Acute lung injury is a complex illness with a high mortality rate (>30%) and often requires the use of mechanical ventilatory support for respiratory failure. Mechanical ventilation can lead to clinical deterioration due to augmented lung injury in certain patients, suggesting the potential existence of genetic susceptibility to mechanical stretch (6, 48), the nature of which remains unclear. To identify genes affected by ventilator-induced lung injury (VILI), we utilized a bioinformatic-intense candidate gene approach and examined gene expression profiles from rodent VILI models (mouse and rat) using the oligonucleotide microarray platform. To increase statistical power of gene expression analysis, 2,769 mouse/rat orthologous genes identified on RG_U34A and MG_U74Av2 arrays were simultaneously analyzed by significance analysis of microarrays (SAM). This combined ortholog/SAM approach identified 41 up- and 7 downregulated VILI-related candidate genes, results validated by comparable expression levels obtained by either real-time or relative RT-PCR for 15 randomly selected genes. K-mean clustering of 48 VILI-related genes clustered several well-known VILI-associated genes (IL-6, plasminogen activator inhibitor type 1, CCL-2, cyclooxygenase-2) with a number of stress-related genes (Myc, Cyr61, Socs3). The only unannotated member of this cluster (n = 14) was RIKEN_1300002F13 EST, an ortholog of the stress-related Gene33/Mig-6 gene. The further evaluation of this candidate strongly suggested its involvement in development of VILI. We speculate that the ortholog-SAM approach is a useful, time- and resource-efficient tool for identification of candidate genes in a variety of complex disease models such as VILI.
- rodent mechanical ventilation model
acute lung injury (ALI) and acute respiratory distress syndrome (ARDS) have been recognized for more than 30 years, and despite major advances in critical care, the mortality rate of ALI/ARDS remains unacceptably high (30–40%). Mechanical ventilation (MV) remains the mainstay of treatment; however, the ARDSnet reports have now confirmed that MV may increase mortality and morbidity due to overextension of lung “volutrauma” associated with the subsequent release of inflammatory mediators (2, 30). A series of animal and human studies has demonstrated the detrimental effects of aggressive ventilation strategies (38, 39, 42). Tremblay et al. (41) reported that ventilation with high-end inspiratory lung tidal volumes of 15 or 40 ml/kg induces lung injury and increases expression of ALI-related genes (TNF-α and IL-6) in pulmonary epithelial cells of ventilated isolated rat lungs. Human studies confirmed that a significant reduction in mortality of mechanically ventilated patients can be achieved by reducing tidal volume from the conventional 12 ml/kg to 6 ml/kg (1). However, the mechanism(s) and the genetic susceptibility of the injurious effects of MV remained unclear.
To investigate the deleterious effects of MV on lung gene expression, multiple animal models of mechanical ventilation-induced lung injury (VILI) have been developed (7). Rodent VILI models (e.g., mouse and rat) are preferred because of the economic feasibility of the studies and the convenient way to administer various specific tidal volumes that are critical in VILI (36). The discovery of heritable differences in the physiological and pathological responses to variety of lung injuries in rodent (20, 29, 46, 49) combined with the availability of mouse and rat gene array platforms increases the appeal of rodent models for genome-wide analyses (22, 51) in the search for genetic factors that underlie VILI.
