Testing an infection model to explain excess risk of preterm birth with long-term iron supplementation in a malaria endemic area

Background and published data on the PALUFER trial, its design and results [2, 5, 7, 8, 13‐15], are summarized in Additional file 1. Laboratory methods are outlined in Additional file 2.

Literature was searched for references illustrative of plausible mechanisms operating for each stage of the model pathways leading to PTB in a malaria endemic area. Search terms focussed on markers of inflammation that could affect birth outcomes. Assumptions for an inflammation-driven model for iron induced excess spontaneous preterm births in the PALUFER trial were formulated derived primarily from published trial results.

Relationships between PTB and biomarkers were assessed using logistic regression models, plotting fitted probabilities with associated 95% CI. The relationship between maternal serum hepcidin and CRP concentrations was strongly non-linear and was fitted using a segmented linear regression model which fits a continuous line composed of straight-line segments with a change of slope at a fitted change point. More complex models were considered including trial arm as a covariate, but as there were no substantive effects of supplementation, these were not more informative and the univariate analyses are presented. Biomarker levels between PTB and term birth groups were compared using Mann–Whitney U-tests. A path model was constructed based on the formulated schematic which incorporated C-reactive protein (CRP) and hepcidin mediation pathways and PTB. Season was used a proxy for malaria infection. The model was fitted to the PALUFER data as described in Additional file 4.

The trial results showed associations between maternal serum CRP concentrations > 5 mg/l at the first (median gestation 18 weeks, IQR 14–23 weeks) and second (median gestation 34 weeks, IQR 33–35 weeks) scheduled antenatal visits (ANC1, n = 282; ANC2, n = 239) and spontaneous PTB incidence (< 37 weeks gestation) [2]. The adjusted relative risk for PTB was higher with CRP > 5 mg/l at ANC1 (1.60, 95% CI 1.00–2.5, P = 0.04) and at ANC2 2.06 (1.04–4.10, P = 0.034). Figure 1 shows the linear association with 95% confidence interval of PTB incidence by log (CRP) for these two visits. There was a strong positive trend with CRP at ANC2 (P = 0.0038), suggesting this inflammatory response was related to spontaneous PTB. Furthermore, the whole gestational age distribution curve was shifted to the left in the iron-supplemented women (see Additional file 3), indicating a population effect on shortening gestation, thereby increasing PTB incidence [2]. Malaria associated inflammation as represented by CRP (geometric mean CRP, mg/l, (n) 1.3 [95% CI 1.0:1.7] (142) without parasitaemia; 9.2 [7.8:10.9] (169) with parasitaemia; relative risk: 1.97 [1.66:2.29], P < 0.001) upregulates hepcidin [8, 16]. At ANC2 (but not ANC1) linear regression showed maternal serum hepcidin concentration was inversely associated with gestational age at delivery (P = 0.002). The linear association and 95% confidence interval of log hepcidin by PTB incidence for these two visits is shown in Fig. 2 with significance only at ANC2 (P = 0.007). Mean ± SD serum hepcidin concentration was higher at both ANC1 and ANC2 in preterm compared to term deliveries (ANC1, 6.9 ± 13.5 vs 5.7 ± 6.9 nmol/l, P = 0.74; ANC2, 4.8 ± 5.3 vs 3.2 ± 6.3 nmol/l, P = 0.004).

In normal pregnancy hepcidin expression decreases physiologically in the second and third trimesters, thereby increasing the supply of iron to the circulation [17]. As malaria risk also alters with gestational age, especially in primigravidae, gestational change in hepcidin may be inhibited by concurrent malaria inflammation. Figure 3 shows plots of the relationship between maternal serum hepcidin and CRP concentrations at the three trial assessment time-points: end assessment for women remaining non-pregnant; for pregnant women at ANC1 and at ANC2. The figure shows a segmented regression model consisting of two segments before and after a sharp rise (post-change) in hepcidin concentration. In the PALUFER trial there was a post-change hepcidin elevation at ANC2 (post-change slope 0.55; 95% CI 0.39–0.92; P < 0.001).

Pathological changes (including detachment of epithelia and shortening of colonic villi) occur in the permeability, leakage of infected erythrocytes into the lumen and dysbiosis of the intestinal gut during P. falciparum infection. Such changes may cause increased intestinal microbiota [6]. The extra enteric iron available with long-term supplementation potentially contributes to disease severity in the gut with a shift in microbiota composition [18, 19] and multiplication of fungal [20] and bacterial [21] pathogens that produce intestinal epithelium biofilm and also contaminate the genital tract. Expression of regulatory inflammatory genes can cause intestinal inflammation, disrupt the intestinal barrier function [22, 23] and prime inflammasome activation. The latter is a multiprotein oligomer responsible for the activation and assembly of the Nod-Like Receptor (NLR)P3 leading to release of pro-inflammatory cytokines, including IL-β [24]. Malaria infection decreases absorption of fortification iron in women [25], which can induce an inflammatory gut reaction [26, 27]. In this way, Plasmodium parasites can have an impact on composition of gut microbiota although bi-directional mechanisms are not well understood, especially in pregnancy [6]. Reactive nitric oxide (NO) intermediates play a prominent role in intestinal barrier damage by inducing enterocyte apoptosis and inhibiting the epithelial restitution processes [28, 29]. NO generation, produced by inducible NO synthase (iNOS), occurs with asymptomatic malaria. Some of the changes related to inflammation from gut infection and which activate pro-inflammatory factors are outlined in Fig. 4. This shows a positive feedback loop with hepcidin expression resulting from gut infection limiting dietary iron absorption, with enteric iron enhancing gut inflammation. Long-term iron supplementation would be expected to exacerbate this cycle as this in addition to food iron increases enteric iron concentration, especially in iron replete women whose iron absorption is already inhibited.

