Integrating multiple ‘omics’ analyses identifies serological protein biomarkers for preeclampsia

As the leading cause of maternal morbidity and mortality, preeclampsia (PE) is a pregnancy-related vascular disorder affecting 5% to 8% of all pregnancies [1, 2]. PE, which often associates with fetal growth restriction and pre-term delivery as well as fetal mortality and morbidity, can be remedied by delivery of the placenta and fetus [3]. The etiology of PE is incompletely understood. Current diagnosis of PE is based on the signs of hypertension and proteinuria [4], but lacks sensitivity and specificity and carries a poor prognosis for adverse maternal and fetal outcomes [5]. Thus, there is a need to identify PE biomarkers that can provide a definitive diagnosis with the opportunity for better monitoring of the condition’s progression and, thus, improved outcomes and economic benefits.

Although the pathophysiology remains largely elusive, PE is a multisystem disorder of pregnancy with the placenta playing a pivotal role. Investigators have used genetic, genomic and proteomic approaches to compare PE and control placental tissues. Transcriptional profiling of case–control samples has identified disease-specific expression patterns, canonical pathways and gene-gene networks [6‐12]. Proteomics-based biomarker studies [13‐15] have also revealed candidate biomarkers for future testing. Placental angiogenic and anti-angiogenic factor imbalance, elevated soluble fms-like tyrosine kinase (sFlt-1) and decreased placental growth factor (PIGF) levels are suggested in the pathogenesis of PE [16‐22], and the sFlt-1/PIGF ratio has been proposed as a useful index in the diagnosis and management of PE [23, 24]. However, no widely applicable, sensitive and specific molecular PE test in routine clinical practice is currently available.

In light of these considerations, there is a strong rationale and need to discover diagnostic biomarkers for PE. It is likely unrealistic that a single biomarker could be used to diagnose multifactorial diseases such as PE. Therefore, we employed a comprehensive unbiased multi-’omics’ approach, integrating results from microarray multiplex analysis and proteomic identification by two-dimensional gel analysis. Our applied parametric method [25, 26] allowed us to identify consistent and significant differential gene expression across experiments to develop biomarkers for downstream experimental validation. Serum proteins are routinely used to diagnose diseases, but sensitive and specific biomarkers are hard to find; this may be due to their low serological abundance, which can easily be masked by highly abundant proteins. Our serum protein marker discovery method [27] combines antibody-based serum abundant protein depletion and two-dimensional gel comparative profiling to discover differential protein gel spots between PE and control sera for subsequent protein mass spectrometric identification. We hypothesized that there would be differential serological signatures allowing PE diagnosis. To validate our discovery findings, we tested all the candidates with available ELISA assays, a higher-throughput method. To construct and optimize a sensitive and specific biomarker panel with the least number of protein analytes, a genetic algorithm was used. Close examination of the biomarkers from comparative transcriptomics and proteomics, and their associated pathways led to new hypothesis about their role in PE pathophysiology.

The results validated our hypothesis that sensitive and specific serological biomarker panels can be constructed to diagnose PE. To our knowledge, this represents the first study to employ a muti-’omics’-based biomarker approach to uncover novel PE biomarkers superior to sFlt-1, PIGF, and the sFlt-1/PIGF ratio in PE discrimination. We believe that the functional significance of these PE biomarkers and their associated pathways will provide new insights into the disease pathogenesis and potentially lead to effective novel therapeutics.

We have applied a multi-’omics’ approach to develop validated PE biomarkers, integrating discoveries from placental mRNA expression multiplex analysis and depleted serological proteome two-dimensional gel comparative profiling. Comparing PE and control sera with commercially available ELISA assays, we have validated 11 protein markers, including sFlt-1 and PIGF, and found that our identified PE biomarkers were superior over the sFlt-1/PIGF ratio in diagnosing PE. The concept of combining a transcriptomic approach in placenta tissue with a proteomic approach in serum is novel. It combines the advantages of a study in tissue which is closer to the focus of the pathophysiology with those of a study in serum which is more appropriate for clinical use. Taking proteins that have been discovered from the discovery phase to an ELISA-based validation phase makes the findings of this study translatable into clinical practice.

