T. brucei cell lines tagged with two drug-resistance markers were generated to select parasites that had stochastically activated an individually tagged BES from the background of highly similar BESs. The active BES1 was tagged with a Puromycin resistance gene (PUR), immediately downstream of the promoter. In the remaining silent BESs, transcription initiates at the BES promoter but is rapidly attenuated. This low level of transcriptional activity allowed the integration of a Neomycin resistance gene (NEO) immediately downstream of a second ‘silent’ BES promoter. In this way, we obtained 30 double-tagged clones. Each randomly NEO-tagged BES was typed according to its ESAG6 hypervariable region, the distance between the ESAG6 gene and the promoter, the number of BES promoters, and the sequence of the expressed VSG upon BES activation. Analysis of the 30 NEO-tagged clones and comparison to the TAR clones showed that 13 different BESs were tagged: BES1, 2, 3, 4, 5, 7, 8, 12, 13, 14, 15a and 15b (a duplicated BES, see below) and 17 (Tables 1 and 2). All tagged BES could be activated in vitro, except for BES8. Characterisation of these clones confirmed many of the apparently anomalous structures revealed by the TAR sequences (see below). The only two cloned BESs that were not tagged were BES10 and BES11 (Tables 1 and 2).

The chromosome containing each BES (indicated by arrows in Figure 1) was identified by Southern blotting of PFGE-separated chromosomal DNA with NEO and VSG probes (Figure 1). It had been shown previously that BES2 is located on an intermediate chromosome [20]. We now show that BES4, BES10, BES11, BES13 and BES17 are also present in this chromosome class and that the megabase-sized chromosomes encode all remaining BES, consistent with previous reports [21].

In some cases, the VSG probe hybridized with more than one chromosome, indicating that the genome contains multiple copies of the corresponding VSG. For instance, BES12 and BES14 contain VSG427-8, and BES15a and BES15b contain VSGbR-2/427-11, on megabase chromosomes of different sizes (Figures S1A and S1B, respectively). VSG121/427-6 is present as a non-BES copy (Dreesen O., personal communication).

The switching assay was designed to select for trypanosomes that had undergone an in situ switch by transcriptionally silencing BES1 and activating another BES without any DNA rearrangements. To assess the efficiency of selecting in situ switchers, Puromycin- and G418-resistant clones were analyzed by PFGE. If an in situ switch had occurred, the chromosomal locations of PUR, NEO and VSG should be unchanged: PUR and VSG221/427-2 should be on chromosome VIa, whereas NEO and the newly identified VSG should be located on a different chromosome. PFGE analysis of 28 NEO-tagged clones that had activated different BES (2, 4, 5, 7, 13, 14, 15b and 17) confirmed that in situ switching was the predominant switching mechanism (15 switch events), but other switching mechanisms were also observed (Figure 2). In BES2- or BES15-tagged clones, for example, we observed that VSG221/427-2 and PUR were deleted, whereas NEO and the new VSG remained in their original chromosomal location. This suggests that activation of BES5 and BES15 was accompanied by the truncation of BES1, a phenomenon previously reported for another BES [22]. In three of the 28 cases studied, the active BES was replaced by duplicative transposition of the entire new BES. In switchers that activated BES7 (Figure 2B), BES13 or BES15b, VSG221/427-2 and PUR were lost and NEO and the new VSG genes were duplicated onto chromosome VIa. This switching mechanism has been previously detected in switchers obtained in mice [23] or by negative-selection in vitro [24]. In three cases, we observed unconventional recombination events that involved more than two chromosomes. These complicated VSG switch events that have undergone multiple DNA rearrangements have been previously documented [25].

In our cell-lines, the frequency of switching between BES1 and a second BES can be measured as the ratio between the number of G418-resistant clones and the total number of starting cells. Switching frequency was measured for clones in which the NEO gene was in BES4, 5, 13, 15b and 17. Consistent with previous reports, the average frequency was around 2±3×10⁻⁶ switchers/cell, with BES15 showing the lowest switching frequency (8.3×10⁻⁸ switchers/cell) [20], [26]. However, it should be noted that these measurements were based on a small number of independent clones (usually 3). A larger-scale study would be necessary to assess if the differences of switching frequency are statistically significant and how those correlate with the DNA sequence of the activated BES.