Global gene expression platforms such as oligonucleotide microarrays are robust techniques for identification of differentially regulated genes that serve as novel markers and potential therapeutic targets in a large variety of human diseases. We recently demonstrated that microarray-driven gene expression profiling followed by gene ontology (GO) analysis is an efficient approach in the evaluation of biological processes of interest and selection of process-related candidate genes in VILI (12, 13). However, the wide application of oligonucleotide microarray approach is limited by the extensive resources required to perform this work. To address potential utilization of fewer numbers of microarrays, the significance analysis of microarrays (SAM) (http://www-stat.stanford.edu/∼tibs/SAM) method was developed by Tusher et al. (44) and applied to analysis of eight expression profiles. This approach demonstrated that even with minimal resources (i.e., 2 controls and 2 experimental radiation treatments of 2 cell lines), biologically plausible results can be produced. SAM not only identified radiation-induced genes involved in cell cycle regulation and apoptosis but also identified novel candidate genes involved in DNA repair (44). Based on these studies, we hypothesized that the SAM method could be duplicated using eight gene expression profiles generated from two rodent VILI models. Moreover, we speculated that these models can be represented by different species’ (mouse and rat) genetic information of which will be linked using an orthologous gene approach (13). To test this hypothesis and explore the effects of MV on gene expression in lung tissues, we utilized well-established mouse and rat VILI models. Mouse and rat exposed to excessive tidal volumes were designed according to the corresponding species physiology and responses to MV. Gene expression profiles generated from mouse and rat lung tissues were linked by orthologous genes as we described previously (13). The resulting ortholog-linked profiles were analyzed using the SAM method, which assigned a score to each gene on the basis of change in gene expression relative to the standard deviation of repeated measurements. For genes with scores greater than an adjustable threshold, SAM uses permutations of the repeated measurements to estimate the percentage of genes identified by chance, the false discovery rate (FDR). In the MV response of rodent models, SAM identified 41 upregulated and 7 downregulated genes that changed at least 1.7-fold with an estimated FDR of <10% (Fig. 1), results validated by real-time or relative RT-PCR in both species. The subsequent GO analysis of candidate genes not only confirmed involvement of the well-known ALI-related biological processes, such as inflammatory and immune responses and regulation of transcription and blood coagulation, etc. (12, 13), but also revealed novel biological processes (cell cycle arrest, cell-cell signaling, chemotaxis, negative regulation of cell proliferation) of potential relevance to VILI. The current study offers a novel bioinformatic approach for the identification of novel targets in VILI, which may provide insight into the mechanisms through which MV exacerbates the pathobiology of ALI.
Animal preparation and treatments.
All experiments were approved by the Johns Hopkins Animal Care and Use Committee and the Johns Hopkins Office of Health, Safety, and Environment. Male C57BL/6J mice (weighing ∼100–120 g) and male Wistar (CRL:WI) rats (∼8 wk of age weighing ∼200 g) were purchased from the Jackson Laboratory and Charles River Laboratory and housed as described previously (13). Mice were exposed to large tidal volume [17 ml/kg, 6 h or 35 ml/kg, 2 h (n = 6)] as described previously (13, 26, 50) using a small animal mechanical ventilator. Nonmechanically ventilated, spontaneous breathing mice were used as controls (n = 6). The severity of lung injury was confirmed by the increases of bronchoalveolar lavage protein concentration (130 ± 46 μg/ml, P < 0.05, ventilation vs. control). In separate experiments, mice were treated with LPS intratracheally (2.5 mg/kg) for 3, 6, or 24 h (n = 3 for each time point). To investigate the combined effect of LPS and MV, mice (n = 3) were treated with LPS (2.5 mg/kg) 24 h before MV (35 ml/kg, 2 h). Rats (n = 5) were mechanically ventilated (12 ml/kg, 5 h) with room air, and the control group (n = 5) spontaneously respired through identical tracheal cannulas without MV. During the experiments, arterial blood pressure, airway pressure, body temperature, and respiratory rate were monitored and recorded as described previously [www.hopkins-genomics.org (13)]. At the end of the experiment, both mice and rats were killed, and the excised lungs were snap-frozen and stored at −80°C as described previously (13). Mouse and rat lung tissues from each group were used for microarray analysis (n = 2) and RT-PCR validation (n = 3). The duration of MV was selected based on changes of bronchoalveolar lavage protein concentration and survival rate of ventilated mice and rats, respectively.
RNA isolation and expression profiling.
Affymetrix GeneChip Expression Analysis Manual protocols (Affymetrix, Santa Clara, CA) recommended by the manufacturer were followed as described previously (13). The signal intensity fluorescent images produced during Affymetrix GeneChip hybridizations were read using the Agilent Gene Array Scanner and converted into GeneChip Cell files (CEL) using MAS 5.0 software (Affymetrix).
Expression data filtering and analysis.
