Background

Plasmodium falciparum is responsible for 85 % of malaria cases and kills over one million people each year [1]. The most prominent virulent surface antigen in P. falcuparum is the protein PfEMP1 (P. falciparum erythrocytic membrane protein 1) encoded by the var multi-copy gene family [2–5]. To escape host immune detection, only small subsets of var genes are expressed at a time [6, 7]. Therefore, understanding the silencing mechanism of var genes is critical in the fight against P. falciparum.

Research over the past 10 years has identified var intron and promoter elements as key for var gene silencing. Daily et al. first observed that a luciferase gene can be silenced by placing a var intron at the gene’s 3′ end [8]. Using transfected plasmids Voss et al. found that a var gene upstream region is capable of silencing a gene thereby promoting mono-allelic expression [9]. Another study showed pairing a var promoter with a var intron or with a different promoter can repress luciferase gene expression suggesting var promoter silencing is similar to var intron-mediated silencing [10]. A model for intron-mediated silencing was proposed by Frank et al., which showed var promoters paired with a var intron always inhibit luciferase gene expression [11]. This result suggests a model that requires a one-to-one pairing requirement between var promoters and var introns for gene silencing. Recent studies using transfected plasmids provide more evidence that promoter and intron interaction is essential for var gene silencing [12, 13].

Other studies have suggested the heterochromatin marker H3K36me3 is involved in var gene silencing. Earlier work using high resolution ChIP-chip analysis demonstrated a link between var gene silencing and H3K9me3 levels [14]. Recent work revealed the heterochromatin marker H3K36me3 is present along the entire var gene body and var gene expression is associated with relatively lower H3K36me3 level [15].

Spatial organization also plays a critical role for var gene silencing [16]. Most var genes are tethered to the nuclear periphery, even though some of them are located on the chromosomal center region [14, 17]. A repeat region within the var intron is crucial for retaining a distinct perinuclear position [18]. Epigenetic research also shows that the heterochromatin marker H3K9me3 is associated with genes localized to the nuclear periphery [14]. Hence, tethering to the nuclear periphery seems to be an intrinsic requirement for var gene silencing.

Interestingly all var genes can produce transcripts [19, 20], but these transcripts are not polysome-associated [20]. This observation implies a post-transcriptional mechanism for var gene silencing. Recent works have proven this hypothesis. Zhang et al. showed an exoribonuclease named as PfRNaseII is involved in nascent var RNA degradation [21]. Long non-coding RNA also participates in var gene regulation. Antisense long non-coding RNA initiating from the var intron is associated with the active var gene [21]. Further experimentation demonstrates these antisense transcripts may contribute to remodeling of chromatin conformation along the gene body.

Although the 5′ upstream and intron regions play a central role in the var gene silencing mechanism, which epigenetic marker on these two regions is involved in var gene silencing remains unknown. Var genes have a long 5′ upstream region (longer than 1 kb). However, the location of the var regulatory region remains unsolved. Next-generation sequencing (NGS) allows screening of the regulation activity for all DNA elements located in P. falciparum. By using FAIRE-Seq (formaldehyde-assisted isolation of regulatory elements–sequencing) [22] and MNase-Seq (MNase-assisted isolation of nucleosomes-sequencing) data [23], we analyzed the chromatin accessibility of each var intron. Chromatin accessibility has been widely used for detecting active regulation elements in mammalian genomes [24–26]. RNA-Seq (RNA-sequencing) [20, 27] and cDNA sequencing data [28] were also incorporated to investigate the relationship between regulatory potential of different DNA elements and gene expression levels. Our results provide evidence that P. falciparum may have developed a specialized nucleosome landscape as a key mechanism to regulate its var gene expression. We have also characterized a specific 5′ upstream region that is associated with var gene expression. This work addresses a controversial issue in the field of var gene silencing and provides insights into the silencing mechanism of native var gene clusters.

