Host matrix metalloproteinases in cerebral malaria: new kids on the block against blood–brain barrier integrity?

The mechanical hypothesis proposes CM is caused by a mechanical obstruction of the cerebral microvasculature, with coma resulting from impaired brain perfusion [9, 14, 18]. Such a hypothesis was made after one of the first pathological studies on human CM showed that brain capillaries were packed with iRBCs [18]. In the mechanical hypothesis, specific interactions between iRBCs and vascular endothelium are thought to mediate sequestration of iRBCs within the brain resulting in removal from peripheral circulation [19‐21]. The molecules involved in these interactions are parasite proteins expressed on iRBC surface, such as P. falciparum erythrocyte membrane protein-1 (PfEMP-1), and specific host receptors in the microvascular endothelium, including intracellular adhesion molecule-1 (ICAM-1), vascular cellular adhesion molecule-1 (VCAM-1), thrombospondin, CD36, and E-elastin [22‐25].

Cytoadherence and decreased pliability are the main mechanisms underlying vascular obstruction [9, 17‐21]. It is speculated that cytoadherence evolved as a mechanism for the parasite to evade triggering a host immune response and being cleared from the spleen. Cytoadherence is also beneficial for the parasite as to provide an optimal environment of low oxygen tension for parasite growth. Decreased deformability along with increased membrane stiffness and rigidity of iRBCs are due to changes in the cytoskeleton triggered by growing intracellular parasites. Cell deformability has been indicated as a predictor of anemia development [26], whereas cell rigidity correlates with a higher fatality rate [27]. Another phenomenon occurring along with iRBC sequestration is rosetting, characterized by iRBCs forming a flower-like cluster around a non-iRBC, making a tight rigid structure [28]. Rosetting is more frequent in patients with CM than in those with uncomplicated malaria. However, rosette formation has also been reported for other Plasmodium strains (P. vivax and P. ovale) which do not cause CM [29]. Since rosetting occurs in all manifestations of the disease, it is not associated with severity or clinical outcome of CM [30]. One question the mechanical hypothesis by itself does not explain is why most patients recovering from CM do not show any evidence of ischemic brain damage [12].

The permeability hypothesis proposes that BBB damage is the underlying mechanism of CM, such that a leaky BBB allows toxic compounds to enter the brain and cause neurological dysfunction [9, 13‐15, 31]. Several animal CM models have confirmed that the BBB is disrupted and that cerebral edema is present in CM, although this is less evident in humans [15]. Nevertheless, iRBCs remain attached to endothelium, without entering the brain parenchyma [7, 9]. Interestingly, Adams and colleagues have suggested that iRBC cytoadherence might activate secondary signaling events similar to those occurring in leukocytes [32]. These secondary signaling events are thought to cause functional alterations in the BBB, which could allow toxic compounds to pass into the CNS. These events might be reversible, therefore explaining why neurological manifestations are just transient in most cases and why a large number of recovering patients lack neurological sequelae [32].

Enwonwu and colleagues implicated histamine as one of these toxic molecules that enters the brain parenchyma after BBB impairment and contributes to the neurological manifestions of CM [33‐37]. The authors observed altered neural histidine uptake in children with severe falciparum malaria providing an explanation for the enhanced cerebral production of histamine [33]. They also found increased plasma levels of histamine in severe malaria patients, further supporting their hypothesis [34]. Moreover, the involvement of histamine in CM has also recently been confirmed in a murine model [35‐37]. In this study, histidine decarboxylase-deficient mice were unable to synthesize free histamine and did not develop CM after infection with P. berghei ANKA. These mice displayed preserved BBB integrity, were void of iRBC aggregation in the brain vessels, and did not sequester CD4+ and CD8+ T cells [36]. Further investigation of histamine receptors revealed histamine-1-receptor (H1R) and histamine-2-receptor (H2R) are associated with severe malaria development [37], whereas histamine-3-receptor (H3R) has a neuroprotective role [36].

The humoral hypothesis is a natural extension of the permeability hypothesis. This hypothesis suggests that host factors such as leukocyte-derived cytokines and chemokines can enter the brain parenchyma after increased BBB permeability, thus causing CM symptoms such as fever and coma [9, 13, 14, 16, 38‐40].

Effector cells including T cells, NK cells, and monocytes, along with inflammatory responses mediated by cytokines such as tumor necrosis factor-α (TNF-α), limphotoxin-α (LT-α), and interferon-γ (IFN-γ), are proposed to contribute to the development of murine CM [41‐48]. However, the extent of their involvement and molecular mechanisms in human CM is still topic of debate [48, 49].

