Whole genome sequencing of Plasmodium falciparum from dried blood spots using selective whole genome amplification

In order to design probes that preferentially bind to the P. falciparum genome, a published PERL script [8] was used to select up to 100 (8–12 mer) primers with a predicted specified melting temperature (≤30 °C). The frequency of these primer sequences in the desired (D) Plasmodium falciparum 3D7 genome were compared to the contaminating (C) human genome (Fig. 1a). Top 50 primers with the highest desired/contaminating (D/C) ratios were selected for further analysis. From these 50, primers with more than three complementary nucleotides at 3′ and 5′ ends were removed to prevent formation of hairpin structures. To prevent primer–primer dimerization, primer pairs with more than three complementary nucleotides at their ends were also removed. A final 28 primers that passed the above quality control were ordered from Integrated DNA Technologies (Coralville, IA) as standard desalting purification with a single modification of phosphorothioate bond between the last two 3′ nucleotides to prevent primer degradation by the Phi29 polymerase exonuclease activity. Individual primers were reconstituted in Tris HCl (pH 8.0) buffer and pooled into three sets (probes) following the D/C ranking described above: the first set consisted of the first 10 primers (Probe_10), the second set consisted of the first 20 primers (Probe_20), and the third set consisted of all 28 primers (Probe_28). Each set was evaluated to determine which should be taken forward for further assessment.

To test whether selected primers would successfully amplify the parasite genome, mock clinical samples were prepared by mixing culture-infected red blood cells (infected with P. falciparum strain 3D7) with uninfected human whole blood to obtain a simulated parasitaemia ranging from 0.0001 to 1%. P. falciparum strain 3D7 parasites for the mock samples were cultured in human O+ erythrocytes with heat-inactivated 10% pooled human serum, as described in [11]. All parasitaemia calculations were based on the estimation of approximately 4 million red blood cells per microlitre of whole blood. DNA was extracted from the samples (N = 8) without leucodepletion. In addition, 6 other mock DNA samples were manually reconstituted by mixing P. falciparum genomic DNA with host (human) genomic DNA to obtain parasite/host DNA mixtures of the ratio 1: 24 (4% parasite and 96% human DNA) that were used to investigate genome coverage following sWGA.

To test the efficacy of sWGA in generating reliable genomic data from DBS, WGS datasets obtained from standard leucodepleted-genomic DNA (gDNA) were compared with sWGA-DBS samples. The gDNA samples were derived from two to three ml of venous blood which were then processed by CF-11 or MN filtration to remove leucocytes [5, 6] prior to DNA extraction. On the other hand, the DBS samples were collected by spotting (on filter paper) 50 μl of whole blood obtained by finger pricking. In total, samples from 156 patients that were positive for P. falciparum clinical malaria based on rapid diagnostic test (RDT) with CareStart™ Malaria kit (Access Bio Inc, USA) were analysed. Eighty-four DBS were collected from the Kassena-Nankana Districts of Upper East Ghana, of which 48 had matching VB pairs; and 72 DBS were collected from Noguchi Memorial Hospital in Accra, Ghana, all of which had matching VB samples.

Two to three ml of VB samples (mock blood or field) were used to extract DNA, using QiAamp DNA blood midi kit (Qiagen) following the kit manufacturer’s instructions. For DBS, DNA was extracted using QIAamp DNA Investigator Kit (Qiagen, Valencia, California, United States). Approximately 1.5 cm (0.6 in) diameter DBS circles from each filter paper were cut out into small pieces of 3 mm diameter using a single-hole paper punch. Punched pieces from each sample were placed into 2 ml micro-centrifuge tubes from which DNA was extracted following the manufacturer’s instructions except for the reagent volumes and incubation times, which were doubled to accommodate the increased amount of DBS used per sample. An average of 116 ng (standard deviation, SD, 116.7) of DNA was obtained from the DBS extracts out of which at least 5 ng was used as template for sWGA amplification reaction.

