Plasmodium falciparum transcription in different clinical presentations of malaria associates with circulation time of infected erythrocytes

Following Plasmodium falciparum infection, individuals can remain asymptomatic, present with mild fever in uncomplicated malaria cases, or show one or more severe malaria symptoms. Several studies have investigated associations between parasite transcription and clinical severity, but no broad conclusions have yet been drawn. Here, we apply a series of bioinformatic approaches based on P. falciparum’s tightly regulated transcriptional pattern during its ~48-hour intraerythrocytic developmental cycle (IDC) to publicly available transcriptomes of parasites obtained from malaria cases of differing clinical severity across multiple studies. Our analysis shows that within each IDC, the circulation time of infected erythrocytes without sequestering to endothelial cells decreases with increasing parasitaemia or disease severity. Accordingly, we find that the size of circulating infected erythrocytes is inversely related to parasite density and disease severity. We propose that enhanced adhesiveness of infected erythrocytes leads to a rapid increase in parasite burden, promoting higher parasitaemia and increased disease severity.

W orldwide over 200 million malaria cases occur yearly, out of which~2 million progress to severe disease, leading in 2019 to more than 400,000 deaths, mostly of African children under the age of five 1 . The major causative agent of malaria, Plasmodium falciparum, causes disease through continuous asexual intraerythrocytic developmental cycles (IDCs), each lasting~48 h and producing 8-30 new parasites 2 . Circulation of young parasite forms within each IDC, called ring stages, is a hallmark of P. falciparum malaria 3 , while more developed stages express parasite antigens at the host cell surface to promote their adhesion to the vascular endothelium and avoid splenic clearance 4,5 . P. falciparum infection can produce a range of clinical outcomes: remaining asymptomatic despite infection, presenting with fever and other non-specific symptoms in uncomplicated malaria, or exhibiting one or several signs of severe disease leading to cerebral malaria, severe anaemia, coma, pulmonary oedema, or metabolic acidosis 6 . This array of malaria presentations associates with parasite burden, and with different stages of a progressively developing protective immunity which increases with parasite exposure and host age 7,8 . Protection from severe malaria is rapidly acquired after a few clinical episodes in areas of high transmission 9 , and years of exposure result in mostly asymptomatic infections in adolescence and adulthood 8,10,11 . Accordingly, in malaria-endemic areas, parasitaemia with high parasite burden is often seen in young children with severe malaria 12 . In contrast, lower parasitaemia typifies uncomplicated malaria or asymptomatic infections in older individuals 13,14 , reflecting cumulative exposure and gradual naturally acquired immunity to malaria. Nevertheless, in agematched individuals with apparently similar parasite exposures, different clinical outcomes have been linked to parasites causing severe malaria with higher or lower parasitaemias 15 , or parasites promoting non-severe vs severe malaria 16,17 , without allowing for major conclusions regarding the contribution of parasite transcription to disease outcome. Recently, in a cohort study in the seasonal transmission setting of Mali, we compared parasites persisting at low parasitaemias in asymptomatic children at the end of the 6-month dry season with parasites causing uncomplicated malaria in age-matched individuals during the ensuing wet season, and found highly different transcriptional signatures. We determined that gene expression differences were strongly associated with the developmental stage of circulating P. falciparum within the~48 h IDC 18 . Indeed, during each IDC, P. falciparum follows a tightly regulated, developmental stagedependent 19 transcriptional pattern from merozoite invasion of erythrocytes, through the ring-and trophozoite-stages, until the multinucleated schizont releases new merozoites at the end of the IDC [20][21][22] , and parasite transcription from patients' blood can reveal the stage composition of circulating parasites 16,17,23,24 . Our recent data showed that during the dry season, P. falciparuminfected erythrocytes circulate longer within the~48 h IDC and low parasitaemia is maintained for several months through increased splenic clearance 18 , hinting that previously reported transcriptomes from P. falciparum field isolates causing distinct clinical outcomes may also have been influenced by differences in adhesion efficiency and circulation times of parasitized erythrocytes.
