Preeclampsia is a multisystem disorder of pregnancy, characterized and indeed defined by the presence of hypertension after 20 weeks’ gestation and before the onset of labor, or postpartum, with either proteinuria or any multisystem complication (1–9). It is a common condition, affecting some 3–5% of nulliparous pregnant women (8, 10) and is characterized by high mortality levels (11–14). There is no known cure other than delivery, and consequently, preeclampsia (PE) also causes significant perinatal morbidity and mortality secondary to iatrogenic prematurity. There are a variety of known risk factors (Table 1) that may be of use in predicting a greater likelihood of developing PE, albeit there are so many, with only very modest correlations, that early-stage (especially the first-trimester) prediction of late-stage PE remains very difficult (8, 15–17).

It is striking that most of the “risk factors” of Table 1 are, in fact, risk factors for multiple vascular or metabolic diseases, i.e., they merely pre-dispose the individual to a greater likelihood of manifesting the disease or syndrome (in this case PE). Indeed, some of them are diseases. This would be consistent with the well-known comorbidities, e.g., between PE and later cardiovascular disease [e.g., Ref. (55–65)], between PE and intracerebral hemorrhage during pregnancy [OR 10.39 (66)], and between PE and stroke postpartum (67, 68). The penultimate row of Table 1 lists a series of diseases that amount to comorbidities, although our interest was piqued by the observation that one-third of patients with antiphospholipid syndrome have PE, and infectious agents with known cross-reacting antigens are certainly one original (external) source of the triggers that cause the antiphospholipid antibodies (43, 44, 47, 69) (and see below). Similarly, in the case of urinary tract infection (UTI), the “risk” factor is a genuine external trigger, a point [following the call by Mignini and colleagues (70) for systematic reviews] that we shall expand on considerably here. [A preprint has been lodged at bioRxiv (71).]

In recent decades, intense investigation has led to the development of a two-stage etiological model for PE, first proposed by Redman (72), in which inadequate remodeling of the spiral arteries in early gestation results in poor placental development (stage one) and the resultant ischemia/reperfusion injury and oxidative stress (73) eventually leads to maternal vascular endothelial cell dysfunction and the maternal manifestations of the disease (stage 2) (72, 74–77). However, many clinical inconsistencies challenge the simplicity of this model. For example, while the association between poor placentation and PE is well established, it is not specific. Poor placentation and fetal growth restriction (FGR) frequently present without maternal signs of PE. Moreover, FGR is not a consistent feature of PE. While it is commonly seen in PE presenting at earlier gestations, in PE presenting at term, neonates are not growth restricted and may even be large for dates (78).

Thus, the two-stage model has been further refined by Roberts and others (77, 79, 80) to take into account the heterogeneous nature of PE and the varying contribution from mother and infant to the disorder. We now appreciate that normal pregnancy is characterized by a low-grade systemic inflammatory response and specific metabolic changes, and that virtually all of the features of normal pregnancy are simply exaggerated in PE (81–83). There is also widespread acceptance that maternal constitutional and environmental factors (such as obesity) can interact to modulate the risk of PE. Thus, with profoundly reduced placental perfusion (or significant “placental loading”), the generation of stage 2 may require very little contribution from the mother to provide sufficient stress to elicit the maternal syndrome. In this setting, almost any woman will develop PE. Conversely, the woman with extensive predisposing constitutional sensitivity could develop PE with very little reduced perfusion, or minimal “placental loading.” As with many complex disorders, multiple factors can affect disease development positively or negatively, with a convenient representation of the two main negative sources (fetal and maternal) being that of a see-saw (84), as in Figure 1.

While this explains the inconsistencies of the two-stage model, the precise mechanisms (1) underlying the initial poor placentation and (2) linking placental stress and the maternal syndrome have still not been fully elucidated.

Much recent research in PE has focused on various angiogenic factors, including the pro-angiogenic factors vascular endothelial growth factor (VEGF) and placental growth factor (PlGF) and the two anti-angiogenic proteins, such as soluble endoglin (sEng) and soluble fms-like tyrosine kinase 1 (sFlt-1). Recent data suggest that alterations in circulating angiogenic factors play a pathogenic role in PE. These angiogenic factors tightly regulate angiogenesis and are also essential for maintenance of normal vessel health. Consequently, the synthesis and action of these factors and their receptors in the uterine bed and placenta are essential for normal placental development and pregnancy (85, 86). In PE, increased levels of the anti-angiogenic sFlt-1 and sEng trap circulating VEGF, PlGF, and transforming growth factor-β (TGF-β), respectively. A myriad of data support the idea that circulating levels of these factors alone, or in combination, can be used to predict PE (87–89) (and see below under PE biomarkers), but in line with the heterogeneous nature of PE, these data are somewhat inconsistent and their performance as biomarkers seems limited to disease with significant placental loading (8). Therefore, angiogenic dysregulation would appear unlikely to be the sole link between the stressed placenta and endothelial dysfunction and the clinical manifestations of the disease.

