Different Stimuli for Inference of Gene Regulatory Network

in Rheumatoid Arthritis

Peter Kupfer

, Sebastian Vlaic

, Ren

e Huber

, Raimund W. Kinne

and Reinhard Guthke

Leibnitz Institute for Natural Product Research and Infection Biology, Hans-Kn

oll-Institute,

Beutenbergstr. 11a, D-07745 Jena, Germany

Institute of Clinical Chemistry, Hannover Medical School, Hannover, Germany

Experimental Rheumatology Unit, Department of Orthopedics, University Hospital Jena, Friedrich Schiller University,

Jena, Germany

Keywords:

Network Inference, Rheumatoid Arthritis, TNF-α, TGF-β1, IL-1, PDGF-D.

Abstract:

Since genetic and epigenetic factors are known to be involved in the pathogenesis of rheumatoid arthritis the

search for key players in this disease is one of the most important challenges. For this purpose gene regulatory

networks are one possibility to reveal underlying interactions for different stimuli. In this study we analyzed

the cellular response of synovial ﬁbroblasts to 4 different stimuli. We infered a gene regulatory network that

is able to explain the observed data for stimulation by TNF-α, TGF-β1, IL-1 and PDGF-D simultaneously.

1 INTRODUCTION

Unveiling the dynamic and interlaced nature of gene

regulation is one of the most important aims in sys-

tems biology. The activity of functional gene products

is on the one hand inﬂuenced by transcription fac-

tors (TFs) and co-factors that inﬂuence transcription,

on the other hand by post-translational modiﬁcation

of proteins as well as by the degradation of proteins

and transcripts. Gene regulatory networks (GRNs)

are a possibility to capture relations between molec-

ular entities. Networks are usually represented as

graphs consisting of nodes (representing genes and/or

proteins) and edges (representing molecular interac-

tions such as protein-protein and protein-DNA inter-

actions). In this publication we present a GRN that

integrates data of 4 different stimuli acting on syn-

ovial ﬁbroblasts (SFBs) of rheumatoid arthritis (RA)

patients. RA is a multifactorial polygenic disease with

inﬂammatory impact of synovial joints. The inﬂam-

matory processes are triggered by cytokines and other

immune system-related genes. Several cytokines play

a critical role as mediators of immune regulation but

the precise molecular mechanisms are still unclear.

To investigate the therapeutic effects, cytokines like

TNF-α, TGF-β1, IL-1 and PDGF-D are used in clin-

ical practice. GRNs which describe the cellular re-

sponse to the individual stimulus are helpful in the in-

vestigation of the effects of cytokines. However, si-

multaneous investigations of multiple stimuli offer

the possibility to investigate the cellular actions from

multiple perspectives and therefore provide more in-

formation and better understanding.

2 MATERIALS & METHODS

2.1 Data

Synovial membrane samples were obtained following

tissue excision upon joint replacement/synovectomy

from RA patients (n = 6; all Caucasian) at the

Clinic of Orthopedics, Waldkrankenhaus ’Rudolf

Elle’ (Eisenberg, Germany) as outlined in Kupfer et

al. (Kupfer et al., 2012). Synovial ﬁbroblasts were

stimulated with TNF-α, TGF-β1, IL-1 or PDGF-D for

0, 1, 2, 4, or 12 hours. By using U133 Plus 2.0 RNA

microarrays (Affymetrix, Santa Clara, CA, USA; to-

tal of 60 microarrays) the analysis of gene expression

was performed (for details see Kupfer et al. (Kupfer

et al., 2012)).

To resolve the problem of choosing reliable and

non-contradictory probesets for each transcript, the

alternative Chip Deﬁnition File (CDF) of Ferrari et

al. was used for annotating the genes (Ferrari et al.,

282

Kupfer P., Vlaic S., Huber R., W. Kinne R. and Guthke R..

Different Stimuli for Inference of Gene Regulatory Network in Rheumatoid Arthritis.

