Prediction of preadipocyte differentiation by gene expression reveals role of insulin receptor...

A RT I C L E S

NATURE CELL BIOLOGY VOLUME 7 | NUMBER 6 | JUNE 2005 601

Prediction of preadipocyte differentiation by gene expression reveals role of insulin receptor substrates and necdinYu-Hua Tseng1,5, Atul J. Butte2,4,5, Efi Kokkotou1, Vijay K. Yechoor1, Cullen M. Taniguchi1, Kristina M. Kriauciunas1, Aaron M. Cypess1, Michio Niinobe3, Kazuaki Yoshikawa3, Mary Elizabeth Patti1 and C. Ronald Kahn1,6

The insulin/IGF-1 (insulin-like growth factor 1) signalling pathway promotes adipocyte differentiation via complex signalling networks. Here, using microarray analysis of brown preadipocytes that are derived from wild-type and insulin receptor substrate (Irs) knockout animals that exhibit progressively impaired differentiation, we define 374 genes/expressed-sequence tags whose expression in preadipocytes correlates with the ultimate ability of the cells to differentiate. Many of these genes, including preadipocyte factor-1 (Pref-1) and multiple members of the Wnt signalling pathway, are related to early adipogenic events. Necdin is also markedly increased in Irs knockout cells that cannot differentiate, and knockdown of necdin restores brown adipogenesis with downregulation of Pref-1 and Wnt10a expression. Insulin receptor substrate proteins regulate a necdin–E2F4 interaction that represses peroxisome-proliferator-activated receptor γ (PPARγ) transcription via a cyclic AMP response element binding protein (CREB)-dependent pathway. Together these define a key signalling network that is involved in brown preadipocyte determination.

Adipogenesis is a complex process that is highly regulated by positive and negative stimuli, including a variety of hormones and nutritional signals1–3. At the cellular and molecular level, the adipogenesis pro-gramme has been well delineated. Preconfluent preadipocytes undergo proliferation and then growth arrest. When triggered by proper stimuli, these preadipocytes undergo clonal expansion, followed by a second stage of growth arrest and then terminal differentiation4,5. The terminal differentiation process is under a complex series of transcriptional con-trols involving CCAAT/enhancer-binding proteins (C/EBPs), PPARγ and others6, leading to a fully differentiated phenotype with expression of insulin receptors, fatty acid synthase (FAS) and glucose transporter-4 (Glut-4). Moreover, PPARγ coactivator-1α (PGC-1α) serves as a unique regulator of brown fat differentiation, involving the induction of uncou-pling protein-1 (UCP-1) expression7.

Less is known about the events prior to the initiation of this transcrip-tional cascade, during which preadipocytes are released from suppres-sion and become committed to terminal differentiation3,8. Some of the known inhibitors of the preadipocyte–adipocyte transition for white fat include the Wnt family of proteins9, Pref-1 (also known as Dlk1)10, Gata3 (ref. 11), and the retinoblastoma (pRb) family of proteins12. Very little is known about whether similar mechanisms have a role in brown adipocyte differentiation, or which factors and mechanisms regulate the production and function of these inhibitors.

Both insulin and IGF-1 have been shown to affect adipocyte dif-ferentiation in vivo and in vitro3,8. These factors exert their biological effects through a complex signalling pathway that involves activation of their respective cell surface receptors and phosphorylation of several intracellular insulin/IGF-1 receptor substrates (IRSs). In the present study, we show that by using a combination of microarray and cell bio-logical approaches, we can define a highly coordinated pattern of gene expression that predicts the potential of brown preadipocytes to become adipocytes. We also identify key roles for IRS, necdin, E2F4 and CREB-mediated pathways in brown adipogenesis.

RESULTSSearch strategyWe previously reported that brown preadipocytes that are derived from wild-type and different Irs knockout animals exhibit a progressive defect in adipocyte differentiation13, such that wild-type and Irs4 knockout cells fully differentiate into mature adipocytes; Irs2 knockout cells show a slight decrease in differentiation; Irs3 knockout cells show a moderate defect; and Irs1 knockout cells show a severe defect in differentiation (Fig. 1b). This occurs not only at the level of lipid accumulation, but also blocks the normal pattern of progression of the transcriptional regulators of adipo-genesis. We hypothesized that these defects in differentiation potential are programmed in preadipocytes by alterations in gene expression, which

1Research Division, Joslin Diabetes Center, and 2Informatics Program, Children’s Hospital, Harvard Medical School, Boston, MA 02215, USA. 3Division of Regulation of Macromolecular Function, Institute for Protein Research, Osaka University, Osaka 565-0871, Japan. 4Present address: Stanford Medical Informatics, 251 Campus Drive, Room X-215, Stanford, CA 94305-5479, USA. 5These authors contributed equally to this work.6Correspondence should be addressed to C.R.K. (e-mail: [email protected])

Published online: 15 May 2005; DOI: 10.1038/ncb1259

print ncb1259.indd 601print ncb1259.indd 601 18/5/05 10:22:48 am18/5/05 10:22:48 am

Nature Publishing Group© 2005

602 NATURE CELL BIOLOGY VOLUME 7 | NUMBER 6 | JUNE 2005

A RT I C L E S

are linked to the defects in insulin/IGF-1-mediated signalling. To test this, we measured the basal gene expression in preadipocytes of each knockout and wild-type cell line using Affymetrix microarrays.

To define the genes that potentially affect the determination of preadipocyte differentiation, we analysed the microarray data using two approaches (Fig. 1). First, we compared genes whose expression was significantly altered in the basal state between wild-type and Irs1 knockout brown preadipocytes, these being the two genotypes that exhibited the most severe difference in phenotype. A total of 284 probe sets corresponding to 276 genes/expressed-sequence tags (ESTs) passed this statistical filter with p value (pirs) < 0.01 using a Student’s t-test (see Supplementary Information, Table S1). To help define further the most relevant changes in gene expression regarding potential for preadipocyte to adipocyte conversion, we added a second more strin-gent criterion by searching for genes whose basal expression patterns across all cell lines correlated with the phenotypic continuum of abil-ity to undergo adipocyte differentiation (Fig. 1c). Genes were defined as meeting this criterion if the ordering of expression levels across the five types of preadipocytes was either wild type ≥ Irs4 knockout ≥ Irs2 knockout ≥ Irs3 knockout ≥ Irs1 knockout, or the reverse; that is, wild type ≤ Irs4 knockout ≤ Irs2 knockout ≤ Irs3 knockout ≤ Irs1 knockout, as defined by Bartholemew’s test of homogeneity for ordered alterna-tives14. The significance of these ordered changes is referred to hereafter as pup, pdown, qup and qdown, where the q value is the calculated estima-tion of false discovery rate15. A total of 190 probe sets corresponding to 181 genes/ESTs passed this second criterion, and 82 genes/ESTs passed both the first and the second criterion (Fig 1 and see Supplementary Information, Fig. S2). Of these 82 genes, 32 showed increasing expres-sion with decreasing ability to differentiate, whereas 50 genes showed the opposite pattern. These genes were then assigned to different func-tional categories (Tables 1 and 2). Interestingly, many of these genes are known to be involved in early adipogenic events, demonstrating that preadipocyte gene expression signatures can be used as a surrogate to predict the outcome of differentiation.

Morphological modifiersThe first hallmark of early adipogenesis is marked morphological changes as the cells convert from fibroblastic to spherical shapes4. This is associ-ated with decreased actin and tubulin expression16. Consistent with this, expression of several components that are involved in cytoskeletal organi-zation and biogenesis, such as actin, myosin light-chain and troponins, is significantly decreased in Irs1 knockout preadipocytes (Table 1).

In addition, a switch in collagen gene expression occurs during early stages of white adipogenesis17,18. Microarray analysis of the brown preadi-pocytes revealed that the levels of expression of procollagens type III α1 (Col3a1), Col4a1, Col4a2, Col6a1 and Col6a3, as well as tenascin XB were decreased by 13–65% in Irs1 knockout cells, whereas Col7a1 and laminin-α4 were increased by about twofold (Table 1). Of these, Col4a2, Col6a1, Col6a3 and Col7a1 were significant progressors (see Supplementary Information, Fig. S2), and expression of Col6a1 was reversed by IRS1 reconstitution in Irs1 null cells (Fig. 2b).

Metalloproteinases (Mmps) are essential for extracellular matrix (ECM) remodelling through degrading the structural matrix and/or adhesion recep-tors19. The expression of several Mmps, including Mmp-14, is increased in adipose tissue from obese mice20,21. Interestingly, expression of Mmp-14 was significantly decreased in Irs1 knockout preadipocytes (Table 1).

Cell–cell communication is another important aspect of early adipogen-esis. Inhibition of gap junctions promotes the adipocyte phenotype22. In Irs1 knockout preadipocytes, the expression of gap junction membrane channel protein α1 (Gja1) was decreased to 23% of the wild-type levels and was also a significant down-progressor (see Supplementary Information, Fig. S2). In contrast, the expression of Gjb3 was increased by 12-fold (Table 1) and was a significant up-progressor (see Supplementary Information, Fig. S2). This overexpression was reduced in IRS1-reconstituted cells (Fig. 2a).

Pref-1 is a member of the EGF-like transmembrane protein family. It is highly expressed in preadipocytes and is undetectable in mature fat cells. Adipocyte differentiation is inhibited in 3T3-L1 cells that constitutively express Pref-1 (ref. 10). Consistent with these findings, expression of Pref-1 in Irs1 knockout brown preadipocyte was increased by 7.8-fold (Table 1) and was a significant up-progressor (see Supplementary Information, Fig. S2). This overexpression was reduced in IRS1-reconstituted cells (Fig. 2a).

