Cyclical regression covariates remove the major confounding effect of cyclical developmental gene expression with strain-specific drug response in the malaria parasite Plasmodium falciparum.

BACKGROUND

The cyclical nature of gene expression in the intraerythrocytic development cycle (IDC) of the malaria parasite, Plasmodium falciparum, confounds the accurate detection of specific transcriptional differences, e.g. as provoked by the development of drug resistance. In lab-based studies, P. falciparum cultures are synchronized to remove this confounding factor, but the rapid detection of emerging resistance to artemisinin therapies requires rapid analysis of transcriptomes extracted directly from clinical samples. Here we propose the use of cyclical regression covariates (CRC) to eliminate the major confounding effect of developmentally driven transcriptional changes in clinical samples. We show that elimination of this confounding factor reduces both Type I and Type II errors and demonstrate the effectiveness of this approach using a published dataset of 1043 transcriptomes extracted directly from patient blood samples with different patient clearance times after treatment with artemisinin.

RESULTS

We apply this method to two publicly available datasets and demonstrate its ability to reduce the confounding of differences in transcript levels due to misaligned intraerythrocytic development time. Adjusting the clinical 1043 transcriptomes dataset with CRC results in detection of fewer functional categories than previously reported from the same data set adjusted using other methods. We also detect mostly the same functional categories, but observe fewer genes within these categories. Finally, the CRC method identifies genes in a functional category that was absent from the results when the dataset was adjusted using other methods. Analysis of differential gene expression in the clinical data samples that vary broadly for developmental stage resulted in the detection of far fewer transcripts in fewer functional categories while, at the same time, identifying genes in two functional categories not present in the unadjusted data analysis. These differences are consistent with the expectation that CRC reduces both false positives and false negatives with the largest effect on datasets from samples with greater variance in developmental stage.

CONCLUSIONS

Cyclical regression covariates have immediate application to parasite transcriptome sequencing directly from clinical blood samples and to cost-constrained in vitro experiments.