DOI: 10.5220/0004196402820287

In Proceedings of the International Conference on Bioinformatics Models, Methods and Algorithms (BIOINFORMATICS-2013), pages 282-287

ISBN: 978-989-8565-35-8

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

2007). The microarray data were preprocessed us-

ing RMA in the default conﬁguration (Irizarry et al.,

2003). Concerning the present study where a combi-

nation of different data sets was used we corrected the

data regarding the creation date were the microarrays

were generated with a modiﬁed version of ComBat

(Kupfer et al., 2012). The Sample Information File

was created as described in the ComBat manual. The

creation date of the microarrays was tagged as ’batch

effect’ and the parameters time point (total of 5), dis-

ease group (RA), and stimulation (TNF-α, TGF-β1,

IL-1 and PDGF-D) were marked as covariates for ev-

ery array.

To detect differentially expressed genes arisen

from microarray experiments, the R-package Limma

was used (Smyth, 2005). By using the expression

data, the contrast and the design matrix, differentially

expressed genes (DEGs) were obtained (Kupfer et al.,

2012). A combination of fold-change and p-value was

recommended by Shi et al. (Shi et al., 2008).

For testing the association between a given gene

list and Gene Ontology (GO) terms the Bioconductor

package GOstats was used (Falcon and Gentleman,

2007).

2.2 Network Inference

The transcription factor binding site integrating

LARS (TILAR) (Hecker et al., 2009) was used to

infer the gene regulatory network shown in Figure

1. The algorithm was modiﬁed to support time se-

ries data. The TILAR concept of modeling permits

gene regulation only via TFs for which the regulated

gene has a transcription factor binding site (TFBS) in

its promoter region and the regulating gene has not.

The advantage of this concept is that a initial prior-

knowledge network template of TF-to-gene relations

is constructed that automatically groups genes with

identical binding sites. This can decrease the num-

ber of possible edges in the network and therefore

lowers the complexity of the inference problem. The

information of the binding sites was extracted from

database Transfac (Matys, 2006). The least angle re-

gression (LARS) algorithm is then used for simulta-

neous selection of the gene-to-TF relations and opti-

mization of their weights. If such a relation is sup-

ported by prior-knowledge (expert knowledge or text

mining programs such as PathwayStudio) the TILAR

algorithm allows its soft integration. This results in

sparse network models with a high content of biolog-

ical prior-knowledge. Optimization of the network

template is performed by a stepwise forward selec-

tion procedure. In that, the algorithm starts with a

network template containing no TF-to-gene relation,

creates the model and calculates the residual sums of

squares (RSS) of the predicted to the measured ex-

pression values. For each TF-to-gene relation, the

new model is created according to the network tem-

plate, LARS is applied and the RSS is calculated. The

TF-to-gene relation that led to the lowest RSS is ﬁxed.

This process is iteratively repeated until the RSS of

the new model is not lower then the RSS of the previ-

ous model.

2.3 Transfac

To extract TFBSs within the promoter regions of the

obtained differentially expressed genes Transfac can

be used (Matys, 2006). One advantage Transfac is

that the database contains experimentally validated

TFBSs.

2.4 PathwayStudio

For the extraction of validation knowledge Path-

wayStudio 9.0 was used with the integrated Mam-

malian database (Nikitin et al., 2003). With Path-

wayStudio it is possible to obtain literature knowl-

edge about the genes, the TFs and the corresponding

associations among each other.

3 RESULTS

3.1 Data Pre-processing

Starting with RMA-normalized data of the 60 arrays

for the time points 0, 1, 2, 4 and 12 hours, standard hi-

erarchical clustering dendrograms (using the R func-

tion hclust with Euclidean distances) were employed

to monitor possible batch effects. Furthermore the ex-

pression values were reviewed regarding differences

in their levels. To correct system biased differences

ComBat was applied.

3.2 Extracting Differentially Expressed

Genes

Limma was used to obtain differentially expressed

genes (DEGs; ﬁltering by the conditions: > 2-fold-

change; p-value of 6 10

−10

) for the question concern-

ing a genetic difference between time point 0 and the

later time points within every single stimulus (TGF-

β1: 507 genes; TNF-α: 582 genes; IL-1: 333 genes

and PDGF-D: 534 genes). Creating the union of the

4 obtained gene lists provided a gene list contain-

ing 1448 genes. For the key question concerning RA

DifferentStimuliforInferenceofGeneRegulatoryNetworkinRheumatoidArthritis

283

Figure 1: GRN describing the cellular response of 4 applied

stimuli.

and osteoarthritis (OA) a previous and published gene

list was used which contains genes differentially ex-

pressed between both diseases for the time point 0

(Kupfer et al., 2012). Using this list and the union

of the differentially expressed genes for all 4 stim-

uli the intersection of 541 genes was extracted. This

constitutes a genetic difference between RA and OA.

