Gene Expression Profiles Distinguish Idiopathic Pulmonary Fibrosis from Hypersensitivity Pneumonitis

Abstract

Rationale: Many of the interstitial lung diseases represent a diagnostic and therapeutic challenge because their clinical and even histologic features are often nonspecific. Likewise, the transcriptional signatures of most of them are unknown.

Objective: To compare the gene expression patterns from patients with idiopathic pulmonary fibrosis (IPF) hypersensitivity pneumonitis (HP), and nonspecific interstitial pneumonia (NSIP) using custom oligonucleotide microarrays.

Methods: We profiled lung biopsies from 15 patients with IPF, 12 with HP, and eight with NSIP. Labeled complementary ribonucleic acid was hybridized to a custom Affymetrix oligonucleotide DNA microarray using standard Affymetrix protocols. The custom array, Hu03, contained 59,619 probe sets representing an estimated 46,000 gene clusters.

Results: We identified statistically significant gene expression signatures that characterize HP and IPF. The HP gene expression signature was enriched for genes that are functionally associated with inflammation, T-cell activation, and immune responses, whereas the IPF signature was characterized by the expression of tissue remodeling, epithelial, and myofibroblast genes. We then compared these gene expression signatures to classify NSIP, a histologic pattern that is often difficult to differentiate consistently from HP and IPF. Two cases exhibited an IPF-like gene expression, another one could be more properly classified as HP, whereas others did not resemble HP or IPF, suggesting that they may represent idiopathic NSIP.

Conclusions: Our results underscore the value of gene expression signatures to classify the interstitial lung diseases and to understand pathogenic mechanisms, and suggest new ways to improve the diagnosis and treatment of patients with these diseases.

Pulmonary fibrosis is the final common outcome of a diverse group of interstitial lung diseases (ILDs). A number of these disorders have known etiologies, arising from exposure to organic or inorganic particles or drugs, or they may have an association to collagen-vascular diseases (1). However, approximately 40 to 50% of these disorders are of unknown etiology and are classified as idiopathic interstitial pneumonias (IIPs) (2). The most common IIP is idiopathic pulmonary fibrosis (IPF), which represents a chronic and usually fatal lung disorder morphologically characterized by a usual interstitial pneumonia (UIP) pattern (2,3). Nonspecific interstitial pneumonia (NSIP) is a more recently recognized morphologic pattern characterized by varying degrees of inflammation and fibrosis. These changes are temporally uniform, suggesting that they have occurred over a single, relatively narrow time span (2,3). Importantly, NSIP can be difficult to differentiate consistently from hypersensitivity pneumonitis (HP) and UIP, and significant interobserver variability exists even among expert pathologists (4).

HP is a complex syndrome with variable clinical presentation and natural history, and represents a diagnostic challenge for clinicians and, in the chronic form, even for experienced pathologists. The diagnosis of HP requires a high index of suspicion and a combination of clinical, environmental, radiologic, physiologic, and pathologic findings (5, 6). Chronic HP often develops into fibrosis with destruction of the lung parenchyma, a process that mimics IPF (7).

It is widely believed that lung fibrosis is triggered by an injury that results in inflammation, which, if unresolved, is followed by fibroblast proliferation/activation and extracellular matrix accumulation. However, recent evidence suggests that this sequence is likely relevant for most ILDs, including HP, but probably not for IPF, which seems to represent primarily an epithelial/fibroblastic disorder (8,9).

To gain a better understanding of the molecular mechanisms that underlie the lung phenotypes in IPF, HP, and NSIP, we sought to identify and compare their gene expression “signatures” using genomewide, custom oligonucleotide microarrays. We hypothesized that these diseases exhibit different gene expression signatures that should be useful to distinguish these disorders and also provide insights regarding the divergent pathways leading to lung fibrosis. We also expected that specific signatures could lead to the identification of new biomarkers to aid in the diagnosis and classification of these diseases.

Our results indicate that IPF can be distinguished from HP by the expression of tissue remodeling and epithelial and myofibroblast genes, whereas HP is characterized by the expression of inflammatory genes representing T-cell activation and immune responses. Using these expression signatures, we were able to reclassify some of the NSIP cases as being closer to IPF or HP, whereas others were likely to represent idiopathic NSIP. Our results highlight the mechanistic differences between ILD subtypes and underscore the potential that transcriptional profiling has for the differential diagnosis of ILD.

METHODS

Fifteen patients with IPF, 12 with HP, and eight with NSIP-like pattern were included in this study. The study received approval from the local ethics committee, and informed consent was obtained from all subjects. Diagnosis of IPF and HP was performed as described elsewhere and confirmed by morphology (3, 10, 11). Diagnosis of NSIP was suspected in patients with an interstitial pneumonia that did not meet the clinical/radiologic criteria for other IIPs and was based primarily on histology (2, 3). Control samples included nonpathologic lung specimens obtained from trauma victims. None of the patients had been treated with corticosteroids/immunosuppressive drugs at the time of biopsy. Additional details are provided in the online supplement. One fragment of the lung samples was frozen in liquid nitrogen for RNA extraction and subsequent expression analysis by DNA microarrays.

RNA Extraction and DNA Microarray Hybridization

RNA extracted from lung tissue was converted into biotinylated cRNA and hybridized to a custom Affymetrix (Santa Clara, CA) oligonucleotide microarray (Hu03 containing 59,619 probe sets representing an estimated 46,000 gene clusters [see online supplement] using standard Affymetrix protocols). The generation of the gene expression data was performed as described (12).

Statistical Methods

Statistical analyses were performed using ScoreGenes software package (Jerusalem, Israel; available from http://www.cs.huji.ac.il/labs/compbio/scoregenes/) (13–15). The most informative genes were defined as those that pass the false-discovery rate of 5% (16) for t test or for threshold number of misclassifications (TNoM) method. To further assess the predictive power of the datasets, we used the Leave-One-Out Cross-Validation (LOOCV) statistical method. This method simulates removal of a single sample every trial and trains on the rest. The procedure is repeated until each sample is left out once and the number of correct and incorrect predictions is counted. For each p value, a classification score was calculated for every sample in a specific class (Reference 15 and the online supplement).