A total of eight CEL files generated from Affymetrix GeneChip (2 control and 2 MV per species) containing image intensities were analyzed by the robust microarray analysis (RMA) module of the Bioconductor package developed at Johns Hopkins University School of Public Health (17, 18). The expression measures of individual probe set were extracted, and background correction, across array normalization, and summarization were performed. A total of 2,769 mouse/rat orthologous genes were identified utilizing RESOURCERER 8.0 (http://pga.tigr.org/tigr-scripts/magic/r1.pl) (19, 43). Further analyses were restricted to these mouse/rat ortholog gene profiles, and their signal intensity values were log transformed and normalized to row-wise mean using CLUSTER tool (10). The global gene expression profile analysis of transformed expression data was conducted using the SAM method (44). Mouse and rat data sets were separated by experimental condition to corresponding blocks as recommended by the SAM procedure. Readers are referenced to the SAM online manual for detailed explanations of the blocking approach (http://www-stat.stanford.edu/∼tibs/SAM). User-definable conditions were set to default values and 1.7-fold change (Supplementary Fig. 1). TM4 Microarray Software Suite developed by TIGR (http://www.tigr.org/software/tm4/mev.html) was used to evaluate the similarity of gene expression profiles. Briefly, data were log transformed, mean centered, and visualized using the Multiexperiment Viewer 2.2 tool. By K-mean clustering analysis, six clusters illustrating genes upregulated in both mouse and rat (cluster_1); upregulated in mouse and marginally regulated in rat (cluster_2); marginally regulated in mouse and upregulated in rat (cluster_3); upregulated in mouse and marginally regulated in rat with smaller signal altitude (cluster_4); downregulated in mouse and marginally regulated in rat (cluster_5); and downregulated in both mouse and rat (cluster_6) were predicted. Relevance network analysis (4, 5) was conducted with threshold ranging between 0.9 and 0.95.
A total of 2,769 orthologous genes were dynamically linked to GO terms using GenMAPP (8) and MAPPFinder (9) tools. Because the mouse GO of GenMAPP is not yet complete, we converted mouse probe_IDs to their human counterparts using RESOURCERER 8.0. A total of 48 genes selected by SAM as significantly differentially regulated were labeled as VILI related. Functional categories of these genes were identified using MAPPFinder with filtered criteria of z-score >2 and “Number Changed Genes” >2. The linkage of selected candidate genes to corresponding biological pathways was conducted using TIGR GO database (ftp://ftp.tigr.org/pub/data/tgi/Resourcerer/Mouse) followed by manual literature curation using PubMed.
To confirm the transcript abundance determined by the Affymetrix platform, relative message levels of selected genes were measured using SuperScript One-Step RT-PCR with Platinum Taq system (Invitrogen, Carlsbad, CA), and 18S rRNA was employed as an internal standard. Specific primer pairs for each gene of interest based on GenBank sequence data were listed (Supplementary Tables S1-S5 are available at http://ajplung.physiology.org/cgi/content/full/00109.2005/DC1). Optimal cycle number, denaturation, annealing, and extension temperatures were determined empirically to achieve amplification within the linear range. Reaction products of control and MV samples for each gene were analyzed by agarose gel electrophoresis, and densitometric analysis was performed with ImageQuant software (Molecular Dynamics). Results were normalized to corresponding 18S rRNA intensity, differential changes were presented as a ratio of normalized MV vs. normalized control, and significance of the ratio was evaluated by unpaired t-test (P < 0.05).
Transcript levels of RIKEN cDNA 1300002F13 (mitogen-inducible gene 6-like or Mig-6) in mouse lungs subjected to either MV, LPS, or LPS+MV as described above were measured (n = 3 per condition) in 96-well microtiter plates with an ABI Prism 7700 Sequence Detector System (Perkin-Elmer/Applied Biosystems). TaqMan 18S rRNA control reagent was used as internal control for normalization. Primers and probes were purchased from Applied Biosystems (cat. no. Mm00505292_m1) in a 20× mixture. All experimental protocols were based on the manufacturer’s recommendation using the TaqMan Gold RT-PCR Core Reagents Kit (P/N 402876, Perkin-Elmer/Applied Biosystems). Experimental parameters were 48°C for 30 min followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. A relative quantitative method was used to analyze changes in gene expression in a given sample relative to an untreated control sample, and specific mRNA transcript levels were expressed as fold difference. ANOVA test was performed on experimental conditions described above followed by a t-test for an individual group. Bonferroni-corrected P < 0.008 was considered significant.
