Ethics statement

All animal work was undertaken under IACUC ref no 2003.19. The ILRI IACUC complies voluntarily with the UK Animals (Scientific Procedures) Act 1986 that contains guidelines and codes of practice for the housing and care of animals used in scientific procedures. All animals on survival experiments were regularly monitored to check for signs of terminal illness, and any showing such signs were euthanised by UK Schedule 1 procedures.

C3H/HeJ × C57BL/6 cross

C57BL/6JOlaHSD (C57BL/6) and C3H/HeJ mice were obtained from Harlan Laboratories. Mice were infected with 4×10⁴ T. congolense strain IL1180 intra-peritoneally (ip) as previously described [6]. Any mice that did not develop a microscopically proven parasitaemia were removed from the study.

345 F2 C3H/HeJ × C57BL/6 mice were phenotyped for survival time after infection with T. congolense strain IL1180. 94 animals that had extreme survival times (≤62 days and >140 days) were selected for genotyping using the markers shown in Table S1 in Supporting Text S1. Selective genotyping significantly reduces genotyping costs with little loss of power to detect QTL, however it does give exaggerated estimates of effect sizes [16]. The F2 mice were also genotyped at the Tlr4 locus since C3H/HeJ carries a proline to histidine mutation at position 712 of the Tlr4 gene that makes this mouse strain insensitive to LPS and might modify response to infection with T. congolense [17].

PCR reactions were performed using Reddymix (Thermo) with 20 ng of template DNA. Cycling conditions were as follows: 95°C, 50 secs; [Tm −5]°C, 50 secs; 65°C, 50 secs; 30× cycles. PCR products, including negative controls, were resolved by ethidium bromide stained agarose-gel electrophoresis and visualised under UV-light. SNP were genotyped by sequencing PCR products using primers shown in Table S2 in Supporting Text S1. Unincorporated primers and residual nucleotides were degraded using ExoSAP-IT (USB Corp, Ohio, USA) and sequencing products generated using Big-Dye v3.1 terminators (Applied Biosystems, Foster City, USA). Cycle sequencing products were ethanol precipitated and subject to electrophoresis on an Applied Biosystems ABI-3130XL capillary sequencer. Microsatellite and SNP genotyping data was viewed using PeakScanner (Applied Biosystems) and GAP4 [18] respectively.

Allocation of strains to haplotypes

Strains were allocated to haplotypes as previously described [19], [20]. Briefly, Perlegen SNP and haplotype boundaries were downloaded from Perlegen [15]. Strains were allocated to haplotypes for each haplotype block using a local Perl script that extracted all alleles from the Perlegen dataset within a haplotype block, substituted them into the C57BL/6 reference sequence and submitted the resulting aligned sequences to the Jukes-Cantor algorithm in DNADIST in PHYLIP to calculate genetic distances between each pair of strains [21]. Strains were given a binary “barcode” with all possible pairs of strains assigned a 1 or a 0 depending on whether the genetic distance for that pair was above or below a threshold value. Strains that had the same “barcode” were allocated to the same haplotype number. C57BL/6 was used as the reference strain for block allocation and assigned to haplotype one; Succeeding strains were allocated to the same haplotype block as another strain that they shared a haplotype with or, if there was none, to the next available haplotype number (Full details are available in Supporting Text S1; Allocation of strains to haplotypes).

CNV discovery

Array CGH was performed using the Agilent Mouse Genome CGH Microarray 244A platform. Dye-flip replicates were carried out on the C57BL/6 reference strain and three test strains (129P3/J, A/J and BALB/cJ) and analysed as previously described [22]. Overlapping aberrations were grouped into CNVR (t-test analysis, P≤0.05, Overlap 0.9) using the Agilent CGH analytics software (v 4.0) and using the ADM-2 algorithm (threshold 6.0) using centralization (threshold 6.0, bin size 1) and Fuzzy Zero [23]. CGH array data have been submitted to the NCBI Gene Expression Omnibus database (GEO) [GEO: GSE9669].

DNA capture and sequencing

Genomic DNA for BALB/cJ (Jackson #000651), 129P3/J (Jackson #000690), A/J (Jackson #000646) and C3H/HeJ (Jackson #000659) were obtained from the Jackson Laboratories and submitted to Nimblegen for sequence capture [24]. Capture probes were designed to cover 4.5 Mbp of non-repetitive sequence between 30,637,692 bp and 36,837,814 bp on Mmu17 (NCBI37). 385,000 60mer probes were tiled at approximately 5 bp intervals leading to a mean of 12 probes over each base. Captured DNA was sequenced on a Roche 454 FLX Genome Sequencer using Titanium chemistry (Roche). Sequence assembly and SNP calling was performed using the Newbler mapping algorithm, which aligned 454 reads against the Ensembl C57BL/6 reference (NCBI37) and outputs lists of SNP and associated coverage metrics.