In this work, we revisit ten studies published between 2007 and 2020 that report parasite transcriptional differences between distinct disease severities, parasite burdens, non-negative matrix factorization (NMF) clustering of expression profiles, or transmission intensities to bioinformatically interrogate transcriptional differences dependent on developmental stage. Our approach centered on the tight transcriptional pattern governing the~48 h IDC 20 ; and based on the simple idea that genes that are differentially upregulated in more vs less-developed parasite stages comprise transcripts peaking at later stages of the IDC, whereas genes that are differentially upregulated in less-developed stages represent transcripts peaking earlier in the IDC. Accordingly, an approach using in vitro grown parasite lines previously assigned 4400 P. falciparum transcripts to a particular IDC stage, allowing predictions of the stage composition of mixed parasite samples 24 , which we also applied here. Furthermore, we use previously described algorithms integrating the parasites' full transcriptome to determine expected stage-compositions 17 , and the maximum likelihood estimate of average parasite age in hours post erythrocyte invasion (hpi) 16 of a given sample. Applying these bioinformatic methods to previously published studies we determine the expression profile of differentially expressed genes (DEGs) 20 , and predict the developmental stages present in each sample 24 . Our re-interpretation of the transcriptional analyses and the calculations of parasites' age in the various studies reveals new transcriptional signatures supporting the hypothesis that shorter circulation time of infected erythrocytes within the~48 h IDC associates with increased parasite growth rates, virulence, and disease severity.

Results
Similarities and differences in P. falciparum transcriptional profiles across studies. To investigate a possible broad signature of P. falciparum transcription associating with parasite burden and/or disease severity, we identified ten published studies between 2007 and 2020 that reported transcriptional differences between malaria parasites isolated from individuals presenting with a spectrum of malaria symptoms and parasitaemias and residing in areas of different transmission intensities (Table 1).
With the exception of one study that compared acute vs chronic P. coatneyi infection in non-human primates in the laboratory setting 25 , all studies analyzed P. falciparum samples from naturally infected humans. Seven studies reported transcriptional differences between P. falciparum samples collected from individuals presenting with distinct disease severities, parasite burdens, transmission intensities, or NMF-generated clusters immediately after blood draw 15,17,18,[26][27][28][29] , and two studies compared cultured late trophozoites or schizont stages of P. falciparum samples from individuals with different disease severities and parasitaemias 16,30 . The studies used different numbers of samples collected from diverse geographic settings, employed microarray or next-generation sequencing methodologies to quantify parasite transcripts, and used various analytical approaches to define differential gene expression. A varying number of DEGs were reported comparing higher vs lower parasitaemias 15,16,18,25,27,30 , severe vs mild malaria 16,17,27,29 , and mild malaria vs asymptomatic parasite carriage 18,25,30 , clusters of transcriptionally similar groups of samples 15,26 , or differing transmission intensities 28 (Table 1). Considering the seven P. falciparum studies that reported DEGs, we observed only limited and not consistent overlap of DEGs; there was not a single DEG that was common across the seven studies ( Fig. 1a and Supplementary Table 1). Moreover, there was no enrichment of common terms identified across studies within gene ontology (GO) or functional category (Fig. 1b). We noted, however, the overlap between DEGs found in lower asymptomatic parasitaemias in the dry season vs higher parasitaemias in clinical malaria cases in the wet season in Mali 18 , and those identified in lower vs higher parasitaemias in cerebral malaria cases in Malawi 15 , and some trend of DEGs connect with the similar directionality between the two studies. Transcripts increased in higher parasitaemias in Andrade  and between severe cases of Tonkin-Hill et al. and higher parasitaemias in Andrade et al. or Milner et al., where we also detected some connections of DEGs aligned with parasitaemia directionality ( Fig. 1a and Supplementary Data 1). In addition, although the GO and functional analyses did not identify significant enrichment of common pathways between studies (Fig. 1b), similar tendencies of gene expression of transcripts belonging to the fatty acid biosynthesis pathway, which appeared significantly increased in higher parasitaemias of Andrade et al., were detected in samples of higher parasitaemias in Milner et al. and Lee et al. (Fig. 1c). Likewise, the spliceosome pathway that was shown to be significantly increased in higher parasitaemias of Milner et al. showed a similar trend of overexpression in higher parasitaemias of Lee et al. and Tonkin-Hill et al. (Fig. 1c). The RNA binding pathway distinguished between groups in Milner et al. and Lee et al. (Fig. 1b, c), but also partially in Andrade et al., and in Tonkin-Hill et al. when samples were ordered by increasing number of housekeeping gene reads (Fig. 1c). Altogether these data highlight the modest overlap of DEGs between the datasets, while hinting that differences in parasitaemia observed across the studies may produce common transcriptional signatures.
