Differences in Gene Transcriptomic Pattern of Plasmodium falciparum in Children with Cerebral Malaria and Asymptomatic Carriers

The mechanisms underlying the heterogeneity of clinical malaria remain largely unknown. We hypothesized that differential gene expression contributes to phenotypic variation of parasites which results in a specific interaction with the host, leading to different clinical features of malaria. In this study, we analyzed the transcriptomes of isolates obtained from asymptomatic carriers and patients with uncomplicated or cerebral malaria. We also investigated the transcriptomes of 3D7 clone and 3D7-Lib that expresses severe malaria associated-variant surface antigen. Our findings revealed a specific up-regulation of genes involved in pathogenesis, adhesion to host cell, and erythrocyte aggregation in parasites from patients with cerebral malaria and 3D7-Lib, compared to parasites from asymptomatic carriers and 3D7, respectively. However, we did not find any significant difference between the transcriptomes of parasites from cerebral malaria and uncomplicated malaria, suggesting similar transcriptomic pattern in these two parasite populations. The difference between isolates from asymptomatic children and cerebral malaria concerned genes coding for exported proteins, Maurer's cleft proteins, transcriptional factor proteins, proteins implicated in protein transport, as well as Plasmodium conserved and hypothetical proteins. Interestingly, UPs A1, A2, A3 and UPs B1 of var genes were predominantly found in cerebral malaria-associated isolates and those containing architectural domains of DC4, DC5, DC13 and their neighboring rif genes in 3D7-lib. Therefore, more investigations are needed to analyze the effective role of these genes during malaria infection to provide with new knowledge on malaria pathology. In addition, concomitant regulation of genes within the chromosomal neighborhood suggests a common mechanism of gene regulation in P. falciparum.

significant difference between the transcriptomes of parasites from cerebral malaria and uncomplicated malaria, suggesting similar transcriptomic pattern in these two parasite populations. The difference between isolates from asymptomatic children and cerebral malaria concerned genes coding for exported proteins, Maurer's cleft proteins, transcriptional factor proteins, proteins implicated in protein transport, as well as Plasmodium conserved and hypothetical proteins. Interestingly, UPs A1, A2, A3 and UPs B1 of var genes were predominantly found in cerebral malariaassociated isolates and those containing architectural domains of DC4, DC5, DC13 and their neighboring rif genes in 3D7-lib. Therefore, more investigations are needed to analyze the effective role of these genes during malaria infection to provide with new knowledge on malaria pathology. In addition, concomitant regulation of genes within the chromosomal neighborhood suggests a common mechanism of gene regulation in P. falciparum.

Background
Of the five malaria species that infect humans, Plasmodium falciparum is the most common in Africa, and is the species that causes the most severe disease. According to the World Health Organization (WHO), the majority (86%) of the estimated 660 000 annual deaths related to malaria are due to P. falciparum infections that occur in children under the age of five years old living in Africa [1][2][3][4]. Despite an adequate antimalarial treatment, 10 to 30% of patients with severe malaria die, and 25% of survivors of cerebral malaria develop chronic neuro-cognitive impairment [5]. Indeed, the dramatic loss of lives due to malaria and the negative economic impact associated with the disease lead to the loss of approximately 412 billion US dollars in African countries yearly, further retarding their economic development [6]. At present, the only means to effectively control malaria in human hosts remains the use of antimalarial drugs. However, malaria parasites are capable of developing resistance to all currently available drugs, and only a few promising novel drug candidates are under development [7]. Furthermore, despite several candidates undergoing early phases of clinical development, a highly effective vaccine against malaria parasites is still unavailable. The identification of new therapeutic targets or vaccine candidates is hampered by the lack of knowledge on various molecular mechanisms used by malaria parasites to survive within the human host.
P. falciparum infection causes a large spectrum of disease manifestations, ranging from asymptomatic infection (AM) to uncomplicated malaria (UM) and severe and complicated malaria. Only a small proportion of individuals develops severe disease, including cerebral malaria (CM), severe malaria-associated anemia (SMA), and other life-threatening clinical syndromes, such as acute renal failure, acute respiratory distress, pulmonary edema, hemoglobinuria, disseminated intravascular coagulation, and circulatory collapse [8,9]. In areas of low transmission, severe malaria also occurs in older children (i.e.. 5 years old) and adults. After repeated malaria episodes, individuals develop natural protective immunity and considerably decrease their risk to develop severe and complicated malaria, suggesting acquired immunity and protection against severe disease [10][11][12]. The heterogeneity of clinical manifestations is due to both human and parasite factors [13][14][15].
Although the biological processes involved in the progression from acute uncomplicated malaria towards severe clinical state are mostly unknown, parasite gene expression changes depending on the clinical manifestations may occur in genes involved in virulence.
The relationship between parasite genetic profiles and malaria-related pathologies needs to be established to further understand the pathogenesis of severe and complicated malaria.
Genomic research on P. falciparum has advanced considerebly since the entire genome sequence of the P. falciparum 3D7 reference clone was determined. P. falciparum 23-Mb genome is AT-rich and comprises 14 nuclear chromosomes, a 6-kb mitochondrial genome, and a 35-kb plastid-like apicoplast genome [16]. The genome sequencing project revealed a large proportion of predicted genes which are likely to be involved in parasite evasion mechanisms from human immune system, and host-parasite interactions. The majority of these latter genes encode amplified gene families, such as the hypervariable variant surface antigen (VSA) genes (var, rif, stevor, and pfmc2TM) that mediate evasion from the host immune system, and genes involved in host cell invasion pathways. The achievement of gene transcription profiling of the intraerythrocytic developmental cell cycle, proteomic studies, and in silico sequence computational analysis have led to a substantial progress in defining gene functions and genetic regulatory networks [17][18][19][20]. However, the absence of data from a large number of P. falciparum field isolates associated with different clinical manifestations, ranging from asymptomatic carriage to uncomplicated malaria and severe disease, tends to limit our knowledge on the biological processes involved in parasite virulence.
It has been shown that parasites isolated from children with severe malaria (SM) express a limited and conserved set of VSA antigens (VSA-SM) that are more strongly and more commonly recognized by IgG from malaria-exposed individuals than VSA expressed by parasites infecting children with uncomplicated malaria (VSA-UM) [21]. Therefore, the identification of VSA-SM encoding genes may reveal new therapeutic and vaccine targets for malaria control.
The aim of this study was to analyze and compare the transcriptomes of P. falciparum parasites isolates from asymptomatic carriers and patients with clinical malaria, including UM and CM, by DNA microarray hybridization, in order to identify potential virulence factors associated with severe clinical manifestations. As controls for our experiments, we analyzed and compared the transcriptome of 3D7 reference clone to that of 3D7 Lib that expresses VSA-SM related to severe pathogenesis [21,22] and differentially transcribes a set of var and rif genes.

