Cerebral malaria is associated with differential cytoadherence to brain endothelial cells

Introduction

Despite the significant reductions in mortality and morbidity of malaria in the last decade, the percentage of patients infected with Plasmodium falciparum that succumb to severe malaria (SM) is not changing (WHO, 2017), with cerebral malaria (CM) contributing to much of the mortality. The overall mortality rate for CM in children is 15–25%, with a recent MRI study showing that brain swelling is strongly associated with fatal outcome in CM (Seydel et al, 2015). The pathology of CM has been studied extensively (Idro et al, 2005; Hawkes et al, 2013) but also debated for many decades, as discussed in numerous reviews (Shikani et al, 2012; Cunnington et al, 2013; Storm & Craig, 2014; Wassmer & Grau, 2017). What is clear is that the pathogenesis is multifactorial, with a role for the immune response to the Plasmodium infection (Hunt & Grau, 2003; Ioannidis et al, 2014; Dieye et al, 2016; Mandala et al, 2017; Wolf et al, 2017) and obstruction of the microvasculature by sequestration and rosetting (Rowe et al, 2009; Craig et al, 2012; Ponsford et al, 2012; White et al, 2013; Milner et al, 2015), leading to endothelial dysfunction. Sequestration of P. falciparum‐infected erythrocytes (IE) in brain microvasculature is a hallmark of human CM as shown in post‐mortem studies (Pongponratn et al, 1991; Taylor et al, 2004), but whether this sequestration is due to differential binding of IE to brain endothelium has been harder to demonstrate.

The major mediator of parasite cytoadherence to endothelium is P. falciparum erythrocyte membrane protein 1 (PfEMP1), a variant surface antigen expressed on knobs on the IE surface and encoded by approximately 60 var genes per parasite genome, with only one PfEMP1 being expressed on the surface of any individual IE (Scherf et al, 2008; Pasternak & Dzikowski, 2009). PfEMP1 is composed of multiple Duffy binding‐like (DBL) and cysteine‐rich interdomain region (CIDR) domains and can be classified into four main groups A, B, C and E based on the 5′ upstream sequence of the encoding var gene (Fig 1; Smith, 2014). PfEMP1 binds to a range of receptors and includes the mutually exclusive CD36 and endothelial protein C receptor (EPCR)‐binding phenotypes, mediated by N‐terminal CIDR domains (Kraemer & Smith, 2006; Semblat et al, 2006, 2015; Rask et al, 2010; Hviid & Jensen, 2015). Approximately half of group A PfEMP1 and a subset of group B/A chimeric PfEMP1, also known as domain cassette 8 (DC8), bind to EPCR via CIDRα1 domains, whereas group B and C PfEMP1 bind CD36 via CIDRα2‐CIDRα6 domains. In addition, binding to intercellular adhesion molecule 1 (ICAM‐1) is mediated via DBLβ domains adjacent to the CIDR domains and in some cases has been associated with a dual‐binding phenotype with EPCR (Lennartz et al, 2017).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. PfEMP1 domain structure

A schematic presentation of PfEMP1 domain structure comprising a N‐terminal head structure, 2‐6 subsequent C‐terminal domains, a transmembrane domain (TM) and an intracellular acidic terminal segment (ATS) with known receptors indicated in bold. Receptor specificity is determined by combined DBL and CIDR domains with mutually exclusive binding to EPCR and CD36 by different CIDRα domains in the head structure. Part of group A PfEMP1 and a particular subset of group B/A chimeric PfEMP1 (DC8) bind to EPCR via CIDRα1 domains, whereas group B and C PfEMP1 bind CD36 via CIDRα2‐6 domains. The atypical group E VAR2CSA PfEMP1 binds placental chondroitin sulphate A (CSA) via DBLpam1 and DBLpam2 domains. The binding phenotype of VAR 1, VAR3 and group A CIDRβ/γ/δ domains is unknown, but they do not bind EPCR or CD36. DBLβ domains can be involved in ICAM‐1 binding and are from both groups A and B. Not much is known about the other DBL domains (γ/δ/ε/ζ), but the DBLε and DBLζ domains are implicated in IgM and α₂‐macroglobulin binding.

