gingivalis version 1 array was placed on top. Hybridization was performed at 65°C for 24 h and 10 RPM in a hybridization oven (G2545A, Agilent Technologies). After the hybridization the backings were removed in LSW (2 × SSC, 0.1% Sarkosyl (L9150, Sigma-Aldrich) at room temperature, washed for 5 min at 42°C in LSW, washed for 10 min at room

temperature in HSW (0.1 × SSC, 0.1% Sarkosyl) and finally washed for 1 min at room temperature in FW (0.1 × SSC). Each array was dipped 5 times in H2O and quickly submerged in isopropanol. Microarrays were spun dry for 1 min at 232 × g and scanned on an Agilent G2505B scanner at 5 μm resolution and data was extracted with Feature Extraction version 9.5.3.1. (Protocol GE2-NonAT_95_Feb07). Experimental design and Microarray data analysis Each strain was cultured in triplicate, in three experimental batches. Selleckchem BVD-523 DNA isolations and hybridizations were therefore performed three times for each strain, each being a biological replicate analyzed in one experimental block. On each array four technical replicate spots were spotted. After log2 transformation, the data was normalized by a global Lowess smoothing click here procedure, omitting the probes with highly divergent intensities because of the bias they induced. A mixed ANOVA model (as described in [61]) with

Bafilomycin A1 mouse group-means-parameterization was used to normalize the data and collapse the technical and biological replicates. The gene specific model was: O-methylated flavonoid (1) y ijklmn represents log2 expression intensities, μ is the gene specific mean, τ represents fixed strain effects

(i = 1, …, 8), ρ is an indicator variable indicating the common reference, S represents random spot effects (j = 1, …, 96), A represents random array effects (i = 1, …, 24), and B represents experimental batch effects (m = 1, …, 3). Normalized average (Cy5) intensities for each strain were calculated as y i * = μ + τ i and normalized average log2-ratio’s with respect to W83 were calculated as Y i * = τ i – τ 1 , for each i ≠ 1 (which represents W83). Hence, each strain was compared with W83, and deviations in log2-ratio’s were interpreted as aberrations. Given j genes divergence from zero were modelled as posterior probabilities of change under a mixture model, where non-divergent Y ij * ~ N(0,s i 2) and divergent Y ij * follows a uniform distribution [62]. Highly variable regions due to mutations or loss were quantified according to [63], using their GLAD (Gain and Loss Analysis of DNA) package with default parameter settings. Finally, we used the negative control probes from Arabidopsis thaliana to define absent calls with the aim to quantify whether an aberration was found more likely due to mutation or loss. The distributions of intensities suggested a distinguishable mixed distribution of intensities from probes interrogating present genes (high) and probes interrogating absent genes (low; Figure 1).