A new method for sequencing the hypervariable Plasmodium falciparum gene var2csa from clinical samples

As part of International Centers of Excellence for Malaria Research (ICEMR) and malaria-in-pregnancy (MIP) studies, samples were collected from children, men and women in Ndirande, a peri-urban township of Blantyre, in Malawi, where malaria transmission occurs throughout the year with a seasonal peak during a rainy season from December to March. Forty-three specimens, including 19 blood-spotted Whatman 3MM filter papers that had been stored at room temperature and 24 cryopreserved parasites, were randomly selected. Genomic DNA was extracted using a Qiagen Midi Kit according to manufacturer’s instructions. Four positive controls were used: 3D7, HB3, and synthetic mixtures of 70% 3D7 plus 30% HB3 and vice versa. 3D7 and HB3 each had an initial concentration of 1 ng/µl.

Primers that flank the region spanning the var2csa upstream promoter (position 500 bp upstream of start codon) to the DBLepam4 domain were designed using 20 aligned var2csa sequences from public databases (GenBank and VarDom) along with 12 var2csa upstream promoter (UPSE) sequences obtained previously [30]. Each secondary PCR primer was labelled using PacBio barcode sequences (listed in Additional file 1: Table S1) to identify sequences from individual samples. High-fidelity Takara LA Taq^® Polymerase (TAKARA BIO INC, Shiga, Japan) was used for the PCR with conditions listed in Additional file 1: Tables S2 and S3. An expected secondary PCR product of 5364 bp was resolved on 0.8% agarose gel and visualized using Gel Doc XR+ System (Bio-Rad, Hercules, CA, USA). Successfully amplified samples were purified using a MultiScreen filter plate (MultiScreen, Tullagreen, Germany). DNA concentration was measured using a Quant-iT PicoGreen dsDNA assay (Life Technologies, Carlsbad, CA, USA) according to manufacturer’s instructions. Subsequently, 48 equimolar barcoded amplicons were pooled to a final total of 2 µg of DNA.

Purified, pooled amplicons were sequenced using Single Molecule Real Time (SMRT) technology on a PacBio RS II sequencer (Menlo Park, Pacific Biosciences, California, USA), at the Genomics Resource Center of the Institute for Genome Sciences, University of Maryland School of Medicine, USA. Briefly, libraries were constructed by ligating SMRTbellTM adapters to the barcoded-pooled amplicon template. One SMRT™ cell with P4-C2 chemistry (P4 polymerase with C2 sequencing chemistry) was used. A 180-min movie was performed on PacBio RS II to generate the reads.

The secondary analysis was performed with the PacBio SMRT Analysis v2.3.0 package, using the long amplicon analysis (LAA) algorithm with the following options—minLength 3250—minSnr 4. This algorithm enables read clustering, phasing and consensus calling, and consensus filtering. The pooled amplicon data were de-multiplexed by barcode. The predicted accuracy of 95%, which is defined as the threshold below which a haplotype (consensus sequence) is considered as noise, was used to filter high-quality consensus sequences for downstream analysis.

To determine the sequence accuracy of the positive controls, amplicon consensus sequences were aligned to coding sequences of the respective strains from which the amplicons were generated. The corresponding test coding sequences were extracted and the phylogenetic tree was built using the Maximum Likelihood method implemented in MEGA 6.0 [31].

Sequences from the same sample that shared 99% identity at the nucleotide level were clustered using CD-HIT [32]. This conservative threshold was chosen assuming that the residual sequencing error was 1% based on data from the resequenced positive controls. Sequences with depth of coverage equal to or greater than 100× were considered for downstream analysis. This 100× threshold was chosen because this was the minimum depth of coverage at which a perfect consensus sequence of the reference 3D7 was obtained. Then the amplicon sequences were fed into OrfPredictor [33] to predict open reading frames, and corresponding amino acid sequences were generated. Amino acid sequences were used to annotate constituent domains using the VarDom 1.0 server [2]. Multiple alignments were performed using MAFFT with L-INS-i accuracy-oriented options. The region in the multiple alignments corresponding to the FCR3 variant of ID1-ID2a was extracted (1158–3075 corresponding to amino acid position N386 to D1025).

For the genetic diversity analysis, sequences corresponding to the coding regions were extracted from the full-length amplicon. Likely frameshift errors were corrected with AlignWise [34]. Sequences were then aligned with MAFFT with the L-iNSi option. DnaSP v5 [35] was used to compute population genetic parameters, including nucleotide diversity, which is the average number of nucleotide differences per site between any two given sequences; haplotype diversity, which is the probability that two randomly sampled alleles are different; and Tajima’s D neutrality test, which is used to infer the selective forces acting on a locus based on the difference between the number of segregating sites and the average number of nucleotide differences.