In choosing which host receptors to study, we took into account the findings that categories of PfEMP1 types are also associated with in vivo expression in SM. A particularly strong example of this is where parasites expressing var genes encoding PfEMP1 containing EPCR‐binding domains have shown a strong association with the development of SM, including CM (Avril et al, 2012; Claessens et al, 2012; Lavstsen et al, 2012; Bengtsson et al, 2013; Bertin et al, 2013; Jespersen et al, 2016; Kessler et al, 2017; Mkumbaye et al, 2017). In vitro, parasites expressing EPCR‐binding PfEMP1 show greater degrees of binding to EPCR, as well as to ICAM‐1 receptors, both of which are expressed on brain microvascular endothelium (Turner et al, 2013; Avril et al, 2016; Lennartz et al, 2017). ICAM‐1 binding has been mapped to some, but not all, DBLβ domains found adjacent to the N‐terminal CIDR domains in about one‐third of all PfEMP1. A subset of ICAM‐1‐binding DBLβ domains were recently shown to be specific for group A EPCR‐binding PfEMP1 and found to be expressed at higher levels in parasites from CM patients than in parasites from non‐CM patients (Lennartz et al, 2017). Parasites expressing CD36‐binding PfEMP1 are found in many patient isolates regardless of symptoms, although some data suggest that they may constitute a smaller proportion of parasites in SM patients (Heddini et al, 2001; Ndam et al, 2017), and are not seen in parasite isolates taken from women with placental malaria (Smith et al, 2013). Several other host receptors for PfEMP1 have been described, however, while PfEMP1 proteins that bind these receptors have been identified (Berger et al, 2013), links between cytoadherence and paediatric CM have not been established, and these were not tested in our study.

Infected erythrocytes binding to specific receptors during an infection may have different functional consequences on the endothelium and hence on disease severity. One clear example is that by binding to EPCR, the IE interfere with production of activated protein C thereby launching the coagulation cascade, leading to increased thrombin production (Moxon et al, 2013) and the potential to cause pro‐inflammatory PAR1‐mediated endothelial activation (Petersen et al, 2015; Gillrie et al, 2016). Other PfEMP1‐receptor interactions have been shown to activate signalling pathways in endothelial cells (Wu et al, 2011; Gillrie & Ho, 2017), but the effect of these events on pathology is unclear. More recent work has also suggested that as well as cytoadherence‐mediated events, the accumulation of sequestered IE in vessels may facilitate endothelial dysfunction caused by the local release of soluble mediators following schizont rupture (Gallego‐Delgado & Rodriguez, 2017).

Thus, it remains unclear why a particular child develops CM at a particular time, as the vast majority of P. falciparum infections do not lead to CM. In addition, most African children who develop CM have had malaria previously without developing CM. One possible mechanism to explain why a child develops CM at a particular time is that they have been infected with a P. falciparum variant that facilitates recruitment of IE to endothelium in the brain. While multiple lines of evidence indicate that specific PfEMP1 variants are associated with severe malaria, that association has not been substantiated by directly measuring the binding of IE to endothelial cells (EC). Thus, the question as to whether IE from children with CM have cytoadherence properties that enable them to bind to brain endothelium and thus enhancing their sequestration in that site has not been tested. The extent of sequestration in the brain is unknown for non‐CM cases, although post‐mortem observations of brain vessels from malaria‐infected children dying from other causes of coma (not CM) show much lower levels of IE sequestration than CM cases (Milner et al, 2015). Therefore, as a comparison, isolates from children with uncomplicated malaria (UM) have also been tested for their binding phenotype in the present study.