Functional Annotations

Enrichment of functional annotations was performed using National Institutes of Health DAVID and EASE online (http://apps1.niaid.nih.gov/david/) (17, 18). Statistical significance was determined using Fisher's exact score and EASE score (17). Additional details on statistical methods and functional annotations are provided in the online supplement.

Bronchoalveolar Lavage

Bronchoalveolar lavage (BAL) was performed as described (10). Cells were used for differential cell counts and fluorescence-activated cell sorter (FACS) analyses. Insulinlike growth factor binding protein 4 (IGFBP-4) was measured by ELISA following the instructions of the manufacturer (Diagnostic Systems Laboratories, Inc., Webster, TX). Results are expressed in micrograms of IGFBP-4/mg protein. Additional details are provided in the online supplement.

Immunohistochemistry

Tissue sections were treated as previously described (10). The following antibodies were used: Human anti–IGFBP-4 and N-cadherin polyclonal antibodies (Santa Cruz Biotech, Santa Cruz, CA), and human matrix metalloproteinase 1 (MMP-1) monoclonal antibody (Oncogene, San Diego, CA). The primary antibody was replaced by nonimmune serum for negative controls. Additional details are provided in the online supplement.

Flow Cytometric Analyses of Surface CD3, CD4, CD8, and CD69 on BAL Lymphocytes

BAL cells (1 × 10⁶) were labeled with fluorescein isothiocyanate–conjugated anti-CD3, phycoerythrin-conjugated anti-CD4, PerCp anti-CD8, and fluorescein isothiocyanate anti-CD69 (Becton Dickinson, San Jose, CA). Flow cytometry was performed using a FACScan (Becton Dickinson) and analyzed with CellQuest software (Becton Dickinson). Additional details are provided in the online supplement.

RESULTS

Baseline Characteristics of Patients with IPF, HP, or NSIP

Demographic data, pulmonary function test results, and BAL differential cell counts are summarized in Table 1. All patients exhibited clinical, radiologic, and functional evidence of ILD, with variable degrees of dyspnea, decreased lung capacities, and hypoxemia at rest that worsened during exercise. In general, the patients with IPF were older, predominantly male, and more likely to have been cigarette smokers than the patients with HP or NSIP. By contrast, all patients with HP were female. This sex prevalence has been previously described in avian antigen-induced HP in Mexico (19). In the HP group, differential cell counts in BAL fluid were characterized by marked lymphocytosis, whereas in IPF, most BAL inflammatory cells were macrophages with a moderate increase in neutrophils and eosinophils.

TABLE 1.

BASELINE DEMOGRAPHIC, CLINICAL, PHYSIOLOGIC, AND BRONCHOALVEOLAR LAVAGE CHARACTERISTICS OF THE THREE STUDY GROUPS

Variable	IPF (n = 15)	HP (n = 12)	NSIP (n = 8)
Age	59.3 ± 7.9	45.6 ± 10.5	45.9 ± 11.1
Sex, male/female	11/4	0/12	1/7
Duration of symptoms before diagnosis, mo	23.4 ± 20.6	18.3 ± 14.3	19.3 ± 12.9
Smoking status
Never	9	12	6
Former	3	0	2
Current	3	0	0
Clubbing	12/15	3/12	4/8
FVC, % predicted	56.7 ± 13.5	60.8 ± 19.4	53.4 ± 17.5
PaO2, mm Hg*	51.5 ± 10.7	53.7 ± 6.6	51.4 ± 7.3
BAL macrophages, %	80.6 ± 6.8	31.9 ± 18.4	54.2 ± 10.9
BAL lymphocytes, %	13.5 ± 6.9	66.6 ± 18.6	44.5 ± 12.2
BAL neutrophils, %	3.5 ± 1.3	0.7 ± 1.2	0.5 + 0.8
BAL eosinophils, %	2.1 ± 2.5	0.8 ± 1.1	0.9 + 1.6

Definition of abbreviations: BAL = bronchoalveolar lavage; HP = hypersensitivity pneumonitis; IPF = idiopathic pulmonary fibrosis; NSIP = nonspecific interstitial pneumonia.

^*Normal values at Mexico City altitude: 67 ± 3 mm Hg.

Gene Expression Patterns Distinguish IPF and HP

IPF is a disease of unknown etiology, and by far the most severe ILD with very poor prognosis. HP is an ILD of known etiology, although its pathogenic mechanisms are not completely understood. Although considered a more benign ILD, some patients with chronic HP also evolve to end-stage lung fibrosis and die of the progressive destruction of the lung parenchyma. We therefore compared the gene expression patterns of these diseases, and asked whether we could find genes specifically expressed in each disease. A well-acknowledged challenge with microarray data is the asymmetry between the number of samples and the number of parameters measured. This asymmetry may lead to spurious p values regardless of which scoring system is used.

To address this challenge, we performed overabundance analysis, which revealed that the actual number of genes for every p value was substantially higher than would be expected under the null hypothesis that the separation of the samples is random (Figure 1A). Using the nonparametric TNoM scoring method (see Methods here and in the online supplement), we identified two genes that had TNoM score p values that passed Bonferroni correction and 407 genes that passed a significance threshold controlling the false-discovery rate at 5% as previously described (16). To identify the genes that best distinguished IPF from HP, we performed LOOCV for every TNoM score p value (Figure 1B). Minimal error rates were observed at TNoM score p values of 0.002762 and 0.000691 (Figure 1B) where classification errors were 1 and 0 for misclassifying HP as IPF and IPF as HP, respectively, strongly suggesting that the signature distinguishing these two diseases is significant. Unsupervised clustering also distinguished the diseases, as did Student's t test (data not shown).

Figure 1.

Hypersensitiviy pneumonitis (HP) and idiopathic pulmonary fibrosis (IPF) can be distinguished using gene expression profiling. (A) Overabundance analysis of the actual and expected number of genes that significantly distinguish whole lung gene expression between patients with IPF or HP and healthy subjects. The x axis denotes the number of genes (probes) in log scale, and the y axis shows the p value (threshold number of misclassifications [TNoM]) in log scale. Actual number of genes is significantly more abundant than would be expected (pink line). (B) Leave-One-Out Cross-Validation (LOOCV) classification graph. The x axis denotes the TNoM p value in log scale, and the y axis denotes the error rate in percentages. The pink arrow is the highest p value that gets the lowest error rate. The orange arrow is the lowest p value that gets the lowest error rate. FDR = false-discovery rate.

