaning and concentrate containing linseed oil fed post-weaning.

The control diet contained 47.5% barley, 27% soybean meal, 10% citrus pulp, 9% soya hulls, 4% minerals and vitamins, and 2.5% cane molasses on a fresh weight basis. For the linseed diet barley inclusion level was reduced by 4% and replaced with 4% linseed oil. Ewes and lambs were housed in straw bedded pens and for the first 2 weeks of the trial, lambs had ad-lib access to the ewes’ milk. Ewes were fed 3 kg/head/day of concentrates which was gradually reduced to 1.5 kg/head/day over several weeks up to weaning. At 2 weeks of age, a creep area was created in the corner of each pen to which only lambs had access. Lambs were offered ad-lib creep meal concentrates according to the dietary treatment they were on.

Average feed intake of lambs was calculated each morning. Daily weigh back measurements of feed were taken and averaged among all trial lambs. Lambs were offered ad-lib concentrates and water. At approximately 6 weeks of age, when average lamb intake was above 550 g per day, lambs were weaned and ewes were removed from the pens. After weaning half the lambs on the control diet (C) were placed on the linseed diet (L) and vice versa. This resulted in the four treatment groups described- -three of which were analyzed in the current study (C, L and L-P). Each of these treatment groups consisted of 15 lambs, 9 of which were analyzed in this study, leading to a total of 27 lambs analyzed.

All groups were balanced for live weight (19 kg ± 3.2) and sex. Lambs were given a 7-day diet acclimation period. At 8 weeks of age, lambs were moved into individual pens where dry matter intake (DMI) was recorded. Lambs were offered ad-lib concentrates at approximately 110% of their DMI from the previous day. During the experimental period, live weight was recorded weekly using a calibrated Tru-Test SR 3000 electronic weighing scales (Tru-Test Ltd., Auckland, New Zealand). The weighing scales were calibrated each time with known weights.

Once they achieved a live weight of 41 kg (± 2.1kg), at approximately 111 (± 14) days, lambs were brought to Kildare Chilling Co. (Kildare Town, Co., Kildare, Ireland) for slaughter. The stomach was dissected and solid rumen digesta was removed, squeezed between two layers of cheesecloth and an 8 ml aliquot from the resulting collected rumen fluid was transferred into a labeled plastic vial containing 2 ml of trichloroacetic acid (50%). This sample was then frozen at -20°C for VFA and NH₃ analysis. Unfiltered rumen solids were collected from 6 sites within the rumen, combined and stored at -80°C for microbial analysis.

Rumen fluid samples were analyzed for VFA concentration by gas chromatography (Varian 3800, gas chromatograph) using a CP-wax 58, 25 m × 0.53 mm capillary column (Varian BC, The Netherlands) as previously described by Hart et al. (2009).

DNA was extracted from freeze-dried, unfiltered rumen solid samples from all 9 animals in each treatment group (27 samples in total) using the PCSA method described by Lueders et al. (2004), with phenol-chloroform added prior to the bead beating step, and stored at -20°C until needed. DNA concentration and purity were assessed using a Nanodrop^TM spectrophotometer (Thermo Scientific, Waltham, MA, United States).

Bacterial and archaeal community structures were analyzed using amplicon sequencing on an Illumina Miseq platform using the method described by Kozich et al. (2013). Briefly, once DNA was extracted and purified 1 μl target DNA was added to a well on a 96-well PCR plate already containing 17 μl of Accuprime Pfx Supermix (Invitrogen, Thermo Fisher Scientific, Dublin, Ireland) and 2 μl of a primer set targeting the V4 region of the 16S rRNA gene (Kozich et al., 2013). PCR conditions consisted of a hot start at 95°C for 2 min, followed by 95°C for 20 s, 55°C for 15 s and 72°C for 5 min (30 cycles) with a final elongation of 72°C for 10 min. PCR products were visualized on a 1.2% (w/v) agarose gel (Roche Diagnostic, Basel, Switzerland).