Established VILI ortholog database.
It is well appreciated that the severity of VILI is dependent on factors such as magnitude of tidal volume as well as the respiratory frequency and duration of ventilation. To synchronize effects of mechanical stretch in various animal models, we identified the terminal stage of VILI and analyzed gene expression profiles of injured lung tissues collected at time points where mortality rate was markedly increased (3–6 h for mouse model and 5 h for rat model) and lung tissue injury confirmed by pulmonary vascular leakage (significant increase in concentration of bronchoalveolar lavage proteins).
To further investigate gene expression profiles of the rodent VILI models, we employed the previously described ortholog linkage approach (13). Orthologous genes of Affymetrix GeneChips MG_U74Av2 and RG_U34A were identified using RESOURCERER 8.0 tool (43) and TIGR Eukaryotic Gene Ortholog database, and a total of 5,293 ortholog pairs, representing 2,769 genes, were identified (Supplementary Table S2). Within this ortholog database, ∼45% of genes were represented by more than one ortholog pair due to the presence of multiple paralogs and orthologs on the mouse and rat GeneChips. SAM was then applied to analyze the data where expression values for each ortholog pair were log transformed and normalized (Supplementary Table S3). A total of 41 significantly upregulated and 7 significantly downregulated genes were identified, where 30 genes were represented by more than one ortholog pair (Table 1). Candidate genes with multiple, concordantly expressed ortholog pairs were considered the most reliable candidates and selected as VILI candidate genes.
Filtering by GO profiling.
We further analyzed our candidate gene list by applying MAPPFinder, a tool that creates a global gene expression profile across all areas of biology by integrating the annotations of the GO Project with the GenMAPP software package (http://www.GenMAPP.org). The GO Consortium created a defined vocabulary of terms describing the biological processes, cellular components, and molecular functions of known genes that facilitates browsing through the complicated genetic network. GO analysis identified eight significantly affected biological processes (Fig. 1, Supplementary Table S4) including four bioprocesses (inflammatory response, blood coagulation, immune response, and apoptosis) that were previously linked to VILI in multispecies models (13), a result we believe validates our candidate gene approach. Differential expression of genes identified by the Affymetrix platform was then validated by relative RT-PCR (n = 15; Fig. 2). Although fold changes presented here generated by both techniques were all statistically significant, the overall fold changes detected by relative RT-PCR were lower than those identified by GeneChip. Hierarchical clustering of selected candidate genes separated control and stretched lung tissues independently of the species (Fig. 3), a finding that further supports our analytical approach and candidate gene selection.
Analysis of ortholog gene expression by K-mean clustering algorithm and gene relevance network.
To further characterize identified VILI-related candidate genes, we employed supervised K-mean clustering procedure that assigned 48 candidate genes to a predicted 6 clusters (Fig. 4A). Cluster_1, consisting of 14 genes, was of particular of interest as nearly 60% (8 of 14 genes) in this cluster were previously identified and associated with VILI (Fig. 4B, gene content of other clusters provided in Supplementary Table S5). Surprisingly, the only unannotated RIKEN cDNA 1300002F13 (Mig6) gene from our candidate gene list (Table 1) was a member of this cluster, suggesting its potential role in VILI. To investigate the relationship of Mig-6 with other candidate genes, relevance network analysis (http://www.tigr.org/software/tm4/mev.html) was conducted and identified a direct positive link of Mig-6 to VILI-related gene coding for plasminogen activator urokinase receptor (PLAUR) and myelocytomatosis oncogene (Myc). The negative correlation with mesenchyme homeobox 2 (Meox2) expression was also predicted by this analysis (Fig. 5). Ventilator-induced changes in Mig-6 expression identified with the Affymetrix platform were comparable with those detected by real-time RT-PCR and demonstrated that Mig-6 expression was significantly (P < 0.05) increased by approximately four- and sixfold, respectively (Fig. 6). Upregulation of Mig-6 was also observed in an LPS-lung injury model at early time points (Fig. 7). Noticeably, 24 h after LPS treatment, the expression of Mig-6 was at the control level, and, moreover, the MV was unable to affect Mig-6 expression.