Notwithstanding these many inconsistencies, the central role of the placenta as a source of “toxin,” in a condition regarded, and indeed often named, as “toxemia of pregnancy” (90–92) cannot be refuted. The uncertainty regarding the nature of the toxin continues, and other placental sources of endothelial dysfunction include syncytiotrophoblast basement membrane fragments (STBM) (93) and endothelial progenitor cells (EPC) (94); an increase of reactive oxygen species over scavenging by antioxidants (95, 96) has also been promoted.

The Bradford Hill criteria for causation of a disease Y by an environmental factor X (97) are as follows: (1) strength of association between X and Y, (2) consistency of association between X and Y, (3) specificity of association between X and Y, (4) experiments verify the relationship between X and Y, (5) modification of X alters the occurrence of Y, and (6) biologically plausible cause and effect relationship.

In general terms (98), if we see that two things (A and B) co-vary in different circumstances, we might infer that A causes B, that B causes A, or that something else (C) causes both B and A, whether in series or parallel. To disentangle temporal relations requires a longitudinal study. The job of the systems biologist doing systems medicine is to uncover the chief actors and the means by which they interact (99), in this way fulfilling the Bradford Hill postulates, a topic to which we shall return at the end.

In infection microbiology, and long predating the Bradford Hill criteria, the essentially equivalent metrics are known [widely, but somewhat inaccurately (100)] as the Koch or Henle–Koch postulates (i.e., criteria). They involve assessing the correlation of a culturable organism with the presence of a disease, the cure of the disease (and its symptoms) upon removal of the organism, and the development of the disease with (re)inoculation of the organism. They are of great historical importance but present us with three main difficulties here. The first is that we cannot apply the third of them to humans for obvious ethical reasons. The second (see also below) and related one is that we cannot usefully apply them in animal models because none of the existing models recapitulates human PE well. Finally, as widely recognized (100–107), they cannot be straightforwardly applied when dealing with dormant bacteria or bacteria that are otherwise refractory to culture.

Our solution to this is twofold: (i) we can assess the first two using molecular methods if culturing does not work and (ii) we exploit the philosophy of science principle known as “coherence” (108–112). This states that if a series of ostensibly unrelated findings are brought together into a self-consistent narrative and that narrative is thereby strengthened. Our systems approach purposely represents a “coherence” in the sense given.

Overall, known biochemical associations with PE come into four main categories, such as abnormal trophoblast invasion, oxidative stress, inflammation and altered immune response, and anti-angiogenic responses (Figure 2). Each of these can contribute directly to PE, and although they can interact with each other (black arrows), no external or causal source is apparent. Figure 2 has been redrawn from a very nice review by Pennington and colleagues (113), which indicates four main generally accepted “causes” (or at least accompaniments) of PE as the four outer colored circles. As illustrated with the black two-way arrows, many of these also interact with each other. What is missing, in a sense, is then what causes these causes, and that is the nub of our argument here. Since we now know (and describe below) that microbes can affect each of these four general mechanisms, we have added these routes to Figure 1 (using pink arrows) where dormant, resuscitating, or growing microbes are known to contribute.

In a similar vein, Magee and colleagues (114) have nicely set down their related analysis of the causes and consequences of PE, with a central focus (redrawn in Figure 3) on endothelial cell activation. While bearing much similarity in terms of overall content to the analysis of Pennington and colleagues (113), and ours above, it again lacks a microbial or infection component as a causative element, but importantly does note that infection and/or inflammation can serve to lower the threshold for PE in cases of inadequate placentation. In our view, microbes can also enter following normal placentation if their dormant microbiome begins to wake up and/or to shed inflammagens.