PCR products were then cleaned and normalized using the SequalPrep Normalization Plate Kit (Invitrogen, Thermofisher, Waltham, MA, United States) according to manufacturer’s instructions. DNA concentration was determined using a Qubit fluorometer and dsDNA kit (Thermofisher Scientific, Dublin, Ireland) to ensure a concentration ≥100 ng was present. The amplicon pool was then sent to the Centre for Genomic Research, University of Liverpool for sequencing on an Illumina Miseq platform. Sequence files associated with each sample have been submitted to the NCBI Sequence Read Archive (Accession no. PRJNA393972).

MiSeq sequencing data was initially processed using the mothur program developed by Schloss et al. (2009). Illumina adapter sequences were trimmed by cutadapt ver.2.1.1 using option -O 3. Sickle ver.1.2 was used to further trim the data with a quality score of ≥20. Reads <10 bp after trimming were removed. Each read was then trimmed to a maximum of 275 bp and ambiguous bases were removed. Sequences which contained homopolymer runs >8 bases were discarded. After trimming, identical sequences were grouped into ‘unique’ sequences. Chimeric sequences were identified using the UCHIME algorithm within mothur and were then removed. Sequences were assigned to operational taxonomic units (OTUs) using the ‘cluster’ command and the average neighbor algorithm. All subsequent OTU-based analyses were performed using a cutoff of 0.03. Taxonomy was assigned to all remaining aligned sequences by comparing processed data to the silva databases for bacteria and archaea independently¹.

Group means were calculated for VFA concentrations, average daily feed intake and average daily gains within treatment groups. Analysis of variance (ANOVA) and post hoc, Tukey HSD tests were carried out within the R software environment using the ‘lsmeans’ package to analyze differences between means. Multivariate analysis was carried out using PRIMER-E v.7 software with the PERMANOVA add on. Similarity matrices were constructed for samples using Bray-Curtis similarities on standardized, fourth root transformed abundance data. Distance-based permutational multivariate analysis of variance (PERMANOVA; Anderson et al., 2001) was then performed to test the null hypothesis that there were no differences in microbial community structure across treatments at a significance level of α = 0.05 based on 9999 possible permutations. nMDS plots were constructed to visualize the data. Similarity percentages (SIMPER) were calculated using Bray-Curtis similarities to evaluate the level to which each family contributed to the difference in community structures between groups. Effects and differences were declared as significant if p ≤ 0.05 and trends if p ≤ 0.2.

Twenty-three distinct bacterial phyla were identified in the rumen of the control group of lambs that were fed standard lamb meal throughout the study. Bacteroidetes (39.43%) and Firmicutes (34.29%) dominated the rumen microbiome of these lambs 16 weeks after birth. The relative abundance of Proteobacteria was also high (16.64%) in the developing rumen (Supplementary Table S1).

Of the bacterial genera detected in the lamb rumen, the relative abundance of the Prevotella was highest, accounting for 31% of total bacterial sequences (Figure 1A), and a further 7% of bacterial sequences were associated with unclassified members of this family. Other bacterial genera that were present in relatively high abundance in the rumen of these lambs were unclassified genera from Gammaproteobacteria (13%), Lachnospiraceae (12%), and Veillonellaceae (9%). Succinivibrio, Succiniclasticum and unclassified genera of Firmicutes and Clostridiales were also present but their relative abundance accounted for less than 3% of all bacterial sequences. It should be noted that 3% of bacterial sequences detected in the rumen of the control lambs could not be assigned taxonomically at any level and 72 OTUs could only be assigned to family level.

Only seven genera of archaea were identified in the rumen of lambs in the control (no linseed supplementation) group. Of these, Methanobrevibacter was by far the most relatively abundant genus followed by Methanosphaera (Figure 1B). These two genera alone accounted for 99.35% of all archaeal sequences identified across all samples within this treatment group. The remaining sequences were assigned to Methanosarcina, Methanomassiliicoccus, unclassified Thermoplasmata, unclassified Euryarchaeota and unclassified Nitrososphaera.

