© 1999 Macmillan Magazines Ltd

matergic receptor may subserve behavioural responses to noxiousheat and cold. Although glutamatergic synapses mediate painsensation at all intensities, recent studies26,27 indicate that substanceP and/or neurokinin A are important in sensing intense pain. Thus,it appears that both intensity and modality of pain are coded bydifferent subtypes of postsynaptic receptor and by the release ofdifferent transmitters from primary afferent ®bres. M. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Methods

Spinal slices. Whole-cell recordings were made from spinal slices of rats at

postnatal days 4±21 as described28. EPSCs were evoked at 0.05 Hz with a bipolar

tungsten electrode (stimulus width 0.1 or 0.4 ms) placed at the DREZ or dorsal

root nerve. Monosynaptic EPSCs were identi®ed as described28. Only mono-

synaptic EPSCs were studied. Currents were ®ltered at 1 kHz and digitized at

5 kHz. Bicuculline methiodide (10 mM) and strychnine hydrochloride (1 mM)

were added to the perfusion solution. Statistical comparisons were made using

one-way analyses of variance (ANOVAs; Dunnett test for post-hoc com-

parison) or Student's t-test. P , 0:05 was considered to be signi®cant. SYM

2206 and SYM 2081 were from Tocris±Cookson; other compounds were from

RBI or Sigma.

Cultured neurons. Neurons were dissociated from the dorsal half of spinal

cord slices and maintained for 7±14 days in culture using standard methods15.

Whole-cell pipettes were ®lled with (in mM): 140 caesium glucuronate, 10

EGTA, 10 HEPES, 5 CsCl, 5 MgCl2, 5 ATP and 1 GTP, pH 7.4. Drugs were

dissolved in (in mM) 160 NaCl, 10 HEPES, 2 CaCl2 plus 500 nM tetrodotoxin

(pH 7.4), and applied by rapid local perfusion from a multibarrelled pipette as

described15.

Labelling with ¯uorescent dye. Rats were anaesthetized with halothane (2±

3%) delivered through a nose cone (with 30% O2 balanced with N2) and were

positioned in a stereotaxic apparatus. Fluorescent tracer (0.5±1 ml DiI, 0.2% in

dimethylsulphoxide, or rhodamine latex microspheres) was injected into one

side of the lateral and medial part of the thalamus24. Two days after injection,

the somata of ascending projection cells were visualized under epi¯uorescent

illumination with a rhodamine ®lter set.

Behavioural tests. The tail-¯ick re¯ex and hot-plate (50 8C) and cold-plate

(0 8C) tests were measured as described25. Cold stimuli (0 8C) are believed to be

noxious and to activate speci®c cold nociceptors25. During intrathecal

injection, mice or rats were anaesthetized with halothane (2%). Injections of

drugs (mice, 5 ml; rats, 15 ml) were made with the steel tip of a 30-gauge needle

connected by a 1-ft length of ¯exible PE-10 tubing to a 50-ml syringe. Saline was

used as a control. After the injection, it took 2±3 min for animals to recover.

Data are presented as maximum possible inhibition �MPI� � �response

latency 2 baseline response latency�=�cutoff time 2 baseline response

latency� 3 100.

Received 22 October; accepted 23 November 1998.

1. Bliss, T. V. P. & Collingridge, G. L. A synaptic model of memory: long-term potentiation in thehippocampus. Nature 361, 31±39 (1993).

2. Lerma, J., Morales, M., Vicente, M. A. & Herreras, O. Glutamate receptors of the kainate type and

synaptic transmission. Trends Neurosci. 20, 9±12 (1997).

3. Castillo, P. E., Malenka, R. C. & Nicoll, R. A. Kainate receptors mediate a slow synaptic current in

hippocampal CA3 neurons. Nature 388, 182±186 (1997).4. Vignes, M. & Collingridge, G. L. The synaptic activation of kainate receptors. Nature 388, 179±182

(1997).

5. Mulle, C. et al. Altered synaptic physiology and reduced susceptibility to kainate-induced seizures in

GluR6-de®cient mice. Nature 392, 601±605 (1998).

6. Yaksh, T. L. & Malmberg, A. B. in Textbook of Pain (edited by Wall, P. D. & Melzack, R.) 165±200(Churchill Livingstone, New York, 1994).