Patients
After obtaining verbal informed consent from the children's parents or guardians, Cameroonian schoolchildren aged ,12 years old were mass-screened in November 2009 to identify asymptomatic P. falciparum malaria parasite carriers (AM) in Ekoundouma, a village located about 5 km west of Yaoundé (Latitude:3.8644652˚, Longitude:11.5138617˚). Thick smears prepared from fingerpricked capillary blood samples were stained with 10% Giemsa, and the parasite density of P. falciparum was determined by microscopy. Children with positive thick blood smear who had not taken antimalarial treatment within the previous two weeks and who were afebrile at the time of mass screening and during the previous one week were enrolled after obtaining verbal informed consent from their parents or their caretakers. Children with gametocytemia, mixed infections with P. malariae, or fever (axillary temperature . 37.5˚C) were excluded. Schoolchildren with .1,000 asexual parasites/mL of blood and symptoms associated with malaria were treated with artesunate-amodiaquine combination and paracetamol, as recommended by the Cameroonian Ministry of Public Health.
Symptomatic patients with acute uncomplicated malaria (UM) were enrolled among patients consulting spontaneously at the Nlongkak Catholic missionary dispensary in Yaoundé from January to December 2009. The inclusion criteria were as follows: age ,12 years old, P. falciparum parasitemia § 0.1% (or . 5,000 asexual parasites/mL of blood), absence of other Plasmodium species, fever (axillary temperature . 37.5˚C) at the time of consultation or history of fever within 24 h preceding consultation, absence of signs and symptoms of severe and complicated malaria, as defined by the WHO, and denial of recent self-medication with an antimalarial drug. The enrolled patients were treated with artesunateamodiaquine and paracetamol.
Children presenting with cerebral malaria (CM) were enrolled at Olembe hospital in Yaoundé, Cameroon, between April 2008 and December 2009, and at the Centre National Hospitalier Universitaire Hubert Koutoucou Mega (CNHU-HKM) in Cotonou, Benin, in 2012 and 2013 between June and August. CM was defined as a Blantyre coma score of # 2 persisting for 30 min and/or at least two seizures within 24 h preceding consultation, and no other obvious cause of coma. The patients were given the appropriate treatment and necessary medical care at the hospital, according to the current guidelines for the treatment of severe and complicated malaria recommended by the Cameroonian and the Beninese Ministries of Health.
Clinical history, physical and neurological examinations, as well as laboratory examinations, including P. falciparum parasitemia, creatinine, glycemia, and Creactive protein, were registered on an ad-hoc data form.
After written informed consent was obtained from symptomatic children's parents or legal guardian, P. falciparum-infected blood was collected by venipuncture. Approximately 3-5 mL of blood was collected in ethylene diamine tetraacetic acid (EDTA)-coated tubes. The study protocol was reviewed and approved by the Cameroonian National Ethics Committee and the Cameroonian Ministry of Public Health (authorization No 028/CNE/DNM/07) as well as by the ethics committee of the Research Institute of Applied Biomedical Sciences, Cotonou, Benin (No 006/CER/ISBA/12 and N˚21/CER/ISBA/13).
We obtained verbal informed consent from all parents or guardiants of schoolchildren who participated in the current study, including those from whom venous blood was not collected.
For all symptomatic children enrolled in the study, written informed consent was obtained from their parents or legal guardians.
Verbal informed consent complied with the national ethics committee rules and recommendations at the time the study was performed.
The written informed consent form and its recording procedures were approved by the two ethics committees.

