When It Comes to Genes-IPF or NSIP, Familial or Sporadic-They're All the Same

Idiopathic pulmonary fibrosis (IPF) continues to intrigue pulmonary physicians and researchers. In the decades since the disease was first described, there have been changes reported in its definition (1), perceived mechanisms (2), and course (3). The availability of gene expression microarrays that allow simultaneous profiling of complete genomes promises to improve the understanding and classification of poorly understood diseases like IPF. In this issue of the Journal (pp. 45-54), Yang and coworkers (4) use microarrays to compare sporadic and familial IPF.

We recently contrasted two distinct approaches to analysis of microarray data: a "cherry picking" reductionist approach and a global "systems biology" approach (5). In the first, the results of a microarray experiment are used to select a few "cute" genes. Their biological relevance is validated by confirming that changes are also translated to alterations in the protein encoded by the gene and by experiments designed to test the role of the protein in fibrosis-relevant models. In the systems approach, investigators use the information available in microarray data and advanced computational tools to gain systems-level insights on disease. This could be by highlighting new pathways involved in a disease, identification of new classes of disease, or by creating models of disease progression and phenotype formation.

The article by Yang and colleagues provides an outstanding example of a combined approach. Although some important global observations are provided, the authors also pick CXCL12 to follow up using immunohistochemistry and mice lacking its receptor (4). Although the article is both compelling and intellectually stimulating, the reader is left with the need to decide what the main conclusions actually are. This combined approach, though compelling and intellectually intriguing, leaves the reader with the need to decide what the main conclusions actually are. In our view, the most exciting and potentially clinically relevant conclusions in the work of Yang and collaborators are gleaned from the global analysis of their microarray data.

One reassuring and important observation is that the results of microarray analysis of fibrotic lungs are highly reproducible. In the past, researchers were discouraged from using microarrays because of claims that the results are not reproducible across institutes and platforms. Recently, these claims have been directly tested and proven wrong (6). Microarray experiments were highly reproducible in identifying differentially expressed genes (6). The work of Yang and colleagues supports this observation in pulmonary fibrosis. Despite a difference in platform, sample collection, and institution, they identify the same functional gene groups that distinguish IPF from normal lung as found by us in much smaller studies (5, 7, 8). This agreement should encourage IPF investigators to use microarray data to generate hypotheses, pulmonary investigators to use microarrays in their research, and reviewers and readers to trust results of microarray experiments.

An additional global observation is that familial IPF represents a more extreme molecular phenotype of the same disease process as sporadic IPF. This is based on the finding that the overall differences are modest and mostly due to changes in expression intensity rather than particular involved genes. The similarity in gene expression patterns is supported by reports on similar risk factors, clinical phenotypes, and outcomes observed in sporadic and familial disease (2, 9-11). The difference in intensity does not have to be specific to familial disease. In fact, it may be explained by the sampling stage: 50% of familial cases were obtained by surgical lung biopsy, whereas 93% of the sporadic cases were obtained from autopsies or explants. It is possible that samples obtained from biopsies were obtained at earlier, more active stages of IPF, compared with end-stage lungs obtained at explant or autopsy, when the decreased cellularity observed in advanced remodeling may have reduced pathologically related transcriptional activity. Regardless of whether the changes in intensity are specific to the difference between familial and sporadic IPF, or are due to different stages in disease, this observation may represent the first evidence that we may be able to monitor IPF activity using gene expression patterns. The genes that are high in sporadic IPF and even higher in familial disease may thus be used as biomarkers for disease activity and response to therapy, whereas their slow spontaneous decline may represent a decrease in transcriptionally active lung tissue that characterizes end-stage lung.

Finally, using rigorous analytic approaches, Yang and colleagues (4) identify only a few genes that are significantly different between samples with the histologic patterns of nonspecific interstitial pneumonia (NSIP) and usual interstitial pneumonia (UIP). This is in agreement with our previous results (12). The intuitive interpretation is that gene expression patterns are similar because NSIP is just an early manifestation of UIP. This interpretation does not really explain how gene expression patterns that underlie such distinct histologic patterns would be missed by a technology as sensitive as microarrays. A technical explanation may suggest that the problem is with our comparisons; that we compare lungs that have pure NSIP with lungs that contain both UIP and NSIP lesions, thus making it impossible to find statistically significant differences. Another, more heretic interpretation would be that at least some of the distinction between UIP and NSIP is visual (histologic) but not substantial (molecular). To test this hypothesis directly, we will need to perform laser capture microdissection combined with microarray analysis on NSIP and UIP lesions in the same lung. A more speculative hypothesis would suggest that alterations in post-transcriptional regulatory mechanisms, such as microRNAs (miRNAs), determine the distinct histologic phenotype of NSIP despite an underlying gene expression profile identical to UIP (13). This hypothesis could be directly tested using miRNA arrays. No matter what the interpretation is, the similarity in gene expression patteras between NSIP and UIP challenges our perceptions of the interaction between gene expression and histologic phenotype.

Gene expression profiles have been highly effective in breaking down and subclassifying disease phenotypes with similar histologic presentation (14-16). In the case of interstitial lung disease, it seems that gene expression profiles suggest similar profiles for the histologically different presentations of UIP and NSIP and for familial or sporadic IPF. The availability of more datasets and approaches for meta-analysis of microarray experiments should allow us to validate these intriguing observations.