Results

Genome wide intron scan reveals all var gene introns maintain high chromatin accessibility

Earlier transgenic works showed that the var intron is required for var gene silencing [12, 29]. Most researchers reasoned that the regulatory activity of the var intron is crucial for var gene silencing [10, 16, 30]. However, no study so far has directly investigated the regulatory activity of the endogenous var introns. To address this problem, we used FAIRE-Seq [22] at seven time point (hours 0, 6, 12, 18, 24, 30 and 36) and MNase-Seq [23] data at three different time points (hour 0, 18 and 36) throughout the parasite erythrocytic cycle to explore this activity. FAIRE-enriched regions have been shown to correspond to open chromatin states with high regulatory activity [24], On the other hand, MNase-Seq is a powerful method to map the genome wide distribution of nucleosome occupancy. MNase-enriched regions indicate lower regulatory activity [31]. We first observed most var genes, no matter which categories (ups A, ups B or ups C), always possess enriched FAIRE-Seq signal in their introns, while all the neighboring regions of introns keep higher enriched MNase-Seq signal during hours 0 and 18 (Additional file 1: Figure S1). However, at hour 36, the neighboring regions of var introns also show enriched FAIRE-Seq signal. Figure 1a shows the average FAIRE and MNase signal coverage around all var introns at hours 0 and 18 (Additional file 2: Figure S2 shows sequencing signals for hours 6, 12, 24 and 30 and 36, MNase-seq data is only available for hours of 0, 18 and 36). We observed that var introns keep significantly higher chromatin accessibility for transcription factor (TF) binding compared with its neighboring regions in ring and trophozoite stages (Wilcoxon-Rank-Sum test p-value < 2.2e-16 in hours 0, 6, 12, 18, 24 and 30 compared with all other genes with one intron). In addition, the var introns always keep significantly lower nucleosome occupancy in ring, trophozoite and schizont stages (Wilcoxon-Rank-Sum test p-value < 2.2e-16 for 0, 18 and 36 compared with all other genes with one intron represented as dash line in Fig. 1a).

Fig. 1
figure 1

Var introns show significantly higher chromatin accessibility. a Average genome-wide sequence read coverage around the intron. Both data sets show higher chromatin accessibility around var introns. Coverage is expressed as the fraction of the highest coverage value among all positions within the plot window. The yellow bar represents average intron length. b Boxplot shows FAIRE-Seq signal around introns grouped according to gene annotation (‘**’ represents p-value < 0.01, FDR <0.01 for all comparisons between different gene groups and var genes). c Average genome-wide heterochromatin sequence read coverage around the intron. Both data sets show lower heterochromatin signal around var introns. Coverage is defined as in Fig. 1a

Full size image

We also investigated whether the high regulatory potential of the intron is specific to var genes. The chromatin accessibility for different gene families was compared. These families comprise nearly all members of variant surface antigens, rifin, Pfmc-2TM and stevor, and other subtelomeric gene families such as CLAG, SURFIN, and many members of exported protein families such as DnaJ, hyp, and PHIST genes. For var genes, we observe a higher enrichment of FAIRE-seq signal (Fig. 1b, Additional file 3: Figure S3 and Additional file 4: Table S1, p-value < 0.05, FDR <0.05 during all time points). The significantly higher chromatin accessibility in var introns indicates var introns exhibit specific regulatory functions. To rule out that this signal enrichment is observed by chance, we assembled all the genes containing a single intron as the control genes. The average sequencing signal around introns of control genes was calculated and plotted as the red dashed line in Fig. 1a and Additional file 2: Figure S2. Compared with the control genes, the var introns always show significantly higher FAIRE-Seq, indicating var genes always keep high regulatory potential and low nucleosome occupancy.