Cultivation of P. falciparum parasites
Two laboratory-adapted parasite lines were used. Cryopreserved stocks of 3D7 reference clone and 3D7-Lib line derived from 3D7 [22] were cultivated for 10 days using the standard method developed by Trager and Jensen [23]. Parasites were synchronized, and ring and late trophozoite stages were isolated using plasmagel [24]. The 3D7-Lib line was obtained by several rounds of incubation with Liberian hyper-immune plasma using anti-IgG coupled with streptavidineconjugated Dynabeads, as previously described [21]. The 3D7-Lib line was shown to express group A var and neighboring rif genes, suggesting that it mainly expresses variable surface antigens (VSAs) associated with severe disease [25][26][27].

Sample collection
A total of 214 blood samples were collected (111 from asyptomatic children, 70 from children with uncomplicated malaria, and 33 from children with cerebral malaria). Three samples contained circulating late trophozoites, one from a patient with uncomplicated malaria and 2 from asymptomatic carriers. Of 111 samples obtained from asymptomatic children, 6 were excluded due to mixed infections with P. malariae and 11 were excluded due to the presence of circulating P. falciparum gametocytes. Forty-four of 61 (72%) UM samples, 20 of 33 (59%) CM samples and 31 of 48 (64%) isolates from asymptomatic carriers grew successfully to late trophozoite stage.

Maturation of Plasmodium falciparum isolates
Venous samples containing late trophozoites were excluded, and only samples with 100% ring stage parasites were used in the present study. Samples were washed in cold phosphate-buffered saline (PBS) by centrifugation and divided into two fractions of 200 mL cell pellets. One fraction was immediately stored at 280˚C in Trizol (Life Technologies) for RNA extraction. The second aliquot was cultivated for 24 to 38 h to obtain late trophozoites. Isolates that developed into at least 85% of late trophozoite stages were harvested and stored at 280˚C in Trizol.

Microarray design and hybridization
For each clinical group, three samples were constituted by pooling equivalent amounts of parasite RNA from six isolates to perform array hybridization. To ensure that the same quantity of RNA from each isolate was analyzed, we used isolates of approximately the same parasitemia. For the controls (3D7 clone and 3D7-Lib line), three samples (at either ring or late trophozoite stage) were obtained from three independent parasite cultures.
Microarrays were manufactured by Agilent Technologies using SurePrint in situ synthesis technology (Agilent Technologies France SAS, Les Ulis, France). Fifteen slides of eight arrays each containing 14,882 oligonucleotides covering 5,534 Open Reading Frame (ORF) sequences derived from the P. falciparum database were used. The oligonucleotides were designed using oligoarray2 software [28] by setting the general parameters to 55-60 bases for oligonucleotides length with a melting temperature between 79 and 81˚C, and with a 39 bias of 1,500 nucleotides since RNases preferentially act from the 59 end. The melting temperature range was relaxed for genes for which no oligonucleotide could be found with the default parameters. The array contained two sets of oligonucleotides. The first set was designed on the basis of the 3D7 sequence genome. These oligonucleotides were chosen within the coding sequences (CDS) of protein-coding genes in the nuclear genome. A minimum of three oligonucleotides per gene were designed when possible. To avoid cross-hybridization with human genes, cross-hybridization was computed using both 3D7 and human genome libraries. The second set of oligonucleotides was designed to hybridize to genes exhibiting large sequence variations, including var, rif, stevor and Pfmc-2TM multigenic families. For these genes, all sequences available in the GenBank and those kindly provided by our collaborators (T. Lavstsen, Centre for Medical Parasitology, University of Copenhagen, Denmark) were retrieved and added to the array. Sequence alignment using Clustal W and Clustal X version 2.0 [29] was performed for each family or sub-family, and specific oligonucleotides were designed. For the other genes with high sequence variations (i.e., genes with high numbers of single nucleotide polymorphisms [SNPs]), sequences were retrieved from PlasmoDB, and specific oligonucleotides were designed.
Total RNA from 200 ml of P. falciparum-infected cell pellet and laboratoryadapted parasites (3D7 and 3D7-Lib) was extracted using chloroform-isopropanol method as recommended by the manufacturer. The integrity of the total RNA from each sample was checked using Agilent 2100 Bioanalyser (Agilent Technologies, France). For reverse transcriptase (RT)-real time PCR (rtPCR), RNA was treated with DNAse at 37˚C for 15 min. The absence of contaminating DNA was assessed by 40 cycles of rtPCR using primers targeting seryl tRNA transferase and fructose biphosphate aldolase genes, which are highly conserved P. falciparum housekeeping genes.
Total RNA (6-10 mg) was transcribed using the SuperScript First-Strand Synthesis System (Life Technologies) and a mixture of random hexamer oligonucleotides. The newly synthesized cDNA was labeled using SuperScript indirect cDNA labeling system (Invitrogen, Carlsbad, CA) to generate green fluorescent Cyanin-3 (Cy3) labeled cDNA according to the manufacturer's instructions. The quantity of labeled cDNA was measured using Nanodrop (Thermo Fisher Scientific, Villebon-sur-Yvette, France) to ensure that the same quantity of labeled cDNA was deposited for each array hybridization.
After quantification of the fluorescent specific activity of Cy3-labeled cDNA, the quantities of targets were normalized at 100 pmol of Cy3, and hybridized for 24 h at 65˚C. The slide was washed with water and dried at room temperature. Slides were scanned using GenePix 4000 A scanner (Axon), and the images were analyzed using GenePix5 software (Axon). Dusted and poorly fluorescent spots (below the background level) were excluded from analysis. After array gridding, data were exported as gpr files for statistical analysis.