7. Yoshimura, M. & Jessell, T. M. Amino acid-mediated EPSPs at primary afferent synapses with

substantia gelatinosa neurones in the rat spinal cord. J. Physiol. (Lond.) 430, 315±335 (1990).

8. Kumazawa, T. & Perl, E. R. Excitation of marginal and substantia gelatinosa neurons in the primate

spinal cord: indications of their place in dorsal horn function organization. J. Comp. Neurol. 177, 417±434 (1978).

9. Light, A. R., Trevino, D. L. & Perl, E. R. Morphological features of functionally de®ned neurons in the

marginal zone and substantia gelatinosa of the spinal dorsal horn. J. Comp. Neurol. 186, 151±171

(1979).10. Bleakman, D. et al. Activity of 2,3-benzodiazepines at naive rat and recombinant human glutamate

receptors in vitro; stereospeci®city and selectivity pro®les. Neuropharmacol. 35, 1689±1702 (1996).

11. Pelletier, J. C., Hesson, D. P., Jones, K. A. & Costa, A.-M. Substituted 1,2-dihydrophthalazines: potent,

selective, and noncompetitive inhibitors of the AMPA receptor. J. Med. Chem. 39, 343±346 (1996).

12. Paternain, A. V., Morales, M. & Lerma, J. Selective antagonism of AMPA receptors unmasks kainaterecceptor-mediated responses in hippocampal neurons. Neuron 14, 185±189 (1995).

13. Wilding, T. J. & Huettner, J. E. Differential antagonism of alpha-amino-3-hydroxy-5-methyl-4-

isoxazolepropionic acid-preferring and kainate-preferring receptors by 2,3-benzodiazepines. Mol.

Pharmacol. 47, 582±587 (1995).

14. Jones, K. A., Wilding, T. J., Huettner, J. E. & Costa, A.-M. Desensitization of kainate receptors by

kainate, glutamate and diasteriomers of 4-methylglutamate. Neuropharmacol. 36, 853±863 (1997).15. Wilding, T. J. & Huettner, J. E. Activation and desensitization of hipocampal kainate receptors. J.

Neurosci. 17, 2713±2721 (1997).

16. Partin, K. M., Patneau, D. K., Winters, C. A., Mayer, M. L. & Buonanno, A. Selective modulation of

desensitization at AMPA versus kainate receptors by cyclothiazide and concanavalin A. Neuron 11,

1069±1082 (1993).17. Todd, A. J., Spike, R. C., Price, R. F. & Neilson, M. Immunocytochemical evidence that neurotension is

present in glutamatergic neurons in the super®cial dorsal horn of the rat. J. Neurosci. 14, 774±784

(1994).

18. ToÈlle, T. R., Berthele, A., ZieglgaÈnsberger, W., Seeburg, P. H. & Wisden, W. The differential expression

of 16 NMDA and non-NMDA receptor subunits in the rat spinal cord and in periagqueductal gray. J.Neurosci. 13, 5009±5028 (1993).

19. Wisden, W. & Seeburg, P. H. A complex mosaic of high-af®nity kainate receptors in rat brain. J.

Neurosci. 13, 3582±3598 (1993).

20. Petralia, R. S., Wang, Y.-X. & Wenthold, R. J. Histological and ultrastructural localization of the

kainate receptor subunits, KA2 and GluR6/7, in the rat nervous system using selective antipeptideantibodies. J. Comp. Neurol. 349, 85±110 (1994).

21. Fitzgerald, M., Butcher, T. & Shortland, P. Development changes in the laminar termination of A ®bre

cutaneous sensory afferents in the rat spinal cord dorsal horn. J. Comp. Neurol. 348, 225±233 (1994).

22. Coggeshall, R. E., Jennings, E. A. & Fitzgerald, M. Evidence that large myelinated primary afferent®bers make synaptic contacts in lamina II of neonatal rats. Brain Res. Dev. Brian Res. 92, 81±90 (1996).

23. Baba, H., Yoshimura, M., Nishi, S. & Shimoji, K. Synaptic responses of substantia gelatinosa neurons

to dorsal column stimulation in rat spinal cord in vitro. J. Physiol. (Lond.) 478, 87±99 (1994).

24. Huang, L.-Y. M., Carlton, S. M. & Willis, W. D. Identi®cation of spinothalamic tract cells in fresh,

un®xed rat spinal cord. J. Neurosci. Methods 14, 91±96 (1985).25. Zhuo, M. NMDA receptor-dependent long term hyperalgesia after tail amputation in mice. Eur. J.

