Predicting functional and regulatory divergence of a drug resistance transporter gene in the human malaria parasite

Abstract

Background: The paradigm of resistance evolution to chemotherapeutic agents is that a key coding mutation in a specific gene drives resistance to a particular drug. In the case of resistance to the anti-malarial drug chloroquine (CQ), a specific mutation in the transporter pfcrt is associated with resistance. Here, we apply a series of analytical steps to gene expression data from our lab and leverage 3 independent datasets to identify pfcrt-interacting genes. Resulting networks provide insights into pfcrt's biological functions and regulation, as well as the divergent phenotypic effects of its allelic variants in different genetic backgrounds.

Results: To identify pfcrt-interacting genes, we analyze pfcrt co-expression networks in 2 phenotypic states - CQ-resistant (CQR) and CQ-sensitive (CQS) recombinant progeny clones - using a computational approach that prioritizes gene interactions into functional and regulatory relationships. For both phenotypic states, pfcrt co-expressed gene sets are associated with hemoglobin metabolism, consistent with CQ's expected mode of action. To predict the drivers of co-expression divergence, we integrate topological relationships in the co-expression networks with available high confidence protein-protein interaction data. This analysis identifies 3 transcriptional regulators from the ApiAP2 family and histone acetylation as potential mediators of these divergences. We validate the predicted divergences in DNA mismatch repair and histone acetylation by measuring the effects of small molecule inhibitors in recombinant progeny clones combined with quantitative trait locus (QTL) mapping.

Conclusions: This work demonstrates the utility of differential co-expression viewed in a network framework to uncover functional and regulatory divergence in phenotypically distinct parasites. pfcrt-associated co-expression in the CQ resistant progeny highlights CQR-specific gene relationships and possible targeted intervention strategies. The approaches outlined here can be readily generalized to other parasite populations and drug resistances.

Figure 1

Overview of the approach applied in this study to predict diverging functional and regulatory interactions associated with the drug resistance transporter gene, pfcrt . Genes significantly correlated to pfcrt (FDR ≤ 0.20) in CQR or CQS recombinant clones were obtained using transcriptional data from the 24 hr developmental stage (trophozoites) [12]. Regulatory candidates associated with the pfcrt co-expression network in CQR or CQS parasites and functional partners of the gene were then predicted by applying the triangle inequality [22] based approach (TrIPI) developed in this study to assess the topological position of pfcrt correlated genes. Validations of the predicted regulatory and functional biological processes associated with pfcrt in CQR or CQS were then performed by measuring dose responses to small molecules targeting the processes. Additional information is provided in Additional file 1: section A.

Figure 2

Co-expression of all genes with pfcrt gene in CQR and CQS parasites. (A) Correlation between the levels of each transcript in the genome to that of pfcrt, determined separately for CQS (x-axis) and CQR (y-axis) parasites. Grey region indicates genes whose correlation to pfcrt passed the threshold of FDR ≤ 0.20. (B) Average correlation between pfcrt and each transcript in 100 pairs of randomly sampled subsets of CQS (x-axis) and CQR (y-axis) parasites. Each subset of CQR or CQS parasites consists of transcriptional data from 8 parasite clones. (C) Average correlation between the transcript level of each gene to that of pfcrt in 100 pairs of randomly sampled subsets of CQS parasites. (D) Comparison of average pfcrt correlations in 100 pairs of randomly sampled subsets of CQR parasites. Like in (B), each randomly sampled subset of parasites consists of 8 parasites.

Figure 3

Prediction and validation of regulatory mechanisms underlying diverging co-expression networks. (A) Potential regulators of the pfcrt co-expression networks by interrogation of the topological relationships between pfcrt partners using the transitivity, t, score. Top scoring candidate regulators- the AP2 transcription factor PF3D7_1007700 (AP2-3) has the highest score in CQS while in CQR the AP2 regulator PF3D7_0420300 (AP2-2) has 3^rd highest score considering all genes correlated to pfcrt (FDR ≤ 0.20). The case of t =0 denotes functional (direct) pfcrt partners which also includes another AP2 transcription factor, PF3D7_0802100 (AP2-1). (B) Top scoring regulators are all part of a previously published high confidence protein-protein interaction sub-network [44] and interact with the histone acetyltransferase (Gcn5). Other transcriptional regulators physically interacting with Gcn5 include CAF1- a component of the CCR4-NOT mRNA deadenylase complex- and adenosine deaminase ADA2, leading to the hypothesis that the Gcn5 protein interaction network could be involved in integration of transcriptional regulation and mRNA stability [44]. (C) Validation of dysregulated histone acetylation as a potential regulatory mechanism using drug response assays. QTL mapping of quantitative dose responses to the HDACi apicidin in progeny of the Dd2 × HB3 genetic cross found significant association to genetic loci on chromosome 5, 57.3 cM (LOD = 5.4) and 8, 77.5 cM (LOD 2.3). The chromosome 5 locus includes a gene encoding CCR4 while the chromosome 8 locus contains CAF1, which physically interacts with Gcn5. (D) Validation of dysregulated histone acetylation using data from previous studies [53]. Promoters of the top 100 genes that are not correlated to pfcrt in CQS but show positive correlation in CQR (gain of positive correlation) carry vastly higher levels of H3K9ac compared to the average levels in all genes (Wilcoxon test P < 2.2 x 10⁻¹⁶). In contrast, H3K9ac levels of the top 100 genes that gain negative correlation are significantly lower compared to the genome-wide promoter baseline (Wilcoxon test P = 3.4 x 10⁻¹⁶).

Figure 4

Validation of dysregulated DNA mismatch repair (MMR) pathway. (A) Divergent co-expression of pfcrt and the msh6 gene in CQR versus CQS parasites. Each dot or triangle represents the transcript level of the 2 genes in CQR and CQS parasites, respectively. The solid line is a linear fit of the data from CQR parasites and the dotted line is for data from CQS parasites. (B) Quantitative dose response variation in the response to the DNA damaging agent MMS across the Dd2 × HB3 genetic cross. (C) QTL mapping results for the MMS dose response reveals two candidate loci- chromosome 4, 60.3 cM (LOD = 2.68, chi-square P = 0.00044) and 5, 20 cM (LOD = 3.02, chi-square P = 0.000194). (D) The predicted regulon of the candidate regulator AP2 in CQR (PF3D7_0420300) is enriched with DNA repair genes including msh2 and rad51, providing further validation of MMR dysregulation. Prediction of potential targets of this AP2 was performed using an independent transcriptional data set as described in methods [68].