Preeclampsia (PE) is a severe disorder of human pregnancies affecting up to 10 % of primiparous women in industrialized countries. This hypertensive disease develops from the second half of pregnancy and is associated with proteinuria, and sometimes with oedemas. The causes of preeclampsia are complex and multiple, with a combination of environmental and genetic effects from maternal [1] as well as paternal origin [2, 3]. At the histological level, preeclampsia is characterized by a shallow colonization of the maternal endometrium (and more specifically of the wall of the uterine vessels) by the invasive cytotrophoblasts. In a normal gestation, this process occurs during the second trophoblastic colonization wave around the end of the first trimester of gestation allowing the invasive cytotrophoblasts to reach the placental bed. In this case, the arterial wall is infiltrated by endovascular trophoblasts, triggering a suppression of the vasomotor control, thus resulting in a very important dilatation of the lumen and a loss of the elastic properties of the arteries [4–8]. By contrast, when the invasion is defective, remaining too shallow [9], fibrinous material accumulates in the arteries, myocytes proliferate. This may lead to local thrombosis and is therefore supposed to impact on the local oxygen pressure leading to placental ischemia/hypoxia, and ultimately to functional anomalies of the maternal vascular endothelium [10, 11]. Finally, the functions of the syncytiotrophoblast, the specific tissue resulting from the fusion of the cytotrophoblasts, may be modified, leading to trophoblast apoptosis. Links between placental pathology and hypoxia are now clearly documented ([12, 13], for a recent review, see Challier et Uzan, 2003; [14]). Intra-Uterine Growth Retardation (IUGR) constitutes another group of complex diseases, a large subset of which is associated with placental malfunction, and often with preeclampsia, although this is not at all systematic. IUGR may be defined as a rupture in the normal growth curve of the foetus, although it is generally defined as a birth weight inferior to the 10^th or to the 3^rd percentile (according to a chosen arbitrary limit) of the smallest babies. This simple definition does not take into account the dynamics of the growth curve, but is very easily workable. This explains at least partly the underlying complexity present under the term IUGR. At least, two very distinct situations are possible: IUGR may be the result of endogenous developmental and growth factors affecting the fetus growth; or, alternatively, a placental defect may inhibit the transfer of nutriments and oxygen from the mother to the fetus (IUGR of vascular origin or not). In the following, vascular IUGR correspond to those exhibiting an abnormal Doppler. In this latter case, the IUGR may be caused by a vascular pathology such as PE. Consequently, there is a clear need for classification of IUGR, and for improving the understanding of its etiology, both issues possibly based upon the characterization of specific marker genes.

In addition, it is now well documented that babies presenting with a small birth weight are at increased risk for developing systemic pathologies at adulthood, such as diabetes or cardiovascular troubles [15, 16]. One clear limitation of the therapeutic possibilities for PE and IUGR resides in the fact that whilst many indications suggest that their causes are very precocious during placental development [17], their symptoms occur late, at least from mid-gestation, and more often from the last trimester of the pregnancy. Evaluating the risk early would allow orienting the medical choices towards a better follow-up of the pregnancy, or even towards some pharmacological options, such as the chronic use of mild doses of aspirin.

In order to explore on a wide basis the links between hypoxia and placental diseases, and to identify early putative markers of the preeclamptic pathology, we combined the Subtractive/Suppressive Hybridization (SSH) methodology starting with cDNA material obtained from first trimester placental purified villi with the construction of high-density nylon macroarrays (Figure 1). A first characterization of the libraries has previously been performed by systematic sequencing, enabling to identify all the genes that were modulated by short-term hypoxia in early placental villi, even when this genes are expressed at a low level [18]. However, this systematic sequencing approach does not indicate which gene could constitute an optimal early marker for disease detection (highly and differentially expressed). Therefore, in the present study, we used the less sensitive approach of hybridization in order to identify genes modulated early by hypoxia in pregnancy (the end of first trimester, which is crucial for trophoblast invasion), and also highly expressed and modified in placental diseases occurring later. We report here on the macroarray hybridization results obtained from 20 different sources of cDNA. We demonstrate a clear association between PE and the early hypoxic induction of a series of genes involved in different metabolic processes. This association was totally absent when cDNAs from isolated IUGR were used as probes. We could also demonstrate that the kinetics of hypoxia is not linear, and that genes transcriptionnally modulated in early placentas after 3 h of hypoxia may return to their basal expression level when the culture is maintained for 48 h in hypoxic conditions. Finally, we demonstrate that genes highly expressed in the placenta are clustered in specific chromosome regions, in particular in regions previously defined as containing imprinted genes, such as 11p15.5.