A number of studies have investigated cytoadherence of specific PfEMP1 variants to human microvascular endothelial cells, but these have been with laboratory strains or PfEMP1‐modified parasites (Madkhali et al, 2014; Gillrie et al, 2015). Patient isolates have also been investigated for their binding phenotype, but mainly on purified protein and mostly under static conditions (Craig et al, 2012; Almelli et al, 2014; Mahamar et al, 2017; Ndam et al, 2017). While providing important evidence, these studies were unable, largely for technical reasons, to combine the most appropriate target (primary brain endothelium) and parasite isolates as close to the patient sample as possible, with a physiologically relevant assay. To address our hypothesis that CM is driven by increased binding of IE to brain endothelium, we assessed whether IE freshly isolated from circulating blood of children with CM preferentially bound TNF‐activated primary brain microvascular endothelium, compared to IE isolated from UM children. We postulated that such a difference might be associated with the expression of particular PfEMP1 variants and with binding to specific endothelial receptors. We collected IE from carefully characterised paediatric CM and UM cases in Malawi and determined cytoadherence to primary human microvascular endothelial cells, with minimal in vitro expansion of the parasite population, using a microfluidic flow device, an experimental design reflecting in vivo physiology. Expression of PfEMP1 variants was investigated by qPCR using the most up‐to‐date set of var domain type‐specific primers available to us.

To our knowledge, this study is the first study to employ such a comprehensive approach to address the question of whether cytoadherence is involved in the pathogenesis of CM.

Results

Cytoadherence of clinical isolates to microvascular endothelial cells under flow

Isolated IE were cultured until the parasites were at the trophozoite stage, when PfEMP1 is expressed on the surface of the IE, and a suspension of 2% parasitaemia and 5% haematocrit was prepared. Using the microfluidic device, cytoadherence to primary human microvascular endothelial cells, derived from brain (HBMEC) and dermis (HDMEC), was determined under flow conditions. Isolates from CM cases demonstrated an average binding of 110 IE/mm² (95% CI: 37–182) to HBMEC which was significantly higher (P = 0.041) than HBMEC binding of UM cases at 43 IE/mm² (95% CI: 28–57; Fig 2). In contrast, there was no difference in binding to HDMEC (P = 0.171) between IE from CM cases (average 165 IE/mm², 95% CI: 81–250) and UM cases (average 110 IE/mm², 95% CI: 71–149). Binding of UM isolates to HBMEC was significantly lower compared to HDMEC (P = 0.002), which was not the case for CM isolates. For isolates from CM patients, avid binding was a common feature; isolates that bound well to HBMEC also bound well to HDMEC with a Spearman's correlation coefficient of 0.83 (P < 0.0001). For UM isolates, however, there was no correlation between binding to HBMEC and HDMEC (r = 0.20, P = 0.28). A recent publication by Azasi et al (2018b) showed that DC8‐PfEMP1 expressing IE do not bind EPCR in the presence of normal human serum. Therefore, we tested whether adding 10% human serum to the binding buffer would decrease the cytoadherence of selected patient isolates to HBMEC (Appendix Fig S2). Human serum did not change the binding of three patient isolates that showed significant EPCR binding to HBMEC. The binding of DC8 variant IT4var19 was also not affected by the addition of human serum in our flow assay system.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Cytoadherence of IE from CM and UM cases to HBMEC and HDMEC

IE were isolated, and binding to HBMEC and HDMEC was determined under flow conditions. Number of IE bound per mm² EC surface was calculated and shown is the mean ± 95% CI for 26 CM and 33 UM cases on HBMEC and 21 CM and 35 UM cases on HDMEC on a log scale. P‐value was calculated by two‐tailed unpaired t‐test. The dotted line is 20 IE/mm², the cut‐off value for inclusion of the inhibition data.

Inhibition of cytoadherence to microvascular endothelial cells under flow

To assess the differential role of the endothelial receptors ICAM‐1, EPCR and CD36, binding was determined in the presence of inhibitory antibodies, αICAM‐1 and αCD36, or recombinant protein, rEPCR (Fig 3). Paired analysis of the inhibition binding data is shown in Fig 3A and significant inhibition was observed for all the EC‐inhibitor combinations, except for inhibition of binding of UM isolates to HDMEC by rEPCR. The data are summarised as percentage inhibition in Fig 3B–D, to compare receptor‐dependent adherence between CM and UM. Approximately half of the IE displayed ICAM‐1‐dependent binding (> 50% inhibition) to both HBMEC and HDMEC, but there was no significant difference between CM and UM isolates nor between the dependency of ICAM‐1 binding to HBMEC and HDMEC. CD36 expression is extremely low on primary HBMEC (Avril et al, 2016), so studies on binding to CD36 were only performed with HDMEC. Inhibition of cytoadherence by αCD36 antibody was variable and not significantly different between the CM and UM isolates (P = 0.23), although there was a trend for higher CD36‐dependent binding in UM isolates. EPCR‐dependent binding also varied, with a subset of isolates binding particularly well to EPCR. This was more pronounced for CM isolates binding to HBMEC, but not significantly different from UM isolates (P = 0.073). There was also no significant difference between rEPCR inhibition of binding to HBMEC and HDMEC. For a few isolates, there was more binding in the presence of αICAM‐1 antibody or rEPCR compared to binding in the absence of inhibitor (Fig 3B and D), and this was more often the case for UM isolates. The reason for this is unclear; we were unable to collect the bound IE to investigate this phenomenon further.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3. Inhibition of cytoadherence of IE from CM and UM cases to HDMEC and HBMEC by αICAM‐1 and αCD36 antibody and rEPCR

