Conceived and designed the experiments: GKAM DM SB. Performed the experiments: GKAM DM MV JEM. Analyzed the data: GKAM DM JEM SB. Contributed reagents/materials/analysis tools: GKAM JEM SB. Wrote the paper: GKAM DM JEM SB.
The authors have declared that no competing interests exist.
Chronic obstructive pulmonary disease (COPD) is a major global health problem. The etiology of COPD has been associated with apoptosis, oxidative stress, and inflammation. However, understanding of the molecular interactions that modulate COPD pathogenesis remains only partly resolved. We conducted an exploratory study on COPD etiology to identify the key molecular participants. We used information-theoretic algorithms including Context Likelihood of Relatedness (CLR), Algorithm for the Reconstruction of Accurate Cellular Networks (ARACNE), and Inferelator. We captured direct functional associations among genes, given a compendium of gene expression profiles of human lung epithelial cells. A set of genes differentially expressed in COPD, as reported in a previous study were superposed with the resulting transcriptional regulatory networks. After factoring in the properties of the networks, an established COPD susceptibility locus and domain-domain interactions involving protein products of genes in the generated networks, several molecular candidates were predicted to be involved in the etiology of COPD. These include COL4A3, CFLAR, GULP1, PDCD1, CASP10, PAX3, BOK, HSPD1, PITX2, and PML. Furthermore, T-box (TBX) genes and cyclin-dependent kinase inhibitor 2A (CDKN2A), which are in a direct transcriptional regulatory relationship, emerged as preeminent participants in the etiology of COPD by means of senescence. Contrary to observations in neoplasms, our study reveals that the expression of genes and proteins in the lung samples from patients with COPD indicate an increased tendency towards cellular senescence. The expression of the anti-senescence mediators TBX transcription factors, chromatin modifiers histone deacetylases, and sirtuins was suppressed; while the expression of TBX-regulated cellular senescence markers such as CDKN2A, CDKN1A, and CAV1 was elevated in the peripheral lung tissue samples from patients with COPD. The critical balance between senescence and anti-senescence factors is disrupted towards senescence in COPD lungs.
Chronic obstructive pulmonary disease or COPD is among the most lethal of respiratory diseases. While this disease has been well characterized, more studies are needed to learn the interaction of macromolecules involved in the progression towards illness. We explored possible interactions involved in the disease process using a compendium of gene expression data from frontline cells of the respiratory airways of the lung. The gene expression data were generated under a variety of experimental conditions. Application of computational schemes, which robustly detect enduring patterns, among sections of the genes represented across the varying experimental perturbations, revealed important regulatory relationships. When gene expression data from lungs of patients with COPD were factored into these networks of regulatory relationships, certain highly connected nodes (hubs) representing differentially expressed genes emerged. Notably included are members of the T-box (TBX) family of genes and CDKN2A, which regulate cellular aging. These findings were confirmed in studies using lung samples from COPD patients. Novel genes linked to TBX and CDKN2A include COL4A3, CFLAR, GULP1, PDCD1, CASP10, PAX3, BOK, HSPD1, PITX2, and PML, which were thus predicted to be involved in the disease process. The balance between senescence and anti-senescence factors is disrupted towards senescence in COPD lungs.