CD8+ T cells have been reported to initiate BBB tight junction disruption and promote CNS vascular permeability under neuroinflammatory conditions [50‐52]. Consistently, CD8+ T cell sequestration in cerebral microvessels and subsequent brain infiltration have been demonstrated in murine CM [43, 44], where Plasmodium antigens can be cross-presented during infection by dendritic cells (DCs) [53, 54] and brain endothelial cells in association with MHC class I molecules [55]. Recent human studies support the idea that malaria antigens can be transferred to endothelial cells [56]. However, it is currently unknown whether Plasmodium-specific CD8+ T cells are involved in the pathogenesis of human CM [57]. Furthermore, lymphocyte infiltration into brain parenchyma remains to be investigated [49].

TNF-α relevance in CM is also unclear. TNF-α involvement in murine CM was first proposed in 1987 [58]. Since then there have been numerous studies investigating TNF-α levels in CM mice albeit the results are inconsistent. For example, some works confirmed the association of high TNF-α levels with murine CM [59‐61], whereas others argued against such correlation, finding LT-α and IFN-γ as more suitable markers [44, 47, 62, 63]. To reconcile such discrepancies, it has been proposed that low concentrations of TNF-α could enhance parasite killing, whereas higher concentrations might be associated with increased incidence of murine CM [46]. However, data on TNF-α also appear inconclusive in human CM studies. Indeed, clinical studies tend to exclude any association between CM and increased plasma, serum or CSF levels of TNF-α [64‐67], although a few works have proposed a correlation in two different Asian populations [68, 69]. Instead, in some of these studies, high CXCL10/IP-10 plasma levels and low angiogenic factors such as vascular endothelial growth factor (VEGF) and angiopoietin-1 (Ang-1) in children with CM, predicted subsequent mortality [65, 66, 68]. Moreover, a protective role for IL-12 has been proposed in human CM [70, 71].

Among soluble factors involved in CM, a critical role for nitric oxide (NO) has also been suggested. It was hypothesized that NO levels correlate with disease severity, since the sequestration of iRBCs might contribute to CM pathogenesis by causing hypoxia, which is related to enhanced production of cytokine-induced NO, compensatory vasodilatation, and subsequent brain volume increase [39]. However, activation of inducible NO synthase (NOS) might also serve protective functions, since NOS inhibits the side effects of brain indoleamine 2,3-dioxygenase (IDO) and quinolinic acid accumulation [72], although IDO systemic distribution is independent of malaria disease severity [73]. In a study performed on Tanzanian children infected with malaria, the plasma levels of NOS-suppressing IL-10 increased with disease severity, suggesting that a reduced NO production may contribute to CM [74]. Moreover, a genetic single nucleotide polymorphism found in the NOS2 promoter region causes elevated NO production and was significantly associated with protection against CM in Tanzanian and Kenyan children [75]. In line with these observations, Anstey and colleagues demonstrated that decreased NO production was associated with endothelial dysfunction in human CM [76, 77]. Similarly, van der Heyde and his group demonstrated that low NO bioavailability was associated with murine CM [78, 79]. Interestingly, prophylaxis with inhaled NO in CM-sensitive mice significantly reduced systemic inflammation and endothelial activation by lowering TNF-α, IFN-γ, monocyte chemotactic protein-1 (MCP-1), sICAM-1 and von Willebrand factor, and by increasing Ang-1 levels in peripheral blood [80]. The protective effect of exogenous NO on mouse CM appears associated with decreased brain vascular expression of inflammatory markers, resulting in attenuated endothelial junction damage and facilitating blood flow [81]. Lastly, treatment with exogenous L-arginine, the substrate for NOS, recently proved to be safe in a pilot study on CM patients, although effective doses still need to be optimized [82].

Furthermore, during malaria infection both host and parasite undergo strong oxidative stress, which leads to increased production of reactive oxygen species (ROS) and subsequent protein and lipid peroxidation [83, 84]. The coexistence of both parasite and erythrocyte is a matter of a delicate balance: low ROS concentrations seem to inhibit parasite growth, whereas larger amounts may damage vascular endothelial cells and increase vascular permeability [85]. Oxidative stress paradoxically has both a pathogenic and protective role in CM [86]. An anti-oxidant diet was shown to reduce BBB damage and counteract CM development in CM-sensitive mice [87], and anti-oxidant adjuvant therapy, provided at the initial stages of murine CM, prevented the development of persistent cognitive damage [88]. In contrast, NADPH-deficient mice were shown to develop CM despite the lack of ROS production, suggesting that ROS did not contribute to CM pathogenesis [89, 90]. To reconcile such an apparent inconsistency, Linares and colleagues have recently shown that glutathione peroxidase and heme oxygenase-1 up-regulation cooperate to suppress superoxide dismutase, catalase, heat shock protein-70 and thioredoxin-1 down-regulation effects in murine CM, counteracting oxidative damage and maintaining redox equilibrium [91]. In human CM, ROS have been associated with a pathogenic role thus far. In vitro, ROS inhibition was shown to protect brain endothelial cells against P. falciparum-induced apoptosis and to decrease iRBC cytoadherence through ICAM-1 down-regulation and iNOS induction [92, 93]. Consistently, in a recent clinical study performed on fifty Indian children with severe malaria, oxidative stress was associated with disease severity [94].