The sWGA reaction was performed in 0.2 ml PCR-tubes or plates. The reaction (50 µl total volume) containing at least 5 ng of template DNA, 1× BSA (New England Biolabs), 1 mM dNTPs (New England Biolabs), 2.5 µM of each amplification primer, 1× Phi29 reaction buffer (New England Biolabs), and 30 units of Phi29 polymerase (New England Biolabs), was placed in a PCR machine (MJ thermal Cycler, Bio-Rad) programmed to run a “stepdown” protocol consisting of 35 °C for 5 min, 34 °C for 10 min, 33 °C for 15 min, 32 °C for 20 min, 31 °C for 30 min, 30 °C for 16 h then heating at 65 °C for 15 min to inactivate the enzymes prior to cooling to 4 °C. Once the product was amplified, it was quantified using Qubit^® dsDNA high sensitivity (Thermo Fisher Scientific) to determine whether there was enough material for sequencing—minimum required is 500 ng of product. Standard whole genome amplified (WGA) products of the test samples were also sequenced as control to determine the extent of enrichment [12].

sWGA products (≥500 ng total DNA) were cleaned using Agencourt Ampure XP beads (Beckman Coulter) following manufacturer’s instructions. Briefly, 1.8 volumes of beads per 1 volume of sample were mixed and incubated for 5 min at room temperature. After incubation, the tube containing bead/DNA mixture was placed on a magnetic rack to capture the DNA-bound beads while the unbound solution was discarded. Beads were washed twice with 200 µl of 80% ethanol and the bound DNA eluted with 60 µl of EB buffer. Cleaned amplified DNA products (~05–1 µg DNA) were used to prepare a PCR-free Illumina library using the NEBNext DNA sample preparation kit (New England Biolabs) for high throughput sequencing. DNA libraries were sequenced at the Wellcome Trust Sanger Institute using Illumina HiSeq 2500 instruments and Illumina V.3 chemistry. Paired-end sequencing was performed with 100-base reads and an 8-base index read. 12-multiplex sample libraries were loaded to target at least 20 million reads per sample.

Sequence data obtained from each sample was subjected to standard Illumina QC procedures and 20 million reads per sample was subjected to detailed analysis for enrichment, quality, content, and coverage. Each dataset was analysed independently by mapping sequence reads to the 3D7 reference genome using BWA [13]. SAMtools [14] was used to generate coverage statistics from the BWA mapping output. For enrichment analysis, the number of reads mapping to either host, or P. falciparum reference sequences was counted. For genotype and concordance analysis, variant calls were generated using SAMtools mpileup (V0.1.1.19; with the following parameters: -DSV -C50 -m2 -F0.0005 -d 10,000 -gu) and bcftools (V0.1.17; with the following parameters: -p 0.99 -vcgN). A list of 1,241,840 (1.2 million) high-quality single-nucleotide polymorphism (SNP) positions, which were not filtered by gene class or region, but on individual properties of SNPs (such as uniqueness of the surrounding region and within an exon) [15, 16] was used. In silico genotyping of both the DBS (sWGA) and VB (leucodepleted and unamplified) samples was performed using mpileup to count alleles present in at least five reads (alleles with less than five reads were discarded). Although P. falciparum is haploid, it is common to find heterozygous calls due to the presence of multiple clonal infections in the same host. In order to genotype heterozygous sites, the 5/2 rule was applied, which requires at least two reads in both reference and alternative alleles, and the sum of both has to be higher than five reads [15]. SNP call concordance analysis between matching DBS and VB samples was performed on sequenced data targeting SNPS present in the core genome as well as key malaria drug resistance genes, such as crt (K76T involved in chloroquine resistance) [17], dhfr (N51I, involved in pyrimethamine resistance) [18], dhps (A581G, involved in sulfadoxine resistance) [19], mdr1 (N86Y, involved in multiple drugs including mefloquine) [20], and kelch13 (C580Y, involved in artemisinin resistance) [1].

