New mechanistic insight into the digestion of complex dietary bre by rumen microbiota using combinatorial high-resolution glycomic and transcriptomic analyses

The rumen microbial community is considered the most ecient anaerobic digestive ecosystem known, yet less than half of the energy in low quality forages is actually metabolized. There is a knowledge gap regarding the specic factors that impede the ruminal digestion of plant cell walls or if rumen microbiota have the functional potential and activities to overcome these constraints. To address these issues, innovative experimental methods may provide a high-resolution understanding of the cell wall chemistries and higher-order structures that are resistant to microbial digestion and how they interact with the functional activities of the rumen microbial community. monoclonal antibodies, NDF: neutral detergent ber, PC: post-chlorite, PL: pectin lyases, RG: rhamnogalacturonan, SC: sodium carbonate, TTIR: total tract indigestible residue, XG: xyloglucan.


Abstract Background
The rumen microbial community is considered the most e cient anaerobic digestive ecosystem known, yet less than half of the energy in low quality forages is actually metabolized. There is a knowledge gap regarding the speci c factors that impede the ruminal digestion of plant cell walls or if rumen microbiota have the functional potential and activities to overcome these constraints. To address these issues, innovative experimental methods may provide a high-resolution understanding of the cell wall chemistries and higher-order structures that are resistant to microbial digestion and how they interact with the functional activities of the rumen microbial community.

Results
With this goal, we characterized the total tract indigestible residue (TTIR) from cattle fed a high-forage diet containing low-quality straw using two comparative glycomic approaches: ELISA-based glycome pro ling and glycosidic linkage analysis. Using these techniques, we successfully detected numerous and diverse cell wall glycan epitopes in barley straw and TTIR and determined their relative abundance pre-and postintestinal digestion. Of these, xyloglucans and heteroxylans were the most recalcitrant to digestion.
Linkage analysis identi ed indigestible linkages consistent with the polysaccharide epitopes identi ed by ELISA-based glycome analysis. To determine if residual plant polysaccharides within TTIR could be metabolised, rumen microbiota from cannulated cattle fed barley straw were incubated with barley straw and TTIR in in vitro batch cultures. Transcript coding for carbohydrate-active enzymes (CAZymes) were identi ed and characterized for their contribution to cell wall digestion based on glycomic analyses, comparative gene expression pro les, and associated CAZyme families. High-resolution phylogenetic ngerprinting of these sequences revealed encoded enzymes with activities predicted to cleave the primary linkages within heteroxylan and arabinan.

Conclusion
This experimental platform provides unprecedented precision in the understanding of forage structure and digestibility, which can inform next-generation solutions to improve the growth of ruminants fed low quality forages and enhance the use of crop residues as a feedstock.

Background
Lignocellulosic biomass can be converted into a broad suite of fuels, chemicals, and biomaterials. The recalcitrance of cellulosic material and high cost of hydrolysis to simple sugars is a major barrier to the commercial viability of bioproducts developed from cellulosic biomass. With an ever-growing human population and more a uent societies, the global demand for food, meat and milk is projected to increase exponentially [1,2]. Ruminant livestock are in a unique position to satisfy the growing demand for high quality protein, as ruminants can produce milk and meat via the microbial fermentation of cellulose-rich forages, crop residues, and food by-products. In this light, the rumen microbiota represents an underexploited repository of carbohydrate active enzymes (CAZymes) and microorganisms for applications in animal nutrition, biofuel, and bio-based chemical industries.
Globally, more than 73.9 million metric tons of crop residues are produced annually [3,4]. Although these residues could serve as feed for ruminants, usually less than half of the biomass they contain is digested.
Supplementing ruminant diets with exogenous enzymes has the potential to improve the utilization of crop residues for meat and milk production [5,6,7,8]. However, current commercial enzyme products have been largely developed for biore nery and bioprocessing applications and were not intended to function in the gastrointestinal tract of ruminants. Physiological conditions within the rumen are not favorable for some enzymes within cellulase mixtures [7,8], with many of these products possessing enzyme activities that are redundant to those already produced by the rumen microbial community. Quantifying the digestibility of neutral detergent ber (NDF) and acid detergent ber (ADF) in ruminants provides only cursory insights into the factors that limit plant cell wall digestion in the rumen. Immuno uorescence microscopy coupled with cell wall speci c monoclonal antibodies (mAbs) has proven to be a more informative method to investigate the polysaccharide composition of plant cell walls [9,10], yet an informed and more tailored approach is needed to investigate ruminal ber digestion. Such an approach should be anchored in a higher understanding of cell wall recalcitrance to ruminal enzymatic digestion and identi cation of rate limiting enzyme activities. Once rate-limiting enzymatic activities within the rumen are identi ed, CAZymes with synergistic activities could be used to enhance the digestion of ber by rumen microbiota.