Wnt signalling pathwayThe Wnt signalling pathway has previously been shown to inhibit the differentiation of 3T3-L1 preadipocytes9. Recently, Kang et al. reported that activation of Wnt signalling early in differentiation also blocked brown adipogenesis, whereas activating this pathway in mature brown adipocytes promoted conversion to white adipocytes23. Microarray analy-sis showed that the levels of expression of several members of the Wnt signalling pathway were coordinately altered in the Irs1 knockout cells

Wild type Irs4 KO Irs2 KO Irs3 KO Irs1 KO0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

4,500

5,000

Criterion 1pirs, qirs

276

46

135

Criterion 2apup, qup

Significant difference

≥ ≥≥ ≥

Criterion 2bpdown, qdown

Arb

itrar

y un

its (A

ffym

etrix

)

32

50

374

a

b

c

≤ ≤≤ ≤

Figure 1 Considering phenotypes as a continuum. (a) An example of Trib3 expression progressively increasing in cells that have increasing defects in differentiation. (b) Plates from cells at day 6 of differentiation were stained with Oil Red O and arranged in order, from highest to lowest differentiation potential (left to right). (c) The Trib3 gene met two statistical criteria: criterion 1, expression was significantly different between Irs1 knockout (KO) and wild-type littermate cells in a Student’s t-test with unpaired values and the Welch correction for unequal variance with pirs < 0.01; criterion 2, this gene showed a significant upward ordering of mean basal expression level across the genotypes (see Methods for details). Each gene was evaluated as being differentially expressed between Irs1 knockout and wild-type littermate cells, and as an up- or down-progressor. The counts of genes/ESTs that meet each criterion are listed, as well as the intersection and union of the criteria (right-hand side).




A RT I C L E S

(Table 2). These included a significantly increased expression of Wnt10a, Wnt5a and the transcription factor Tcf7. Expression of Wnt4 and Wnt6 was also increased in Irs1 knockout cells by 1.24- and 2.46-fold, but did not quite meet our criteria of significance with pirs = 0.0197 and 0.0165, respectively. Among these genes, Wnt10a and Tcf7 also showed significant up-progression in cells that showed decreasing abilities to differentiate (see Supplementary Information, Fig. S2). Interestingly, expression of the natu-rally occurring Wnt antagonist Sfrp2 (ref. 24) was significantly decreased in the IRS1-deficient cells to 27% of the wild-type level and was also a

significant down-progressor (see Supplementary Information, Fig. S2). Moreover, expression of phosphatidic acid phosphatase type 2B (Ppap2b or Lpp3), which was previously shown to regulate Wnt signalling by inhib-iting β-catenin-mediated transcription25, was significantly decreased in Irs1 knockout cells to half that of the wild-type level (Table 2) and was a down-progressor (see Supplementary Information, Fig. S2). Associated with, and probably as a result of, these changes, two Wnt-inducible genes, Wisp1 and Rhou (also known as Wrch1), increased in expression by 3.2- and 9.3-fold, respectively, and were significant up-progressors.

Table 1 Genes involved in morphological modifi cation

Probe set Gene name Symbol R†

Cytoskeleton organization

101028_i_at Actin, α, cardiac Actc1 0.05

101029_f_at Actin, α, cardiac Actc1 0.41

97904_at Actin-related protein 3 homologue (yeast) Actr3 1.29

103430_at Drebrin 1 Dbn1 1.66

95529_at Drebrin-like Dbnl 2.37

92541_at Myosin, light polypeptide 1, alkali; atrial, embryonic Myl1 0.11



92541_at Myosin light chain, alkali, fast skeletal muscle Mylf 0.11

92881_at Myosin light chain, phosphorylatable, fast skeletal muscle Mylpf 0.46

101063_at Troponin C, cardiac/slow skeletal Tncc 0.06

98561_at Troponin I, skeletal, slow 1 Tnni1 0.19

100593_at Troponin T2, cardiac Tnnt2 0.25

92885_at Troponin T3, skeletal, fast Tnnt3 0.26

Cell adhesion

101637_at CEA-related cell adhesion molecule 10 Ceacam10 2.60

98331_at Procollagen, type III, α1 Col3a1 0.87

101093_at Procollagen, type IV, α1 Col4a1 0.40

101039_at Procollagen, type IV, α2 Col4a2* 0.44

162459_f_at Procollagen, type VI, α1 Col6a1* 0.44

95493_at Procollagen, type VI, α1 Col6a1* 0.55

101110_at Procollagen, type VI, α3 Col6a3* 0.37

93383_at Procollagen, type VII, α1 Col7a1* 2.07

104587_at Laminin, α4 Lama4 1.97

95016_at Neuropilin* Nrp* 0.44

96272_at Protein tyrosine phosphatase, receptor-type, F Ptprf* 2.95

102669_at Sialoadhesin Sn 4.49

160319_at SPARC-like 1 (mast9, hevin) Sparcl1* 0.09

97519_at Secreted phosphoprotein 1 Spp1 1.94

92877_at Transforming growth factor, β induced, 68 kDa* Tgfbi* 0.60

161907_s_at Tenascin XB Tnxb 0.35

ECM proteolysis and peptidolysis

160118_at Matrix metalloproteinase 14 (membrane-inserted) Mmp14 0.71

Cell communication

101975_at Preadipocyte factor-1 (Pref-1) Dlk1* 7.79

160857_at Ephrin B2 Efnb2* 4.22

100064_f_at Gap junction membrane channel protein α1 Gja1* 0.23

100065_r_at Gap junction membrane channel protein α1 Gja1* 0.23

104232_at Gap junction membrane channel protein β3 Gjb3* 12.32

These genes are signifi cantly altered between wild-type and Irs1 knockout cells, as defi ned by criterion 1. A full list of all 276 genes/ESTs can be found in the Supplementary Information, Table S1 and all raw microarray data are available at http://www.ncbi.nlm.nih.gov/geo/ (accession number GSE2556) and www.diabetesgenome.org. †Ratio of mean expression level in Irs1knockout cells over mean expression level in wild-type littermate cells. *Genes that show a signifi cant up- or down-progression, as defi ned in the text and Supplementary Information, Fig. S2.




A RT I C L E S

Regulators of cell cyclePreadipocytes have to pass through stages of growth arrest and mitotic clonal expansion before becoming committed to terminal dif-ferentiation8. Irs1 knockout cells showed a marked defect in mitotic clonal expansion following treatment with the induction cocktail. Not surprisingly, expression of many genes that are involved in the regulation of the cell cycle and cell growth was altered in the Irs knockout cells (Table 2). In particular, expression of the genes spe-cific to growth arrest, Gas2 and Gas5, as well as two CDK4 inhibitors,

Cdkn2b and Cdkn2c, were significantly altered between wild-type and Irs1 knockout brown preadipocytes, suggesting a dynamic change in the control of exit from and re-entry to the cell cycle, regulated by insulin or IGF-1 in adipogenesis (Table 2). Among these genes, Gas2 and Cdkn2c were significant down-progressors, and Gas5 and Cdkn2b were significant up-progressors. Furthermore, expression of Trib3 (also known as Trb-3) was progressively increased in preadi-pocytes that had increased defects in differentiation (Fig. 1 and see Supplementary Information, Fig. S2), implicating a potential role

Table 2 Genes involved in the Wnt signalling pathway and cell growth

Probe set Gene name Symbol R†

Wnt signalling pathway

96662_at Phosphatidic acid phosphatase type 2B Ppap2b* 0.51

96747_at Ras homologue gene family, member U (Wrch1) Rhou* 9.31

162011_f_at Ras homologue gene family, member U (Wrch1) Rhou* 9.10

93503_at Secreted frizzled-related sequence protein 2 Sfrp2* 0.27

97994_at Transcription factor 7, T-cell specifi c Tcf7* 3.16

102044_at WNT1 inducible signalling pathway protein 1 Wisp1* 3.20

98862_at Wingless-related MMTV integration site 10a Wnt10a* 2.37

99390_at Wingless-related MMTV integration site 5A Wnt5a 4.81

Cell cycle

104097_at Budding uninhibited by benzimidazoles 1 homologue (S. cerevisiae) Bub1 1.65

101900_at Cyclin-dependent kinase inhibitor 2B (p15, inhibits CDK4) Cdkn2b* 1.40

160638_at Cyclin-dependent kinase inhibitor 2C (p18, inhibits CDK4) Cdkn2c* 0.71

101429_at DNA-damage inducible transcript 3 Ddit3 1.28

97327_at Flap structure specifi c endonuclease 1 Fen1 1.39

94338_g_at Growth arrest specifi c 2 Gas2* 0.31

98530_at Growth arrest specifi c 5 Gas5 2.40

98531_g_at Growth arrest specifi c 5 Gas5* 5.64

92323_at Mitogen-activated protein kinase 12 Mapk12* 2.21

100023_at Myeloblastosis oncogene-like 2 Mybl2 3.30

102715_at Nuclear receptor subfamily 2, group F, member 1 Nr2f1 4.99

94932_at Platelet-derived growth factor, α Pdgfa 1.45

92348_at Thyroid hormone receptor α Thra 0.74

Other cell growth and maintenance

161028_at Bone morphogenetic protein 6 Bmp6* 6.41

100127_at Cellular retinoic acid binding protein II Crabp2* 3.56

100567_at Fatty-acid binding protein 4, adipocyte Fabp4* 0.46

97500_g_at Four and a half LIM domains 1 Fhl1* 1.90

101991_at Flavin-containing monooxygenase 1 Fmo1* 0.30

104449_at Glycine receptor, β subunit Glrb* 2.20

95546_g_at Insulin-like growth factor 1 Igf1 1.44

101571_g_at Insulin-like growth factor binding protein 4 Igfbp4 0.46

160832_at Low density lipoprotein receptor Ldlr 1.79

104590_at Myocyte enhancer factor 2C Mef2c 7.72

101059_at Necdin Ndn* 15.33

100697_at Paired box gene 3 Pax3* 0.47

94855_at Prohibitin Phb 2.28

95516_at RAB9, member RAS oncogene family Rab9* 0.46

161067_at Mammalian homologue of Drosophila tribbles Trib3* 6.09

94367_at Uridine monophosphate kinase Umpk* 2.33

These genes are signifi cantly altered between wild-type and Irs1 knockout cells, as defi ned by criterion 1. A full list of all 276 genes/ESTs can be found in the Supplementary Information, Table S1 and all raw microarray data are available at http://www.ncbi.nlm.nih.gov/geo/ (accession number GSE2556) and www.diabetesgenome.org. †Ratio of mean expression level in Irs1knockout cells over mean expression level in wild-type littermate cells. *Genes that show a signifi cant up- or down-progression, as defi ned in the text and Supplementary Information, Fig. S2.