The question of the extent of heritability of PE (susceptibility) is of interest. Although this seems to vary widely in different studies (Table 1), a number of candidate gene studies (37, 115–118) imply that a susceptibility to PE is at least partly heritable, consistent with the variance in all the other “risk factors” of Table 1 [and see Ref. (6)]. As with all the other gene association studies where phenotypic (“lifestyle”) information is absent (119–121), it is not possible to ascribe the heritability to genetics alone, as opposed to an interaction of a genetic susceptibility (e.g., in the HLA system) with environmental factors (117), such as cytomegalovirus infection (122).

Preeclampsia is accompanied by oxidative stress (123, 124) and inflammation and, thus, shares a set of observable properties with many other (and hence related) inflammatory diseases, be they vascular (e.g., atherosclerosis), neurodegenerative (e.g., Alzheimer’s, Parkinson’s), or “metabolic” (type 1 and 2 diabetes). It is thus at least plausible that they share some common etiologies, as we argue here, and that knowledge of the etiology of these diseases may give us useful clues for PE.

As well as raised levels of inflammatory cytokines that constitute virtually a circular definition of inflammation, we and others have noted that all of these diseases are accompanied by dysregulation of iron metabolism (84, 125, 126), hypercoagulability and hypofibrinolysis (127, 128), blood microparticles (126), and changes in the morphology of fibrin fibers [e.g., Ref. (129–134)] and of erythrocytes [e.g., Ref. (127, 132–137)].

In addition, we and others have recognized the extensive evidence for the role of a dormant blood and/or tissue microbiome in these (138–143) and related (144–147) diseases, coupled in part with the shedding of highly inflammagenic bacterial components such as Gram-negative lipopolysaccharide (LPS) and their Gram-positive cell wall equivalents such as lipoteichoic acids (148). (We shall often use the term “LPS” as a “shorthand,” to be illustrative of all of these kinds of highly inflammagenic molecules.)

The purpose of this review, outlined as a “mind map” in Figure 4, is thus to summarize the detailed and specific lines of evidence suggesting a very important role of a dormant microbial component in the etiology of PE [and see also Ref. (138)]. To do this, we must start by rehearsing what is meant by microbial dormancy.

In microbiology, we usually consider microbes as being in one of the three “physiological macrostates” (Figure 5). The definition of a “viable” bacterium is normally based on its ability to replicate, i.e., “viability” = culturability (149–151). In this sense, classical microbiology has barely changed since the time of Robert Koch, with the presence of a “viable” microorganism in a sample being assessed via its ability to form a visible colony on an agar plate containing suitable nutrients. However, it is well known, especially in environmental microbiology [“the great plate count anomaly” (152)], that only a small percentage of cells observable microscopically is typically culturable on agar plates. In principle, this could be because they are/were “irreversibly” non-culturable (operationally “dead”), or because our culture media either kill them (153) or such media lack nutrients or signaling molecules necessary for their regrowth (154, 155) from an otherwise dormant state (156, 157). These statements are true even for microbes that appear in culture collections and (whose growth requirements) would be regarded as “known.”

However, it is common enough in clinical microbiology that we detect the existence or presence of “novel” microbial pathogens with obscure growth requirements before we learn to culture them; this is precisely what happened in the case of Legionella pneumophila (158–161), Tropheryma whipplei [Whipple’s disease (162, 163)], and Coxiella burnetii [the causative agent of Q fever (164, 165)]. Even Helicobacter pylori was finally brought into culture on agar plates only because an unusually long Easter holiday break meant that the plates were incubated for an extended period of 5 days (rather than the normal 2) before being thrown out (166, 167)! Consequently, there is ample precedent for the presence of “invisible” microbes to go unremarked before they are discovered as the true cause of a supposedly non-infectious disease, even when they are perfectly viable (culturable) according to standard analyses.

Dormancy for a microbe is defined operationally as a state, commonly of low metabolic activity, in which the organism appears not to be viable in that it is unable to form a colony but where it is not dead in that it may revert to a state in which it can do so, via a process known as resuscitation (156, 157). However, an important issue (and see above) is that dormant bacteria do not typically fulfill the Koch–Henle postulates (100–103), and in order for them to do so it is necessary that they be grown or resuscitated. This is precisely what was famously done by Marshall and Warren when they showed that the supposedly non-infectious disease of gastric ulcers was, in fact, caused by a “novel” organism called H. pylori (168, 169). One of the present authors showed in laboratory cultures of actinobacteria that these too could enter a state of true dormancy (170, 171) [as is well known for Mycobacterium tuberculosis, e.g., Ref. (172–176)], and could be resuscitated by a secreted growth factor called Rpf (177–181). This RPF family has a very highly conserved motif that is extremely immunogenic (182, 183), and it is presently under trials as a vaccine against M. bovis.