Prevotella (27.48%), unclassified Gammaproteobacteria (23.16%), unclassified Veillonellaceae (9.97%) and unclassified Lachnospiraceae (6.57%) also dominated the rumen of lambs receiving the diet supplemented with linseed oil throughout the study (L) (Figure 1A). n-MDS ordination revealed that bacterial communities in these lambs (L) appeared to cluster together (Figure 2A) and PERMANOVA analysis confirmed there were significant (p = 0.033) differences in rumen bacterial community structure of these animals compared to the control animals.

Similarity percentage (SIMPER) analysis revealed that the main contributors to the dissimilarity in bacterial community structures between the lambs fed the linseed diet and those fed the control diet included Succinivibrio, unclassified Gammaproteobacteriaceae, Selenomonas, unclassified Veillonellaceae, Dialister, Bifidobacterium, unclassified Prevotellaceae, Roseburia, Ruminococcus, unclassified Erysipelotrichaceae, Erysipelotrichaceae_incertae_sedis, and unclassified Firmicutes (Table 1). Together these bacterial genera explained 21.16% of the dissimilarity in bacterial community structures.

Clear clustering of archaeal communities in the control lambs from the lambs on the linseed supplemented feed was also observed (Figure 2B) and PERMANOVA confirmed these communities were significantly different and this difference was most pronounced at genus level (p = 0.005). Although the same two genera of archaea, Methanobrevibacter and Methanosphaera, dominated the rumen of lambs receiving the supplement, the relative abundance of these genera differed significantly (p = 0.005) from the non-supplemented lambs. The relative abundance of Methanobrevibacter was 58.75% in the lambs receiving linseed compared to 73% in those that didn’t. In contrast, the relative abundance of Methanosphaera was significantly higher in the animals receiving the supplement compared to animals receiving no supplement (40.82 and 26.67%, respectively).

No significant differences were observed in the concentrations of propionic, butyric or acetic acid produced by lambs fed the standard meal and those receiving the linseed oil supplementation (Figure 3).

The rumen bacterial community structure of lambs that only received linseed oil during the pre-weaning stage (L-P) was also found to be significantly (p = 0.015) different from the control group (Figure 2A). SIMPER analysis revealed that 14 bacterial genera including Succinivibrio, Bifidobacterium, Dialister, Lachnospiraceae_incertae_sedis and unclassified Gammaproteobacteria were responsible for 21.28% of the dissimilarity (Table 2).

Although there appeared to be increased archaeal species diversity in the group receiving linseed oil pre-weaning (L-P group) (Table 3B), community structure was not significantly (p = 0.12) different to that of the control group. There were no significant differences recorded in VFA concentrations between the L-P group and the control group (Figure 3).

Comparison of lambs fed the linseed oil supplemented feed for ∼14 weeks (L group) and lambs fed linseed oil pre-weaning only (∼4 weeks; L-P group) revealed significant (p = 0.003) differences between their rumen bacterial communities. Clear clustering of bacterial communities in these two groups of lambs can be seen on the nMDS plot (Figure 2A) and bacterial species diversity was higher (p = 0.02) in the group that only received the linseed oil during the pre-weaning period compared to the group that received it throughout (Table 3A). Furthermore, some differences were detected in VFA concentrations between these two groups of lambs, with a trend toward increased propionate and decreased butyrate in the animals receiving the supplement during pre-weaning only (p = 0.16 and p = 0.083, respectively). However, no significant differences in archaeal diversity (Table 3B) or community structure (p = 0.99 and 0.29, respectively) were observed.