As discussed below and summarized in Table 1, evidence showing differential phenotypes between neural and non-neural endothelial cells after co-culture with Plasmodium iRBCs comes from several in vitro studies [56, 93, 99‐108].

First, the effects of P. falciparum infection were investigated in a BBB model of cultured primary porcine brain capillary endothelial cells (PBCECs) [99]. In this study, membrane-associated malaria antigens obtained from lysed P. falciparum schizont-iRBCs increased endothelial E-selectin and ICAM-1 expression, reduced the trans-endothelial electrical resistance (TEER), and promoted the disruption of tight junctions, indicative of increased BBB permeability.

Consistently in various types of human brain endothelium, including HMBEC primary cultures and HBEC-5i or hCMEC/D3 cell lines, iRBCs were also shown to increase ICAM-1 expression [93, 99, 104, 105], to reduce TEER [56, 101, 102], to alter tight junction expression and distribution [56, 105], and to enhance BBB permeability to 70-kDa dextran [101]. Interestingly, platelets were suggested to play a key role in iRBC-dependent increase in BBB permeability, releasing microparticles and causing cell apoptosis in TNF-α- and LT-α-activated HBEC-5i [101]. In hCMEC/D3 cells, iRBC-increased cell adhesion and paracellular permeability correlated with ZO-1 disorganization, but the latter effect appeared mediated by parasite-induced metabolic acidosis, independent from cytoadherence [105]. Moreover, differential global gene expression in HBMEC after interacting with iRBCs revealed significantly up-regulated transcripts related to immune and inflammatory responses, apoptosis, cell-cell signaling, signal transduction and nuclear factor-kB (NF-kB)-activation cascade [104]. After co-culturing with iRBCs, the mRNA expression of neural endothelial pro-inflammatory chemokines (IL-6, CXCL-8/IL-8, CXCL-1/GRO-α, CXCL-2/MIP-2α, and CCL-20/MIP-3α) increased more than 100-fold, highlighting the strong inflammatory component and the active role of the endothelium in CM pathogenesis [104]. Furthermore, in TNF-α-activated subcutaneous fat tissue-derived endothelial cells, a model comparable to cerebral endothelium, P. falciparum iRBCs induced several CM-specific effects, including up-regulation of ICAM-1, VCAM-1, and CD61, enhancement of microparticle, MCP-1 and IL-6 release, and higher caspase-3 activation [106]. Increased levels of inflammatory cytokines may have direct systemic effects and adversely affect the clinical outcome by increasing the cytoadherence of infected RBCs to venular endothelium through up-regulation of adhesion molecules, such as ICAM-1 [93].

To assess the specificity of these effects for human cerebral endothelium, additional comparative studies were also performed using non-neural endothelial cells. Interestingly, P. falciparum iRBCs did not affect the expression and distribution of tight junctions (as measured by claudin-5 and ZO-1) and did not induce pro-inflammatory response or cell death in human dermal or lung microvascular endothelium, although parasite sonicates did [103]. Additionally, the up-regulating effects of iRBCs on ICAM-1 expression observed in HBMEC were not reproduced in human umbilical vascular endothelial cells (HUVEC) from healthy donors [93]. An increase in ICAM-1, VCAM-1, and E-selectin mRNA was found in HUVEC from patients with different degrees of malaria (uncomplicated, severe, or CM) after co-culturing with iRBC-fed mononuclear cells, however such increase did not appear specific for CM. On the contrary, reduced mRNA levels of tight junction proteins (occludin, vinculin, and ZO-1) were strictly associated with CM [100].

Genetic differences between Plasmodium strains might also play a role in CM development. Indeed, it has been shown that different strains of P. falciparum display variable degrees of cytoadherence to HBEC-5i [107]. Additionally, P. berghei ANKA, a murine CM-associated Plasmodium strain, induces a higher VCAM-1-mediated cytoadherence compared to P. berghei K173 (non-CM strain) in either brain or lung mouse vascular endothelial cells [108].

Several in vivo animal models have reported alterations in BBB after exposure to Plasmodium parasites or malaria products such as hemozoin (Hz, malarial pigment) [47, 109‐124]. As summarized in Table 2 and described below, these studies provide insightful findings regarding BBB breakdown in animal CM models.