A RT I C L E S

for this protein in regulation of brown adipogenesis. Trib3 is the mammalian homologue of the Drosophila melanogaster gene tribbles that inhibits mitosis in early development26, and has been shown to act as an inhibitor of insulin signalling by directly binding to Akt27. Taken together, these data suggest a potential role for these proteins in the regulation of adipogenesis, perhaps in the proliferative phase of preadipocytes and/or the post-mitotic clonal expansion stage of cells committed to adipogenesis.

IRS1 reconstitution reverses most gene expressionTo confirm that the changes in gene expression were due only to the lack of IRS1, we characterized Irs1 knockout cells that were reconsti-tuted with IRS1, and in which differentiation was fully restored28. The genes that were overexpressed in Irs1 knockout cells, such as Pref-1, Gjb3, Trib3, Mef2c, Wnt6, Wnt10a and necdin (Ndn), were significantly reduced by IRS1 re-expression (Fig. 2a). Likewise, the reduced levels of expression of Col6a1, Gata3, Fabp4 and Sfrp2 were reversed in the reconstituted cells (Fig. 2b). Interestingly, not all the changes in gene expression in the Irs1 knockout cells could be reversed by IRS1 re-expression at this level (Fig. 2a). Whether this reflects that the level of IRS1 re-expression that is achieved by retroviral infection is not suf-ficient to reverse some gene expression, or represents epigenetic altera-tions in the Irs1 knockout cells, is unclear.

Concordant changes in gene expression in vivoTo determine whether the changes in gene expression in cultured preadipocytes that are described above were applicable to the in vivo mechanisms of adipogenesis, we measured gene expression in brown and white adipose tissues that had been isolated from wild-type, Irs1 knockout and Irs1/3 double knockout mice by quantitative RT–PCR (Q-RT–PCR) (Fig. 2c, d). The Irs1 knockout mice exhibit embryonic and postnatal growth retardation29,30, and the Irs1/3 double knockout mice exhibit marked lipoatrophy, with severe insulin resistance31. Consistent with these phenotypes, expression of Pref-1 and Wnt10a was progres-sively increased in brown fat that was taken from mice with increasing defects in the development of adipose tissues. Expression of Pref-1 also showed a progressive increase in white adipose tissue of the knockout and double knockout mice. In addition, expression of PGC-1α in brown adipose tissue was significantly and progressively decreased in mice that had an increasing deficiency in the development of mature adipose tissues. Trib3 and necdin also showed trends towards increased expres-sion in brown adipose tissue that was isolated from knockout animals, but did not reach statistical significance. Taken together, these data reveal that many of the changes in gene expression that were observed in cultured brown preadipocytes derived from mice with defects in insulin signalling also ocur in in vivo brown adipose tissues of mice with similar genetic defects.

Sfrp2

a Upregulated genes

Wnt10a necdin0

1,000

2,000

3,000

4,000

20,000

25,000

**

***

0

500

1,000

1,500

2,000

2,500

3,000

Pref-1 Gjb3 Trib3 Mef2c Wnt6 Wnt5a TCF7 Wisp1

Q-R

T−P

CR

(arb

itrar

y un

its)

Wild type

Wild type

Irs1 KOIrs1 KO + IRS1

*

***

*

* *

0

50

100

150

200

250

300

350

400

Col6a1 Gata3

***

*

**

**

Fabp4

b Downregulated genes

0

400

800

1,200

2,400

2,800

3,200

*

***

Pref-1 Wnt10a

Q-R

T−P

CR

(arb

itrar

y un

its)

necdin Trib3 PGC-1α0

40

80

120

160

200

240

c Brown adipose tissue d White adipose tissue

Pref-1 Wnt10a necdin Trib3 PGC-1α0

100

200

300

400

500

600

700

800***

***

***

***

*

*

Irs1 KOIrs1/3 DKO

Figure 2 Q-RT–PCR analyses confirm changes in gene expression in vitro and in vivo. (a–d) Total RNAs isolated from wild-type, Irs1 knockout and Irs1 knockout cells reconstituted with IRS1. (a, b) as well as brown (c) and white (d) adipose tissues from wild-type, Irs1 knockout and Irs1/3 double knockout (DKO) mice (n = 3–5) were analysed by Q-RT–PCR. Data are presented as mean ± s.e.m. Significance was determined

by an unpaired, unequal t-test. Expression of genes shown in a and b significantly differed between wild-type and Irs1 knockout cells. Asterisks depict statistically significant differences between Irs1 knockout and Irs1 knockout reconstituted with IRS1 (a, b) or between adipose tissues isolated from wild-type and knockout or double knockout mice (c, d). *P < 0.05; **P < 0.01; ***P < 0.001.




A RT I C L E S

The role of necdin in brown preadipocyte differentiationThe NDN gene (human homologue of nectin) is one of a cluster of genes that is deleted in some patients with Prader–Willi syndrome, a neu-rodevelopmental disorder that is characterized by global developmental delay, mental retardation, feeding problems, hypogonadism, hyperphasia and gross obesity32. Necdin functions as a growth suppressor in postmi-totic neurons33 and promotes differentiation and survival of neurons34. Necdin also interacts with viral proteins and transcription factors of the E2F family, demonstrating a functional similarity with the pRB family. However, necdin’s role in other tissues is unknown.

In Irs1 knockout brown preadipocytes, expression of necdin mRNA and protein was significantly and progressively increased in cells with decreasing ability to differentiate (Fig. 3a, b and see Supplementary Information, Fig. S2). IRS1 reconstitution of the Irs1 knockout cells significantly reduced this high level of necdin mRNA and protein as it restored the ability of these cells to differentiate (Figs 2a and 3b).

Because necdin has been shown to interact with E2F4 and E2F1 in the control of gene transcription34, we first examined the expression of E2F4 and E2F1 proteins in wild-type and Irs knockout cells during differentiation (Fig. 3b). E2F4 was expressed at similar levels in wild-type and all Irs knock-out cells at day 0, and this expression was almost completely diminished

at days 2 and 6 in the cells that could differentiate (that is, wild type, Irs4 knockout and Irs2 knockout). By contrast, E2F4 protein levels remained high in cells with impaired ability to differentiate (that is, Irs1 knockout and Irs3 knockout cells). E2F1 protein, on the other hand, was present at similar levels in all cells throughout the whole differentiation course.

To determine the molecular mechanisms by which necdin might regulate brown fat differentiation, we co-transfected wild-type and Irs1 knockout cells with expression constructs of E2F4, E2F1, necdin, and a truncated mutant of necdin that lacked the carboxy-terminal E2F-interacting domain34, together with a luciferase reporter gene construct that contained 3,000 base pairs (bp) of regulatory sequence of the human PPARγ1 gene35 (Fig. 3c, d). In wild-type cells, the E2F4/DP-1 heterodimeric complex was able to stimulate PPARγ1 promoter activity twofold. This was significantly abolished by expression of full-length necdin. The decrease in reporter gene activity seen in wild-type cells that were co-transfected with truncated necdin and E2F4/DP-1, as compared with that induced by E2F4/DP-1, suggests that additional E2F4-interact-ing sites may be present in the amino-terminal region of necdin. The stimulatory effect of E2F4 on PPARγ1 promoter activity was blunted in the Irs1 knockout cells, presumably due to the high level of necdin protein present in these cells. Forced expression of full-length necdin,

a b c

d

3,500

3,000

2,500

2,000

1,500

1,000

500

0

Wild

type

Irs4

KO

Irs2

KO

Irs3

KO

Irs1

KO

Wild

type

Irs4

KO

Irs2

KO

Irs3

KO

Irs1

KO

Irs1

KO+ I

RS1

Arb

itrar

y un

its (A

ffym

etrix

)

Necdin

E2F4

E2F1

p44/42MAPK

Day

0

2

6

0

2

6

0

2

6

0

2

6

PPA

Rγ1

pro

mot

er a

ctiv

ity(a

rbitr

ary

units

)

200

150

100

50

0

PPA

Rγ1

pro

mot

er a

ctiv

ity(a

rbitr

ary

units

)

200

150

100

50

0

Wild type

Wild type

Irs1 KO

Irs1 KO

PGL3γ1

E2F4/DP-1Necdin

Truncated necdin

+−−−

++−−

+++−

++−+

+−−−

++−−

+++−

++−+

PGL3γ1

E2F1/DP-1Necdin

Truncated necdin

+−−−

++−−

+++−

++−+

+−−−

++−−

+++−

++−+

Figure 3 Role of necdin and E2Fs in brown adipogenesis. (a) Expression of necdin progressively increased in preadipocytes with increasing defects in differentiation. (b) Protein expression of necdin, E2F4 and E2F1 in wild-type, various Irs knockout and IRS1-reconstituted cells at day 0, day 2 and day 6 of differentiation. The bottom panel shows similar levels of p44/42 MAP kinase expression among different samples to ensure equal loadings. (c, d) Regulation of PPARγ1 promoter activity

by interactions of necdin and E2F4 (c) or E2F1 (d) in wild-type and Irs1 knockout brown preadipocytes. Relative luciferase activities were measured after transfection with reporter gene construct pGL3γ1 and combinations of expression vectors for DP-1, E2F4, E2F1, necdin and a truncated necdin mutant as indicated. Data are presented as means ± s.e.m. Significance was determined by an unpaired, unequal t-test. *P < 0.05; **P < 0.01; ***P < 0.001.