Pearson’s correlation coefficient (r) was used to determine if any significant correlations could be observed between VFA profiles in the rumen of the lambs fed the different diets and any of the ten most relatively abundant bacterial genera (Figure 4). In lambs that received the unsupplemented diet throughout the study period (control), all the dominant genera except the unclassified Prevotellaceae were positively correlated with propionic acid. The unclassified Prevotellaceae were positively correlated with acetic and butyric acid (r = 0.15 and 0.75, respectively) while the other abundant genera had no correlation or were negatively correlated with these two VFAs.

A similar pattern was observed with the lambs fed the linseed oil supplemented diet in the pre-weaning period only (L-P group), with the relatively most abundant bacterial genera positively correlated with propionic acid with one exception, an unclassified Firmicutes (r = -0.1). However, in contrast to the control group, many of these genera were also positively correlated with butyric acid. Similar to the control group, most were negatively correlated with acetic acid.

A different profile was observed for the lambs fed the supplemented diet throughout the study period (L group). These lambs had less bacterial sequences positively correlated with propionic acid and negatively correlated with acetic acid. Although four of the most relatively abundant sequences were positively correlated with propionic acid, this correlation was weak. Relative abundances of Prevotella, Succinivibrio and an unclassified genus of Veillonellaceae were all negatively correlated with propionic acid concentrations (r = -0.33, -0.49, and -0.28, respectively). Relative abundances of Succinivibrio, and an unclassified Veillonellaceae were positively correlated with acetic acid (r = 0.79 and 0.47, respectively) while Succiniclasticum, Prevotella and an unclassified Gammaproteobacteria were positively correlated with butyric acid (r = 0.41, 0.25, and 0.35, respectively).

Similar to other reports (Stevenson and Weimer, 2007; Lee et al., 2012; Henderson et al., 2015; Lopes et al., 2016), Prevotella, was the most relatively abundant genus in the rumen of lambs fed a standard lamb meal, accounting for almost one third of total bacterial sequences present. This genus is considered part of the core rumen bacterial community and important for organic matter degradation (Henderson et al., 2015). Based on the physiology of culturable members of this genus, Prevotella spp. are thought to be propionate and succinate producers (Bryant and Small, 1956; Strobel, 1992). Unclassified genera of Gammaproteobacteria also had high relative abundance in the lambs’ rumen; this finding was also reported by Wang et al. (2017). Unclassified Lachnospiraceae, a family reported to be positively correlated with gross feed efficiency (Jewell et al., 2015) were also present at high relative abundance. Similarly, Veillonellaceae, another family in relatively high abundance in lambs fed the unsupplemented diet, were identified as gain-associated bacteria in the rumen of steers (Myer et al., 2015), although Li and Guan (2017) reported that members of this family were associated with low feed efficiency. The Veillonellaceae family are known to produce propionate as a major fermentation product (Myer et al., 2015). Prevotellaceae, Gammaproteobacteria, Lachnospiraceae, and Veillonellaceae have also all been recently identified as part of the core rumen microbiome in cattle (Li and Guan, 2017).

The high number of unclassified sequences detected in this study has also been reported by other groups (Creevey et al., 2014; Liu et al., 2016) and is not surprising given the large percentage of rumen bacteria that still remain unculturable. The generation of a reference set of rumen microbial genome sequences by initiatives such as the Hungate 1000 should facilitate more accurate classification in the future (Creevey et al., 2014).