The Gene Expression Signatures That Characterize HP or IPF Are Composed of Distinctly Different Functional Groups

To better understand the mechanistic nature of the genes that distinguish IPF from HP, we looked for Gene Ontology functional annotations that were enriched in genes that were higher in IPF lungs or in HP lungs using the National Institutes of Health DAVID and EASE online (http://apps1.niaid.nih.gov/david/) (17, 18). This analysis was only performed on transcripts that could be found in LocusLink (http://www.ncbi.nlm.nih.gov/projects/LocusLink/). Most of the genes that were significantly increased in IPF were enriched for genes involved in development, extracellular matrix structure and turnover, and cellular growth and differentiation. The 354 upregulated genes that include the complete list of statistically significant annotations are listed in Table E1 on the online supplement. Some relevant upregulated genes are shown in Table 2. Furthermore, in addition to the statistically significant p values, many of these genes were associated with the relevant annotations (Figure 2A), suggesting that this represents a biologically meaningful phenomenon. The functional profile was completely different in HP lungs. The 595 genes that were overexpressed in HP compared with IPF and had locus-link information were enriched with a variety of host defense and inflammatory annotations (Table E2 and Figure 2B). Some relevant upregulated genes are shown in Table 3.

TABLE 2.

UPREGULATED GENES IN IDIOPATHIC PULMONARY FIBROSIS

Gene Name	Locus-Link No.	TNoM p Value	t Test p Value
NM_017680: asporin (LRR class 1; ASPN), mRNA	54829	4.0382E-05	0.00028842
NM_033285: tumor protein p53 inducible nuclear protein 1 (TP53INP1), mRNA	94241	0.00033652	2.5438E-06
NM_000029: angiotensinogen (serine [or cysteine] proteinase; AGT), mRNA	183	0.00033652	3.6249E-06
NM_000861: histamine receptor H1 (HRH1), mRNA	3269	0.00033652	1.6201E-05
Hs.31386: secreted frizzled-related protein 2	6423	0.00033652	0.00020623
NM_001792: cadherin 2, type 1, N-cadherin (neuronal) (CDH2), mRNA	1000	0.00033652	0.00026976
NM_005214: cytotoxic T-lymphocyte–associated protein 4 (CTLA4), mRNA	1493	0.00033652	0.00039581
NM_000577: interleukin-1 receptor antagonist (IL1RN), transcript variant 3, mRNA	3557	0.00033652	0.00194308
NM_033326: SRY (sex-determining region Y)-box 6 (SOX6), mRNA	55553	0.00033652	0.0019824
NM_004425: extracellular matrix protein 1 (ECM1), transcript variant 1, mRNA	1893	0.00033652	0.00223442
NM_001847: collagen, type IV, α6 (COL4A6), transcript variant A, mRNA	1288	0.00033652	0.00342796
NM_006475: osteoblast-specific factor 2 (fasciclin I-like; OSF-2), mRNA	10631	0.00033652	0.00359433
NM_130445: collagen, type XVIII, α1 (COL18A1), transcript variant 2, mRNA	80781	0.00033652	0.00457581
NM_002423: matrix metalloproteinase 7 (matrilysin, uterine; MMP7), mRNA	4316	0.00033652	0.00486756
NM_002991: chemokine (C-C motif) ligand 24 (CCL24), mRNA	6369	0.00033652	0.0099309
NM_000582: secreted phosphoprotein 1 (osteopontin, SPP1), mRNA	6696	0.00201911	2.6178E-05
Hs.156369: tenascin N	63923	0.00201911	7.8339E-05
NM_005268: gap junction protein, β5 (connexin 31.1; GJB5), mRNA	2709	0.00201911	8.237E-05
NM_000494: collagen, type XVII, α1 (COL17A1), transcript variant long, mRNA	1308	0.00201911	8.792E-05
NM_032888: collagen, type XXVII, α1 (COL27A1), mRNA	85301	0.00201911	0.00012678
Hs.9914: follistatin	10468	0.00201911	0.0001307
NM_022097: hepatocellular carcinoma antigen gene 520 (LOC63928), mRNA	63928	0.00201911	0.00014831
NM_001723: bullous pemphigoid antigen 1 (BPAG1), transcript variant 1e, mRNA	667	0.00201911	0.00020654
NM_006142: stratifin (SFN), mRNA	2810	0.00201911	0.00030087
NM_152999: six transmembrane epithelial antigen of prostate 2 (STEAP2), mRNA	261729	0.00201911	0.00039041
Hs.306692: epithelial membrane protein 1	2012	0.00201911	0.00042533
NM_014799: hephaestin (HEPH), transcript variant 2, mRNA	9843	0.00201911	0.00051385
Hs.512555: collagen, type XIV, α1 (undulin)	7373	0.00201911	0.00058874
Hs.398100: mucin 6, gastric	4588	0.00201911	0.00087085
NM_003289: tropomyosin 2 (β; TPM2), mRNA	7169	0.00201911	0.00090799
NM_002546: tumor necrosis factor receptor superfamily, member 11b (osteoprotegerin; TNFRSF11B), mRNA	4982	0.00201911	0.00106604
NM_002785: pregnancy-specific β1-glycoprotein 11 (PSG11), mRNA	5680	0.00201911	0.00148639
NM_005786: serologically defined colon cancer antigen 33 (SDCCAG33), mRNA	10194	0.00201911	0.00134524
NM_002421: matrix metalloproteinase 1 (interstitial collagenase; MMP1), mRNA	4312	0.00201911	0.00180712
NM_014476: α-actinin-2–associated LIM protein (ALP), mRNA	27295	0.00201911	0.00185632
NM_153756: fibronectin type III domain containing 5 (FNDC5), mRNA	252995	0.00201911	0.00225902
NM_004385: chondroitin sulfate proteoglycan 2 (versican; CSPG2), mRNA	1462	0.00201911	0.00284683
NM_001552: insulinlike growth factor binding protein 4 (IGFBP4), mRNA	3487	0.00201911	0.003664
Hs.443625: collagen, type III, α1	1281	0.00201911	0.00486324
NM_000599: insulinlike growth factor binding protein 5 (IGFBP5), mRNA	3488	0.00201911	0.00619535
NM_001855: collagen, type XV, α1 (COL15A1), mRNA	1306	0.00201911	0.00778896
NM_004363: carcinoembryonic antigen-related cell adhesion molecule 5 (CEACAM5), mRNA	1048	0.00201911	0.00834501
Hs.169849: myosin binding protein C, slow type	4604	0.00928793	0.00042598
NM_003882: WNT1 inducible signaling pathway protein 1 (WISP1), transcript variant 1, mRNA	8840	0.00928793	0.0004304
Hs.172028: a disintegrin and metalloproteinase domain 10	102	0.00928793	0.00046155
Hs.58324: a disintegrin-like and metalloprotease (reprolysin type) with thrombospondin type 1 motif, 5 (aggrecanase-2)	11096	0.00928793	0.00105437
Hs.56255: keratin 6 irs4	121391	0.00928793	0.00155258
Hs.443567: transitional epithelia response protein	29914	0.00928793	0.00190983
NM_001793: cadherin 3, type 1, P-cadherin (placental; CDH3), mRNA	1001	0.00928793	0.00217983
Hs.352156: a disintegrin-like and metalloprotease (reprolysin type) with thrombospondin type 1 motif, 14	140766	0.00928793	0.00277865
NM_015444: Ras-induced senescence 1 (RIS1), mRNA	25907	0.00928793	0.00397939
Hs.28005: transforming growth factor, β receptor I (activin A receptor type II–like kinase, 53 kD)	7046	0.00928793	0.00403024
NM_002273: keratin 8 (KRT8), mRNA	3856	0.00928793	0.00463512
NM_007159: sarcolemma-associated protein (SLMAP), mRNA	7871	0.00928793	0.00467059
NM_053025: myosin, light polypeptide kinase (MYLK), transcript variant 1, mRNA	4638	0.00928793	0.00584138
NM_000640: interleukin-13 receptor, α2 (IL13RA2), mRNA	3598	0.00928793	0.00666183
NM_012134: leiomodin 1 (smooth muscle; LMOD1), mRNA	25802	0.00928793	0.00847851
NM_000094: collagen, type VII, α1 (COL7A1), mRNA	1294	0.00928793	0.00937536
NM_001202: bone morphogenetic protein 4 (BMP4), transcript variant 1, mRNA	652	0.00928793	0.00953431