• A. IE were isolated, and binding to HBMEC and HDMEC was determined under flow conditions in the absence and presence of 5 μg/ml αICAM‐1 or αCD36 antibody or 50 μg/ml rEPCR. Number of IE bound per mm² EC surface was determined.

• B. Using the same data, percentage inhibition by αICAM‐1 antibody was calculated relatively to binding in the absence of antibody.

• C. Using the same data, percentage inhibition by rEPCR was calculated relatively to binding in the absence of inhibitor.

• D. Using the same data, percentage inhibition by αCD36 antibody was calculated relatively to binding in the absence of antibody.

Data information: (A) shown is the paired analysis between the absence and presence of inhibitor, with number of cases (n) indicated. Statistical significance was determined by two‐tailed paired t‐test, and the P‐value is shown; ns is not significant. (B–D) shown are the mean ± 95% CI, and no significant differences were determined with a two‐tailed unpaired t‐test. Each assay was only conducted once for each isolate. The number of isolates tested can be seen from the dot plot.

Discussion

Post‐mortem studies have shown that IE sequestration in the brain is a feature of CM (Taylor et al, 2004; Ponsford et al, 2012) and have shown a strong association between IE sequestration and key pathogenic processes occurring in CM: endothelial activation/inflammation (Turner et al, 1994), microvascular thrombosis (Dorovini‐Zis et al, 2011; Moxon et al, 2013, 2015) and endothelial barrier disruption (Dorovini‐Zis et al, 2011). Children who died of CM also have sequestered IE in the retinal microvessels, which correlates with the amount of sequestration in the brain (Barrera et al, 2015). To assess the role of cytoadherence in CM pathology, some studies have found correlations between the in vitro adherent properties of patient isolates with specific host receptors (Newbold et al, 1997; Rogerson et al, 1999; Heddini et al, 2001; Mayor et al, 2011; Ochola et al, 2011; Almelli et al, 2014), and the importance of specific PfEMP1 variants has been identified using parasite isolates from patients with different disease syndromes. Based on the hypothesis of cytoadherence to brain endothelium being involved in CM pathogenesis and following a process involving selecting and panning IE on HBMEC (Claessens et al, 2012; Lavstsen et al, 2012; Turner et al, 2013), EPCR was identified as the main receptor of PfEMP1 DC8 and DC13 variants responsible for HBMEC binding. It is worth noting that despite extensive evidence of the involvement of these PfEMP1 variants in SM by qPCR genotyping, a recent publication has questioned whether EPCR acts as a primary adhesion receptor to endothelium for some of these PfEMP1 variants (Azasi et al, 2018b). Taken together, these data, showing that severe malaria is associated with parasite variants with particular cytoadherence characteristics, provide strong evidence that sequestration may be a critical process in the pathogenesis of CM.

To look more directly at the role of cytoadherence in disease, we assessed whether parasite isolates from well‐defined clinical cases of CM, refined by observation of malaria retinopathy (Maccormick et al, 2014), bind human brain microvascular endothelium under physiological flow conditions more readily than do those from UM controls. Our main finding is that parasite isolates from CM and UM patients exhibit differential binding capacities to primary brain endothelial cells, in particular that UM isolates bind less well to HBMEC than CM isolates. For CM patient isolates, binding intensity also correlated with peripheral parasitaemia and the degree of thrombocytopenia. We examined whether this cytoadherence phenotype is determined by the expression of specific PfEMP1 variants by performing var gene analysis by qPCR. Higher transcript levels of ICAM‐1‐ and EPCR‐binding var gene domains were detected in CM isolates, which for the EPCR‐binding var domains seems to correlate with EPCR‐dependent binding to HBMEC. We also observed potential associations between cytoadherence and the transcript levels of the DBLζ and DBLε domains, indicating new roles for these domains in receptor‐mediated cytoadherence, but focussed studies on these are needed to confirm our initial observations.