Here, we have developed a platform for coupling glycomic analyses to comparative metatranscriptomes of rumen microbiota in batch cultures with barley straw and total tract indigestible residue (TTIR) as substrates. We aimed to characterize recalcitrant cell wall structures and identify candidate enzyme activities with the potential to enhance the rumen digestion of barley straw. Using this approach, we identi ed indigestible cell wall moieties within barley straw, and characterized the linkages that contributed to recalcitrance. A suite of target genes was selected for downstream production of recombinant enzymes that could potentially overcome the constraints to the digestion of barley straw plant cell walls. Although this study focused on the digestion of barley straw, the experimental pipeline holds high promise for the improvement of lignocellulosic feedstock from diverse plant sources in support of the sustainable conversion of lignocellulose biomass into industrial feedstock.

Glycome pro ling of barley straw and total tract indigestible residue
The sequential extraction of cell wall material leads to selective enrichment of various cell wall glycans into fractions based on the degree of integration into the plant cell wall [11]. Oxalate (AO) and carbonate (SC) fractions contain loosely bound pectic and arabinogalactan polysaccharides, while extraction with 1 M and 4 M KOH solubilizes hemicelluloses (e.g., xyloglucans and xylans) and tightly bound pectic polysaccharides. Chlorite (CH) treatment releases lignin and lignin-associated polysaccharides, while a subsequent 4 M KOH extraction after deligni cation recovers remaining recalcitrant hemicellulose and pectin fractions embedded within crystalline cellulose. Chemical fractionation of barley straw and TTIR followed by enzyme-linked immunosorbent assay (ELISA) identi ed differences in the glycome pro le of these substrates (Fig. 1). One major difference between barley straw and TTIR is the amount of carbohydrate recovered in the oxalate (AO) extracts of these two materials; about two and half times as much carbohydrate was recovered in the AO extract from TTIR as compared with barley straw. Yet, with the exception of the de-esteri ed homogalacturonan epitopes, which were enriched in the TTIR AO extract in comparison with the AO of barley straw ( Fig. 1; box 1), there was very little antibody reactive material in the AO extract of TTIR. This suggests that most of the carbohydrate extracted by AO is too small to bind to the ELISA plate, or too small to include the full epitopes recognized by the antibodies used in this study. In general, the AO and SC fractions of TTIR had minimal or lacked most xyloglucan (XG), xylan, type I rhamnogalacturonan (RG-I), and arabinogalactan (AG) moieties, suggesting that these carbohydrates were digested during passage through the ruminant digestive tract. In addition, there was a notable reduction in the AG-3 and AG-4 epitopes and in the lignin-associated MeGlcA-substituted xylans ( Fig. 1, Box 7) in TTIR compared with barley straw. In contrast, the somewhat loosely associated MeGlcA-substituted xylans present in the SC extract appear resistant to degradation during passage through the bovine digestive tract (Fig. 1, Box 2).
In the remainder of the glycome pro les, there appear to be selective enrichments in the TTIR for particular epitopes in comparison with barley straw. For example, there was a relative increase in the abundance of Gal-XG epitopes in the 1M and 4 M KOH extracts of TTIR compared with barley straw (Fig. 1, Boxes 3, 4), as well an increased enrichment of the Xylan-3 and 2-Ara-substituted xylans (Fig. 1, Boxes 5, 6). With diverse 6-linked β-galactan epitopes (common in RG-I and AGPs), there appears to be an enrichment in these epitopes in the 4M KOH extract (Fig. 1, Box 9), but a decreased presence of these epitopes in the 1 M KOH extract (Fig. 1, Box 8) when compared with barley straw. There also appears to be enrichment in some RG-I epitopes in the 4 M KOH extract from TTIR ( Fig. 1, Box 10) compared with the corresponding extract from barley straw. Lastly, there is an overall enrichment in the TTIR biomass for the most tightly bound glycans released in the 4 M KOH post-chlorite extraction (PC) compared with barley straw. It is important to note that CH and PC represent the most recalcitrant cell wall fractions. The persistent presence of crosslinked (glucurono) arabinoxylan, xyloglucan, and cellulose-embedded pectins (galactans, RG-I, and AG-2) in the KOH, CH and PC extracts of TTIR indicates their recalcitrance to microbial digestion within the ruminant digestive tract.
In order to evaluate the recalcitrance of barley straw to digestion by the rumen microbiota and identify ratelimiting enzymatic activities, the ten most abundant indigestible linkages were calculated as a ratio of abundance in TTIR over that of barley straw, and ranked highest to lowest (Fig. 2B). In the AIR fraction, t-Xylp was found to be highly abundant (1.6 ± 0.6), whereas 3,4-Xylp (5.4 ± 0.4), t-Manp (2.4 ± 0.7), and 2-Araf (2.6 ± 0.9) were the most indigestible linkages identi ed within the EDTA + Na 2 CO 3 , 4 M KOH, and cellulosic residue fractions, respectively.
