Genome-wide analysis of the mouse lung transcriptome reveals novel molecular gene interaction networks and cell-specific expression signatures

The lung is the first line of defense against many pathogens and inhaled xenobiotics and is therefore a key part of the immune system. Host defense is strongly influenced by genetic differences and several studies have shown that the genetic background and sequence difference among humans and other host species modulate susceptibility and resistance to infectious diseases, allergens, and xenobiotics. Systems genetics is a modern extension of complex trait analysis that jointly analyzes and integrates large sets of genotypes and phenotypes to explain and predict variation in outcome measures and disease severity (for review see [1, 2]). A typical systems genetics study relies on extensive single nucleotide polymorphism (SNP) data sets, matched data on RNA expression in key cells, tissues, or organs and a core set of key dependent measures such as disease susceptibility [3]. These data are collected across a panel or population of genetically diverse individuals or strains. This group of individuals represents a natural genetic perturbation, with well defined genotype and haplotype differences comprising the "treatment." The independent measurements in this case can consist either of the genotype or of crucial intervening variables such as the expression of genes and proteins.

In this study, we exploited a very well characterized panel of inbred strains of mice (a mouse genetic reference panel) that consists of two parts--a small set of standard inbred strains and a larger family of BXD type recombinant inbred strains. The genome of each BXD strain represents a mixture of the C57BL/6J and DBA/2J parental background and is homozygous at almost every genomic location. The genomic make-up of each BXD line has been determined by extensive mapping with molecular markers. After performing microarray expression analysis for each of the BXD mice, the expression level of each gene can be used as a quantitative trait (e.g. [4‐6]). By comparing these expression values for all BXD mice with their molecular markers data along the genome, genomic expression quantitative trait loci (eQTL) can be identified that are likely to regulate the expression of one or several genes [2, 5, 7‐12]. When an eQTL is located at the same genomic position as the gene itself (within a 10Mb interval of the gene) it is considered as a cis-eQTL. In this case, variations in the promoter sequence or in elements that determine the stability of the mRNA of the gene are the most likely causes for the observed differences in expression levels. If the eQTL is at a distant location from the regulated gene, the eQTL is referred to as a trans-eQTL.

Here, we performed a global gene expression analysis from the lungs of 47 BXD and eight widely used inbred strains. The aim of our study was to reveal genes and gene networks in mouse lung in steady state condition. We found that many genes had high variation in expression and that often this variation was highly heritable. This allowed us to identify many cis- and trans-eQTLs. In addition, we used the correlation structure in the data to obtain expression signatures for specific cell types within the lung.

Here, we performed global gene expression profiling in eight inbred mouse strains and a cohort of BXD recombinant inbred strains from whole lung tissues. Our studies identified several lung-specific genes, large variations in gene expression levels, and a strong heritability in many gene expression traits. Correlation analysis of gene expression and genotypes identified potential gene interaction networks, pairs of trans- and cis-eQTLs, and genes with cis-eQTLs that may represent candidate genes involved in susceptibility to respiratory infections. In addition, one specific gene interaction pathway was identified in which Ahr regulates the Cyp1a1 gene.