Definition of abbreviation: TNoM = threshold number of misclassifications.

Figure 2.

The distribution of differentially expressed genes among functional categories. (A) Genes that had locus-link information and were significantly increased in IPF compared with HP. (B) Genes that were increased in HP compared with IPF. Note the large percentage of genes that belong to the distinct functional annotations. All enrichments are statistically significant (Fisher's exact score p value < 0.05).

TABLE 3.

UPREGULATED GENES IN HYPERSENSITIVITY PNEUMONITIS

Gene Name	Locus-Link No.	TNoM p Value	t Test p Value
NM_005516: major histocompatibility complex, class I, E (HLA-E), mRNA	3133	3.1063E-06	8.581E-08
NM_002985: chemokine (C-C motif) ligand 5 (CCL5), mRNA	6352	3.1063E-06	1.1854E-07
Hs.351874: major histocompatibility complex, class II, DOα	3111	3.1063E-06	2.0498E-07
Hs.368409: major histocompatibility complex, class II, DP β1	3115	3.1063E-06	2.6687E-06
NM_014387: linker for activation of T cells (LAT), mRNA	27040	3.1063E-06	2.9103E-06
NM_000732: CD3D antigen, δ polypeptide (TiT3 complex; CD3D), mRNA	915	4.0382E-05	1.1558E-06
NM_002209: integrin, αL (antigen CD11A [p180], lymphocyte function–associated antigen 1; α polypeptide; ITGAL), mRNA	3683	4.0382E-05	1.3539E-06
NM_002198: interferon regulatory factor 1 (IRF1), mRNA	3659	4.0382E-05	2.2803E-06
NM_005601: natural killer cell group 7 sequence (NKG7), mRNA	4818	4.0382E-05	2.5394E-06
NM_001767: CD2 antigen (p50), sheep red blood cell receptor (CD2), mRNA	914	4.0382E-05	3.4159E-06
NM_001558: interleukin-10 receptor, α (IL10RA), mRNA	3587	4.0382E-05	1.0847E-05
NM_006120: major histocompatibility complex, class II, DMα (HLA-DMA), mRNA	3108	4.0382E-05	1.141E-05
NM_002118: major histocompatibility complex, class II, DMβ (HLA-DMB), mRNA	3109	4.0382E-05	1.3627E-05
NM_000246: MHC class II transactivator (MHC2TA), mRNA	4261	4.0382E-05	2.1963E-05
NM_001109: a disintegrin and metalloproteinase domain 8 (ADAM8), mRNA	101	4.0382E-05	3.0077E-05
NM_000565: interleukin-6 receptor (IL6R), transcript variant 1, mRNA	3570	4.0382E-05	3.5004E-05
NM_004355: CD74 antigen (invariant polypeptide of major histocompatibility complex, class II antigen-associated; CD74)	972	4.0382E-05	4.5692E-05
NM_000206: interleukin-2 receptor, γ (severe combined immunodeficiency; IL2RG), mRNA	3561	4.0382E-05	5.3729E-05
NM_001778: CD48 antigen (B-cell membrane protein; CD48), mRNA	962	4.0382E-05	0.00019398
NM_002339: lymphocyte-specific protein 1 (LSP1), mRNA	4046	4.0382E-05	0.0004061
NM_006144: granzyme A (granzyme 1, cytotoxic T-lymphocyte-associated serine esterase 3; GZMA), mRNA	3001	0.00033652	3.5408E-06
NM_003874: CD84 antigen (leukocyte antigen; CD84), mRNA	8832	0.00033652	2.1163E-05
Hs.181244: major histocompatibility complex, class I, A	3105	0.00033652	2.3716E-05
NM_000265: neutrophil cytosolic factor 1 (47 kD, chronic granulomatous disease, autosomal 1; NCF1), mRNA	4687	0.00033652	2.7125E-05
NM_001768: CD8 antigen, α polypeptide (p32) (CD8A), transcript variant 1, mRNA	925	0.00033652	3.6982E-05
NM_004355: CD74 antigen (invariant polypeptide of major histocompatibility complex, class II antigen-associated; CD74)	972	0.00033652	3.7304E-05
Hs.387679: major histocompatibility complex, class II, DQ α1	3117	0.00033652	7.4311E-05
Hs.387787: killer cell lectinlike receptor subfamily K, member 1	22914	0.00033652	0.00010199
NM_003019: surfactant, pulmonary-associated protein D (SFTPD), mRNA	6441	0.00033652	0.00015899
NM_016326: chemokine-like factor (CKLF), transcript variant 3, mRNA	51192	0.00033652	0.00027442
NM_001242: tumor necrosis factor receptor superfamily, member 7 (TNFRSF7), mRNA	939	0.00033652	0.00038993
NM_012292: minor histocompatibility antigen HA-1 (HA-1), mRNA	23526	0.00033652	0.00046328
NM_000878: interleukin-2 receptor, β (IL2RB), mRNA	3560	0.00033652	0.00531621
NM_003879: CASP8 and FADD-like apoptosis regulator (CFLAR), mRNA	8837	0.00033652	0.00648208
NM_002984: chemokine (C-C motif) ligand 4 (CCL4), mRNA	6351	0.00201911	6.3196E-06
NM_002416: chemokine (C-X-C motif) ligand 9 (CXCL9), mRNA	4283	0.00201911	5.4339E-05
NM_004221: natural killer cell transcript 4 (NK4), mRNA	9235	0.00201911	9.8112E-05
NM_021181: SLAM family member 7 (SLAMF7), mRNA	57823	0.00201911	0.00016428
NM_000063: complement component 2 (C2), mRNA	717	0.00201911	0.00024703
NM_172369: complement component 1, q subcomponent, gamma polypeptide (C1QG), RNA	714	0.00201911	0.00040263