Inclusion of linseed oil in the lambs’ diet for ∼14 weeks significantly changed ruminal bacterial and archaeal community structures. Similar results were observed by Lillis et al. (2011) when they analyzed the ruminal bacterial communities of adult cattle fed a diet supplemented with soya oil and those not. Soya oil, like linseed oil, contains polyunsaturated fatty acids and has also been shown to reduce enteric methane emissions when included in ruminant diets (Petrie et al., 2010). One of the top contributors to dissimilarity in bacterial communities between the control lambs and those fed linseed oil (L group) in the current study was the succinate producing (Russell and Rychlik, 2001) bacterial family, Succinivibrionaceae, which had a higher relative abundance in the L group. A previous study on tammar wallabies also reported high relative abundances of Succinivibrionaceae in the foreguts of wallabies and this was suggested to be an important contributing factor to the animals’ low enteric methane emissions (Pope et al., 2011). Veillonellaceae, known to be associated with propionate production (Reichardt et al., 2014; Lecomte et al., 2015; Myer et al., 2015), were present in a higher relative abundance in the L group while Ruminococcaceae, a family linked with higher CH₄ emitters (Kittelmann et al., 2014), was present in a lower relative abundance in the L group compared with the control group. Higher relative abundances of Veillonellaceae have been reported in the rumen of low methane emitting cattle (Wallace et al., 2015). Linseed oil suppressed the relative abundance of Erysipelotrichaceae, a family commonly found in the rumen of sheep. These results suggest that the lower levels of methane production associated with diets supplemented with linseed oil recorded in previous studies may be related to the selective promotion and suppression of specific bacterial families.

Although archaeal community structure was significantly different in lambs fed linseed oil (L) and the control group, the same two methanogens, Methanobrevibacter and Methanosphaera dominated both communities, albeit their relative abundance differed between treatments, with Methanosphaera relatively more abundant in the rumen of the L group. These two genera differ in their substrate specificity; Methanosphaera, spp., use CO₂, H₂ and formate, while Methanobrevibacter spp., use CO₂, H₂ and methanol (Liu and Whitman, 2008; Zhou et al., 2010). Thus differences in their relative abundance may reflect changes in substrate availability. There may also have been significant differences at species level that were not detected in the current study as certain substrates have been reported to favor certain species of methanogens which often have a narrow range of substrates (Danielsson et al., 2014; Henderson et al., 2015). Additionally, species-level differences in methanogenesis have been previously reported (Danielsson et al., 2017). Although methanogens are the sole producers of methane in the rumen, they are dependent on substrates, namely H₂ and CO₂, provided by bacterial and protozoal species to proliferate and produce CH₄ (Hook et al., 2010). Thus, changes in other rumen microbial communities will impact on methanogens. Indeed, acetate and hydrogen producing bacteria have been linked with higher methanogen abundance in the rumen, while propionate and succinate producers are correlated with lower methanogen abundance (Van Nevel and Demeyer, 1996; Knapp et al., 2014).

Volatile fatty acid concentrations were similar in the control and linseed fed lambs despite these significant changes in microbial community structure. Like other microbiomes, the rumen microbiome contains a high level of redundancy (Weimer, 2015) which could explain why differences in bacterial and archaeal diversity and bacterial VFA correlation profiles did not impact VFA concentrations/profile.

Lambs in the L-P group received linseed oil during the pre-weaning stage only and reverted to the control diet after weaning. Despite being on the same diet for approximately 10 weeks prior to sampling, the rumen bacterial community structure of the lambs in the L-P group differed significantly from those that never received any linseed oil (control group) and thus did not revert back to a structure similar to the control group once supplementation ceased. This is an interesting result, as manipulation of rumen microbiomes in later life typically only last days to weeks and this has been attributed to the redundancy and resilience present in the rumen microbiome (Hart et al., 2008; Weimer, 2015). There have been other reports of pronounced and long-lasting effects of early life dietary experiences (Distel et al., 1994; Yáñez-Ruiz et al., 2015). Both Yáñez-Ruiz et al. (2010) and Abecia et al. (2014b) reported persistently altered microbiomes and possible early-life microbial programming with diets supplemented with BMC. Primary colonizers of the gut have been shown to influence sequential colonizers (Chaucheyras-Durand and Fonty, 2002; Danzeisen et al., 2013) and thus early life diet may exert an effect by influencing primary colonizers which are important in determining the mature microbiome composition.

Even when the lambs only received linseed oil during the pre-weaning phase, most of the dissimilarity observed in bacterial community structure compared to the control group was once again due to the Succinivibrionaceae, which were relatively more abundant in the L-P group. Likewise, the Veillonellaceae were relatively more abundant in the L-P group compared to the control group whereas the Ruminococcaceae family were relatively more abundant in the control lambs.