A RT I C L E S

but not the deletion mutant of necdin, in the Irs1 knockout further reduced PPARγ1 promoter activity. In contrast to E2F4, the E2F1/DP-1 did not have any effect on PPARγ1 transcription in the wild-type cells. Overexpression of full-length necdin, but not the truncated mutant of necdin in these cells decreased PPARγ1 promoter activity, suggesting that this inhibitory effect was mediated via the C-terminal region of necdin. This repressive effect of necdin on PPARγ1 gene transcription was fur-ther demonstrated in the Irs1 knockout cells, because the E2F1/DP-1 complex significantly reduced PPARγ1 promoter activity in these cells with a high level of necdin protein. Co-transfection of necdin together with the E2F1/DP-1 constructs further decreased the reporter gene activity. These data suggest that in brown preadipocytes necdin inhib-its PPARγ1 gene expression via interaction with the E2Fs, providing a potential molecular mechanism through which it regulates adipogenesis. Interestingly, stable overexpression of E2F4 protein in Irs1 knockout cells resulted in a partial restoration of differentiation (data not shown).

Knocking down necdin expression restores differentiationTo examine further the role of necdin in regulation of brown adipocyte differentiation, we knocked down the high level of necdin expression in

Irs1 knockout cells by isolating clones that were stably transfected with a construct containing inhibitory short hairpin RNA (RNAi) that targets necdin. This resulted in an almost 90% reduction of necdin protein expres-sion (Fig. 4a) and led to an about 80% recovery in differentiation compared with wild type (Fig. 4b). In addition, downregulation of necdin in Irs1 knockout cells reversed the defect in mitotic clonal expansion (Fig. 4c) and almost completely suppressed the expression of Pref-1 and Wnt10a, two inhibitors of early adipogenesis, as described above (Fig. 4d). Western blotting analysis confirmed the reduction of Pref-1 protein in Irs1 knock-out preadipocytes with necdin RNAi (Fig. 4a). Furthermore, the markedly decreased levels of PGC-1α and PPARγ mRNA that were observed in the Irs1 knockout preadipocytes were partially restored by knockdown of nec-din (Fig. 4e). This led to a profound restoration in protein expression of the brown-fat-specific marker UCP-1, and mitochondrial proteins ubiquinol-cytochrome c oxidoreductase and the β subunit of ATP synthase (Fig. 4f). Irs1 knockout cells that stably express an unrelated RNAi had no effect on differentiation, mRNA and protein expression, or cell growth (data not shown). These data suggest that downregulation of necdin expression in Irs1 knockout cells not only reverses differentiation but also restores characteristics of mature brown adipocytes.

ab Irs1 KO +

necdin RNAiWild type Irs1 KO

Necdin

Pref-1

Wild

type

Irs1

KOIrs

1 KO + n

ecdin

RNAi

Pref-1 Wnt10a

d

10

102

103

104

105

106

Q-R

T-P

CR

(arb

itrar

y un

its)

**

**

c

0.0

0.5

1.0

1.5

2.0

2.5

3.0

Hours after induction

Rel

ativ

e ce

ll nu

mb

ers

Wild typeIrs1 KO Irs1 KO + necdin RNAi

Wild typeIrs1 KO Irs1 KO + necdin RNAi

Wild-typeIrs1 KO Irs1 KO + necdin RNAi

f

Ctrl Ndn

Day 2 Day 4 Day 7Irs1 KO+ RNAi:

eQ

-RT−

PC

R (a

rbitr

ary

units

)

Ctrl Ndn NdnCtrl

0

20

40

60

80

100

120

PGC-1α PPARγ

***

* UCP-1

Cyto C

β-ATP synthase

GAPDH

0 604020 80

Figure 4 Necdin RNAi restores differentiation, mitotic clonal expansion, and mRNA and protein expression. (a) Western blotting analysis of necdin and Pref-1 in preadipocytes of wild-type, Irs1 knockout and Irs1 knockout cells that stably express necdin RNAi. (b) Oil Red O staining of wild-type, Irs1 knockout and Irs1 knockout cells that stably express necdin RNAi at day 6 of differentiation. (c) Wild-type, Irs1 knockout and Irs1 knockout cells that stably express necdin RNAi were plated at a density of 1.5 × 105 per well in 12-well plates and grown to confluence. Cells were then induced to differentiate using the induction cocktail described in the Methods (0 h). After 48 h of induction, cells were returned to a regular medium supplemented with 20 nM insulin and 1 nM T3. At the times indicated, cells were trypsinized and counted in a haemocytometer. Data are plotted as the relative cell number at each time point to those at 0 h for each

cell line. Error bars represent s.e.m. from triplicate wells. The results are representative of two independent experiments. (d, e) Q-RT–PCR analysis for Pref-1, Wnt10a, PGC-1α and PPARγ using total RNAs isolated from preadipocytes of wild-type, Irs1 knockout and Irs1 knockout cells stably express necdin RNAi. Data are presented as mean ± s.e.m. Asterisks depict statistically significant differences between Irs1 knockout and Irs1 knockout + necdin RNAi by an unpaired, unequal t-test (*P < 0.05; **P < 0.01; ***P < 0.001). (f) Western blotting analysis of mitochondrial proteins UCP-1, ubiquinol-cytochrome c oxidoreductase (Cyto C) and the β subunit of ATP synthase in Irs1 knockout cells that stably express an unrelated control RNAi (Ctl) or necdin RNAi (Ndn) at days 2, 4 and 7 of differentiation. Blots were stripped and re-probed with GAPDH to normalize for variation in loading and transfer of proteins.




A RT I C L E S

Surprisingly, overexpression of necdin in wild-type cells did not block lipid accumulation (see Supplementary Information, Fig. S3), suggesting that either additional factors that are produced by the altered insulin signalling must interact with necdin to produce the full phenotype of inhibition of adipogenesis, or that necdin overexpression must act prior to preadipocyte determination.

The role of CREB in the regulation of gene expressionOne of the most challenging tasks in many studies of expression profil-ing is to identify the transcription factors that drive the changes in gene expression that are observed in the microarray studies. To this end, we examined the 2-kilobase (kb) promoter region of the genes that were

regulated in the Irs1 knockout cells using the programs SRMS (Silico Informatics Systems, Santa Clara, CA) and TRANSFAC. One of the motifs that was revealed by this approach was the response element for CREB. Previously, we found that IGF-1/insulin-stimulated CREB phosphorylation on Ser 133 was markedly diminished in IRS1-deficient cells36. Here we found that the levels of IGF-1-stimulated CREB phos-phorylation were progressively decreased in cells with increasing defects in differentiation (Fig. 5a, b).

To determine whether CREB might be the link between the insulin/IRS pathway and the regulation of gene expression in brown preadi-pocytes, we stably expressed a constitutively active CREB in Irs1 knock-out preadipocytes and examined whether this protein was able to reverse gene expression and/or differentiation. Constitutively active CREB sig-nificantly decreased the high levels of expression of Wnt10a, necdin and Gjb3 (Fig. 5c). Interestingly, however, constitutively active CREB was not able to reverse Pref-1 expression, although there was a slight increase in lipid accumulation in these cells. Thus, insulin/IGF-1-induced CREB activation seems to be required for some of the changes in gene expres-sion that are involved in adipocyte differentiation, but CREB alone is not sufficient to trigger the whole differentiation process.

DISCUSSIONMarked changes in gene expression occur as preadipocytes differenti-ate into adipocytes1,6. However, the factors that influence preadipocyte determination remain poorly understood. In this study, we demonstrate that the insulin/IGF-1/IRS pathway has an important role in the regu-lation of genes that are involved in multiple early adipogenic events in brown preadipocytes, and that this involves a highly coordinated series of gene regulatory events that can be identified even in the preadipocyte. On the basis of the present and previously published data13,28,36, we pro-pose a model for the role of the insulin/IGF-1/IRS pathway in the regu-lation of brown adipogenesis, as illustrated in Fig. 6. The present study indicates that CREB and necdin may serve as potential links between upstream signalling events and downstream gene expression. In par-ticular, necdin seems to have a coordinated role in mediating the CREB effect on downstream gene expression.

Taking advantage of gene expression profiling and brown preadi-pocytes that lack IRS1, -2, -3, or -4 and show varying degrees of defects in differentiation, we have identified 276 genes/ESTs that differ in expres-sion between wild-type and IRS1-deficient brown preadipocytes, and have shown that 82 of these have expression patterns that parallel the defect in differentiation in all five genotypes. Most notably, these changes in gene expression can already be detected in preadipocytes, indicating that the alterations to the insulin/IGF-1 signalling pathway have affected preadipocyte determination.

