Direct whole-genome sequencing of Plasmodium falciparum specimens from dried erythrocyte spots

Blood samples (N = 199) were collected in Bandim, Guinea-Bissau, which represents many general obstacles encountered when setting up patient sampling in sub-Saharan Africa: the infrastructure is poor, and the connection to the electrical grid is unstable and expensive or completely lacking. The samples were collected from patients with uncomplicated malaria from October 2014 to October 2016. Inclusion criteria were: Informed consent, axillary temperature above 37.5 °C or a history of fever within the previous 24 h. Plasmodium falciparum mono-infection, parasite density ≥ 1000 P. falciparum/µl, age ≥ 6 months and absence of signs of severe malaria infection. Giemsa-stained thick and thin films were prepared and malaria species identified using a microscope. Parasite densities were calculated by counting the number of P. falciparum per 200 white blood cells, or up to 500 parasites. Approximately 2–3 ml of venous blood was drawn from each patient in EDTA-containing vacuum tubes.

Leukocyte depletion and dried erythrocyte spots (depicted in Fig. 1).

Venous blood samples were left to precipitate on the counter. Plasma and buffy coat were removed when the erythrocytes had precipitated using a Pasteur pipette. PBS was added to the remaining RBCs, creating approximately a 1:1 dilution and inverted 3–4 times. The RBC fraction + PBS mixture was sucked up using a sterile syringe and needle, and the syringe was then applied to the Plasmodipur filter (Europroxima, Arnhem, The Netherlands, Cat. 8011Filter25U). The mixture was filtered according to manufacturer’s protocol. The filtered mixture was left standing at room temperature for 3 h, to let the erythrocytes precipitate from the PBS. The PBS was carefully removed using a Pasteur pipette, and the erythrocytes were spotted on Whatman paper #3, as 3 Pasteur pipette-drops per spot, and left to dry overnight in closed drawers. Finally, the blood spots were packed in individual zip-lock bags containing desiccant, as well as sample ID numbers. Samples were stored at room temperature (approximately 25–30 °C) in a dark box, for 2–4 months, before shipment to Copenhagen, Denmark. In Copenhagen, the samples were kept at − 20 °C, for 6–8 months before DNA extraction.

DNA was extracted using the QIAamp DNA Mini Kit (Qiagen, Limburg, Netherlands, Cat. 51306), according to WWARN procedures [15], but eluting in 35 μl elution buffer.

A previously described qPCR method [16] comparing the presence of P. falciparum seryl-tRNA synthetase to Human Beta-2-microglobulin, correcting for the size difference between the P. falciparum and human genomes was done.

DNA concentrations in extracts were measured on the Qubit double-stranded DNA (dsDNA) HS assay kit (Invitrogen). Libraries for paired-end sequencing were constructed from DNA extracts ranging from < 50 ng/ml to 0.2 ng/µl, using the Illumina NexteraXT (Illumina, California, USA) Guide 150319425031942 and following protocol revision E. The Pooled NexteraXT libraries were loaded onto an Illumina MiSeq reagent cartridge using MiSeq reagent kit v3 and 500 cycles with a standard flow cell.

Paired end reads were analysed with MGmapper v. 2.0 [17, 18], available from the Center for Genomic Epidemiology (CGE). Reads were trimmed from bases with a quality score below 30 (Phred), and paired reads were mapped to the following libraries: (1) malaria, (2) protozoa, (3) bacteria, (4) viruses, (5) fungi and (6) humans, in “bestmode”. The malaria database consists of the 3D7 reference genome, 21 other accessible Plasmodium genomes from NCBI, and contigs generated from P. falciparum specimens obtained from malaria patients in Tanzania. The protozoa database does not contain Plasmodium species.

Alignment of malaria reads to the 3d7 genome was performed using BWA [19], which was then sorted and piled with SAMtools [20], using the–aa option for “absolutely all” bases, to ascertain complete listing of positions with 0 reads. SNP calling was performed with Assimpler as described previously [21].

Genomic position information and read depth was cut from the pileup file generated with SAMtools, to generate a separate coverage file, from which average read depths across 2000 bp were calculated. The result was used to visualize read distribution using the Circos software [22].

Figure 1 illustrates the sampling protocol applied, which was developed in order to allow sampling for direct WGS of P. falciparum specimens collected in Bandim, Guinea-Bissau, where electricity is scarce, unstable and expensive. The sampling procedure is described in detail in the methods section. Local laboratory assistants were shown once how to perform the procedure, and were equipped with Plasmodipur filters, Pasteur pipettes, falcon tubes, PBS-tablets, Whatman filter paper #3, desiccant, zip lock bags and both a written- and video-presented protocol. The cost of sampling was highly affected by the use of Plasmodipur filters for leukocyte depletion, which were chosen due to the unavailability of the cheaper alternative, CF11 cellulose. An alternative cellulose-product has since been identified [23], which can be used according to the CF11 cellulose protocols and, therefore, represents an inexpensive, yet efficient alternative to Plasmodipur filters [9, 23] for future implementation of this sampling procedure.