For the resulting list a gene enrichment analysis was

done with GOstats (p-value ≤ 0.05). As shown in Ta-

ble 1 the highest ranked GO term was cartilage devel-

opment with a p-value of 1.42

−07

and 18/134 genes.

With regard to the computational complexity of net-

work modeling we have chosen 8 out of the 18 ob-

tained genes of this GO term as highlighted in Table

Table 1: Overrepresented GO categories in the set of DEGs

and the obtained p-values.

TermID Count Size Pvalue Term

GO:0051216 18 134 1.42e-07 cartilage development

GO:0002062 12 58 1.49e-07 chondrocyte differentia-

tion

GO:0048518 133 2898 1.87e-07 positive regulation of bi-

ological process

GO:0009888 61 1016 2.76e-07 tissue development

GO:0048522 122 2649 6.73e-07 positive regulation of cel-

lular process

GO:0048705 18 150 7.87e-07 skeletal system morpho-

genesis

GO:0001503 23 243 1.70e-06 ossiﬁcation

GO:0006357 56 964 2.56e-06 regulation of transcrip-

tion from RNA poly-

merase II promoter

GO:0001649 15 116 2.56e-06 osteoblast differentiation

GO:0008284 38 551 2.61e-06 positive regulation of cell

proliferation

Figure 2: Measured and simulated expression proﬁles.

Dashed lines represent the measured log

-FC of the 4 stim-

uli. The simulated results are shown with solid lines.

3.3 Knowlege Extraction

By using Transfac we extracted all experimental vali-

dated TFBSs for the obtained genes for the following

network modeling. In the TILAR concept of mod-

elling, these TF-to-gene relations are used as a net-

work structure template. Furthermore, we extracted

literature knowledge for the genes, the TFs and the

extracted TFBSs by using Pathway Studio 9.0. There-

fore, we used the genes to be modeled, the obtained

regulating TFs and collected prior knowledge about

BIOINFORMATICS2013-InternationalConferenceonBioinformaticsModels,MethodsandAlgorithms

284

Table 2: Differentially expressed genes of the top-ranked

GO category (GO:0051216 cartilage development). Genes

used for network inference are highlighted in bold.

SYMBOL ENTREZ GENENAME UNIPROT

OSR1 130497 odd-skipped related 1 Q8TAX0

COMP 1311 cartilage oligomeric matrix

protein

P49747

DLX2 1746 distal-less homeobox 2 Q07687

FGF2 2247 ﬁbroblast growth factor 2 P09038

GHR 2690 growth hormone receptor P10912

GLI2 2736 GLI family zinc ﬁnger 2 P10070

GLI3 2737 GLI family zinc ﬁnger 3 P10071

SMAD3 4088 SMAD family member 3 P84022

CHST11 50515 carbohydrate (chondroitin 4)

sulfotransferase 11

Q9NPF2

PTHLH 5744 parathyroid hormone-like

hormone

P12272

CREB3L2 64764 cAMP responsive element

binding protein 3-like 2

Q70SY1

BMP4 652 bone morphogenetic protein

P12644

BMP6 654 bone morphogenetic protein 6 P22004

SNAI2 6591 snail homolog 2 O43623

SNAI1 6615 snail homolog 1 O95863

TGFB1 7040 transforming growth factor,

beta 1

P01137

WNT9A 7483 wingless-type MMTV inte-

gration site family, member

O14904

FGF18 8817 ﬁbroblast growth factor 18 O76093

edges by Pathway Studio. The extracted knowledge

was veriﬁed manually.