As pyrosequencing is known to miscall sequences either across, or either side of, homopolymeric tracts (long stretches of a single nucleotide), discrepancies were removed from subsequent analysis if they were within 13 bp of a homopolymeric tract ≥5 bp [25]. SNP were additionally filtered to those with at least an eight-fold coverage and occurring in at least 87.5% of the reads sequenced across any polymorphic position. 14,440 high-confidence genotypes were submitted to dbSNP with SSIDs ss159831440-ss159845897. 454 reads were submitted to the European Short Read Archive under Accession number ERA000179.

SNP were aligned against coding sequences and non-synonymous SNP were identified. SNP positions were compared to the mouse regulatory build to test for SNP that may alter transcription factor binding sites or promoter regions [26], [27]. A 24-bp insertion in Mdc1 in susceptible strains was amplified by PCR and verified by agarose gel electrophoresis, but could not be shown to have any functional effect (data not shown).

Identification and annotation of single nucleotide polymorphisms (SNP)

SNP outside the Tir1 region were obtained from the 8 million Perlegen SNP set [15]. phastCons conservation scores for SNP positions [28] were obtained from UCSC [29]. These scores are a measure of how conserved a position is amongst 30 mammalian species and are on a scale between 0–1 with the most conserved positions scored as 1.

SNP within exons were annotated using the Ensembl SNP annotation API to identify non-synonymous SNP (nsSNP) and SNP in splice sites. nsSNP in the 454 data were identified with a local Perl script. Publically available functional SNP were also obtained from BioMart and the Wellcome Trust Sanger Institute website [12].

nsSNP were annotated with Polyphen [30] using the Polyphen batch submission tool. Publicly available functional SNP identified at QTL for which complete genotypes were not available were confirmed in C57BL/6, A/J, BALB/cJ and 129P3 mice using PCR and dideoxynucleotide sequencing as described for genotyping. Sequences which showed evidence of multiple copies were cloned using TOPO-TA cloning kit (Invitrogen) and sequenced.

Measurement of gene expression

Gene expression was measured for A/J, BALB/c and C57BL/6 mice before infection and at four time points post infection on Affymetrix 450_2 microarrays as previously described [11]. All microarray data has been deposited at ArrayExpress under the accession number E-MEXP-1190. The expression data and plots like those presented here are also available for all genes on the microarrays from the authors' website [31].

Refining numbers of candidate genes within the Tir QTL

Determination of QTL boundaries and initial candidate gene identification.

Different locations of the Tir2 and Tir3a,b,c QTL have been published at the F6 and F12 generations [7], [8]. QTL have also been physically mapped using congenic mice [32]. The congenic data supports the F6 location in one case (Tir2) and the F12 location in one other (Tir3a). Consequently we have annotated genes under both definitions of QTL positions and discuss their relative merits, case by case, below. In order to refine the number of candidate genes within Tir QTL it is necessary to first convert the 95% confidence intervals of the QTL from centiMorgan (cM) positions to megabase (Mbp) positions. Whilst the exact assignment of physical boundaries to the QTL is not possible, we have used the physical position of the peak marker in the F6 and F12 advanced intercross studies [7], [8] as the most likely position of the peak of the QTL. We estimated the physical size of the 95% confidence interval (CI) by using Mouse Genome Informatics data to find the median Kbp/cM ratio for the intervals between the ten flanking markers (which were spaced at ∼0.3 Mbp intervals). This ratio varied between 0.69–5.43 Mpb/cM and was used to convert the 95% CI in cM to Kbp. These positions are then used to identify the candidate genes contained within the QTL prior to further refinement (Table 1).

Download:

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

Table 1. Physical locations of QTL and counts of candidate genes.

https://doi.org/10.1371/journal.pntd.0000880.t001

Identification of QTL in C3H/HeJ mice

By increasing the number of breeds known to carry susceptible alleles at the QTL, candidate gene lists can be refined to remove those genes that are in QTL for T. congolense infection response but have the same ancestral haplotype as the resistant strain in at least one susceptible mouse breed. The three major Tir QTL have only been identified in C57BL/6, A/J and BALB/c mice, with C57BL/6 carrying the resistant allele at each locus. To that end, we measured survival after infection in an inter-cross between another susceptible breed, C3H/HeJ, and C57BL/6 mice. For the cross, the mean survival times of parental founder lines for the C3H/HeJ × C57BL/6 F2 cross were 63 days for C3H/HeJ and 87 days for C57BL/6. Out of the 345 F2 C3H/HeJ x C57BL/6 mice that were phenotyped, we selectively genotyped the 94 mice (51♂ and 43♀; p = 0.41) that had the most extreme survival times (Table S1 in Supporting Text S1) with microsatellite and SNP markers across the three known QTL. Table 2 shows that C3H/HeJ carries alleles that reduce survival time at the Tir3 QTL on Mmu1 and the Tir1 QTL on Mmu17. No QTL was discovered on Mmu5 in the region of Tir2.

