ABSTRACT: Porphyromonas gingivalis is a periodontal pathogen that is also associated with preterm low birth weight delivery. We investigated the transcriptional responses of human extravillous trophoblasts (HTR-8) to infection with P. gingivalis. Over 2000 genes were differentially regulated in HTR-8 cells by P. gingivalis. In ontology analyses of regulated genes, overpopulated biological pathways included MAP kinase signaling and cytokine production. Immunoblots confirmed over expression of the MAP kinase pathway components MEK3, p38 and Max. Furthermore, P. gingivalis infection induced phosphorylation and activation of MEK3 and p38. Increased production of IL-1β and IL-8 by HTR-8 cells was demonstrated phenotypically by ELISA of HTR-8 cell lysates and culture supernatants. Thus infection of trophoblasts by P. gingivalis can impact signal transduction pathways and modulate cytokine expression, outcomes that could disrupt the maintenance of pregnancy. P. gingivalis were reacted with HTR-8 cells at an MOI of 200 for 2 hours at 370C in 5% CO2. Co-cultures were carried out in quadruplicate. The HIGK cells were lysed with Trizol (Invitrogen) prior to RNA extraction. RNA isolation, cDNA synthesis, labeled cRNA synthesis and chip hybridization were conducted as previously described (Handfield et al., 2005). Briefly, total RNA was extracted from Trizol-lysed cells, treated with DNase I, purified and quantified according to standard methods (Qiagen, Valencia, CA; and Affymetrix, Santa Clara, CA). Complementary DNA (cDNA) synthesis was performed according to the Affymetrix protocol (SuperScript double-stranded cDNA synthesis kit; Invitrogen,) with 8 µg of total cellular RNA used as a template to amplify mRNA species for detection. Double-stranded cDNA was purified, and used as a template for labeled complementary RNA (cRNA) synthesis. In vitro transcription was performed using a BioArray high-yield RNA transcript labeling kit (T7) (Enzo Life Science, Farmingdale, NY), to incorporate biotinylated nucleotides. cRNA was subsequently fragmented and hybridized onto Genechip Human Genome (HG) U133-A Plus 2.0 oligonucleotide arrays (Affymetrix) with proper controls. Each sample was studied in parallel, and the samples were not pooled. The microarrays were hybridized for 16 h at 45°C, stained with phycoerythrin-conjugated streptavidin and washed according to the Affymetrix protocol (EukGE-WS2v4) using an Affymetrix fluidics station, and scanned with an Affymetrix GeneChip 3000 scanner. Microarray data analysis was performed as previously described (Mans et al., 2006, Hasegawa et al., 2008). Briefly, expression filters were applied to remove Affymetrix controls and probe-sets whose signal was undetected across all samples. The signal intensity values of the resulting dataset were variance-normalized, mean-centered and ranked by their coefficients of variation. Normalization was performed to give equal weight to all probe-sets in the analysis, regardless of the order of magnitude of the raw signal intensity. To reduce the confounding effect of background signal variation on the analysis, only the half of the dataset demonstrating the most variation across samples was used to perform unsupervised hierarchical cluster analysis using Cluster software (Eisen et al., 1998). The resulting heat-map and cluster dendrograms were visualized with Treeview software Eisen to reveal the extent of characteristic host cell responses to each infection state, defined as identical treatments clustering together. Additional quality control data for the arrays is provided in the supplemental material. Following initial assessment of the host cell response to each condition, supervised analysis was performed to investigate differences in gene regulation among experimental conditions. For this analysis, the raw signal intensities were log2-transformed for all probe-sets that passed the initial expression filters, and were correlated using BRB Array Tools (Simon and Peng-Lam, National Cancer Institute, Rockville, MD). In each supervised analysis, biological replicates were grouped into classes according to their infection state during co-culture experiments and probe sets significant at the p < 0.001 level between classes were identified. To test the ability of these significant probe sets to truly distinguish between the classes, leave-one-out-cross-validation (LOOCV) studies were preformed. In these LOOCV studies each array was left out in turn and a classifier was derived between the groups by selecting probe sets significant at p < 0.001. The significant probe sets were then used with several prediction models (compound covariate predictor, nearest neighbor predictor, and support vector machine predictor) to predict the class identity of the array that was left out and not included when the classification model was built. The significance (p<0.001) of the LOOCV analysis was estimated using a Monte Carlo simulation with 2,000 permutations of the dataset. KEGG pathways were populated using Pathway Express (Khatri et al., 2005), available at http://vortex.cs.wayne.edu/projects.htm