CREB activation by insulin and other agents is required for 3T3-L1 differentiation37,38. In brown preadipocytes, the level of IGF-1-stimulated CREB phosphorylation was progressively decreased in the Irs knockout cells with increasing defects in differentiation. Overexpression of the constitutively active CREB in Irs1 knockout cells was able to reverse many of the changes in gene expression, but alone only partially restored differentiation. Similarly, overexpression of the dominant-negative CREB in wild-type cells did not completely block brown adipogenesis (data not shown), suggesting that CREB activation in brown preadipocytes is essential for some aspects of gene expression, but is not sufficient to initiate the whole adipogenic programme.

a

b

0

20

40

60

80

100

120

Wild type Irs4 KO Irs2 KO Irs3 KO Irs1 KO

CR

EB

pho

spho

ryla

tion

(per

cent

age

of m

axim

um)

BasalIGF-1

*

Phospho-CREBPhospho-ATF-1

Wild typeIrs4KO

Irs2KO

Irs3KO

Irs1KO

CREB

IGF-1:

c

0

20

40

60

80

100

120

140

160

Q-R

T− P

CR

(arb

itrar

y un

its)

Wnt10a Necdin Gjb3 Pref-1

Irs1 KO+ GFP

Irs1 KO+ CA-CREB

***

*****

180

+− +− +− +− +−

Figure 5 Regulation of gene expression by CREB. (a) A representative blot of basal and IGF-1-stimulated (10 nM; 5 min) Ser 133 phosphorylation of CREB in wild-type and different Irs knockout brown preadipocytes. The blots were stripped and re-probed with a non-phospho-specific anti-CREB antibody to normalize for variation in loading and transfer of proteins. (b) Data from three independent experiments were quantified and presented as means ± s.e.m. (c) Q-RT–PCR analysis for Wnt10a, necdin, GJB-3 and Pref-1 using total RNAs isolated from preadipocytes of Irs1 knockout that stably express a constitutively active form of CREB (CA-CREB), or the control green fluorescent protein (GFP). Inserts show Oil Red O staining of these cells at day 6 of differentiation. Data are presented as means ± s.e.m. Significance was determined by an unpaired, unequal t-test. *P < 0.05; **P < 0.01; ***P < 0.001.




A RT I C L E S

Expression profiling of cells with varying adipogenic potential has allowed us to identify important new players in the regulation of adipogenesis; one of these is necdin, a protein that is functionally reminiscent of the pRb protein family. Necdin is highly upregulated in preadipocytes that have defects in differentiation, suggesting that necdin may function as a negative regulator in the early stages of brown adipogenesis. During the normal adipogenic process, downregulation of necdin expression by insulin or IGF-1 in the preadipocyte seems to be essential for brown adipogenesis. We find that necdin acts upstream of other inhibitors of the preadipocyte–adipocyte transition, such as Pref-1 and Wnt10a, and knocking down necdin expression decreases the overexpression of these genes and restores differentiation.

We also find that at the molecular level, necdin may exert this regula-tory effect via interaction with the E2F family of transcription factors that act on PPARγ gene transcription. E2Fs have been previously shown to have a role in white adipogenesis35,39. The patterns of E2F1 and E2F4 protein expression during brown adipocyte differentiation, however, are quite distinct from those observed in white adipocyte differentiation. In 3T3-L1 cells, E2F4 is increased during differentiation35, whereas in normal brown adipogenesis, expression of E2F4 rapidly declines as dif-ferentiation proceeds. However, in cells with high levels of necdin that

differentiate poorly, such as Irs1 knockout and Irs3 knockout cells, the level of E2F4 protein remains high. This suggests that, as with pRb protein that binds and stabilizes E2F1 by protecting it from ubiquitination40, nec-din may stabilize E2F4 protein during the late stage of differentiation.

Although the exact function of necdin remains unclear, it seems to have an important role in differentiation. Necdin was initially isolated from embryonal carcinoma cells that were induced to undergo neuronal differ-entiation, and is highly expressed in most of the terminally differentiated neurons32,33. However, expression of necdin is not restricted to neuronal tissues41. Boeuf et al. have shown that necdin is one of the genes that is dif-ferentially expressed between white and brown preadipocytes42. Recently, necdin has been found to interact with transcription factor Msx2 to specify myogenic differentiation43,44. Interestingly, the human necdin gene maps to chromosome 15q11.2-q12, a region deleted in Prader–Willi syndrome. The role of these pathways in more common forms of obesity and in diabetes in humans remains to be determined. However, low IRS1 expression in fat has been identified as a marker of insulin resistance, a risk for type 2 diabetes, and is associated with evidence of early vascular complications45. Thus, genes that are altered in expression in the Irs1 knockout preadipocytes provide not only additional cellular markers but also potential molecular targets for drug development to treat obesity, diabetes and its complications.

Insulin/IGF-1receptors

IRS proteins PP

CREB MAP kinase

Ras

Akt

PI(3)K

FoxO1

Necdin + other transcription factors

PPARγ, C/EBPs, PGC-1α and others

Glut-4, FAS, UCP-1, mitochondrial proteins

Regulators of adipogenesis

Components of differentiated state

Morphologicalmodifiers

Inhibitors of preadipocyte toadipocyte transition

Regulators ofcell cycleCol6a1, MMP-14, Gjb3

Pref-1, Wnt-pathway+ E2Fs

Cdk4 inhibitors, Trib3

PPPP

P

Figure 6 Proposed model of regulation of brown adipogenesis by the insulin/IGF-1/IRS pathway. In brown preadipocytes, activation of the cell surface receptors of insulin or IGF-1 results in phosphorylation of the IRS proteins and leads to activation of the PI(3)K and the MAP kinase pathways. These in turn modulate the activity of transcription factors, such as CREB, FoxO1 and others, via a phosphorylation-dependent mechanism. One potential function of these signalling cascades is to suppress expression of necdin, which acts upstream of other inhibitors of the preadipocyte–adipocyte transition, such as the Pref-1 and Wnt pathways. Necdin also represses PPARγ promoter activity

via interaction with E2Fs. The insulin/IGF-1/IRS pathway also regulates expression of several genes that are involved in other early adipogenic events, such as morphological modifiers and regulators of the cell cycle. These lead to the initiation of a transcriptional cascade that involves C/EBPs, PPARγ and PGC-1α, which programs the final changes that are required for full differentiation of the brown adipocytes with expression of Glut-4, FAS, UCP-1 and mitochondrial proteins that are involved in oxidative phosphorylation. Arrows indicate activation and T-shaped lines indicate repression. A dashed line indicates that the connection has not yet been completely defined.




A RT I C L E S

METHODSMaterials. The pGL3γ1 reporter gene construct and expression vectors for E2F1, E2F4 and DP-1 were gifts from L. Fajas (Metabolism and Cancer Laboratory, INSERM, France)35. Plasmid that encodes constitutively active CREB was a gift from M. Montminy (Salk Institute, La Jolla, CA)46. The E2F1, E2F4, Pref-1 and UCP-1 antibodies were purchased from Santa Cruz Biotechnology (Santa Cruz, CA). Anti-phospho-CREB (Ser 133) and anti-CREB antibodies were from Cell Signaling Technology (Beverly, MA) and antibodies against ubiquinol-cyto-chrome c oxidoreductase (core I subunit) and ATP synthase (β subunit) were from Molecular Probes (Eugene, OR). Human recombinant IGF-1 was obtained from Pepro Tec. Inc. (Rocky Hill, NJ). Chemicals were obtained from Sigma (St Louis, MO) unless otherwise specified.

Cell culture and differentiation. Generation of brown preadipocyte cell lines that were derived from newborn wild-type and Irs knockout mice, and differentia-tion of these cells, was as described previously13,28,36. IRS1 reconstitution in Irs1 knockout cells was generated using retroviral-mediated gene transfer28.

To evaluate CREB phosphorylation in response to acute IGF-1 stimulation, cells were grown on a 100-mm dish to about 90% confluent, were serum deprived overnight in medium containing 0.1% BSA and then treated for 5 min with IGF-1 at a final concentration of 10 nM.

Microarray analysis. Brown preadipocytes were grown to confluence and synchronized by overnight serum starvation. Total RNA was isolated with ULTRASPEC RNA Isolation System (Biotecx Laboratories Inc., Houston, TX) following the manufacturer’s instructions. Four independent RNA samples were analysed from each Irs knockout cell line, and in all cases RNAs from two inde-pendent knockout cell lines were pooled for each chip analysis. In addition, three independent clones of wild-type cells (littermates for Irs1 knockout, Irs2 knockout and Irs3 knockout) were separately analysed as controls. This resulted in a total of 28 microarrays being used in this study. Total RNA (28 µg) was subjected to cDNA and cRNA preparations. Adjusted cRNA (15 µg) was hybridized to Affymetrix U74A-v2 arrays (Santa Clara, CA). Intensity values were quantified by MAS 5.0 software (Affymetrix). All chips were subjected to global scaling to a target intensity of 1, 500 to take into account the inherent differences between the chips and their hybridization efficiencies.

Two criteria were used in data analysis to select genes with the potential for strong impact on adipogenesis. The first criterion compared genes whose expres-sion was significantly altered at the basal state between wild-type (four arrays) and Irs1 knockout (four arrays) brown preadipocytes — the two genotypes that exhibited the most severe difference in phenotype. For each gene, we examined the hypothesis H1: mg, wt ≠ mg, irs1ko against H0: mg, wt = mg, irs1ko, where mg, x indicates the mean expression level for gene g in samples obtained from mice with geno-type x. Wild-type littermates for the Irs1 knockout (irs1ko) mice are denoted by wt (four samples). Significance was determined for each gene through the calculation of a p value (pirs) using the Student’s t-test with unpaired values and the Welch correction for unequal variance, implemented as described in ref. 47. As recommended in ref. 48, we did not assume equal variance between the two groups, because previous publications have demonstrated that variance in gene expression measurements might be a function of expression level49.

The threshold of pirs < 0.01 was chosen on the basis of the data, considering multiple comparisons in the following manner: the sample labels (wild type and Irs1 knockout) for the data set of gene expression values were randomly permuted 100 times, and the genes were subjected to the same t-tests each time. A q value (qirs) was then calculated for each gene, based on the ratio of the number of false positives we would expect over the total number of significant genes if the gene were called significant15. We plotted qirs versus the number of significant genes if each q value were used as the threshold, and noted several regions of the curve where the number of significant genes increased with little change in qirs. Using a previously described heuristic, we noted one of these regions at qirs < 0.44, cor-responding to pirs < 0.01, and used this to set our pirs threshold (see Supplementary Information, Fig. S1)15. The pirs and qirs values for the significant genes are reported in the Supplementary Information, Table S1.