Download:

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

Table 2. Loci regulating survival after T. congolense infection in the C3H/HeJ × C57BL/6 cross.

https://doi.org/10.1371/journal.pntd.0000880.t002

Refining numbers of candidate genes by allocation of alleles to haplotype blocks

Over eight million SNP and haplotype block boundaries derived from them have been published for the whole mouse genome [15], however the haplotype alleles carried by each strain are not available on a genome-wide basis. Full details of the allocation of strains to haplotype blocks and associated figures are available in Supporting Text S1; Allocation of Strains to Haplotypes. Results are briefly presented here:

In order to identify haplotype alleles that correlated with phenotype we obtained the Jukes-Cantor distance between each pair of the four mouse strains (C57BL/6, A/J, BALB/c and C3H/HeJ) for each haplotype block across each QTL. The distribution of the natural logarithm of Jukes-Cantor distances was approximately normal and the fifth percentile of the distribution corresponding to a distance of 5×10⁻⁵ was selected as a threshold (Figure S2 in Supporting Text S1). Strains were allocated to the same haplotype allele if the Jukes-Cantor distance between them at that block was less than 5×10⁻⁵.

Under the assumption that where QTL coincide in multiple crosses, it is likely that it is due to the same polymorphism in all breeds tested, there are only two possible distributions of resistant and susceptible haplotypes. In the present case, C57BL/6 was the only strain carrying a haplotype for longer survival at any QTL, therefore either: all susceptible breeds have the same haplotype that is different from C57BL/6 (hypothesis one); or C57BL/6 has a unique haplotype that differs from all susceptible breeds but that these might differ amongst themselves (hypothesis two).

A list was compiled of the genes for which the susceptible strains were on a different haplotype or immediately upstream or downstream of a different haplotype from the resistant (C57BL/6) strain (Supplementary Data S2: GenesAndHaplotypes.xls). Table 1 shows the number of candidate genes in each locus under the two nested hypotheses: H1) Short survival time is caused by a common deleterious allele in all three susceptible strains; or H2) that the difference in survival is attributable to a beneficial allele in the single long surviving line (C57BL/6) and the susceptible lines may carry any non C57BL/6 haplotype. Under H1 the number of candidate genes was reduced from 1193 to 283 and under H2 the number was reduced from 1193 to 651.

A null allele of the Tlr4 gene in C3H/HeJ does not affect survival

A functional toll like receptor 4 (Tlr4) gene is necessary for maximal control of Trypanosoma cruzi in mice [33] and there is evidence that the GPI anchor of T. brucei VSG has endotoxin like properties that could stimulate Tlr4 [34]. C3H/HeJ has a polymorphism in the Tlr4 gene, on mouse chromosome four, which ablates its function, making these mice insensitive to LPS [17]. We used this spontaneous mutation to discover whether Tlr4 was as important in the response to T. congolense as to T. cruzi. Since all previous mapping had been done in mice with intact Tlr4 genes, no QTL could have been detected at this locus even if Tlr4 does modulate the response to infection. The C3H/HeJ × C57BL/6 mapping population could therefore be used to discover whether this gene (or a closely linked one) is involved in the regulation of survival time after infection. Mice were genotyped with a microsatellite marker linked to the functional polymorphism and sequenced across the polymorphic position. There was no association with either of these markers and survival time, indicating that the Tlr4 pathway does not affect survival after T. congolense infection in mice (Table S1 in Supporting Text S1).

Comparative genomic hybridisation and gene expression

To assess the impact of copy number variation regions (CNVR) upon the expression of genes that may influence response to T. congolense infection we performed array-based comparative genomic hybridisation (aCGH) on the complete genome of three mouse strains: 129P3, A/J and BALB/c, relative to C57BL/6. The expression of genes within CNVR in A/J, BALB/c and C57BL/6 mice over the course of infection was evaluated using a previously described dataset [11].

Genome-wide, one hundred and twenty-nine CNVR involving three or more probes were common to A/J, BALBc/J and 129P3/J. These encompassed a total of 317 genes, and ranged in size from 400 bp to 6.4 Mbp, although 96% were smaller than 1 Mbp. Twelve CNVR containing the complete coding sequences of genes and that had corresponding differences in gene expression, were common in all susceptible breeds of mice tested. Lists of the genome-wide CNVR is shown in Table S8 in Supporting Text S1.

One significant CNVR was detected close to the peak of Tir3c in the F6 population (D1Mit113: 173,734,611 bp). A two to four-fold reduction in C57BL/6 copy number relative to A/J, BALB/c and 129P3/J encompassed, or overlapped with, the coding sequences of Itln1 (intelectin 1), Cd244, and AC083892.19-1 and may affect the nearby Ly9 (lymphocyte antigen 9) (173,441,746-173,499,029 bp; 11 probes; p = 0.0003; Figure 1A). There were expression differences in Cd244 (Figure 2A), but not Itln1 or Ly9 [31], over the course of infection between resistant C57BL/6 and susceptible A/J and BALB/c. AC083892.19-1 was not on the expression microarray. This CNV region has also been previously reported by Graubert et al [14] who showed that an additional susceptible strain, C3H/HeJ, carries the same variant as A/J and BALB/c.