The goal of the present studies was to extend our current understanding of the pathogenesis of VILI by selecting and validating novel ALI-related candidate genes using gene expression profiling. The manifestation of VILI is universal among mammals exposed to excessive mechanical stress, and, similar to other causes of ALI, characterized by increased lung weight gain reflecting pulmonary edema, a major feature of the ALI. Because there is a diverse response to vascular leakage in VILI in different rodent species, we were unable to synchronize injury in mouse and rat models based on vascular permeability characteristics. Instead, we chose to utilize mean survival time (28) to measure the injury severity, and, therefore, the injured lung tissues were collected at the preterminal time points. We hypothesized that excessive tidal volumes delivered by MV may lead to activation of genes responsible for production and release of proinflammatory factors and thus initiate an inflammatory cascade in the lung tissues, including production of chemical mediators and neutrophil infiltration. We speculated that global genome analysis of mouse and rat models will allow us to identify VILI-related genes and evolutionarily conserved bioprocesses in rodents even though the degree and magnitude of mechanical deformation observed with injurious forms of ventilation do not naturally occur. It has been shown that SAM method can be employed for the simultaneous analysis of different biological models exposed to the same stimulus (44). We adapted this method for the analysis of two rodent species exposed to the same stimulus (MV), speculating that this approach will allow us to identify functionally related genes in both species affected by MV. The mouse and rat gene expression profiles were first linked using the ortholog technique developed by our group (13) and then analyzed using SAM and GO software. This approach offers not only resource-efficient (4 microarrays per model) global genome analysis but also the advantage of the multispecies over the single-species analysis, which can identify the injury-related evolutionarily developed responses triggered by a nonevolutional stimulus such as VILI.
This ortholog approach identified 48 genes significantly affected by MV in both species. A number of identified genes represented the two major biological processes, inflammation and coagulation, involved in ALI (12, 13). The inflammatory response ontology was represented by inflammatory cytokines and factors including IL-1β, IL-6, C-C chemokine 2 (CCL-2), and COX-2, whereas blood coagulation ontology included plasminogen activator inhibitor type 1 (PAI-1), tissue factor (TF), and PLAUR. We also identified several novel genes with a potentially important role in VILI (ATF-3, GADD-45α, IL-1RII, and LAT) that have not yet been directly linked to VILI. Expression levels of candidate genes were validated by real-time (Fig. 6) and relative RT-PCRs (n = 15; Fig. 2), which confirmed reliability of our findings. Hierarchical clustering of candidate genes grouped gene expression profiles, based on the experimental condition rather than on species and clustered arrays of ventilated rat lungs with arrays of ventilated mouse lungs (Fig. 3), further validated our ortholog approach (Fig. 3). These results suggest that similarity in ventilator-affected changes in orthologous candidate gene expression between rat and mouse outweighs the species-specific gene expression differences. Subsequent K-mean clustering identified a particularly interesting cluster_1 consisting of 14 genes (Fig. 4A), including the most prominent ALI-related gene, IL-6 (Fig. 4B). Clinical studies have shown that IL-6 concentrations in bronchoalveolar lavage fluid (BALF) from patients with established ARDS is significantly higher than in BALF from normal volunteers (25), and the blood concentration of IL-6 is higher in patients at risk for ALI who subsequently developed ALI compared with patients at risk who do not develop ALI (40). Another ALI-related gene, the CCL-2, in this cluster plays an important role in multiple lung inflammatory disorders and ALI (31) as does the inflammation-related enzyme prostaglandin-endoperoxide synthase 2/COX-2. Ecosanoids appear to be important to edemagenesis in experimental ALI, and expression of this enzyme correlates with increased vascular permeability (11, 15). Cluster_1 also contains three members of the ALI-related GO bioprocess “blood coagulation” (Fig. 1, Supplementary Table S4). Increased levels of coagulation factor III (tissue factor) and PAI-1 were reported in patients with ALI (14, 27, 32) and VILI (35, 37), and PLAUR is upregulated by inflammatory cytokines (16).