To define further the relationship between the altered gene expression and the potential to differentiate, we added a second, more conservative criterion. This involved selection of those genes whose expression patterns in all four Irs knockout brown preadipocytes and all wild-type littermates correlated with the continuum of phenotypes for the potential for adipocyte differentiation (Fig. 1).

Specifically, for each gene g we examined the following hypothesis: H1: mg, pooled wt ≤ mg, irs4ko ≤ mg, irs2ko ≤ mg, irs3ko ≤ mg, irs1ko (up-progression)

againstH0: mg, pooled wt = mg, irs4ko = mg, irs2ko = mg, irs3ko = mg, irs1ko

where mg, x indicates the mean expression level for gene g in samples obtained from mice with genotype x, and where one of the inequality signs must be a strict inequality. ‘pooled wt’ indicates all wild-type littermates (12 samples). We calculated the test statistic Fk using a program that was implemented exactly as described by Bartholemew and was validated using Bartholemew’s sample data14. Similarly, we calculated a separate test statistic for the following hypothesis:

H1: mg, pooled wt ≥ mg, irs4ko ≥ mg, irs2ko ≥ mg, irs3ko ≥ mg, irs1ko (down-progression)againstH0: mg, pooled wt = mg, irs4ko = mg, irs2ko = mg, irs3ko = mg, irs1ko

Two separate calculations were performed scoring both an up- and down-progression for each gene. p values were determined using the resultant F

—

k scores and the F distribution. To test the underlying assumptions in our data using Bartholemew’s test of homogeneity for ordered alternatives, we tested for homo-geneity of variances using Levene’s test and determined that for 15% of the genes, the p value from Levene’s test was under 0.05, indicating that the null hypothesis of homogenous variances would be rejected50. In addition, we also note that the pooled wild-type group contains significantly more samples than any of the other genotype groups. Both of these limitations commonly restrict the interpretation of an analysis of variance (ANOVA) test, and must be considered when interpreting the results of the Bartholemew’s test.

To control for multiple comparisons, q values (qup and qdown) were calculated in a process similar to the one described above. The five sample labels (wild type and Irs1, -2, -3 and -4 knockout) were randomly permuted 100 times. The genes were subjected to the same Bartholemew’s test each time, and qup and qdown were computed as the ratio of the number of false positives that we would expect over the total number of significant genes if the gene were called significant. We plotted qup and qdown versus the number of significant genes if each q value were used as the threshold, and chose the region qup and qdown < 0.03, which corresponds to pup < 0.27 and pdown < 0.47 (see Supplementary Information, Fig. S1). The qup and qdown values for significant genes are reported in the Supplementary Information, Table S1.

All permutation was done by assigning each sample label to be permuted a random number from a uniform distribution, and sorting by this number to shuffle those labels.

Eighteen genes were selected for Q-RT–PCR validation (see below) on the basis of pirs value or biological interests. Of these, 11 genes show concordant (that is, same direction) and significant changes (61.1%). In the remaining seven selected genes, expression of four genes was still concordant with the array data. Thus, a total of 15 out of the 18 genes that were assessed (83.3%) showed concordant changes in Q-RT–PCR and the microarray analysis.

Q-RT-PCR analysis. cDNA was prepared from 1 µg of RNA using the Advantage RTEPCR kit (BD Biosciences, Palo Alto, CA) according to the manufacturer’s instructions. Five microlitres of cDNA were used in a 40 µl PCR reaction (SYBR Green, PE Biosystems, Foster City, CA) containing primers at a concentration of 300 nM each. PCR reactions were run in triplicate and were quantified in the ABI Prism 7700 Sequence Detection System. Results were expressed as arbitrary mRNA units. Sequences of primers used in this study are available upon request.

Transfection and reporter gene assay. Brown preadipocytes were plated at 1 × 105 cells per well on 12-well plates, 18 h before transfection. Cells were trans-fected by LF2000 reagent (Invitrogen, Carlsbad, CA) with 2 µg of pGL3γ1 reporter construct, 1 µg of expression vectors for DP-1, E2F1, E2F4, necdin and the trun-cated necdin mutant, and 0.5 µg of a promoterless Renilla luciferase construct (pRL-0). Cells were lysed 48 h after transfection with 200 µl per well of 1 × passive lysis buffer (Promega, Madison, WI). Luciferase activity was measured by using 20 µl of lysates and the Dual-Luciferase Reporter Assay System (Promega), fol-lowing the manufacturer’s instructions. Relative light units were determined by




A RT I C L E S

quantification of the signal from the Firefly luciferase, normalized with co-trans-fected Renilla luciferase activity in the same sample. Finally, these relative values were normalized against mock transfection. Each expression construct was trans-fected in triplicate wells. The experiments were repeated three times.

Inhibitory RNAi of necdin. The necdin mRNA sequence was analysed for selecting inhibitory RNA sequences using generally accepted design rules, and a 19-nucle-otide (nt) sequence run that starts at nt 1203–1222 was selected. This sequence was synthesized as a double-stranded oligonucleotide that contained the indicated sequence shown below, an 8-nt loop sequence, the reverse complement of the sequence, followed by a pentathymidine termination sequence. To allow for cloning into the U6 vector, a XhoI-compatible end was added to the 5′ end and an XbaI site was added to the 3′ end. The oligonucleotide sequences used are as follows: TCGAGgagtatcccaaatggacagttcaagagactgtccatttgggatactcTTTTT (forward) and CTAGAAAAAgagtatcccaaatggacagtctcttgaactgtccatttgggatactcC (reverse). The forward and reverse oligonucleotides were annealed and ligated into a vector that contained the human U6 promoter (Imgenex, IMG-800, San Diego, CA). This construct, together with a vector encoding an unrelated RNAi, was co-transfected along with a vector encoding a bleomycin-resistant gene into Irs1 knockout brown preadipocytes at a ratio of 3:1 by LF2000 reagent (Invitrogen) and selected with 250 µg ml−1 of the bleomycin analogue Zeocin (Invitrogen) 48 h after transfection.

Western blot analysis. Cells grown on a 100-mm dish were washed twice with ice-cold phosphate-buffered saline and scraped into 0.5 ml of lysis buffer as pre-viously described13. Protein concentrations were determined using the Bradford protein assay (Bio-rad, Hercules, CA). Lysates (30–50 µg) were subjected to SDS–PAGE, followed by immunoblotting using specific antisera and detection with chemiluminescence (ECL, Amersham Pharmacia Biotech, Piscataway, NJ).

Note: Supplementary Information is available on the Nature Cell Biology website.

ACKNOWLEDGEMENTSWe acknowledge A. Norris for constructive comments on the manuscript. We thank M. Montminy (Salk Institute for Biological Studies, La Jolla, CA) and L. Fajas (Metabolism and Cancer Laboratory, INSERM, France) for providing the plasmids used in this study. We acknowledge J. Klein and M. Fasshauer for preparation of cell lines, P. Laustsen, S. Crunkhorn, B. Emanuelli, S. Gesta, D. Espinoza, P. Lin and H. Gami for technical assistance, and J. Marr for excellent secretarial assistance. This work was supported in part by the National Institutes of Health grants DK33201, DK60837 (to C.R.K.), DK101183 (to Y.-H.T.), DK63696 (to A.J.B.) and DK07260 (to A.M.C.) as well as grants from the Lawson Wilkins Pediatric Endocrine Society and the Harvard/MIT Health Sciences and Technology (to A.J.B.).

COMPETING FINANCIAL INTERESTSThe authors declare that they have no competing financial interests.

Received 7 March 2005; accepted 28 April 2005Published online at http://www.nature.com/naturecellbiology.

1. Rangwala, S. M. & Lazar, M. A. Transcriptional control of adipogenesis. Annu. Rev. Nutr. 20, 535–559 (2000).

2. Koutnikova, H. & Auwerx, J. Regulation of adipocyte differentiation. Ann. Med. 33, 556–561 (2001).

3. MacDougald, O. A. & Mandrup, S. Adipogenesis: forces that tip the scales. Trends Endocrinol. Metab. 13, 5–11 (2002).

4. Gregoire, F. M. Adipocyte differentiation: from fibroblast to endocrine cell. Exp. Biol. Med. (Maywood) 226, 997–1002 (2001).

5. Cowherd, R. M., Lyle, R. E. & McGehee, R. E. J. Molecular regulation of adipocyte differentiation. PMID 10, 3–10 (1999).

6. Rosen, E. D., Walkey, C. J., Puigserver, P. & Spiegelman, B. M. Transcriptional regula-tion of adipogenesis. Genes Dev. 14, 1293–1307 (2000).

7. Puigserver, P. et al. A cold-inducible coactivator of nuclear receptors linked to adaptive thermogenesis. Cell 92, 829–839 (1998).

8. Gregoire, F. M., Smas, C. M. & Sul, H. S. Understanding adipocyte differentiation. Physiol. Rev. 78, 783–809 (1998).

9. Ross, S. E. et al. Inhibition of adipogenesis by Wnt signaling. Science 289, 950–953 (2000).

10. Smas, C. M. & Sul, H. S. Pref-1, a protein containing EGF-like repeats, inhibits adi-pocyte differentiation. Cell 73, 725–734 (1993).

11. Tong, Q. et al. Function of GATA transcription factors in preadipocyte-adipocyte transi-tion. Science 290, 134–138 (2000).

12. Chen, P. L., Riley, D. J., Chen, Y. & Lee, W. H. Retinoblastoma protein positively regu-lates terminal adipocyte differentiation through direct interaction with C/EBPs. Genes Dev. 10, 2794–2804 (1996).