Download:

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

Figure 1. CNV plots from Agilent DNA Analytics software.

A: Reduced copy numbers in C57BL/6 of Itlnb and Cd244 near Tir3c relative to two susceptible breeds of mice (Chr 1: 172,831,532–173,931,532 bp). B: CNV data at the proximal end of Tir1 showing a deletion of Glo1 and Dnahc8 in C57BL/6 and BALB/c relative to A/J and 129P3. (Chr 17: 29,854,972–30,954,972). Probes are plotted at their genomic position relative to their respective log₂ fluorescence intensity ratios (Y-axis) along with genes on the x-axis (filled blue rectangles). Green dots are negative ratios and red dots positive ratios (threshold 0.5). Lines are a moving average over a 10 Kbp window for A/J (blue); 129P3 (red) and BALBc (yellow). Genomic positions are based on mouse build mm8 (NCBI36).

https://doi.org/10.1371/journal.pntd.0000880.g001

Download:

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

Figure 2. Expression of A/J OlaHsdnd (A/J), BALB/cJ OlaHsdce (BALB/c) and C57BL/6JOlaHSD (C57BL/6) mouse genes in the Tir3c locus at five time points in the course of infection (0 days; 3 days; 5days; 9 days; 17 days).

Graphs include a small x-axis offset to improve spatial clarity. A Cd244 in the spleen, B Cd48 in the liver, C Apcs in the liver D Ifi202b in liver and spleen. Cd244 expression was low in liver in all strains until Day 7 when it rose above background and C57BL/6 had slightly lower levels than A/J or BALB/c (data not shown).

https://doi.org/10.1371/journal.pntd.0000880.g002

No common CNVR were detected within Tir1 or Tir2. The CNVR that was previously reported to be the cause of differential expression of Glyoxalase 1 (Glo1) [35] and is 2.8 Mbp from the peak of Tir1, was detected as a two to fourfold reduction in copy number for C57BL/6 and BALB/c relative to A/J and 129P3 (Chr17: 30,176,153 bp–30,650,413 bp; 68 probes; p<0.001; Figure 1B). Since the CNVR did not correlate with phenotype, this polymorphism is unlikely to contribute to the difference in response to infection.

Identification of functional SNP

Lists of published non-synonymous SNP (nsSNP), SNP in splice sites; and regulatory regions and SNP that cause gain or loss of stop codons were obtained from BioMart. nsSNP were annotated using Polyphen [30] in order to identify those most likely to modify gene function. A complete list of annotated SNP is in Supplementary Data S1: AnnotatedFunctionalSNP.xls. Polyphen classifies nsSNP as benign, possibly damaging or probably damaging according to the likelihood that the polymorphism will modify protein activity. ‘Damaging’ implies a change of activity or function but this change could be beneficial to the animal.

Sequence capture and sequencing of Tir1

DNA from across the Tir1 QTL was sequenced in order to characterise novel SNP and to improve the identification of alternate alleles for each haplotype block. DNA from four mouse breeds: 129P3, A/J, BALB/c and C3H/HeJ; was captured on Nimblegen arrays with probes for a 6.2 Mbp region of mouse chromosome 17 between 30,637,692 and 36,837,814 (NCBI37). 1.7 Mbp of repetitive sequence was excluded. Captured DNA was sequenced on a Roche 454 Genome Sequencer FLX using Titanium chemistry. 1,308,175 reads were mapped to the C57BL/6 reference sequence giving an average ∼15× coverage of each sequenced strain (mean read length 282 bp; total sequence ∼370 Mbp).

As 454 pyrosequencing is known to suffer from sequencing errors within, or close to long homopolymeric tracts, SNP were filtered to exclude those that were within a 13 bp window of homopolymeric tracts ≥5 bp. Furthermore, SNP were additionally filtered for those that were not outside regions covered by capture probes even if they were within the Tir1 region. After filtering, 14,440 SNP loci were identified, 3,618 of which were not in dbSNP build 128. 1,588 loci were common to A/J, BALB/c and C3H/HeJ, but differed from C57BL/6. Furthermore, upon adding data for 129P3, there were 466 SNP loci common to all four sequenced mouse strains. Summary statistics for all SNP are available in Table S4 in Supporting Text S1.

Figure 3 shows a circular plot of all SNP called by the Roche/454 mapping algorithm (Newbler) against the C57BL/6 reference. Haplotype blocks can be seen as clusters of high-densities and low-densities of SNP. Whilst at this resolution it is not easy to see haplotype blocks in the A/J, BALB/c or C3H/HeJ data, one haplotype block stands out in the 129P3 data where 81 common SNP clustered within a 430 Kbp region (33,245,853–33,675,688 bp).