Quantitative rtPCR
Quantitative rtPCR was performed to confirm array results. For each target gene, a pair of primers was designed to obtain specific DNA sequence fragments that span less than 350 bp. Primer sequences are presented in Table S1. Individual real-time rtPCR amplifications were carried out in a final volume of 20 ml in a Frame Star 384-well plate (4titude France, Bagneux, France) containing 16 final concentration of SYBR Green JumpStart Taq ReadyMix (Sigma-Aldrich, Saint-Quentin Fallavier, France), 100 nM gene-specific primers, and 1 ml cDNA. rtPCR was performed using 7900 HT (Applied Biosystems, Saint Aubin, France). The thermal cycle program was as follows: 10 minutes of pre-incubation at 95˚C followed by 95˚C 6 15 sec and one min at 60˚C for 40 cycles. All samples were run in triplicate. Threshold cycles (Ct) and melting curves were analyzed using Applied Biosystems ABI SDS software 2.3. The level of amplification referred to the number of cycles at which the PCR product was detectable and confirmed by the melting curve for each gene. Normalized data were used to quantify the relative levels of a given mRNA between samples using the DDCt analysis [30]. Before proceeding to the relative quantification using this method, the similarity in PCR amplification efficiency of the target gene and a house keeping gene (argenyl tRNA synthetase, PF3D7_1218600/PFL0900c) used as internal controls was checked.
To validate the array results by rt-qPCR, a set of 14 genes were chosen randomly from genes that were differentially expressed between the clinical groups and between 3D7 and 3D7-Lib laboratory-adapted parasites. These genes were grouped into three categories, depending on their fold change with the arrays: down-regulated genes defined as Fold change [FC] value ,2 units of expression), moderately up-regulated genes (FC value, 2-7 units of expression), and highly up-regulated genes (FC value . 7 units of expression). To further confirm our findings, additional rt-qPCR was performed for a set of var gene family found to be up-regulated by arrays analysis using a new batch of parasite RNAs isolated from children with CM or UM. The transcript abundance of var genes was expressed as Transcript units (T u ), defined as T u 5 2(5-DCt) [31].

Statistical analysis
Data generated from the arrays were analyzed using the R software 2.14.1 R Core Team 2012 http://www.R-project.org/ [32] and Bioconductor limma package [33].
Since hybridizations included only one condition per array, the resulting onecolor hybridization signal was first loaded in a limma-like RG structure with the red foreground signal set to 1.

Background signals (green signal/red signal)
Since the Agilent microarrays (which are usually used as two-color cDNA arrays) were used as single color arrays, the standard loess normalization method for twocolor microarrays did not apply in our study. Thus, a between-array quantile normalization which is the standard method for one-color Affymetrix arrays was applied to all arrays.
Background signals were set to 0 for both colors, and the RG data structure was transformed into MA using the RG.MA function, with logarithmic ratios in log 2 (green signal/red signal). A quantile normalization between arrays was applied to the full data set [34]. After exclusion of flagged (invalid) and control spots, differentially expressed genes were extracted using linear models and the empirical Bayes method implemented in limma [35], followed by a Benjamini and Yekutieli (BY) P-value adjustment with error value set to 0.05 [36].
The mean fold changes of differential expression between pairs of conditions were computed and a cut-off value of 2 was applied. Transcripts were considered as differentially expressed when all their probes had a fold change higher than 2 and a p-value lower than 0.05. Correlation (Rho) between qPCR and array results was determined by the rank-based non-parametric Spearman's test, and the Kendall test was used to test the correlation between our results and the published transcriptomic data.

Bioinformatics analysis
Functional analysis of the differentially expressed genes was performed using the Gene Ontology (GO) GOstats package in Bioconductor with default settings [37] and Sanger GeneDB P. falciparum gene annotations. The supplementary file lists the top five most overrepresented GO terms in each set (Table S2).
Differentially expressed genes were clustered according to their fold change using the neighbor joining algorithm with Lance-Williams dissimilarity updates (hclust function from R 2.15.1) starting from euclidian distance on fold changes.
Stage-matching analysis was performed by Canonical Correlation Analysis (CCA) to search for potential correlation between the transcriptomic data of our samples at ring and late trophozoite stages and data of 3D7 obtained hourly during the 48-h intra-erythrocytic life cycle [18]. A multivariate statistical model was used to facilitate the study of linear interrelationships between two sets of variables and find a set of basis vectors for two multidimensional variables such that the projections of variables, called canonical correlation, onto these basis vectors are maximally correlated. Each CCA component is associated with the corresponding canonical correlation that characterizes the strength of dependency [38]. We used an R package, CCA, to develop numerical and graphical outputs [39].