Parasite circulation time drives transcriptional difference between asymptomatic dry season infections and mild malariacausing parasites in the wet season. Recently, we have shown that differences in circulation time of infected erythrocytes prior to cytoadhesion between persisting low parasitaemias at the end of the dry season in asymptomatic children and malaria-causing P. falciparum parasites in the wet season associated with major transcriptional differences that were linked to the developmental stage of circulating parasites 18 . Now, we quantitatively characterized those associations with a series of bioinformatic approaches centered on the tight transcriptional pattern governing P. falciparum during the~48 h IDC 20 . First, we outlined the expression pattern of DEGs upregulated in the low dry season parasitaemias or in the higher malaria cases' parasitaemias along the~48 h IDC in the reference HB3 P. falciparum parasite line in vitro, according to Bozdech et al. 20 , and obtained very different heatmaps and associated expression curves ( Fig. 2a and Supplementary Data 2). While transcripts upregulated in low dry season parasitaemias mostly showed high expression between 12 and 36 h post invasion in the reference HB3 P. falciparum parasite in vitro, the upregulated transcripts in the higher parasitaemias of clinical malaria cases in the wet season presented the opposite trend, higher expression between 0 and 12 h post invasion, or at very late stages after 40 h (Fig. 2a), when genes of merozoite invasion and early ring stages start to be transcribed again 20 . This agrees with the presence of more developed parasite stages in the low parasitaemias in dry season samples, and young ring stages in the higher burden malaria cases 18 . Interestingly, this gene-level metric of DEGs along the IDC in the reference HB3 P. falciparum in vitro was independent of the fold change in expression observed between dry season and malaria cases (Supplementary Fig. 1 and Supplementary Data 2). We then used another pre-established gene-specific metric, the time of peak expression of a transcript within the~48 h IDC determined through a periodogram analysis 24 . Examining the time of peak expression of upregulated DEGs in the dry season (low parasitaemias), and in malaria cases (high parasitaemias), we observed later peaks (24.1 h 95% CI 23.9, 24.4) in upregulated genes in the dry season, and earlier peaks (12.4 h 95% CI 11.1, 15.5) for the upregulated genes in malaria cases (p < 0.0001, Mann-Whitney) ( Fig. 2b and Supplementary Data 2). The DEGs between low parasitaemias in the dry season and higher parasitaemias in the wet season were also assessed with another approach proposed by Painter et al. 24 , which previously assigned a particular IDC stage to~4400 P. falciparum transcripts, thus allowing for the prediction of stage composition of  16 applied to the transcriptomes of parasites collected in the dry season and malaria cases determined that the expected age in hpi was higher in parasites circulating in the dry vs the wet season (p < 0.0001, Mann-Whitney) ( Fig. 2e and Supplementary Data 2). Altogether, these analyses using DEGs and the whole transcript data of low vs high parasitaemias highlight the strong link between the transcriptional signature observed in low parasitaemias and the increased age of the circulating parasites.
DEGs from earlier studies show increased parasite circulation time in low vs high parasitaemias. To determine if parasite circulation time could also associate with transcriptional differences identified in other studies, we applied these same    . 3b), likely due to the small number of DEGs found in the study and also possibly due to the largely overlapping parasitaemias between individuals included in nonsevere and severe cases, which were not significantly different.