Download:

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

Figure 3. Array-based sequence capture and next generation sequencing of a 6.2 Mbp region of Mmu17 in four breeds of mice: A/J; BALB/c; C3H/HeJ and 129/J (Mmu17:30,637,692–36,837,814 bp).

Plot is circular for ease of display [56]. Tir1 is highlighted in black on the inside track. Genomic positions are in Mbp. The outer tracks (blue and brown) show genes and designed capture probes, respectively. The four, coloured, inner tracks show SNP called in each of the four sequencing experiments, with the black tick marks highlighting areas of common SNP. Haplotype blocks can clearly be seen as clustering of high- and low- density regions of SNP. A magnified region around Tir1 is displayed underneath the circular plot. Tracks are identically coloured and include a moving average (window 1 Kbp) of sequence read coverage across the region (top). Genes in the region are displayed for the forward strand (above) and reverse strand (below).

https://doi.org/10.1371/journal.pntd.0000880.g003

In order to validate SNP calls, 454-generated SNP were compared against those called in a recently published set sequenced on the Solexa/Illumina platform from flow-sorted mouse chromosome 17 for A/J [13], and similar, publicly available SNP from the concurrent Mouse Genomes Project (Wellcome Trust Sanger Institute) for BALB/c, C3H/HeJ and 129P2 mouse breeds [12]. Only 3 out of 36,784 (0.014%) of the homozygous calls (coverage >1; alternative allele frequency (AAF) >80%) were discordant between the two datasets Table S7 in Supporting Text S1. The 454 data included 53–71% of SNP in the Illumina data depending on the coverage required to call a SNP and the Illumina data contained 94–97% of SNP in the 454 data (Figure S5 in Supporting Text S1). Full details of the comparison are available in Supplementary Data S3; SNP validation.xls.

Structural polymorphisms

Using all available data, the Tir1 region contained 80 nsSNP loci that correlated with phenotype. There were seven “possibly damaging” (Polyphen) nsSNP and “probably damaging” nsSNP in PML-retinoic acid receptor alpha regulated adaptor molecule 1 (Pram1) (rs33399614), Rgl2 (Ral guanine nucleotide dissociation stimulator-like 2) and CR974462 (Table 3). Nine genes contained splice site polymorphisms (See Supplementary Data S1: AnnotatedFunctionalSNP.xls).

Download:

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

Table 3. nsSNP loci within the extended Tir1 definition.

https://doi.org/10.1371/journal.pntd.0000880.t003

Regulatory polymorphisms

Differences between the susceptible strains and C57BL/6 were aligned to the Ensembl mouse regulatory build (NCBI37: Ensembl 54). Ten differences were predicted to fall within regions of accessible chromatin and may affect transcription factor binding regions. Furthermore, 13 differences mapped to within 2500 bp of the upstream region of genes that may be associated with promoter regions. In total, 14 genes may be affected by SNP in this way (Table S5 in Supporting Text S1). Of the 13 genes for which microarray data was available, however, only phosphodiesterase 9A (Pde9a) showed any differences in gene expression, and these correlated with alleles of a SNP (rs33223038). A/J differed from C57BL/6 and BALB/c at this locus in both SNP genotype and Pde9a expression, but since this did not correlate with phenotype, it was discounted as a candidate SNP. There were also SNP in non-essential splice sites in nine genes that may modify their exon usage (Supplementary Data S1: AnnotatedFunctionalSNP.xls).

Correlation of haplotype assignments using 454 and Perlegen data

Jukes Cantor distances were calculated for each haplotype block in the Tir1 region using the 454 and Perlegen datasets. A more detailed description of the analysis is presented in Supporting Text S1. Shared haplotypes had high positive predictive value and specificity for shared SNP alleles but low negative predictive value and sensitivity (Table S3b in Supporting Text S1), indicating that having shared haplotypes is a good indicator of shared SNP alleles but that the converse is not true. This means that assignments will be accurate where C57BL/6 has been assigned the same haplotype allele as susceptible strains but less accurate where C57BL/6 has been assigned to a different haplotype block allele from the susceptible strains. Therefore the data may be reliable way of excluding loci as candidate QTL regions but less accurate for including loci. The correlation between the distances calculated from the 454 and Perlegen SNP sets was modest (r = 0.63). The slope of the regression line was 0.67 reflecting the greater number of SNP in the 454 dataset. A high degree of scatter was observed in a plot of distances based on Perlegen and 454 data (Figure S3 in Supporting Text S1). The scatter suggests that SNP coverage is uneven in one or both datasets, and therefore increasing SNP density should increase the reliability of haplotype calls. Inspection of a plot of SNP coverage in the two data sets shows that the ratio of the number of SNP that were found in the two data sets varied substantially between haplotype blocks (Figure 4 and Figure S4 in Supporting Text S1).