Cerebral malaria isolates adhered equally well to HBMEC and HDMEC, and binding levels were highly correlated, in agreement with earlier work for laboratory strains that were selected for binding to HBMEC and which bound well to other types of non‐brain microvascular EC (Claessens et al, 2012; Avril et al, 2013). This is also reflected by comparable levels of IE sequestration across many organs in the autopsy studies of children with CM (Seydel et al, 2006; Milner et al, 2014). The UM isolates bind well to HDMEC too, but less avidly to HBMEC, and there is no correlation between their binding to HBMEC and HDMEC. The overall significantly higher binding of CM isolates to HBMEC supports the role of intracerebral cytoadherence and sequestration of infected erythrocytes in CM. The reduced binding of UM isolates to HBMEC compared to binding to HDMEC would result in the majority of IE binding in locations other than the brain and thus avoiding CM pathology and high mortality risk. We hypothesise that UM isolates, which are thought to bind to CD36, do not adhere well to HBMEC because brain endothelial cells constitutively express little CD36 (Avril et al, 2016). The binding of patient isolates to purified receptors, including CD36, has been investigated in many studies, but mostly under static conditions, and the results are conflicting (Rogerson et al, 1999; Yipp et al, 2000; Heddini et al, 2001; Mayor et al, 2011; Ochola et al, 2011; Almelli et al, 2014). In our study, binding in the presence of αCD36 antibodies was only determined for HDMEC and there was no significant difference in CD36‐dependent binding of CM or UM isolates. However, it is notable that there are two populations of UM isolates, with high and low CD36‐dependent binding. There were no differences between the two patient populations in terms of clinical parameters or var gene transcripts with the available primers, but it would be interesting to analyse the var gene usage of the two sub‐populations further.

Previous studies, including ours, suggested independent associations between binding to ICAM‐1 (Ochola et al, 2011) and EPCR (Smith et al, 2013; Turner et al, 2013) with CM. Therefore, we used specific inhibitors of binding to these receptors to measure the dependency of binding via ICAM‐1 and EPCR to endothelium in our assay system. There were no significant differences either in ICAM‐1‐ or in EPCR‐dependent binding between the UM and CM isolates, but there was a trend of higher EPCR‐dependent binding of CM isolates to HBMEC (P = 0.073). The number of observations for binding inhibition by rEPCR was relatively low as we only had access to rEPCR from 2015. Paired analysis showed that there was no significant reduction in binding of UM isolates to HDMEC in the presence of rEPCR, indicating that EPCR does not play a major role in the adhesion of the UM isolates to HDMEC. EPCR binding by IE expressing the DC13 and DC8 variants has recently been questioned with data showing that DC13 expressing IE do not bind EPCR in vitro and binding of DC8 expressing IE to EPCR and brain endothelium was reduced in the presence of 10% human serum (Azasi et al, 2018). This effect of human serum was not seen with three patient isolates and It4var19 in our assay (Appendix Fig S2). This discrepancy may be explained by different assay conditions; Azasi et al determine static binding to the CD31‐negative HBEC‐5i cell line, while we used primary HBMEC under flow conditions.

This highlights a number of limitations in using in vitro methods to investigate the relationship between cytoadherence and disease. Besides cell and assay type, in standardising the in vitro assay parameters at 2% parasitaemia and 5% haematocrit under constant shear stress, we can only mimic in vivo dynamics, for which parasitaemia is variable and haematocrit is much higher at 30–50%. In our assay system, we could not collect the bound IE, hindering further molecular analysis of the adherent IE population. Refined flow adhesion techniques would be needed to overcome these limitations.

