SUPPLEMENTAL METHODS
The authors wish to take this opportunity to provide additional details and contextual information pertaining to the studies desribed herein.
Motivation and Study Design Overview
We conducted an expression microarray experiment in order to identify the transcriptional changes associated with the reduced rate of metastasis formation by cells ectopically expressing MKK4 [1]. Each mouse was injected with SKOV3ip.1 human ovarian cancer cells which were engineered to ectopically express either HA-tagged MKK4 (SKOV3ip.1-HA-MKK4 cells) or an empty vector as a control (SKOV3ip.1-pLNCX2 cells). Five omenta were harvested at each of the timepoints indicated in Figure 1 Panel A, and RNA was extracted from cancer cells collected from lesions by LCM (depicted in Figure 1 Panel B). RNA samples were then amplified, labeled, assessed for quality (via Agilent Bioanalyzer and gel electrophoresis) and finally hybridized to Affymetrix HG-U133 Plus 2.0 microarrays (see below). The specific experimental methods employed for each of these procedures is discussed in detail in the following sections.
As shown in Figure 1 Panel A, the complete experiment consists of 34 expression arrays corresponding to five biological replicates of SKOV3ip.1-pLNCX2 and SKOV3ip.1-HA-MKK4 lesions at 3 dpi and 14 dpi, five biological replicates of SKOV3ip.1-pLNCX2 lesions at the vector endpoint (43 dpi), three biological replicates of SKOV3ip.1-HA-MKK4 lesions at the MKK4-expressing endpoint (65 dpi), and finally three in vitro replicates of the respective SKOV3ip.1-HA-MKK4 and SKOV3ip.1-pLNCX2 cell lines.
Cell Culture
The human ovarian carcinoma cell line SKOV3ip.1 was grown in DMEM containing L-glutamine (584 mg/L) and glucose (4.5 g/L; Mediatech, Herndon, VA), supplemented with 5% FCS (Atlanta Biologicals, Norcross, GA), 1% penicillin (100 units/mL)/streptomycin (100 Ag/mL) mixture, 10 mmol/L sodium pyruvate, 1x nonessential amino acids, and 2x MEM vitamin solution (all from Mediatech). All cell lines were a gift from Dr. Gordon Mills, M.D. Anderson Cancer Center, Houston, TX, and were maintained at 37°C in 5% CO2.
The development of clonal cell lines has been previously described [2, 3].In brief, the coding region of MKK4 (termed JNKK1) was provided by our collaborator Dr. Anning Lin of tthe University of Chicago [4]. The Hemagglutinin (HA)-tagged MKK4 coding region was subcloned into the pLNCX2 (Clontech) expression vector. Infectious, replication incompetent retroviruses carrying either the pLNCX2 empty vector or pLNCX2-HA-MKK4 construct were genenerated by transient transfection of Retro Pack PT67 packing cells using Effectene transfection reagent (Qiagen, Santa Clarita, CA). Conditioned medium containing viral particles was collected 48 hours later and passed through a 0.45 µm filter. SKOV3ip.1 cells (2 x 106) were plated 24 hours before infection and incubated for 24 hours with filtered virus containing 8 mg/mL Polybrene (SpecialtyMedia, Phillipsburg, NJ). For selection and establishment of stable SKOV3ip.1 clones, the above medium was additionally supplemented with 500 mg/mL (active concentration) of G418 sulfate (Mediatech). Transgene expression was confirmed in clonal cell lines by immunoblotting for expression of the HA tag using the HA.11 antibody from Covance (catalog #MMS-101P) at a dilution of 1:1000.
.
Animals and Experimental Metastasis Assays
All animal procedures were carried out in accordance with institutional animal research guidelines. For experimental metastasis assays, 6-8 week old female Hsd: Athymic Nude-Fox1nu mice were obtained from Harlan Laboratories (Madison, WI). Animals were injected intraperitoneally (i.p.) with 1 x 106 cells prepared in sterile phosphate-buffered saline (PBS). Prior to injection, cells were cultured for three days to approximately 80% confluence, and then trypsinized, washed, and resuspended in cold PBS at a final concentration of 2 x 106 cells/mL. In parallel, a protein lysate was extracted from an aliquot of each cell culture and ectopic expression of HA-MKK4 confirmed via immunoblotting (see below). Injected mice were weighed and observed 2-3 times per week, and sacrificed at the timepoints indicated in the study design by CO2 asphyxiation.
Immunoblotting
To verify expression of HA-MKK4, lysates from cells used for experimental metastasis assays were immunoblotted with anti-HA antibody (Co-Vance) at 1:1000 dilution for 1 hour followed by anti-mouse IgG secondary antibody (Sigma) at 1:5000 dilution for 1 hour (Supplemental Figure S1, Panel A). Membranes were stripped and re-probed for β-actin as a loading control using anti-actin Ab-1 antibody (Calbiochem) at 1:10,000 for 1 hour followed by goat anti-mouse IgM secondary antibody at 1:20,000 for 1 hour. Membranes were developed using SuperSignaling West Femto substrate (ThermoScientific) and developed on Thermo Pierce CL-Xposure film (Fisher).