Figure 4 highlights the importance of release of endotoxins and lipopolysaccharide (LPS) into the circulation with inflammation [30, 31]. LPS is a central component of the outer membrane in Gram-negative bacteria and frequently plays a key role during host–pathogen interaction and the establishment of chronic infection in the gut, genital tract and other mucosa. LPS-mediated virulence resides in its endotoxic activity [32]. LPS induces release of acute phase blood proteins associated with pro-inflammatory signalling pathways. Studies in experimental animals have shown intestinal bacteria, especially Enterobacteriaceae and Clostridium perfringens, may influence the level of lipopolysaccharide binding protein and CRP in blood plasma [33] Strong activation is reported in macrophages in response to LPS of NF-κB, a highly conserved transcription factor that regulates inflammatory responses, cellular growth and apoptosis [34]. Placental mediated responses to LPS have been reported in experimental animals with greater cytokine responses in preterm placentas and increased phosphorylated NF-κB induction which may occur by activation of the Toll-like receptor (TLR) pathway [35]. TLRs are essential for an effective host cell response to LPS. Circulating LPS binding to membrane-bound TLR4 results in a series of kinase cascades that phosphorylate NFκB into the nucleus where it acts as a transcription factor stimulating production of proinflammatory cytokines. Placental TLRs can result from transient clinically inapparent maternal bacteraemia [36].

Hepcidin expression is reported to negatively affect proliferation of intrahepatic sporozoites [37], but conversely it may increase susceptibility to iron dependent gut pathogens, e.g., Salmonella. This is because iron-loaded macrophages, which increase with hepcidin expression, have an impaired potential to kill various pathogens. This is partly attributed to reduced formation of NO which is essential to fight infection, as iron blocks transcription of inducible NO synthase (iNOS) [38]. Regulatory mechanisms would depend on the intracellular or extracellular location of bacteria.

Concurrent effects of pregnancy malaria are shown in Fig. 4. The primary events are sequestration of P. falciparum in the placenta and parasite adherence to gut endothelial cells with altered intestinal barrier function [6, 39]. Adherence of infected erythrocytes containing late developmental stages of the parasite (trophozoites and schizonts) to the endothelium of capillaries and venules, is characteristic of P. falciparum infections. Malaria in pregnancy is characterized by the accumulation of infected red blood cells, leukocyte infiltration, and excessive fibrin deposition in the intervillous space of the placenta [40], and is associated with increased risk of low birthweight, attributed to PTB and fetal growth restriction with early development of clinical malaria in infants [41]. Placental intervillositis occurs [36], with induction of oxidative stress biomarkers [42]. Oxidative stress, is an imbalance between oxidants and antioxidants in favour of oxidants, and leads to disruption of redox signalling and physiological function [43]. Placental parasites result in haemozoin formation leading to stimulation of cytokine responses [44, 45], regulation of iNOS expression [46] and NLRP3 inflammasome priming [47]. Haemozoin is an iron-containing pigment which accumulates as cytoplasmic granules in malaria parasites and is a breakdown product of hemoglobin.

The induction of inflammatory cytokines can modulate pregnancy outcomes with beneficial and/or detrimental effects [48]. Maintenance of an appropriate ratio of pro-and anti-inflammatory responses at the feto-maternal interface is a hallmark of successful pregnancy A systemic inflammatory response to malaria during pregnancy leads to increased interleukin (IL)-1β, IL-6, IL-8, IL-10, IL-22, interferon (INF)-Ɣ and soluble tumour necrosis factor (sTNF)-RII in maternal blood [3, 41, 45‐51]. Placental IL-6 and IL-8 have been associated with pregnancy loss and PTB [3, 47‐49]. These biomarkers are predominantly pro-inflammatory, but some have variable anti-inflammatory effects. Expression of CRP, the synthesis of which is rapidly upregulated with inflammation, occurs principally in hepatocytes, under the control of cytokines originating at the site of pathology [52]. The pro-inflammatory cell signalling upregulates hepcidin [53], with increased iron accumulation in macrophages [16, 37], and reduces iron absorption.

