(1) Sequence view.

Sequences are displayed above the alignment location indicator. The dynamic yellow bar associated with the latter represents the region of the alignment that is currently visible. The green dot on the right hand side indicates what proportion of the sequences are currently visible. The consensus sequence of the alignment is displayed along the top of the sequence area, and directly under this the “+” indicates columns where all characters agree with the consensus character. Sequences and their titles are clickable and when an individual sequence is selected it will be traced through the corresponding network as a yellow line. If a user clicks on a node within the network area, all sequences that pass through that node will have a red dot placed next to their titles. These red dots correspond to the red paths that will become visible on the network; the latter indicative of through which other nodes the sequences pass. Sequences possessing a red dot next to their titles in this manner will be placed on top of the sequence display list, i.e. the order of sequences is sorted with these on top. If a different node is subsequently selected, the red title dots are reallocated to sequences passing through the newly selected node, paths redrawn and the sequence list is resorted. Basic parameters associated with the sequence display include the masking of characters that are the same as consensus, thus making it easier to identify non-consensus residues by eye, altering font size, altering space allocated to display titles and scroll speed; these are achieved through the “Navigation and Control” panel that is located on the bottom right of the display. A button within the control panel area can be used to reset all display parameters to default values. Site locations are highlighted in yellow along the top of the interface.

(2) Network view.

The network depicting sequence diversity within the alignment is displayed directly below the alignment location indicator. The associated orange bar of the latter represents the region of the alignment that is currently represented by the network. The region begins from the current sequence view starting site and extends to the right-hand side in a manner that is dependent on the number of consecutive windows placed along the alignment, as well as their width (Fig 2, step i); windows being regions from which nodes reflecting diversity are created. Both these parameters, as well as that of the clustering threshold that is applied within windows, can be altered using sliders located within the control area, thus allowing the user to visually explore the variation present. Once again there is a reset button to reset all network parameters to default values. Window locations are highlighted in orange along the bottom of the interface.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Network construction.

Coloured bars indicate unique sequence id’s relative to the corresponding sequences (dotted lines). Within each window the identification representing each full sequence are associated with individual sequence fragments spanning that window (i), and fragments within windows are clustered (ii). Edges are placed between neighbouring clusters where they share one or more sequence identification, i.e. differing regions of the same sequence (iii). Clusters are represented visually on the network by grey dots. If a single cluster is selected the paths of all sequences passing through in relation to all other clusters (red lines) can be traced (vi).

https://doi.org/10.1371/journal.pone.0259726.g002

(2.1) Nodes. For a given window clustering fragments of sequences that span that window creates nodes (Fig 2, step ii). Each cluster is created using an iterative approach. Initially a fragment is randomly selected to be a seed for a newly created empty cluster. All related fragments to that seed are then added to the cluster and become seeds for the next iteration. The metric used to define relatedness is hamming distance, i.e. the number of different characters between two aligned sequence fragments are counted. The default threshold value is a lenient 0.3, indicating that fragments that have less than 30% divergence from a seed are included within the cluster. More advanced measures of genetic distance exist that account for proposed models of sequence evolution at both nucleotide and amino acid levels [40, 41], but for the rapid clustering across windows placed along an alignment for the purpose of visualization hamming distance works well [42]. Iterations continue until no more next-round seeds can be identified. If unclustered fragments within the window still exist, a new cluster is initiated by selecting another random seed from the remaining fragments and the process is repeated.