Pharmacol. 349, 211±220 (1998).

26. Cao, Y. Q. et al. Primary afferent tachykinins are required to experience moderate to intense pain.

Nature 392, 390±394 (1998).

27. De Felipe, C. et al. Altered nociception, analgesia and aggression in mice lacking the receptor forsubstance P. Nature 392, 394±397 (1998).

28. Li, P. & Zhuo, M. Silent glutamatergic synapses and nociception in mammalian spinal cord. Nature

393, 695±698 (1998).

Acknowledgements. We thank D. Leander for GYKI 53655. This work was supported in part by grantsfrom NIDA (to M.Z.) and NINDS (to J.E.H.) of the NIH.

Correspondence and requests for materials should be addressed to M.Z. (e-mail: zhuom@morpheus.wustl.edu).

letters to nature

164 NATURE | VOL 397 | 14 JANUARY 1999 | www.nature.com

TheoncogeneandPolycomb-groupgene bmi-1 regulatescellproliferationandsenescencethrough the ink4a locusJacqueline J. L. Jacobs*, Karin Kieboom*, Silvia Marino²,Ronald A DePinho³ & Maarten van Lohuizen*

* Division of Molecular Carcinogenesis and ² Division of Molecular Genetics,The Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam,

The Netherlands³ Dana Farber Cancer Institute, Harvard Medical School,

44 Binney Street (M463), Boston, Massachusetts 02115, USA. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The bmi-1 gene was ®rst isolated as an oncogene that cooperateswith c-myc in the generation of mouse lymphomas1,2. We subse-quently identi®ed Bmi-1 as a transcriptional repressor belongingto the mouse Polycomb group3±6. The Polycomb group comprisesan important, conserved set of proteins that are required tomaintain stable repression of speci®c target genes, such as homeo-box-cluster genes, during development7±9. In mice, the absence ofbmi-1 expression results in neurological defects and severe pro-liferative defects in lymphoid cells, whereas bmi-1 overexpressioninduces lymphomas4,10. Here we show that bmi-1-de®cient pri-mary mouse embryonic ®broblasts are impaired in progressioninto the S phase of the cell cycle and undergo premature senescence.In these ®broblasts and in bmi-1-de®cient lymphocytes, theexpression of the tumour suppressors p16 and p19Arf, which areencoded by ink4a, is raised markedly. Conversely, overexpressionof bmi-1 allows ®broblast immortalization, downregulatesexpression of p16 and p19Arf and, in combination with H-ras,leads to neoplastic transformation. Removal of ink4a dramaticallyreduces the lymphoid and neurological defects seen in bmi-1-

© 1999 Macmillan Magazines Ltd

letters to nature

NATURE | VOL 397 | 14 JANUARY 1999 | www.nature.com 165

de®cient mice, indicating that ink4a is a critical in vivo target forBmi-1. Our results connect transcriptional repression by Poly-comb-group proteins with cell-cycle control and senescence.

To investigate the role of bmi-1 in cell proliferation in biochemicaldetail, we used primary mouse embryonic ®broblasts (MEFs)derived from bmi-1-/- embryos. Bmi-1-/- MEFs have a signi®cantlyreduced proliferation rate compared with bmi-1+/+ MEFs (Fig. 1a).This is because of an S-phase defect, as shown by a reduced rate ofincorporation of the DNA-labelling dye bromodeoxyuridine(BrdU) (Fig. 1b). Cytoplasmic enlargement, ¯attening of the cells,unresponsiveness to growth factors and expression of acidic b-galactosidase11 in 80% of arrested passage 3 bmi-1-/- MEFs indicatespremature entry into senescence, in contrast to wild-type cellswhich ®rst arrest at passage 7 (Fig. 1e). Re-expression of wild-type bmi-1 by retroviral transduction, but not expression of Bmi-1with a mutation in the RING-®nger domain that is incapable ofinducing tumours in vivo10, prevents premature senescence andcompletely restores the proliferative capacity of bmi-1-/- cells,indicating speci®city of bmi-1-/- arrest (Fig. 1c, d). Wild-type MEFsthat overexpress bmi-1 proliferate signi®cantly faster and to higher celldensities than control cells, and have an extended lifespan: they eitherbecome immortal immediately or enter a slow growth period at aboutpassage 13, after which they easily become immortalized (Fig. 1cand Supplementary Information). Similar data were obtained usinga de®ned 3T3 culture scheme (see Supplementary Information).Thus, lack of bmi-1 expression causes premature entry into senescence,whereas overexpression of bmi-1 allows immortalization.