Selected 28 primers were analysed individually (Fig. 1a) to determine their expected binding sites and distribution pattern across the P. falciparum genome. Each 1 or 2 kb block had at least one primer binding (Additional file 1: Figure S1). These 28 primers were pooled into three different sets (probes): Probe_10 (consisting of the first 10 primers), Probe_20 (consisting of the first 20 primers), and Probe_28 (a pool of all the 28 primers). In separate reactions, the three probes were used to amplify 5 ng of simulated mock samples (a mix of 3D7-infected red blood cells with uninfected human whole blood; N = 8) to determine which set gives optimal genome amplification and coverage. The amplified products were cleaned and the DNA quantified using Quant-iT™ PicoGreen^® dsDNA assay kit (Invitrogen) to determine the yield for each primer pool (Fig. 1b). Different yields were observed between the three primer pools: Probe_10 produced the highest average yield (2.5 ± 0.87 µg) followed by Probe_20 (1.85 ± 0.81 µg) and Probe_28 (1.2 ± 1.0 µg) (Fig. 1b). Whole genome sequencing of the amplified products were used to compared the quality of genome coverage (number of bases with at least 5x coverage) by each set (pool) and no significant difference was found (Spearman’s correlation: Probe_10 and Probe_20, R² = 0.97, p < 0.001; Probe_10 and Probe_28, R² = 0.96, p < 0.001; Probe_20 and Probe_28, R² = 0.97, p < 0.001). Probe_10 (Additional file 1: Table S2) was therefore chosen for all subsequent sWGA reactions based on amplification yield and cost.

To perform a more in-depth analysis on Probe_10, a mock sample containing a mixture of human (96%) and P. falciparum (4%) DNA was amplified and sequenced, as described in Methods. Using the Illumina short read sequence data obtained, the primer binding positions as well as the short-read sequence alignments was plotted against the reference genome using Circos software [21] for data visualization. Figure 2 shows the probe binding sites (middle circle) as well as the sequence reads coverage profile (outermost circle) on all 14 chromosomes (inner circle) of the P. falciparum genome. Probe_10 successfully amplifies the majority of the parasite genome, but the variable subtelomeres are not adequately covered (Fig. 2; Additional file 1: Figure S1). However, the coverage profile is uneven and does not correlate properly with the primer binding sites (Pearson’s correlation R² = −0.007, p = 0.3).

Previous analysis [15] revealed accessible and inaccessible regions in the P. falciparum genome. Inaccessible regions, mainly the telomeres, centromeres and sub-telomeres, are comprised of hypervariable and/or highly repetitive sequences that are difficult to assemble or map. The remaining parts (core genome) consist of mainly the coding sequences of relatively balanced-base composition, and are generally accessible in most genome analysis. In order to test whether sequences generated from sWGA samples would successfully cover the core genome, coverage profile of P. falciparum strain 3D7 samples sequenced as gDNA (3 samples without amplification), WGA DNA (3 samples amplified using optimized whole amplification method [12]) or sWGA (3 samples consisting of a mixture of 4% parasite and 96% human DNA, amplified using selective whole genome amplification method) was plotted. Figure 3 shows the coverage profile of the samples on chromosome 1, highlighting regions corresponding to the core genome. Unamplified DNA (gDNA) provided the most even and uniform coverage across the entire genome. Both WGA and sWGA samples produced relatively spiky and uneven coverage, with the sWGA producing higher coverage depths of uneven distribution.

The sequence data was analysed by comparing datasets from standard whole genome amplified (WGA) samples against their selectively amplified (sWGA) counterparts to determine the level of enrichment. An average of 73.2% (sd 4.4; N = 5) of the reads in the sWGA-treated samples mapped to P. falciparum. More than 18-fold enrichment of parasite DNA was achieved, depending on the extent of host contamination in the original sample (Table 1; Fig. 4). In contrast, data obtained from DBS extracts and amplified by standard WGA (no selective amplification) had <1% of reads mapping to P. falciparum and the rest (>99%) mapping to the host genome (Table 1; Fig. 4), demonstrating the efficacy of sWGA in selective amplification of parasite DNA.

To investigate the sensitivity of the sWGA application, genome coverage threshold by sequence data generated from samples with different levels of parasitaemia was analysed. In vitro infected red blood cells were mixed with human whole blood to simulate different levels of clinical parasitaemia ranging from 1.0 to 0.0001%. For samples with a parasitaemia of ≥0.005% [~6.25 parasites per 200 white blood cells (WBC)], ≥70% of the core nuclear 3D7 genome was covered at depth of ≥5× reads. However, the coverage dropped sharply for samples with parasitaemia below 0.005% (Fig. 5a; see Additional file 1: Figure S2 for detailed coverage distribution). The same dataset was used to analyse coverage of known important drug resistant loci in the genome [22]. As shown in Fig. 5b, a similar coverage profile was observed where all the 7 specified drug resistant loci were covered 100% at depths of ≥5× reads for samples with parasitaemia ≥0.005%.