Heterochromatin markers along the var gene body also play a key role in regulation [15]. We checked if the var intron keeps high levels of heterochromatin marker similar to the whole gene body. By using recent H3K36me3 and H3K9me3 ChIP-Seq data [15] (Data are only available on hours 18 and 42), we showed the var intron maintains low levels of H3K36me3 and H3K9me3 (Fig. 1c, Wilcoxon-Rank-Sum test p-value < 2.2e-16, compared with neighboring region). The level of heterochromatin makers around var introns is consistent with their open chromatin state, suggesting the var intron is involved in the var gene regulation mechanism.

2 kb upstream region of var genes keep high chromatin accessibility

Earlier works have shown that the var 5′ upstream flanking region is another important regulatory element for var gene regulation [9, 29, 30]. However, var genes have long (>1 kb) 5′ upstream regions compared with other genes (Fig. 2a, Wilcoxon-Rank-Sum test p-value < 2.2e-16). Since P. falciparum exhibits long rang interactions [32], the regulatory site of var genes might be far from the transcription start site (TSS). We used FAIRE-Seq [22] and MNase-Seq [23] data to investigate possible locations of 5′ upstream regulatory elements. We found most var genes exhibit high chromatin accessibility around the 2 kb upstream region in ring, trophozoite and schizont stages (Fig. 2b for hours of 0 and 18, Wilcoxon-Rank-Sum test p-value < 0.05 compared with all other genes with one intron). Additional file 5: Figure S4 shows FAIRE-Seq for hours 6, 12, 24, 30 and 36. Specific examples are shown in Additional file 6: Figure S5. Except for the 2 kb region, most of the 5′ upstream region of var genes shows lower chromatin accessibility and higher nucleosome occupancy during the ring and trophozoite stages (Additional file 7: Figure S6, Wilcoxon-Rank-Sum test p-value < 0.05 compared with all other genes with one intron). To test whether the regulatory ability of the 2 kb upstream region is unique to var gene regulation, we also compared FAIRE-Seq signal with other gene families (Fig. 2c for hours 0 and 18, and Additional file 8: Figure S7 for hours 6, 12, 24, 30 and 36). The results show the 2 kb upstream region of var genes exhibit significantly higher chromatin accessibility during ring and trophozoite stages (Wilcoxon-Rank-Sum test p-value < 0.05, FDR <0.05 in hours 0, 6, 12, 18, 24 and 30, Additional file 9: Table S2). The heterochromatin ChIP-Seq signal around the 5′ upstream region also supports our conclusion for hour 18 (Fig. 2d), with var genes showing a clear valley at the 2 kb upstream region. All these results suggest that the var 2 kb upstream regions are potential regulation regions and may be involved in the var gene regulation mechanism.

Fig. 2
figure 2

Var 2 kb upstream regions show higher chromatin accessibility. a Boxplot distribution shows var genes have longer 5′ upstream region (Control genes indicate all P. falciparum genes except var genes). Ends var indicates the 25 var genes located on the end of chromosome, while central var indicates the 37 of var genes located on the central region of chromosome (‘**’ represents p-value < 0.01, while ‘***’ represents p-value < 2.2e-16. P-value was calculated based on Wilcoxon-Rank-Sum test). b Average genome-wide sequence read coverage on 5′ upstream region. Both data sets show higher chromatin accessibility around 2 kb upstream region. c Boxplot shows FAIRE-Seq signal around the 2 kb upstream regions grouped according to gene annotation (‘*’ represents p-value < 0.05, FDR <0.05 for all comparisons between different gene groups and var genes). d Average genome-wide heterochromatin sequence read coverage on 5′ upstream region. Both data sets show lower heterochromatin signal around 2 kb var upsteam regions. Coverage is expressed as Fig. 1a

Full size image

The chromatin accessibility around the 2 kb upstream region and intron is associated with var gene expression level