Polysaccharide composition was estimated by assigning the linkage composition data to classes of polysaccharides according to Pettolino et al. [12] (Fig. 2C, Table 1, Additional File 1b). Differences in polysaccharide composition were observed between barley straw and TTIR samples. TTIR contained more arabinan (AB), HX and XG than barley straw in AIR and the nal cellulosic residual fractions. The EDTA + Na 2 CO 3 fraction of TTIR showed higher HX and XG content. AG-2 content within the TTIR 4 M KOH fraction was observed to be higher as compared to barley straw. Note: "t.r." means trace amount (mol%<0.1%). "n.a." means linkage not assigned to polysaccharide. No 4-Glcp linkage was assigned to cellulose as EDTA + Na 2 CO 3 , and 4 M KOH solutions do not extract this fraction [60]. 1 The estimation of polysaccharide was according to Pettolino et al. 2012 [12].
RNA-seq output and de novo assembly RNA-seq and differential gene expression analyses were conducted to study the composition of rumen microbial communities in vitro, and to compare microbial CAZyme expression pro les of rumen microbes cultured on TTIR or barley straw. Rumen microbial samples were batch-cultured in triplicate in vitro for each substrate (TTIR and barley straw), as reported previously [13], and transcriptomics analyses were performed. A total of 364,583,860 reads were used for quantitative RNA-Seq analysis. The number of reads per sample ranged from 56.95 million to 66.20 million. The transcriptome was assembled de novo using the Trinity assembler [14] and led to 1,998,343 distinct transcripts with a median transcript length of 336 (bp) and N50 of 533 (bp) (Additional File 2).
Microbial taxonomic classi cation based on putative mRNA.
RNA-seq raw reads were submitted as input to the Kaiju [15] webserver for taxonomical assignment at the whole transcriptome level. The results of the taxonomic classi cation with Kaiju were further processed and visualized with Krona [16]. Bacteria represented the majority (88%) of rumen transcripts (Table 2).
Note: Taxonomical a liations of whole transcriptomic data was obtained from taxonomic analyses of mRNA by Kaiju [15]. Relative abundance (%) at the domain level were calculated as a percentage of total mRNA transcripts, whereas relative abundance at the phyla and family level are calculated as a percentage of mRNA transcripts within the a liated domain. Only those phyla and families with relative abundance higher than 0.6% are reported.

Differential gene expression at individual transcript level
Identifying transcripts that catalyze the cleavage of carbohydrate linkages that limit ruminal digestion could provide insight into improving the digestibility of barley straw. Therefore, we focused on differentially expressed transcripts between barley straw and TTIR that are classi ed as CAZymes. Differential expression analyses identi ed a total of 88 glycoside hydrolases (GH) that were upregulated when rumen microbiota were cultured with TTIR, while in the presence of barley straw, 130 GH transcripts were (p < 0.01) upregulated (Table 3, Additional File 7). In TTIR samples, 7 PL and 9 CE transcripts were upregulated (p < 0.01), whereas 9 PL transcripts and 21 CE transcripts were upregulated (p < 0.01) with barley straw. At the family level (the total collection of differentially expressed transcripts within a given CAZy family), higher expression was observed in the presence of TTIR of CAZy families known to be putatively involved in degradation of cellulose (GH5), xylan (CE15, GH67), pectin (GH105, PL11) and mannan (GH26), but were not found to be statistically signi cant due to the high variation in expression levels across replicate batch cultures (Fig. 5, Additional File 7).

Discussion
Developing e cient technologies to enhance the conversion of low-quality forages into available energy within the rumen is pivotal for sustainable dairy and beef production [21]. Potentially, such technologies can reduce feed costs and improve the competitiveness of cattle management, while reducing the environmental footprint of meat and milk production [22]. Supplementing ruminant diets with exogenous enzymes has the potential to enhance feed e ciency [7,8]. However, current commercial enzymes have been tailored for bio-re nery and bioprocessing applications and as such are not optimized for the physiological conditions of the rumen [23,24,25]. Likewise, commercial enzymes often contain redundant enzyme activities already encoded within the rumen microbiota. Understanding the rate-limiting steps in ruminal ber digestion from the perspective of plant cell wall chemistry and endogenous catalytic potential is critical for the successful design of technologies based on enzyme addition, programming of the microbiome, and breeding plant cell walls that are more amenable to digestion by microorganisms within the ruminant digestive tract. Here we present an informed experimental platform (Fig. 8) that couples highresolution cell wall structural analysis with targeted meta-transcriptomic analysis of rumen microbiota so as to expand the present understanding of the factors that limit plant cell wall degradation.