Nineteen TAR clones [19] were sequenced, manually finished and analysed. This set represents 14 unique BESs, with TAR clones from BESs 9 and 16 having been re-classified (Table 2). For each BES group, at least one clone was sequenced (Figure 3, Table 1), choosing a larger sized clone wherever possible. Additional clones were sequenced for BES3 and BES4 to verify the sequence and clone integrity. Global alignments showed that TAR clones 2 and 15 (BES3) as well as 3 and 28 (BES4) are 100% identical at the DNA level with the exception of an ESAG7 triplication in TAR28. In addition, TAR clones from BES1 (TAR40), BES2 (TAR129) and BES17 (TAR59) were compared to the same BES previously sequenced from a BAC library [27] and were 99%,100% and 99% identical, respectively, indicating that TAR cloning produced stable clones.

Eighteen of the TAR clones contain full-length BESs, ending in telomeric hexamer repeats, and the 14 BES types show a striking conservation in overall structure and order of ESAGs (Figure 3). The 5′ end of each TAR clone contains the BES promoter that was used as the recombination target during cloning. Half of the BESs have a second promoter approximately 13 kb upstream from the first [28] which accounts for the differences in clone sizes observed in some of the BES groups. The region between the two promoters encodes ESAG10 and a variety of pseudogenes that are commonly found in T. brucei subtelomeric regions (Figure 3). The presence of two promoters in some BES was confirmed by restriction mapping of the NEO-tagged clones. BES5 and BES8 were the only two discrepancies: NEO-tagged clones suggested the presence of 2 promoters, whereas TAR clones only had one, presumably because the downstream promoter was the target of the TAR cloning (Figure S3B). Downstream of the second promoter, the majority of sites encode polymorphic variants of ESAG7, ESAG6, ESAG5, ESAG3, ESAG4, ESAG8, ESAG3, ESAG2, ESAG11 and ESAG1, in this conserved order. The only exception was an inversion between ESAG7 and ESAG6 in two BES groups (TAR29/BES12 and TAR56/BES13).

Nine BESs encode at least one copy of each ESAG (excluding ESAG10). We also observed frequent duplication of ESAGs, including a large-scale duplication in BES1 (TAR40), an ESAG7 triplication in BES4/TAR28 and three cases of divergent additional copies of ESAG8 (BES1, BES13 and BES15).

The terminal VSGs are flanked upstream by tracts of 70-bp repeats, whose lengths vary in our assemblies between 0.2–7.1 kb, although this repeat region was only manually finished in 10 clones (Figure 3). The distance between the VSG and the telomeric repeats could be assessed in 17 TAR clones and ranged from 200–1590 nucleotides.

As described above, the analysis of the NEO-tagged clones was consistent with the analysis of TAR clones except in one case. Although VSG221/427-2 is a single-copy gene (Figure 1), we found it in both BES1/TAR40 and BES16/TAR128. The presence of VSG221/427-2 in BES1 has been firmly established from sequencing one BAC and two TAR clones. The promoter-proximal region (approx 40 kb) of TAR128 is identical to TAR126. Taken this data together, we conclude that TAR128 may be an artefact that resulted from recombination between BES1 and BES15. PCR analysis of additional TAR clones of BES16 failed to detect VSG221/427-2 (data not shown), further suggesting that TAR128 is not representative of BES16.

A defining feature shared by most BESs is the strict conservation of the 5′ copy of ESAG3 and ESAG11 as pseudogenes (Figure 3). The predominance of the latter is corroborated by a previous study describing the characterisation of this gene family [7]. In addition to these two pseudogenes, single frameshifts cause the presumed inactivation of ESAG4 and ESAG5 in some BESs. An analysis of ESAG processing signals gave further indication that some of the ESAGs may not localise correctly within the cell, including a number of ESAG4 genes which lack sequences for surface targeting.

In five BESs (BESs 1, 2, 4, 5, 7) we identified a novel highly conserved putative protein coding sequence with a predicted signal peptide that we have designated ESAG12 (Figure 3). No functional clues could be discerned from sequence similarity or protein domain search results. A single homologue is found in the TREU 927 genome (Tb927.7.7510), although this is likely to be an under-representation due to the lack of subtelomeric sequence in the current TREU 927 genome assembly [18].

