Advances in short- and long-read sequencing and assembly over the last decade (Salzberg et al., 2011; Chin et al., 2013; Hackl et al., 2014) have made whole genome sequencing a routine task for biologists in various fields. Public sequence databases already contain several thousand of draft and finished genomes (Benson et al., 2013), with many more on the way (Pagani et al., 2012). In particular, high throughput sequencing projects of pathogen strains related to recent outbreaks (Rasko et al., 2011), and large-scale ecological studies targeting microbial communities and pan genomes of populations using metagenome and single cell sequencing approaches contribute in this process (Turnbaugh et al., 2007; Kashtan et al., 2014). These rich data sets can be explored for large-scale evolutionary processes using comparative genomics and whole genome alignments, revealing genomic recombinations (Didelot, Méric & Falush, 2012; Namouchi et al., 2012; Yahara et al., 2014), islands and horizontal gene transfer (Avrani et al., 2011; Coleman et al., 2006; Langille, Hsiao & Brinkman, 2008) as well as the often related dynamics of mobile or endogenous viral elements (Fischer, 2015; Touchon & Rocha, 2007). Other applications of whole genome comparisons include the analysis of paleopolyploidization events (Vanneste et al., 2014) and quantitative measurements of intra-tumour heterogeneity (Schwarz et al., 2015).
However, to facilitate proper interpretation of the obtained whole genome comparisons, visualization is key. One of the first tools to provide an interactive graphical representation of aligned genomes is the multiple whole genome alignment program Mauve (Darling et al., 2004). Mauve represents genomes in a co-linear layout with homologous syntenic blocks indicated by colors and connecting lines. The interactive stand-alone viewer ACT (Carver et al., 2008), in addition to alignment blocks, supports the representation of genomic annotations, such as genes. The R library genoPlotR (Guy, Kultima & Andersson, 2010) and the Python based application EasyFig (Sullivan, Petty & Beatson, 2011), both also based on a co-linear layout and supporting feature annotations, lack interactive analysis features as they are designed to generate static figures.
In addition to co-linear layouts, tools using circular representations of genomes have been developed. BLASTatlas (Hallin, Binnewies & Ussery, 2008) and BRIG (Alikhan et al., 2011) use multiple concentric rings to represent data of individual genomes, with BRIG also providing an interactive graphical interface. GenomeRing (Herbig et al., 2012) uses a circular representation as well, however, places all genomes on the same ring and syntenic blocks are connected with arcs extending into the center of the ring.
The web-based comparative genomics software Sybil (Riley et al., 2012) provides interactive co-linear visualization of multiple whole genome alignments with feature annotations and also supports a phylogenetic tree alongside the alignments. The software builds on a relational Chado database schema and, therefore, requires upload and import of custom data sets prior to analysis.
During our analysis of existing software, we found that interactive tools are useful for data exploration, but offer limited support for the figure export and at low qualities. Scripting-based tools provide higher levels of customization and figure quality, however, require familiarity with the respective language, thus often rendering the generation of figures time-consuming. For web- and database-based suites, such as Sybil, the upload and import procedure complicate utilization and limit applicability.
Our tool AliTV is divided into two parts. The first non-interactive part is required for the generation of the input files for our interactive viewer. The second part represents that interactive viewer in the form of a SVG file embedded in a HTML5 website. The latest version of our code can be obtained from GitHub (https://github.com/AliTVTeam/AliTV). It is planned to adjust AliTV in order to integrate it into the BioJS registry (https://biojsnet.herokuapp.com/, Corpas et al. (2014)). The general design of AliTV assures, that AliTV runs on different hard- and software platforms, e.g., Linux, MacOSX, and Windows. The following sections describe those parts in more detail.
The data preparation is performed by a single Perl script named alitv.pl. This script uses a set of different Perl modules to import incoming data and generate valid JSON input data for our visualization engine described in the next paragraph. One of our aims is to support as many different input formats for sequence and annotation information as possible. Therefore, we used the well tested and broadly accepted BioPerl as basis for our modules (Stajich et al., 2002).
The script alitv.pl uses a YAML file to specify the different input files. Moreover, an easy-to-use-mode is available which requires only a couple of input files and generates the required YAML file on the fly. This generated YAML settings file might be used to reproduce AliTV results or can be used as starting point to alter configuration parameters.
During the preparation step, AliTV requires all-vs-all alignments of the complete sequence set. Those alignments are generated or user provided. The current version of alitv.pl requires lastz to generate all alignments in MAF format (Harris, 2007). Nevertheless, BioPerl supports a broad range of alignment formats. Therefore, other programs can easily be added to the list of supported alignment programs. Moreover, the ability to use existing alignments allows a huge time benefit, when AliTV parameters are changed to optimize the visualization via YAML settings file in a non-interactive manner. Thus future versions of alitv.pl will support caching of alignments based on checksums to avoid unnecessary recalculations.
The final result of our alitv.pl is a JSON file, which can be load into our interactive visualization page.