Previous works showed var gene transcripts could be detected in blood-stage P. falciparum [19, 20], though these transcripts were not polysome-associated [20]. In line with these reports [19, 20], we observed most var genes produced transcripts using three published expression data sets [20, 27, 28]. We compared the expression of each var gene with the expression distribution of all other genes. More than 90 % of var genes have a p-value bigger than 0.05 indicating var gene expression levels are not significantly different than other genes. However, most var genes have low expression levels with only one or two of them exhibiting high expression levels (Additional file 10: Figure S8). In addition, the expression pattern between these three data sets is similar (Pearson expression correlation between three dataset >0.33). Therefore, we expected that the var genes with low expression levels would be consistent between the independent datasets. To investigate this assumption, var genes with the bottom 80 % expression levels were extracted independently from three data sets [20, 27, 28] and most of them consistently appeared in the bottom 80 % expression group (~78 % var genes were in the lowest 80 % expression group in all 3 datasets, Fisher’s exact test p-value = 2.58e-2 compared with expectation). We investigated whether the bottom 80 % of expressed var genes tend to exhibit higher chromatin accessibility around their intron and 2 kb upstream regions. Var genes with the bottom 80 % expression level were extracted and their chromatin accessibility around the two possible regulatory regions was compared with the remaining var genes. Figure 3a, b shows var genes with the bottom 80 % expression level (Otto TD et al. RNA-Seq data [27]) have significantly higher chromatin accessibility compared with the remaining var genes in ring and trophozoite stages. Additional file 11: Figure S9 and Additional file 12: Figure S10 shows similar result using Bunnik et al. [20] and Lopez-Barragan et al. [28] expression data. In addition, we also investigate whether other 5′ upstream regions including TSS, 0.5 kb, 1 kb, 1.5 kb and 2.5 kb upstream regions of var genes are also involved in regulation (Data are available in Additional file 13: Table S3 and Additional file 14: Table S4). The Pearson correlation values between FAIRE-Seq signals around these upstream regions and var gene expression was calculated. Var genes having higher expression levels show lower chromatin accessibility on two regulatory regions (2 kb upstream region and intron) during ring and trophozoite stages, seven correlation values for these two blood stages were calculated from three different expression data sets (2 for Bunnik et al. data [20], 2 for Otto TD et al. data [27] and 3 for Lopez-Barragan et al. data [28]). As shown in Fig. 3c, the chromatin accessibility only in the 2 kb upstream region shows negative correlation with var gene expression (Wilcoxon-Rank-Sum test p-value < 0.05 compared with all other upstream regions), suggesting only the 2 kb upstream region of var genes is associated with var expression level.

Fig. 3
figure 3

The chromatin accessibility of introns and 2 kb upstream region are associated with var gene expression. a, b Vioplots show the FAIRE-Seq signals of different var gene regions that are classified based on gene expression data (Otto TD et al. RNA-Seq data [27]). c The bee swarm plot shows Pearson correlation value between FAIRE-Seq signal on different var 5′ upstream regions and var gene expression from three independent expression data sets. The median value and standard deviation is represented by the black bar (‘*’ represents Wilcoxon-Rank-Sum test p-value < 0.05 compared with the second most significant correlation value). d Scatterplot shows FAIRE-Seq signal around var intron and 2 kb upstream region. X-axis and Y-axis represent log(FAIRE-Seq RPKM) value on var 2 kb upstream and intron region. e The bee swarm plot shows FAIRE-Seq signal correlation value between different var regions and exon during ring and trophozoite stages (‘*’ represents Wilcoxon-Rank-Sum test p-value < 0.05 compared with second significant correlation value)

Full size image

The “promoter-intron pairing model” suggests var gene regulation requires the presence of both the var promoter and intron [10, 11]. To test this model, we checked the correlation of chromatin accessibility between the var 5′ upstream region and the intron. As shown in Fig. 3d, the var 2 kb upstream region and the intron have significant positive Pearson correlation values during ring and trophozoite stages (Pearson correlation >0.31 for all time points, Bootstrap and Wilcoxon-Rank-Sum test p-value < 2.2e-16 compared with other genes with only one intron). Other 5′ upstream regions are also considered and the correlation value with the var intron calculated (Additional file 15: Figure S11). The result shows only the 2 kb upstream region exhibits significant chromatin accessibility correlation with the var intron (Wilcoxon-Rank-Sum test p-value < 0.05 compared with all other upstream regions), suggesting a “punctuated chromatin” status on intron and 2 kb upstream regions.