Chronic obstructive pulmonary disease (COPD) is characterized by a progressive decline in lung function, with an irreversible airflow obstruction, caused either by chronic bronchitis, emphysema or both
Cigarette smoking remains the primary preventable environmental risk factor for COPD
Oxidative and nitrosative stress induced by cigarette smoking is thought to be responsible for corticosteroid resistance in COPD
Although, the understanding of the underlying mechanisms of COPD is constantly evolving, the absence of any novel or effective therapy aimed at this irreversible disease presents a significant challenge
To explore the molecular definition of COPD, transcriptional regulatory networks were derived from airway gene expression data. Large collections of gene expression data provide regulatory patterns that potentially bear valuable insights regarding disease mechanisms. A number of predicted molecular participants involved in COPD etiology are identified in this report, concurring with the aging hypothesis for COPD
The experimental approach is summarized in
A combination of network inference and other algorithms applied to the datasets as described in the
For the purpose of identifying the most influential nodes within the overall network, two features were used. First, the size of each node is an indication of its connectedness within the network: the large size nodes are more connected in the network. The larger nodes include HMOX1, TGFB1, TBX3, CDKN2A, PML, NME1, NPM1, SMAD3, RELA, FOXL2, STAT1, IL1B, TP63, NOTCH2, NFX1, ELF3, HIF1A, NLRP3, NFRKB, E2F1, TIAL1, AATF, TBX5, TCF7L2, HTAPIP2, TNF, ITCH, NFAM1, and CREB1. Second, the color of each node is an indication of the alterations in gene expression in COPD. For this purpose, a study on the differential gene expression in 15 COPD cases and 18 controls was used
Combining these two features facilitated the identification of the most connected nodes that were also differentially expressed in COPD. We hypothesize that these nodes represent genes that may be critical regulators in the etiology of the disease. Among these, the T-box (TBX) transcription factors, TBX3 and TBX5 (
The extent of the suppression is highly significant between patients with mild and those with severe COPD. Patient diagnosis was based on the National Heart, Lung, and Blood Institute/World Health Organization Global Initiative for Chronic Obstructive Lung Disease (GOLD)
For confirmation purposes, a second transcriptional regulatory network was generated using the same data and an alternative algorithm, the Algorithm for the Reconstruction of Accurate Cellular Networks (ARACNE)
GENE | CLR | ARACNE | DIRECT NETWORK NEIGHBORS FOUND BY BOTH ALGORITHMS |
|
80/84 | 80/347 | ACTN1 ACTN3 ACVR1B AIFM2 AVEN BAX BCL2L1 BCL2L10 BCL3 BCLAF1 BID BIRC7 BOK CASP10 CD74 CDK5R1 CDKN2D CECR2 CIDEA CLCF1 COL4A3 CRYAA DAPK2 DAPK3 DIABLO DLC1 DOCK1 F2 FOXO1 FURIN GLRX2 HIP1 IGF1R IL3 IL4I1 KIAA1967 KNG1 KRT18 LYST MAP3K10 MAPK1 MAPK8IP2 NCR1 NLRP12 NME1 NME2 NME5 NOL3 NPM1 PAX3 PCSK6 PDCD5 PDIA2 PHLDA2 PRKAA1 PRKCA PRKCZ PRLR RNF7 RTKN SCIN SEMA6A SERPINB2 SERPINB9 SFRP1 SMAD3 SOCS3 SPATA3 SST SSTR3 TBX5 TGFB1 TGFB2 TIA1 TIAL1 TLR2 TNFRSF19 TNFSF10 TP53 TRIAP1 |
|
37/45 | 37/180 | ACTN1 ACTN3 ADORA1 AGTR2 ALOX15B BBC3 BCL2 BCL2L1 BOK CD74 CDK5R1 CDKN2D CECR2 COL4A3 DNM2 GRM4 HSPD1 KNG1 LYST MAPK1 MAPK8 MPO NLRP12 PAX3 PDCD1 PDCD6 PDIA2 PRKCA PTEN RHOB RHOT1 SEMA6A SNCA SOCS3 TBX3 TIA1 TRAF7 |
|
44/60 | 44/189 | ADAMTSL4 ANXA1 ANXA4 API5 BBC3 BCL2 BCL2A1 BCLAF1 BTG1 CADM1 CASP8AP2 CD38 CDC2L2 CFLAR CRADD CYFIP2 DAPK2 DAXX DEDD2 EIF5A FASLG GCLC GULP1 HDAC3 HMGB1 IL1A MAPK8IP2 MCL1 NLRP1 NOTCH2 NOX5 PCSK6 PDCD1 PDIA2 PIK3R2 PML PRKCA PRKCE SPHK1 TAOK2 TGFB1 TNFRSF25 TNFRSF6B TP63 |