Having established sWGA efficacy in mock blood samples, DBS field isolates collected from two sites in Ghana, with a parasitaemia ranging from 0.001 to 8.9% (1.25–11,125 parasites per 200 WBC or 40–356,000 parasites per µl of blood) were used to test the method. DNA was extracted from 205 DBS samples (average yield 116 ng, SD 116.7), which were subsequently subjected to sWGA (average yield 1399 ng, SD 502). From those, 156 (76%) passed the threshold of 500 ng for library preparation and were, therefore, whole genome sequenced.

A total of 156 DBS samples were analysed, excluding those with <50% of the core genome covered at 5× reads or less (N = 25). On average only 2.3% (SD 2.3) of the core genome of the 131 DBS samples was not covered at all (Fig. 6a), whereas 85% (SD 13) of the core genome was covered at 5× or more (Fig. 6b). The median coverage of the core genome was 29× (Fig. 6c).

As expected, samples with higher parasitaemia (above 0.1%) produced sequence data with better coverage at depths of ≥5×, whereas samples with parasitaemia lower than 0.03% had many positions covered at depths <5× (Fig. 7, F _(1,150) = 135.5, p < 0.001). In this dataset all samples with parasitaemia lower than 0.03% (N = 25) had more than 50% of the genome covered at 5× or less. There was one exception; a sample with 0.001% parasitaemia had 51.4% of the core genome covered at 5×. The samples with low parasitaemia had a much larger proportion of missing bases in the core genome (Fig. 7). Coverage of genes that are either responsible for, or associated with, anti-malarial drug resistance (Additional file 1: Figure S3) were analysed, and a general tendency of better coverage in samples of higher parasitaemia (>0.02%) was observed, while those with parasitaemia lower than 0.01% showed poor coverage across the genes.

Taken together, our data establishes 0.03% parasitaemia (40 parasites per 200 WBC) as the minimum threshold on which sWGA technology is capable of generating quality sequence data with coverage suitable for most genetic analyses on DBS field samples (Fig. 7, vertical dotted line marks the 0.03% parasitaemia threshold). The data also show that at least 180 parasite genomes per sample is required for efficient sWGA processing.

In order to further evaluate sWGA efficiency and suitability for genetic studies from DBS samples, a concordance analysis was performed using the set of 120 field samples with matched pairs of VB and DBS filter papers. Sequence data from both VB and DBS samples were analysed in parallel and genotyped against the ~1.2 million high quality SNP positions previously identified in the P. falciparum genome [15]. Genotype calls from matching VB (gDNA) and DBS (sWGA) sample pairs were analysed. In the gold standard VB samples, a median of more than 98% (N = 1,217,003) of SNPs were called in all the samples (Fig. 8). Overall, 93% (N = 1,154,911) of SNPs were called in the DBS samples, with a slight reduction at a lower parasitaemia (Fig. 8). The accuracy of the SNPs called was investigated by performing a concordance analysis between SNP calls made from VB and DBS samples. Samples that had <50% of the genome covered at 5× (N = 7) as well as all missing calls of the remaining DBS samples (total samples analysed N = 113) were excluded from this analysis. There was high concordance between the SNPs called in both VB and DBS samples, with an average of more than 99.9% (out of 1,241,840 SNPs, SD 10.97%) of calls being concordant (either Ref/Ref or Alt/Alt; Table 2). Only 0.04% (out of 1,241,840 SNPs, SD 0.08%) of calls were discordant (Ref/Alt or Alt/Ref).

The accuracy of SNP calls from sWGA-generated data was further tested using allele frequency concordance metrics. Using the VCF files targeting the 1.2 million high quality-biallelic SNPs, the population-level allele frequencies was analysed from matching VB and DBS samples, and strong correlation of non-reference allele frequencies (NRAF) was found between VB (gDNA) and DBS (sWGA) samples (Fig. 9; Pearson’s correlation Ρ = 0.99, p < 0.001). For more detailed analysis, allele frequencies of specific mutations for key malaria drug resistance genes—dhfr, mdr1, crt, dhps and kelch13—were analysed. Once again, high concordance between VB and DBS samples was observed (Fig. 9; Additional file 1: Table S1). In summary, after excluding lower quality samples with missing calls, very high concordance in population genetic data between VB and DBS samples was observed.