In this study we designed a new transcriptome resource directed at evaluating the effects of hypoxia on human placenta. This tool is particularly original if compared to the existing commercial membranes, since the distribution of clones into two subgroups makes it possible to analyze statistically whether one condition is connected to hypoxia-induced stimulation or inhibition of gene expression. This was clearly shown for PE, a pathologic condition that correlated very well with the induction of "hypoxic" genes. Since the placenta is one of the organs presenting the greatest abundance of diversified transcripts, these membranes can also be useful for characterizing either other biological systems, or the effects of hypoxia on other tissues. The use of the SSH approach to construct the membranes may also generate some biases as the cloning is dependent on the existence of Rsa1 restriction sites in the starting material (driver and tester cDNAs). Nevertheless, our tool, focused on early hypoxia, is a useful complement to other DNA arrays experiments, based on commercial membranes [30]. As shown in our study, this tool was used to analyze at the mRNA level the consequences of the two most frequent placental pathologies, preeclampsia and Intra-Uterine Growth Retardation.

Some of the genes found associated with PE in the present study had already been evaluated as putative markers of placental pathological status, such as human chorionic gonadotrophin (hCG) and human placental lactogen (hPL, also known as CSHL1), as well as pregnancy specific glycoproteins (PSGs). A significant serum increase in hCG was found more prevalent in preeclamptic women [23]. In another recent study, hPL and PSGs were found reduced at 17 weeks in the serum of patients who develop later a preeclamptic condition, albeit serum levels are restored later [31]. hCG is composed of two polypeptidic chains, α and β, encoded by CGA (common to four polypeptide hormones) and CGB, respectively. It is known to play important roles in placental physiology [32]. In our study CGA was indeed found induced in several PE cases, but could not be associated with a specific differential oxygen status. CGB was present in the membranes, but was not specifically induced in preeclampsia. The only CGB clone was located in plate 7H12 (normoxia). The over-expression of CGA could lead to an overall increase of hCG in the patient serum however the mechanisms involved for regulating the expression of the two polypeptidic chains constitutive of hCG seem to display opposite modes of regulation.

Recently, Bersinger and Odegard [31] have demonstrated that in IUGR, hPL is continuously lowered later in pregnancy. This is consistent with our results, as we could not detect any trace of expression of this gene in the IUGR probes that were hybridized on the membranes. Similarly, the expression level was low in control term placentas (FTP). However, hPL appeared strongly induced in PE, especially in severe cases. These results at the mRNA level differ from the findings of Bersinger and Odegard [31] concerning the hPL protein levels, which were going back to normal at weeks 28 and 33 of preeclampsia. Possibly, the accumulation of hPL mRNA would not be followed by translation in PE.

Only three PSG were found in the membranes amongst the 11 existing genes. PSGs are supposed to play an essential immunomodulatory effect in pregnancy [33–35]. Consistently with our result, in a recent microarray study [36] the authors used a commercial membrane to identify differentially expressed genes in preeclamptic and normal placentas. Among the ten most highly expressed genes in the membranes, the authors have identified PSG4, PSG5 and PSG7. Interestingly, those are exactly the highly expressed genes that we have found in term placentas suggesting that only these PSGs are specifically expressed at a high level in term pregnancies. This observation raises the question of the regulation of the entire cluster of PSG genes which spans roughly 550 kilobases on 19q13 in the order 3, 1, 6, 7, 11, 2, 5, 4, 9 (the precise localization of the two other genes PSG 8 and PSG10 is not yet known). To address the question of PSG regulation, we performed a Clustal alignment of 3 kilobases upstream of the first codons of PSG 1, 2, 3, 4, 5, 6, 7, 9, and 10. The clustering of the promoters indicated a very high level of conservation and was not able to group PSG4, PSG5 and PSG7 as more similar together (not shown). This indicates that the specifically high expression level of 3 genes, which are not contiguous, depends on specific long-range acting chromatin factors.

ANGPTL4, also called FIAF (fasting-induced adipose factor) is another 3 h hypoxia-activated gene recurrently found in PE cases. It is a downstream target of PPARγ (peroxisome-proliferator-activated receptor γ), and is therefore supposed to regulate lipid metabolism and glucose homeostasis [37]. Although belonging to a family of 9 genes, it is the only one that we could detect by hybridization, suggesting a highly specific mode of regulation, consistently with what has been described of its specific regulation compared to ANGPTL3 [38]. In mice, ANGPTL4/FIAF is increased in the plasma by fasting and decreased by high fat diet [39], demonstrating its involvement in lipid capture in difficult physiological conditions, of which PE may represent a paradigmatic case.

Another unexpected actor of the onset of PE could be H19. This RNA-encoding gene is of ill-defined function. It is located in the 11p15.5 cluster of imprinted genes and expressed by the maternal allele, in apparent opposition with IGFII. Both genes appear regulated by a common DMR (differentially methylated region), and have opposite effects on fetal growth in mice [20, 40]. Both IGFII and H19 were found in the set of hypoxia-induced genes, consistently with the existence of HIF1α binding sites in their promoters suggesting the ability to respond almost instantaneously to variations in oxygen concentration. For IGFII, this is also consistent with its described although controversed angiogenic properties [41, 42]. However, while IGFII was not PE-specific (and was found expressed in term placentas, pathologic placentas (either PE or vascular IUGR) as well as in early placentas exposed to hypoxia), H19 was specifically expressed in PE and strongly in severe PE.

In our study, we could observe that true vascular IUGR results in a drastic reduction of the transcriptional activity of the placenta. In these cases, only mRNAs for IGFII, PSG4, PSG5 and PSG7, and mitochondrial genes were present, and no polyA could be detected. Therefore, we could conclude that in vascular IUGR, only a minimal set of genes was transcriptionnally maintained in the placental tissue in order to prevent spontaneous abortion, (i) transcripts from genes of the respiratory cascade, (ii) IGFII, one isoform of which is the essential growth factor in placenta [20], and (iii) three genes of the PSG cluster. The absence of polyA in the cDNAs from purely vascular IUGR suggests that the remaining clones are maintained by stabilization of the transcripts rather than at the transcriptional level. This may be substantiated by the observation of IMP3 cDNA in the transcripts specifically expressed in less severe cases of IUGR. IMP3 encodes a factor interacting with the IGFII mRNA, possibly stabilizing the transcript, and belonging to an imprinted region on chromosome 7.

In non-purely vascular IUGR, FTH1 (ferritin) was expressed at a high level, suggesting the existence of genetic adaptative mechanism to a restricted supply of nutrients. However, ferritin was found twice in the library, without correlation with the oxygen status, showing that this gene does not participate directly in the rapid response to low oxygen concentrations. Ferritin is a major factor for ensuring a sufficient iron store to the neonate at birth. This echoes to a study showing that children presenting a low iron store at birth had low serum ferritin concentrations at 9 months [43], suggesting a risk of iron deficiency in the second postnatal year.