Download:

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

Figure 4. SNP plots of Tir1 between 31 and 31.65 Mbp.

The C57BL/6 row represents the reference allele for all loci that are polymorphic in either the Perlegen set or our 454 set. The SNP density is much greater in the 454 data set, in which haplotype blocks are clearly identifiable by eye. It appears that SNP are much better represented in the Perlegen data set in some regions than in others. Between 31.2 and 31.30 the two data sets are very similar with high density of SNP in BALB/c and 129 substrains in each dataset. However in the region between 31.1 and 31.20 the SNP in BALB/c and 129 are relatively much sparser in Perlegen than in the 454 data.

https://doi.org/10.1371/journal.pntd.0000880.g004

Plots like those shown in Figures 4 and S3 can be obtained for any region of Tir1 from our website [31]. Plots of SNP and haplotypes and tables of Jukes Cantor distances between alleles at each haplotype block based on Perlegen data can be obtained for any part of the mouse genome at the same site.

QTL mapping

The mapping studies showed that C3H/HeJ mice carry susceptible alleles at the Tir1 and Tir3 loci. No QTL were observed at the Tir2 locus. The Tir1 locus as defined by previous fine mapping studies is just proximal to the major histocompatibility complex (MHC) (Table 2), and the conversion of genetic distances to physical positions presented here shows that Tir1 includes three classical class I MHC H2K genes. However previous studies have found no correlation between MHC haplotype and response to infection [4] consistent with the QTL gene not being a classical MHC molecule.

The mapping population was also screened for an association between Tlr4 and survival; no association was found. This observation implies that the presence or absence of a functional Tlr4 gene has no effect on survival, but does not preclude the pathway from Tlr4 to Nfkb (nuclear factor kappa-B) from responding to infection. Tlr4 could still participate in the regulation of anaemia and parasitaemia, which are not correlated with survival [11].

Identification of physical boundaries of QTL

The two independent F6 and F12 mapping populations have reduced the 95% CI of the QTL to exceptionally small regions, particularly at Tir1, the QTL of largest effect, where the physical size of the 95% CI was 930 Kbp for the combined data from the A/J × C57BL/6 and BALB/c × C57BL/6 F6 crosses (Table 1). This was only twice the mean distance between markers at this locus (400 Kbp) and consequently the main limitation in identifying the boundaries of the QTL is in estimating the position of the peak.

Identification of functional nsSNP

Resequencing of the QTL region on the Roche 454 platform at Liverpool to 15× coverage discovered 3,618 novel SNP loci that were deposited in dbSNP. Comparison with a resequencing project on the Illumina platform at the Wellcome Trust Sanger Institute to 22× coverage [13] showed 99.98% consistency in SNP calls even when no minimum coverage criterion was applied for calling a SNP. Both data sets contained large numbers of SNP called as heterozygotes with alternative allele frequencies between 25–80%. These loci from both data sets were associated with significantly higher sequence coverage in our data indicating that the majority were likely to be due to mapping artefacts probably caused by CNV. The 454 data contained only 71% of the SNP discovered by the higher coverage Illumina data but both methods discovered the same set of nsSNP. The 454 data discovered an additional 3% of SNP that were not in the Illumina data.

Utilising all SNP from the 454, Perlegen and Illumina data sets, three probably damaging nsSNP were identified in genes at the peak of the Tir1 QTL that correlated perfectly with phenotype (Table 3). Two nsSNP were in Pram1; the Pram1^537L/P polymorphism was scored as probably damaging by Polyphen. The Pram1^103R/K polymorphism was classed as benign by Polyphen but lies within a proline rich domain (PRINTS: PR01217) that is involved in binding the “SH3 domain of hematopoietic progenitor kinase 1 (HPK-1)-interacting protein of 55 kDa (HIP-55),” which is known to stimulate the activity of HPK-1 and c-Jun N-terminal kinase (JNK)” [39]. Pram1 is almost exclusively expressed in myeloid cells [36] and specifically in granulocytes in terminal stages of differentiation [40] where it is induced by retinoic acid. It was thought that Pram1 might be a negative regulator of neutrophil differentiation since it is repressed in acute myeloid leukaemia. The deletion of Pram1, however, has no effect on neutrophil differentiation and maturation but does disrupt reactive oxygen intermediate production and degranulation by neutrophils [41]. This may affect the early, pro-inflammatory response to infection or downstream TNFa signalling, which has been shown to be differentially expressed in susceptible and resistant mice [42]. C57BL/6 appears to have the derived allele of Pram1^537L/P since A/J, BALB/c and C3H/HeJ had the same allele as Hominidae and dogs. Since C57BL/6 tend to have a more inflammatory phenotype, it is possible that the polymorphisms lead to a gain of function with stronger binding to HIP55 leading to faster and more persistent ROI induction and a more inflammatory state.