13. Tseng, Y. H., Kriauciunas, K. M., Kokkotou, E. & Kahn, C. R. Differential roles of insulin recep-tor substrates in brown adipocyte differentiation. Mol. Cell. Biol. 24, 1918–1929 (2004).

14. Bartholomew, D. J. A test of homogeneity for ordered alternatives. Biometrika 46, 36–48 (1959).

15. Storey, J. D. & Tibshirani, R. Statistical significance for genomewide studies. Proc. Natl Acad. Sci. USA 100, 9440–9445 (2003).

16. Spiegelman, B. M. & Farmer, S. R. Decreases in tubulin and actin gene expression prior to morphological differentiation of 3T3 adipocytes. Cell 29, 53–60 (1982).

17. Weiner, F. R., Shah, A., Smith, P. J., Rubin, C. S. & Zern, M. A. Regulation of collagen gene expression in 3T3-L1 cells. Effects of adipocyte differentiation and tumor necrosis factor alpha. Biochemistry 28, 4094–4099 (1989).

18. Yi, T., Choi, H. M., Park, R. W., Sohn, K. Y. & Kim, I. S. Transcriptional repression of type I procollagen genes during adipocyte differentiation. Exp. Mol. Med. 33, 269–275 (2001).

19. Vu, T. H. & Werb, Z. Matrix metalloproteinases: effectors of development and normal physiology. Genes Dev. 14, 2123–2133 (2000).

20. Chavey, C. et al. Matrix metalloproteinases are differentially expressed in adipose tissue during obesity and modulate adipocyte differentiation. J. Biol. Chem. 278, 11888–11896 (2003).

21. Maquoi, E., Munaut, C., Colige, A., Collen, D. & Lijnen, H. R. Modulation of adipose tissue expression of murine matrix metalloproteinases and their tissue inhibitors with obesity. Diabetes 51, 1093–1101 (2002).

22. Schiller, P. C., D’Ippolito, G., Brambilla, R., Roos, B. A. & Howard, G. A. Inhibition of gap-junctional communication induces the trans-differentiation of osteoblasts to an adipocytic phenotype in vitro. J. Biol. Chem. 276, 14133–14138 (2001).

23. Kang, S. et al. Effects of Wnt signaling on brown adipocyte differentiation and metabo-lism mediated by PGC-1 . Mol. Cell. Biol. 25, 1272–1282 (2005).

24. Kawano, Y. & Kypta, R. Secreted antagonists of the Wnt signalling pathway. J. Cell Sci. 116, 2627–2634 (2003).

25. Escalante-Alcalde, D. et al. The lipid phosphatase LPP3 regulates extra-embryonic vasculogenesis and axis patterning. Development 130, 4623–4637 (2003).

26. Mata, J., Curado, S., Ephrussi, A. & Rorth, P. Tribbles coordinates mitosis and morphogen-esis in Drosophila by regulating string/CDC25 proteolysis. Cell 101, 511–522 (2000).

27. Du, K., Herzig, S., Kulkarni, R. N. & Montminy, M. TRB3: a tribbles homolog that inhibits Akt/PKB activation by insulin in liver. Science 300, 1574–1577 (2003).

28. Fasshauer, M. et al. Essential role of insulin receptor substrate 1 in differentiation of brown adipocytes. Mol. Cell. Biol. 21, 319–329 (2001).

29. Araki, E. et al. Alternative pathway of insulin signaling in mice with targeted disruption of the IRS-1 gene. Nature 372, 186–190 (1994).

30. Tamemoto, H. et al. Insulin resistance and growth retardation in mice lacking insulin receptor substrate-1. Nature 372, 182–186 (1994).

31. Laustsen, P. G. et al. Lipoatrophic diabetes in Irs1-/-/Irs3-/- double knockout mice. Genes Dev. 16, 3213–3222 (2002).

32. Goldstone, A. P. Prader-Willi syndrome: advances in genetics, pathophysiology and treatment. Trends Endocrinol. Metab. 15, 12–20 (2004).

33. Taniura, H., Taniguchi, N., Hara, M. & Yoshikawa, K. Necdin, a postmitotic neuron-specific growth suppressor, interacts with viral transforming proteins and cellular tran-scription factor E2F1. J. Biol. Chem. 273, 720–728 (1998).

34. Kobayashi, M., Taniura, H. & Yoshikawa, K. Ectopic expression of necdin induces dif-ferentiation of mouse neuroblastoma cells. J. Biol. Chem. 277, 42128–42135 (2002).

35. Fajas, L. et al. E2Fs regulate adipocyte differentiation. Dev. Cell 3, 39–49 (2002).36. Tseng,Y. H., Ueki, K., Kriauciunas, K. M. & Kahn, C. R. Differential roles of insulin

receptor substrates in the anti-apoptotic function of insulin-like growth factor-1 and insulin. J. Biol. Chem. 277, 31601–31611 (2002).

37. Reusch, J. E., Colton, L. A. & Klemm, D. J. CREB activation induces adipogenesis in 3T3-L1 cells. Mol. Cell. Biol. 20, 1008–1020 (2000).

38. Klemm, D. J. et al. Insulin-induced adipocyte differentiation. Activation of CREB res-cues adipogenesis from the arrest caused by inhibition of prenylation. J. Biol. Chem. 276, 28430–28435 (2001).

39. Landsberg, R. L. et al. The role of E2F4 in adipogenesis is independent of its cell cycle regulatory activity. Proc. Natl Acad. Sci. USA 100, 2456–2461 (2003).

40. Martelli, F. & Livingston, D. M. Regulation of endogenous E2F1 stability by the retino-blastoma family proteins. Proc. Natl Acad. Sci. USA 96, 2858–2863 (1999).

41. Yang, T. et al. A mouse model for Prader-Willi syndrome imprinting-centre mutations. Nature Genet. 19, 25–31 (1998).

42. Boeuf, S. et al. Differential gene expression in white and brown preadipocytes. Physiol. Genomics 7, 15–25 (2001).

43. Brunelli, S. et al. Msx2 and necdin combined activities are required for smooth muscle differentiation in mesoangioblast stem cells. Circ. Res. 94, 1571–1578 (2004).

44. Kuwajima, T., Taniura, H., Nishimura, I. & Yoshikawa, K. Necdin interacts with the Msx2 homeodomain protein via MAGE-D1 to promote myogenic differentiation of C2C12 cells. J. Biol. Chem. 279, 40484–40493 (2004).

45. Jansson, P. A. et al. A novel cellular marker of insulin resistance and early atherosclero-sis in humans is related to impaired fat cell differentiation and low adiponectin. FASEB J. 17, 1434–1440 (2003).

46. Asahara, H. et al. Chromatin-dependent cooperativity between constitutive and induc-ible activation domains in CREB. Mol. Cell. Biol. 21, 7892–7900 (2001).

47. Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. Numerical Recipes in C: The Art of Scientific Computing (Cambridge Univ. Press, Cambridge,1993).

48. Pan, W., Lin, J. & Le, C. T. How many replicates of arrays are required to detect gene expression changes in microarray experiments? A mixture model approach. Genome Biol. 3, research0022 (2002).

49. Ideker, T., Thorsson, V., Siegel, A. F. & Hood, L. E. Testing for differentially-expressed genes by maximum-likelihood analysis of microarray data. J. Comput. Biol. 7, 805–817 (2000).

50. Levene, H. Contributions to Probability and Statistics: Essays in Honor of Harold Hotelling (eds Olkin, I. et al.) 278–292 (Stanford Univ. Press, Stanford, CA, 1960).



S U P P L E M E N TA RY I N F O R M AT I O N

WWW.NATURE.COM/NATURECELLBIOLOGY 1

Figure S1 Student’s t-test with Welch correction for unequal variance between WT and IRS-1 KO cells. We compared genes whose expression was significantly altered at the basal state between IRS-1 KO (4 arrays) and WT littermate control (4 arrays) brown preadipocytes, the two genotypes that exhibited the most severe difference in phenotype. Significance was determined by selection of genes with pirs < 0.01 in a Student’s t-test with unpaired values, with Welch correction for unequal variance. To assess the effect of multiple comparisons, a q-value (qirs) was calculated for each gene, based on the proportion of false positives we would expect if the gene were called significant. Graph (B) shows that at qirs ≈ 0.44, one finds a region of

the curve where one can add a number of significant genes at little expense to q-value. The point of qirs < 0.44 corresponds to pirs < 0.01, as shown in (A). Each gene was assigned a pup and pdown based on the likelihood of an ordering of increasing or decreasing mean expression levels corresponding to increasing degree of difficulty of adipogenesis across the genotypes, using Bartholemew’s test of homogeneity for ordered alternatives, and qup and qdown were in a manner similar to qirs. The point of qup and qdown < 0.03 (D), corresponding to pup < 0.27 and pdown < 0.47 (C), was used as our threshold.