NM_003467: chemokine (C-X-C motif) receptor 4 (CXCR4), mRNA	7852	0.00201911	0.00043211
NM_001079: ζ-chain (TCR)–associated protein kinase 70 kD (ZAP70), mRNA	7535	0.00201911	0.00061774
NM_020125: SLAM family member 8 (SLAMF8), mRNA	56833	0.00201911	0.00062281
NM_001565: chemokine (C-X-C motif) ligand 10 (CXCL10), mRNA	3627	0.00201911	0.00109636
NM_004513: interleukin 16 (lymphocyte chemoattractant factor; IL16), transcript variant 1, RNA	3603	0.00201911	0.00188205
NM_005781: activated Cdc42-associated kinase 1 (ACK1), mRNA	10188	0.00201911	0.00213368
NM_007261: leukocyte membrane antigen (CMRF-35H), mRNA	11314	0.00201911	0.00214496
NM_153461: interleukin-17 receptor C (IL17RC), transcript variant 1, mRNA	84818	0.00201911	0.0026955
NM_000733: CD3E antigen, ɛ polypeptide (TiT3 complex; CD3E), mRNA	916	0.00201911	0.00294939
NM_021798: interleukin-21 receptor (IL21R), transcript variant 1, mRNA	50615	0.00201911	0.003582
NM_001953: endothelial cell growth factor 1 (platelet-derived; ECGF1), mRNA	1890	0.00201911	0.00438273
NM_000579: chemokine (C-C motif) receptor 5 (CCR5), mRNA	1234	0.00928793	9.6163E-05
NM_019111: major histocompatibility complex, class II, DR α (HLA-DRA), mRNA	3122	0.00928793	0.00036044
NM_003820: tumor necrosis factor receptor superfamily, member 14 (herpesvirus entry mediator; TNFRSF14), mRNA	8764	0.00928793	0.00041457
Hs.409934: major histocompatibility complex, class II, DQ β1	3119	0.00928793	0.00055793
NM_006926: surfactant, pulmonary-associated protein A2 (SFTPA2), mRNA	6436	0.00928793	0.0007712
NM_015259: inducible T-cell costimulator ligand (ICOSL), mRNA	23308	0.00928793	0.00281981
NM_001774: CD37 antigen (CD37), mRNA	951	0.00928793	0.00288764
NM_001784: CD97 antigen (CD97), transcript variant 2, mRNA	976	0.00928793	0.00541578
NM_014312: cortical thymocyte receptor (X. laevis CTX) like (CTXL), mRNA	23584	0.00928793	0.00621325
NM_003810: tumor necrosis factor (ligand) superfamily, member 10 (TNFSF10), mRNA	8743	0.00928793	0.00811714

For definition of abbreviation, see Table 2.

IPF Can Be Distinguished from HP by the Expression of Proliferation, Tissue Remodeling, and Myofibroblast Genes

IPF lungs strongly express genes of a variety of extracellular matrix components, as well as muscle-specific genes and genes involved in cell motility and muscle contraction, suggesting significant activity in tissue remodeling and reorganization (Table 2). Likewise, many of these genes are implicated in epithelial cell adhesion/migration. Genes encoding extracellular matrix molecules that were highly expressed in IPF samples included many collagens (types III, IV, VII, XIV, XV, XVII, XVIII, and XXVII), tenascin N, versican, and asporin. Other extracellular matrix components increased in IPF lungs included a disintegrin-like and metalloprotease (reprolysin type) with thrombospondin type 1 motif (ADAMTS)-14 and ADAMTS-5 (aggrecanase-2), and two MMPs, MMP-1 and MMP-7.

Several muscle-specific and actin-binding genes were upregulated, including myosin-binding protein C, myosin heavy polypeptide 11, tropomyosin 2, calponin 1, leiomodin 1, light polypeptide kinase of myosin, and α-actinin-2–associated LIM protein. Likewise, several other genes related to cell motility and muscle contraction were also increased.

Two genes involved in the Wnt pathway, the WNT1-inducible signaling pathway protein 1 (WISP-1) and the secreted frizzled-related protein 2, were increased in IPF lungs.

A number of genes implicated in tissue fibrogenesis were upregulated in IPF lungs when compared with HP. Among them, angiotensinogen was one of the most overexpressed. Likewise, the profibrotic cytokine osteopontin, IGFBP-4 and IGFBP-5, and the interleukin (IL)-13 receptor were increased.

Several epithelial-related genes were upregulated in IPF lungs. These include keratins 6 and 8, mucin 6, stratifin, which regulates cell growth and is also a tumor suppressor gene, and extracellular matrix protein 1, a gene located in chromosome 1q21 (close to the epidermal differentiation complex), which is believed to play an important role in epithelial cell differentiation (20). N-cadherin, a member of a large family of calcium-dependent cell adhesion molecules, was also increased. Taken together, these data indicate that IPF involves profound changes within epithelial cells of the lung and in the connective tissue, which together mediate morphologic restructuring of the lung.