The stage composition prediction according to Painter et al. 24 identified more developed parasites from upregulated transcripts in low parasitaemia groups across the different studies, while transcripts upregulated in the high parasitaemia groups associated with higher proportions of ring stages (p < 0.001, Fisher's exact test) (Fig. 3b); and again the comparison between nonsevere and severe cases without strong parasitaemia differences included in Lee et al. or Tonkin-Hill et al. resulted in much less pronounced differences in predicted stages (Fig. 3b). Rono et al., a study that did not segregate sample groups based on parasitaemia or severity of disease but instead compared samples from high vs low transmission settings across all parasite burdens, did not show pronounced differences in heatmaps or expression curves of DEGs upregulated in low and high transmission settings along the~48 h IDC in HB3 P. falciparum reference in vitro ( Fig. 3c and Supplementary Data 3), nor were the peak expression times according to Painter et al. 24 (Fig. 3c). Accordingly, the predicted differences of developmental stages associating with the DEGs upregulated in high or low transmission settings although significant were not particularly striking ( Fig. 3d and Supplementary Data 3).
Interestingly, in Daily et al., a study clustering samples from a range of clinical presentations through NMF according to their transcriptional similarity into three groups of samples without significant differences in parasite densities, our analyses nevertheless highlighted the presence of more developed parasites in samples of cluster 1 and cluster 3 than in cluster 2. Heatmaps and expression curves along the~48 h IDC showed later expression in DEGs upregulated in cluster 1 and 3 vs cluster 2, and the same was true for cluster 1 vs cluster 3, with DEGs upregulated in cluster 3 showing high expression in early hours of the IDC (Fig. 3e). Also, the peak expression times within the IDC according to Painter et al. 24 (Fig. 3f and Supplementary Table 3). This categorization is in line with what Daily et al. refer in their original study, namely, that cluster 2 is the closest to early ring-stage profiles and also the cluster with the highest average parasitaemia and thus likely younger in hpi than the other two clusters, but also that cluster 3 is the one with most severe cases and presenting higher parasitaemias than what is observed in cluster 1 26 is also in line with the predicted stage composition with later forms in the cluster 1 and 3 vs 2, and cluster 1 vs 3 (Fig. 3f).  16 .
The particular case of the study by Cordy et al. analyzing P. coatneyi in four rhesus macaques is the only study involving longitudinal parasite sampling, including acute malaria with high parasitaemias and chronic low-level infection following subcurative treatment 25 . Applying the MLE method to the whole transcriptome of the different groups of samples, we observed that a higher proportion of younger parasites were present in acute infection/higher parasitaemias than in low parasitaemias immediately or long after the sub-curative treatment (Fig. 4c and Supplementary Data 4), and that the estimated parasite age in hpi increased as the macaques' parasitaemias decreased (Fig. 4d and Supplementary Data 4). Low parasitaemia is associated with longer circulation time of infected erythrocytes. We then aimed to determine if parasite developmental stages in circulation correlated with parasitaemia. Some but not all studies included in the current analyses originally reported individual parasite densities (in parasites/μL of blood), which allowed us to seek associations between parasitaemia and the predicted proportions of non-ring stages according to the mixture model by Tonkin-Hill et al. 17 as a proxy for parasite development, or whenever possible, the calculated parasite age in hpi through the MLE algorithm by Lemieux et al. 16 . In accordance with our hypothesis of longer circulation of infected erythrocytes with more mature parasites in low parasitaemias, we found significant negative correlations between the proportion of non-rings ( Fig. 5a and Supplementary Data 5) or the predicted age in hpi ( Fig. 5b and Supplementary Data 5) and parasite density of samples in different studies (r = -0.64 and r = -0.59, respectively; p < 0.0001). We were also able to measure the size of infected erythrocytes on thick smears collected at the time of blood draw in two of these studies (Andrade et al. and Milner et al.), and we confirmed that the average parasite size was significantly larger in low than in high parasitaemias (Fig. 5c, d and Supplementary Data 5). Of note, we illustrated with 3D7 P. falciparum parasites grown in vitro after merozoite isolation and synchronized invasion of erythrocytes, that differences in parasite sizes at early hours of development can be detected (Supplementary Fig. 2). Accordingly, despite an obvious overlap close to overall median size within each study, the proportions of parasites falling into the 1st and 4th quartiles were very much linked with parasite burden; with high parasitaemias showing larger proportions of parasites in the 1st quartile and low parasitaemias showing increased proportions of 4th quartile sized parasites ( Fig. 5e and Supplementary Data 5). For these two studies we could also assemble a correlation matrix including six quantified