The IE that bind from the parasite suspension to HBMEC under flow conditions in our assays are a representation of the parasite types that might be recruited to brain endothelium from peripheral blood flow in vivo. In addition, parasite populations in peripheral blood, the only population readily accessible in patients, have been shown to resemble the same parasite populations as the sequestered IE (Montgomery et al, 2006). To ensure we were investigating brain‐specific binding, we used TNF‐activated primary HBMEC, with HDMEC as a comparator. That we did not find significant differences in receptor usage between the UM and CM isolates could be explained by the infection consisting of multiple P. falciparum genotypes, including those underlying var gene expression (Montgomery et al, 2006, 2007). It is also not as simple as a one‐to‐one interaction with one PfEMP1 variant binding to one specific receptor. Lately, dual receptor binding by one PfEMP1 variant has been reported, with some PfEMP1 being able to bind both EPCR and ICAM or CD36 and ICAM‐1 simultaneously (Avril et al, 2016; Lennartz et al, 2017). We also determined binding of the patient isolates to HBMEC and HDMEC in the presence of αICAM‐1 antibody and rEPCR combined, but the limited data (Appendix Fig S3) showed no significant difference between UM and CM isolates, nor between HBMEC and HDMEC binding. It is clear that HBMEC binding of some isolates is not affected by inhibition with αICAM‐1 and rEPCR, indicating that other receptors may play a role in cytoadherence to brain endothelium, a phenomenon also seen with selected laboratory strains (Yipp et al, 2007; Avril et al, 2016; Mahamar et al, 2017; Metwally et al, 2017).

The domain structure of var genes expressed by the patient isolates was determined with a set of extensive qPCR primers. Transcripts encoding EPCR‐binding domains, both when assessed by individual CIDRα1 domain subsets or summarised in the group B/A and A (CIDRa1.DC8, CIDRα1.A), as well as transcripts encoding the CIDRα1‐associated DBLα domains, were all expressed higher in CM isolates. This was also the case for transcripts encoding group A non‐EPCR‐binding domains and the ICAM‐1‐binding domains DBLβ1/3‐1 (group A) and DBLβ5 (group B), all in line with previous reports (Lavstsen et al, 2012; Bertin et al, 2013; Bernabeu et al, 2016; Kessler et al, 2017; Mkumbaye et al, 2017; Shabani et al, 2017). In our study, transcripts encoding other domains, found C‐terminally in both group A and B PfEMP1, showed higher levels in CM compared to UM cases. This may be attributed to the inclusion of retinopathy in our CM cases, producing more definitive case definition. Two recent studies compared the var gene expression between retinopathy‐positive and retinopathy‐negative children and found a number of differences, concluding that group A var genes are more commonly found in the CM patient population that are retinopathy‐positive (Abdi et al, 2015; Shabani et al, 2017). Higher transcripts of the DBLζ2a, DBLζ2c, DBLζ3, DBLζ5 and DBLε2 domains were detected in the CM cases, of which DBLζ2a, DBLζ3 and DBLε2 were also reported in a recent study (Kessler et al, 2017; Mkumbaye et al, 2017). The functions of the C‐terminal DBLε and DBLζ domains are not fully understood, but some are involved in non‐immune IgM and α₂‐macroglobulin binding (Semblat et al, 2006, 2015; Jeppesen et al, 2015; Stevenson et al, 2015a; Pleass et al, 2016). The variant nature of var genes makes it difficult to design qPCR primers with high enough specificity and selectivity to cover transcripts encoding CD36‐binding CIDR domains (group B and C var genes). For this reason and due to the different specificities and amplification bias of the primers, exact determination of var transcript distribution cannot be determined for each isolate (Mkumbaye et al, 2017).

We attempted to discover whether specific var domains contribute to the binding phenotype of the patient isolates by determining the correlation between IE binding and qPCR data. However, our study was not powered to examine these correlations with sufficient statistical significance after correction for multiple comparisons. The uncorrected results from this analysis, shown in Appendix Table S2, require further studies to identify statistically significant associations.