3.4 Network Inference

Prior to the network inference, the gene expression

data of the three different experiments (TGF-β1 &

TNF-α, PDGF-D and IL-1) was scaled to a maximum

of 1. This was done to remove the experiment-speciﬁc

bias due to separate pre-processing and analysis. Sub-

sequently, for each gene which did not exceed a abso-

lute log

- fold change of 0.6 in the expression proﬁle

of a treatment, all values of the corresponding proﬁle

were substituted by 0. This way, small changes in the

expression which are likely to resemble noise were re-

moved preventing the algorithm from modeling artiﬁ-

cial rather then the true biological signal. Linear inter-

polation was used to obtain measurements at equidis-

tant time points. We used a constant input function

to simulate the input of cell stimulation. TILAR was

then used for network inference. Identiﬁcation of the

best algorithm parameters was performed in a param-

eter study optimizing the number of integrated prior-

knowledge edges with respect to the error between

the observed and the predicted log

- fold-changes.

The ﬁnal network (Figure 1) was visualized in Cy-

toscape (Smoot et al., 2011) and is composed of 21

nodes (8 modeled genes, 9 bridging TFs and 4 in-

put perturbations) and 51 edges out of which 42 were

inferred by the algorithm (26 gene-to-TF edges and

16 input-to-gene edges). The remaining edges are

prior-knowledge TF-to-gene edges. Furthermore, 8

known direct gene-to-gene relations which were not

used by the algorithm during the inference were iden-

tiﬁed in the constructed network. The node size of the

modeled genes resembles their out-degree as visual-

ized in Figure 1. Together with the highest input-to-

gene edge weights, this property identiﬁes especially

WNT9A as a signal distributing gene. The only gene

which is strongly affected by all four inputs is BMP4.

Regarding the ﬁt of the simulated data compared to

the measured once we got akin dynamics as shown in

Figure 2.

3.5 Discussion

To our knowledge, this is the ﬁrst network model

simulating the initial regulatory steps in SFBs dur-

ing stimulation with four different cytokines/growth

factors. An early event in ﬁbroblast activation by

TNF-α, TGF-β1, IL-1, and PDGF-D is the expres-

sion of genes coding for additional secreted factors

also modulating cellular responses. In this context,

a variety of factors is synergistically induced in re-

sponse to two or more stimuli. For instance, both

TGF-β1 and PDGF-D are able to induce the expres-

sion of members of the TGF family (i.e., BMP4

and TGFB1), WNT9A (also known as WNT14), and

PTHLH (also known as PTHRP). PTHLH may also

be driven by TNF, whereas BMP4 could be induced

by all applied stimuli. Our model conﬁrms other

studies reporting TGF- or TGF signalling-dependent

TGFB1, PTHLH, and WNT9A expression (Bascom

et al., 1989) (Kiriyama et al., 1993) (Spagnoli et al.,

2007) and TNF-dependent PTHLH or BMP4 expres-

sion (Funk et al., 1998) (Horiguchi et al., 2000) in

several cell types.

BMP4, WNT9A, and PTHLH are secreted factors

involved in tissue development, especially cartilage

and bone formation (Bramlage et al., 2006) (Hart-

mann and Tabin, 2001) (Karaplis et al., 1994). In

addition, PTHLH has been shown to mediate anti-

proliferative effects and to induce matrix-degrading

enzymes (Maioli et al., 2002) thus potentially in-

ﬂuencing matrix and bone remodelling. The stim-

ulatory effects of these factors may trigger cellular

characteristics of SFB and other cell types in the

joint such as chondrocytes or osteoblasts (Tsumaki

et al., 2002)(Guo et al., 2004)(Amizuka et al.,

2000)(Ikegame et al., 2001). This cascade reﬂects

DifferentStimuliforInferenceofGeneRegulatoryNetworkinRheumatoidArthritis

285

the inﬂuence of activated SFB on development, func-

tion, and maintenance of the joints or, pathophysio-

logically, on joint destruction, synovitis, and ﬁbrosis,

e.g., in the course of rheumatoid arthritis (Karouza-

kis et al., 2006)(Huber et al., 2006). Another set

of stimulation-dependent genes consists of selected

TFs, e.g., TGF- and IL-1-inducuble GLI3 and TGF-

inducible SNAI2 (also known as SLUG) which are

involved in regulating a variety of developmental pro-

cesses (Johnson and Tabin, 1997) (Nieto et al., 1994)