It is well known that [with the possible exception of Borrelia (217, 218)] a lack of free iron normally limits microbial growth in vivo [e.g., Ref. (219–243)], and we have reviewed previously (84, 125, 126) the very clear iron dysregulation accompanying PE [e.g., Ref. (90, 244–256)].

This has led to the recognition (128, 139, 141) that the source of the continuing inflammation might be iron-based resuscitation of dormant microbes that could release well known and highly potent inflammagens such as LPS. Indeed, we have shown that absolutely tiny (highly substoichiometric) amounts of LPS can have a massive effect on the blood clotting process (257), potentially inducing β-amyloid formation directly (258, 259) [something, interestingly, that can be mimicked in liquid crystals (260, 261)]. The overall series of interactions envisaged [see also Kell et al. (139)] is shown in Figure 6.

As pointed out by a referee, worldwide, iron deficiency anemia is associated with increased perinatal morbidity and mortality, predominantly in low resource settings. This is most recently and comprehensively reviewed by Rahman and colleagues (262). This meta-analysis (262) found that in low- and middle-income countries, maternal hemoglobin concentrations <10 or <11 g dL⁻¹ or hematocrit values <33 or <34% accounted for 12% of low birth weight, 19% of preterm births (PTBs), and 18% of perinatal mortality. However, the independent contribution of iron deficiency is difficult to define, especially since in low resource settings anemia is often not found in isolation but as part of a wider spectrum of nutritional insufficiencies (especially vitamins and folic acid) or as a result of parasitic infection. These can increase susceptibility to other insults such a hemorrhage or infection, also common in low resources settings. In addition, the exact speciation and hence availability of iron may be important (not just the total amount). Thus, the hypothesis presented here is that iron in excess or under certain physiological conditions can be, and indeed is, equally detrimental to maternal health. There is a likely a “Goldilocks” concentration of iron which is optimal for maternal and neonatal health, and the authors are proposing the selected use of treatments such as statins or iron chelators in the at-risk population, rather than as a broad panacea. Furthermore, the use of statins to prevent pregnancy complications is far from controversial (see also below). There are at least two registered current trials of statins to prevent PE (ISRCTN17787139 and ISRCTN23410175).

By definition, dormant bacteria escape detection by classical methods of assessing viability that involve replication on agar plates. Other growth-associated methods include measurements involving changes in turbidity (263), including an important but now rather uncommon technique referred to as the “most probable number” (MPN). The MPN involves diluting samples serially and assessing by turbidity changes in the presence of growth/no growth. Look-up tables based on Poisson statistics enable estimation of the number of cells or propagules that were present. A particular virtue is that they allow dormant and “initially viable” cells to be discriminated via “dilution to extinction” (171), thereby avoiding many artifacts (157). As mentioned earlier, preincubation in a weak nutrient broth (171, 264) was instrumental in allowing the discovery (177) of an autocrine “wake-up” molecule necessary for the growth of many actinobacteria.

Other more classical means of detecting microbes, but not whether they were culturable, involved microscopy (190, 265–268) or flow cytometry (269) with or without various stains that reflected the presence or otherwise of an intact cell wall/membrane (170, 270–277). These stains are sometimes referred to as “viability” stains, but this is erroneous as they do not measure “culturability.” Readers may also come upon the term “viable-but-not-culturable”; however, since viable = culturable, this is an oxymoron that we suggest is best avoided (157). Other methods involved measurement of microbial products, e.g., CO₂ (278, 279), or changes in the conductivity or impedance of the growth medium (263, 280–282).

Most importantly, however, dormant (as well as culturable) cells may be detected by molecular means, nowadays most commonly through PCR and/or sequencing of the DNA encoding their small subunit ribosomal RNA (colloquially “16S”) (283–297) or other suitable genes. It is clear that such methods will have a major role to play in detecting, identifying, and quantifying the kinds of microbes that we argue lie at the heart of PE etiology.

