Methodological approach from the Best Overall Team in the sbv IMPROVER Diagnostic Signature Challenge

Molecular subclasses of preeclampsia characterized by a longitudinal maternal proteomics study: distinct biomarkers, disease pathways and options for prevention

OBJECTIVES

The heterogeneous nature of preeclampsia is a major obstacle to early screening and prevention, and a molecular taxonomy of disease is needed. We have previously identified four subclasses of preeclampsia based on first-trimester plasma proteomic profiles. Herein, we expanded this approach by using a more comprehensive panel of proteins profiled in longitudinal samples.

METHODS

Proteomic data collected longitudinally from plasma samples of women who developed preeclampsia (n=109) and of controls (n=90) were available from our previous report on 1,125 proteins. Consensus clustering was performed to identify subgroups of patients with preeclampsia based on data from five gestational-age intervals by using select interval-specific features. Demographic, clinical, and proteomic differences among clusters were determined. Differentially abundant proteins were used to identify cluster-specific perturbed KEGG pathways.

RESULTS

Four molecular clusters with different clinical phenotypes were discovered by longitudinal proteomic profiling. Cluster 1 involves metabolic and prothrombotic changes with high rates of early-onset preeclampsia and small-for-gestational-age neonates; Cluster 2 includes maternal anti-fetal rejection mechanisms and recurrent preeclampsia cases; Cluster 3 is associated with extracellular matrix regulation and comprises cases of mostly mild, late-onset preeclampsia; and Cluster 4 is characterized by angiogenic imbalance and a high prevalence of early-onset disease.

CONCLUSIONS

This study is an independent validation and further refining of molecular subclasses of preeclampsia identified by a different proteomic platform and study population. The results lay the groundwork for novel diagnostic and personalized tools of prevention.

Maternal whole blood mRNA signatures identify women at risk of early preeclampsia: a longitudinal study

PURPOSE

To determine whether previously established mRNA signatures are predictive of early preeclampsia when evaluated by maternal cellular transcriptome analysis in samples collected before clinical manifestation.

MATERIALS AND METHODS

We profiled gene expression at exon-level resolution in whole blood samples collected longitudinally from 49 women with normal pregnancy (controls) and 13 with early preeclampsia (delivery <34 weeks of gestation). After preprocessing and removal of gestational age-related trends in gene expression, data were converted into Z-scores based on the mean and standard deviation among controls for six gestational-age intervals. The average Z-scores of mRNAs in each previously established signature considered herein were compared between cases and controls at 9-11, 11-17, 17-22, 22-28, 28-32, and 32-34 weeks of gestation.Results: (1) Average expression of the 16-gene untargeted cellular mRNA signature was higher in women diagnosed with early preeclampsia at 32-34 weeks of gestation, yet more importantly, also prior to diagnosis at 28-32 weeks and 22-28 weeks of gestation, compared to controls (all, p < .05). (2) A combination of four genes from this signature, including a long non-protein coding RNA [H19 imprinted maternally expressed transcript (H19)], fibronectin 1 (FN1), tubulin beta-6 class V (TUBB6), and formyl peptide receptor 3 (FPR3) had a sensitivity of 0.85 (0.55-0.98) and a specificity of 0.92 (0.8-0.98) for prediction of early preeclampsia at 22-28 weeks of gestation. (3) H19, FN1, and TUBB6 were increased in women with early preeclampsia as early as 11-17 weeks of gestation (all, p < .05). (4) After diagnosis at 32-34 weeks, but also prior to diagnosis at 11-17 weeks, women destined to have early preeclampsia showed a coordinated increase in whole blood expression of several single-cell placental signatures, including the 20-gene signature of extravillous trophoblast (all, p < .05). (5) A combination of three mRNAs from the extravillous trophoblast signature (MMP11, SLC6A2, and IL18BP) predicted early preeclampsia at 11-17 weeks of gestation with a sensitivity of 0.83 (0.52-0.98) and specificity of 0.94 (0.79-0.99).

CONCLUSIONS

Circulating early transcriptomic markers for preeclampsia can be found either by untargeted profiling of the cellular transcriptome or by focusing on placental cell-specific mRNAs. The untargeted cellular mRNA signature was consistently increased in early preeclampsia after 22 weeks of gestation, and individual mRNAs of this signature were significantly increased as early as 11-17 weeks of gestation. Several single-cell placental signatures predicted future development of the disease at 11-17 weeks and were also increased in women already diagnosed at 32-34 weeks of gestation.

