The Bordetella pertussis strains, including the virulent wild type laboratory strain Bp3381, the clinical isolate Minnesota1 (gift from Minnesota State Department of Health), and isogenic mutant strains2 derived from Bp338, Bp5373, BpTox64, and BpA2-65 were grown at 37°C on Bordet-Gengou agar (Difco), supplemented with sheep blood (13% v/v), and then transferred to Stainer-Scholte liquid media6 and grown to late logarithmic phase. For experiments with heat-killed bacteria, the liquid culture was heated to 56°C for 30 minutes. The other bacterial strains, an E. coli clinical isolate and two Staphylococcus aureus clinical isolates, were isolated from patient blood cultures (VAPAHCS clinical lab) and were grown in the same medium, and heat-killed for parallel experiments in the same manner, as the Bordetella strains. Bordetella pertussis lipopolysaccharide (LPS) (List Biologicals), phorbal 12-myristate 13-acetate (PMA) (Sigma), and ionomycin (Sigma) were diluted in fresh Stainer-Scholte media for treatment time courses.
Human primary peripheral blood mononuclear cells (PBMCs), including all circulating non-red cell and non-granulocyte populations (primarily lymphocytes and monocytes), were acquired by apheresis or from whole blood from healthy donors, and purified using endotoxin tested Ficoll-Paque PLUS (Amersham Pharmacia). Before stimulation, cells were cultured overnight at a density of 2x107 per ml in RPMI 1640 supplemented with 10% endotoxin-free fetal bovine serum (Life Technologies). Cells of the U937 monocyte/macrophage cell line (ATCC) were also maintained in RPMI 1640 supplemented with 10% fetal bovine serum and were induced to differentiate into macrophage-like cells with 10ng/ml PMA for 48hrs, washed with phosphate-buffered saline, and incubated in RPMI without PMA for an additional 24hrs before infection.
After heat-killing, bacteria were diluted in their original growth media and added to 1x108 PBMCs at a ratio of 4 to 0.002 bacteria per cell, as described in the figure legends. LPS was added to the same number of PBMCs at a concentration ranging from 1mg/ml to 0.01mg/ml and ionomycin was added at 1mM in combination with PMA at 25ng/ml. For each of the two large sets of parallel time courses, at least 4 replicate samples were harvested at the beginning of treatment (0 hour) and subsequent samples were harvested, after incubation at 37°C, at 0.5, 1, 2, 4, 6, 12 and 24 hours. For the comparison of live versus heat-killed B. pertussis, the bacterial culture was split, half of the bacteria heat-killed and added to PBMCs, the other half added live to PBMCs; samples were harvested at 0, 0.5, 2, 4, 6, and 12 hours. At harvesting, cells were scraped from the flask, diluted in cold PBS, pelleted at 400 x g for 5 minutes, and snap-frozen on liquid nitrogen. U937 cells were infected at a concentration of 50 live bacteria/cell for all time courses with Bordetella mutant and wild type strains.
Poly-(A)+ mRNA was extracted from cell pellets with the FastTrack 2.0 kit (Invitrogen). Poly-(A)+ mRNA from samples for the data set comparing the live versus heat-killed bacteria was amplified according to the method of Wang et al.7. The microarrays employed for the bacterial diversity and dose response data sets comprised 18,432 cDNAs, representing ~7,619 unique genes (estimated based on NCBI UniGene build #136), and were constructed as previously described8. Fluorescently-labeled cDNA probes were prepared by incorporation of Cy3- and Cy5-conjugated dUTP during cDNA synthesis using a standard reference pool of mRNA and the experimental sample of mRNA, respectively. Equal amounts of the two probes were pooled and hybridized to microarrays. The reference pool of mRNA was prepared from several immune cell lines (U937, THP-1, HL-60, VDSO, MonoMac6, and Jurkat) under a variety of stimulatory conditions, including PMA treatment and bacterial infection. Comparison of all experimental samples to the same reference allowed the relative expression level of each gene to be compared across all of the experiments9. Hybridized arrays were scanned using a GenePix 4000A microarray scanner (Axon Instruments) and the images analyzed and tabulated with the GenePix Pro software package.
Microarray elements (i.e., spots representing unique arrayed cDNA clones) were considered for analysis in two stages. In the first stage, we applied a high stringency data selection, allowing only elements for which at least half the measurements within a set of experiments had fluorescence intensity in both channels at least 4-fold over background intensity. In the second stage, additional array elements with low signal intensity but high quality were selected, if their expression profile had a correlation of at least 0.7 to any gene selected in the first stage. Only cDNAs that met these criteria at 80% or more of the measured time points across all of the experiments were used in the analysis. Measurements from the multiple samples taken at the t=0, pre-treatment time point for each time course hour were averaged, with the exclusion of those measurements that fell beyond 1 standard deviation from the mean. This average t=0 measurement was then subtracted from each subsequent time point measurement in order to depict the temporal response patterns of expression relative to t=0 as the baseline. We then filtered the data based on the variation of each element from the baseline across all experiments, as described in the figure legends. Hierarchical clusters and self-organizing maps were generated by the Cluster program and analyzed with the TreeView program (M. Eisen, http://www.microarrays.org/software)10. For the comparison of live versus killed bacterial stimuli, the Euclidean distance was calculated as , where Xlive,i and Xkilled,i represent log2 of the ratio of expression level of a given gene at time ti , in the live and killed time series, respectively (ti= 0.5, 2, 4, 6, 12 hours.