Of course, actual bacteremia, the presence of replicable bacteria in blood, is highly life-threatening (298), but – as emphasized – viability assays do not detect dormant bacteria. When molecular detection methods are applied to human blood, it turns out that blood does indeed harbor a great many dormant bacteria [e.g., Ref. (299–309)]; they may also be detected ultramicroscopically [e.g., Ref. (139–141, 190, 267, 300, 310)] or by flow cytometry (311), and dormant blood and tissue microbes probably underpin a great many chronic, inflammatory diseases normally considered to lack a microbial component (139–141, 144–147, 190, 267, 268, 302, 312–321). Multiple arguments serve to exclude “contaminants” as the source of the bacterial DNA (141): (1) there are significant differences between the blood microbiomes of individuals harboring disease states and nominally healthy controls, despite the fact that samples are treated identically; (2) the morphological type of organism (e.g., coccus vs. bacillus) seems to be characteristic of particular diseases; (3) in many cases, relevant organisms lurk intracellularly, which is hard to explain by contamination; (4) there are just too many diseases where bacteria have been found to play a role in the pathogenesis, that all of them may be caused by contamination; (5) the actual numbers of cells involved seem far too great to be explicable by contamination; given that blood contains ~5 × 10⁹ erythrocytes mL⁻¹, if there was just one bacterial cell per 50,000 erythrocytes this will equate to 10⁵ bacteria mL⁻¹. These are big numbers, and if the cells were culturable, that number of cells would be the same as that ordinarily defining bacteriuria.

A recent study by Damgaard and colleagues (306) is of particular interest here. Recognizing the strong mismatch between the likelihood of an infection post-transfusion [very high (306)] and the likelihood of detecting culturable microbes in blood bank units (negligible, ca. 0.1%) (306, 322), Damgaard et al. reasoned that our methods of detecting and culturing these microbes might be the problem. Certainly, taking cells from a cooled blood bag and placing them onto an agar plate at room temperature that is directly exposed to atmospheric levels of gaseous O₂ is a huge stress leading to the production of “reactive oxygen species” (125, 323), that might plausibly kill any dormant, injured, or even viable microbes. Thus, they incubated samples from blood on a rich medium (trypticase soy agar) for a full week, both aerobically and anaerobically. Subsequent PCR and sequencing allowed them to identify specific microbes in some 35–53% of the samples. Thus, very careful methods need to be deployed to help resuscitate bacteria from physiological states that normally resist culture, even when these bacteria are well-established species. This is very much beginning to happen in environmental microbiology [e.g., Ref. (154, 324–326)], and such organisms are rightly seen as important sources of novel bioactives (327, 328).

As reviewed previously (139–143), the chief sources of these blood microbes are the gut microbiome, the oral microbiome [periodontitis (329)], and via UTIs. Consequently, if we are to argue that there is indeed a microbial component to PE, we should expect to see some literature evidence for it (53, 54, 138, 330–332). In what follows we shall rehearse the fact that it is voluminous.

It might be natural to assume that the placenta is a sterile organ, like blood is supposed to be. However, various studies [modulo the usual issues of contamination (418)] have shown the presence of microbes in tissues including the placenta (400, 409, 419–432), vagina (393, 433–440), uterus (391, 441, 442), amniotic fluid (430, 443–448), and follicular fluid (449, 450), and how these may vary significantly in PE [we do not discuss other pregnancy disorders such as small for gestational age (SGA) and intrauterine growth restriction (IUGR)]. We list some of these in Table 3.

We have recently rehearsed these issues elsewhere (139–141), so a brief summary will suffice. Clearly, the gut holds trillions of microbes, with many attendant varieties of LPS (461), so even low levels of translocation [e.g., Ref. (462–464)], typically via Peyer’s patches and M cells, provide a major source of the blood microbiome. This may be exacerbated by intra-abdominal hypertension and overeating (465–467) that can indeed stimulate the translocation of LPS (468). For reasons of space and scope, we do not discuss the origins and translocation of microbes in breast milk (469) nor the important question of the establishment of a well-functioning microbiome in the fetus and neonate (470), and the physiological role of the mother therein.

One potential origin of microbes that might be involved in, or represent a major cause of, PE is the oral cavity, and in particular when there is oral disease (such as periodontitis and gum bleeding) that can allow microbes to enter the bloodstream. If this is a regular occurrence one would predict that PE would be much more prevalent in patients with pre-existing periodontitis [but cf. Ref. (471) for those in pregnancy] than in matched controls; this is indeed the case (Table 4). As with many of the tables herein, the odds ratios are far beyond anything that might weakly be referred to as an “association.”