Interestingly, in both pathologies, we found the induction of FEM1-C. This gene, discovered as a homologous of a Caeborhabditis elegans gene contains KH domains, themselves highly present inside FGIF, the principal inducer of the foetal globin. Again, its presence in the library was not correlated with short hypoxia (3 h). However, we observed that its mRNA concentration was lowered at 48 h hypoxia. Its occurrence in pathological situations could refer to a specific adaptative mechanism aiming at increasing the oxygen capture for the foetus. Its induction fits well with the observed increase in ferritin mRNA, both genes FEM1-C and FTH1, aiming at building the proteic and prosthetic part of the globin polypeptide, respectively.

Some particularly appealing genes for the diseases studied were present in the membranes but were not revealed by hybridization with specific pathologic samples, as they appear down-regulated in pathological conditions. Among those, we found two clones corresponding to SERPINE2 and one to SERPING1. These two genes are inhibitors of Serine-Proteases that may play a role in thrombus clearance, and therefore are necessary for adequate circulatory functions. SERPING1 (also known as C1-inhibitor) is particularly interesting, since several mutations of this gene are involved in the development of hereditary angioedema (MIM&606860, [44–47]). Oedemas frequently accompanies the preeclamptic condition, therefore specific malfunctions of SERPINE2 and SERPING1 could represent risk factors for placental diseases. SERPINE2 also appeared in our previous study [18] as a gene present in the hypoxic library at the highest number of occurrences. Therefore, it suggests that a gene induced by short-term hypoxia, may be on the contrary down-regulated by chronic hypoxia, such as observed in PE. Recently, we have shown that the same type of regulation is also true for SERPING1 (F. Quetin and S. T Chelbi, unpublished results).

We performed a systematic analysis of the 5' regions of the PE and IUGR induced genes. In many of them, we found CpG islands that may be modulators of gene expression. In a future work we shall analyze several of these regions by bisulphite analysis of CpG methylation in normal and pathological cases, to try to give a molecular basis to the observed differences in gene expression. In most cases, we discovered more than one HIF binding site in the various promoters identified. Experimental verification will be needed to evaluate the significance of these binding sites.

A surprising result in our study was the demonstration that genes expressed at a level sufficiently high to be detectable in one at least of our hybridization experiments, are mapped to specific chromosome regions. This is particularly the case for clusters of imprinted genes. In these respects, the involvement of the 11p15.5 cluster in the physiopathology of PE fits well with the recent observation that the invalidation of p57/kip2, another gene of this cluster results in "preeclampsia-like" symptoms in mice [48].

In recent studies, several groups have demonstrated that the human genome is organized in large clusters of highly expressed genes [49, 50]. This high level organization of the human genome is conserved in other mammalian species, such as cattle and pigs [51]. In the present study, we demonstrate that beside expression level, clustering also exists for functional purposes, such as placental physiology. This vision of the mammalian genome is consistent with the hypotheses and experimental demonstrations developed by Cremer and co-workers, indicating that the presence of entire chromosomes or chromosome regions inside sub-compartments of the nucleoplasm triggers variations in expression levels [52].

In conclusion, our study has:

• Demonstrated highly significant differences between isolated IUGR and PE concerning the effects of placental oxygen pressure on gene expression. This finding suggests that very different mechanisms are involved when IUGR originates from a fetus-borne developmental dysfunction, and when IUGR results from a vascular defect such as PE.

• Provided the scientific community with a directly available tool, making the link between hypoxia and placental diseases

• Confirmed the importance of several risk factors for PE and IUGR (such as CGA, PSGs and hPL)

• Suggested new possible targets for diagnosing early these pathologies, and possibly for alleviating their effects (such as COX1, ANGPTL4, H19, FTH1, FEM1c, IMP3, SERPING1 and SERPINE2). Since the SSH was carried out from an early placenta, our study makes the link between potential markers of early oxygen depletion, and the late development of preeclamptic lesions, strengthening the idea that PE may be caused by early alterations of placental function.

• Demonstrated the existence of a genomic organization of placental function in placental mammals.

This work opens the way to characterize at the molecular level the physio-pathological mechanisms of very complex situations that often perturb normal pregnancies.