Patients characteristics
Blood samples from 54 patients were used in the study: 18 children with cerebral malaria (CM), 18 children with acute uncomplicated malaria (UM), and 18 children with asymptomatic malaria (AM). The clinical and biological characteristics of these patients are summarized in Table 1. The mean age differences between patients of the three clinical groups were statistically significant (P ,0.05). At enrolment, the mean body temperatures were comparable between CM and UM groups (P . 0.05). As expected, the asymptomatic carriers were all afebrile, and their mean body temperature was significantly lower (P ,0.05) than that of the other two clinical groups. The CM group had significantly lower hemoglobin values than the UM group (P ,0.05). The geometric mean parasite densities of CM and UM groups were not significantly different (P . 0.05). Asymptomatic carriers had a significantly lower parasite density compared with that of other patient groups (P ,0.05). These differences in parasite densities were corrected before performing the arrays. To avoid bias in the expression pattern of abundant transcripts due to high parasitemia, RNAs from highly parasitized CM and UM isolates were diluted to obtain similar concentrations of RNA as in AM isolates.
Gene expression pattern of field isolates For each clinical group, three pools of six isolates were prepared and hybridized on arrays at both ring and late trophozoite stages. These isolates were collected from children with CM, UM, or AM. The transcriptome of isolates after maturation was compared to that of late trophozoites of 3D7 reference clone. Furthermore, the transcriptome of clinical isolates from children with CM was also compared to that from children with AM and to that from children with UM. We also compared the transcriptome of the parent 3D7 clone to that of the selected 3D7-Lib. Figure 1 presents the MA plot of differentially expressed genes  The Transcriptome of Plasmodium falciparum Cerebral Malaria Isolates between these two conditions and Figure S1 presents the raw p-values obtained from limma for the comparison of cerebral malaria with asymptomatic malaria. Our transcriptomic data from 3D7 reference clone was also compared to those of the published database [40] which yielded a high correlation (r 5 0..56, P 5 0.004) between the two transcriptomic data. The transcriptome of the rings was obtained by hybridization of cDNA from a mixture of 80% rings, 5% late trophozoites, and 15% schizonts for laboratoryadapted parasite lines. All field isolates consisted of 100% ring stages. In parallel, the transcriptome of late trophozoites of both laboratory-adapted parasites and clinical isolates was analyzed by cDNA hybridization using a mixture of 85% late trophozoites, 10% schizonts, and 5% rings.
Using one-color labeling method, the mean green fluorescence intensity after removing the background signal reflects the presence or absence of cDNA corresponding to each gene among the 5,534 analyzed ORFs. The analysis of transcriptomic data of 3D7 and 3D7-Lib ring stage parasites provided 168 transcripts with a mean fluorescence level below the background, which were undetected. Likewise, in late trophozoite stages, 61 of these 168 transcripts were also not detected. However, the transcriptome data of field isolates at late trophozoite stage showed only 38 transcripts with fluorescence intensities below the background level. Transcriptomic data from 3D7 reference clone and field isolates at the same developmental stages were highly correlated (r 5 0.79) in most (4,837 to 5,344) transcribed ORFs. These results suggest that we detected the majority of isolate transcripts at both ring and late trophozoite stages. The lists of non-transcribed genes in 3D7, 3D7-Lib, CM-associated isolates, and AMassociated isolates are provided in supplementary information Table S2, S3, S4, respectively.

Stage matching analysis results
CCA for stage-matching analysis showed that the transcriptome of our samples at the ring stage is positively correlated with the transcriptome of 3D7 between 1 and 8 hr and between 43 and 48 hr post-invasion. Similarly, the transcriptome of late trophozoite stage in our samples matches with that of 38 -48 hr post-invasion parasites ( Figure S3).

Differentially expressed genes in 3D7-Lib and 3D7
3D7-Lib expresses VSAs that are potentially implicated in severe and complicated malaria, including cerebral malaria. The comparison of this selected 3D7-Lib to the parent 3D7 reference clone showed some differences in the expression profile. A total of 395 transcripts were up-regulated in 3D7-Lib late trophozoites, as compared to the corresponding developmental stage of 3D7. Of these 395 genes, 52 displayed 3-to 9-fold higher expression in 3D7-Lib. Some of these genes encode ring exported proteins REX1, REX2, REX3, erythrocyte membrane protein PfEMP1 members, membrane-associated histidine-rich protein (MAHRP1), histidine-rich protein II (HRPII), Plasmodium exported protein (PHISTa), (PHISTb, and PHISTc), DnaJ protein, putative serine/threonine protein kinase family (FIKK4.2), and hypothetical proteins. A total of 296 genes were slightly down-regulated, exhibiting 2-to 3-fold change in the expression level. The comparison of ring-stage transcriptomes revealed 30 up-regulated genes, ranging from 4-to 30-fold increased expression in 3D7-Lib, and five down-regulated genes. The majority of the overexpressed genes encoded proteins belonging to PfEMP1 and RIFIN families, exported proteins, knob-associated histidine-rich protein (KAHRP), proteins with DNAJ domain, and some hypothetical proteins. The group A var genes [PF3D7_1100200/PF11_0008, PF3D7_1150400/ PF11_0521), PF3D7_1300300/PF13_0003), and PF3D7_0425800/PFD1235w], and neighboring rif genes [PF3D7_1100300/PF11_0009, PF3D7_1150300/PF11_0520, PF3D7_1300400/PF13_0004, and rif PF3D7_0425700/PFD1230c] were up-regulated in 3D7-Lib, as compared to 3D7, with an expression level ranging from 2.5 to 26-fold (Table S5). These genes were found to be up-regulated in 3D7-Lib in a previous study [22].