We observed a severe downregulation of cyclin A and cyclin E

protein levels in bmi-1-/- MEFs, which did not occur in bmi-1-/-

MEFs infected with the bmi-1 retrovirus (Fig. 2a). Although nodifferences were observed in amounts of the cyclin-dependentkinase inhibitors p21 and p27 or the tumour suppressor p53, anupregulation of p16 protein in bmi-1-/- MEFs compared with wild-type MEFs was clearly seen (Fig. 2a, b, and data not shown).Conversely, overexpression of bmi-1, but not of the RING-®ngermutant protein, led to rapid downregulation of p16 levels in bmi-1-/-

MEFs and in wild-type MEFs (Fig. 2a, b). Signi®cantly, p16 isthought to be involved in the induction of replicative senes-cence12±14. In agreement with the highly conserved structure andfunction of Polycomb-group (PcG) genes7,8, overexpression of bmi-1 also downregulates p16 protein levels in TiG-3 human primary®broblasts (Fig. 2c). This downregulation is accompanied bydelayed entry into senescence, although the bmi-1-overexpressinghuman cells are not fully immortalized and arrest after 67 morepopulation doublings than normal (Fig. 2d). Human cells showmore complex and multifactorial regulation of the senescencecheckpoint than murine cells12,13,15,16. In agreement with this weobserved no detectable telomerase reactivation upon bmi-1 over-expression in TiG-3 cells using the sensitive telomeric-repeat-ampli®cation protocol (data not shown).

The tumour suppressor p19Arf, which, like p16, is also encoded bythe ink4a locus but is controlled by a separate promoter, is alsoimportant in cell proliferation and senescence17. p16 and, to a lesserextent, p19Arf steady-state transcript levels were upregulated onaverage eightfold and two- to threefold, respectively, in bmi-1-/-

MEFs (Fig. 3a) and splenocytes (Fig. 4a), whereas the neighbouring

bmi-1+/+ bmi-1–/–

p1 p3 p1 p3e

108Rel

ativ

e nu

mbe

r of

cel

ls

6420002

468

1012

1+/+3+/+16–/–21–/–

Days in culture +/+

–/–

Brd

U-p

ositi

ve c

ells

(%

)

30

20

10

0

642000

2

4

6

+/+ C+/+ B–/– C–/– B

Days in cultureRel

ativ

e nu

mbe

r of

cel

ls

64Rel

ativ

e nu

mbe

r of

cel

ls

2000

1

2

3

4

5

+/+ C+/+ Z–/– C–/– Z

Days in culture

a b

c d

Figure 1 Impaired proliferation of bmi-1-/- MEFs is rescued by wild-type but not

mutant bmi-1. a, Growth curves for two independent bmi-1+/+ and bmi-1-/- MEF

cultures at passage 2. The numbers refer to independently derived MEF cultures.

b, Level of BrdU incorporation by asynchronously growing passage 3 bmi-1+/+

and bmi-1-/- MEFs. c, d, Growth curves for passage 2 bmi-1+/+ and bmi-1-/- MEFs

infected at passage 1 with `empty' control LZRS retrovirus (C), LZRS±bmi-1 virus

(B) or LZRS±BZN virus (Z): wild type Bmi-1 rescues proliferation, whereas the BZN

Bmi-1 RING-®nger mutant is unable to do so. e, b-Galactosidase (pH 6) activity in

passage 1 and 3 bmi-1+/+ and bmi-1-/- MEFs.

U C B Zp16

Tub.

T1 T2

p21

Tub.p16

–/–C

–/–B

+/+

C

+/+

B

–/–C

–/–B

+/+

C

+/+

B

a

b

c

d

8642000

4

8

12

16

CB

Rel

ativ

e nu

mbe

r of

cel

ls

Days in culture

C. ECDK4

p16

1+/+

3+/+

4–/–

16–/

–

1+/+

3+/+

4–/–

16–/

–

C. D1

p21

C. A

Tub.