Localization of IGFBP-4, N-Cadherin, and MMP-1 Immunoreactive Proteins in IPF Lungs

To establish whether the information obtained using gene expression profiling reflects protein expression, we selected three representative genes found elevated in the IPF lungs (Table 2). The cellular sources of N-cadherin, IGFBP-4, and MMP-1 were examined by immunohistochemistry on tissue sections of both IPF and control lungs. As shown in Figure 3, MMP-1 was detected in hyperplastic type 2 pneumocytes, whereas IGFBP-4 was found expressed by alveolar and bronchiolar (basal) epithelial cells. N-cadherin was detected in flattened alveolar epithelial cells covering fibroblast foci. MMP-1 was also revealed in macrophages in HP lungs (not shown). These findings strongly suggest that the gene expression levels correlate with protein expression in the tissues studied and support the notion that epithelial cells are highly active in IPF.

Figure 3.

Immunolocalization of insulinlike growth factor binding protein 4 (IGFBP-4), N-cadherin, and matrix metalloproteinase 1 (MMP-1) in IPF and normal lungs. (A) MMP-1 immunoreactive protein was found in alveolar epithelial cells. It can be noticed that the enzyme is virtually absent in fibroblastic foci (original magnification, 40×). (B) IGFBP-4 immunoreactive bronchiolar basal epithelial cells in IPF lungs (40×). (C) N-cadherin immunoreactive protein was observed in alveolar epithelial cells usually covering fibroblastic foci (40×). (D) Normal lungs were usually negative for the three molecules as exemplified for N-cadherin (40×). (E) A negative control omitting the primary (in this case, IGFBP-4) antibody (40×). Results illustrate the analysis of six IPF and five control lungs.

To explore this further, we quantified IGFBP-4 in BAL fluids of patients with IPF or HP as well as suitable control subjects. As shown in Figure 4, IGFBP-4 protein was undetectable in BAL from control individuals, and was significantly elevated in BAL from patients with IPF compared with patients with HP (2.84 ± 1.35 vs. 1.17 ± 0.33 μg/mg protein; p < 0.01).

Figure 4.

ELISA for IGFBP-4 protein in bronchoalveolar lavage (BAL) fluid samples. The protein was not detected in BAL from healthy individuals (controls). An increased concentration of IGFBP-4 was detected in BAL obtained from patients with IPF compared with patients with HP. The data represent the mean ± SD. For the IPF samples, p < 0.01.

HP Is Characterized by the Upregulation of Genes Related to T-Cell Activation and Immune Responses

In HP lungs, microarray analyses identified a variety of genes typically associated with inflammation, T-cell activation, and immune responses (Table 3). Genes related to T-lymphocyte activation included Src-like adaptor 2, CD2, components of the T-cell receptor complex (CD3-D, and CDR-E), and the α chain of CD8. Likewise, major histocompatibility complex (MHC) class II transactivator, the master regulator of MHC class II expression, and several genes encoding MHC class I and II molecules were upregulated.

Several chemokines were increased in HP, including CXCL9 and CXCL10. These recruit activated T and natural killer cells through the chemokine receptor CXCR3 and have been associated with T-helper type 1 (Th1) immune responses (21). CXCR4 and CCR5 and their ligands CCL5 and CCL4 were also upregulated, suggesting that the recruiting/homing program for lung lymphocytes involves multiple chemokines. Another increased gene was CD84, which is associated with follicular T-helper cells, a third major subset of nonpolarized effector T cells that provides help to B cells (22).

Also, several genes involved in host defense were overexpressed, including granzyme A, which reflects the presence of activated CD8 T lymphocytes (see below). Other immune genes increased in HP included several members of the tumor necrosis factor superfamily, IL-16 and several cytokine receptors, including IL-2R, IL-6R, IL-10R, IL-17R, and IL-21R.

FACS Analysis of BAL Cells Confirmed the Presence of Activated T Cells in HP but Not in IPF

Patients with HP showed a significant increase in total T lymphocytes as well as in the number of activated CD4⁺ and CD8⁺ T cells in the BAL when compared with BAL cells from patients with IPF (Table 4).

TABLE 4.

FLUORESCENCE-ACTIVATED CELL SORTER ANALYSIS OF LYMPHOCYTES OBTAINED BY BRONCHOALVEOLAR LAVAGE FROM PATIENTS WITH HYPERSENSITIVITY PNEUMONITIS OR IDIOPATHIC PULMONARY FIBROSIS

	HP (n = 7)	IPF (n = 9)	p Value
Total cells, ×106	30.1 ± 14.6	10.8 ± 4.2	< 0.01
Lymphocytes, %	59.9 ± 13.2	14.7 ± 7.9	< 0.01
Total CD3 T cells, ×106	21.7 ± 6.6	1.6 ± 0.7	< 0.01
Total CD4/CD69, ×106	10.1 ± 3.6	0.8 ± 0.4	< 0.01
Total CD8/CD69, ×106	7.3 ± 4.2	0.7 ± 0.4	< 0.01

Definition of abbreviations: HP = hypersensitivity pneumonitis; IPF = idiopathic pulmonary fibrosis.

HP and IPF Gene Expression Signatures Identify NSIP Diagnostic Subgroups

We then analyzed the patients with NSIP in the context of the molecular signatures that distinguish IPF from HP. To avoid overfitting, we opted to use the larger set of genes (1,058 genes; TNoM p value, 0.002762; Figure 1B) with the minimal error rate for classification purposes. We used the 12 HP cases and the 15 UIP samples as the training set. The NSIP-like samples were classified according to them. Figure 5A shows the Infogram of these 1,058 genes. The test samples are in the middle, whereas IPF lungs are on the left and HP on the right. Samples were ranked by their classification score. Samples in the test set closest to IPF classify as IPF, whereas samples closest to HP classify as HP. The classification graph reveals that, in the training set, all IPF samples were correctly classified as IPF, whereas one HP sample was classified as IPF. In the test set, one NSIP sample was classified as HP (Figure 5B). This patient had a history of previous exposure to home birds, a hobby common in Mexico (where the samples were collected), but without cause–effect relationship, positive precipitins, or apparent exposure to other organic particles. Two other NSIP samples were classified as IPF, supporting the notion that fibrotic NSIP may be undistinguishable from UIP. The other five NSIP samples did not classify as either IPF or HP, suggesting that they indeed represent true idiopathic NSIP. Four of these cases were evaluated by a panel of experts (American Thoracic Society workshop on NSIP) and classified as definite (three cases) or probable (one case) NSIP.