The exploratory study was then expanded to involve all probe sets available on the Affymetrix U133A platform (removing the focus away from apoptosis, oxidative stress, and inflammation genes). All 22,283 probe sets represented after robust multi-array analysis
A) Quantitative PCR data indicate TBX2 gene expression is suppressed, while senescence factors, CDKN2A, CDKN1A, and caveolin-1 are induced in the lungs of patients with COPD compared to lung tissue from normal smokers. Fifteen normal, nine mild COPD, and six severe COPD samples were used for this analysis. The data is represented as Mean ± S.D. The data was analyzed using student's t-test for comparing mRNA expression in the respective groups. B) Representative Western blots showing suppressed TBX2, HDAC2, SIRT1 proteins and increased expression of CDKN2A, CDKN1A and caveolin-1 proteins in samples from patients with COPD. C) Densitometry analysis of Western blot data. Four normal, four mild COPD, and four severe COPD samples were used for this analysis. Densitometry analysis was carried out using image-J software. The data is represented as Mean ± S.D. The data was analyzed using student's t-test for comparing protein expression in the respective groups. *represents a significance of p-value<0.01.
For confirmatory purposes, the entire dataset of 22,283 probe sets and 109 observations (Gene Expression Omnibus datasets GDS534 and GDS999) was also subjected to a bicluster analysis. Each bicluster consists of a subset of probe sets and a collection of the observations (conditions) within which they are similar
Number of Biclusters Learned | Bicluster Number | Probe sets Present | Arrays Present | Some Genes Present | Description of Array Samples In Bicluster |
|
Bicluster #4 | 924 | 35 |
|
|
|
Bicluster #3 | 1749 | 30 |
|
|
|
Bicluster #2 | 2096 | 30 |
|
|
|
Bicluster #8 | 1991 | 44 |
|
|
|
Bicluster #13 | 567 | 56 |
|
Bronchoalveolar lavage samples obtained from lung transplant recipients whose biopsies had a perivascular score of between 0 and 2, and a bronchiolar score of between 0 and 1
Lung epithelial cell transcriptome study of 34 current smokers, 18 former smokers, and 23 subjects who had never smoked
Furthermore, the Inferelator algorithm
BICLUSTER∧ | NUMBER OF PROBESETS | NUMBER OF OBSERVATIONS | CLUSTER PROBABLY POSITIVELY REGULATED BY*** | CLUSTER PROBABLY NEGATIVELY REGULATED BY*** |
|
2084 | 32 | TBX3, ELF3 | ATF2_with_HSF1 |
|
2096 | 30 | CEBPG, ATF1 | PML, TP63, TBX5, TBX5_with_NFKBIB, PPARD_with_SMAD3 |
|
1560 | 31 | CEBPG_with_TP63, NME1-NME2, TCF7L2, TIAL1 | |
|
1569 | 38 | NFE2L2_with_CDKN1A, STAT1 | TCF7L2 |
|
2619 | 35 | STAT5B_with_TP63, SMAD3 | HTATIP2 |
|
1705 | 35 | RUNX3 | E2F1_and_TCFL2 , HTATIP2 |
|
1381 | 38 | STAT1, CREBBP | E2F1, LOC652346, PML_with_TBX2, TBX5 |
|
1991 | 44 | STAT5B, NFATC3 | TCF7L2, TCF7L2_with_TIAL1 |
|
1089 | 33 | NLRP3, SMAD3, RUNX3 | |
|
967 | 47 | SIGIRR | |
|
806 | 32 | RUNX3, RUNX3_with_SMAD3 | ATF2, EP300_with_TBX2, TCF7L2 |
|
566 | 34 | TP63_and_STAT1 | CEBPB, TCF7L2 |
|
567 | 56 | FOXO3, SMAD3_with_ERC1, ELF3 | STAT5A |
|
493 | 50 | ARHGDIA_with_RUNX3, RUNX3 | PML, E2F1 |
|
754 | 45 | ELF3 | CREB1_with_TIAL1 |
|
356 | 43 | PML_with_HSF1, LOC161527, HSF1, SBNO2 | |
|
510 | 46 | SBNO2, JUN, RUNX3_with_CREB1,JUN_with_TP53, TP63 | NFATC4_with_HTATIP2 |
|
380 | 61 | ELF3_with_BCL10 | ATF1 |
|
347 | 47 | SMAD3 | TGFB1_with_GOLGA6L4, STAT5A |
|
497 | 50 | HIF1A, STAT1, STAT5B, RELA | HSF1_with_TBX5 |