+ Bmi-1

Figure 2 Effects of absence or overexpression of bmi-1 on cell-cycle-regulatory

proteins in primary ®broblasts. a±c, Western blots of cell lysates. a, Cell lysates

from control- or LZRS±bmi-1-infected passage 3 bmi-1+/+ and bmi-1-/- MEFs.

Tubulin (Tub.) serves as loading control. C.D1, cyclin D1; C.A, cyclin A; C.E, cyclin

E; CDK4, cyclin-dependent kinase4. b, Cell lysates from control (C) or LZRS±bmi-1

(B)-infected bmi-1+/+ and bmi-1-/- MEFs, showing p16 and p21 expression 48h

(T1) and 6d (T2) after infection. c, Downregulation of p16 protein in primary human

TiG-3 cells after infection with LZRS±bmi-1 virus (B), but not after infection with

control (C) or BZN-mutant (Z) virus. U, uninfected. d, Growth curves for TiG-3 cells

infected at PDL 50 with control (C) or Bmi-1-expressing (B) retrovirus; growth was

monitored for control-infected cells starting at PDL 68 (nearly senescent) and for

Bmi-1-overexpressing cells at PDL 70 (still proliferating).

© 1999 Macmillan Magazines Ltd

letters to nature

166 NATURE | VOL 397 | 14 JANUARY 1999 | www.nature.com

p15ink4b gene was only modestly affected (1.6-fold upregulation; Fig.3a). This gradual effect on the ink4a locus is in line with transcrip-tional repression by PcG proteins in Drosophila, which is thought tospread over a limited distance from cis elements into adjacentchromatin18±20. As PcG proteins act in multiprotein complexes5,21,22,we analysed senescence entry by MEFs de®cient in another PcGgene. Signi®cant upregulation of p16 messenger RNA levels (three-fold) was also observed in mel18-/- MEFs23; this upregulation wasaccompanied by premature senescence entry at passage 5 (Fig. 3b).The passage of arrest correlated well with relative levels of p16induction. This indicates that the effects on the cell cycle andsenescence are intrinsic to the interaction of PcG proteins inrepressive complexes. Consistent with transcriptional repressionby PcG proteins, downregulation of steady-state levels of p16(fourfold) and p19Arf (2.4-fold) mRNA were easily detected inbmi-1-infected MEFs, but not in cells infected with control virus(Fig. 4b).

Unlike wild-type MEFs but as reported for ink4a-/- MEFs, MEFsoverexpressing bmi-1 could be transformed by oncogenic ras alone,as detected by a focus-formation assay; serial dilutions of the bmi-1retrovirus led to a dose-dependent reduction in the number oftransformed foci (Fig. 4c). This indicates that bmi-1 and rascooperate in neoplastic transformation of primary MEFs; in asimilar way, ras and the loss of p16 and p19Arf expression cooperateto induce neoplastic transformation17,24.

To determine in vivo to what extent the upregulation of p16 andp19Arf is needed to produce the observed proliferative defects inbmi-1-/- mice, mice heterozygous for the bmi-1 gene and for exons 2and 3 of ink4a were intercrossed to generate double-knockout miceas well as heterozygous and wild-type control littermates. In the ®rstlitters (37 offspring out of 5 litters), we observed two mice of anintermediate size, two very small mice and a large group of mice ofnormal size. The two small mice developed ataxia at about 3±4weeks after birth, unlike the intermediate-sized mice. Genotypingcon®rmed that the two small mice were bmi-1-/- ink4a+/+ whereasthe mice of intermediate size were bmi-1-/- ink4a-/-. At 5 weeks ofage, the bmi-1-/- ink4a+/+ mice had progressed to severe ataxia,whereas the two double-knockout mice were still indistinguishablein behaviour from wild-type mice.

Histopathological analysis of bmi-1-/- ink4a-/- mice showed adramatic rescue of the cerebellar defects normally observed in bmi-1-/-

mice4 (Fig. 5a). The size of the cerebellum was comparable to the wild-type cerebellum, the width of the granular cell layer was restored andthe cellularity in the granular and molecular layers increased(Fig. 5b). Notwithstanding the rescued cellularity, the reactiveastrogliosis observed in bmi-1-/- mice4 was only slightly reducedin the double-knockout mice (see Supplementary Information).This result indicates that there may be only a partial rescue of theneurological defects, resulting in a delayed onset and slowerprogression of the pathological process. These results, which are