The amplification protocol was developed using genomic DNA from two laboratory parasite reference strains, 3D7 and HB3. The primers are designed to bind to the VAR2CSA-specific upstream promoter, UPSE, and to the DBLepam4 domain, to amplify a segment close to 5 kb in length (Fig. 1). This primer combination readily amplified var2csa from both of these lines. To test the robustness of the protocol, it was applied to clinical samples including DNA extracted from blood-spotted filter papers and from cryopreserved specimens obtained from 81 clinical malaria samples from Malawi. A long PCR amplicon of the expected size was obtained from both of these sample types. The PCR was successful for all cryopreserved specimens (24/24), whereas only 33% (19/57) of the tested filter samples yielded visible PCR products by gel electrophoresis. Therefore, the starting combined pool consisted of 43 amplicons, in addition to the four positive controls from 3D7 and HB3 (described in “Methods” section).

All 43 amplicons and the 4 controls (see “Methods”) were multiplexed in one PacBio SMRT cell, yielding 95,220 raw polymerase reads, with average and median length of 5.3 and 4.4 kb, respectively (Additional file 2: Figure S1). The raw reads were processed by removing the circularization (barbell) adaptors, resulting in 161,534 sub-reads. Applying a filter with quality score cut-off of 0.75 (Phred score ≥20) and read length ≥50 bp, reduced this set to 143,603 sub-reads, with length ranging from 50 to 30,000 base pairs, for an average of 3055 sub-reads per sample. The presence of sub-reads longer than 5 kb indicates that some of the reads contain missing or unidentified barbell adaptors. The sub-read length distribution was bimodal with one peak around 2 kb and a second peak at the expected amplicon size of approximately 5 kb, and a median length of 2728 bp (Fig. 2).

To validate this amplification, sequencing and assembly approach, the nucleotide sequence of the positive controls was compared to the published VAR2CSA sequences for each strain (PFL0030c, HB3var2csaA, and HB3var2csaB). The sequences recovered from positive controls were virtually identical to the respective reference var2csa sequences in the 3D7 and HB3 genomes, with nucleotide sequence identity from 99.8 to 100% between the re-sequenced consensus and the corresponding reference allelic sequences (Table 1).

To evaluate the method’s performance with polyclonal infections, a synthetic mixture containing var2csa sequences from the two reference strains was sequenced. Each allele from the mixture was successfully reconstructed (Fig. 3). Sequences recovered from HB3 alone and from synthetic mixtures had between 99.82 and 99.98% identity to HB3 reference sequences (Table 1). The lack of complete sequence identity was related to low depth of coverage. The non-mixed 3D7 (referred to as 3D7_pacbio100 in Table 1) had 445-fold coverage, whereas the HB3 and the mixed-controls have a coverage range between 46- and 167-fold. The only difference observed between reconstructed and the reference sequence were deletions that occurred at homopolymeric regions (Additional file 1: Table S4).

To determine the minimum number of reads required for an accurate assembly, reads were subsampled to simulate the 3D7 assembly at different coverage levels. At 100-fold coverage, a sequence identical to the reference was obtained. However, when a larger number of reads was used, some artifacts were observed in the assemblies, such as one deletion at 125-fold and two deletions at 150-fold coverage, suggesting that as read coverage increases so does the likelihood of repeated erroneous sequences. Therefore, the relationship between accuracy and coverage was not always linear (Table 2).

The parameter settings used for sub-read assembly, optimized during the reconstruction of the reference 3D7 allele, were then applied to the clinical samples. After filtering, read clustering, and chimeric sequence removal, 135 consensus sequences were generated from the 43 samples. After collapsing nearly identical sequences (see “Methods”) from the same sample using a threshold for minimum nucleotide sequence identity of 99%, which is a cut-off based on the re-sequenced positive control (Table 1), the number of sequences was reduced to 99. A subsequent filter, using a minimum coverage threshold of 100-fold, yielded 51 sequences from 40 samples, with a mean coverage of 280-fold (100–500) (Additional file 2: Figure S2). All samples but one yielded one or more unique amplicon sequences (Additional file 3).