features in our analyses (circulating parasite size, estimated  NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-25062-z ARTICLE parasite age according to Lemieux et al. 16 , median peak expression time of DEGs according to Painter et al. 24 , and proportion of rings estimated by DEGs, and by the mixture model according to Painter et al. 24 and Tonkin-Hill et al. 17 , respectively, and parasite density), and we observed that while the area size of circulating parasites correlated positively with estimated age in hpi and peak expression times of DEGs (all indicative of more developed parasites); it correlated negatively with the proportion of ring stages predicted by Painter categorizations of DEGs or by the full transcriptome mixture model method, and it also correlated negatively with parasite density (Fig. 5f and Supplementary Data 5). Finally, we aimed to cover a comprehensive range of parasite densities and investigate a possible correlation with size of circulating parasites. We compiled the measurements of parasites from the two studies mentioned above, Milner et al. comparing cerebral cases with high and low parasitaemias, and Andrade et al. comparing asymptomatic vs mild malaria cases (Fig. 5c, d), and we added measurements from samples from a third study (Coulibaly et al. in preparation, see Methods) comparing severe vs mild malaria. The correlation between parasite densities found in the blood of children and the parasite sizes across the studies was statistically significant and negative (r = -0.726, p < 0.0001), and particularly evident at low parasitaemias where circulating parasites are larger ( Fig. 5g and Supplementary Data 5).

Discussion
Although the parasite and host factors that determine the clinical outcome of P. falciparum infections are not fully understood, higher parasitaemia is frequently associated with poorer prognoses 6,15,16,18,30,31 . In this study, we accessed publicly available transcriptional profiles of parasites from reports across the malaria clinical spectrum and parasitaemia range. We showed a clear relation between circulation of older parasites within each 48 h asexual replication cycle and lower parasitaemia. Our data support a model in which higher parasite burdens result from parasites sequestering from peripheral circulation by competently binding to endothelial cells and thus evading splenic clearance, contributing to more rapid parasitaemia growth and worse clinical presentations. Asymptomatically infected children typically present with low-level parasitaemias, which are often submicroscopic 32,33 , mild, or uncomplicated malaria cases typically associated with higher parasite burdens than the ones observed in asymptomatic infected children 18,30 , but lower than what is frequently seen in severe or complicated malaria cases 31 . This gradient led to the hypothesis that intrinsic replication rates of P. falciparum may contribute to the level of parasitaemia. However, in vitro culture of parasites collected from uncomplicated and severe malaria cases fails to consistently explain parasitaemia differences found in clinical settings 34,35 , and our recently published data show similar in vitro replication ability between parasites causing mild malaria and asymptomatic infection 18 . Furthermore, we recently reported transcriptional differences between parasites causing mild and subclinical malaria associated with increased circulation time of parasites causing subclinical malaria 18 , which we confirmed here (Fig. 2). Cumulative parasite exposure and the acquisition of humoral immunity are surely at the center of clinical protection in malaria-endemic areas, but the factors that determine whether similarly exposed individuals present with symptoms or not, or which parasites will increase to disease-causing levels in each individual are unknown. We propose that the binding ability to vascular endothelium of infected erythrocytes that are not cleared by antibodies affects the rate of increase of parasite load and hence malaria severity, which is supported by the ordered acquisition of antibodies recognizing PfEMP1 on the surface of iRBCs with earlier responses observed to pathogenic domain variants with particular binding phenotypes 36,37 . In agreement with this concept, our present data reanalyzing multiple published studies comparing parasite transcription of higher vs lower parasitaemias, or more vs less severe malaria-causing parasites identified transcriptional profiles of younger parasites in circulation in higher parasitaemias and more severe malaria cases than in lower parasitaemias or more mild malaria cases (Figs. 2-4). A simple overlay of DEGs lists could not show extensive parallels between the studies, and affected molecular pathways showed only moderate similarities (Fig. 1a-c), which may be in part be explained by the many differences in methodology and analytical pathways and overlapping parasitaemias between groups of samples in some studies (Table 1). However, using analytical approaches described earlier to infer the stage composition and parasite developmental age from DEGs 20,23,24 , and from whole transcriptome 16,17 , we showed within each study comparing parasite of lower vs higher parasitaemias that we could detect more mature parasites in the lower parasitaemia samples and less mature signatures in the higher parasitaemia samples. Transcripts