Several BESs diverge from this conserved structure either by encoding additional sequences or because they represent truncated sites, lacking members of individual ESAG families. Most BES harbour only the VSG between the 70 bp-repeat array and the telomere (Figure 3). BESs 2, 4, 7 and 15 are exceptions to this rule, showing additional sequences (ingi element, VSG or ESAG3 pseudogene). A further three sites encode duplicated regions of 70-bp repeats (BES1, 14 and 15), in each instance sandwiching VSG pseudogenes (see below). Five of the BES groups do not contain one or more of the expected ESAG repertoire. Amongst this set are:

BES7 (TARs 65,153) lacks ESAG7, but contains tandem copies of ESAG6, with different hypervariable regions, with ESAG6a resembling an ESAG6/ESAG7 mosaic. This BES could be activated in vitro at the same frequency as other BESs. In order to investigate whether any DNA rearrangements have donated a copy of ESAG7 to the newly activated BES7, DNA from cells with an active or silent BES7 were compared by PCR and restriction mapping (Figure S2). No differences were observed, suggesting that a BES lacking ESAG7 can indeed be activated, confirming previous observations [29].

BES8 (TAR64) is unusually small and only contains a single copy of ESAG7 followed by the 70-bp repeat region and the VSG (Figure 3). Long-range analysis of chromosomal DNA from a BES8-tagged clone digested with Apa I supported this finding, with the NEO and VSG427-14-hybridizing restriction fragment being only ∼20 kb in length (Figure S3A). Genotyping of the NEO-tagged clone confirmed the presence of ESAG7 and absence of ESAG6 (data not shown). Attempts to activate this BES in two independent clones were unsuccessful, suggesting that this mini-BES may not be functional.

BES10 (TAR134) lacks the central portion (ESAG 4-8-3-2-11-1). TAR134 also encodes only a 23-bp fraction of a 70-bp repeat. This site was not among those that were tagged with a drug resistance marker, so its potential for activation could not be assessed.

BES14 (TAR10), also lacks the ESAG4-8-3 gene cluster and has two VSG pseudogenes sandwiched between 70-bp repeat arrays. An inconsistency was found between the NEO-tagged clone and TAR sequence. The NEO-tagged clone did not harbour the VSG predicted by the TAR clone. Instead, the NEO tagged cell line expressed VSG427-8 upon activation, rather than VSG427-19, as predicted by the TAR sequence. We confirmed by PFGE that the BES activation did not involve gene rearrangements (data not shown), indicating that VSG427-8 was at the BES14 locus prior to activation. Further examination of the TAR clones in the BES14 set revealed TAR117 to correspond to the NEO-tagged clone. An interesting consequence of these results is that two BESs (BES12 and BES14) contain the VSG427-8 gene (Figures 1 and S1A).

In BES11 (TAR122), the ESAG2-11-1 cluster has been replaced by a VSG-related gene (VR, Figure 3) and a full-length copy of ESAG9, the only copy found to date in a T. brucei BES. The VR gene is believed to be a VSG that has already evolved a new function and, like other VRs, has no 70-bp repeat sequences in its upstream flank [30]. Similarly to BES10, this site was not tagged with a drug selectable marker and therefore, activation in vitro could not be assessed.

BES12 (TAR29) has a deletion of the ESAG4-8-3 cluster, but successful tagging and activation demonstrated that this site was functional in vitro (Table 1).

In order to investigate the role of recombination in the evolution of bloodstream expressions sites, we examined changes in phylogenetic relationships along the BESs using two methods.

The BES TAR library was constructed from T. brucei Lister 427 strain clone 221a [45], in which a hygromycin-resistance gene had been previously introduced immediately downstream of the promoter of the active VSG221/427-2 expression site and a neomycin-resistance gene downstream of the promoter of the silent VSGVO2 expression site [22]. BES-tagged cell lines were also derived from wild-type T. brucei 221a. Bloodstream-form trypanosomes were grown in HMI-9 medium [46]. Stable transfections were performed using the BTX Electroporator [47].

Yeast artificial chromosomes (YACs) containing BESs, TAR-cloned using the conserved VSG BES promoter as a recombination targets, have been previously described [19]. A representative clone was chosen from each group, selecting a larger sized clone wherever possible. Clones were isolated from yeast transformants through separation of the YAC DNA from endogenous chromosomes by CHEF (contour clamped homogeneous electric field) gel electrophoresis [48]. The recovered DNA was fragmented by sonication and end-repaired with mung bean nuclease prior to preparative agarose gel electrophoresis. DNA fragments of 2–4 kb and 1.4–2 kb were purified and cloned into pUC18 plasmid vector.