The first animal studies on BBB permeability in malaria date back to 1968, when Migasena and Maegraith demonstrated the movement of albumin across the BBB in Macaca mulatta monkeys infected with P. knowlesi[109‐111]. However, P. knowlesi does not induce CM. As such, the rhesus monkey infected with primate malaria parasites, P. coatneyi and P. fragile, is considered to be a more valid primate model to study in the context of severe malaria with cerebral involvement [112‐114].

Of the four species of rodent malaria parasites (P. berghei, P. yoelii, P. chabaudi, P. vinckei), only a few P. berghei strains can induce experimental CM in mice, with the ANKA strain being the most widely studied. Symptoms of experimental CM in P. berghei ANKA-infected susceptible mice include paralysis, ataxia, head deviation, convulsion and coma [98]. In P. berghei K173-infected mice an excessive movement of water, albumin and other proteins into the brain, as well as severe brain edema, microthrombosis, sludging of mononuclear cells, arteriolar spasms, scattered disturbances of the microcirculation, and occasional proliferation of gliocytes were observed, suggesting a progressive deterioration of BBB integrity culminating in endothelial lesions and haemorrhages [31, 115‐118]. Of note, mouse CM models present neurological signs (ataxia, hemiplegia and coma) similar to the clinical features reported in human CM [119].

In a recent work, Penet and colleagues presented the first in vivo magnetic resonance study of mouse CM, demonstrating BBB breakdown in CM. Multimodal magnetic resonance neuroimaging techniques (imaging, diffusion, perfusion, angiography, spectroscopy) of P. berghei ANKA-infected mice revealed vascular damage, including BBB disruption and haemorrhages, major edema formation, reduced brain perfusion and ischemic metabolic profile, with reduced high-energy phosphates and enhanced brain lactate. These data strongly point to the coexistence of inflammatory response and ischemic lesions [120].

Other recent works illustrated a complex strain-dependent relationship between leukocyte recruitment, BBB permeability and chemokine production. Major pathological consequences of malaria arise from inappropriate or excessive immune response mounted by the host in an attempt to eliminate the parasite. In P. berghei ANKA-infected mice, inflammation of the cerebral microvasculature and leukocyte recruitment were clearly evident and found to be driven by production of pro-inflammatory cytokines (IL-12, IFN-γ) and CM development [121]. On the other hand, P. berghei NK65-infected mice showed enhanced production of LT-α and several chemokines (CXCL-9/MIG, CCL-2/MCP-1, CCL-3/MIP-1α and CCL-5/RANTES), but no neurological symptoms [47, 122]. A complementary study performed on the same model proposed a concurrent role for Transforming Growth Factor-β (TGF-β) and TNF-α in promoting splenocyte apoptosis [123].

It should be noted that the cerebral microvascular tree contains two functionally distinct BBB: i) the physiological BBB, formed by capillaries 4–8 mm in diameter, consisting of a single layer of endothelia, gliovascular membrane, and astrocyte endfeet; and ii) the neuroimmunological BBB, formed by postcapillary venules 10–60 mm in diameter and encompassing two layers - the endothelium with its basement membrane and the glia limitans with associated astrocyte endfeet - separated by the perivascular space [125]. The physiological BBB serves as a tight diffusion barrier for small solutes while the neuroimmunological BBB permits transport of macromolecules and diapedesis of immune cells [125]. In a very recent study comparing different mouse models of experimental CM (P. berghei ANKA infection), human CM-like histopathology (P. yoelii 17XL) and non-CM (P. berghei NK65 and P. yoelii YM), Nacer and colleagues observed that the physiological BBB in the experimental CM model remained intact, whereas regulated fluid transport across the neuroimmunological BBB led to brain swelling, intracranial hypertension, coma, and ultimately death due to dysfunction of respiratory centers in pons and the medulla oblongata as a result of brain stem compression [124]. Thus, they proposed that CM may occur in two steps: 1) induction of coma based on regulated, preventable and reversible opening of the neuroimmunological BBB; and 2) endothelial death-associated haemorrhaging, which is difficult to reverse by treatment and eventually fatal [124]. A similar mechanism for neuroimmunological BBB opening in human CM would explain the reversibility of coma with treatment, the scarce traces of tissue necrosis in surviving patients, and the different neurological outcomes of patients despite similar clinical presentation [6, 8, 9, 13, 102].

BBB functional impairment during human CM has been investigated in several clinical and post mortem studies [126‐143]. Table 3 summarizes the most relevant results. Here, the investigations on human CM patients were performed using albumin CSF/serum ratio as an indicator of BBB integrity [126‐128], by post mortem immunohistochemical analysis [129‐135], or through brain imaging techniques [136‐144]. Interestingly, the BBB seems to be more impaired in children than in adults. Moreover, it appears that African and Asian patients display a different degree of BBB damage, with BBB breakdown being more likely to occur in African than Asian populations.