or TGF- and TNF-inducible CREB3L2 (also known

as BBF2H7) which participates in regulating cell sur-

vival and chondrogenesis (Sheng et al., 2010)(Saito

et al., 2009). In part, these ﬁndings are in good

agreement to the literature, since it has already been

shown that SNAI2 may be induced by TGF (Ao-

matsu et al., 2011). Further TFs are predominately

induced in response to secondarily secreted factors,

e.g., BMP4-induced HOXD13 which is also a regu-

lator of tissue/organ development (Goodman, 2002)

or WNT9A-induced EGR1 which contributes to ba-

sic processes such as tissue repair, (Braddock, 2001),

cellular growth regulation, and apoptosis (Liu et al.,

1998). Here, our model provides new insights into the

intricate successive regulation of TF-induction, since

indirect activation pathways are still inadequately

characterized in the literature. Following expression

and activation, these TFs mediate the regulation of

further target genes (which are not included in Figure

1) deﬁning the superordinate cellular response of SFB

to the (combination of) different stimuli. However, in

the presented network, negative regulatory (feedback)

mechanisms occurring during TNF/TGF/IL-1/PDGF

stimulation are also predicted. They are mediated ei-

ther directly in response to the primary stimuli (e.g.,

inhibition of WNT9A TF, RelA, and GLI3 by TGF-

and IL-1-inducible GLI3) or indirectly in response to

secondarily secreted factors (e.g., POUF1, TRPS1,

ETS2, and VDR or TWIST1, SIM2, SMARCA4,

MITF, SPDEF, and RBPJ in response to WNT9A).

In consequence, a variety of genes is regulated via a

complex network of positively or negatively regulated

TFs representing the interplay between activating and

deactivating features during stimulation with several

cytokines/growth factors.

4 CONCLUSIONS

In this study we were able to present a single net-

work model that describes TGF-β1, TNF-α, IL-1 and

PDGF-D stimulation simultaneously. The ﬁt to all ex-

pression proﬁles of genes included was excellent and

the robustness analysis showed that the obtained net-

work is reliable. Moreover, the biological meaning

of the infered GRN shows new insights like the intri-

cate successive regulation of TF-induction as well as

already published results regarding the single stimuli.

These results are now combined in one network.

ACKNOWLEDGEMENTS

This work was supported by grants from the German

Federal Ministry of Education and Research (BMBF

FKZ 0315719A (PK); 0315736 (SV); ERASysBio

PLUS; LINCONET and VIRTUAL LIVER).

REFERENCES

Amizuka, N., Henderson, J. E., White, J. H., Karaplis,

A. C., Goltzman, D., Sasaki, T., and Ozawa, H.

(2000). Recent studies on the biological action of

parathyroid hormone (pth)-related peptide (pthrp) and

pth/pthrp receptor in cartilage and bone. Histol

Histopathol, 15(3):957–70.

Aomatsu, K., Arao, T., Sugioka, K., Matsumoto, K.,

Tamura, D., Kudo, K., Kaneda, H., Tanaka, K., Fu-

jita, Y., Shimomura, Y., and Nishio, K. (2011). Tgf-

induces sustained upregulation of snai1 and snai2

through smad and non-smad pathways in a human

corneal epithelial cell line. Invest Ophthalmol Vis Sci,

52(5):2437–43.

Bascom, C. C., Wolfshohl, J. R., Coffey, Jr, R. J., Madisen,

L., Webb, N. R., Purchio, A. R., Derynck, R., and

Moses, H. L. (1989). Complex regulation of trans-

forming growth factor beta 1, beta 2, and beta 3 mrna

expression in mouse ﬁbroblasts and keratinocytes by

transforming growth factors beta 1 and beta 2. Mol

Cell Biol, 9(12):5508–15.

Braddock, M. (2001). The transcription factor egr-1: a po-

tential drug in wound healing and tissue repair. Ann

Med, 33(5):313–8.

Bramlage, C. P., H

aupl, T., Kaps, C., Ungeth

um, U., Krenn,

V., Pruss, A., M

uller, G. A., Strutz, F., and Burmester,

G.-R. (2006). Decrease in expression of bone morpho-

genetic proteins 4 and 5 in synovial tissue of patients

with osteoarthritis and rheumatoid arthritis. Arthritis

Res Ther, 8(3):R58.