For windows where six or less clusters are created, all clusters are displayed as grey circles, or nodes, on the network. For windows with more than six clusters, the largest six are displayed as grey circles, whilst sequences from any remaining smaller clusters are placed into a holding structure that is used for visualization purpose and that is displayed as a black circle. This holding structure is treated in the same as any other node when it comes to tracking sequences that are within it, i.e. if it is clicked then the paths of all sequences passing through it will be highlighted across the rest of the network. Here six was chosen to be the upper display limit so that following edge placement (next section), and associated edge crossover minimization, the maximum number of nodes that need order re-arrangement within any one window is seven, including the holding node (if present). This is because for a set containing n items, there are n factorial different order permutations [43], and during edge crossover minimization the number of edge crossovers produced by each permutation, in relative to nodes within a neighbouring window, must be counted. For a given window if there are the maximum of seven nodes present, 7! permutations (5040) must be identified during crossover minimization and this can be done in a reasonable time (< 1 second on an average laptop). If on the other hand there are fifteen nodes allowed within a window, then there are 15! permutations (1307674368000) requiring a time of many days. We arbitrarily felt that six clusters, and the additional holding node, was a sufficient number to graphically indicate the main variant groups present whilst minimizing edge crossover calculation time; the latter maintaining parameter updates in real time for the user as they scroll through the alignment. However, this limitation is for interface visualization purposes only and a higher resolution of all clusters defined by up to a 1% divergence threshold is possible using the “All Sequences” -> “Cluster Sequences” option. The default number (10) and width (50 bp) of windows, as well as the pairwise distance threshold, can be altered using the slider bars within the “Navigation and Control” area. The number of sequences passing through each node is indicated in green on the left and right hand sides of the network, as well as within the hint box of the control panel area if a particular node is clicked on.

(2.2) Edges. Edges are placed between nodes of neighbouring windows where they possess fragments that are derived from the same underlying sequence(s) (Fig 2, iii). Consequently, individual sequences can be traced through nodes across different windows. As indicate in the previous section, edge crossover between nodes within adjacent windows is minimized. Starting at the second most right-hand-side window, this is done by calculating all possible node order permutations, following which for each permutation, the number of edge crossovers to nodes within the adjacent right-hand window is counted, node layout order in the latter being kept constant (Fig 3). The permutations that produce the minimum number of crossovers are selected (Fig 3, red numbers), and from these a random one is used. The process is then repeated one window to the left, until the first window of the alignment is reached. Crossover minimization, whilst resulting in a visually more pleasing network, has no effect on the sequence information or underlying node connections. Following the connection of edges it is possible to click on nodes within the graph and track the sequences that pass through them (Fig 2, iv). On the interface, and as described in detail within the “Sequence View” section, such sequence paths are displayed in red, and a red dot is placed next to the titles of the sequences within the clicked on node.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Minimization of edge crossovers between nodes of the two right most windows of the alignment.

This process is repeated until the left most window (anchored on site 1) is reached. Clusters within the two windows are labelled with integers and required edges, based on the sequence ids (coloured bars), are listed (i). All order permutations of the current left window are identified and for each permutation the required edges are placed relative to the constant cluster order of the right window (ii). Crossovers are then counted (red numbers). Of the permutations that produce the minimum number of crossovers a random one is selected for graphical node layout order.

https://doi.org/10.1371/journal.pone.0259726.g003

(3) Navigation and control.

Access to all the network and display parameters is provided through the slider bars associated with the navigation and control panel (Fig 1), and in each case reset to default buttons have been provided. The buttons labelled with then red directional arrows are used to scroll through the alignment. These were implemented to remove the need for flat scroll bars as future developments will be aimed more at tablets and mobile devices. The red dot, at the centre of the four scroll arrows, immediately jumps a viewpoint at the centre of the alignment. In addition to the directional arrows the user can click directly on the grey squares along the alignment location indicator bar to immediately move to a particular location. Within this control area there are also buttons associated with for printing the network to a.png formatted file and clearing highlighted paths as well as a hint text section. The latter displays hint text from other areas of the display that the user clicks on such as sequences, nodes, location bars and general areas.

(4) Menu driven output.

Output options are accessed through the top-level menu bar and can be applied to (i) all sequences within the alignment, (ii) a subset of sequences whose titles match a user search criteria, (iii) a subset of sequences that pass through a selected node and (iv) a subset of sequences defined by the user based on a supplied file of titles. Interacting with sequences through these predefined options increases output data robustness; as opposed to allowing the random the inclusion of sequences through mouse selection into the subsets that the various sub-menu options are applied to. In addition to exporting subsets of sequences and/or specified regions of the alignment CView can generate summary statistics such as frequencies of residues and kmers as well as tertiary information such as pairwise distance matrix’s, variant count information and clustered sets of sequences. A description of sub-menu options is available on the wiki associated with the SourceForge project page, within the software itself under the menu option “Help” -> “Menu” (also accessible via the help button within the “navigation and control” area) and is outlined within the demonstration video [35].

(2.1) Background.

