Plasmodium falciparum antigenic variation. Mapping mosaic var gene sequences onto a network of shared, highly polymorphic sequence blocks

Abstract

Plasmodium falciparum erythrocyte membrane protein 1 (PfEMP1) is a potentially important family of immune targets, encoded by an extremely diverse gene family called var. Understanding of the genetic organization of var genes is hampered by sequence mosaicism that results from a long history of non-homologous recombination. Here we have used software designed to analyse social networks to visualize the relationships between large collections of short var sequences tags sampled from clinical parasite isolates. In this approach, two sequences are connected if they share one or more highly polymorphic sequence blocks. The results show that the majority of analysed sequences including several var-like sequences from the chimpanzee parasite Plasmodium reichenowi can be either directly or indirectly linked together in a single unbroken network. However, the network is highly structured and contains putative subgroups of recombining sequences. The major subgroup contains the previously described group A var genes, previously proposed to be genetically distinct. Another subgroup contains sequences found to be associated with rosetting, a parasite virulence phenotype. The mosaic structure of the sequences and their division into subgroups may reflect the conflicting problems of maximizing antigenic diversity and minimizing epitope sharing between variants while maintaining their host cell binding functions.

Fig. 1

The rationale for the approach. A and B. A Clustal alignment of four var sequences from 3D7 genome. Comparison of genes would normally be based on an alignment of the regions that are most shared between different sequences. The alternative used here (B) is to align polymorphic blocks of sequence (orange) to fixed reference points that are known to be invariant (black shaded sequence). For this, alignment of some conserved residues (one example is highlighted in blue) takes lower priority than alignment with respect to the chosen anchor points. These ‘position specific polymorphic blocks’ (PSPBs) are defined at up to four positions, PSPBs1–4. The default start positions (positions closest to the anchor point) are shown with vertical arrows. We chose default positions for the PSPBs that were adjacent to, but did not overlap with previously defined ‘positions of limited variability’ (PoLV1-4) marked 1, 2, 3 and 4 respectively (Bull et al., 2005; 2007). C–E. A hypothetical recombination network. Two hypothetical recombination events (1 and 2) are shown (C) together with a summary of the PSPBs that would be shared between the recombining genes and their products (D), and the resulting recombination network that would be obtained if all the products of recombination were sequenced (E). Each line a-g represents a portion of a chromosome corresponding to the sequenced regions of a several hypothetical var genes. Black portions of each line represent the three islands of homology D, F and H (A) that were used as anchor points. Thick coloured portions represent the four position specific polymorphic blocks (PSPBs) used in the analysis as distinct markers for ancestral var gene fragments. Dotted portions represent regions that were not used to align sequence. The first recombination event (1) between variants a and b occurs between PSPB1 and PSPB2 giving rise to two different var gene variants c and d. Recombination of one of these products c with another variant e between PSPB2 and PSPB3 (event 2) generates variants f and g. The relationships between each of the seven variants can be expressed as a network (E). Though variants a and b share no PSPBs they are connected indirectly through sharing PSPBs with the var sequences that resulted from recombination event 1. The same can be said for variants c and e in relation to recombination event 2. Thus all the hypothetical genes shown in (C–E) could be considered to be in the same community of genes that are capable of sharing blocks of sequences through recombination.

Fig. 2

Optimization of the approach. A and B. Variation in network structure with length, number and position specific polymorphic blocks (PSPBs). (A) The largest number of sequences that form an unbroken network (giant component) was determined for different numbers of PSPBs (1–4) and different PSPB length. PSPB length was varied while the distance between the proximal ends of the PSPBs (their start positions) and their respective anchor residues were kept constant. Start positions are indicated with vertical arrows in Fig. 1B. (A) shows representative results for PSPB1 alone, PSPB1 + 2, PSPB1 + 2 + 3 and all four PSPBs. (B) The effect of varying the distance between the start positions and the anchor positions. Distances are shown relative to the default positions used in (A) and shown in Fig. 1B. For this analysis, we used four PSPBs and a window length of 10 aa. C. The frequency of each variant observed at PSPBs1–4 was determined among the 1420 sequences from the Kilifi network. A large proportion of PSPB variants only occurred once among these sequences. D. The basic structure of the Kilifi network containing 1420 sequences and constructed using four PSPBs and the default PSPB start positions shown in Fig. 1B. See Fig. S1 for the structures of networks drawn with only three PSPBs.