Falcon, S. and Gentleman, R. (2007). Using GOstats to test

gene lists for GO term association. Bioinformatics,

23(2):257–8.

Ferrari, F., Bortoluzzi, S., Coppe, A., Sirota, A., Safran, M.,

Shmoish, M., Ferrari, S., Lancet, D., Danieli, G. A.,

and Bicciato, S. (2007). Novel deﬁnition ﬁles for hu-

man GeneChips based on GeneAnnot. BMC Bioinfor-

matics, 8:446.

Funk, J. L., Cordaro, L. A., Wei, H., Benjamin, J. B., and

Yocum, D. E. (1998). Synovium as a source of in-

creased amino-terminal parathyroid hormone-related

protein expression in rheumatoid arthritis. a possible

BIOINFORMATICS2013-InternationalConferenceonBioinformaticsModels,MethodsandAlgorithms

286

role for locally produced parathyroid hormone-related

protein in the pathogenesis of rheumatoid arthritis. J

Clin Invest, 101(7):1362–71.

Goodman, F. R. (2002). Limb malformations and the hu-

man hox genes. Am J Med Genet, 112(3):256–65.

Guo, X., Day, T. F., Jiang, X., Garrett-Beal, L., Topol, L.,

and Yang, Y. (2004). Wnt/beta-catenin signaling is

sufﬁcient and necessary for synovial joint formation.

Genes Dev, 18(19):2404–17.

Hartmann, C. and Tabin, C. J. (2001). Wnt-14 plays a piv-

otal role in inducing synovial joint formation in the

developing appendicular skeleton. Cell, 104(3):341–

51.

Hecker, M., Goertsches, R. H., Engelmann, R., Thiesen,

H.-J., and Guthke, R. (2009). Integrative modeling of

transcriptional regulation in response to antirheumatic

therapy. BMC Bioinformatics, 10:262.

Horiguchi, M., Akiyama, H., Ito, H., Shigeno, C., and

Nakamura, T. (2000). Tumour necrosis factor-alpha

up-regulates the expression of bmp-4 mrna but in-

hibits chondrogenesis in mouse clonal chondrogenic

ec cells, atdc5. Cytokine, 12(5):526–30.

Huber, L. C., Distler, O., Tarner, I., Gay, R. E., Gay, S.,

and Pap, T. (2006). Synovial ﬁbroblasts: key play-

ers in rheumatoid arthritis. Rheumatology (Oxford),

45(6):669–75.

Ikegame, M., Ishibashi, O., Yoshizawa, T., Shimomura, J.,

Komori, T., Ozawa, H., and Kawashima, H. (2001).

Tensile stress induces bone morphogenetic protein 4

in preosteoblastic and ﬁbroblastic cells, which later

differentiate into osteoblasts leading to osteogenesis

in the mouse calvariae in organ culture. J Bone Miner

Res, 16(1):24–32.

Irizarry, R. A., Hobbs, B., Collin, F., Beazer-Barclay, Y. D.,

Antonellis, K. J., Scherf, U., and Speed, T. P. (2003).

Exploration, normalization, and summaries of high

density oligonucleotide array probe level data. Bio-

statistics, 4(2):249–64.

Johnson, R. L. and Tabin, C. J. (1997). Molecular models

for vertebrate limb development. Cell, 90(6):979–90.

Karaplis, A. C., Luz, A., Glowacki, J., Bronson, R. T., Ty-

bulewicz, V. L., Kronenberg, H. M., and Mulligan,

R. C. (1994). Lethal skeletal dysplasia from targeted

disruption of the parathyroid hormone-related peptide

gene. Genes Dev, 8(3):277–89.

Karouzakis, E., Neidhart, M., Gay, R. E., and Gay, S.

(2006). Molecular and cellular basis of rheumatoid

joint destruction. Immunol Lett, 106(1):8–13.

Kiriyama, T., Gillespie, M. T., Glatz, J. A., Fukumoto,

S., Moseley, J. M., and Martin, T. J. (1993). Trans-

forming growth factor beta stimulation of parathyroid

hormone-related protein (pthrp): a paracrine regula-

tor? Mol Cell Endocrinol, 92(1):55–62.