qPCR was performed on all 199 samples to analyse the ratios of parasite:human DNA. This was done in order to assess the applicability of the samples for direct WGS, as defined by a minimum ratio of parasite:human DNA of 2:1. It was decided that samples with human DNA content above this ratio would require excessive sequencing resulting in excessive costs. Samples were categorised as “applicable” or “inapplicable”, according to the threshold. qPCR analysis revealed that 73% (N = 145) of the samples were applicable for WGS. Parasitaemia for the 199 samples varied from 800 parasites/µl to > 81,633 parasites/µl, and logistic regression was performed to establish whether increasing parasitaemia would increase the odds of a sample being applicable for WGS, the input data are listed in Table 1. The acquired odds ratio (OR = 1.29) confirms that, the likelihood of the sample being applicable for direct WGS increases with increasing parasitaemia. The contamination risk is inherently higher when applying this protocol, as it is not performed under sterile conditions. It was therefore assumed that lower-parasitaemia samples would be more difficult to sample successfully, which is the reason for an inclusion criteria of minimum parasitaemia of 1000 parasites/μl. Contamination may also affect higher parasitaemia infections, and the risks include lysis of leukocytes prior to filtration (if for example the blood sample was left for longer than indicated at ambient temperatures), applying too much force during filtration and contamination by anyone handling the samples prior to DNA-extraction.

For the current study, five samples of varying parasitaemia (0.1–1.2%, see Table 2) were subject to paired-end sequencing on the Illumina Miseq. Raw sequences were quality-trimmed and mapped to a variety of databases, including a human database and a custom-made malaria database (see “Methods”), using MGmapper [18]. Table 2 lists the percentages of quality trimmed raw reads mapping to human, malaria and other databases. On average, the parasite content was 61% across the five samples, ranging however from 46 to 82%. In comparison, initial demonstration of leukocyte depletion with Plasmodipur filters revealed a median parasite content of 36.6% for samples ranging in parasitaemia from 0.7 to 9.9% [8], while demonstration of a comparable protocol applying CF11 cellulose resulted in an average parasite content of 66%, for samples with parasitaemias ranging from 0.4 to 7.3% [9]. Studies applying sWGA have demonstrated an average parasite content of 70% [10, 11]. The percentages of malaria, human and other reads in the samples analysed in the current study are not clearly correlated with parasitaemia of the infection, as has also been seen before [8]. The discrepancies may mainly be due to suboptimal leukocyte depletion in some samples (sample 1 contains 35% human reads) and/or the presence of environmental contamination of the sample, such as bacteria (sample 5 contains 36% “other”), as the samples are not processed under sterile conditions. The possibility of environmental contamination was anticipated, and illustrates the necessity of filtering the raw reads bioinformatically.

Reads mapping to the malaria-database were aligned to the 3d7 reference genome, to assess the coverage and depth obtained for the individual samples (Table 2). The data clearly demonstrate the expected relationship between parasitaemia of the infection and resulting coverage and read depth, illustrating that lower parasitaemia infections will require more sequencing to attain the same depth of coverage. The sample with lowest parasitaemia (sample 1, corresponding to 0.1%) resulted in a coverage of 87.5% of the 3d7 reference genome, and an average read depth of 16× (≈ 370 million bp). These results are comparable to results obtained using sWGA [10], where an infection with 0.1% parasitaemia gave coverage of ~ 90% with 400 million bp sequenced (depth = 17.5×). The samples with highest parasitaemia (sample 4 and 5) averaged on a coverage of the 3d7 reference genome of 98.4% and a read depth of 83× (sample 4 (1% parasitaemia) = 67 × depth and 97.8% coverage and sample 5 (1,2% parasitaemia) = 99× depth and 98.9% coverage). Although similar read depths have not been demonstrated for sWGA studies on P. falciparum, the sWGA studies indicate that a 3d7 reference genome coverage above 90% is difficult to achieve at 1% parasitaemia [10], or solely the core genome coverage is assessed, also just surpassing 90% coverage [11], which is likely due to low amplification of certain regions in the genome. The overall distribution of read depth acquired in the current study, is depicted in Fig. 2 as circular diagrams representing each of the five samples across all 14 3d7 reference chromosomes. Together with the chromosomal-percentage of uncovered bases (percentage of chromosome size not covered, Fig. 3), the data indicate a relatively evenly distributed sequencing depth across the genome, including subtelomeric regions, with lower-parasitaemia samples capable of generating comparable data to higher-parasitaemia samples, given the extra sequencing capacity. Also for sWGA studies, it has been shown that lower-parasitaemia samples mimic higher-parasitaemia samples in read distribution across the genome [10, 11].

Relatively small “dents” in the read depths can be seen centrally on chromosomes 4 and 7 (Fig. 2), mirrored by peaks of uncovered bases for these chromosomes on Fig. 3. These areas correspond to clusters of var genes located in these chromosomes, and may be caused by difficulty in mapping to these highly variable regions, but may also be caused by difficulties in sequencing these regions due to their increased tendency to form secondary structures in general [24]. The same “dents” have been shown for sWGA protocols [10].

The samples subject to whole-genome sequencing in this study have previously been subject to targeted sequencing [21], for analysis of resistance-conferring mutations in pfmdr1, pfdhfr, pfcrt, pfdhps and pfk13. The WGS data were, therefore, compared to the targeted sequencing data in order to assess the reliability of the WGS data. The results are listed (see Additional file 1: Table S1A), and the percentages of the genes covered by the WGS data are listed (see Additional file 1: Table S1B). All resistance-conferring SNP data for these five samples were confirmed. The only gene to not be covered 100% in the samples with lowest parasitaemia was pfcrt, which contains 12 AT-rich introns and, therefore, is expected to be more difficult to sequence, align and assemble. This not only confirms the reliability of the WGS data, but also illustrates that reliable SNP-analyses can be performed based on samples with coverage around 90% and an average read depth of 16×.