When comparing the discoveries from expression multiplex analysis and two-dimensional gel serum proteomics, only A2M showed up in both analyses. This could be due to the following reasons: (1) the discordant expression of protein and mRNA as previously characterized [38‐41]; (2) the lack of translation of the placental expression into circulation protein level abundance; and (3) two-dimensional gel technology detection limit of 0.5 to 5 ng. The optimized two-dimensional gel technique has a dynamic range of approximately five orders of magnitude in protein concentration [42], whereas serological protein concentrations vary over approximately ten orders of magnitude, with the highest concentrations reaching mg/mL [43]. Even with the depletion step, protein detection by our two-dimensional gel is limited to proteins whose serological concentrations are >10 ug/mL, clearly influencing the composition of the protein biomarkers we detected. In addition, potentially informative low molecular weight proteins may bind to albumin and, thus, be removed at the depletion step [44], which could be a potential disadvantage. Thus, candidates with pg/mL concentration, for example, sFlt-1 and PIGF, would not be found when applying the two-dimensional gel serum proteomics based approach. Publicly available genome-wide gene expression data on disease tissues can be effectively mined to provide significant synergies to complement our two-dimensional serum proteomics efforts to unveil differential PE biomarker candidates of low serum abundance (pg/mL). Notably, our productive PE discovery efforts support the notion that the multi-’omics’ approach for biomarker analyses are comprehensive, complementary and effective in identifying candidates of a broad dynamic range of serological protein expression, varying from pg/mL to ug/mL.

As summarized in Figure 2, the validated biomarkers’ placental expression, and the early and late gestation maternal serum analyses revealed a similar trend of up- or down-regulation between PE and control samples. However, our study did not explore the extent (percentage) of the contribution by the placenta or other maternal cells to the overall differential serum expression between PE and control subjects. Future expression analysis is needed to characterize the tissue expression pattern of these PE markers and their expression kinetics as a function of the gestational age to understand the tissue specific expression contribution to the differential serum expression pattern observed in this study.

Additional pathway analyses of the protein markers corroborate growing evidence implicating roles for the lipid homeostasis, IL-12 and coagulation canonical pathways in PE pathophysiology. The LXR/RXR activation pathway was identified as the most significant pathway. This supports recent findings [45] that PE is associated with hyperlipidemia and that the regulators of lipid homeostasis are important in the PE pathophysiology. The previous evidence [46‐48] of IL-12, in PE patients, with less activity in placenta and more abundance in sera was reflected as in line with our PE biomarker panel pattern pathway analysis.

A previous multicenter case–control study [24] with an automated assay, demonstrating the utilities of sFlt-1 and PIGF for PE assessment, reported serum abundance of sFlt-1 (PE: 12,981 ± 965 versus control: 2,641 ± 100.5 pg/mL) and PIGF (PE: 76.06 ± 10.71 versus control: 341.5 ± 13.57 pg/mL). Although with greater variation, possibly due to different sample cohorts or assay platforms, the trend of alteration reflected in our results, sFlt-1 (PE: 16,398.02 ± 5,142.32 versus control: 4,282.63 ± 2,532.90 pg/mL) and PIGF (PE: 161.83 ± 118.98 versus control: 383.75 ± 343.84 pg/mL) was in line with their report. As shown in Additional file 2: Figure S1 and summarized in Additional file 1: Table S2, sFlt-1 and PIGF protein abundance differs significantly between early and late gestational age samples in both normal (p value: sFlt-1 0.003, PIGF 0.020) and PE (p value: sFlt-1 0.017, PIGF 0.022) groups. Our biomarker [see Additional file 1: Table S2] RBP4 (p value: normal 0.029, PE 0.176), ADAM12 (p value: normal 0.035, PE 0.777) and pikachurin (p value: normal 0.049, PE 0.502) differs marginally between early and late gestational age samples in normal (p value <0.05) but not in PE (p value >0.05) groups. For HPX, APO C-III, HP, APO-E and APO A1, there was no significant difference (p value >0.05) between early and late gestation sera in both normal and PE groups. The sFlt-1/PIGF ratio was found to be important for the prediction of both preeclampsia and intrauterine growth restriction (IUGR) [49]. Therefore, the previous observation of the sFlt-1 and PIGF expression difference between early and late onset cases may be due to the recruited sample difference between early (both IUGR and early PE) and late (PE only) cases. Given that our sample cohort excluded IUGR, the serum markers identified in this study may be more specific to PE rather than to both IUGR and PE. To summarize, our results here indicate that sFlt-1 and PIGF are regulated during placental development as a function of gestation, and differential expression between PE and control might be due to placental adaptation during PE. The PE biomarkers found in this study are not significantly different between early and late gestation in either PE or control sera. Therefore, their differential expression in PE might directly gauge the disease activity of PE and disease development or reflect features that are present at fairly advanced stages of the pathogenesis, for example, proteinuria and high blood pressure.