upregulated in higher parasitaemias tended to be early-expressed genes in the~48 h IDC, while upregulated transcripts in lower parasitaemias were laterexpressed genes within the asexual cycle; across the whole transcriptome through algorithms previously established, more developed parasites were consistently predicted in lower vs higher parasitaemias (Figs. 2 and 4). A study not involving parasite burden or clinical severity in the original parasite transcriptome analysis, but instead comparing P. falciparum from higher and lower transmission settings across all parasitaemias 28 , did not reveal any signature related to parasite circulation time when we applied our analytical approach (Fig. 3c, d). None of the human parasite studies available for this reanalysis reported longitudinal data of P. falciparum infections, which would help determine if the progression of infection and consequent immunity would lead to a signature of more developed circulating parasites. However, a non-human primate malaria study analyzing transcription of P. coatneyi during acute malaria, and following sub-curative antimalarial drug treatment that allowed immunity to control infection to low parasitaemias 25 provided us longitudinal data of high parasitaemia during symptomatic malaria and low parasitaemias in the absence of symptoms. We found that the orthologous transcriptional signature of the higher symptomatic parasite burden seen in acute malaria aligned with less-developed, younger parasites forms than the low parasite burden following sub-curative treatment (Fig. 4c, d).
In two studies included in this reanalysis, data supporting further developed parasites circulating in lower parasitaemias and less severe malaria cases were already reported at that time, albeit possibly not clearly or knowingly. Tonkin-Hill et al. described a bias toward early trophozoite transcription in the uncomplicated compared to the ring-stage transcriptional profile of the severe malaria cases analyzed ex vivo 17 , and Lemieux et al. reported that a shorter period of hours in culture was needed to achieve schizont stages in parasites collected from uncomplicated cases (24 h 95% CI 23, 44) than the period needed to have parasites reach the same stage from severe malaria cases (38 h 95% CI 24, 48) (p = 0.05 t test) 16 . These are two clues in two different publications, of what we now put in evidence across multiple studies, and is possibly a major driver of P. falciparum ability to grow within a host. Early stage adhesion of infected erythrocytes results in effective evasion of splenic clearance and a subsequent rapid increase in parasitaemia and severity of disease. With two studies 15,18 we could relate the parasite transcriptional signatures with size measurements of circulating parasites within the lower and higher parasitaemia groups, which confirmed the presence of more mature parasites in individuals with lower parasitaemias (Fig. 5d), validating longer circulation within the~48 h IDC detectable through transcriptomic analyses leads to less rapid increase in parasite burden. Adding a third study (Coulibaly et al. in preparation) we were able to cover a wide range of parasite densities across an array of clinical presentations and quantify parasite sizes of freshly isolated parasites from multiple studies evidencing more developed parasites in samples from lower parasitaemias and milder malaria forms, and showing a negative association between parasitaemia and developmental stage of circulating parasites (Fig. 5g). It remains to be explained how P. falciparum alters its adhesive properties driving increased splenic clearance and maintaining lower parasitaemias. We propose that circulating antibodies control parasites with stronger binding abilities and thus potential fast-growth, leading to the biased presence in the circulation of immune or semi-immune individuals of parasites binding less efficiently. These parasites would not be able to increase to high levels because their longer presence in circulation promotes splenic clearance. Although very difficult due to the variable nature of the parasite gene families promoting cytoadhesion, the study of sequential expression and humoral responses to variant surface antigens in P. falciparum may inform how less-adhesive parasites appear gradually in individuals, and how virulence is regulated favoring persistence. Within the ten studies reanalyzed here, only Andrade et al. and Tonkin-Hill et al. discussed possible associations between disease severity and variant gene family transcripts, but none of the studies included longitudinal sampling. In the small number of samples analyzed in Andrade et al., there was no significant enrichment of particular vars, and only a trend of higher expressed var genes in individuals with clinical malaria vs asymptomatic in the dry season 18 . Tonkin-Hill et al. detected no differences between severe and uncomplicated malaria cases in total number of var gene reads, but identified segregation at the multidomain and individual domain level between severe and uncomplicated disease. However, Tonkin-Hill et al. also found genes involved in PfEMP1 transport and regulation to be downregulated in severe malaria leading the authors to suggest that var gene expression was reduced in severe cases 17 , which in light of our current data could also simply indicate less-developed parasites.