The time averaged induction was calculated from the area under the curve (AUC) for expression time courses, by normalizing the AUC over the total time of exposure for each condition (i.e.,). The AUC was calculated using the trapezoid method, based on the relationship , where ti represents exposure time, and Xi represents the log2 of the ratio of expression level at time ti (ti= 0, 0.5, 2, 4, 6, 12 hours for PBMC, and ti= 0, 0.5, 1, 2, 4, 6 hours for U937 experiments).
We used the following function to model simultaneously the effects of bacterial type, treatment dose, and exposure time on the expression response: the variables b, d, and t denote, respectively, bacterial type (b=0 for B. pertussis and b=1 for S. aureus), log10 treatment dose (d= 0, 1, 2, 3), and exposure time (t= 0.5, 2, 4, 6, 12 hours, omitting the 1- and 24-hour time points, which were measured only for the 100X dose). For a particular gene, let represent the expected gene expression response for given levels of the three factors (Equation 1). This model specifies a linear log-dose response and a quadratic time response, and also allows for 2-way interactions between each of the factors. Using the model in Equation 1, the difference in the expected response for the two types of bacteria is for a given treatment dose d and exposure time t (Equation 2). We used the method of least squares to fit the function in Equation 1 to the zero-transformed expression data of a subset of genes with multiple representations on the array. For each gene, this method identified parameter values , which minimize the sum of squared deviations between the observed responses and the fitted responses.
Using the bacterial diversity data set, 1208 unique genes were selected based on having a LocusLink assignment ((http://www.ncbi.nlm.nih.gov/LocusLink). p-values for the frequency of 184 GO annotation terms (those that had at least 5 occurrences in the data set) were calculated based on the hypergeometric distribution. Those annotation terms found to have p-values less than 1% within the Common Induction (red) and Common Repression (green) clusters are displayed with their GO identification numbers and example genes from the 1208-gene data set.
The Gene Ontology annotations for genes in the entire bacterial diversity data set were extracted in batch from SOURCE (http://genome-www.stanford.edu/source), using the Locus Link identification number for each gene. The resulting 1208 unique annotated genes were used for the analysis, among which 184 annotation terms occurred for at least 5 genes. We compared the frequency of these 184 annotations in the data set as a whole to that in the common induction and common repression clusters, using the hypergeometric distribution to calculate p-values (Tavazoie, 1999 Nature Genetics) (see equation below). Let N=1208 denote the total number of genes under consideration and A the number of these genes with a particular annotation. The chance of observing at least x genes with that annotation in a random subset of n genes is given by , where .
1. Stibitz, S., Aaronson, W., Monack, D. & Falkow, S. Phase variation in Bordetella pertussis by frameshift mutation in a gene for a novel two-component system. Nature 338, 266-9. (1989).
2. Boschwitz, J. S., Batanghari, J. W., Kedem, H. & Relman, D. A. Bordetella pertussis infection of human monocytes inhibits antigen- dependent CD4 T cell proliferation. J Infect Dis 176, 678-86. (1997).
3. Relman, D. A., Domenighini, M., Tuomanen, E., Rappuoli, R. & Falkow, S. Filamentous hemagglutinin of Bordetella pertussis: nucleotide sequence and crucial role in adherence. Proc Natl Acad Sci U S A 86, 2637-41. (1989).
4. Relman, D. et al. Recognition of a bacterial adhesion by an integrin: macrophage CR3 (alpha M beta 2, CD11b/CD18) binds filamentous hemagglutinin of Bordetella pertussis. Cell 61, 1375-82. (1990).
5. Gross, M. K., Au, D. C., Smith, A. L. & Storm, D. R. Targeted mutations that ablate either the adenylate cyclase or hemolysin function of the bifunctional cyaA toxin of Bordetella pertussis abolish virulence. Proc Natl Acad Sci U S A 89, 4898-902. (1992).
6. Stainer, D. W. & Scholte, M. J. A simple chemically defined medium for the production of phase I Bordetella pertussis. J Gen Microbiol 63, 211-20. (1970).
7. Wang, E., Miller, L. D., Ohnmacht, G. A., Liu, E. T. & Marincola, F. M. High-fidelity mRNA amplification for gene profiling. Nat Biotechnol 18, 457-9. (2000).
8. Alizadeh, A. et al. The lymphochip: a specialized cDNA microarray for the genomic-scale analysis of gene expression in normal and malignant lymphocytes. Cold Spring Harb Symp Quant Biol 64, 71-8 (1999).
9. Alizadeh, A. A. et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503-11 (2000).
10. Eisen, M. B., Spellman, P. T., Brown, P. O. & Botstein, D. Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 95, 14863-8. (1998).