Tissue Collection and Laser Capture Microdissection
The omenta and pancreas of sacrificied animals were embedded in Tissue-Tek optimal cutting temperature (OCT) compound (Sakura, Torrance, CA) and frozen in liquid nitrogen within 1 hour after sacrifice. Tissue samples were then kept at -80°C until use. For Laser Capture Microdissection, 30-50 8 µm sections per sample were prepared on PEN-membrane-coated slides (W. Nuhsbaum, Inc. ) and kept at -80°C overnight. Slides were individually brought to room temperature and stained rapidly with Hematoxylin and Eosin Y (both from Surgipath) using the following procedure: 70% Ethanol (30 dips), water (wash), Hematoxylin (30 dips), water (wash), Eosin (10 dips), 95% Ethanol (10 dips), 100 % Ethanol (10 dips). All water and ethanol solutions were prepared with Diethyl pyrocarbonate (DEPC)-treated water (Sigma). Water wash solutions were changed after each slide. Ethanol solutions were changed every 5-10 slides. Hematoxylin and Eosin were changed each day. Staining procedures were carried out in Falcon 50-mL polypropylene centrifuge tubes (Fisher Scientific) that were discarded at the end of each day.
UV-assisted Laser capture microdissection (LCM) on a Leica AS LMD 6000 was used to isolate cells of interest from the surrounding omental tissue. An average of approximately 40,000 cells were collected (range of 9,748 to 115,328 cells, 2-36 x 106 µm2) were collected from each sample into RLT lysis buffer (Qiagen) containing b-mercaptoethanol in a Nuclease-free thin-walled 0.5 mL collection tube (Labsource). RNA was extracted the same day using RNEasy Micro (Qiagen) according to the manufacturer’s instructions under RNAse-free conditions. The optional 5-minute on-column treatment with DNAse I (Qiagen) was performed to remove contaminating genomic DNA from the RNA sample. Total RNA was eluted in RNAse free water (Qiagen). Samples were aliquoted for analysis and stored at -80°C until further use.
RNA quality control Analysis
To determine that cryosection and staining conditions were suitable for extraction of intact RNA, spleen tissue from an untreated nude mouse was embedded, stored, and sectioned as above, except the sections were placed onto precleaned uncharged glass slides (Fisher Scientific). Individual 5 µm-thick sections were scraped directly into 1.5 mL Trizol (Gibco) in a 15-mL polypropelene centrifuge tube and vortexed at top speed for 30 seconds. RNA was extracted by adding 0.3 mL chloroform (Sigma). After centrifugation, RNA was precipitated from the aqueous phase with 0.75 mL isopropanol, and the pellet washed with 70% Ethanol made with DEPC-treated water. RNA was resuspended in 20 µL of molecular grade water. Two µL of 10X nucleic acid loading dye was added to 18 µL of each sample, and the samples were run on a 1% Tris-Acetate EDTA (TAE) gel and visualized by staining with Ethidium Bromide (0.2 µg/mL final concentration in gel) as a basic quality control (Supplemental Figure S1, Panel B). TAE gel and buffers were made with DEPC-treated water.
RNA Extraction and Amplification
RNA from microdissected samples was extracted using RNEasy Micro kit (Qiagen) under RNAse-free conditions and treated with DNAse I. RNA from cultured cells was extracted from 5x105 cells of each type by adding 350 µL of RLT buffer supplemented with β-Mercaptoethanol to cells in three wells of a 12-well tissue culture plate (Fisher Scientific). Cell lysates of each cell type were were homogenized by passing them through a 20-gauge needle five times and then aliquoted into three tubes and extracted in triplicate using RNEasy Micro. RNA was eluted in nuclease free water (Qiagen) and stored at -80°C until further use.
After all total RNA samples had been collected and extracted, one round of whole genome amplification was performed using Arcturus RiboAmp Plus (Molecular Devices, MDS, Sunnyvale, CA) RNA Amplification kit using 8 hours for in vitro transcription. The quality of both the total RNA and amplified RNA were determined using the Agilent Bioanalyzer RNA Pico or Nano platforms, respectively (Supplemental Figure S1, Panel C), and concentration measured with a Nanodrop (Thermo Scientific) to ensure equal loading onto the arrays. Only intact samples were used for hybridization. Amplified samples were biotin-labeled, sheared and hybridized onto Affymetrix HG U133 Plus 2.0 Arrays. Amplification and array hybridization were performed at two separate times, and all arrays were from the same manufacturer batch. Five samples were used for an initial pilot study, and the remainder of the samples (29) were amplified and hybridized at a later time. Amplification and hybridization was carried out in a randomized fashion to avoid introduction of sample-order bias. Samples for the pilot study were amplified with Arcturus RiboAmp RNA Amplification Kit (Molecular Devices), an earlier version of the same kit. No evidence for a batch effect was observed during correlation, MA, and other quality control analyses (see below), and all arrays were combined for the final analysis.