“Chromatin spreading” is another model for explaining var gene regulation, which states lower expressed var genes require the presence of heterochromatin, such as H3K36me3 along the entire gene body [15]. “Pairing” and “chromatin-spreading” models are not necessarily mutually exclusive. Here, we investigated whether the chromatin accessibility around var regulatory regions (2 kb upstream and intron) is associated with the heterochromatin marker along var exons. As shown in Fig. 3e, only the 2 kb upstream region of var genes shows significant negative chromatin accessibility correlation with var exons during ring and trophozoite stages (Wilcoxon-Rank-Sum test p-value < 0.05 compared with all other var regions), implying the regulatory ability around 2 kb upstream region may be associated with the chromatin accessibility around the var exon region. This result is also supported by H3K36me3 ChIP-Seq data [15] from the trophozoite stage, the FAIRE-seq signal of 2 kb upstream regions is positively correlated with heterochromatin H3K36me3 level around the var exons (Pearson correlation = 0.25, Bootstrap and Wilcoxon-Rank-Sum test p-value < 0.05 compared with other var regions). Interestingly, the chromatin accessibility around the 1 kb upstream region of var genes exhibits positive correlation with its exon during stages of ring and trophozoite (Fig. 3e), though it is not significant compared with TSS region (Wilcoxon-Rank-Sum test p-value = 0.15). This region was previously reported to have the H2A.Z marker enriched in this region of active var genes [33]. However, the 1 kb upstream region does not show significantly higher chromatin accessibility compared with its neighboring region (Wilcoxon-Rank-Sum test p-value > 0.35). Higher resolution chromatin accessibility data is required to test whether the chromatin accessibility around this region is associated with the chromatin marker of the var gene exon region.

The GC content of var regulation regions is associated with their chromatin accessibility

In an earlier study, Bunnik and coworkers reported AT-repeats segments strongly inhibit nucleosome occupancy (nucleosome occupancy is measured by MNase-seq) [23], implying that a correlation between GC content and their potential for chromatin accessibility. To see if the chromatin accessibility of the P. falciparum genome is uniformly correlated with GC content the correlation between GC content and chromatin accessibility during ring and trophozoite stages was calculated (see Methods, Fig. 4a). The FAIRE-Seq signal along the majority of P. falciparum genome region is anti correlated with its GC content (79 and 83 % genome region during ring and trophozoite stage have a correlation value lower than −0.1). The correlation value between chromatin accessibility and GC content is conserved during different blood stages (Pearson correlation value = 0.54, Wilcoxon-Rank-Sum test p-value < 2.2e-16 compared with the correlation value of shuffled genome, see Methods). However, not all genome regions show a negative correlation between chromatin accessibility and GC content (Fig. 4a). It is necessary to check whether the functional region of var genes also exhibit this anti-correlation between GC content and an open chromatin state. We first plotted the landscape of GC content around two var regulatory regions (Fig. 4b, c). The results show var 500 bp upstream, 2 kb upstream and intron regions show relatively lower GC content (Wilcoxon-Rank-Sum test p-value < 2.2e-16 compared with their neighboring region, Wilcoxon-Rank-Sum test p-value < 0.05 compared with control genes), implying the higher chromatin accessibility along var regulatory regions might be encoded in their sequence information. To further support this conclusion, the Pearson correlation between chromatin accessibility along two var regulatory regions during ring and trophozoite stages and their sequence GC content was calculated. There is a significant anti-correlation value on the var 2 kb upstream and intron regions (Fig. 4d, Wilcoxon-Rank-Sum test p-value < 0.05 compared with all other regions), suggesting the sequence GC content is associated with the chromatin accessibility around var regulatory regions.

Fig. 4
figure 4

Sequence GC content on two var regulatory regions is associated with their chromatin accessibility. a Scatterplot shows the correlation between sequence GC content and chromatin accessibility of genome 1 kb bins during ring and trophozoite stages (see Methods). Each point represents a 1 kb bin in P. falciparum genome. X-axis and Y-axis represent the correlation value between GC content of each 1 kb bin (each bin is divided again into 10 bins with length 100 bp) and its FAIRE-Seq signal during ring and trophozoite stages. Green histogram plot represents the distribution of correlation value. b Average sequence GC content around var intron and 2 kb upstream regions is plotted. c The sequence GC content distribution of different genome regions are plotted (‘C_2kb’ and ‘C_intron’ indicate 2 kb upstream and intron region of P. falciparum genes with one intron, ‘*’ represents Wilcoxon-Rank-Sum test p-value < 0.05). d The bee swarm plot shows Pearson correlation between sequence GC content of different genome regions and their FAIRE-Seq signal during different blood stages (‘*’ represents Wilcoxon-Rank-Sum test p-value < 0.05 compared with other elements)

Full size image

To exclude the possibility that the higher chromatin accessibility in the var intron and 2 kb upstream regions maybe due to the intrinsic lower GC content, we extracted intron and 2 kb upstream regions from P. falciparum genes based on their GC content distribution in the var intron and 2 kb upstream regions. This procedure was repeated 100 times and Additional file 16: Figure S12a and Additional file 17: Figure S13a show the same distribution of GC content on intron and 2 kb upstream regions between control group and var genes. We observed var intron and 2 kb upstream regions exhibit significantly higher FAIRE-Seq signal compared with control groups during all stages (Additional file 16: Figure S12b and Additional file 17: Figure S13b, p-value < 0.01 for all stages).

Discussion

The silencing of var genes is regulated via multiple mechanisms: (i) DNA regulatory elements such as var promoters and var introns [9, 34], (ii) histone modifications and epigenetic memory [15, 35, 36], (iii) change in subnuclear localization [14, 17], and (iv) post-transcriptional non-coding RNA [21, 37]. These different regulation mechanisms are not necessarily mutually exclusive, and all of them may be at work. It has been demonstrated that the var intron is involved in mechanism (iii) [18] and (iv) [37]. However, which epigenetic marker along the intron is involved in var gene silencing remains elusive. Similarly, although the var 5′ upstream region is an essential element for var gene silencing, it is difficult to identify an exact regulatory site in such a large region (>1 kb). Using next-generation sequencing (NGS) we can directly monitor how the epigenetic environment changes, in particular during the different stages of P. falciparum’s erythrocytic cycle. We can use NGS data to study the general properties of var gene regulation. By using FAIRE-Seq [22] and MNase-Seq [23], we found both var intron and 2 kb upstream regions show high chromatin accessibility during ring and trophozoite stages, suggesting high regulatory activity on these two regions. Further investigation shows the chromatin accessibility during these two stages is anti-correlated with var gene expression levels with the schizont stage being the only exception. One possible explanation for this exception is that an uncharacterized transcriptional repressor stage specifically binds these potential regulatory regions during ring and trophozoite stages, but not during the schizont stage. A future study using ChIP-Seq (Chromatin Immunoprecipitation Sequencing) could prove this hypothesis. Here, we provide DNA sequence evidence for such a regulator. The MEME algorithm [38] was used to search sequence motifs on var intron and 2 kb upstream regions. Two GC-rich motifs were found on the var introns (e-value = 1.1e-192) and two GC-rich motifs were also found on the 2 kb upstream regions (e-value = 8.1e-282). Additional file 18: Table S5, Additional file 19: Table S6, Additional file 20: Figure S14 and Additional file 21: Figure S15 list the most statistically significant locations (p-value < 1e-8) for these motifs.

Interestingly, var genes tend to exhibit open chromatin on both the var intron and 2 kb upstream regions. This result partly supports the “promoter-intron pairing” model were the promoter and intron need to be in an open chromatin state to repress var gene expression (Fig. 5a). Except for the var intron and 2 kb upstream regions, most var regions including the exons and TSS regions contain heterochromatin marker and exhibit lower chromatin accessibility (Fig. 5b). Both the “promoter-intron pairing” and “chromatin spreading” model explain aspects of var gene regulation. Since the var intron and 2 kb upstream regions tend to show punctuated chromatin states, we propose a new regulation mechanism called the “punctuated chromatin” model (Fig. 5c) that combines the “promoter-intron pairing” and “chromatin spreading” models. Although we observed epigenetic marker association between the 2 kb upstream region and this var exon this result still requires more evidence for validation. The “punctuated chromatin” model is consistent with the model proposed by, Avraham et al., which is based on the “promoter-intron pairing” model, were a chromatin loop between the var intron and promoter regions plays a role in maintenance of var gene silencing [13]. Simultaneous open chromatin states on both the intron and 2 kb upstream regions is in line with this loop model, which creates an ideal epigenetic environment for transcription factor binding. But, directly proving the existence of looping requires higher resolution chromosome conformation capture (3C) data.

Fig. 5
figure 5

Summary of var gene silencing model. a “Promoter-intron pairing” model. The var intron and promoter are required for var gene silencing. b “Chromatin-spreading” model. Lower expressed var gene exhibits higher heterochromatin marker levels along the gene body. c “Punctuated chromatin” model. Both “Chromatin-spreading” and “Promoter-intron pairing” work for var gene regulation. Lower expressed var genes tend to exhibit higher chromatin accessibility on both intron and 2 kb upstream regions, whereas lower chromatin accessibility can be observed on higher expressed var genes

Full size image

In this work, we show that the GC content throughout the genome is not associated with nucleosome occupancy everywhere. In var genes, only the GC content of the introns and 2 kb upstream regions is anti correlated with their chromatin accessibility. This result suggests the regulatory ability on these two regions is encoded in their sequences.

Conclusions

In conclusion, the chromatin accessibility of all var regions was analyzed systematically in our work to study the general properties of var gene regulation as a group. By using FAIRE-Seq and MNase-Seq data, we first showed that relatively lower chromatin accessibility in both the intron and 2 kb upstream regions is associated with higher var gene expression. The chromatin accessibility in these two var regulatory regions is positively correlated with their GC content, suggesting the level of var gene silencing may be encoded in their sequences. Interestingly, the chromatin accessibility of var introns and promoters shows a positive correlation, indicating some transcription factors might bind both. This result agrees with the previous hypothesis that a chromatin loop between the var intron and promoter might maintain gene silencing. We found that chromatin punctuation at a novel upstream element and the intron contributes to the var gene silencing mechanism promoting malaria antigenic variation.

Methods

Dataset and software package

In this work, two complementary high-throughput sequencing datasets, one created via formaldehyde-assisted isolation of regulatory elements sequencing (FAIRE-Seq) [22] and the other created using MNase-mediated purification of mononucleosomes sequencing(MNase-Seq) [23], were used to detect the chromatin accessibility of different genome regions. The data sets can be downloaded from the Short Read Archive (SRA010122 and SRA091359; http://www.ncbi.nlm.nih.gov/sra). H3K36me3 and H3K9me3 Chromatin immunoprecipitation followed by high-throughput DNA sequencing (ChIP-Seq) data [15] were used to detect heterochromatin markers. All the sequencing reads were mapped to P. falciparum 3D7 genome v 9.0 using BWA [39]. Using MNase-Seq procedure [40] and approximate nucleosome length [41], all the MNase-Seq sequencing reads are extended from the 5′ end for 150 bp (~1nucleosome length) toward the 3′ direction. The middle 75 nucleotides were extracted to represent the enrichment of the nucleosome signal. This extending procedure was also implemented in recent work [42]. Three independent gene expression data sets [20, 27, 28] normalized using the method described in Ferhat et al. [32] was used in our work to investigate gene expression level during P. falciparum’s erythrocytic cycle. The flanking regions [−200 bp, +200 bp] centered at different 5′ upstream sites TSS, 500 bp, 1 kb, 1.5 kb, 2 kb and 2.5 kb were extracted for chromatin accessibility investigation.