Although we favor a hypothesis where lower parasitaemias and milder malaria cases are promoted by decreased cytoadhesion of longer circulating parasites, achieved through imposed switching following humoral immunity against better-binding parasite variants, we cannot exclude alternative scenarios. Other mechanisms affecting circulation of infected erythrocytes linked to decreased expression levels of PfEMP1 18,30 , the potential effect of the host febrile temperatures 38,39 and cytokine environment [40][41][42][43][44] , other host cues independent of humoral immunity 45 and promoted by different clinical presentations may be at place; or even the possibility that factors independent and parallel to longer circulating parasites should be investigated.
In conclusion, we revealed through transcriptional signatures that parasite circulation time associates with growth potential and parasite virulence, influencing disease outcome, and potentially highlight cytoadhesion dynamics as a major force driving the clinical prognosis of malaria.

Methods
Description of studies and transcriptional datasets included. The studies included in the analyses were selected through a PubMed search using "Plasmodium" "transcription" "field samples," and similar terms, and we further identified studies cited in the search results. Ten studies were included in the transcriptional analyses.
Andrade et al. 2020: 12 P. falciparum RDT + samples from asymptomatic children in the dry season, and 12 malaria case samples from the wet season in Mali. DEGs were reported.
Milner et al. 2012: 58 P. falciparum cerebral malaria samples clustered into low (24) and high (34)  An 11th study was included to investigate the relation of parasite sizes on thick blood smear and parasite density, which was not included in any gene expression analyses.
Coulibaly et al. in preparation: a case-control study of severe malaria in Bandiagara, Bamako, Sikasso, and satellites villages with 6-month to 10-year-old participants enrolled and followed from October 2014 to December 2018 in Mali. Cases were recruited among children hospitalized or seeking care with cerebral malaria (Blantyre score ≤2) or severe anaemia (hemoglobin level ≤5 g/dl) at the dedicated health facilities. Controls were children suffering from uncomplicated malaria seeking care at the same health facilities and matched by age class, residence, sex and ethnicity to the index case. The study protocol obtained ethical clearance from the Ethics Committee of Faculty of Pharmacy and Faculty of Medicine and Odonto-stomatology, University of Sciences, Techniques and Technologies of Bamako, Mali; letter of approval #2014//97/CE/FMPOS. Individual informed consent was obtained from parents or guardians. Data were anonymized to guarantee confidentiality of volunteers' identities.  48 to only use reads that uniquely mapped to P. falciparum. FeatureCount (-g ID --primary -C -Q 30 -p -t gene) 49 54,55 were excluded from analysis. Genes of all studies were filtered by more than 2 counts per million in ten or more samples.
Shared reported DEGs between different studies were assessed and highlighted through the online platform "circos table viewer" (http://mkweb.bcgsc.ca/ tableviewer/visualize/) 56 , and the Venn diagrams obtained through the platform from Bioinformatics and Evolutionary Genomics at Ghent University (http:// bioinformatics.psb.ugent.be/webtools/Venn/) available online. each study subject were ordered by increasing parasite density (parasite/μL) (Andrade et al., Milner et al., and Lee et al.); by %Pf tags (Yamagishi et al.); or by the expression of the P. falciparum housekeeping gene glycine-tRNA ligase (PF3D7_1420400) (Tonkin-Hill et al.). Within each GO group genes were ordered by their peak time of transcription in vitro according to Painter et al. 24 .