Array processing and Quality Control
We were initially concerned that a large degree of between-individual variation might exist in this experiment, as a result of the asynchronous development of metastases within each animal combined with the LCM methodology for isolating cells of interest from omental tissue. To address this we conducted a thorough quality control analysis of the arrays to assess the degree of between-individual variation observed within the experiment and to inform the methodology used to identify gene expression differences correlated with ectopic expression of MKK4.
All of the arrays were hybridized and scanned at the Functional Genomics Facility (The University of Chicago) according to the manufacturer’s protocols, and all processing and analysis was conducted in R (http://www.r-project.org). The raw CEL files were background corrected using the normexp method implemented in the package RMA [5], and then normalized and summarized using the Invariant Set method [6]. A model-based normalization method was used to allow for the possibility that the overall transcriptional effect of ectopic MKK4 is large and that some of the arrays could have distinct intensity distributions that would be inappropriately corrected with a standard quantile normalizaion. All background correction, normalization, and summarization methods are implemented in the R package affy [7]. As shown in Supplemental Fig. 1, Panel D2 MKK4 expression estimates derived from the microarray data show persistent up-regulation of MKK4 through out the experiment.
We first looked at broad signatures of the raw and normalized data to assess the consistency of the replicated arrays and to verify that there were minimal batch effects present in our data. Boxplots characterizing the raw and normalized intensity distributions are shown in Supplemental Figure S2. As expected, the overall distributions of the probe intensities on each array are comparable, and that these distributions converge (albeit slightly) as a result of normalization. It should be noted that the normalized distributions are not identical as a consequence of the Invariant Set normalization method, which only uses a subset of “invariant” probes to normalize arrays. This approach is useful when there is not a strong prior assumption about the underlying similarity of the intensity distributions across experimental conditions. Nevertheless, further investigations using other normalization methods (quantile, MAS 5.0) indicated that the major results presented here are robust to the particular normalization methodology (data not shown).
We further investigated the quality of our data by generating MA plots comparing each pair of replicated arrays within each condition (Supplemental Figure S4). In these plots, each point represents the intensities of a single probe measured on two arrays, with the y-axis coordinate representing the log2 difference between two observed intensities, and the x-axis coordinate corresponding to their mean. Ideally, points should lie on the M=0 axis, where deviations correspond to variable probe performance between replicates or normalization artifacts. A loess curve is plotted through each MA plot to indicate whatever intensity-based effects may exist on the arrays, and the median difference in probe intensity and interquartile ranges are reported for each comparison. In each case the medians are very close to zero and the interquartile ranges, which represent the dispersion observed in the middle 50% of the M values, are typically small (0.3-0.4 except at the endpoints, where the metastases are large and might be expected to exhibit increased dissimilarity).
A third, less precise way to assess the quality of our findings is to investigate the correlation structure of the data. We generated pairwise Spearman Rank Correlation coefficients using the expression estimates for each array pair, and compared these statistics across and within array conditions (Supplemental Figure S3). The pairwise correlations were uniformly high across all arrays, consistent with the high data quality observed in the MA plots and suggesting that the overall transcriptional effect of MKK4 activation is generally small relative to overall intensity distributions observed on the arrays (see Supplementary Table ST1). The highest correlations were observed between replicate arrays, followed by correlations between arrays within the same time point, then arrays within the same cell type, with the lowest correlations observed between arrays from discordant timepoints and cell types.
Transcriptional Characterization of Cell Lines and Mouse Effects
Once we were satisfied that our data was internally consistent and of high quality, we investigated the overall transcriptional variability that exists between SKOV3ip.1-HA-MKK4 and SKOV3ip.1-pLNCX2 cell lines grown in vitro using a two-sample t-test, as discussed in the main text. As expected, the overall p-value distribution was generally uniform and no genes were found to be differentially expressed (FDR = 0.1), indicating that the overall transcriptional
difference between the two cell lines in culture is small despite the ectopic expression of HA-MKK4.
The finding that SKOV3ip.1-pLNCX2 cells and SKOV3ip.1-HA-MKK4 cells grown in vitro were highly similar at the transcriptional level is consistent with previous studies showing that these cell lines have similar in vitro doubling times and MKK4 is not activated during in vitro growth. Indeed, the suppressive effects of MKK4 can only be detected in vivo. To evaluate the transcriptional profiles of cells in vivo, we fit the following mixed-effects linear model to the summarized expression values observed for each probeset in a balanced subset of the expression arrays, consisting of the six arrays hybridized with RNA isolated from the cell cultures and three randomly-selected arrays from the day 3 timepoint from each of the injected cell lines: