A Multi-Omic Huntington’s Disease Transgenic Sheep-Model Database for Investigating Disease Pathogenesis

The 5-year-old OVT73 sheep cohort

All datasets were generated from samples obtained from the same cohort of 5-year-old South Australian Merino (n = 12). These animals were generated using juvenile in-vitro embryo transfer (JIVET), involving artificial insemination of eggs harvested from 6-week-old transgenic ewes, with all animals born as singletons from surrogate mothers. The cohort consists of two groups of siblings; six G1 generation animals (Table 1) were the progeny from crossing the founder (G0) transgenic ewe with a control ram (Ram A), six G2 generation animals (Table 1) were the progeny from crossing another G1 transgenic ewe (a full sibling to the six G1 animals in the cohort) with another control ram (Ram B). The sheep were maintained at the South Australian Research and Development Institute (SARDI) in accordance with the SARDI/PIRSA (Department of Primary Industries and Regions South Australia) Animal Ethics Committee (Approval number 19/02). Animals were kept as part of mixed wild-type/transgenic flocks in large paddocks typical of South Australian farming conditions, grazing pasture ad libitum with feed supplementation during dry season periods when there was little or no pasture feed on offer. Rams and ewes were kept in separate flocks but in the same locality. In June 2012, the animals were euthanized, and tissue necropsy was performed at the Large Animal Research and Imaging Facility (LARIF), Adelaide, Australia in accordance with the SARDI/PIRSA Animal Ethics Committee (Approval number 05/12) and LARIF/SA Pathology Animal Ethics Committee (Approval number 31/12). Sheep were killed by rapid intravenous injection of pentobarbitone sodium solution (Lethabarb, 1 ml/2 kg body weight). Brain and peripheral tissues were immediately sampled and snap frozen in liquid nitrogen as described in [10]. Transgenic status of each sheep was confirmed by PCR amplification of the transgene from tail-tissue genomic DNA using methods described in [16]. All samples were stored at –80°C until use. Necropsy and biometric data for each animal is summarised in Table 1.

Datasets used in this research

Seven individual datasets sourced from OVT73 researchers were incorporated into the HD sheep platform/database (Table 2). These datasets were generated from brain and peripheral tissue samples taken from the 5-year-old cohort (OVT73 = 6, Control = 6). Included are two high-throughput RNA-Seq datasets collected from sub-regions of the striatum, a targeted transcription dataset validating the quantification of transcripts from 24 selected striatal genes via NanoString, proteomic datasets generated from striatum, cerebellum, and motor cortex, and two metabolic datasets quantifying metabolite abundance in brain (cerebellum, motor cortex and hippocampus) and peripheral (liver and plasma) tissues. In addition, a targeted dataset was sourced from a follow-up study which quantified the metabolite urea in 11 tissues. A description of each dataset including experimental and data normalisation procedures, and previous publications, is provided in the Supplementary Material. Individual datasets have been assigned a unique number for reference in this report (Table 2).

Assembly of the HD sheep platform, initial analyses, and data visualisation

R programming language (R Core Team 2018) was chosen as a tool to assemble the ‘multi-omic’ platform due to the abundance of add-on packages relevant to this research, and publication-quality graphics. R is freely available from the Comprehensive R Archive Network (CRAN, https://cran.r-project.org/), under the GNU General Public License. This work was undertaken using RStudio (RStudio Team 2018), a powerful integrated development environment (IDE) where visible panes display useful outputs simultaneously. The R package, ‘tidyverse’ (version 1.2.1, downloaded 2017) [18], containing six additional core packages, was installed on base R software, for data manipulation, analysis, and visualisation.