The other probably damaging SNP at Tir1 were CR974462 and Rgl2. There is no annotation for CR974462. Rgl2 (Rif) is a small GTPase that is most highly expressed in macrophages and B cells and appears to be involved in Ras mediated signalling [43]. The Rgl2^147H→Y polymorphism could affect the Ras pathway that plays a key role in leukocyte activation and is therefore a plausible candidate gene.

The Fas death domain-associated protein (Daxx) gene, which we have previously reported to contain a deletion of a single aspartate residue in susceptible mice [44], is also under the peak of Tir1. Daxx is within the MAPK pathway, which was found to respond to T. congolense infection in microarray data. However a new Polyphen analysis of the aspartate deletion in Daxx indicates that this polymorphism will be benign in effect. The aspartate deletion is within a run of 11 aspartate residues and a region where 35/41 residues are acidic [44]. Therefore this polymorphism is probably less significant than the probably damaging ones reported here.

Regulatory polymorphisms could also cause the phenotypic difference: one SNP (rs33223038) was identified in Ensembl as being in a regulatory region upstream of Pde9a but although this SNP correlated with differential expression it did not correlate with survival differences between susceptible and resistant mouse breeds. There were also SNP in non-essential splice sites in nine genes.

Copy number variation at Tir QTL

CNV have previously been shown to be a major cause of quantative trait differences [10]. We used Agilent whole mouse genome aCGH arrays to identify CNV between C57BL/6 mice and A/J, BALB/c and 129P3 mice. The aCGH data highlighted a CNVR containing three genes close to the peak of the Tir3c QTL: Cd244; Ly9 and Itln1 (Figure 1A). A nearby gene Cd48 had a probably damaging nsSNP. Cd244, Cd48 and Ly9 are important genes involved in the production and regulation of IFNg by NK and T cells. CD244 binds CD48 on lymphocytes and is involved in NK:NK cell and NK:T cell interactions leading to NK and T cell proliferation [45], which are important mechanisms in innate resistance to protozoan infection [46], [47].

Splenic expression of Cd244 differed between strains with the resistant C57BL/6 mice having the lowest expression consistent with the low copy number of Cd244 in C57BL/6. Cd48 expression increased 16-fold in liver after infection with T. congolense, but this occurred in all strains tested (Figure 2). Since CD48 and CD244 directly interact, it is possible that the QTL is a consequence of the combined effect of the probably damaging nsSNP in Cd48 and the CNV in Cd244. Differences in expression could not be seen in Itlnb or Ly9.

The large number of genes in Tir3c that had CNV, nsSNP or haplotypes that correlated with phenotype may make it difficult to identify the QTL gene at this locus. It is possible that the QTL is not a consequence of a single polymorphism but the combined effect of multiple polymorphisms in an extended haplotype. However the CNV at Cd244 was the most substantial DNA polymorphism in the region making Cd244 a strong candidate QTL gene. Inserting an additional copy of Cd244 into the C57BL/6 background, so that it had a similar gene dosage to the susceptible strains, could test the effect of this CNV on the response to infection.

Haplotype block analysis

We have previously used this strategy to show a strong association between upstream haplotype differences and high confidence (p<0.005) differences in gene expression [19] and also short listed genes under QTL for differences in response to Heligmosomoides bakeri infection [20]. We reduced the number of candidate genes in this study by about 76% and 45% under hypotheses H1 and H2 from the 1193 genes that were under the 95% confidence intervals of the QTL. There were 283 genes where A/J, BALB/c and C3H/HeJ had the same haplotype different from C57BL/6 and 651 genes where C57BL/6 differed from the other three. The large number of genes that had haplotypes that correlated with phenotype is mainly because: 1) C3H/HeJ, A/J and BALB/c are more similar to each other than to any other strain based on analysis of 673 genome wide SNP in 55 strains [48]; 2) we used the stringent criterion that a gene was included if any haplotype block between the two neighbouring genes correlated with phenotype; 3) The high positive predictive power of the method means that whilst it is probably very reliable for excluding loci where susceptible strains share a haplotype block with the C57BL/6 resistant strains, it assigns too many haplotype blocks to different alleles.

Whilst there are large numbers of reported SNP for A/J, 129X1/SvJ and 129S1/SvImJ due to the Celera sequencing project [49] and for BALBc/ByJ and C3H/HeJ from the Perlegen project [15], relatively few SNP are publicly available for the 129P3 strain. The 454 resequencing of the Tir1 region indicated that approximately 50% of the resequenced region could be excluded from the QTL if the allele carried by 129P3 mice at this locus was known. If a QTL was identified at Tir1 in a 129P3 × C57BL/6 cross then the QTL gene could be assumed to be within the three blocks where 129P3 differed from C57BL/6. If no evidence of a QTL was found then these regions could be excluded from the QTL on the assumption that 129P3 carried the same allele as C57BL/6 at this locus. This analysis indicates that mapping QTL for response to infection in a 129P3 × C57BL/6 cross should significantly refine the list of candidate genes. The availability of this haplotype data makes it possible to make more rational choices about the selection of strains for mapping experiments. This strategy has been used before with a much more limited SNP set [50] but the corresponding online resource is no longer available.