Fig. 3

Locations of different groups of sequences within the Kilifi network. A–F. Locations of sequences falling in each of 6 previously defined cys/PoLV groups (Bull et al., 2005). G. Locations of 3D7 genes (Gardner et al., 2002). ‘Group A’ genes (genes associated with an upsA upstream element) are highlighted with an asterisk (*). var1 is indicated with an arrow. However, var1 is dimorphic within the tag region (Bull et al., 2005). The other var1 sequence type, present in parasite line FCR3, is in cys/PoLV group 1 and shares PSPBs with other cys/PoLV group 1 sequences within the small lobe of the network (data not shown). H. Location of 102 group A reference sequences (Trimnell et al., 2006). I. Location of DBLα sequences from a P. reichenowi (chimpanzee malaria) genome (Wellcome Trust Sanger Institute).

Fig. 4

Identification of putative recombining groups by varying PSPB length (see Fig. 2A). As PSPB length was increased the giant component of the network broke down into smaller components (unbroken networks of sequences). Using the giant component of the network structure generated using 10 aa PSPBs as a framework, the positions of each of these smaller components was mapped. This was used as an approach to identifying putative recombining groups within the network. For clarity, only components containing 20 or more sequences are highlighted. PSPB lengths: (A) 12 aa (B) 13 aa (C) 14 aa (D) 15 aa (E) 16 aa (F) 17 aa (G) 18 aa (H) 19 aa (I) 20 aa. Components obtained using a PSPB length of 14 aa were numbered as shown in C and are referred to in the text as block-sharing groups 1–7.

Fig. 5

Comparison with sequences collected worldwide. A total of 2257 sequences collected worldwide were used to construct a new network. The vertices were coloured according to their cys/PoLV groups (A). A perl script (Folder S2) was used to identify sequences containing any of the 14 aa PSPBs carried by sequences within block-sharing groups 1 and 2. These were called ‘block-sharing group 1-like’ or ‘2-like’ sequences. Block-sharing group 1-like sequences are highlighted in (B). Block-sharing group 2-like sequences are highlighted in (C). The sequences matching these two sets of PSPBs tend to be located in different parts of the network. (This is much more clearly seen in 3D versions of the networks, see Folder S1.) (D) focuses on the cys/PoLV group 2 sequences. Vertices corresponding to hybrid sequences carrying PSPBs from both block-sharing groups 1 and 2 (‘1 and 2-like’) are coloured in yellow. Overall, fewer hybrid sequences occurred than would be expected by chance (see text).

Fig. 6

Comparisons of var gene expression in rosetting and non-rosetting isolates. Expression levels of each gene were assessed by sequencing multiple clones from a library of RT-PCR amplified DBLα sequences from parasite RNA (see Table S1). The percentage representation of each sequence was determined within each isolate. The mean percentage was then determined for two pools of isolates, one pool of seven parasite isolates with high rosetting (13–94%) and one pool of seven parasites with low rosetting (0–6%). The area of each vertex is proportional to this mean percentage representation. (A and C) Pooled data from seven parasite isolates exhibiting low rosetting. (B and D) Pooled data from seven parasite isolates exhibiting high rosetting. Vertices are coloured according to block-sharing groups obtained using PSPB length of 14 aa (A and B) and by cys/PoLV groups (C and D).

Fig. 7

Comparison with a global sequence alignment. The sequences used to build the Kilifi network were aligned using the default parameters in MUSCLE. A. Sequences were coloured according the six cys/PoLV groups. B. Only group A reference sequences are shown (Trimnell et al., 2006). A region of the tree which lacks group A reference sequences is shown with a bracket. C. The positions of block-sharing group 1 (black) and 2 (blue) sequences are shown. There was a good correspondence between the block-sharing group 2 sequences and the gap in the group A reference sequences. Two block-sharing group 1 sequences (tags 993 and 1111) that fall outside the upsA region are indicated. D. Though tags 993 and 1111 are more like cys4 sequences they are linked to block-sharing group 1 because sequence 993 shares a PSPB with sequence 992.