Figure 5.

Infogram and classification score. (A) Infogram of 1,058 genes that made the least classification errors by LOOCV. The expression levels for each gene were normalized to the geometric mean of all the samples for each gene. Increased genes are shown in progressively brighter shades of yellow, and decreased genes are shown in progressively darker shades of blue. Genes shown in gray are not different between the groups. The genes were ranked according to their significance level. (B) Classification score of every sample in the dataset. Samples below the zero line were considered as usual interstitial pneumonia. Samples above the separation line were classified as HP. Note that only one HP sample was misclassified. NSIP = nonspecific interstitial pneumonia.

DISCUSSION

This study used microarrays to identify disease-specific gene expression signatures and to directly compare gene expression profiles from various ILDs of known and unknown etiology on a genomewide scale. Our results demonstrate that gene expression patterns clearly distinguish IPF from HP, and indicate that these disease-specific gene expression signatures are biologically meaningful and provide significant information on the mechanisms underlying these diseases.

IPF is a specific and progressive form of chronic interstitial pneumonia limited to the lung and characterized by epithelial cell injury and activation followed by a fibroproliferative response and subsequent scarring with only mild to moderate signs of inflammation. However, this disease has generally been considered a chronic, insidious, inflammatory disease, and consequently, research in this disorder has primarily focused on its inflammatory component and, more recently, on its fibroproliferative component.

Our findings indicate that IPF lungs do not exhibit a “typical” inflammatory pattern (characterized by leukocyte/lymphocyte genes, and mediators such as cytokines). In this study, only CCL24 was found to be significantly upregulated when compared with HP lungs. In a previous study, CCL7 was found to be upregulated in IPF in comparison with other forms of IIPs (23). These differences may reflect different sensitivities of the gene array platforms used.

By contrast, it is precisely the absence of many of these proinflammatory lymphoid/leukocyte-associated genes that distinguishes IPF from HP. Instead, IPF is characterized by the expression of genes that indicate an active tissue remodeling program, including extracellular matrix and basement membrane components as well as by a large number of myofibroblast/smooth muscle cell–associated and epithelial cell–related genes. Thus, several fibrillar and nonfibrillar collagens were increased in IPF. One of these is type VII collagen, the major structural component of the anchoring fibrils at the dermal–epidermal junction in the skin (24). Interestingly, these anchoring fibrils have been described in fibrotic lungs where their presence seems to correlate with the severity of the fibrosis (25). Other upregulated collagens included type XV and type XVIII, which are expressed in specialized basement membranes, and their C-terminal parts, endostatin and restin, have been reported to have a strong antiangiogenic effect (26). Another overexpressed gene is transmembrane collagen XVII, an epithelial adhesion molecule that belongs to the group of collagenous transmembrane proteins, which has been recently implicated in the regulation of keratinocyte migration (27). Interestingly, collagen XXVII, a new member of the vertebrate fibrillar collagen family, was also upregulated. This peculiar collagen is expressed by a variety of epithelial cell layers in developing tissues, including lung (28).

MMP-1 and MMP-7 were also upregulated. As previously shown, MMP-7 seems to play an important role in the pathogenesis of IPF (29).

Angiotensinogen, which is produced by fibroblasts from fibrotic but not from normal tissues (30, 31), was strongly upregulated, supporting the notion that this mediator may be important in the pathogenesis of the disease. Angiotensin has been implicated not only in profibrotic activity as a result of induction of extracellular matrix accumulation but also in alveolar epithelial cell apoptosis, an event considered one of the earliest alterations in IPF (9).

To confirm the RNA profiling analyses, we selected three genes overexpressed in IPF for additional studies at the protein level, and all of them were mainly localized in epithelial cells. IGFBP-4 belongs to a family of at least six specific high-affinity IGF binding proteins that modulate the local bioavailability of IGF-1, a fibroblast mitogen and stimulator of collagen synthesis by fibroblasts, which may also contribute to the fibrogenic process (32, 33). IGFBP-5 gene expression was also upregulated, and it has been recently reported that this protein together with IGFBP-3 are increased in IPF lungs and localized to fibroblasts surrounding airspaces, and to epithelial cells in bronchioles and alveoli (34). The functional roles of IGFBP are poorly understood. They usually inhibit IGF-1 effects by blocking binding to its receptor, but they can also enhance IGF-stimulated cell growth (35, 36). Moreover, it has been suggested that IGFBP-3 may increase the bioactivity of IGF-1, a process that may contribute to lung fibrosis (37). More recently, it was found that IGFBP-3 and IGFBP-5 induce extracellular matrix production by normal lung fibroblasts (34).

Another important finding in IPF lungs was the overexpression of N-cadherin, a mesenchymal cadherin preferentially expressed in migratory cells and in connective tissue. In our study, this molecule was localized in flattened and attenuated alveolar epithelial cells usually overlaying fibroblastic foci and occasionally in hyperplastic type 2 pneumocytes. N-cadherin belongs to a family of calcium-dependent cell adhesion molecules that mediate cell–cell adhesion and also modulate cell migration and tumor invasiveness, thus showing the opposite effect of E-cadherin (38). Actually, it has been hypothesized that a cadherin switch (i.e., a change from E- to N-cadherin expression), is involved not only in the detachment and migration of epithelial cells during embryonic development but also in the transition from a benign to an invasive, malignant tumor phenotype (38, 39). Interestingly, WISP-1 and the secreted frizzled-related protein 2 were also increased in IPF (40). WISP-1 belongs to a recently described connective tissue growth factor/cysteine-rich 61/nephroblastoma (CCN) family of multifunctional signaling regulators that comprises six cysteine-rich proteins, including connective tissue growth factor (41), and is strongly increased in aggressive fibromatosis (42). WISP-1 can activate the antiapoptotic Akt/protein kinase B signaling pathway protecting cells from p53-dependent cell death (43). Collagen type XVIII, another gene overexpressed in IPF lungs, has been recently shown to colocalize with Wnt molecules, and its frizzled-containing N-terminal fragment may be implicated in Wnt signaling (44). In addition, several target genes of the canonic WNT/β-catenin pathway, such as MMP-7 and osteopontin, were also increased in IPF lungs, further supporting the concept that this pathway may be aberrantly induced in IPF but not in HP. Moreover, we have recently demonstrated that, in IPF, osteopontin colocalizes with MMP-7 in alveolar epithelial cells, and application of weakest-link statistical models to microarray data suggests a significant interaction between both molecules (45).