Identification of differentially expressed genes in CM compared with AM
Only the comparison between late trophozoites from CM and AM patients showed statistically significant differentially expressed genes, leading to 99 upregulated transcripts and 135 down-regulated transcripts (Table S5). However, the comparison between parasites from CM and UM patients showed no significant differences in the transcriptomic pattern.
The up-regulated genes in CM parasites compared to AM parasites exhibited 3 to 14 higher expression rates, while down-regulated genes showed 3-to 22-fold change. The 15 most up-regulated genes encode Plasmodium exported proteins, glycophorin-binding proteins (GBP), hypothetical proteins, VSA proteins, such as PfEMP1, RIFIN, other surface proteins like tryptophan-and threonine-rich antigen, and ring-infected erythrocyte surface antigen (RESA). Of the 99 upregulated genes in CM-associated parasites compared to AM-associated parasites, 39% were also up-regulated in 3D7-Lib, and included members of rif A1 and var A group, some of them being topologically located in the neighborhood. In parallel, 135 transcripts were down-regulated in CM parasites, as compared to AM parasites. Among these 135 transcripts, 36 showed a 5-to 22-fold decrease in transcription and corresponded to genes encoding rhoptry-associated proteins (RAP1, RAP2, and RAP3), merozoite surface proteins (MSP3, MSP7, and MSP9), Maurer's cleft 2 transmembrane domain protein, and hypothetical proteins.
We then compared the transcriptome of CM-and AM-associated late trophozoite isolates to that of cognate laboratory-adapted 3D7. A significant difference in gene expression between field maturated trophozoites from individuals with CM and those with AM, compared to 3D7 was observed. In CMassociated isolates, 182 transcripts were up-regulated and 85 were down-regulated, while in AM-associated parasites 40 were up-regulated and 41 were down-regulated. The most up-regulated transcripts in CM parasites presented 2-to 9fold changes, and the most down-regulated genes presented fold change ranging from 2 to 16 for the majority of transcripts. Most (50 to 67%) of the genes found to be transcribed in both CM-and AM-associated parasites compared to 3D7, belonged to phist rifin, stevor, pfMc2-TM and pfemp1 encoding genes. In addition to VSA-encoding gene detection, we observed the up-regulation of genes encoding Skeleton Binding Protein1 (SBP1), HRPIII, glycophorin binding protein (GBPH2), as well as genes encoding Plasmodium conserved proteins for which no function has been assigned yet. Down-regulated genes in CM-associated parasites included other members of Plasmodium exported proteins such as Pfmsp2, Pfmsp11 as well as other genes [PF11_0442, PF13_0090, PFC0185w, PF11_0381] encoding transcription factors with AP2 domain, ADP-ribosylation factor, membrane skeletal protein IMC1, respectively. PF10_0343 coding for S antigen was the most repressed transcript in CM-and AM-associated parasites. This gene might not be transcribed in these isolates.

Confirmation of array results using quantitative rt-PCR
The differential transcription level obtained by DNA hybridization arrays was further confirmed by qPCR using cDNA from the samples used for array hybridization ( Figure 2) and cDNA from a second set of samples (Figure 3). A highly significant correlation (rho 5 0.775; P ,0.0011, Spearman's rank correlation) between array and qPCR results was found, indicating that both methodological approaches yielded similar results and identified the genes that were either down-or up-regulated with accuracy ( Figure S2).