HIV-1 viruses can be characterized into two phenotypes that are dependent on cellular tropism and that are as a result of differences in co-receptor usage [48]. The macrophage tropic phenotype, often referred to as R5, requires the CCR5 co-receptor, whilst the T-cell tropic phenotype (X4) uses the CXCR4 co-receptor, the latter often emerging later on during infection [49]. Co-receptor usage can be detected by computational analysis based on specific genetic alterations within the V3 loop of the gp120 gene [29, 30]. Genetic variation within this region, of approximately 105 nt in length, lead to structural shifts that result in optimized binding to one co-receptor or the other [50]. For demonstrating the applicability of CView in aiding the characterization of sequence diversity, we have prepared a number of easy-to-follow steps involving the preparation and alignment of HIV-1 subtype B gp120 sequences representing each of the phenotypes described. Towards the end of these steps the V3 region of the alignment is extracted and per-site sequence variation between the two phenotypes is summarized. In this test case scenario the exact co-ordinates of the V3 loop are known prior to analysis, as are the phenotypes of each of the sequences present. However, this will often not be the case and a more exploratory approach could be adopted as indicated following the steps below.

(2.2) Method.

1. All North American subtype B gp120 sequences, verified to be CCR5-using sequences (n = 636), were downloaded from the Los Alamos HIV sequence database in aligned fasta format [37]. These were loaded into CView, using the “All Sequences” -> “Load Fasta (Alignment)” menu option, following which they were saved in unaligned format using the “All Sequences” -> “Save Unaligned” option. Additionally, the titles of these sequences were saved to a separate file using “All Sequences” -> “Save Titles”.
2. Step 1 was repeated for CXCR4-using sequences (n = 76).
3. In order to make sites directly comparable between the two sets of unaligned sequences, they were combined into a single file (by simply copying and pasting using a text editor) and aligned using MUSCLE [1]. Note: steps 1 to 3 were performed in order to ensure the quality of the final alignment. Alignment containing both CCR5- and CXCR4-using sequences could have been downloaded directly from the database, but these will have been extracted from a larger alignment containing many more sequences and so may not have been optimized.
4. The alignment created by MUSCLE, available at [51], was loaded into CView and the consensus sequence of the region spanning the V3 loop was saved using the “All Sequences” -> “Consensus Sequence” option. Within this alignment the co-ordinates of the region spanning the V3 loop were from 1436 to 1568. Although the exact location of the V3 loop within the gp120 region is known relative to the HXB2-LAI-IIIB-BRU reference strain (accession: K03455), the coordinates will vary depending on the alignment due to the placement of gap characters (‘-‘) introduced during the alignment process. The exact co-ordinates of the V3 region within our alignment were identified by eye using the V3 sequence of the HIV-1 reference strain where the start residues of the loop are TGTACAAGACCC and the end residues are CAAGCACATTGT [37].
5. The proportion of the alignment corresponding to R5 sequences was saved to a separate file in aligned format; and labelled accordingly with R5 in the title. This was done using the “User Group” -> “Save Region” menu option, where the titles used to define the R5 group of sequences to be saved were those obtained during step 1. The latter was used as in the case of this data the sequence titles were not labelled with any motif/tag that could have been used to identify them using the “Search Titles” top-level menu option.
6. Step 5 was repeated for the titles corresponding to the X4 sequences (obtained during step 2) and the output file was labelled accordingly.
7. Steps 5 and 6 resulted in two sub-alignments, contained within separate files, where although they represented different phenotypes (R5 and X4) the per-site co-ordinates were compatible as they represented sequences from a common underlying alignment. Each of these two files were loaded into CView, and nucleotide frequencies spanning the V3 loop were obtained using the “All Sequences” -> “Residue Frequencies” menu option. The co-ordinates entered for the V3 region within the subsequent popup box were those used in step 3. Once again each output file was labelled accordingly, for example the frequencies obtained from the R5 alignment were labelled R5_frequencies.txt.
8. For each of the extracted R5 and X4 alignments obtained in steps 5 and 6, CView was also used to output a list of variants spanning the V3 loop along with their frequency of occurrence. This was done by loading each sub-alignment into CView and applying the “Variant Frequencies”option of the “All Sequences” menu and using the co-ordinates specified above.
9. Each of the two output files containing variants were then translated using the EMBOSS Transeq tool [52].