in line with the signi®cant upregulation of p16 and p19Arf expres-sion in RNA isolated from bmi-1-/- brains (data not shown) andwith the absence of behavioural disorders in double-knockout miceat 5 weeks of age, indicate that ink4a is a critical target for bmi-1 inthis pathological setting too. Signi®cant rescue of thymocyte andsplenocyte cell numbers to up to 50±70% of wild-type numbers wasseen in bmi-1-/- ink4a-/- mice, in contrast with bmi-1-/- ink4a+/+

littermates which retained only 2±4% of wild-type cell numbers(Fig. 5c). Fluorescence-activated cell sorting (FACS) analysis, usingstandard T- and B-cell differentiation cell-surface markers, ofbmi-1-/- ink4a-/- thymocytes and splenocytes showed pro®lesindistinguishable from those of wild-type cells; bmi-1-/- ink4a+/+

littermates, however, showed increased populations of CD4- CD8-,CD3 dull CD25+ or sIg- B220+ immature cells (Fig. 5d). Indepen-dent crosses con®rmed the rescue of cerebellar and lymphoiddefects in another ink4a-/- bmi-1-/- mouse. Full rescue of theproliferative defects and premature senescence entry of bmi-1-/-

MEFs was also observed in bmi-1-/- ink4a-/- MEFs, which grew

p16

+/–+/+ –/–

+/–

p15

+/+ –/–

+/– +/+ –/–

p19ArfProbe: p16

+/+ –/–

b

Probe:

GAPDH

a

GAPDH

Figure 3 p15, p16 and p19Arf mRNA levels in bmi-1+/+, bmi-1-/- and bmi-1+/- MEFs

and in mel-18+/+ and mel-18-/- MEFs. a, Northern blot analysis of p15, p19Arf exon

1b and p16 exon 1a transcripts in passage 3 bmi+/+, bmi-1-/- or bmi-1+/- MEFs. b,

p16mRNA levels in passage 3 mel-18+/+ and mel-18-/- MEFs. GAPDH was

included as a loading control. All signals were quanti®ed relative to the loading

control using a phosphorimager.

M 0x 5x 25x

125x

625x

3,12

5x

0x 5x 25x

125x

625x

3,12

5x

bmi-1+/+ bmi-1–/–

– +

p19Arf

HPRT

p15

HPRT

p16

HPRT

0.5kb

0.5kb

0.2kb

0.2kb

0.5kb

0.2kb

a

b

c

p15 p19Arf p16Probe:

GAPDH

U C B U C B U C B

E1ARas

Bmi-1Ras

1:3 Bmi-1Ras

1:6 Bmi-1Ras

Bmi-1control

Rascontrol

E1AcontrolControl

Figure 4p15, p16 and p19Arf mRNA levels in bmi-1+/+ and bmi-1-/- splenocytes and

control or bmi-1-virus-infected MEFs, and cooperation between Ras and Bmi-1 in

neoplastic transformation. a, Semiquantitative RT-PCR analysis of speci®c p15,

p16 and p19Arf transcripts in bmi-1+/+ and bmi-1-/- spleens. Simultaneous

ampli®cation of hypoxanthine guanine phosphoribosyl transferase (Hprt)

served as an internal control. Water (-) and total MEF cDNA (+) also served as

controls. M, markers; kb, kilobases. The numbers above the lanes represent

serial dilution of ®rst-strand cDNA template used. b, Northern blot analysis of p15,

p19Arf exon 1b and p16 exon 1a transcripts in uninfected (U), control (C) or

LZRS±bmi-1-infected (B) bmi-1-/- MEFs. GAPDH served as a loading control c,

Focus-formation assay using bmi-1+/+ MEFs infected with `empty' retrovirus (top

left) or with retroviruses encoding the proteins indicated. The plates in the top row,

except top left, were infected with retroviruses encoding 12S E1A, Ras or Bmi-1,

and were superinfected with an `empty' retrovirus. The plates in the bottom row

were infected with retroviruses encoding 12S E1A, undiluted Bmi-1, or Bmi-1

diluted 1:3 or 1:6; they were then superinfected with H-ras V12 retrovirus.

© 1999 Macmillan Magazines Ltd

letters to nature

NATURE | VOL 397 | 14 JANUARY 1999 | www.nature.com 167

faster, readily bypassed senescence arrest and were indistinguishablefrom ink4a-/- MEFs24 in this respect (Fig. 5e).