Nineteen TAR clones were sequenced by random sequencing of small insert libraries using dye-terminator chemistry on an ABI 3730 sequencing machine. Sequence reads were assembled using Phrap [www.phrap.org; P. Green, unpublished)]. Manual base calling and finishing was carried out using Gap4 software (http://www.mrc-lmb.cam.ac.uk/pubseq/manual/gap4_unix_1.html). Gaps and low quality regions of the sequence were resolved by primer walking and targeted polymerase chain reactions (PCR). A number of gaps, all exclusively located in the 70-bp repeat regions, remain in the assemblies and are indicated by 100 N's (Figure 3).

Each clone was annotated using the Artemis software [49]. Protein coding sequences were predicted as previously described and analysed to assign putative functions as previously described [18]. The full annotation of 19 expression sites can be viewed and searched via GeneDB (http://www.genedb.org/genedb/tbrucei427/) and sequences have been submitted to EMBL with the following accession numbers: FM162566 - FM162583.

Current VSG nomenclature reflects multiple naming schemes. We therefore propose a new systematic nomenclature based on sequential numbering with reference to the T. brucei Lister 427 strain (Table 1).

The Needle program, an implementation of Needleman-Wunsch algorithm in the EMBOSS [50] software package was used to perform global alignments using default parameters. The region of the imperfect 70-bp repeats was excluded from each alignment. Local alignments were performed using ClustalW (http://www.ebi.ac.uk/clustalw/index.html) with default parameters

BES-tagged clones were generated by introducing a Puromycin resistance gene (PUR) downstream of the promoter of a wild-type VSG221/427-2 expressing cell line. NEO was subsequently integrated downstream of the promoters of random BESs conferring resistance to G418. Clones obtained from this transfection were named BF-LF17.x or BF-LF18.x. depending on the presence or absence of the 50-bp repeat array upstream of BES1 promoter (Figueiredo et al., unpublished) (see also Data S1).

To prevent premature growth of in situ switchers, NEO-tagged clones were grown under Puromycin selection. After 5 days, cells were subcloned by limiting dilution in the absence of any drug to allow independent switching in each subclone. 2.4×10⁷ cells of 2–5 subclones were subsequently diluted in HMI-9 containing 100 µg/mL G418, and distributed in a 24-well plate. After 6–8 days, the number of G418-resistant clones was counted. Switching frequency was calculated by dividing the number of G418-resistant clones by the total number of plated cells. One or two clones from each plate were tested for Puromycin sensitivity by diluting 10⁵ cells in 5 mL of HMI-9 containing Puromycin at 1 µg/mL. Cell growth was scored after 2 days. Western Blotting was performed to confirm that G418-resistant cells no longer expressed VSG221/427-2.

The ESAG6 hypervariable region of each NEO-tagged BES was PCR amplified using a NEO-specific primer and one of two ESAG6-specific primers (5′-TAAAGAGAGTTGTTCACTCAC-3′ or 5′-TGTTCACTCACTCTCTTTGAC-3′) and the products gel-purified and sequenced. From clones with two consecutive ESAG6 genes, the ESAG6 hypervariable region from the most telomeric ESAG6 gene, was reamplified from the largest PCR fragment by nested-PCR with primers: 5′-CACTAATGATCAGCTTTACG-3′ and 5′-GACTCTTTTACACGTGAATC-3′. The resulting 1-kb fragment was purified and sequenced.

Total RNA was extracted from ∼10⁸ G418-resistant cells with RNA STAT-60 reagent (TEL-TEST, Inc.), following the manufacturer's instructions. cDNA was synthesized using a PolyT-primer and the StrataScript™ First Strand Synthesis System (Stratagene). VSG cDNA was amplified with primers that bind to the spliced leader (5′-GACTAGTTTCTGTACTATAT-3′) and to a conserved sequence found in the 3′UTR of all VSG (5′-GTGTTAAAATATATC-3′). Using the spliced leader primer ensured that the amplified product originated only from trans-spliced RNA and not from a DNA contaminant. PCR products were subsequently cloned in pGEM-T Easy (Promega) and sequenced.