∧Obtained using FABIA algorithm | ****Per Inferelator predictions |
On the basis of these results, samples obtained from patients with COPD and normal subjects without COPD were examined for the relative levels of TBX2 and CDKN2A mRNA and protein expression. As shown in
It is well known that COPD clusters in families
A) The state of the Type IV collagen alpha 3 subunit, COL4A3, depends on the states of both TBX2 and CDKN2A in human lung epithelial cells. Following Robust Multi-Array Analysis of a compendium of 109 Affymetrix arrays on the U133A platform, the Context Likelihood of Relatedness (CLR) algorithm was used to generate a transcriptional regulatory network involving all available probe sets (at a CLR likelihood estimate cut-off of 2.5). Olive-green nodes represent genes whose median probe set expressions are suppressed in COPD. White nodes represent genes whose median probe set expressions are elevated in COPD. COL4A3, whose expression is suppressed in the COPD lung, is thus statistically dependent on both TBX2 and CDKN2A. B) Evolutionarily conserved probable protein domain-domain interactions corresponding to the predictions of
Taken together, these results indicate a critical balance between senescence and anti-senescence factors in normal smokers, which is disrupted towards senescence in COPD lungs. There are previous reports of decline in telomere length, which is a hallmark of senescence in samples from patients with COPD
Using a variety of computational approaches, a number of regulatory genes important in COPD have been identified in these studies. First, using gene expression data of genes associated with three Gene Ontology biological processes implicated in COPD, transcriptional regulatory networks were learned by way of CLR and ARACNE. The most highly connected nodes of the networks which simultaneously represented genes differentially expressed in COPD were noted as important. The basic findings were also present in an expanded CLR study of all 22,283 probesets on the U133A Affymetrix platform. Differentially expressed and highly connected nodes of note included TBX2, TBX3, TBX5, and CDKN2A.
Bicluster analyses of the entire U133A platform dataset, which were unbiased in terms of a prior determination of genes of focus, found that the genes related to those noted in the CLR and ARACNE studies clustered together, often co-occurring in more than one bi-cluster (
T-box proteins are an important family of transcription factors. Over 20 genes in vertebrates have a region of homology to the DNA-binding domain of the transcription factor encoded by Brachyury, the T gene
CDKN2A is a mechanistic marker for cellular senescence
Thus, the activities of both T-box proteins and the CDKN2A products converge on the p53 pathway. Our findings (
TBX2 and its close relative, TBX3, negatively regulate cell cycle control genes and CDKNs, specifically CDKN2A, CDKN2B, and CDKN1A
As reported here, the expression of the anti-senescence T-box transcription factors are suppressed in the COPD lungs, and there is a concomitant rise in the expression of the cellular senescence markers, CDKN2A, CDKN1A, and CAV-1 (
Genome-wide studies in patients with COPD indicate that the chromosomal locus spanning 2q33.3–2q37.2 is associated with COPD
Hub | Direct Neighbor | Chromosomal Location |
CDKN2A | ANXA4 | 2p13 |
|
|
|
|
|
|
IL1A | 2q14 | |
|
|
|
TBX3 |
|
|
|
|
|
RTKN | 2p13.1 | |
TIA1 | 2p13 | |
TBX5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
PRKCE | 2p21 | |
RHOB | 2p24 | |
TIA1 | 2p13 |
Genes in bold character are in or border the region of the susceptibility locus.