The overall improved health status of the double-knockoutanimals, the normal lymphocyte FACS pro®les and the increasedcellularity of the cerebellar layers indicate a remarkable restorationof lymphoid homeostasis and neurological functions. This, togetherwith the results of our studies of MEFs, provides evidence that a

common, critical cell-cycle-control locus (ink4a) lies at the heart ofproliferative defects in very different affected cell types in bmi-1-/-

mice. As expected for a major downstream target, overexpression ofeither p16 or p19Arf from a retroviral vector in MEFs alreadyoverexpressing bmi-1 still causes cell-cycle arrest (data notshown), indicating that bmi-1 overexpression does not lead tomajor activations of signalling pathways downstream of ink4a. Ingood agreement with ink4a-mediated arrest acting before M2 crisis,karyotype analysis of arrested bmi-1-/- MEFs showed no obviousabnormalities and no signi®cant reduction in telomere length wasobserved by ¯uorescence in situ hybridization analysis (data notshown). These results, together with our observations of humanprimary cells, indicate that the wearing away of telomeres ortelomerase activity are not involved in the observed ink4a-mediatedarrest in bmi-1-/- cells.

Our results place bmi-1 at an early step in tumorigenesis as animmortalizing oncogene that is capable of cooperating with H-rasin transformation of primary cells, and provide a convenient in vitroassay for the effects of PcG dose on cell-cycle regulation. Moreover,our results offer a plausible explanation for the strong cooperationbetween bmi-1 and c-myc or H-ras. Whereas oncogenes such asH-ras, c-myc and E1A induce p19Arf/p53 (refs 25, 26) and/or p16(ref. 15) in primary cells to prevent immortalization during ectopicmitogenic signalling, bmi-1 acts primarily by suppressing p16/cyclinD/retinoblastoma protein (Rb) and/or p19Arf/MDM2/p53 tumour-suppressor pathways, thereby allowing progression through the cellcycle. Both of these latter tumour-suppressor pathways, whendefective, cooperate ef®ciently with c-myc or activated cyclin E inthe induction of mouse lymphomas27,28. The ef®cient in vivocooperation of these oncogenes, and our observations that prolif-erative defects in c-myc-/- rat TGR-1 cells29 cannot be rescued byoverexpression of bmi-1 and that bmi-1-/- MEFs cannot berescued by c-myc-expressing retroviruses, indicates that these onco-genes may ful®l separate, cooperative functions in immortalization(data not shown).

In conclusion, our in vivo and in vitro results identify the tumoursuppressors p16 and p19Arf as critical downstream targets for thePcG gene and oncogene bmi-1, with regard to its effects on cellproliferation and senescence. This establishes for the ®rst time, toour knowledge, a connection between PcG-mediated silencing, cell-cycle regulation and the senescence checkpoint, and shows thatPcG-mediated transcription repression is not only crucial forregulation of Hox genes, but also controls the expression of criticalcell-cycle regulators. PcG-dependent regulation of ink4a may re¯ectthe fact that two of the most frequently disrupted tumour-suppres-sor pathways30 in a variety of neoplasias, p16/cyclin D/Rb andp19Arf/MDM2/p53, each include a protein encoded by the samegene, ink4a16. Thus a tight and coordinated transcriptional regula-tion of this locus is required. The signi®cance of regulation of ink4aby bmi-1 is underscored by our observation that this regulation isconserved in primary MEFs and human ®broblasts, indicating thatoverexpression of Bmi-1 may contribute to human neoplasias thatretain wild-type ink4a. M. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Methods