The functional enrichment of IPF lungs with genes associated generally with development and cellular differentiation strongly suggests that a global process involving a change in cellular phenotypes occurs in IPF but not in HP. Such a process may contribute to a transition of epithelial cells to a mesenchymal phenotype. Recently, while this article was under review, findings suggesting an epithelial–mesenchymal transition in IPF have been reported (46).

Our findings on IPF gene expression contrast strongly with those of HP. In the latter disease, the main finding is a strong T-cell response that reflects a strong proinflammatory mechanism. Importantly, similar findings have been made in some chronic cases in which septal fibrosis was present in the biopsy. Taken together, these results support the existence of at least two different routes for developing diffuse pulmonary fibrosis. One of them is through an inflammatory pathway represented by HP, where there is an early, clearly defined phase of alveolitis, which may be followed by a late fibrotic phase. The second is represented by IPF (9) and may be mainly an epithelial/fibroblastic pathway, which may not include a “typical” inflammatory phase.

HP lungs showed a strong inflammatory component, with many genes of infiltrating activated T cells, including cytotoxic T lymphocytes and natural killer cells. The presence of activated CD8 T cells strongly suggests a viral component in this disease, a possibility that has been suggested previously (47, 48). Although the factors that trigger the onset of HP have not been defined yet, genetic susceptibility associated with the MHC (49) and/or viral infections may participate. Common respiratory viruses are often present in the lower airways of patients with HP, and viral infection strongly augments both the early and late inflammatory responses in experimental HP (47, 48). Interestingly, CCL5, which is commonly induced by acute respiratory viral infection, was also upregulated in the HP lungs (19, 50). In addition, CCL5 has been found expressed in the cell membranes of granuloma cells from noncaseating granulomas, which also characterize HP (51). On the other hand, the upregulation of the chemokines CXCL9 and CXCL10, known to be IFN-γ− inducible and ligands of CXCR3, supports the hypothesis that HP is mediated by a Th1-like cell-mediated host response.

The term “nonspecific interstitial pneumonia” is currently applied for lung biopsy specimens from some patients with IIPs that do not fit into any well-defined histopathologic pattern. It is important to emphasize that NSIP was originally described by the absence of morphologic characteristics of other IIPs rather than by exhibiting distinguishing features of its own. Although NSIP is often idiopathic, the same morphologic pattern has been associated with a number of other diseases, including collagen vascular diseases, drug-induced pneumonitis, and importantly, HP (2, 52, 53). Therefore, finding an NSIP pattern should prompt the clinician to carefully consider potentially causative exposure or the presence of a systemic disease. In general, idiopathic NSIP represents a problem for the clinician because its clinical features have been poorly described and thus there is as yet no distinctive clinical description for these patients. Morphologically, two subtypes of NSIP are recognized: cellular and fibrotic. Differential diagnosis of cellular NSIP includes HP, and differential diagnosis of fibrotic NSIP includes primarily UIP but also fibrotic forms of other types of ILDs, including advanced forms of HP (54). Our findings indicating that the transcriptional pattern of three of the eight NSIP-like cases behaves as UIP or HP support this notion.

The most important obstacle in the differential diagnosis of IPF/UIP is NSIP, which differs clinically by a more favorable response to corticosteroid therapy and a better prognosis (51, 55, 56). When fibrosis is prominent in NSIP, the changes can usually be difficult to distinguish from UIP. In addition, areas resembling NSIP are present in many UIP cases in both biopsy and explant specimens, and they are extensive in some of them (57). These findings suggest that an NSIP-like pattern may be found even randomly in the biopsy specimens of many patients with IPF/UIP, making the identification of the latter problematic. Supporting this point of view, it has been demonstrated that patients with IPF may exhibit histologic variability in their lung specimens when biopsy samples are taken from multiple lobes (58, 59). Interestingly, patients with discordant UIP—that is, UIP morphology in one part of the lung and NSIP in the other—show survival, clinical, and physiologic features similar to those with concordant UIP. Moreover, the prognosis in both patients with concordant and discordant UIP is significantly worse than that of the patients with concordant NSIP (NSIP in multiple biopsies).

Although our study includes the largest number of human samples reported in microarray analysis of IPF (or any other nonmalignant lung disease), the sample size still prevents generalization of our results. However, given the statistical magnitude of the differences between IPF and HP, the magnitude and statistical significance of the difference in functional groups enriched within each disease gene expression signature, and the biological plausibility of the results, it is unlikely that larger numbers of samples will change our conclusions significantly. Although, the results of this study should be tested on other independent datasets, the identification of statistically significant, biologically meaningful, disease-specific gene expression signatures suggests that we could use these signatures to predict whether patients would respond to immunosuppressive therapy or to future specific antifibrotic therapies. Currently, it is recommended (with little scientific support) that a therapeutic trial with corticosteroids and some immunosuppressive drugs would be administered to any patient with IPF/UIP and definitely to patients with the other more traditional inflammatory ILDs. Our data support the view that gene expression signatures can be a better predictor of patient responsiveness to antiinflammatory therapy, thus preventing the side effects of an unindicated therapeutic trial from other patients. In addition, when specific antifibrotic therapies may be ready to be tested in clinical trials in the near future, the best way to select patients may be to use their tissue functional gene expression signatures.

In summary, genomewide gene expression profiling of individual IPF, HP, and NSIP lungs has allowed the identification of genetic signatures for these diseases. This may speed up the understanding of the molecular mechanisms underlying these disorders and may also potentially aid in their differential diagnosis. Advances in such diagnoses could potentially be invaluable to, first, distinguish HP from IPF, and second, to provide a more accurate diagnosis in cases currently diagnosed as NSIP.