One of the first studies on Asian patients was conducted in Thailand [126]. In this work, albumin CSF/serum ratios were higher in CM patients than in controls, but it did not correlate with coma and mortality. Thus, the authors concluded that their data did not support the idea that cerebral edema might be the cause of coma. More than a decade later, albumin and Immunoglobulins G plasma/CSF ratios were found to be only mildly impaired in Vietnamese patients, suggesting only minimal degree of BBB breakdown in few CM cases [127]. Therein, human CM appeared to cause only subtle functional changes in BBB integrity, with minimal intra-parenchymal inflammatory response compared with other neurologic infections, such as cryptococcal, tubercular, and acute bacterial meningitis [127].

Regarding African populations, a study on Zairean children showed no difference in CSF albumin compared to controls [128]. However, in Malawian children with CM, the activation of endothelial cells and macrophages, along with the disruption of endothelial intercellular junctions in vessels containing sequestered iRBCs, and subtle but measurable changes in albumin CSF versus albumin serum levels were observed. Nevertheless, negligible leakage of plasma proteins was still apparent [129]. In Kenyan children with CM, protein and amino acid levels in paired plasma and CSF samples were measured, showing that BBB was mildly impaired in some children with severe falciparum malaria [130]. However, this impairment was not confined to CM, as it was also reported in children with prostration-associated malaria and, to a lesser extent, in children with malaria and seizures. Evidence of intrathecal immunoglobulin synthesis in children with malaria was also observed [130]. Finally, data obtained in a recent work performed on Malawian children are consistent with the proposed link between iRBCs sequestration and intravascular/perivascular pathology in fatal pediatric CM, resulting in myelin damage, axonal injury, and BBB breakdown; however, no Hz-laden monocyte extravasation was found [131].

Pathological studies on post mortem samples of CM patients showed cerebral edema and raised intracranial pressure in 50% of West African children [132] but not in South Asian adults [133, 134] or Malawian children [129]. Nevertheless, an important correlation between sequestration of iRBCs in the brain microvessels and the malaria-related encephalopathy was shown in Asian patients [133]. The adhesion of iRBCs to brain microvessels is mediated by specific receptors on the host endothelium, including ICAM-1, CD36 and CD31 [22‐25]. Immunohistochemistry showed altered distribution of the cell junction proteins occludin, vinculin and ZO-1 in Vietnamese adults and Malawian children with CM [129, 135]. Sequestration of iRBCs in cerebral microvessels was significantly higher in the brains of patients with CM compared with non-CM patients in all parts of the brain (cerebrum, cerebellum, and medulla oblongata), and was quantitatively associated with pre mortem coma [129].

In recent years, several imaging studies have been also conducted on the brains of CM patients during disease progress or after recovery [136‐144]. Using magnetic resonance or computed tomography, several common features implicating BBB damage have been observed, including cerebral edema, increased brain volume, ischemia and large vessel infarcts, hemorrhagic cortical lesions, focal and multifocal atrophy, and limited CSF circulation [136‐139, 141‐144]. Interestingly, magnetic resonance in a recently published case-report of a 37-year-old French patient with malaria travelling back from Equatorial Guinea, showed that he developed posterior reversible encephalopathy syndrome, which is characterized by diffuse symmetric signal-intensity abnormalities of white matter in the posterior circulation territory [140]. Since data from previous literature suggest one of the mechanisms of posterior reversible encephalopathy involves capillary leakage and acute disruption of the BBB, the authors concluded that this case-report supports the theory of BBB disruption as a key factor for CM development [140].

Clearly there is much discrepancy on the extent of BBB damage between animal and human models of CM. While several studies performed on mouse CM models suggest a strong BBB breakdown [115‐124], data on increased BBB permeability in human CM are somehow less evident, generally suggesting the occurrence of only mild BBB impairment, characterized by a relevant degree of tight junction disruption, but lacking molecule exchange between serum and CSF [126‐135].

In this context, it should be noted that the relevance of murine CM models for studying CM pathophysiology has been a topic of big debate in the recent years [49]. Being clearly an inflammatory syndrome with local vascular endothelial activation, murine CM displays obvious differences and some similarities to the clinical and pathological features of human CM, such as signs of vascular inflammation/damage [145]. A recurring issue concerns the degree of iRBC sequestration in the brain and other organs of P. berghei ANKA–infected mice. Although recent data find increased iRBC accumulation during murine CM in multiple organs including the brain [146], P. berghei infection is generally acknowledged to promote marked accumulation of leukocytes (particularly monocyte, macrophages and T cells), which is in stark contrast to human CM [147]. Thus, despite several processes shared either by murine or human CM, the changes in the endothelial cell microenvironment induced by cytoadherence and inflammation are not the same [104]. Additionally, mouse studies suggesting associations between high levels of cytokines and CM [58‐61] have been recently challenged by works showing that high levels of pro-inflammatory cytokines such as TNF-α are poor indicators of human CM in African children [64‐67]. Thus, future experimental studies on alternative animal models (non-human primates and other mouse models) are encouraged and urgently necessary to better understand the pathological processes underlying human infection [49].