A particular feature of UTIs is the frequency of reinfection (497–504). This is because the organisms can effectively “hide” in bladder epithelial cells as the so-called “quiescent intracellular reservoirs” (212, 501, 503, 505–509) of (presumably) dormant cells that can resuscitate. This is why reinfection is often from the same strains that caused the original infection (510–514). Other complications can include renal scarring (515). Bacteriuria (often asymptomatic) is a frequent occurrence in pregnancy [e.g., Ref. (373, 375, 473, 516–522)], and the frequency of UTI as a source of microbes causing PE is clear from Table 2.

We and others have noted the fact that many chronic, inflammatory disease are accompanied by the shedding of various antigens and other factors; typically they pass through the bloodstream as microparticles (126, 140, 523–530), sometimes known as endosomes (345, 347, 348, 524, 531) [and see later under microRNAs (miRNAs)]. Similarly, LPS is normally bound to proteins such as the LPS-binding protein and apoE (140). Given their prevalence, their role in simply finding their way from maternal blood to placenta, and the fact that we discussed them extensively in two previous reviews (126, 140), we do not discuss them further here.

Activin A is a member of the TGF-β superfamily. Its levels are raised significantly in PE (118, 689–691). However, activin A is also well established as a biomarker of infection (692–695).

In a proteomic study of preclamptic vs. normal placentas (696), calretinin was one of the most differentially upregulated proteins (p = 1.6 × 10⁻¹³ for preterm PE vs. controls, p = 8.9 × 10⁻⁷ for term PE vs. controls), and in a manner that correlated with the severity of disease. While calretinin [normally more expressed in neural tissue and mesotheliomas (697)] is not normally seen as a marker of infection, it is, in fact, raised significantly when Chlamydia pneumoniae infects human mesothelial cells (698).

Chemerin is a relatively recently discovered adipokine, whose level can increase dramatically in the first-trimester of preeclamptic pregnancies (699), and beyond (700). Its levels are related to the severity of the PE (701–703). Specifically, an ROC curve (704) analysis showed that a serum chemerin level >183.5 ng mL⁻¹ predicted PE with 87.8% sensitivity and 75.7% specificity (AUC, 0.845; 95% CI, 0.811–0.875) (699). Papers showing that chemerin is also increased by infection (hence inflammation) include (705, 706); it even has antibacterial properties (707, 708) and was protective in a skin model of infection (709, 710). In a study of patients with sepsis (711), circulating chemerin was increased 1.69-fold compared with controls (p = 0.012) and was also protective as judged by survival. These seem like particularly potent argument for a role of chemerin as a marker of infection rather than of PE per se, and for the consequent fact that PE follows infection and not vice versa.

Copeptin, a glycosylated polypeptide consisting of the 39 C-terminal amino acids of arginine vasopressin, has been suggested as “a new biomarker that is specific for preeclampsia” (712), and certainly changes during its development (713). However, it turns out that it is also essentially a measure of all kinds of stresses and adverse events (714–719), including those caused by infection (720–729).

Not least because kidney function is impaired in PE, low molecular weight (MW) proteins may serve as biomarkers for it. To this end, cystatin C (13 kDa) has been found to be raised significantly in PE (730–736); it also contributed to the marker set in the SCOPE study (8, 17). Notably, although it certainly can be raised during infection (737), it seems to be more of a marker of inflammation or kidney function (738, 739).

“d-dimer” is a term used to describe quite varying forms of fibrin degradation products (740). Given that PE is accompanied by coagulopathies, it is probably not surprising that d-dimer levels are raised in PE (741–745), although this is true for many conditions (746), and some of the assays would bear improvement (747, 748). Needless to say, however, raised d-dimer levels are also a strong marker for infection (749, 750).

Endothelial cell-specific molecule-1 (ESM-1), known as endocan, is a cysteine-rich dermatan sulfate proteoglycan expressed (and sometimes released) by endothelial cells. It has been suggested to be a new biomarker for endothelial dysfunction and PE (751, 752). It would appear, however, to be a rather less specific inflammatory biomarker (753, 754) and is associated with a variety of diseases, including chronic kidney disease (755, 756) and cardiovascular disease (755). Most pertinently from our perspective, it is also raised strongly during sepsis (757–760).