Bioinformatics analysis of differentially expressed genes
GO term analysis of up-regulated genes in CM-associated parasites compared to AM-associated parasites, as well as ring and late trophozoite stages of 3D7-Lib compared to those of 3D7, revealed GO terms of pathogenesis as the most over represented term (Table 2). Other over represented GO terms included cytoadherence to microvasculature mediated by symbiont protein, regulation of cell adhesion, and rosetting.
Clustering analysis performed for up-regulated genes according to their fold change exhibited a network of two clusters in up-regulated gene set of the three compared conditions, CM versus AM, 3D7-Lib T (trophozoite) versus 3D7 T, and 3D7-Lib R (rings) versus 3D7 R. For CM up-regulated genes, the first cluster contained three genes with PF3D7_0201700/PFD0085c; encoding acylcoA gene separated from PF3D7_1301200/PF13_0010 gene encoding GBPH2 and PF3D7_0701900/PF07_0004 encoding Plasmodium exported protein. Similarly, comparison of the first cluster of 3D7-Lib T versus 3D7 T contained PF10_0038; gene encoding 40S ribosomal protein (S20e) separated from PF3D7_0935900/ PFI1735c REX1, PF3D7_1301700/MAL13P1.61 Plasmodium exported protein, (GEXP07), PF3D7_1001500/PF10_0019 early transcribed membrane protein 10.1 (ETRAMP10). The first cluster of 3D7-Lib R up-regulated genes also contained four genes with KAHRP-encoding gene with the highest fold change, followed by three PfEMP1 encoding genes; PF13_0003, PF11_0521 and PFD1235w.
In three conditions, the second cluster contained the remaining up-regulated genes within two sub-networks in which some of the connected genes were grouped according to their chromosomal location or sequence and functional similarities, as PF3D7_0201600/PFB0080c, encoding PHISTb, PF3D7_0201800/ PFB0090c, encoding RESA-like protein with PHIST and DNAJ domains.