Another interesting point emerging from clinical data is that the BBB appears more impaired in children than in adults [126‐143]. Since CM often strikes children at a critical time in brain development, Hawkes and colleagues have nicely hypothesized that developmental changes in the cerebral vasculature may account for some of the differences in disease presentation and outcome between children and adults, including mortality, seizures and neurocognitive sequelae, rates of associated anemia and renal dysfunction, retinal vessel changes, frequency of ring haemorrhages, and inflammatory cell accumulation in brain microvessels [148]. To avoid any misunderstanding, it has been pointed out that the widespread belief amongst neurotoxicologists that BBB is immature or even absent in the newborn is contrasted by a substantial body of evidence supporting the concept of well-developed barrier mechanisms in the developing brain [149, 150]. For instance, inter-cellular tight junctions between cerebral endothelial cells and between choroid plexus epithelial cells are functionally effective as soon as they differentiate [150]. Nevertheless, the brain develops within an environment that is different from that of the rest of the body, and the developing brain possesses a number of unique features not generally present in the adult [149]. Interestingly, certain genes coding for influx/efflux proteins are expressed at much higher levels early in development than in the adult, and there is physiological evidence that these transport systems are functionally more active in the developing brain [150]. How such differences between the pediatric and adult BBB can affect CM pathogenesis and correlate with enhanced BBB permeability in pediatric CM is still unknown. Future research aimed at shedding light on this topic will certainly be useful.

In vivo mouse models of CM have recently implicated MMP dysfunction in disease pathology, although it should be kept in mind that experimental CM presents important differences compared to human CM, such as leukocyte sequestration in cerebral microvessels and subsequent migration into brain parenchyma [49]. An excellent study performed by Van den Steen and his group comprehensively investigated mRNA expression levels of MMPs and protein release or pro-enzyme activation in five different organs (brain, lung, spleen, liver, and kidney) from CM-sensitive C57B1/6 mice infected with P. berghei ANKA (CM model) or P. berghei NK65 (non-CM model) and CM-resistant Balb/C mice infected with P. berghei ANKA (CM-resistant model) [172]. Importantly, they observed enhanced expression and activation of monocytic (CD11b+) MMP-9 in brains of CM mice [172] specific to CM, as suggested by comparison with non-CM models, such as lung pathology [173]. Additionally, tissue-specific increases in mRNA expression were found for several MMPs, including MMP-3, -4, -8, and -13 in spleen, MMP-8, -12, -13, and -14 in liver, and MMP-8 and -13 in brain. All of these increases were more pronounced in the CM model. In a CM-resistant model, MMP-3 expression was significantly enhanced, suggesting a protective role for this MMP in CM [172]. In another study, CM mice showed increased neural MMP-7 protein levels [151]. Interestingly, urokinase-type plasminogen activator (uPA) -/- or urokinase-type plasminogen activator receptor (uPAR) -/- knock-out mice with CM displayed enhanced survival and attenuated thrombocytopenia [174].

A parasite molecule, malarial pigment Hz - a lipid-bound ferriprotoporphyrin IX produced by P. falciparum after hemoglobin catabolism [175] - is proposed to play a role in experimental CM. Indeed, using a sensitive fluorometric method to determine Hz content in blood and tissue samples from mice infected with P. berghei NK65 (non-CM model) or ANKA (CM model), Sullivan and colleagues observed increased Hz levels in tissue correlating with the duration of infection, with neural Hz levels being higher in CM than non-CM mice, raising the possibility that Hz presence may be associated with cerebral pathology [176].

Interestingly, in vitro, Hz appears to play a major role in MMP dysfunction. Phagocytosis of Hz by RAW 264.7 rat macrophage cell line was shown to impair expression of several inflammatory molecules [177] and, after an early inhibitory peak, to increase the long-term mRNA expression of MMP-9 [178]. This effect was related to the lipid moiety of Hz, since lipid-free synthetic Hz (β-hematin) did not modulate MMP-9 expression. The Hz-dependent enhancement of MMP-9 transcription and protein release was mimicked by 4-hydroxy-2-nonenal (4-HNE) [178], a molecule generated by Hz from polyunsaturated fatty acids [179].