Endoglin is the product of a gene implicated (761, 762) in the rare disease Hereditary Hemorrhagic Telangiectasia. The role of endoglin remains somewhat enigmatic (763). However, endoglin levels were 2.5-fold higher in preeclamptic placentas compared to normal pregnancies (15.4 ± 2.6 vs. 5.7 ± 1.0, p < 0.01). After the onset of clinical disease, the mean serum level of sEng in women with preterm PE was 46.4 ng mL⁻¹, as compared with 9.8 ng mL⁻¹ in controls (p < 0.001) (88). Women with a particular endoglin polymorphism (AA) were 2.29 times more likely to develop PE than those with the GG genotype (p = 0.008) (764), and endoglin is seen as a reasonably good marker for PE (88, 691, 765–768) [cf. Ref. (769)]. Again, endoglin levels are raised following infection by a variety of organisms (770–773), with a particularly clear example that it is a marker of infection coming from the fact that there is raised endoglin only in infected vs. aseptic loosening in joints following arthroplasty (774). In general, it seems likely that these circulating (anti)angiogenic factors are more or less markers of endothelial cell damage, just as we have described for serum ferritin (126).

The natural iron transporter in blood is transferrin [e.g., Ref. (775–780)], present at ca. 1–2 g L⁻¹, with ferritin being an intracellular iron storage molecule, so one is led to wonder why there is even any serum ferritin at all (126, 781). The answer is almost certainly that it is a leakage molecule from damaged cells (126), and when in serum it is found to have lost its iron content (782–785). Serum ferritin is, as expected, raised during PE (244, 246, 249, 253, 255, 786, 787) and in many other inflammatory diseases (126), including infection [e.g., Ref. (788, 789) and above].

microRNAs are a relatively novel and highly important class of ~22 nt non-coding, regulatory molecules (790–793). Some are placenta specific, and those in the circulation [often in endo/exosomes (794–796)] can be identified during pregnancy (797–800), potentially providing a minimally invasive readout of placental condition (801–803). There is aberrant expression of placenta-specific miRNAs in PE including miR-517a/b and miR-517c (804–810) and miR-1233 (811). C19MC is one of the largest miRNA gene clusters in humans, maps to chromosome 19q13.41, and spans a ~100 kb long region. C19MC miRNAs are processed from the cluster (812), are primate-specific, conserved in humans, and comprise 46 miRNA genes, including the miR-517 family (813). miR-517 is known to be antiviral (814, 815), while miR-517a overexpression is apoptotic (816) and can inhibit trophoblast invasion (817). Importantly for our argument, miR-517 molecules are overexpressed following infection (818, 819).

Although, as its name suggests, neuropeptide Y is a neurotransmitter, it is also correlated with stress. Certainly, it is related to noradrenaline (see below) that may itself be responsible for the raised blood pressure (BP) in PE (820). It is also raised in sepsis, where it is considered to counterbalance the vasodilation characteristic of septic shock [e.g., Ref. (821, 822)]. The apparent paradox of a raised BP in PE and a lowered one in septic shock is considered to be related to the very different concentrations of endotoxin involved (Figure 7).

Neutrophil gelatinase-associated lipocalin (NGAL) is a lipocalin that is capable of binding catecholate-based siderophores (125, 823, 824). As such it is antimicrobial and is also an inflammatory or sepsis biomarker (825, 826). Given our interest in iron, it is not surprising that it is changed during PE. While one study suggested it to be decreased in PE (827), a great many other studies showed it to be increased significantly in PE, and typically in a manner that correlated with PE severity (735, 828–837). Pertinently to PE, it is also well established as an early biomarker of acute kidney injury (AKI) (838–841). However, it is not a specific biomarker for AKI vs. sepsis (839, 842–850) and its origin in sepsis differs (851, 852). Of course, it can be the sepsis that leads to the AKI (853, 854). Fairly obviously, while it does tend to be increased during PE, we again see its direct role as an antimicrobial and marker of sepsis as highly supportive of our present thesis.

This is a member of the VEGF Family that despite its name has a great many activities (855). It is often considered in parallel with endoglin and sFlt, with a high sFlt:PlGF ratio being considered as especially discriminatory for PE (89, 856–869), i.e., a lower PlGF can be diagnostic of PE (769, 870–873). PlGF tends to be raised in sepsis unrelated to pregnancy (874, 875), while its lowering in PE may be due to the excess sFLT that decreases it (855, 876, 877). In one study of a patient with CMV infection and PE, it was, in fact, raised (878), while, in the case of IUGR, it was massively lowered (879). PlGF alone is thus probably not a useful general marker for either PE or sepsis if one is trying to disentangle them, although it has clear promise when PE is superimposed on CKD (872, 880).

Procalcitonin is the 116 amino acid polypeptide precursor of calcitonin, a calcium regulatory hormone. It is another marker that has been observed to be raised (according to severity) in preeclamptics (742, 881, 882) [but cf. Ref. (713)]. However, it is also a known marker of bacterial infections or sepsis (881, 883–891).