T. brucei chromosomes were separated using a CHEF electrophoresis apparatus (CHEF-DR III, Bio-Rad), loading DNA from approximately 2×10⁷ cells per lane. Electrophoresis conditions for the panel separating T. brucei intermediate chromosomes were: 25 s pulse time, 6 V/cm, for 20 h, in 0.5×TBE, at 14°C, using a gel of 1% high-strength agarose (Helena Biosciences). T. brucei megabase chromosomes were separated using 1400–700 s linear ramped pulse, 2.5 V/cm, for 144 h in 1×TBE at 14°C in a gel with 1.2% high-strength agarose.

Agarose-embedded chromosome plugs were digested with 40 units of Apa I or Apa I and Xma I (New England Biolabs), as previously described [51]. Restriction products were separated on a 1.0% agarose gel in 0.5×Tris-Borate-EDTA buffer (TBE) in the Stratagene Rotating Agarose Gel Electrophoresis variant of Pulsed Field Gel Electrophoresis (PFGE) at 12°C and with a rotation angle of 120°. The running program for Apa I restriction fragments consisted of a 5 to 15 sec linear ramp at a constant 155 V for 18 h. For Apa I and Xma I restriction fragments the program was 1 to 4 sec linear ramp at a constant 150 V for 18 h. After alkaline transfer and hybridization with a radioactive probe, the Hybond-N+ Nylon membranes (Amersham Biosciences) were typically washed at 65°C as follows: twice for 20 min with 2×Sodium Chloride/Sodium Citrate (SSC), 0.1% SDS and twice for 20 min with 0.5×SSC, 0.1%SDS.

ClustalX was used to align homologs of 7 ESAG loci from 11 different expression sites [52]. The maximum likelihood phylogenetic tree topology was estimated for each locus using the program PHYML [53], with a GTR+Γ model and parameters estimated from the data. By constraining the resulting trees with the optimal topology of each other locus in turn a further six tree topologies were produced for each locus. The differences in likelihood value between the optimal tree and each constrained tree were calculated and the significance evaluated using the Shimodaira-Hasegawa test [54] thus quantifying the difference in phylogenetic signal between different loci along the expression site. Such likelihood comparisons require that two trees have equal taxon sets. For this reason a consistent gene complement from each expression site was used and ESAG10 and ESAG7 were therefore excluded. ESAG3 was also not used because it is present in three positions along the expression site and does not cluster by position; hence, its phylogenetic signal was not amenable to positional comparisons.

Sequence alignments of each ESAG family and selected intergenic regions were generated with ClustalX as described above, manually edited and gap-stripped. Each alignment was tested for evidence of recombination using two approaches. (i) The likelihood-based method implemented in the program GARD ([31]; http://www.datamonkey.org/GARD) was used to identify putative recombination and gene conversion breakpoints, employing a nucleotide substitution model specified by the HyPhy software package ([55]; http://www.hyphy.org). (ii) The program Phi [32]; http://www.mcb.mcgill.ca/trevor) was used to assess the significance of the pairwise homoplasy index (Φ_w) with a window size of 100 bases and 10,000 permutations per test.

ESAG protein-coding sequences were analysed for adaptive evolution by estimating the relative rates of non-synonymous and synonymous substitutions within each ESAG family. Positive selection is indicated by a non-synonymous/synonymous rate ratio (ω) greater than 1. The method was modified for recombinant sequences by allowing the genealogy to vary across the alignment while sharing the parameters of the codon substitution model between tracts [56]. The dimensions and genealogies of putative recombination tracts were provided by the GARD analyses described previously. HyPhy was used to obtain maximum likelihood estimates for two nested models of the distribution of ω. A likelihood ratio test for positive selection was performed with a χ²-distribution on two degrees of freedom by comparing twice the log-likelihood difference between a nearly-neutral model (M1a, in which ω cannot exceed 1) and a selection model (M2a) that included an additional category for ω>1 [57]. Evidence for adaptive evolution at individual codons was tested using an empirical Bayes method [57]: model M2a, with the maximum likelihood estimated parameters, was taken as a prior distribution for ω, and Bayes' formula was used to calculate the posterior distribution of ω at each codon, given the sequence data.