Discussion
Microarray-based transcriptomic studies are powerful approaches that have been successfully applied to identify many virulence genes in bacterial and cancer cell systems [41,42]. Due to the limited amount of RNA that can be obtained from clinical samples of P. falciparum-infected individuals, only a few studies have investigated whole transcriptomes of clinical isolates [43][44][45][46][47]. These studies have demonstrated substantial differences in gene expression patterns between clinical isolates and laboratory-adapted P. falciparum strains.
The present study aimed to compare and analyze the whole transcriptome of P. falciparum parasites isolated from patients with distinct clinical features, including asymptomatic carriage, acute uncomplicated malaria, and cerebral malaria, in order to gain new insights into the regulation of gene transcription underlying malaria pathology. Our results indicate noticeable variation in gene transcription pattern of parasite isolates compared to 3D7, as well as between the parent 3D7 and selected 3D7-Lib. The observed differential in gene transcription may result from the difference in the transcription level of conserved genes but may also reflect the transcription of other vsa genes that belong to the same group in the case of field isolates since several variants could be detected by the same probe. Therefore, more vsa genes were detected in field parasites compared to 3D7, which is consistent with other findings [43,44].
The observed variation in gene transcription pattern between parasites isolated from children with CM and parasites isolated from asymptomatic children may be The Transcriptome of Plasmodium falciparum Cerebral Malaria Isolates related to the parasite growth rate as well as host environmental factors, such as fever and inflammation that differ drastically between these two groups. Likewise, the absence of differentially expressed genes between CM and UM transcriptomes suggests similar transcription pattern since these parasite groups multiply efficiently in a similar host environment during malaria episodes characterized by fever which is absent in AM-associated parasites. This result was not in agreement with our rt-qPCR finding concerning the up-regulation of UPs A1, A2, A3, B1, DC8 and DC13 var genes in CM-associated parasites compared to UM-associated parasites using more isolates (Figure 3, A, B). Therefore, the similar transcriptomic pattern between CM and UM isolates may not include var genes and their cognate partners like phist genes which may be concomitantly transcribed [50]. To confirm the equivalence between two transcriptomes, more array experiments using a higher number of isolates are needed.
We then focused our attention on genes which showed the same regulation of transcription in both CM-associated parasites and 3D7-Lib compared to AMisolates and 3D7, respectively. Among these genes, there were genes encoding conserved proteins, exported protein, and variant surface proteins as phist, rifin, and UPs A1 A3 and B1 var. The difference in the transcription profile of a set of these genes confirmed by rt-qPCR showing the up-regulation of specific var gene groups targeted by UPs A1, A2, A3, B1, var2, var3 and DC8, DC13 in CM-associated parasites, compared to AM-and UM-associated parasites, might be of biological significance (Figure 3). These assumptions are reinforced by our test for enrichment on gene ontology categories of up-regulated genes in 3D7-Lib and CM-associated parasites which yielded the over representation of GO terms related with pathogenesis, regulation of cell or cell-cell adhesion, rosetting, and interaction with host. We also compared our transcriptomic results to published data and found that the majority of transcripts that were not detected in our study in ring and trophozoite stages of 3D7 and 3D7-Lib were also absent in previous array data [17,18], suggesting untranscribed ORFs in in vitro cultivated parasites. However, many transcripts that were absent from the 3D7 transcriptome were present in our transcriptomic data of field isolates and in the transcriptome of other clinical isolates [47], suggesting that these transcripts are not essential for asexual in vitro survival within the red blood cells. Among the undetected transcripts in ring and trophozoite stages in 3D7 clone, most were amplified gene families that encode PfEMP1, RIFIN, STEVOR, and PFMC2TM as well as unknown hypothetical proteins.
We also compared our transcriptomic data from ring and late trophozoite stages of 3D7, as well as from parasite isolates, to those of the corresponding stages of 3D7 obtained by RNA sequencing [40] and found a highly significant correlation (p 5 2.2e-16, Kendall test) between these data. These results demonstrate the robustness of the array hybridization method to provide consistent data that are comparable to RNA sequencing and reflect the conserved nature of basic biological processes, including general metabolism and cell cycle, in both in vivo and in vitro conditions. This high correlation is also consistent with the predominance of ring stage parasites in the peripheral blood of infected individuals and changes in the transcription pattern leading to the progression and development towards the late trophozoite stage.
We subsequently investigated the function of 99 up-regulated genes in parasites from children with CM compared to the parasites from children with AM, and found that 28% of up-regulated genes encode exported proteins, with the majority (67%) of them being VSA-encoding genes, such as var, rif, phist, and surfin. These genes are implicated in parasite sequestration and immune evasion. Interestingly, var group A/Ups A1, A2, A3, and/or UpsB1, [PF3D7_1150400/PF11_0521, PF3D7_1300300/PF13_0003, PF3D7_0937600/PFI1820w, PF3D7_0632800/ PFF1595c, PF3D7_0425800/PFD1235w] as well as CD8 and DC13 were confirmed to be up-regulated in CM-associated isolates and in the 3D7-Lib. 3D7 PF3D7_1150400/PF11_0521 displays DBLa-1.7 and CIDR-a1.2, while PF3D7_0400400/PFD0020c displays CIDRa-1.1, DBLb12, DBLc6. These architectural domains characterize Domain Cassettes (DC) DC13 and DC8, respectively, which were shown to be transcribed at higher levels in parasites from children with severe malaria in Tanzania [31,48], as compared to the parasites from children with uncomplicated malaria. Similarly, PF3D7_1300300/ PF13_0003, PF3D7_0425800/PFD1235w contain DC5 which was associated with severe malaria in one study [31] and also with parasite binding to the human receptor PECAM1 [49]. Furthermore, transcription of PF3D7_0400400/PFD0020c PF3D7_1220100/PFL0970w PF3D7_1231600/PFL1525c pre-mRNA splicing factors were transcribed at higher levels in CM-associated parasites, compared to AMassociated parasites and 3D7. PF3D7_1220100/PFL0970w was highly expressed in parasites from children and pregnant women [47]. However, the transcription factors with ApiAp2 domains PF3D7_1439500/PF14_0374, PF3D7_1143100/ PF11_0442 were slightly down-regulated in CM-and AM-associated parasites, compared to 3D7 which is consistent with the findings of Vignali et al. [47].
Several differentially expressed genes in the present study were similarly regulated in other studies [58], supporting the variable nature of expression pattern of these genes between field isolates and 3D7 clone. For instance, (REX1) ring-exported protein 1, PF3D7_0935900/PFI1735c, (REX2) ring-exported protein 2, PF3D7_0936000/PFI1740; (GIG) gametocytogenesis-implicated protein PF3D7_0935600/PFI1720w, (STARP) sporozoite threonine-and asparagine-rich protein PF3D7_0702300/PF07_0006; lysophospholipase, putative PF3D7_0702200/PF07_0005, conserved Plasmodium protein with unknown function PF3D7_0702400/PF07_0007, (Pfg27) gamete antigen 27/25 PF3D7_1302100/PF13_0011 were all up-regulated in AM-and CM-associated isolates analyzed in our study and were also among the top 20 up-regulated genes in clinical isolates in a previous study [59]. REX1, REX2 and GIG genes are located in close proximity in chromosome 9, while STARP and putative lysophospholipase, as well as a conserved Plasmodium gene are located within a close distance in chromosome 7, suggesting co-regulation of gene transcription due to their proximity. Similarly, when examining the topological location of all up-regulated genes in parasites from CM patients compared to those from AM carriers, three to 12 genes were located on the same chromosome at a distance ranging from 5 to 100 kb, a distance within the variation of expression can be controlled by a single promoter or the same epigenetic mechanism.
In conclusion, we identified a set of specifically up-regulated genes in CMassociated parasites. Many of these genes encode proteins of diverse functions including remodeling of host cell, protein exports, surface antigens proteins, and cytoadherence which are likely to take part in parasite survival in the human host and pathogenesis. Due to the hypervariability of variant amplified genes as var, phist and rifin families the accuracy of their up-regulation in CM parasites has to be confirmed by whole RNA sequencing methods. In addition, some discrepancies in the stages of in vitro matured parasites may be at the origin of gene differential transcription, as gene expression during intra-erythrocytic cycle of the parasite is tightly regulated.
The potential role of these genes in CM deserves further investigations to identify new targets for malaria control. Figure S1. Histogram of raw p-values obtained from limma for the comparison of cerebral malaria with asymptomatic malaria. doi:10.1371/journal.pone.0114401.s001 (TIF) Figure S2. Spearman's rank correlation (r 5 0.775; P ,0.001) between arrays hybridization and qPCR methods for gene expression level. doi:10.1371/journal.pone.0114401.s002 (TIF) Figure S3. Correlation circle plot: the variables are represented through their projections onto the plane defined by the new components (score vectors). The variables being assumed to be of unit variance, their projections are inside a circle of radius 1 centered at the origin of the circle. Strongly associated (or correlated) variables are projected in the same direction from the origin. The greater the distance from the origin, the stronger is the association. Two circumferences of radius 1 and 0.5 are plotted to reveal the correlation structure of the data.