Kupfer, P., Guthke, R., Pohlers, D., Huber, R., Koczan, D.,

and Kinne, R. W. (2012). Batch correction of microar-

ray data substantially improves the identiﬁcation of

genes differentially expressed in rheumatoid arthritis

and osteoarthritis. BMC Med Genomics, 5(1):23.

Liu, C., Rangnekar, V. M., Adamson, E., and Mercola, D.

(1998). Suppression of growth and transformation and

induction of apoptosis by egr-1. Cancer Gene Ther,

5(1):3–28.

Maioli, E., Fortino, V., Torricelli, C., Arezzini, B., and

Gardi, C. (2002). Effect of parathyroid hormone-

related protein on ﬁbroblast proliferation and collagen

metabolism in human skin. Exp Dermatol, 11(4):302–

10.

Matys, V. (2006). TRANSFAC(R) and its mod-

ule TRANSCompel(R): transcriptional gene regu-

lation in eukaryotes. Nucleic Acids Research,

34(90001):D108–D110.

Nieto, M. A., Sargent, M. G., Wilkinson, D. G., and Cooke,

J. (1994). Control of cell behavior during vertebrate

development by slug, a zinc ﬁnger gene. Science,

264(5160):835–9.

Nikitin, A., Egorov, S., Daraselia, N., and Mazo, I. (2003).

Pathway studio–the analysis and navigation of molec-

ular networks. Bioinformatics (Oxford, England),

19(16):2155–2157. PMID: 14594725.

Saito, A., Hino, S.-i., Murakami, T., Kanemoto, S., Kondo,

S., Saitoh, M., Nishimura, R., Yoneda, T., Furuichi,

T., Ikegawa, S., Ikawa, M., Okabe, M., and Imaizumi,

K. (2009). Regulation of endoplasmic reticulum

stress response by a bbf2h7-mediated sec23a path-

way is essential for chondrogenesis. Nat Cell Biol,

11(10):1197–204.

Sheng, Z., Li, L., Zhu, L. J., Smith, T. W., Demers, A.,

Ross, A. H., Moser, R. P., and Green, M. R. (2010).

A genome-wide rna interference screen reveals an es-

sential creb3l2-atf5-mcl1 survival pathway in malig-

nant glioma with therapeutic implications. Nat Med,

16(6):671–7.

Shi, L., Jones, W. D., Jensen, R. V., Harris, S. C., and

Perkins, e. a. (2008). The balance of reproducibil-

ity, sensitivity, and speciﬁcity of lists of differentially

expressed genes in microarray studies. BMC Bioin-

formatics, 9(Suppl 9):S10. PMID: 18793455 PMCID:

2537561.

Smoot, M. E., Ono, K., Ruscheinski, J., Wang, P., and

Ideker, T. (2011). Cytoscape 2.8: new features for data

integration and network visualization. Bioinformatics

(Oxford, England), 27(3):431–432. PMID: 21149340.

Smyth, G. K. (2005). Limma: linear models for microarray

data. In Gentleman, R., Carey, V., Dudoit, S., Irizarry,

R., and Huber, W., editors, Bioinformatics and Com-

putational Biology Solutions using R and Bioconduc-

tor, pages 397–420. Springer, New York.

Spagnoli, A., O’Rear, L., Chandler, R. L., Granero-Molto,

F., Mortlock, D. P., Gorska, A. E., Weis, J. A., Lon-

gobardi, L., Chytil, A., Shimer, K., and Moses, H. L.

(2007). Tgf-beta signaling is essential for joint mor-

phogenesis. J Cell Biol, 177(6):1105–17.

Tsumaki, N., Nakase, T., Miyaji, T., Kakiuchi, M., Kimura,

T., Ochi, T., and Yoshikawa, H. (2002). Bone mor-

phogenetic protein signals are required for cartilage

formation and differently regulate joint development

during skeletogenesis. J Bone Miner Res, 17(5):898–

906.

DifferentStimuliforInferenceofGeneRegulatoryNetworkinRheumatoidArthritis

287