The prediction of early preeclampsia: Results from a longitudinal proteomics study

OBJECTIVES

To identify maternal plasma protein markers for early preeclampsia (delivery <34 weeks of gestation) and to determine whether the prediction performance is affected by disease severity and presence of placental lesions consistent with maternal vascular malperfusion (MVM) among cases.

STUDY DESIGN

This longitudinal case-control study included 90 patients with a normal pregnancy and 33 patients with early preeclampsia. Two to six maternal plasma samples were collected throughout gestation from each woman. The abundance of 1,125 proteins was measured using high-affinity aptamer-based proteomic assays, and data were modeled using linear mixed-effects models. After data transformation into multiples of the mean values for gestational age, parsimonious linear discriminant analysis risk models were fit for each gestational-age interval (8-16, 16.1-22, 22.1-28, 28.1-32 weeks). Proteomic profiles of early preeclampsia cases were also compared to those of a combined set of controls and late preeclampsia cases (n = 76) reported previously. Prediction performance was estimated via bootstrap.

RESULTS

We found that 1) multi-protein models at 16.1-22 weeks of gestation predicted early preeclampsia with a sensitivity of 71% at a false-positive rate (FPR) of 10%. High abundance of matrix metalloproteinase-7 and glycoprotein IIbIIIa complex were the most reliable predictors at this gestational age; 2) at 22.1-28 weeks of gestation, lower abundance of placental growth factor (PlGF) and vascular endothelial growth factor A, isoform 121 (VEGF-121), as well as elevated sialic acid binding immunoglobulin-like lectin 6 (siglec-6) and activin-A, were the best predictors of the subsequent development of early preeclampsia (81% sensitivity, FPR = 10%); 3) at 28.1-32 weeks of gestation, the sensitivity of multi-protein models was 85% (FPR = 10%) with the best predictors being activated leukocyte cell adhesion molecule, siglec-6, and VEGF-121; 4) the increase in siglec-6, activin-A, and VEGF-121 at 22.1-28 weeks of gestation differentiated women who subsequently developed early preeclampsia from those who had a normal pregnancy or developed late preeclampsia (sensitivity 77%, FPR = 10%); 5) the sensitivity of risk models was higher for early preeclampsia with placental MVM lesions than for the entire early preeclampsia group (90% versus 71% at 16.1-22 weeks; 87% versus 81% at 22.1-28 weeks; and 90% versus 85% at 28.1-32 weeks, all FPR = 10%); and 6) the sensitivity of prediction models was higher for severe early preeclampsia than for the entire early preeclampsia group (84% versus 71% at 16.1-22 weeks).

CONCLUSION

We have presented herein a catalogue of proteome changes in maternal plasma proteome that precede the diagnosis of preeclampsia and can distinguish among early and late phenotypes. The sensitivity of maternal plasma protein models for early preeclampsia is higher in women with underlying vascular placental disease and in those with a severe phenotype.

The prediction of late-onset preeclampsia: Results from a longitudinal proteomics study

BACKGROUND

Late-onset preeclampsia is the most prevalent phenotype of this syndrome; nevertheless, only a few biomarkers for its early diagnosis have been reported. We sought to correct this deficiency using a high through-put proteomic platform.

METHODS

A case-control longitudinal study was conducted, including 90 patients with normal pregnancies and 76 patients with late-onset preeclampsia (diagnosed at ≥34 weeks of gestation). Maternal plasma samples were collected throughout gestation (normal pregnancy: 2-6 samples per patient, median of 2; late-onset preeclampsia: 2-6, median of 5). The abundance of 1,125 proteins was measured using an aptamers-based proteomics technique. Protein abundance in normal pregnancies was modeled using linear mixed-effects models to estimate mean abundance as a function of gestational age. Data was then expressed as multiples of-the-mean (MoM) values in normal pregnancies. Multi-marker prediction models were built using data from one of five gestational age intervals (8-16, 16.1-22, 22.1-28, 28.1-32, 32.1-36 weeks of gestation). The predictive performance of the best combination of proteins was compared to placental growth factor (PIGF) using bootstrap.