To further evaluate the specificity of the new protocol, in particularly whether in fact var2csa sequences were amplified, the translated sequences were annotated using the VarDom 1.0 server [2], which uses a Hidden Markov Model to identify PfEMP1 protein domains. var2csa-specific homology blocks (HB) and domains (DBLpam1, DBLpam2, CIDRpam, and DBLpam3) were detected in all samples with a single, intact open reading frame (30 sequences with length ≥1400 amino acids). The remaining 21 amplicons were also var2csa sequences, validated as such using blast searches against NCBI’s non-redundant sequence database, with frameshift mutations that resulted in in-frame stop codons. These results validate our approach to generating var2csa sequences (Additional file 2: Figure S3).

To characterize the genetic variation in the 5′ region of the locus, the 5 kb segment amplified from the var2csa gene was used. The analyses were based only on sequences with an intact open reading frame that spanned the whole 5 kb region; partial sequences were excluded. A total of 30 full-length sequences were included in the analysis. A total of 4613 nucleotide sites (sites with gaps/missing data were excluded) were analysed, of which 1408 were polymorphic. The overall nucleotide diversity across this region was 11.16% (π = 0.1116). When analysed by constituent domains, DBLpam3 showed the least variation, with a nucleotide diversity of π = 0.056, followed by CIDRpam (π = 0.104) and DBLpam2 (π = 0.106), whereas DBLpam1 was the most diverse domain, with π = 0.138 (Table 3). To further determine the distribution of genetic variation across this region of the gene, we calculated nucleotide diversity using a sliding window of 99 nucleotides in length, and sliding step of 30 nucleotides, across the coding region of the 5 kb segment. The most polymorphic regions corresponded to the interdomain region between DBLpam1 and DBLpam2, as well as a central region of DBLpam1 and the N-terminal region of DBLpam2 (Fig. 4a).

To better understand the evolutionary forces that shape the genetic variation of var2csa, we estimated the distribution of two different statistics across the this segment, namely the ratio π _N/π _S, which measures the ratio of non-synonymous to synonymous mutations per site, with expected value of zero when the types of mutations are randomly distributed; and Tajima’s D statistic, which assesses the evolutionary forces operating on a locus, with expected value of zero under neutral evolution. A π _N/π _S ratio greater than 1 is suggestive of positive selection, whereas a ratio of less than 1 suggests purifying selection. On antigenic loci, positive values of Tajima’s D are suggestive of balancing selection, while negative values are likely due to purifying selection. None of the statistics showed a statistically significant departure from neutrality either across the whole 5 kb segment or on individual domains (Table 3). The overall ratio π _N/π _S was 0.89, indicating a slight excess of synonymous over non-synonymous polymorphism.

To determine whether specific short regions of the gene show evidence of positive selection, these two statistics were estimated using a sliding window analysis, with a window size of 99 nucleotides and a sliding step size of 30 nucleotides (Fig. 4). The Tajima’s D analysis identified some regions where the distribution of genetic variation departed from neutrality (p < 0.05), corresponding to positions 666–764 in the DBLpam1 domain; 3563–3694 in the CIDRpam domain; and 3866–3964 and 4391–4573 in the DBLpam3 domain (Fig. 4a). One striking π _N/π _S peak was present, which mapped to DBLpam3 (alignment positions 4041–4139 with πN/πS ratio of 6.47) but with relatively low overall π, suggesting that whatever polymorphism there is in this region is non-synonymous. Two intermediate peaks were also observed, which mapped to the C-terminal regions of the CIDRpam domain (alignment positions 3246–3344 with πN/πS of 2.73) and the DBLpam3 domain (4707–4856 with πN/πS ratio of 3.92), each with a significantly high value of Tajima’s D, suggesting that balancing selection is acting at these regions (Fig. 4). Taken together, these results showed that the N-terminal of var2csa is highly polymorphic. To put var2csa genetic diversity in context, its nucleotide diversity is ten times higher than that in the P. falciparum apical membrane antigen 1 (AMA1), which itself is one of the most diverse loci in P. falciparum [14]. The high values of Tajima’s D in the DBLpam1, the CIDRpam and the DBLpam3 domains of the protein are consistent with balancing selection. These findings are in line with those from other studies that found that variable blocks in DBLpam1, DBLpam2 and DBLpam3 are evolving under positive selection [26, 36, 37]. Taken together, these analyses suggest that host immune pressure drives the diversity in these segments, likely under a selective regimen of frequency-dependent selection that favours rare alleles.

Sequence diversity of the CSA binding region, located between coordinates 1185–3297 in Fig. 4, was characterized since this region is a target of the current VAR2CSA sub-unit vaccine [17]. The overall amino acid sequence identity was 74% and the average nucleotide diversity was 14.6% (Table 3), with a proportion of segregating sites of 53%. The nucleotide diversity in this region is higher than any of the values calculated for individual VAR2CSA domains.