Organizational inconsistencies between each data-set, specifically sample naming and the arrangement/format of data, were adjusted according to Hadley Wickham’s ‘Tidy Data’ [19]. Within each dataset, the data was rearranged so that variables formed columns and samples were represented by rows. Categorical information, including transgenic/control status (T or C) and ewe/ram sex (E or R) was also added to each dataset as two new variable columns. For datasets that contained measurements from multiple tissue types a third categorical variable was included signifying the sample region (e.g., striatum, motor cortex). All other variables were numeric, consisting of values representing data-specific measurements. A unique identifier name was created for each of the 12 sheep and was used to denote the animal that each sample in a dataset was derived from. These sheep ID numbers were used consistently across all datasets to allow for pairwise comparisons. Specific R functions for data analyses and visualisation are described in the results section and can be found in the raw R scripts.

Development of the web-based HD sheep database and a summary of the statistics and analyses as presented under each database tab

R Shiny [20] was used to build a web application to serve as a public interface for query-based HD sheep data exploration and analyses. Our local HD sheep R platform was converted into a queryable web-based interface (the HD sheep database) and is accessible at https://hdsheep.cer.auckland.ac.nz. This online database hosts all the data described in Table 2 and allows each dataset to be queried using statistical functions implemented in the specific analysis tabs, summarised below.

Implementation of the ‘exCorr’ R package, for differential correlation analysis within the HD sheep database

To investigate differential correlations between any two variables and compare OVT73 and control groups in the HD sheep database, we developed the R package, exCorr, and implement it under the differential correlation analysis tabs. The exCorr package implements six functions for exploratory diff-erential correlation analysis, calculating pairwise correlation coefficients (r) and associated p-values for variable-variable combinations, with separation of two groups for differential comparison (e.g., control and test). The exCorr R package is open-source and available for download (including documentation) via the GitHub repository: https://github.com/emily5/exCorr.

Validation of data integrity in the multi-omic HD sheep R platform

As the datasets were re-formatted to enable integration in R, we first sought to confirm that the integrity of each dataset was intact within the platform. To do this, each dataset was analysed for known results, including our previously published observations. A selection of these validation results is described.

We previously reported that expression of the urea transporter SLC14A1 is upregulated in the OVT73 str-iatum (Ref. data 1 and 2) [13]. Repeated investigation of this transcript in the same dataset (Ref. data 1 and 2) in the HD sheep platform confirmed this result, validating the accurate transfer of data into the database context (Fig. 2A, C, D). Moreover, we report here that SLC14A1 is also upregulated in a separately generated RNA-Seq dataset derived from laser captured striatal matrix-derived neurons obtained from the same animals (Ref. data 3) (Fig. 2B). This finding provides additional support to our hypothesis that there is a urea cycle defect in HD, where resulting elevated levels of urea in the brain [13, 15] could be contributing to cell dysfunction and death.

OPEN IN VIEWER

Fig. 2

Comparison of SLC14A1 expression in four different transcriptomic datasets within the multi-omic HD sheep platform. SLC14A1 is significantly upregulated (p < 0.05, two tailed T-test) in the OVT73 brain (n = 6) relative to control animals (n = 6). A) Ref. data 1: An RNA-Seq analysis conducted on striatum samples, B) Ref. data 3: An RNA-Seq analysis conducted on striatal matrix-derived neurons preferentially captured via laser-captured microdissection. C, D) Ref. data 2: nanoString quantification of 24 genes conducted on samples taken from dorsolateral (C) and dorsomedial (D) striatum. In all graphs the x-axis displays the transgenic status of the sheep (Control or OVT73) and the y-axis displays normalised SLC14A1 levels (FPKM for RNA-Seq data and normalised counts for nanoString data). Direct visual comparison of datasets as shown was performed using the ggplot2 package in R.

Another previously reported observation is a hy-per-regulation of metabolites measured in the OVT73 cerebellum and liver compared to control samples (Ref. data 4) [10]. This newly formatted metabolite data within our HD sheep R platform, was analysed using the R package, ggcorrplot, generating pairwise correlation matrices and heatmaps for visualisation and interpretation (Fig. 3) (Ref. data 4). An excess of strong metabolite pairwise correlations were again seen in the OVT73 cerebellum and liver samples compared with controls, supporting a hypothesis that metabolic pathways are more tightly regulated in the prodromal OVT73 sheep model.