We identified an association between the expression of EPCR‐binding PfEMP1 and IE cytoadherence to EPCR, but only for the CM isolates and mainly to HBMEC. While positive correlations provide straightforward evidence for associations, negative correlations can also be informative. The negative correlation between ICAM‐1‐dependent EC binding and levels of transcript encoding CIDRα1 domains (EPCR‐binding) in UM isolates suggests that dual EPCR and ICAM‐1 binding is not a prominent phenotype in UM. This indicates that EPCR is a major component of cytoadherence to HBMEC in CM, but not necessarily as the only primary receptor, as suggested by a recent study (Azasi et al, 2018). Recruitment of IE to the EC could also occur via other receptors and then subsequently bind to EPCR causing pathology linked to degradation of the control of coagulation pathways.

To validate the relevance of binding characteristics to pathogenesis, we assessed whether the level of cytoadherence in vitro corresponded with clinical characteristics in the patients from whom the parasites had been isolated. There was a correlation between the avidity of binding of CM isolates and peripheral parasitaemia and platelet counts at the time of admission of the CM cases. The increased binding (at standardised assay conditions of 2% parasitaemia and 5% haematocrit) to EC correlated with increased parasitaemia, but with decreased platelet numbers. The negative correlation between peripheral platelets and binding intensity is interesting as thrombocytopenia has been used as a predictor for malaria (Lampah et al, 2015; Thachil, 2017) and specifically for P. falciparum SM (Gerardin et al, 2002; Cserti‐Gazdewich et al, 2012). A recent study in children recruited from the same hospital in 2015–2016 showed that low platelet levels were also associated with retinopathy‐positive CM cases and brain swelling (Kessler et al, 2017). Some studies have shown that CM cases have more platelets localised in the brain microvasculature (Grau et al, 2003; Dorovini‐Zis et al, 2011) and show increased platelet‐mediated clumping of IE (Pain et al, 2001; Wassmer et al, 2008) and platelet involvement in the adhesion of IE to human microvascular EC lacking CD36 (Wassmer et al, 2004). We postulate that these processes lead to platelet sequestration and consumption, which is augmented by the presence of IE with high binding capabilities, and therefore a decrease in peripheral platelet counts. Activation of platelets and platelet consumption leads to a pro‐coagulant state characterised by an increase in thrombin and von Willebrand factor, which are demonstrated in SM (Dorovini‐Zis et al, 2011; O'Sullivan et al, 2016; Thachil, 2017). In addition, anti‐coagulation and endothelial protective pathways are affected in CM through a decrease in EPCR and thrombomodulin and thus dysfunction of the activated protein C pathway (Moxon et al, 2013, 2015; O'Sullivan et al, 2016).

A number of mechanisms have been proposed to explain the pathology of CM, and it is unlikely that a single molecular process is involved. However, the identification of EPCR as a cytoadherence receptor and its association with CM in several studies, including this one, provides a compelling direction for future research.

Binding of IE to EPCR inhibits EPCR interaction with activated protein C, leading to deregulation of inflammation, coagulation and loss of endothelial barrier integrity (Gillrie et al, 2015; Sampath et al, 2015; Bernabeu et al, 2016; Kessler et al, 2017). This might be one of the reasons why the consequences of sequestration in the brain can be catastrophic. From our data and others, we do not find specific brain‐binding IE in CM as CM isolates bind well to a variety of EC, which is also demonstrated by the sequestration of IE in many other organs (Avril et al, 2013; Milner et al, 2014, 2015). IE sequestration in the brain, irrespective of EPCR binding, leads to loss of EPCR function (Moxon et al, 2013); thus, the combined low levels of EPCR on HBMEC and binding of IE to EPCR lead to reduced levels of activated protein C and subsequent increase in thrombin, impacting on inflammation and coagulation, as reviewed in Bernabeu and Smith (2017). Furthermore, the endothelium responds locally to release of soluble content of sequestered IE leading to the disruption of barrier function (Gallego‐Delgado & Rodriguez, 2017). Thus, cytoadherence of IE to brain endothelium may alter the balance of endothelial activation and protective pathways via a number of mechanisms, all driven by the binding properties of CM isolates to brain endothelium.

This work has shown that CM is associated with increased adhesion of IE to brain endothelium in vitro and that this adhesion is not specific to cerebral vascular endothelium. Indeed, it is the reduced ability of binding to brain EC that characterises isolates from UM patients that strengthens the case for the role of sequestration in the pathology of CM.

Materials and Methods