In vitro studies using human monocytes and endothelial cells [152, 180‐195] provide convincing and homogeneous evidence for Hz-dependent mechanisms underlying aberrant MMP-9 function (see Figure 4). In a series of works performed with human adherent or immunopurified monocytes from peripheral blood, the phagocytosis of free Hz or Hz-containing trophozoites enhanced MMP-9 mRNA levels, protein expression, and activity [180‐186]. This observation was also investigated using THP-1 monocyte cell line [187]. Hz-fed monocytes display increased total gelatinolytic activity [186] and invasiveness [180] caused by MMP-9 - but not MMP-2 [188] - enhancement.

Increased MMP-9 function in human monocytes appears to be mediated by Hz-dependent over-production of several pro-inflammatory molecules, including TNF-α [180], IL-1β [181], and CCL-3/MIP-1α [184]. Further investigation revealed increases in MMP-9 [181], TNF-α [189, 190] and IL-1β [181, 190], but not CCL-3/MIP-1α [183, 190], were dependent on the lipid moiety of Hz. These studies unveiled a major role for 15-HETE, a potent lipid peroxidation derivative generated by Hz autocatalysis. Hz was also causally related to increased TIMP-1 and lysozyme release from human adherent monocytes, two molecules stored in gelatinase granules along with MMP-9 [186, 190‐192]. Further studies also showed that Hz-induced monocyte degranulation was mediated by TNF-α, IL-1β and MIP-1α/CCL-3 [191, 192] and dependent on Hz lipid moiety, suggesting a major role for 15-HETE [190]. The heme core of Hz was shown to bind MMP-9 hemopexin domain and to prime the activation of the zymogen by other MMPs, such as MMP-3 [193]. The mechanisms underlying Hz-dependent enhancement of MMP-9, TNF-α, IL-1β, CCL-3/MIP-1α, TIMP-1 and lysozyme appear to involve NF-kB activation, as suggested by results from parallel works performed with adherent monocytes from peripheral blood [182, 186, 190, 191] and THP-1 cell line [187]. In these works, Hz-induced enhancement of MMP-9, TNF-α, IL-1β, CCL-3/MIP-1α and TIMP-1, as well as total gelatinolytic and lysozyme activity were abrogated by using different NF-kB inhibitors showing anti-malarial properties (quercetin, artemisinin, parthenolide, and extracts from Punica granatum). Moreover, results from experiments with SB203580, a known inhibitor of p38 MAPK pathway suggest that concurrent activation of p38 MAPK pathway seems to be mandatory for Hz- and 15-HETE-dependent increased MMP-9 [185] and related molecules TNF-α, IL-1β, CCL-3/MIP-1α, TIMP-1 and lysozyme [186, 190, 191]. On the contrary, ERK and JNK MAPK pathways do not seem to be activated by Hz [185].

Additional evidence on Hz-dependent MMP dysregulation is also derived from studies using human endothelial cells. In the human microvascular endothelial cell line HMEC-1, either free Hz [194] or Hz-containing-iRBCs [195] induced the release of pro-MMP-9 and active MMP-9 proteins de novo without altering pro-MMP-2 basal levels. Interestingly, Hz also enhanced the protein levels of basal MMP-1 and MMP-3, two MMPs sequentially involved in pro-MMP-9 activation. Consequently, total gelatinolytic activity and cell invasion were increased. More evidence of Hz-triggered enhancement of MMP-9 protein release emerges from studies using human endothelial cells from large calibre vessels [152]. Similar to human monocytes, a role for the Hz lipid moiety also appears likely in endothelial cells, although the molecules responsible for these effects have not yet been identified [194].

Little evidence exists on the in vivo involvement of MMPs in human CM from organ autopsy studies or from fluid (blood, serum) analysis of patients with uncomplicated or severe malaria [65, 196‐201]. The little data that does exist are somehow conflicting and still incomplete.

The accumulation of pro-angiogenic uPAR [196], MMP-1 and VEGF [197] in Durck granulomas, along with the lack of anti-angiogenic endostatin/collagen XVIII in ring haemorrhage areas [198] has been observed in the brains of patients with fatal CM, suggesting that the proMMP-9 proteolytic machinery is activated in areas of intense parasite sequestration and vascular damage. Consistently, microarray analysis performed on whole blood from Kenyan children with severe malaria showed P. falciparum activation of the human MMP-9 gene [199].

However, sera of Gabonese and Ghanaian children with uncomplicated or severe malaria did not display altered MMP-9 levels [65, 200]. Moreover, serum MMP-8 levels were elevated in Gabonese children with either uncomplicated malaria or CM [200], but not in Nigerian children with uncomplicated malaria [201]. Nonetheless, it has been argued that serum levels might not be an ideal source for reliable data concerning MMP levels, since MMP release from blood cells during sample processing might give a highly non-specific background result, thus hindering the ability to assess true concentrations of circulating MMPs [202].