Our genetic algorithm-based biomarker panel construction led to final early and late gestational age biomarker panels for PE assessment. Compared to the benchmark sFlt-1/PIGF ratio in PE assessment, our biomarker panels perform comparably during early gestational age but clearly outperform at later gestational weeks. Although the sFlt-1 and PIGF imbalance used for PE diagnosis has been demonstrated, there is mounting evidence to support the notion that normal sFlt-1 and PIGF expression actually characterizes healthy pregnancies [50]. Therefore, sFlt-1 and PIGF may really be general markers for failed pregnancies, for example, ectopic pregnancies, missed abortions, rather than specific to PE. Our multi-’omics’ approach discovered panels of multiple biomarkers, reflecting the multifaceted aspects of PE disease, and have the potential both to provide a definitive diagnosis of PE patients and to be used to monitor disease progression.

We also recognize several limitations to our study. Samples were collected after the clinical diagnosis of PE with disease onset. The outcome information after the sample collection, including the time of delivery, and the birth weight and growth percentile of the babies, is not available. Therefore, the biomarker panels’ utility in risk patient identification remains to be demonstrated. Nevertheless, confirmatory diagnosis is also valuable as it has the benefit of objective diagnosis, reducing over and under diagnoses. To translate our innovative PE markers to clinics, a clinical trial of the prospective cohort design is needed. As one of the limitations of this study, we used commercially available ELISA kits, the antibodies of which may cross-react with other homologous proteins. For example, the R&D sFlt-1 ELISA kit antibodies can recognize both sFlt-1 and full trans-membrane VEGF-R1, as well. To support future prospective trials to test the clinical utility of our PE panel, analyte specific antibody reagents may need to be developed.

Although preeclampsia is diagnosed when a pregnant woman develops both elevated blood pressure and proteinuria, these symptoms tend not be specific and preeclampsia can be asymptomatic as well. Therefore, the clinical definition alone is insufficient to predict adverse maternal and/or neonatal outcomes [51, 52] caused by preeclampsia. Previous prospective cohort studies have found the utility of the elevated sFlt1/PIGF ratio in the prediction of the subsequent adverse maternal and prenatal outcomes within two weeks [53]. The scatter plot analysis (Figure 3) as a function of gestational age suggests that, for our early onset panel, the best performance in PE assessment was obtained near 24 to 25 weeks when comparing to gestation towards 34 weeks. Certain changes in our biomarker panel of serum protein profiles may occur at the first trimester and in advance of clinically-detectable PE disease activity. Thus, we hypothesize that our PE biomarkers can predict impending PE disease activity, and/or adverse outcomes in pregnant women with suspected preeclampsia, especially in a pre-specified group of patients presenting at less than 24 weeks gestation. To test these hypotheses, future prospective cohort studies will be required to address the potential clinical usefulness of our PE biomarkers in predicting impending PE or adverse maternal and/or neonatal outcomes.