Straw residues that escape ruminal total tract digestion (TTIR) can provide valuable insight into nutrient utilization by rumen and lower-tract microbes and into plant cell wall components that confer recalcitrance to the hydrolysis of straw. Presented here is the rst report using ELISA-based glycome pro ling to study the progression of ber utilization and recalcitrance during total tract digestion of ber in ruminants. This methodology has been extensively used to characterize plant cell walls including those of genetically modi ed plants, and microbially and chemically treated plant biomass [9,10,26]. Glycome pro ling provided us with an unprecedented ability to decipher the architecture of native barley straw and its residual fraction that is recalcitrant to digestion in cattle (TTIR) as it can detect diverse structural features within plant cell wall polysaccharide classes (Fig. 1). For example, a comprehensive suite of more than 25 mAbs that recognize different substructures, including each of the major side-chains of xyloglucans was applied, while another set of approximately 30 mAbs that recognize diverse xylan epitopes including Ara, GlcA and MeGlcA substitutions as well as various lengths of backbone residues [27] were used to assay xylans (Fig. 1). This not only allowed us to identify recalcitrant polymers with high precision, but also to differentiate polysaccharide subclasses that were removed during passage through the ruminant digestive tract or that exhibit high recalcitrance and were enriched in the TTIR. For instance, 4-O-Me-GlcA xylan epitopes that are lignin bound, as well as those xylans loosely associated with the cell walls in barley straw were removed during passage through the digestive tract (Fig. 1). Alternatively, 2-Ara-substituted xylans, as well as xylan epitopes recognized by the Xylan-2 and Xylan-3 groups of antibodies, were enriched in TTIR, suggesting that these epitopes directly contribute to recalcitrance. Relative proportions of xyloglucan and xylan epitopes in TTIR remained largely unchanged from the proportions of these epitopes in native barley straw (Fig. 1). However, it was clear that glycome pro ling alone could not fully describe the glycan composition of TTIR. For example, there was a signi cant fraction of glycans that were loosely associated with the TTIR (AO extract), yet, with the exception of de-esteri ed HG epitopes (DP > 4), there was no antibody binding to this extracted fraction from TTIR. This lack of antibody binding may be attributed to the fact that short-chain oligosaccharides released during digestion fail to bind to ELISA plates [10]. Alternatively, while the antibody collection used here recognizes a large and diverse set of plant cell wall glycan epitopes, there may be other epitopes native to barley straw or generated during passage through the ruminant digestive tract that are not recognized by the current antibody library. Therefore, to con rm glycome pro ling results we quanti ed individual glycosidic linkages using partially methylated acetyl alditol derivatization, which measures all glycan linkages present. The linkage data showed that the TTIR was enriched in HX, XG, and AB when compared with barley straw at the whole cell wall level (Fig. 2, Table 1), suggesting that these components were more recalcitrant than other components within this biomass. Fractionation of the walls showed that the EDTA + carbonate extract was strongly enriched in HX and XG (Fig. 2, Table 1), which was in contrast to the results from glycome pro ling, where the equivalent extract (AO) only showed enhanced HG content (Fig. 1). This may re ect the composition of the large fraction of cell wall glycans that are loosely associated with TTIR, but which were not assayable in the glycome pro ling ELISAs for the reasons mentioned above. Consistent abundance of HX and XG within the whole cell wall and the EDTA/carbonate and KOH extracts of TTIR suggest that these epitopes contribute to the recalcitrance of barley straw within the ruminant digestive tract. Apart from cell wall composition, linkage analysis also provided useful insight into the predominant bonds within identi ed recalcitrant polysaccharides. The results presented in this study could prove to be a valuable resource for crop breeders to develop feed stocks that exhibit increased digestibility in ruminants.
Identi cation of predominant recalcitrant bonds coupled to transcriptomic and differential expression analyses, provided two complementary approaches to identify candidate linkages to target for improved plant cell wall digestion. To identify potential enzymes tailored for the digestion of recalcitrant linkages, TTIR and barley straw were used as substrates to culture rumen microbial communities in batch cultures. Incubation with TTIR provided an unique experimental approach that enabled the rumen microbial digestion of recalcitrant plant cell walls to be studied. This is the rst such attempt where recalcitrant residues like TTIR have been used to study rumen community composition and CAZyme expression. Nutritional value and effective fermentation of TTIR was con rmed by ELISA results and through comparison of total gas production using barley straw and TTIR as substrates in in vitro cultures (Additional File 8). Firmicutes (Lachnospiraceae, Ruminococcaceae, Clostridiaceae), Bacteroidetes (Prevotellaceae, Bacteroidaceae and unclassi ed Bacteroidiales), Actinobacteria, Fibrobacteres, Proteobacteria and Spirochaetae were found to be dominant bacterial phyla at the transcriptome level (Table 2). These results are in agreement with previously reported metatranscriptomic analysis, such as Dai et al. [28], that reported Firmicutes, Bacteroidetes, Spirochaetes, Proteobacteria, Actinobacteria and Fibrobacteres are the most abundant bacterial phyla. Likewise, Comtet-Marre et al. [29] found that bacteria represented 77.5% of rumen ssu rRNA reads with Firmicutes, Bacteroidetes, Fibrobacteres, Proteobacteria, Spirochaetae and Lentisphaerae as predominant phyla. Although the present study found a similar core microbiome, there was variation in the abundance of taxonomic families as compared to previous literature [28,29]. This variation may be attributed to the shorter incubation time (48 h), and absence of host effects in closed in vitro batch culture system as compared to in vivo animal-based studies. Shorter incubation times favor rapidly growing bacteria over the slower growing fungal and protozoal members of ruminal microbial communities. In contrast to the host environment where fermentation by-products like VFA and ammonia are absorbed across the intestinal epithelium, closed batch culture systems accumulate by-products, which can alter the fermentation process towards certain types of microbial communities.