DEGs from studies expression in P. falciparum HB3 in vitro: heatmaps and expression curves. For transcripts found up-or downregulated in each study comparison, we plotted the relative expression in P. falciparum HB3 parasite line in vitro along the 48 h IDC defined by Bozdech et al. 20 , which reported microarray gene expression for 3719 P. falciparum genes over the 48 timepoints of the IDC. Heatmaps were generated with online application Morpheus (https://software. broadinstitute.org/morpheus/). Expression curves show the median ± IQR of the log2 expression values along the 48 h IDC.
DEGs from different studies expression in P. falciparum 3D7 in vitro: peak of transcription and developmental stage categorization. For transcripts found up-or downregulated in each study comparison, we plotted the hours post invasion at which transcription peaked in P. falciparum 3D7 parasite line in vitro along the IDC defined by Painter et al. 24 with a Lomb-Scargle periodogram approach as previously described 58 , to define the peak time of total abundance for 5428 genes over the 48 timepoints throughout the 48 h IDC.
Painter et al. 24 further used the expression data of P. falciparum 3D7 parasite line in vitro along the IDC to bin transcripts based on their peak times of transcription, stabilization, and decay into groups representing six developmental stage: early ring (0-10 hpi), mid-ring (11-15 hpi), late ring/early trophozoite (16-21 hpi), mid-trophozoite (22-26 hpi), late trophozoite (27-32 hpi), and schizont (33-48 hpi). We used this categorization to show the proportion of DEGs found up-and downregulated in each study comparison.
Whole transcriptome on a mixture model of developmental stage categorization. The mixture model described by Tonkin-Hill et al. 17 , which used RNAseq reference data of López-Barragán et al. 23 , was applied to the different studies. In brief, proportions of parasite life cycle stages were estimated in a constrained linear model by fitting to transcriptomic profiles of known stages. Data with old gene names were converted and any duplicated genes were collapsed by average expression value. Genes with less than 2 counts per million in ten or more samples were excluded.
Whole transcriptome and maximum likelihood estimation of parasite hours post invasion. Described by Lemieux et al. 16 with microarray reference data of Bozdech et al. 20 , the maximum likelihood method was applied as to the different studies transcriptome. In brief, hpi was estimated by the peak in probability of parasite age given by hourly gene expression data of the HB3 strain. Data with new gene names were converted according to the reference set and any duplicated genes were collapsed by average expression value.
Measurement of parasite sizes and parasite density correlation. P. falciparum parasites were measured on Giemsa-or Field-stained thick blood films made at the time of the blood draw for RNA extraction and/or parasite density quantification of samples from Andrade et al., Milner et al., and Coulibaly et al. in preparation. Slides were imaged to obtain 3-20 photomicrographs blindly acquired from each case with a 63× or 100× objective. Fiji software Image J was used to independently quantify the parasite areas by three to four researchers. The average measure of the interclass correlation for Malian and Malawian samples was 0.963 and 0.889, respectively. Conversion from 63× sizes to 100× was obtained using the πr2 formula. Parasite densities were personally communicated.
Synchronized 3D7 P. falciparum culture. 3D7 P. falciparum parasites synchronized to schizont stage were added E64 compound (Sigma-Aldrich) to prevent egress of merozoites for 6-8 h, and later merozoites were purified through filtration and cultured with non-infected RBCs and RPMI medium supplemented with Albumax-II for up to 16 h.
Statistical analysis. For continuous variables two-tailed Mann-Whitney or Kruskal-Wallis was used to test for differences between two or more groups respectively. Fisher's exact test was used on absolute DEG counts assigned to different developmental stages according to Painter et al. One-way ANOVA with Sidak multiple comparisons test was used to compare proportion of similar stages between groups of each study according to Tonkin-Hill et al.'s mixture model. Pearson correlation or Spearman's rank correlations were obtained accordingly. Statistical significance was defined as a two-tailed p value of ≤0.05. All analyses were performed with GraphPad Prism software version 8.0 or 9.0,or R (http://www.R-project.org).