© 2005 Nature Publishing Group


2 WWW.NATURE.COM/NATURECELLBIOLOGY

Up progression

Probe set name Gene name (symbol) qup WTIRS-4KO

IRS-2KO

IRS-3KO

IRS-1KO

101975_at Delta-like 1 homolog (Drosophila) (Dlk1) 0.004100046_at Methylenetetrahydrofolate dehydrogenase (NADP+ dependent) 2 (Mthfd2) 0.03093866_s_at Matrix gamma-carboxyglutamate (gla) protein (Mglap) 0.025101900_at Cyclin-dependent kinase inhibitor 2B (p15, inhibits CDK4) (Cdkn2b) 0.02698531_g_at Growth arrest specific 5 (Gas5) 0.00496717_at DEAD (Asp-Glu-Ala-Asp) box polypeptide 47 (Ddx47) 0.014102044_at WNT1 inducible signaling pathway protein 1 (Wisp1) 0.02695529_at Drebrin-like (Dbnl) 0.00494367_at Uridine-cytidine kinase 2 (Uck2) 0.006160857_at Ephrin B2 (Efnb2) 0.004104048_at Cysteinyl-tRNA synthetase (Cars) 0.03094238_at RIKEN cDNA 2310046G15 gene (2310046G15Rik) 0.008161067_at Tribbles homolog 3 (Trib3) 0.00493383_at Procollagen, type VII, alpha 1 (Col7a1) 0.010100127_at Cellular retinoic acid binding protein II (Crabp2) 0.00492323_at Mitogen-activated protein kinase 12 (Mapk12) 0.004100600_at CD24a antigen (Cd24a) 0.00497500_g_at Four and a half LIM domains 1 (Fhl1) 0.004104449_at Glycine receptor, beta subunit (Glrb) 0.01896272_at Protein tyrosine phosphatase, receptor type, F (Ptprf) 0.01398862_at Wingless related MMTV integration site 10a (Wnt10a) 0.018101059_at Necdin (Ndn) 0.00497994_at Transcription factor 7, T-cell specific (Tcf7) 0.004104509_at Cholesterol 25-hydroxylase (Ch25h) 0.00496747_at Ras homolog gene family, member U (Rhou) 0.00494815_at 2,3-bisphosphoglycerate mutase (Bpgm) 0.00497733_at Adenosine A2b receptor (Adora2b) 0.02697520_s_at Neuronatin (Nnat) 0.014160905_s_at RIKEN cDNA A030009H04 gene (A030009H04Rik) 0.004161028_at Bone morphogenetic protein 6 (Bmp6) 0.024104232_at Gap junction membrane channel protein beta 3 (Gjb3) 0.012162011_f_at Ras homolog gene family, member U (Rhou) 0.01294776_f_at Killer cell lectin-like receptor, subfamily A, member 4 (Klra4) 0.005

Down progression

Probe set name Gene name (symbol) qdown WTIRS-4KO

IRS-2KO

IRS-3KO

IRS-1KO

93503_at Secreted frizzled-related sequence protein 2 (Sfrp2) <0.00195493_at Procollagen, type VI, alpha 1 (Col6a1) 0.003101110_at Procollagen, type VI, alpha 3 (Col6a3) <0.00199586_at Cystatin C (Cst3) 0.00393077_s_at Lymphocyte antigen 6 complex, locus C (Ly6c) 0.00395516_at RAB9, member RAS oncogene family (Rab9) <0.001100064_f_at Gap junction membrane channel protein alpha 1 (Gja1) 0.00499552_at Snail homolog 2 (Drosophila) (Snai2) 0.00795016_at Neuropilin (Nrp) <0.001101039_at Procollagen, type IV, alpha 2 (Col4a2) 0.00394338_g_at Growth arrest specific 2 (Gas2) <0.001162459_f_at Procollagen, type VI, alpha 1 (Col6a1) 0.00297942_g_at Calpain 6 (Capn6) <0.00197943_at Calpain 6 (Capn6) 0.00294345_at Interleukin 6 signal transducer (Il6st) 0.00292877_at Transforming growth factor, beta induced (Tgfbi) 0.00396662_at Phosphatidic acid phosphatase type 2B (Ppap2b) <0.001160638_at Cyclin-dependent kinase inhibitor 2C (p18, inhibits CDK4) (Cdkn2c) <0.00193188_at Dickkopf homolog 3 (Xenopus laevis) (Dkk3) <0.00198136_at Spermine synthase (Sms) <0.001100435_at Endothelial differentiation, lysophos. acid G-prot-coupled rcptr 2 (Edg2) 0.006103756_at cDNA sequence BC023829 (BC023829) 0.003102223_at Periplakin (Ppl) 0.00293728_at Transforming growth factor beta 1 induced transcript 4 (Tgfb1i4) 0.007104003_at Phosphorylase kinase alpha 2 (Phka2) 0.00293743_at Heat shock factor binding protein 1 (Hsbp1) 0.02793914_at Interleukin 1 receptor, type I (Il1r1) 0.00292555_at Transmembrane 4 superfamily member 6 (Tm4sf6) <0.00193913_at RIKEN cDNA 3110018A08 gene (3110018A08Rik) <0.00195731_at Sestrin 1 (Sesn1) 0.00395019_at Glutathione S-transferase, theta 1 (Gstt1) 0.007104174_at Ectonucleotide pyrophosphatase/phosphodiesterase 1 (Enpp1) <0.001162081_f_at RIKEN cDNA 0610010E05 gene (0610010E05Rik) 0.00499444_at Receptor (calcitonin) activity modifying protein 2 (Ramp2) <0.001104601_at Thrombomodulin (Thbd) <0.001102327_at Amine oxidase, copper containing 3 (Aoc3) 0.003160319_at SPARC-like 1 (mast9, hevin) (Sparcl1) <0.00192357_at A disintegrin and metalloprotease domain 23 (Adam23) 0.00392722_f_at Sine oculis-related homeobox 1 homolog (Drosophila) (Six1) <0.00198924_at ADP-ribosyltransferase 3 (Art3) <0.001102657_at H2.0-like homeo box 1 (Drosophila) (Hlx) 0.00698960_s_at UDP-Gal:betaGlcNAc beta 1,3-galactosyltransferase (B3galt3) 0.006102871_at Eph receptor B6 (Ephb6) 0.00396122_at RIKEN cDNA 2310016A09 gene (2310016A09Rik) 0.003100065_r_at Gap junction membrane channel protein alpha 1 (Gja1) 0.00797941_at Calpain 6 (Capn6) 0.006160530_at Growth hormone inducible transmembrane protein (Ghitm) 0.022100697_at Paired box gene 3 (Pax3) 0.007104519_at Inter-alpha trypsin inhibitor, heavy chain 2 (Itih2) 0.012161173_f_at Interferon activated gene 202B (Ifi202b) 0.018101991_at Flavin containing monooxygenase 1 (Fmo1) 0.003103484_at Popeye domain containing 3 (Popdc3) 0.003100567_at Fatty acid binding protein 4, adipocyte (Fabp4) 0.007103493_at Neuromedin B (Nmb) 0.017

Figure S2 Genes show up- or down-progression correlated with the phenotypes. 33 genes were defined as up-progressors and significantly different between IRS-1 KO cells and their WT controls (upper panel) and 54 genes were similarly different but were down-progressors (lower panel) using the criteria described in Fig. 1 and in Methods. The columns indicate that gene expression in WT and IRS KO cells ordered left-to-right by increasing defects in adipogenesis. Green indicates expression higher than the mean for each gene; red indicates expression lower than the mean. The colour-row at

the top of each of the two sections indicates +2, +1, 0, -1 and -2 standard deviations. The qup or qdown values indicate the degree of significance of the up- and down-progression, respectively, in terms of the expected false discovery rate if each gene were significant. The q-values are listed in italics if the p-value from Levene’s test for the gene was < 0.05, indicating that the gene may not have homogenous variances, which may limit interpretation of the gene using Bartholemew’s test of homogeneity for ordered alternatives. The genes are sorted in descending order by mean expression in WT.



WWW.NATURE.COM/NATURECELLBIOLOGY 3

Figure S3 Effect of overexpression of Necdin in WT cells. (A) Western blotting analysis of necdin in preadipocytes of WT overexpressing necdin or empty vector, and IRS-1 KO cells. (B) Oil Red O staining of WT cells overexpression empty vector or necdin at day 0, 2, 4, and 7 of differentiation. (C) Western blotting analysis of PPARγ, UCP-1 and fatty acid synthase (FAS)

WT overexpressing empty vector [V] or necdin [N] at day 0, 2, 4, and 7 of differentiation. The blots were stripped and reprobed with GAPDH to normalize for variation in loading and transfer of proteins. The results are representative of two independent experiments.

Additional discussion regarding selection of threshold and calculation of statistical power used in microarray analysis:We acknowledge that the threshold qirs close to 0.5 is higher than used in some microarray analyses, and reflects the number of microarrays meas-ured in each category. This was part of our decision to use the power of the genetics and phenotypes by measuring more genotypes with fewer microarrays, rather than more microarrays in fewer genotypes. As would be expected, there are more significant genes at the lower thresholds qup and qdown for the progression analysis because of the increased number of samples available for that analysis. In addition, we performed extensive biological functional validation for these genes, leading to the findings in this study. Since experimental validation was necessary, as well as studies of the function of some of the genes in adipogenesis, we were willing to accept a somewhat higher false-positive rate on the basis that distinguish-ing true positives from false positives was a task we were willing to undertake a priori.

In addition, power is an issue that must be considered. The a priori use of power to calculate the number of replicates needed has been described in an increasing number of publications, but even the authors of these publications note that it is difficult to answer questions about sample size and power when the criteria for selection of genes (for example, a p-value threshold based on a false-discovery rate) and the variance of gene expression measurements are not known in advance1, 2. We had no a priori estimates on the number of genes whose expression may be different between genotypes, and we have no concrete information about the power of the study to detect effects of interest or those plausibly hypothesized to exist, so our study may have been grossly underpowered, markedly overpowered, or adequately powered.1. Pan,W., Lin,J. & Le,C.T. How many replicates of arrays are required to detect gene expression changes in microarray experiments? A mixture model approach. Genome Biol 3,

research0022 (2002).2. Yang,Y.H. & Speed,T. Design issues for cDNA microarray experiments. Nat. Rev. Genet. 3, 579-588 (2002).

Supplementary Table 1: List of 276 genes/Ests significantly altered at basal state between WT and IRS-1 KO cells.


Prediction of preadipocyte differentiation by gene expression reveals role of insulin receptor...

Documents

Transcript of Prediction of preadipocyte differentiation by gene expression reveals role of insulin receptor...