COL4A3
MLE Probability | ||
CDKN2A | 0.99 | ETS1 |
CDKN2A | 1 | NFKBIB |
CDKN2A | 0.94 | NLRP3 |
CDKN2A | 1 | NOTCH2 |
CDKN2A | 1 | PML |
CDKN2A | 0.99 | TIAL1 |
|
|
|
|
|
|
ETS1 | 0.99 | NFKBIB |
ETS1 | 0.99 | NOTCH2 |
HSF1 | 0.9 | NFKBIB |
HSF1 | 0.9 | NOTCH2 |
JUN | 0.99 | NEUROD1 |
NEUROD1 | 0.99 | PITX2 |
NLRP3 | 0.94 | NFKBIB |
NLRP3 | 0.94 | NOTCH2 |
NLRP3 | 0.99 | PITX2 |
NLRP3 | 1 | TIAL1 |
NOTCH2 | 0.99 | ETS1 |
NOTCH2 | 1 | NFKBIB |
NOTCH2 | 1 | PML |
PML | 1 | JUN |
PML | 1 | NFKBIB |
PML | 1 | NOTCH2 |
PML | 1 | PITX2 |
TIAL1 | 0.99 | NFKBIB |
TIAL1 | 0.99 | NOTCH2 |
TIAL1 | 0.99 | NOTCH2 |
TIAL1 | 1 | PITX2 |
TIAL1 | 1 | PITX2 |
TIAL1 | 1 | PML |
MLE = Maximum likelihood estimation.
Probable interactions involving COL4A3, which is in the susceptibility locus, are in bold character.
The importance of TBX2 and CDKN2A in the network has been highlighted. Other gene products associated with senescence were found via the CLR (using all available probe sets) to be statistically associated with, and thus probably dependent on, TBX2 and/or CDKN2A (
TBX2 | CDKN2A | TGFB1 | |
|
CTNNA1, CTNND1 | CTNNBIP1 | |
|
FOXA2, FOXB1, FOXD3, FOXH1, FOX01, FOXO3 | FOXA1, FOXJ1, FOXO4 | |
|
IGF2, IGF2BP3, IGFBP7, IGFBP5 | IGFALS | IGFALS, IGFBP4, IGF2PB2, IGFBP7, IGFBP5, IGF2R, IGFBP3 |
|
IL4, IL13, IL13RA1, IL27RA, IL1RL1, IL17RC, IL10RB, IL2RA, IL12RB1 | IL1RN, IL6ST, IL1RAPL1 | IL20RA, IL27RA, IL13RA1, IL2RG, IL10RA, IL4I1 |
|
WNT3, WNT4, WNT6, WNT7B, WNT11 | WNT10B |
Our study furthers the paradigm on cellular senescence as its effectors and their regulation are the cross-roads of smoking-induced lung cancer or COPD
The study protocols were approved by the Institutional Review Board for human studies, and patients' lung function data from each of the contributing centers were obtained for this study.
The overall experimental approach is summarized in
Expanding on the findings of the first phase, the entire set of probe sets represented on 109 arrays of the U133A platform was used during the second phase in which the CLR algorithm was executed (Gene Expression Omnibus datasets GDS534 and GDS999). Subsequently to assure the reliability of the regulatory relationships just described, biclusters were identified within the dataset using the FABIA algorithm. The Inferelator algorithm was then also used to predict regulators of those biclusters.