It is likely that MMPs play an active role in several steps during CM development as they can process a large repertoire of substrates [169‐171], including pro-inflammatory molecules, tight junctions, and hemostatic factors likely involved in CM (see Figure 5 and Tables 4 and 5).

First, MMPs proteolytically cleave the pro-forms of many cytokines and chemokines reportedly enhanced in CM, including TNF-α, IL-1β, CXCL-8/IL-8. As a result of cleavage, these molecules are shed, activated or functionally potentiated, respectively. Thus, MMPs could contribute to the uncontrolled inflammatory response typical of CM. Consistently, an in vitro study using adherent monocytes demonstrated a Hz-dependent pathological auto-enhancing loop established between MMP-9 and TNF-α [180].

Moreover, a growing number of inter-endothelial tight junctions, including occludin, ZO-1, claudin-1 and claudin-5 are known MMP substrates (e.g. MMP-2, MMP-3, MMP-7 and MMP-9), which raises the possibility that MMP activity on these substrates can enhance the permeability of endothelial barriers by destroying these junctions [203‐209]. Therefore, enhanced MMP levels in malaria might facilitate BBB leakage. This hypothesis is consistent with data showing that Hz-containing iRBCs reduced human BBB permeability in vitro[102], and that Hz and iRBCs enhanced MMP-1, MMP-3 and MMP-9 produced by human microvascular endothelium [194, 195].

Finally, MMPs can also influence hemostasis. Both tPA and uPA molecules, responsible for plasminogen conversion to plasmin, as well as tPA/uPA inhibitor PAI-1 can be processed by MMPs [210‐212]. As a consequence, MMPs might affect fibrin degradation by plasmin, perhaps explaining thrombotic events that frequently occur during CM.

Interestingly, some molecules such as 4-aminoquinolines and artemisinins, which are currently used for primary therapy of uncomplicated malaria, have displayed MMP-inhibiting properties. Chloroquine treatment reduces MMP-9 serum levels in patients with systemic lupus erythematosus [213]. Artemisinin down-regulates MMP-2 levels in human melanoma cells [214] and MMP-1, MMP-2 and MMP-9 levels in mouse embryonic stem-cells derived from embryoid bodies [215]. Dihydroartemisinin inhibits MMP-2, MMP-9 and MMP-14 expression/activity in human fibrosarcoma cells [216] and MMP-9 expression in human umbilical vein endothelial cells [217]. Finally, artesunate down-regulates MMP-2 and MMP-7 expression in human non-small cell lung cancer [218]. However, it should be noticed that the drug concentrations used in such studies need to be verified and optimized for human clinical trials. This might explain why antimalarial drugs alone cannot prevent CM development.

It is intriguing to explore the idea of targeting MMPs with broad spectrum or specific MMP inhibitors as adjuvant therapy in CM. In the last two decades, a large number of synthetic MMP inhibitors have gone through clinical trials and largely failed as anti-cancer and anti-arthritis drugs due to serious long-term side effects, with only one currently commercially available [219‐223]. Hopefully, using combinations of MMP inhibitors with antimalarials could justify lower therapeutic doses of both drugs, thereby reducing their potential side effects whilst still enhancing anti-MMP properties by drug synergy. To date, the effects of MMP inhibitors in CM remain scarce. In vitro, the use of a specific synthetic inhibitor of MMP-9 was shown to abrogate Hz-dependent increase of TNF-α in human monocytes, suggesting that MMP-9 inhibition might be useful to counteract pathological inflammation in CM [180]. However, MMP-9 knock-out mice infected with P. berghei ANKA did not display any protection from CM development, probably due to the redundant functions of other MMPs which might compensate for the loss of MMP-9 [172]. On the contrary, treatment with broad-spectrum MMP inhibitor BB-94 significantly improved survival of CM mice [172].

Future research aimed at determining the exact role (protective or detrimental) of each MMP during malaria infections will be highly informative. Unfortunately, with the exception of a few cases, specific inhibitors against individual MMPs are currently lacking [224‐226]. Some metalloproteinases are also produced by malaria parasites, for example to perform hemoglobin degradation [227]. Therefore, MMP inhibitors may not only influence host but also parasitic pathways. Another issue to be taken in account is represented by the effects of MMPs on other organs than brain. However, it should be noted that the adverse effects of MMP inhibitors documented in other pathologies such as cancer were associated with long-term treatment [220], whereas the time course of drug administration in CM therapy should be reasonably shorter, possibly limiting the development of side effects. A detailed analysis of the role of each protease in physiology and pathology, along with the development of specific inhibitors, could yield novel insights to assess whether specific MMP inhibition might be considered as new adjuvant therapies.