The underlying alignment described for this use-case scenario, consisting of all the HIV-1 SUBTYPE B sequences spanning the gp120 region of the genome that have been verified as either being a CCR5-using (n = 636) or CXCR-using (n = 76), is available from the zenodo repository [51]. Within this alignment the exact co-ordinates of the region of the gene that harboured the diversity associated with either the X4 or the R5 phenotypes were known prior to the analysis. However, for many alignments a more exploratory approach may be preferable. One way that this can be achieved is by toggling the number of windows, their size and the clustering threshold used in order to obtain a more global view of the diversity present, and by subsequently using the “Node Path” top-level menu option to extract portions of sequences, and/or information from sequences such as kmer frequencies, that pass though nodes on the network that are of visual interest.

(2.3) Result and discussion.

Fig 4A displays the consensus sequence of the V3 region from the MUSCLE generated alignment prior to being divided by phenotype. The top ten most frequent variants from each of the two phenotypes are also displayed. The seqPublish tool [37], located at https://www.hiv.lanl.gov/content/sequence/SeqPublish/seqpublish.html, was used to format these alignments from the CView output such that characters identical to those of the consensus sequence were hidden. A similar text-based formatting is available at the bottom of the output file that is generated by the “Variant Frequency” option of CView, where residues that are identical to those present of the most frequent variant are represented by a “|” character. Here we used the seqPublish tool as variants from both phenotypes were being compared to the combined consensus and not the most frequent variant from a single alignment representing a single phenotype. The translation of each of these variants is presented within Fig 4B. Fig 4C shows a summary of the translations, obtained using sequence logos [53], and it can be observed that at site 11 the positively charged amino acid residues R (arginine) and H (histidine) are present within the sequences that were known to be CXCR4-using, whilst they are absent within the sequences obtained from the CCR5-uisng strains. At site 26 a similar observation is made in relation to positively charged residues, this time including a K (lysine) residue; although there is a minority K also present at 26 within the CCR5-using variants. This is a known observation where the presence of positively charged amino acids at sites 11 and 26 result in a structural alteration that optimizes CXCR4 co-receptor binding [29, 30]. The defined steps outlined in the use-case scenario presented here led to the summarization of the variation present within the V3 loop partially demonstrate the utility of CView when exploring aligned sequence data. The correctness of the summarization is confirmed by the presence of the known variation seen at sites 11 and 26. Complete per-site nucleotide frequencies for both R5 and X4 sequences spanning the V3 region are presented in supplementary S1 Table.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Summarization of variation present within the V3 loop.

(A) Green residues represent non-consensus residues from the ten most frequent variants associate with the CCR5-using phenotype. Brown represents those of the CXCR4-using phenotype. The consensus sequence (black) is shown. (B) Translations of the most frequent ten variants from each phenotype. (C) Sequence logos summarizing these translations. The top logo is from represents the CCR5-using sequences whilst the bottom represents the CXCR4-using ones.

https://doi.org/10.1371/journal.pone.0259726.g004

In the absence of pre-defined phenotypes, site co-ordinates and subtype, a more exploratory approach could be preferred. Fig 5 displays a picture of the network diversity observed across the entire gp120 alignment were the number and size of the windows have been maximized (to 50 and 75 respectively) and the clustering threshold set to 0.20. The user can click on individual nodes within the network, in order to identify subsets of sequences passing through, in order to export a range of information types that are implemented within CView for subsequent analysis relative to those subsets.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Network based diversity across the full gp120 case-study alignment: Thirty non-overlapping windows of length 75 nt fitted along the length of the alignment.

The clustering threshold within each was 0.2. The orange numbers along the top indicate window locations, whilst the green numbers on each side indicate the number of sequences observed within nodes (clusters) at the ends of the network. Grey dots are clickable nodes, that when clicked will highlight the paths (in red) of the sequences passing through across the remainder of the network. Here the highlighted paths are a result of clicking the bottom left most node (containing 33 sequences). Information from sequences highlighted in this manner can be extracted using the sub-options of the “Node Path” top-level menu within the software as indicated within the demonstration video [35], and under the software help menus. This network was printed using the “Print PNG” button located within the control panel of the software.

https://doi.org/10.1371/journal.pone.0259726.g005