A compendium of microarray data was generated from the Gene Expression Omnibus (GEO). GEO record numbers GDS534, GDS999, GDS2604, and GDS2486 (
The CLR algorithm
Thus a likelihood estimate:
Given two random variables, X and Y, the mutual information between them is given by:
In other words, the shared information between X and Y corresponds to the remaining information of one party if we remove the information of that party that is not shared with the other party. For two genes, X and Y, the mutual information is given by:
The code implementation provided by Faith et al. was used from a Linux command line
In the second phase, all 22,283 probe sets represented after background correction and normalization on the U133A platform were used (Gene Expression Omnibus datasets GDS534 and GDS999). A likelihood estimate cut-off value of 2.5 used generated a network consisting of 17,396 nodes and 127, 331 transcription regulatory links.
Like CLR, ARACNE uses mutual information
The algorithm
The Inferelator
The human lung transcription regulatory networks generated were subsequently analyzed in the light of GEO datasets GSE1122, GSE1650, and GSE8581, representing studies on changes in gene expression between emphysema subjects and control subjects
Proteins interact with each other via their component domains. An accurate prediction of domain-domain interactions would facilitate the prediction of protein-protein interactions. The Pfam database
Frozen peripheral lung tissue samples used in this study were obtained from two tissue banks: (1) the NHLBI Lung Tissue Research Consortium (University of Colorado Health Sciences Center, Denver, CO); and (2) the iCAPTURE (James Hogg iCAPTURE Centre for Cardiovascular and Pulmonary Research, St. Paul's Hospital, University of British Columbia, Vancouver, BC, Canada). We obtained data on patients' lung function from both established patient registries. Clinical information, samples size, and classification based on Global Initiative for Obstructive Lung Disease (GOLD) for Chronic Obstructive Lung Disease stages of patients with COPD and normal control subjects are summarized in
Characteristic | Normal Samples | COPD Samples |
GOLD Stage (1/2/3/4) | 15 | 0/9/6/0 |
Sex (Male/Female) | 15/0 | 15/0 |
Age (mean ± SD), years | 69.9±14.4 | 68.4±15.1 |
Pack years smoked, (mean ± SD) | 51.4±11.1 | 58.3±8.3 |
FEV1 % predicted, (mean ± SD) | 95.1±9.3 | 31.3±29.9 |
FVC % predicted, (mean ± SD) | 88.7±8.9 | 60.1±18.1 |
FEV1 = Forced expiratory volume at 1 sec; FVC = Function vital capacity; SD = Standard deviation; COPD = Chronic obstructive pulmonary disease.
GOLD (Global Initiative for Chronic Obstructive Lung Disease) Stages:
1 – mild COPD: FEV1≥80% predicted, FEV1:FVC<70%.
2 – moderate COPD: 50%≤FEV1≤80% predicted, FEV1:FVC<70%.
3 – severe COPD: 30%≤FEV1≤50% predicted, FEV1:FVC<70%.
4 – very severe COPD: FEV1<30%predicted or FEV1<50% predicted with chronic respiratory failure, FEV1:FVC<70%.
Pack years: (Packs smoked per day)×(years as a smoker).
Selected genes (TBX2, TBX3, TBX5, CDKN2A, CDKN1A, HDAC2, HDAC5, SIRT1, SIRT5, and CAV1) from our analysis were validated by qRT-PCR. Total mRNA from the peripheral lung tissues from patients with COPD and non-COPD individual's lungs were purified using the Qiagen RNeasy kit (Qiagen, Valencia, CA). qRT-PCR was then performed using inventoried Assay-on-Demand primers and probe sets from Applied Biosystems (Foster City, CA). We used the ABI 7000 Taqman system (Applied Biosystems) to perform these assays. β-actin was used as a normalization control. The analysis was run as previously described
Immunoblots were performed using antibodies for TBX2, CDKN2A, CDKN1A, SIRT1, CAV1, HDAC2, and ACTIN-B (Santa Cruz Biotechnology, Santa Cruz, CA). ACTIN-B was used as a loading control. These immunoblots were performed using protocols as described previously
Fifteen normal, nine mild COPD, and six severe COPD samples were used for q-RTPCR analysis. Four samples per group were used for immunonoblots. All immunoblots were quantified by measuring scanned photographs in ImageJ software (NIH). All statistical analyses were done with student's t-test for comparisons of COPD groups with normal samples as control. Data in graphs were represented as mean values and error bars in the graphs represent standard deviation (SD).
