*4.5. Microbiological Analyses*

#### 4.5.1. Microbial DNA Extraction

The extraction of microbial genomic DNA was conducted using a QIAamp DNA stool kit (Qiagen, West Sussex, UK) according to standard protocols. After DNA extraction DNA was evaluated using the Nanodrop spectrophotometer (Thermo Scientific, Wilmington, DE, USA).

## 4.5.2. Illumina Sequencing

The V3–V4 hypervariable region of the bacterial 16S rRNA gene was performed on an Illumina MiSeq platform according to their standard protocols (Eurofins, Wolverhampton, UK). The V3–V4 region was PCR-amplified with universal primers taking in adapters including nucleotide sequences for forward and reverse index primers. Amplicon purification was conducted with AMPure XP beads (Beckman Coulter, Indianapolis, IN, USA) and prepared for index PCR using Nextera XT index primers (Illumina, San Diego, CA, USA). The purification step was repeated on the indexed samples using AMPure XP beads and assessed using a fragment analyzer (Agilent, Santa Clara, CA, USA). Following this step a pool was made using equal quantities from each experimental sample. The library was then analysed using the Bioanalyzer 7500 DNA kit (Agilent) and sequenced using the V3–V4 chemistry (2 × 300 bp paired-end reads).

#### *4.6. Bioinformatic and Statistical Analyses*

Quantitative Insights into Microbial Ecology (Qiime) was used to examine the sequencing data [42]. Sequencing primers were removed using the cutadapt package and the resulting paired end reads were merged using the paired-end reads function within Qiime using the standard criteria. Demultiplexing of paired end raw reads was through the split libraries function and quality filtering was conducted utilizing default QIIME parameters. Only reads that contained no ambiguous characters, no non-exact barcode matches, a sequence length > 225 nucleotides and a read-quality score of >27 were retained. The uclust function in Qiime was used to pick OTUs based on a sequence similarity of 97%. Singletons were removed, as only OTUs that were present at the level of at least two reads in more than one sample were retained while chimeric sequences were removed using ChimeraSlayer [43,44]. The GreenGenes database assigned OTUs to different taxonomic levels. A combination of the mormalized OTU table, experimental phenotypic data and the phylogenetic tree were combined to produce the phyloseq object for further analysis (http://www.r-project.org; version 3.5.0, accessed on 25 March). This phyloseq R package was used to examine measures of richness and diversity using the Observed, Chao1, ACE, Shannon, Fisher and Simpson measures as described by Maurer et al. [45]. The PROC GLIMMIX procedure of Statistical Analysis Software (SAS) 9.4 (SAS Institute, Cary, NC, USA) was used to examine differences between experimental treatments at phylum, family and genus level with pig being the experimental unit and *P*-values presented using a Benjamini–Hochberg (BH) correction.

#### *4.7. Nanostring nCounter Analysis*

Tissue from the duodenum, jejunum and ileum were used for analysis of gene expression profiles using the Nanostring nCounter Analysis System (Nanostring Technologies, Seattle, Washington, USA). Gene lists contained six positive and eight negative controls, eight reference targets and thirty-two target genes. A single multiplexed hybridisation reaction, as originally described by Geiss et al. [46] was used to analyse all target genes. Samples were initially standardized to 20 ng/μ<sup>L</sup> using a Qubit fluorometer (Thermo Fisher Scientific). A master mix was created by combining 70 μL of hybridisation buffer and the reporter codeset, as per the manufacturer instructions. For the hybridization reaction each tube contained the master mix (8 μL), the sample (5 μL) (total RNA concentration 100 ng) and capture probeset (2 μL). Each reaction tube was inverted to mix and spun down before incubation at 65 ◦C for 20 h in a Bio-rad thermocycler.

The Nanostring nCounter prep station liquid handling robot which was used for post-hybridisation processing. This involved the removal of excess unbound probes and immobilisation of samples onto the internal surface of the sample cartridge to allow imaging to be conducted using the digital analyser, which collects data by taking images of the immobilised fluorescent reporters in the sample cartridge with a CCD camera through a microscope objective lens.

The analysis and normalisation of the raw Nanostring data were performed using the nSolver Analysis Software v4.0 (Nanostring Technologies) which was used for initial data normalisation and analysis. The background threshold value was estimated using the average count of the negative control probes in every reaction plus 2 standard deviations to which all samples were adjusted [47]. Gene targets with raw counts below the threshold in more than two thirds of samples were excluded from the analysis. Raw counts were normalised using a combination of positive control normalisation and codeset content normalisation. The former accounts for errors such as pipetting errors, lot-to-lot variation in nCounter preparation plates and nCounter cartridges, while the latter uses reference/housekeeping genes to account for variability in the quantity and quality of sample RNA.

#### *4.8. Analysis of Performance and Gene Expression Data*

The univariate procedure of SAS 9.4 (SAS Institute, Cary, NC, USA) was used to check performance, gene expression and VFA data for normality. Experimental data were analysed as a complete randomized design using the mixed procedure of SAS with the fixed effect of treatment. The initial weight was used as a covariate for the performance data with pen being the experimental unit. For all other data, the pig was the experimental unit. The probability level that denoted significance was *P* < 0.05, while *P*-values between 0.05 and 0.1 are considered numerical tendencies. Data are presented as least-square means with their standard errors of the mean.