The inflammasome promotes maturation and secretion of pro-inflammatory cytokines IL-1β and IL- 18, which results in pro-inflammatory cell death [54]. Nod-Like Receptor (NLR) P3/P12 dependent activation of caspase-1 is likely to be a key event in mediating systemic production of IL-1β with hypersensitivity to secondary bacterial infection during malaria [55]. Bacterial super-infection is well described in acute malaria, and low-density asymptomatic malaria infections are a risk factor for non-typhi salmonella bacteraemia in children [56] and young adults in Burkina Faso [57]. Salmonella typhimurium during malaria may facilitate bacterial colonization [58], and experimental studies in animals indicate shifts in enteric microbiota and increase in susceptibility to intestinal colonization by S. typhimurium [59]. In vivo evidence indicates that S. typhimurium benefits from iron availability during malaria infection [60]. Higher mean body iron stores, as observed in malaria-positive primigravidae in the PALUFER trial [8], and also reported in a Congolese study [61], may favour colonization with S. typhimurium. Evidence is required to substantiate this association in young adults and pregnant women.

A maternal cortisol response occurs with inflammation. In non-endemic malaria areas a majority of studies have reported a consistently negative association between maternal cortisol and infant birthweight [62‐64]. In contrast studies in malaria endemic areas have reported significant positive associations between cortisol concentration and P. falciparum infection prevalence [65, 66]. The process of parturition is associated with inflammation within uterine tissues and it is generally accepted that inflammatory stimuli from multiple extrinsic and intrinsic sources induce labor [67], associated with increases in IL-1β, IL-6 and IL-8 gene expression [68]. Inflammatory stimuli induce labour by affecting progesterone transcriptional activity in uterine cells and causing functional progesterone withdrawal [69]. Progesterone plays a critical role in successful pregnancy and low progesterone concentration is reported in preterm delivery [69], with functional involvement of hormonal and inflammatory stimuli [70, 71], or in experimental animals 24 h following LPS exposure [72]. Hepcidin synthesis in experimental animals has been linked to progesterone receptor membrane component-1 [73]. Young maternal age is important because acquisition of progesterone responsiveness depends on endometrial maturation. Persistence of partial progesterone resistance in adolescents could compromise deep placentation increasing risk of PTB [74].

The characteristics of pregnancy malaria and enteric infection outlined in Fig. 4 indicate a state of chronic inflammation, a maternal cortisol response, progesterone withdrawal and increased risk of PTB. The pathways involved include inflammation-induced cytotoxic effects on progesterone production [75, 76], with progesterone modulation of LPS-induced responses [77, 78]. Metabolic endotoxaemia would impair progesterone production and increase progesterone receptor sensitivity [76, 79, 80]. Concurrent infections including lower genital infections and chorioamnionitis would be expected to be contributory by enhancing activation of the NLRP3 inflammasome, especially with intra-amniotic infection [81]. While the path model is not limited to iron supplemented women, supplementation would influence enteric inflammation before and during pregnancy.

A path model was constructed based on the schematic in Fig. 4 which incorporated the CRP and hepcidin mediation pathway. Full details are provided in Additional file 4. Figure 5 shows the fitted model along with the estimated coefficients. This model confirmed the seasonal malaria effects on PTB, with mediation through CRP and (non-linear) hepcidin induction. As there was no data on the enteric mediators, this process manifests itself as a strong direct treatment effect on PTB. As in previous analyses, any direct induction of hepcidin from iron supplementation is negligible. An additional effect was detected of season on PTB (broken line in Fig. 5). This may reflect a residual effect due to incomplete measurement of the mediating variables (a single measurement at one time may not capture the full extent of the mediation). Alternatively, it may capture other seasonal effects that are not related to malaria, possibly associated with other infections or nutritional changes.

There was no statistically convincing evidence for any other mechanistic paths beyond the model presented.

Interventions to prevent PTB focus mainly on managing risk factors [89]. Improving the biologic understanding of nutritional and inflammatory mediators, and their seasonal patterns in malaria endemic areas may provide novel ways to identify interventions to reduce inflammatory stimuli. A more clinical perspective requires a focus on female adolescents who face high risk of malaria exposure in their first pregnancy, and in whom inflammation exposure may remain unrecognized due to chronic asymptomatic malaria infection in areas with high malaria transmission. The use of long-term iron supplementation, as recommended by WHO, may be detrimental—even when available on an intermittent seasonal basis, as synergistic enteric inflammatory responses would preclude safety. The occurrence of combined exposures with malaria, enteric, as well as genital infections provides an opportunity to consider other co-infection models [90] and double-hit mechanisms [10], which would necessitate different options for infection control. The role for anti-inflammatory agents, which have been evaluated only for therapeutic use, such as targeting the NLRP3 axis might be considered [91]. However, when malaria and enteric infection co-exist, and if the latter are enhanced by iron supplementation, then this model strongly suggests that malaria control in areas with comparable infection exposures should take precedence to iron supplementation to avoid increasing PTB risk.