CLR-Generated transcriptional regulatory network of human lung epithelial cells. Following Robust Multi-Array Analysis of a compendium of 158 Affymetrix arrays, the Context Likelihood of Relatedness (CLR) algorithm was used to generate a transcriptional regulatory network (false discovery rate, 0.05). A) A synoptic view of the overall lung epithelial transcriptional regulatory network generated using Gene Ontology genes associated with apoptosis, response to inflammation, and response to oxidative stress. B) An up-close view of TBX3 and nodes directly connected to it in the network generated. C) An up-close view of TBX5 and nodes directly connected to it in the network generated. Larger-sized nodes represent hubs within the network, i.e. human lung epithelium cell genes more highly connected to other genes associated with apoptosis, response to inflammation, and response to oxidative stress. Olive-green nodes represent genes whose median probe set expressions are suppressed in COPD. White nodes represent genes whose median probe set expressions are elevated in COPD.
(TIF)
The states of a large cross-section of human epithelial cell genes differentially expressed in COPD depend on the states of (A) TBX2 and (B) CDKN2A. Following Robust Multi-Array Analysis of a compendium of 109 Affymetrix arrays on the U133A platform, the Context Likelihood of Relatedness (CLR) algorithm was used to generate a transcriptional regulatory network involving all available probe sets (at a CLR likelihood estimate cut-off of 2.5). Olive-green nodes represent genes whose median probe set expressions are suppressed in COPD. White nodes represent genes whose median probe set expressions are elevated in COPD. TBX2 gene expression is suppressed while CDKN2A gene expression is elevated in COPD.
(TIF)
TBX2 is statistically associated with a large cross-section of genes differentially expressed in the COPD lung. Following Robust Multi-Array Analysis, ten biclusters were identified using FABIA from a compendium of 109 Affymetrix human lung epithelial cell microarrays. Focusing only on genes present in each cluster, the Context Likelihood of Relatedness (CLR) algorithm was used to generate Transcriptional Regulatory Networks (false discovery rate, 0.05). The ten networks are merged in this figure. Olive-green nodes represent genes whose median probe set expressions are suppressed in COPD. White nodes represent genes whose median probe set expressions are elevated in COPD. TBX2, the olive node in the center, is either directly or indirectly (by way of one or two intervening nodes) linked with a significant cross-section of genes differentially expressed in the COPD lung.
(TIF)
TBX gene products (captured by red arrows in figure) are predicted to be involved in the direct regulation of 40% of biclusters in the dataset. Following Robust Multi-Array Analysis, ten biclusters were identified using FABIA from a compendium of 109 Affymetrix human lung epithelial cell microarrays. Inferelator version 1.0 was used to infer a minimal set of regulators that explain the expression levels of each of 10 biclusters. TBX2 was predicted to be involved in the direct regulation of biclusters 6 and 10. TBX3 was predicted to be involved in the direct regulation of bicluster 10. TBX5 was predicted to be involved in the direct regulation of biclusters 2 and 8.
(TIF)
Interacting nodes in CLR-generated network and their corresponding likelihood estimates. The data in this table correspond to
(DOC)
Direct Connections to TBX2 in the CLR-Generated Network. The data in this table correspond to
(DOC)
Direct Connections to CDKN2A in the CLR-Generated Network. The data in this table correspond to
(DOC)
A: Membership of
(DOC)
Details of CLR Execution that Generated Results Found in
(DOC)