RESULTS

1) At 8-16 weeks of gestation, the best prediction model included only one protein, matrix metalloproteinase 7 (MMP-7), that had a sensitivity of 69% at a false positive rate (FPR) of 20% (AUC = 0.76); 2) at 16.1-22 weeks of gestation, MMP-7 was the single best predictor of late-onset preeclampsia with a sensitivity of 70% at a FPR of 20% (AUC = 0.82); 3) after 22 weeks of gestation, PlGF was the best predictor of late-onset preeclampsia, identifying 1/3 to 1/2 of the patients destined to develop this syndrome (FPR = 20%); 4) 36 proteins were associated with late-onset preeclampsia in at least one interval of gestation (after adjustment for covariates); 5) several biological processes, such as positive regulation of vascular endothelial growth factor receptor signaling pathway, were perturbed; and 6) from 22.1 weeks of gestation onward, the set of proteins most predictive of severe preeclampsia was different from the set most predictive of the mild form of this syndrome.

CONCLUSIONS

Elevated MMP-7 early in gestation (8-22 weeks) and low PlGF later in gestation (after 22 weeks) are the strongest predictors for the subsequent development of late-onset preeclampsia, suggesting that the optimal identification of patients at risk may involve a two-step diagnostic process.

Classification of Non-Small Cell Lung Cancer Using Significance Analysis of Microarray-Gene Set Reduction Algorithm

Among non-small cell lung cancer (NSCLC), adenocarcinoma (AC), and squamous cell carcinoma (SCC) are two major histology subtypes, accounting for roughly 40% and 30% of all lung cancer cases, respectively. Since AC and SCC differ in their cell of origin, location within the lung, and growth pattern, they are considered as distinct diseases. Gene expression signatures have been demonstrated to be an effective tool for distinguishing AC and SCC. Gene set analysis is regarded as irrelevant to the identification of gene expression signatures. Nevertheless, we found that one specific gene set analysis method, significance analysis of microarray-gene set reduction (SAMGSR), can be adopted directly to select relevant features and to construct gene expression signatures. In this study, we applied SAMGSR to a NSCLC gene expression dataset. When compared with several novel feature selection algorithms, for example, LASSO, SAMGSR has equivalent or better performance in terms of predictive ability and model parsimony. Therefore, SAMGSR is a feature selection algorithm, indeed. Additionally, we applied SAMGSR to AC and SCC subtypes separately to discriminate their respective stages, that is, stage II versus stage I. Few overlaps between these two resulting gene signatures illustrate that AC and SCC are technically distinct diseases. Therefore, stratified analyses on subtypes are recommended when diagnostic or prognostic signatures of these two NSCLC subtypes are constructed.

Using ILP to Identify Pathway Activation Patterns in Systems Biology

We show a logical aggregation method that, combined with propositionalization methods, can construct novel structured biological features from gene expression data. We do this to gain understanding of pathway mechanisms, for instance, those associated with a particular disease. We illustrate this method on the task of distinguishing between two types of lung cancer; Squamous Cell Carcinoma (SCC) and Adenocarcinoma (AC). We identify pathway activation patterns in pathways previously implicated in the development of cancers. Our method identified a model with comparable predictive performance to the winning algorithm of a recent challenge, while providing biologically relevant explanations that may be useful to a biologist.

Strengths and limitations of microarray-based phenotype prediction: lessons learned from the IMPROVER Diagnostic Signature Challenge

MOTIVATION

After more than a decade since microarrays were used to predict phenotype of biological samples, real-life applications for disease screening and identification of patients who would best benefit from treatment are still emerging. The interest of the scientific community in identifying best approaches to develop such prediction models was reaffirmed in a competition style international collaboration called IMPROVER Diagnostic Signature Challenge whose results we describe herein.

RESULTS

Fifty-four teams used public data to develop prediction models in four disease areas including multiple sclerosis, lung cancer, psoriasis and chronic obstructive pulmonary disease, and made predictions on blinded new data that we generated. Teams were scored using three metrics that captured various aspects of the quality of predictions, and best performers were awarded. This article presents the challenge results and introduces to the community the approaches of the best overall three performers, as well as an R package that implements the approach of the best overall team. The analyses of model performance data submitted in the challenge as well as additional simulations that we have performed revealed that (i) the quality of predictions depends more on the disease endpoint than on the particular approaches used in the challenge; (ii) the most important modeling factor (e.g. data preprocessing, feature selection and classifier type) is problem dependent; and (iii) for optimal results datasets and methods have to be carefully matched. Biomedical factors such as the disease severity and confidence in diagnostic were found to be associated with the misclassification rates across the different teams.

AVAILABILITY

The lung cancer dataset is available from Gene Expression Omnibus (accession, GSE43580). The maPredictDSC R package implementing the approach of the best overall team is available at www.bioconductor.org or http://bioinformaticsprb.med.wayne.edu/.