Potential ALI-related genes, including growth arrest and DNA damage-inducible, alpha (13, 45), and amphiregulin (47), were also found in cluster_1. Therefore, cluster_1 was chosen as a primary source for novel ALI candidate gene selection. Surprisingly, most of the remaining genes in this cluster were implicated into stress response mechanisms. Beside well-known stress-responsive genes coding for myelocytomatosis oncogene (Myc), cluster_1 contained novel stress-associated genes, including suppressor of cytokine signaling 3 (Socs3) and cysteine-rich protein 61 (Cyr61), whose upregulation has been linked to biomechanical stress (34, 52). Given that the only unannotated gene in cluster_1 was RIKEN_1300002F13 EST, it became an attractive ALI-related candidate, and we investigated its expression pattern in more detail using real-time RT-PCR (Figs. 6 and 7). Sequence analysis revealed that this gene was a close ortholog of rat Gene33 and human Mig-6 and therefore was named mouse Mig-6-like gene (Mig-6). It has been reported previously that Gene33 is a stress-related gene, the expression of which is rapidly triggered during diabetic nephropathy (21) and experimental hypoxia (33), respectively. Makkinje et al. (21) have also shown that the Gene33 activation was followed by induction of stress-activated protein kinases (SAPKs). Although it has been shown that SAPKs can be activated by mechanical stress in the cardiovascular system and by LPS in bronchoalveolar cells (24), the activation of Mig-6/Gene33 expression by these stimuli in mouse lungs has not been reported. We hypothesized that mechanical stress and LPS can activate Mig-6 expression in our mouse models and that this gene can be a common transducer of signals generated by mechanical stress and LPS in mouse lungs. To test this hypothesis, we treated mice with LPS and measured the Mig-6 gene expression at different time points. The time-dependent decrease in expression levels of Mig-6 after the initial activation by LPS was observed with peak expression at the 3-h time point (Fig. 7A). Interestingly, 24 h after LPS treatment, the Mig-6 expression returned to its basal levels, and subsequent MV challenge lost its effect on Mig-6 expression (Fig. 7B). A similar LPS-induced cardiac cross-resistance to ischemia has been reported previously (23) and had shown that LPS preconditioning induces cardiac resistance to subsequent ischemia. Our findings suggested a similar LPS preconditioning of mouse lung-specific Mig-6 expression in response to MV. Additionally, we speculated that there could be a lung-specific pattern of Mig-6 expression in response to stressors with the unknown mechanism of Mig-6-negative regulation. The relevance network analysis suggested that Mig-6 can mediate not only stress-induced responses but also blood coagulation processes. We believe that further studies of this novel stress-related gene in mouse ALI models have great potential to unravel new mechanisms of lung injury and identify new targets for therapeutic intervention.
The studies presented here offer a new approach to candidate gene selection combining multiple genomics tools in one analytical streamline. Ortholog linking of rat and mouse Affymetrix platforms followed by RMA and SAM analyses allowed us to generate a reliable candidate gene list with minimal resources (4 gene arrays per model). Consequent analysis of identified candidate genes using clustering and relevance network techniques narrowed our search to potential ALI-related genes, including the as yet unannotated Mig-6 gene. Real-time RT-PCR analysis of this gene demonstrated that Mig-6 is a valid ALI-related target for further investigation, thus validating our novel candidate gene-searching technique. Progress in understanding disease heterogeneity through the use of evolving biological, genomic, and genetic approaches should provide major new insights into the pathogenesis and treatment of ALI.
This work is supported by National Institutes of Health/National Heart, Lung, and Blood Institute Grant P01 HL-69340 and Specialized Centers of Clinically Oriented Research P50 HL-073994.
We acknowledge the support of the Center for Translational Respiratory Medicine. We thank Dr. David Pearse and Ian Miller for invaluable efforts in preparing animals.
↵* S.-F. Ma and D. N. Grigoryev contributed equally to this work.
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
- Copyright © 2005 the American Physiological Society