Furthermore, host genetics have also been reported to in uence microbial populations [30], as a region on chromosome 6 in cattle has recently been shown to be associated with Actinobacteria, Euryarchaeota and Fibrobacteres densities [31]. While whole animal experiments can not speci cally study the digestion of TTIR, in vitro batch cultures provided us with the level of control over experimental variables required to investigate those factors that limit plant cell wall digestion.
Gene expression analyses of rumen microbial communities cultured on barley straw and TTIR demonstrated substrate-speci c modes of digestion, as evidenced by the expression levels of CAZy family transcripts. At the family level, higher expression of putative endo-glucanases GH5, GH9, and GH8 on TTIR indicated an abundance of metabolizable cellulose within TTIR, as con rmed in the linkage analyses. Furthermore, GH48 cellobiohydrolases were the dominant exo-glucanases expressed during the digestion of TTIR by rumen microbes. In contrast, GH6 and GH48 were the predominant families associated with the microbial digestion of barley straw (Additional File 6B). Both GH6 and GH48 are known to e ciently hydrolyze the crystalline regions of cellulose [28,32]; while cellobiohydrolases of family GH48 act on the reducing ends of cellulose chains, and GH6 attack non-reducing ends, generating cellobiose [33]. The high expression levels of these exo-glucanases suggest there is a greater abundance of amorphous cellulose at the reducing ends of micro-brils in TTIR as compared to barley straw, as result of its exposure to CAZymes during passage through the ruminant digestive tract. The level of expression of GH43 hemicellulases in both TTIR and barley straw was higher than other hemicellulose-targeting activities (Additional File 6E). Among the 501 bacteria in the Hungate catalogue, the GH43 family was reported as the most abundant hemicellulase [34]. Metagenomic [35][36][37] and metatransciptomic [28,29,[38][39][40] studies have suggested that members of the GH43 family are the principal hemicellulases within the rumen. GH43 encode a range of debranching enzymes, including arabinofuranosidases, xylanases, galactanases, arabinanases and β-xylosidases, which aid in the degradation of arabinoxylan [41]. Indeed, glycomics revealed that arabinoxylan was enriched in TTIR, highlighting its resistance to ruminal digestion. Additionally, abundance of branched linkages like 3,4-Xylp, t-GlcAp, 2,3,4-Xylp, t-Araf in TTIR also con rm high resistance of heteroxylan to degradation by rumen microbiota (Fig. 2). Metatranscriptomic analyses re ected the efforts of the rumen microbiota to degrade this polysaccharide, with the GH10 family being the dominant family of xylanases (Additional File 6D). This family has been previously reported to digest heteroxylan and release xylotriose with a methylglucuronic acid (MeGA) moiety at the non reducing end which is hydrolyzed by members of the GH5 family of xylanases [42]. The GH5 xylanases have been shown to be active against glucuronoxylan and MeGA substituents on the xylan backbone [42]. Pectin polysaccharides differed between barley straw and TTIR (Figs. 1 and 2). Accordingly, expression levels of pectin-targeting CAZy families differed between rumen microbiota incubated with TTIR vs barley straw (Additional File 6F). Extracted rumen microbial communities cultured on barley straw expressed more GH78s, a family that contains α-l-rhamnosidases that act on rhamnogalacturonan and arabinogalactan-protein linkages [43]. In contrast, with TTIR there was a higher expression of GH28, a family that encodes polygalacturonases and rhamnogalacturonases that hydrolyse the α-1,4 galacturonate linkage in HG and RG-I respectively; and GH53s, which encode for β-1,4-galactanases [44].
Metatranscriptomics of barley straw and TTIR cultures provided a broad picture of the enzymatic potential of the rumen in terms of CAZy families. However, to gain more insight into the enzymatic processes of ruminal digestion, differential expression analyses, a more nuanced approach with in-depth analyses was used. Based on differential expression, several GH families putatively involved in cellulose, hemicellulose, and pectin digestion were identi ed (Fig. 5). Informed by glycomics, the relative contribution of dominant DE transcripts was determined for candidate CAZyme families with a focus on those activities important for digestion of recalcitrant plant cell walls (Fig. 6). Since candidate transcripts were expressed by the microbial community in the presence of TTIR, it is highly likely that these transcripts coded for enzymes targeted at inaccessible or non-hydrolyzable linkages within recalcitrant polymers of TTIR. As identi ed transcripts represent uncharacterized microbial genes, phylogenetic analysis using SACCHARIS was employed in an attempt to postulate enzyme function and speci city for DE transcripts from the selected families GH5, GH11, GH43, and CE15 important for XG, HX, and AB sacchari cation. No signi cant difference was seen between barley straw and TTIR in the relative contribution of GH5 and GH9 family DE transcripts (Fig. 6), suggesting that rumen microbes may lack specialized xyloglucanases for the utilization of XG, despite its abundance within TTIR. Thus, xyloglucanases may represent a rate-limiting enzyme activity in the rumen, and exogenous supplementation of xyloglucanases may promote straw digestion within this environment. Although no xyloglucanases were identi ed, substrate-speci c gene expression targeted towards HX and AB digestion was evident from phylogenetic trees of GH11, GH43 and CE15 DE transcripts, whereby the expression of specialized xylanases and arabinanases were induced within the rumen microbial community for the digestion of TTIR. Interestingly, GH11 transcripts that were overexpressed in TTIR cultures clustered distantly from the barley straw overexpressed GH11 transcripts, and some of the early diverging sequences, are GH11-like, and may represent members of a new family (Fig. 7A). The divergence of TTIR-overexpressed transcripts also suggests that changes in expression levels is a result of changes of the predominant bacterial species, differing between TTIR and barley straw microbes for utilizing recalcitrant HX. Furthermore, the TTIR-overexpressed GH11 transcript TR83509|c0_g1 was observed to contain an unique arrangement with four tandem GH11-like domains (Fig. 7A), possibly the rst report of a novel repetitive xylanase architecture. The phylogenetic analyses of the overexpressed GH43 transcripts revealed that both TTIR (TR712918|c0_g1) and barley straw (TR941931|c2_g1 and TR300921|c2_g2) clustered together with endo-α-1,5-l-arabinanases (Fig. 7B), suggesting differential regulation of arabinanase activity in rumen microbiota cultured on barley straw vs TTIR. The recalcitrance of arabinan to microbial digestion was also re ected by linkage analysis and glycome pro ling (Figs. 1 and 2; Table 1). Transcripts TR400300|c0_g2 and TR451123|c0_g1 were overexpressed in incubations with barley straw and clustered with arabinofuranosidases, refective of the cross-linked nature of HX in barley straw (Fig. 7B). Likewise, the three overexpressed CE15 transcripts with TTIR were embedded among acetyl xylan esterase activity, further re ecting the recalcitrant nature of HX.
The results shown in this study also identi ed 5-linked Araf within arabinan, 3-linked Araf within arabinoxylan and Xyl-3AC and Xyl-2AC within acetyl-xylan as linkages that resisted microbial digestion (Fig. 7). These results suggest that rumen microorganisms are quite e cient in utilizing loosely bound pectin and hemicellulose, but are limited in their ability to digest cross-linked core pectin and hemicellulose.
Previous metatranscriptomic studies have been mostly limited to GH family levels with limited or absent information regarding substrate speci city of identi ed transcripts [28,29,[38][39][40]. In this study, transcriptlevel SACCHARIS analyses in combination with glycome pro ling and linkage analysis suggested that substrate-speci c CAZyme transcript expression is inducible. The TTIR-overexpressed transcripts and putative activities may identify those CAZymes with the greatest potential to improve the rate and extent of recalcitrant plant cell wall residues by the rumen microbial consortia. Identi ed target CAZymes of rumen origin are likely to stand a better chance of success when applied as additive enzymes as they are optimized for the complex physiological conditions within the rumen. Future recombinant production and characterization of these candidate CAZymes and their screening is needed to ascertain their synergy to the natural compliment of CAZymes produced by rumen microbiota. In addition, the barley straw overexpressed transcripts and their closely clustered CAZymes signi es important enzyme activities for effective barley straw digestion and hold high intellectual value for their wide application to industrial sectors such as bioenergy, and green chemistry. Here we have shown that integrating high-resolution analytical methods, such as cell wall structural characterization, in vitro batch culture, RNA-sequencing and DE analysis (Fig. 8), can be successfully used to identify recalcitrant polymers and resistant linkages within feeds, as well as inform the biological processes involved in ruminal digestion of polysaccharides.

Conclusion
A comparative study involving changes in cell wall glycan composition, glycosidic linkages, and the relative extractability of polysaccharides within TTIR and barley straw provided an unprecedented level of understanding of nutrient utilization by rumen microbes and helped to delineate the role of matrix polysaccharides in cell wall recalcitrance towards microbial enzymatic digestion. Adopting TTIR as a substrate for batch culture of rumen samples and the application of next generation sequencing to study gene expression by the rumen microbial community, successfully identi ed a number of CAZyme genes displaying potential for downstream enzyme product development and degradation of recalcitrant residues. Furthermore, incubation of rumen microbial communities on TTIR has proven a powerful approach for enzyme discovery. In this regard, characterization of the small oligosaccharide fraction associated with TTIR may allow targeted searches from other organisms for hydrolytic enzymes capable of digesting these oligosaccharides, whose addition may enhance the rumen digestion of recalcitrant forages. Taken together, the results presented in this study suggest that the cell wall layered structure hinders enzymatic accessibility to recalcitrant cell wall components like HX and XG. Considering the rumen is one of the most e cient biomass digestive systems known, the results of this study hold high value for bioproduct and bioprocess advances with lignocellulosic feedstocks. The integration of high-resolution glycomics with targeted metatranscriptomics analyses is an innovative experimental platform that is expandable to other feed-host systems and industrial fermentation processes as it holds vast promise for addressing rate-limiting reactions in digestive ecosystems.

Methods
Experimental setup and collection of samples. In vitro batch culture was used to culture mix rumen microbes on barley straw and total tract indigestible residues (TTIR) as substrates, as described previously [12]. TTIR was obtained from faecal samples collected from heifers (six) fed barley straw-based diets containing 70:30 forage-to-concentrate on a dry matter (DM) basis [45]. Collected faecal samples were pooled and extensively washed (6-7 times) with 50 mM citrate buffer to remove soluble particles. Given the high digestibility of the concentrate, it was assumed that the majority of TTIR originated from barley straw. The fecal residue obtained was dried for 72 h in an oven at 40 °C and de ned as TTIR for use in batch culture experiments. Barley straw was dried at 40°C and ground to pass through a 1 mm screen.
In vitro batch cultures were set up in 3 replicate serum vials. Barley straw and TTIR were weighed (0.7g/bag) into acetone-washed and pre-weighted lter bags (F57 ANKOM bag, Ankom Technology Corp., Macedon, NY) that were then heat sealed. Individual bags were then placed in 125 mL serum vials. Rumen uid from 4-5 different sites within rumen was collected from four ruminally cannulated Angus x Hereford heifers fed 50% grass hay, 30% barley straw, 15% corn dried distillers grains plus solubles and 5% mineral/vitamin supplement (DM basis) [13]. Collected rumen uid was strained through cheesecloth and equal volumes from each cow were combined. The inoculum was prepared by mixing rumen uid 1:4 with mineral buffer [13]. Inoculum (65 mL) was transferred to each vial under a stream of O 2 -free CO 2 . Vials were sealed with rubber stoppers and placed in a rotary shaker (120 rpm) in an incubator at 39 °C. Gas pressure in each vial was measured at 3, 6, 12, 24 and 48 h of incubation by inserting a 23 gauge (0.6 mm) needle attached to a pressure transducer (model PX4200-015GI, Omega Engineering, Inc., Laval, QC, Canada) and used to estimate gas production according to Romero-Pérez and Beauchemin (2018) [46]: Gas volume = 4.7047 × (gas pressure) + 0.0512 × (gas pressure 2 ).

RNA extraction and sequencing
Total RNA from the solid bound ruminal microbes were extracted as described previously [47]. Brie y, the solid contents within nylon bags from in-vitro batch cultures were recovered upon completion of 48 h of incubation and manually ground to a ne powder using a mortal and pestle for 5 min in liquid nitrogen.
Ground samples (~ 200 mg) were placed in 2 mL microfuge tubes and each was mixed with 1.5 mL of TRIzol reagent. The samples were thawed, incubated at room temperature for 5 min, and subsequently the RNA was extracted using the acid guanidinium-phenol-chloroform (AGPC) method [48]. Total RNA was further puri ed with MEGAclear kit according to manufacturer instructions (Applied Biosystems/Ambion).
The RNA concentration and integrity were estimated using an Agilent 2100 bioanalyzer (Agilent Technologies, Mississauga, Ontario, Canada) and RNA 6000 Nano kit (Agilent Technologies) according to the manufacturer recommendations. RNA sequencing was conducted on rRNA depleted library using the HiSeq 4000 PE100 platform at McGill University-Genome Quebec Innovation center.
Data analysis and bioinformatics RNA-seq raw reads were submitted as input to the Kaiju [15] webserver for taxonomical assignment at the whole transcriptome level. Kaiju was run in greedy mode with a minimum match length of 11 and a minimum match score of 55, with 5 mismatches allowed. The results of the taxonomic classi cation with Kaiju were visualized with Krona [16].
The raw sequenced reads were further processed using MUGQIC RNA-Seq De Novo assembly and differential analysis pipeline [49]. Raw sequenced reads were trimmed and clipped for sequencing adapters, low quality and short sequences were ltered using Trimmomatic [50], normalized and assembled using Trinity normalization utility inspired by the Diginorm algorithm [51]. The sequenced reads were assembled by the de novo transcriptomic assembly program Trinity. Transcripts from Trinity were aligned against the uniprot_sprot.trinotate_v2.0 protein database using the BLASTx program. The Trinotate suite was used for homology searches to known sequence data (BLAST+/SwissProt/Uniref90), protein domain identi cation (HMMER/PFAM), protein signal peptides and transmembrane domain prediction (signal/tmHMM), and for comparison to currently curated annotation databases (EMBL Uniprot eggnog/GO Pathways databases). The derived functional annotation data were integrated into SQLite database to prepare the annotation report. To identify putative carbohydrate active enzymes, all sequences were pro led against the CAZy database [16]. Gene abundance was estimated using RSEM and differentially expressed transcripts between substrates were identi ed using DEseq [52] and edgeR [53] package from Bioconductor. Sequence assemblies were deposited in GenBank under BioProject PRJNA673210.
For DE transcript functional analyses, DE transcript nucleotide sequences were re-submitted to the dbCAN2 meta server [19], manually curated, and categorized by CAZy family. The amino acid sequences from the dbCAN-annotated CAZymes were submitted as input to SACCHARIS for phylogenetic analyses [20]. Sequences and accession numbers of characterized GH11, GH43, and CE15 enzymes were extracted from the CAZy database [17], and ProtTest [54] was used for best-t model selection using the sequence alignment. FastTree [55] was used to generate the phylogenetic trees which were then annotated using iTOL [56].  [57]. These plant cell wall extracts were probed with a collection of 154 cell wall glycandirected mAbs [58] using an enzyme-linked ELISA, and binding intensities from this experiment presented as a heat map as described previously [57].
Linkage analysis of barley straw and total tract indigestible residue Alcohol insoluble residues (AIRs) were prepared according to Wood et al. [59]. The dried AIRs were destarched by incubation with porcine pancreatic α-amylase in 10 mM Tris-maleate buffer (pH 6.9) following Pettolino et al. [11]. The de-starched AIRs were further fractionated based on the literature [60][61][62][63] with modi cations. Brie y, the cell walls (~55 mg) were treated with 10 mL of 0.25 M sodium borodeuteride (NaBD 4 , 99 atom % D, Alfa Aesar) deionized water solution at 4°C for 24 h, followed by quenching the excess reductant by dropwise addition of 10% (v/v) acetic acid and adjusting pH to neutral. The suspensions were then centrifuged (3000 × g, 0.5 h), and resulting supernatants were pooled with three deionized? water washes of the pellets (with centrifugation between each wash), and a pooled solution (Solution 1) was stored at 4 °C. The pellets were then subjected to sequential extraction of polysaccharides using 15 mL of each of the following solutions: 50 mM EDTA (pH 6.5), 50 mM Na 2 CO 3 (containing 25 mM NaBD 4 ) , and 4 M KOH (containing 25 mM NaBD 4 ) over 24 h at room temperature with gentle magnetic stirring. After extraction, soluble fractions were separated from residues by centrifugation (3000 × g, 0.5 h), followed by neutralizing the supernatants, washing the residues three times with deionized water (with centrifugation between washes), and pooling all washes into corresponding fractions. The EDTA and Na 2 CO 3 extracts and the aforementioned Solution 1 were pooled (designated F EDTA+Na2CO3 ). The 4 M KOH fraction and the nal residue left after strong alkaline extraction were designated F 4MKOH and F Residue , respectively. All fractions were dialyzed against deionized water and lyophilized. For linkage analysis of the whole cell walls and the isolated fractions, uronic acids in the samples (~2 mg) were rst converted to their 6,6-dideuterio neutral sugars using carbodiimide activation at pH 4.75 followed by NaBD 4 reduction at pH 7.0. Samples were then dialysized against deionized water, and freeze-dried. The carboxyl-reduced samples were converted to their partially methylated alditol acetate (PMAA) derivatives by permethylation with iodomethane and sodium hydroxide in dimethyl sulfoxide [64,65] with 2 M tri uoroacetic acid at 121°C for 1.5 h, reduction with NaBD 4 , and peracetylation with acetic anhydride [12]. Deutero-methylation using iodomethane-d3 (≥99.5 atom % D, Sigma) was applied to the whole cell wall and F 4MKOH fractions in order to identify and quantify the linkages from endogenously O-methylated sugars (e.g., 4-O-methyl glucuronic acids) [66][67][68]. The PMAAs were then subjected to GC-MS and GC-FID analyses on an Agilent 7890A-5977B GC-MS/FID system (Agilent Technologies, Santa Clara, CA). The PMAAs were separated using a medium polarity Supelco SP-2380 capillary column (30 m × 0.25 mm × 0.20 μm, Sigma-Aldrich) with a constant column outlet helium ow rate of 0.8 mL/min. Sample solutions were injected at an inlet temperature of 250°C with a split ratio of 10:1. The oven temperature was programmed to start at 120 °C (hold 4 min) followed by increasing at 8 °C/min to 175 °C, 0.5 °C/min to 183 °C (hold 8 min), 0.5 °C/min to 195 °C, 4 °C/min to 210 °C, and 20 °C/min to 255 °C (hold 8 min). The transfer line temperature was kept at 280 °C. The mass spectrometer was operated in electron ionization (EI), full-scan mode (ionization energy, 70 eV; source temperature, 230 °C; quad temperature, 150 °C). The FID detector was operated at 300 °C (H 2 ow, 30 mL/min; air ow, 400 mL/min; N 2 makeup ow, 25 mL/min). The PMAAs were then identi ed by comparing their MS fragmentation patterns with those of reference derivatives and the literature [69], and quanti ed based on the previously reported FID response factors calculated using the effective carbon number concept [70]. Glyosidic linkages were assigned to relevant cell wall polysaccharides, and the relative compositions of each type of polysaccharides were then estimated by summing up corresponding linkage compositions [12,71]. Six separate experiments were conducted for the whole cell walls and the F 4MKOH fractions, of which half were used for deuterio-methylation analyses and the other half for methylation analyses. The F Residue fractions were subjected to six separate experiments (methylation analysis only). Three separate methylation analyses were conducted for the F EDTA+Na2CO3

Consent for publication
No consent was required for this publication Availability of data and material Sequenced transcripts have been deposited into the NCBI SRA database BioProject accession number: PRJNA673210.
Competing interests