Results and Discussion
To demonstrate the capabilities of AliTV we describe a short case study using seven published chloroplast genomes (Table 1). Four of the chloroplasts belong to parasitic plant species and three to non-parasitic ones. Parasitic plants rely much less or not at all on photosynthetic activity, a trait that should be reflected in the genomic structure of their chloroplast genomes. To assess this hypothesis the chloroplast genomes were downloaded from NCBI and processed with alitv.pl. For demonstration purposes, the chloroplast genome of Nicotiana tabacum was split in two pieces to represent an unfinished genome with more than one contig, and the genome sequence of Schwalbea americana was reverse-complemented (flipped). The pair-wise whole genome alignments are visualized by AliTV (Fig. 1A). The left-hand side of the display panel shows the phylogenetic tree for the seven species with species names as tip labels (parasitic plants are highlighted with an asterisk). The tree has been created provided in accordance to NCBI taxonomy (Sayers et al., 2009). Next to the tip labels, each genome is drawn as a scaled and annotated horizontal bar. The orientation of the S. americana genome was swapped back to match the orientation of the other genomes, indicated by the tick coordinates in reverse order (0 on the right side). N. tabacum is represented by two bars as the sequence has been split into two parts. On those bars features (e.g., genes or (IRs)) are shown as either rectangles or arrows. Alignments between adjacent genomes are represented as colored ribbons. The bottom legend shows the default color scale from red to green corresponding to low and high identity respectively.
|Olea europaea||NC_013707||Non-parasitic||Messina (2010)|
|Lindenbergia philippensis||NC_022859||Non-parasitic||Wicke et al. (2013)|
|Cistanche phelypaea||NC_025642||Holo-parasitic||Wicke et al. (2013)|
|Epifagus virginiana||NC_001568||Holo-parasitic||Wolfe, Morden & Palmer (1992)|
|Orobanche gracilis||NC_023464||Holo-parasitic||Wicke et al. (2013)|
|Schwalbea americana||NC_023115||Hemi-parasitic||Wicke et al. (2013)|
|Nicotiana tabacum||NC_001879||Non-parasitic||Kunnimalaiyaan & Nielsen (1997)|
The most striking observation is that three of the chloroplast genomes have drastically reduced sizes. All of those are parasitic (Table 1). Interestingly the chloroplast genome size of S. americana is similar to that of the non-parasitic plants. This can be explained by the life style of S. americana which is hemi-parasitic in contrast to the other parasitic plants which are holo-parasites. The features shown are the IR regions as arrows, the hypothetical chloroplast open reading frames as orange and the genes of the ndh family as pink rectangles. First, it can be seen that there is a big variation in size of the inverted repeats. While the IR of Orobanche gracilis is the shortest with roughly 5,000 bp, that of S. americana is the largest with roughly 35,000 bp. Second, there are less genes of the ndh family on Cistanche phelypaea, Epifagus virginiana, O. gracilis, and S. americana. Members of the ndh gene family encode subunits of the NADH dehydrogenase-like complex, which is involved in chlororespiration (Martín & Sabater, 2010). However, they are not required for plant growth under optimal conditions (Burrows, 1998). The absence of ndh genes in chloroplasts of parasitic plants has been studied in detail in Wicke et al. (2013). Loss of ndh genes has also been reported for photosynthetic plants such as some conifers and orchids (Wakasugi et al., 1994; Kim et al., 2015). Looking at the pairwise similarities of adjacent genomes, it is apparent that the non-parasitic plants (e.g., Olea europaea and Lindenbergia philippensis) have high overall sequence identity. In contrast, the sequence similarity within parasitic plants is lower. This observation can help framing a hypothesis about the evolutionary pressure on chloroplasts of parasitic plants. Another interesting observation is the distribution of missing regions of C. phelypaea in comparison to L. philippensis. Missing regions are distributed all over the genome and the order of the remaining parts remains stable. Wicke et al. (2013) describe an inversion in the large single copy region of S. americana compared to non-parasitic plants which is clearly visible by the link to N. tabacum around the 115 kbp position. All these observations can be made by simply looking at the raw figure created by alitv.pl and visualized by AliTV. However the figure can be analyzed interactively in more detail. One shortcoming of the linear representation of whole genome alignments is the limited comparability of non-adjacent sequences. Therefore, AliTV provides a way for the user to re-order the genomes on the figure (Fig. 1B). If reordering causes inconsistencies with the phylogenetic tree, the tree is hidden and a warning message is displayed. Furthermore, the links can be filtered by their alignment identity. The default setting is to display only links with minimal identity of 70%. But sometimes it might be interesting to look at regions with less similarity. To see these regions it is also important to hide large regions with high similarity. This can be achieved by changing the identity via a slider (Fig. 1C). After setting the identity range to 50%–90% red ‘X’-shaped links between N. tabacum, O. europaea, L. philippensis, and S. americana become apparent. For detailed inspection of regions of interest, AliTV provides a zoom function (Fig. 1D). This way the exact location of the alignments can be traced to the locations of psaA and psaB. Moreover AliTV provides functions like alignment length filtering, selective hiding of sequences, links and features, change of orientation (reverse complement) and rotation of circular chromosomes. Finally, it is possible to tweak many graphical parameters, such as colors, labels or spacing, directly via the interface to produce a publication quality figure which can be saved in SVG format. Furthermore, the current state can be saved in JSON format in order to share it with collaborators or continue the work with AliTV at a later time.
The case study demonstrates the suitability of AliTV as a tool for visualizing and analyzing whole genome comparisons. AliTV can be used to easily create a figure that show cases many genomic features at once. Furthermore, the rich interactive features enable the exploratory analysis and discovery of previously unknown features. Thus, novel hypotheses can be generated that can then be validated with experimental methods. Therefore, AliTV is a useful tool that will help scientists to find biologically meaningful information in the vast amount of genomic data.