OPEN IN VIEWER

Fig. 3

OVT73 have more significant metabolite-metabolite correlations than controls. The ggcorrplot() function in R was applied to GC-MS metabolomics dataset (Ref. data 4), producing correlation coefficients, p-values and visual plots, as shown, for every variable-variable pair, with comparison of OVT73 and control groups. Significantly positively correlated metabolites are displayed as red squares and significantly negatively correlated metabolites are displayed as blue squares (p < 0.05) according to the figure legend. As shown by the number of coloured squares, there were more significantly correlated metabolite-metabolite pairs in the OVT73 cerebellum (B) and liver (D) compared to their respective control plots; control cerebellum (A) and control liver (C).

These initial validation analyses were conducted on each individual dataset within the multi-omic HD sheep platform and confirmed that the integrity of each dataset was intact and were directly comparable to previously reported results. Moreover, the platform enables data to be visualised in a way that can be easily interpreted, especially when comparing multiple variables.

To expand on the dataset validation, two tables are presented in the supplemental data showing add-itional comparisons between the striatal gene expression results. The nominal significance values and direction of change of gene expression of OVT73 versus control sheep measured by three methods (selected gene nanostring RNA quantification, striatal RNA-Seq, and the LCM striatal neurone RNA-Seq) are given in Supplementary Table 1. In addition, the intersect between the nominally significant differentially expressed genes in both the striatal RNA-Seq and the LCM striatal neurone RNA-Seq data sets are presented for comparison in Supplementary Table 2.

Development and utilisation of the online HD transgenic sheep database

To promote data sharing and accelerate the dis-covery of mechanisms involved in HD pathogen-esis, we converted our local HD transgenic sheep platform in R, into an interactive web-based RShiny database that can be utilised by HD researchers. The online database, which is accessible at https://hdsheep.cer.auckland.ac.nz/, contains detailed information regarding the OVT73 model, the 5-year-old cohort specifically, and includes a list of previously published journal articles for reference. The database hosts all the data described in Table 2 and contains specific analysis tabs for various statistical tests with visual outputs. The user can query the database without programming knowledge and use it to investigate specific transcripts/proteins/metabolites within the HD sheep system, providing another layer of HD data for the investigation of potential disease mechanisms.

Each ‘Analysis’ tab presents a basic statistical tool that allows selected data comparisons between OVT73 and control samples. These analyses include: student’s t-test, bootstrap and permutation tests, principle component analyses and differential correlation analyses, each with refined visualisation outputs for interpretation of the results. The structure of the website allows researchers to query user-specified variables and tissues of interest, producing outputs for interpretation. An example of the website interface highlighting two ‘Analysis’ tabs is shown in Fig. 4. The outputs from each analysis can be downloaded as PDF or CSV files for simplicity of use.

OPEN IN VIEWER

Fig. 4

The HD sheep database https://hdsheep.cer.auckland.ac.nz/. Examples of two analyses tabs within the HD sheep database: A) Student’s T-test and B) Differential correlation plots. These analyses can be used to investigate data in a query-based manner, comparing OVT73 sheep to controls, with variable selection. Results are displayed as informative graphical outputs for interpretation by the researcher. A shows a significant difference in OVT73 vs. control SLC14A1 expression in the dorsomedial striatal tissue (p < 0.007). B shows differential correlation structures in OVT73 vs. control for SLC14A1 and RHCG transcript expression in dorsomedial striatal tissue.

Due to the computationally demanding analysis that is implemented in some of the functions on the website, the more complex datasets have been reduced in size. This includes the three proteomics datasets, the striatum RNA-Seq and striatal neurone RNA-Seq datasets. The full datasets can be downloaded from the website. These datasets (Ref. data 1, 3, and 6) were partitioned based on the individual OVT73 versus control statistical significance to include the comparisons unadjusted T-test: p < 0.05 threshold. However, to continue our ethos of data sharing, all the raw datasets can be downloaded for user defined analyses.