For sequence motif discovery, the MEME algorithm [38] was used to search for the most significant motifs in the var intron and 2 kb upstream regions.

Sequencing signal analysis

The reads per kilo-base per million reads (RPKM) is calculated for comparing the sequencing signal between different genome elements.

$$ RPKM=\left(10\hat{\mkern6mu} 9\times R\right)/\left({R}_{total}\times L\right) $$
(1)

Where R is the sequencing reads number in a specific genome segment, L represents the length of the segment and R total is the total mapped reads in the experiment.

The normalized coverage (NC) is calculated for plotting the sequencing signal around the var intron and 5′ upstream regions. It is expressed as the fraction of the highest coverage value among all bins within the plot window.

$$ N{C}_t=\frac{E_t\left({R}_i\right)}{\mathbf{ma}{\mathbf{x}}_{i=1n,t=1,2}\left({E}_t\left({R}_i\right)\right)}i=1,2,3\dots n;\kern1.5em t=1,2 $$
(2)

“t” is the gene type of sequencing signal, which t = 1 for var genes and t = 2 for control genes. “i” indicates the bin id and “n” indicates the total number of investigated bins. R i is the reads number in a specific bin i, E t (R i ) is the average number of reads for a specific gene type t in bin i. t = 1,2 max i = 1 : n (E t (R i )) is the maximum average reads number of all gene type t = 1,2 on all bins i = 1:n.

Whole genome scale correlation between sequence GC content and chromatin accessibility

For the correlation between GC content and chromatin accessibility on the whole genome scale, the P. falciparum genome (genome length 23,292,622 bp) is divided into equal length bins of 1 kb. Each 1 kb bin is represented as B i , where i = 1 to 23,292. Each B i is divided again into smaller bins b i with equal length 100 bp.

$$ {B}_i=\left({b}_1,{b}_2,\dots {b}_j\dots {b}_{10}\right) $$
(3)

The GC content (gc) and FAIRE-Seq normalized reads number (NRN) is calculated for each small bin b i . The GC content and FAIRE-Seq signal for each B i can be represented as two vectors.

$$ g{c}_{B_i}=\left(g{c}_{b_1},g{c}_{b_2},\dots g{c}_{b_j}\dots g{c}_{b_{10}}\right) $$
(4)
$$ NR{N}_{B_i}=\left(NR{N}_{b_1},NR{N}_{b_2},\dots NR{N}_{b_j}\dots NR{N}_{b_{10}}\right) $$
(5)

Then the Pearson correlation value for each bin B i can be calculated between \( g{c}_{B_i} \) and \( NR{N}_{B_i} \).

Correlation comparison

To compare the correlation value with control genes, the bootstrap method [43] was implemented to extract random samples of var genes and controls with replacement. Then the Wilcoxon signed-rank test is used to compare the median value for the two samples.

Shuffled genome generation

A Shuffled genome was generated for each chromosome C i , i = 1…14 with length L i . We assume each nucleotide n ij , j = 1… L i, independently follows the same distribution. This distribution comes from the percentage of each nucleotide type in P. falciparum.

$$ \left(\begin{array}{c}\hfill A=0.4\hfill \\ {}\hfill T=0.4\hfill \\ {}\hfill C=0.1\hfill \\ {}\hfill G=0.1\hfill \end{array}\right) $$
(6)