Spirochaetaceae, a bacterial family commonly found in the rumen (Bekele et al., 2011) had a higher relative abundance in the control group than in the animals receiving the supplement during the pre-weaning period. Fibrobacter and Sharpea, found in higher abundance in low CH₄ emitters by Kittelmann et al. (2014) were found to be relatively more abundant in the control group in the current study than in either group of lambs receiving the supplemented diet pre-weaning or throughout. Overall these results may indicate that the linseed fed lambs had a microbiome more likely to be associated with reduced methane emissions.

Bifidobacteriaceae, reported to confer numerous health benefits on a host including regulation of intestinal microbial homeostasis, inhibition of pathogens, modulation of local or systemic immune responses, production of vitamins and bioconversion of dietary components to bioactive compounds (Mayo and Van Sinderen, 2010) were present in very low relative abundance (0.06%) in bacterial communities in control lambs compared to both lamb groups that received linseed oil, even those that received it pre-weaning only. This may be significant for the long-term health of the linseed oil fed groups.

Once again, despite the impact of linseed oil on microbial community structure and differences in the VFA correlation profiles, no difference was observed in the VFA profiles of the L-P and control lambs, again suggesting that overall fermentation was not impacted. These results suggest that dietary treatment early in life can cause alterations in the microbiome which persist even after the diet has been changed, suggesting that a window of opportunity may exist when microbial communities could be manipulated to facilitate improved health and/or production characteristics. However, the differences observed in the microbial composition of lambs receiving the linseed supplement pre-weaning only and those that received it throughout suggest that the effects of supplementation on the microbiome may be less predictable than desired.

Although rumen archaeal community structure was similar in lambs fed linseed oil throughout the study (L) and those fed the supplement pre-weaning only (L-P), significant differences were observed in bacterial community structures between these groups. This highlights the difficulty in predicting outcomes of interventions in such complex microbial networks. One possible explanation for the differing bacterial community profiles between the two groups is the effect of a second disturbance, in the form of a dietary change during a crucial milestone (weaning), on microbial communities. Multiple stressors can combine in different ways to give an overall effect which may be equal to the sum of the stressors (additive), equal to the effect of the most detrimental impact (simple comparative) or greater (synergic) or less than (antagonistic) the sum of individual disturbances. This makes the effect of multiple disturbances difficult to predict (Folt et al., 1999; O’Gorman et al., 2012; Vye et al., 2015). It is possible that introduction of an additional disturbance during early life microbial colonization of the rumen led to greater stress on the microorganisms and therefore an overall greater effect on microbial community structure. If dietary treatment had been continued through weaning (for an extra 2 weeks) the microbial profiles of the lambs fed linseed oil throughout the study and the lambs fed linseed oil short-term may have been more similar.

Although not significant, there did appear to be a trend of higher propionate and lower butyrate levels in lambs in the L-P group compared to those fed linseed oil throughout (L group). This may be relevant in the long-term for methane production. High butyrate concentrations in the rumen have been positively correlated with high methane emissions, while concentrations of propionate have been negatively correlated with methane emissions (Hungate, 1996; Lana et al., 1998; Russel, 1998; Mohammed et al., 2011). The fact that the relative abundance of three of the dominant bacteria (Prevotella, Succinivibrio, and unclassified Veillonellaceae) was negatively correlated with propionate concentration in lambs in the L group, but positively correlated with this VFA in the lambs fed the diet during pre-weaning only, suggests that either different processes were occurring in the two treatment groups or different bacteria were mediating these processes. It would be interesting to compare methane emission levels from lambs fed linseed oil short-term to those that continued to receive linseed oil for the entire study to see whether the microbial alterations achieved through short-term supplementation translated into persistent decreased methane emissions once supplementation had ceased.