Our objective was to identify the SNP that were most likely to have an impact on function. These were considered to be nsSNP that altered the physical properties of the protein as judged by Polyphen analysis, SNP in essential splice sites and CNV and regulatory SNP that correlated with changes in expression. It should be emphasised, however, that many types of SNP can underlie QTL, for example the QTL SNP at the Idd5 locus appears to be a synonymous SNP that gives rise to a splice variant [51]. This SNP would not have been identified as a high priority by our pipeline. Furthermore, although we have substantially complete sequence coverage of the Tir1 locus, at other loci we have used the Perlegen data, which is estimated to be about 45% complete [15]. Therefore although the candidate QTL SNP presented here are the most likely given the available data and annotation, both SNP data and annotation is incomplete and other candidates may be discovered in the future.

The correlation of Jukes-Cantor distances calculated from our 454 data and the published Perlegen dataset was only modest (r = 0.63). 32% of our 454 SNP loci were also in the Perlegen set, however the low correlation between the two sets shows that SNP discovery was uneven in one or both sets and inspection of the SNP distribution suggests that this was certainly the case in the Perlegen set. The uneven distribution of SNP discovery makes it much harder to undertake a consistent analysis across the genome using a single threshold for assigning alleles to haplotype blocks. However the high positive predictive value (Table S3 in Supporting Text S1) for identifying shared haplotypes suggests that this procedure should reliably exclude regions where C57BL/6 shares haplotypes with the susceptible strains. Nevertheless other more robust data types such as CNV and potentially functional SNP should still be surveyed in regions where haplotype does not correlate with phenotype. The more complete mouse resequencing projects that are currently underway should increase the predictive power of this approach substantially.

QTL involved with resistance to other parasitic diseases overlap with the Tir QTL, raising the possibility that polymorphisms discovered here may be involved in the response to other parasites. Leishmania resistance 1 (Lmr1) [52], Plasmodium chabaudi resistance QTL 3 (Char3) [53] and Heligmosomoides bakeri nematode resistance 2 (Hbnr2) [54] all overlap with Tir1. Similarly, the Tir3c QTL overlaps with a QTL for murine resistance to Plasmodium berghei-driven experimental cerebral malaria (Berr1) [55].

Thirteen genes around the peak of Tir1 show conserved order and sequence homology to a ∼311 Kbp region of BTA7 (15,412,179∶15,723,462 bp) where there is a QTL in cattle that regulates the level of parasitaemia in cattle infections with T. congolense [2]. This region includes Pram1, which has a probably damaging mutation that correlates with phenotype in mice and was the most plausible candidate gene in Tir1 and is therefore a candidate QTL gene in cattle as well. However since trypanotolerance QTL cover approximately 15% of the bovine genome it would be expected that at least one of the five murine QTL would coincide with a bovine QTL by chance (p = 0.56).

Conclusions

By linking genes to haplotypes, we have reduced the number of candidate genes in Tir1 to 43. Within these there were three genes with probably damaging nsSNP; CR974462.5, Rgl2 and Pram1. CR974462.5 is an anonymous gene in which the effects are hard to predict. Pram1 regulates oxidative stress in neutrophils and Rgl2 is involved in Ras signalling which can regulate inflammation. Pram1 is the closest to the peak of the QTL and has the best understood function making it the most attractive candidate at Tir1 however Rgl2 is also a plausible candidate. Probably damaging polymorphisms were identified in Srp72 in Tir2 and Thsd7b in Tir3a but little is known of their functions so it is hard to interpret these observations. Ptprc (Cd45) and Soat1 in Tir3b had probably damaging polymorphisms in conserved nucleotides, CD45 is the common leukocyte antigen and has multiple roles in cytokine signaling and cell regulation making it plausible candidate. Tir3c has a CNVR encompassing Cd244, which is differentially expressed and has a haplotype that correlates with phenotype in the four strains tested. Since gene dosage is lower in C57BL/6 it will be possible to test this hypothesis by inserting an extra copy of the Cd244 gene into the C57BL/6 background. Several other genes in Tir3c had haplotype and nsSNP that correlated with phenotype but none had such a distinct CNV and such strong differential expression.

By combining haplotype analysis, array-CGH, gene expression and next-generation DNA capture and sequencing, we have identified a small number of genetic polymorphisms that may be responsible for differences in response to T. congolense infection, demonstrating that this approach can systematically reduce the number of candidate genes under QTL to generate a short enough list of genes to test for function.