Using tissue correlations of gene expression patterns across the BXD strains, we identified 16 genes with a highly restricted expression in the lung of which 14 could be validated by comparison to the BioGPS database [19]. The second most strongly expressed gene in the lung tissues was Sftpc which has been shown to play a role in lung development and the prevention of pneumonitis and emphysema [32, 33]. Also, Sftpc deficiency increases the severity of respiratory syncytial virus-induced pulmonary inflammation [34]. Furthermore, Scgb1a1 and Ager were amongst the five most strongly expressed genes. Scgb1a1 is expressed in lung clara cells and its deficiency results in enhanced susceptibility to environmental agents [35]. Scgb3a1 (secretoglobin, family 3A, member 1) and Scgb3a2 (secretoglobin, family 3A, member 2) were shown by others to be highly expressed in the lung and lower levels in other organs [36]. Scgb3a2 is down-regulated in inflamed airways [37] and plays an important role in lung development [38]. Sftpb (surfactant associated protein B (non-ciliated bronchiolar and alveolar type 2 cell signature) is a hydrophobic peptide which enhances the surface properties of pulmonary surfactant and is expressed in non-ciliated bronchiolar and aleveolar type 2 cells [39]. Maintenance of Sftpb expression is critical for survival during acute lung injury [40] and reduction of alveolar expression causes surfactant dysfunction and respiratory failure [41]. Plunc (palate, lung, and nasal epithelium carcinoma associated) is expressed in the oral, lingual, pharyngeal and respiratory epithelia [42] and members of the Plunc gene family are thought to pay a role in the innate immune response [43]. The presence of Plunc protein in the lung decreases the levels of Mycoplasma pneumoniae and its levels are reduced in allergic inflammatory conditions [44]. Thus, the lung data set allowed us to find important genes that are expressed primarily in the lung and are important for lung homeostasis and prevention of disease.

It should be noted that our analysis of genes with "restricted expression to the lung" is not ex-clusive; it only refers to the tissues that are represented in GeneNetwork and BioGPS. Also, the analysis performed here should not be considered to be comprehensive. More sophisticated approaches may be employed to identify additional genes which also fulfill the criterion of "lung-restricted" expression.

Furthermore, genes may not be apparent in the lung transcriptome because they are expressed only in a small fraction of cells within the lung. This issue of dilution of expression signals is an important one and we have studied it in several tissues with considerable care (eye, retina, and numerous brain regions) using the same genetic methods and the same array platform. We were consistently able to detect expression of genes that are only expressed in very small cell subpopulations (<0.1%) such as rare amacrine cell subclasses in the retina [8] or very rare oxytocin-expressing neurons (<2000) in whole brain samples. The reason for the increased sensitivity is that with such large sample sizes (~70 lung arrays) the signal-to-noise ratios are much better than standard studies using Affymetrix arrays. These stuides typically use far fewer arrays and do not use genetic methods to "validate" the source of signal.

The strong signal for hemoglobin and lymphocyte-specific genes clearly showed that gene ex-pression patterns of circulating blood cells are readily detectable in the lung transcriptome. This raises the question if an organ should be studied with or without containing blood. The correct answer to this question depends of course on the particular circumstances. However, we feel strongly that a global systems and genetic approach requires the analysis of the entire organ. The expression of genes is not cell-autonomous and depends on cellular micro envi-ronment, physical factors (gas pressure and gradients, etc), pathogen exposure, and many types of interactions. These factors also influence the expression of genes in blood cells. Therefore, we think that it is imperative to look simultaneously at all cells in a function unit: in this case the whole lung plus its containing blood.

In conclusion, the combined analysis of expression levels and correlations in a variety of tissues tissue allowed us to determine genes with restricted or preferential expression in the lung. For several of these genes, an important function in the lung has been described and the same may be assumed for the others. This information will also contribute to a better understanding of the biological function of these genes.

Many phenotypic traits have been studied for the BXD mouse populations and several QTLs were identified which influence diseases or vulnerability in the lung. The detection of cis eQTLs in the very same tissue is one method to identify potential candidate genes under the QTL which may causally influence the trait. Here, we investigated two traits in more detail, susceptibility to influenza virus and susceptibility to mycoplasmosis. Several cis-eQTLs were found in the corresponding QTL regions and in each case, genes could be identified with a presumed role in the host immune defense (discussed already in the results section). Thus, the study of cis-eQTLs in our data set may provide valuable candidates for other quantitative trait genes that influence important lung phenotypes. Furthermore, we found 13 BXD lines with low expression signals for Krt4, Krt13 and Krtdap. Krt4 and Krt13 have been shown to be responsible for White sponge nevus (WSN), also known as Cannon's disease, which is an autosomal dominant skin condition in humans [45‐47]. We propose that the 13 mouse strains have genetic alterations which result in low transcript levels of these genes and they may represent a good model for Cannon's disease. It should be noted, however, that no cis-eQTLs found were found for any of the Krt genes.