This is an inflammatory biomarker, which was shown to increase fourfold in PE in one study (892), was significantly raised in another (882), but not in a third (893). However, it is a well-established (and potent) biomarker for infection/sepsis [e.g., Ref. (894–907)]. Defective amyloid processing may be a hallmark of PE more generally (908), and of course amyloid can be induced by various microbes (317, 319, 909, 910) and their products (257).

The sFlt receptor is a splice variant of the VEGF receptor (766). It is raised considerably in PE (691, 767, 856, 862, 864, 911–914) and may be causal (545, 590, 915–918). Needless to say, by now, we can see that it is also a very clear marker of infection (767, 919, 920), whose levels even correlate with the severity of sepsis (921–923). Of particular note is the fact that sFLT is actually anti-inflammatory (922).

Soluble thrombomodulin was recognized early as an endothelial damage biomarker and is raised in PE (924–934). Interestingly, it has been found to have significant efficacy in the treatment of sepsis [-based disseminated intravascular coagulation (DIC)] (935–943).

TLR4 upregulation in preeclamptic placentas (944) is entirely consistent with infection and the “danger model” as applied to PE (945). As well as LPS activation [reviewed in Kell and Pretorius (140)], the heat shock protein 60 of Chlamydia also activates TLR4 (138).

Visfatin is another adipokine that is raised in PE, approximately 2-fold in the study of Fasshauer and colleagues (946), and 1.5-fold in that of Adali and colleagues (947). However, it was little different in a third study (948), while in a different study, it was rather lower in PE than in controls (949). This kind of phenomenon rather lights up the need for excellent quality studies, including ELISA reagents, when making assessments of this type.

Fairly obviously, the conclusion that this long list of biomarkers that are raised in PE might be specific “PE” biomarkers is challenged very strongly by the finding that they are, in fact, all known markers of infection, a finding that in our view strongly bolsters the case for an infectious component in PE.

In a similar vein, there are a number of other sepsis markers (where sepsis is varied via, or occurs as, an independent variable) that we would predict are likely to be visible as raised in PE patient. These might include (680, 950) PAI-1, sE-selectin (951), and sVCAM-1 (921). In particular, Presepsin looks like a potentially useful marker for sepsis (888, 889, 952–961), but we can find no literature on its use as a PE biomarker, where we predict that it may also be raised.

An interesting early study (1002) found that venous plasma noradrenaline was raised by 67% in preeclamptics vs. controls. Similar data were found by others (1003). This is of particular interest in the present context since noradrenaline is well established as highly growth stimulatory to Gram-negative microorganisms [e.g., Ref. (1004–1008)], in part by acting as a siderophore (1009–1011). It also raises the levels of neuropeptide Y (820), and as a stress hormone (1012), is of course well known for its role in raising BP, a hallmark of PE.

There is relatively little metabolomics work in sepsis, but in one study, carnitine and sphingolipid metabolism were also modified during sepsis (1013), while in another (1014), a suite of molecules were decreased during acute sepsis. However, the patients involved here were quite close to death, so it is not clear that comparisons between the metabolome in PE and in dying patients are that worthwhile.

We also note a recent and rather interesting suggestion by Eggers (1015) that the maternal release of adrenaline (rather than noradrenaline) may have an important etiological role in PE, although as with the rest of our thesis here it is not there indicated as to what causes the adrenaline to rise (although infection and inflammation can of course do so).

Hyperuricemia is a moderately common finding in preeclamptic pregnancies and may even be involved in its pathogenesis [see, e.g., Ref. (1016–1022)]. However, it does not seem to be very specific (1023–1027) and is seemingly not an early biomarker [and it did not appear in our own study (15)]. Its lack of specificity is illustrated by the fact that there is considerable evidence for the roles of purinergic signaling (1028), and especially the role of uric acid, in Alzheimer’s and Parkinson’s disease (1029–1031), as well as in a variety of other kinds of inflammatory processes, including pro-inflammatory cytokine production (1032, 1033), the Plasmodium falciparum-induced inflammatory response (1034), the mechanistic basis for the action of alum as an adjuvant (1035), and even peanut allergy (1036–1038). As is common in case–control studies when just one disease (e.g., PE) is studied, artificially high levels of sensitivity and (especially) specificity may appear when other patients with other diseases are not considered.