We also identified a set of genes for which the expression levels correlated highly with members of the Klr gene family. Klra3 and Klrg1 are killer cell lectin-like receptors that are exclusively expressed on natural killer cells (NK cells). NK cells form a major component of the innate immune system and kill cells by releasing small cytoplasmic granules of proteins called perforins and granzymes [48]. Both Gzma and Prf1 were in the gene network that we identified. In addition, correlations can also be used to expand already known gene networks in specific cell populations. When starting with the Cd3 T cell marker and calculating correlations with all other transcripts measured, we identified a strongly correlated network of genes, in which most of the genes were known as T cell markers or to be involved in T cell activation or homeostasis. In a similar way, when starting with the Cd19 B cell marker, we could identify a strongly correlated network of B cell signature genes. The analysis of these T cell and B cell expression signatures in the Bi-oGPS data base with expression profiles in mouse tissues revealed that indeed >90% of the T and B cell markers were specifically expressed in either T or B cells. Furthermore, most of the T and B cell signature genes represented genes with known function in B and T cell differentiation, activation and homeostasis. For example, the T cell signature included genes encoding subunits of the T cell receptor: Cd3d (CD3 antigen, delta po-lypeptide), Cd3g (CD3 antigen, gamma polypeptide), Tcra (T-cell receptor alpha chain) and Tcrb-V13 (T-cell receptor beta, variable 13) and Lat (linker for activation of T cells) which are involved in T cell activation. The B cell signature contained components of the B cell antigen receptor complex, Cd19 (CD19 antigen) and Cd79a (CD79A antigen (immunoglobulin-associated alpha)), as well as Blk (B lymphoid kinase) tyrosine kinase which is associated with the receptors. Also, the correlations for both signatures in the spleen expression data set in GeneNetwork could indeed confirm that the signatures were strongly correlated (data not shown). In summary, these studies demonstrate that correlation analyses are able to identify genes which very likely interact in a common network or biological process. The approach used here may thus have a great potential to identify new networks and biological processes in the lung. In addition, starting with a known bona-fide cell-specific gene and then analyzing gene expression values across strains, it is possible to identify a set of highly correlated genes. These gene sets genes can now be used as cell-specific signature genes in complex transcriptome studies, e.g. to detect infiltrating immune cells in the lungs after infection.

The genetic mapping of lung expression profiles revealed many cis- and trans-eQTLs, indicating that many gene expression patterns in lung have a strong genetic component. Trans-eQTLs allow the identification of gene-gene regulatory networks. As an example, we found that the transcription factor Ahr was present in a trans-eQTL region detected for the Cyp1a1 gene. Ahr is a transcription factor known to induce Cyp1a1 transcription levels after ligand binding [49‐51]. Six binding sites for the Ahr receptor ligand have been revealed in the 700-basepair DNA domain upstream of Cyp1a1 [52]. However, a critical leucine-to-proline substitution in Ahr results in a 15 to 20-fold reduction in the binding affinity of the proline variant found in DBA/2J compared to the leucine variant found in C57BL/6J [53]. Indeed, in our data set, expression values for Cyp1a1were low for BXD strains carrying the DBA/2J allele at the Ahr locus and high for the strains carrying the C57BL/6J allele. Since Ahr is not cis-regulated in lung, the downstream effects appear to be only caused by changes in Ahr protein binding affinity. Although the interaction between Cyp1a1 and Ahr as such is not a new finding, it is quite remarkable that the interaction becomes apparent in lungs which were not exposed to an inducing xenobiotic. Furthermore, we do not see this relationship in several other tissues, such as liver. Therefore, our observation suggests that in the lung, which is potentially exposed to many xenobiotics, the Ahr receptor may always be activated at a low level. Alternatively, Ahr expression may be stimulated by yet unknown ligands that are also present under normal environmental conditions.