Cell culture, growth curves and retroviral infection. Cells were maintained

in DMEM medium supplemented with 10% fetal bovine serum (Gibco). Head

and organs of day 14.5 embryos were dissected; fetal tissue was rinsed in PBS,

minced, rinsed twice in PBS and kept overnight on ice with 100 ml trypsin/

EDTA (Gibco). The next day another 100 ml trypsin/EDTA was added and fetal

tissue was incubated for 30 min at 37 8C and subsequently dissociated in

medium. After removal of large tissue clumps, the remaining cells were plated

out in a 175 cm2 ¯ask. After 48 h, con¯uent cultures were frozen down. These

cells were considered as being passage 1 MEFs. For continuous culturing, MEF

cultures were split 1:3, 1 passage being equivalent to 1.8 population doublings

(PDLs). TiG-3 primary human diploid ®broblast cultures were split 1:4, 1

passage being comparable to 2 PDLs. For growth curves, 2:5 3 104 cells were

bmi-1+/+

ink4a+/–bmi-1–/–

ink4a+/+bmi-1–/–

ink4a–/–a

b

M-

G-

P

c

864200

4

8

12

16

Rel

ativ

e ce

ll nu

mbe

r

bmi-1

ink4

a–/–

–/–

+/++/+

–/– –/––/––/–

+/++/+

Days in culture

e

bmi-1

+/+

ink4

a+/–

bmi-1

+/+

ink4

a–/–

bmi-1

–/–

ink4

a+/+

bmi-1

–/–

ink4

a+/+

bmi-1

–/–

ink4

a–/–

bmi-1

–/–

ink4

a–/–

0

20

40

60

80

100

Nuc

leat

ed c

ells

(%

) SpleenThymus

Figure 5 Rescue of bmi-1-/--associated proliferative defects in bmi-1-/- ink4a-/-

mice and MEFs. a, b, Haematoxilin-stained cerebellar sagittal sections. a,

Cerebellum of 5-week-old bmi-1+/+ ink4a+/-, bmi-1-/- ink4a+/+ and bmi-1-/- ink4a-/-

littermates, showing up to 90% restoration of cellularity in double knockouts. b,

Highermagni®cation of molecular (M) and granular (G) layers, showing increased

cellularity and more regular spacing of Purkinje cells (P) in bmi-1-/- ink4a-/- mice

compared with bmi-1-/- ink4a+/+ mice. c, Per cent nucleated cells in spleen and

thymus of mice generated by intercrossing bmi-1+/- ink4a+/- mice. d. Flow-

cytometric analysis of thymocytes and splenocytes of `wild-type' bmi-1+/+ ink4a+/-

,bmi-1-/- ink4a+/+ andbmi-1-/- ink4a-/- mice.Note thealmost complete restoration

to wild-type levels of B- and T-lymphocyte subpopulations in double knockouts. e,

Growth curves: bmi-1-/- ink4a-/- MEFs proliferate like bmi-1+/+ ink4a-/- MEFs.

d

CD

3

CD25

19.3 1.3

1.7

28.1 9.0 17.9 1.2

2.0

Thy

moc

ytes

sIg

B220

0.2 43.1

7.2

0.4 32.0

8.8

0.2 43.0

4.3

bmi-1+/+

ink4a+/–bmi-1–/–

ink4a+/+bmi-1–/–

ink4a–/–

CD

8

CD4

3.1 81.9

2.1

12.9

4.5 29.7

21.2

2.9 81.9

1.6

13.6

Thy

moc

ytes

44.6

77.7 28.7

34.2

78.9

58.8 52.549.5

Spl

enoc

ytes

© 1999 Macmillan Magazines Ltd

letters to nature

168 NATURE | VOL 397 | 14 JANUARY 1999 | www.nature.com

plated in triplicate into 12-well plates. At various time points cells were stained

with crystal violet (Sigma) and the optical density at 590 nm was determined15.

Values were normalized to the optical density at day 0 (20 h after plating).

Phoenix producer cells were used to generate retroviral stocks as described15.

For infection, subcon¯uent passage 1 MEF cultures were incubated at 37 8Cwith viral supernatant in the presence of 4 mg ml-1 polybrene (Sigma). After 6 h

the viral supernatant was diluted 1:3 with complete medium and left on the

cells for 42 h. Appropriate dilutions of viral supernatants were used such that

100% of MEFs were infected.

BrdU-incorporation assay and western blotting. MEFs were grown on

coverslips and incubated for 4 h with 10 mM BrdU (Amersham). Cells were

washed in PBS, ®xed for 15 min at -20 8C in 5% acetic acid/95% ethanol and

incubated in PBS. Fixed cells were incubated for 20 min in 2 M HCl/0.5%

Triton-X100, for 30 min in blocking solution (5% fetal calf serum, 5% normal

goat serum in PBS/0.02% Triton-X100), for 1 h with 1:10 diluted anti-BrdU

monoclonal antibody (DAKO) in blocking solution, and overnight at 4 8C with

1:50 diluted ¯uorescein isothiocyanate (FITC)-conjugated goat anti-mouse

antibody (Jackson Immuno Research Labs) and 4,-6-diamidino-2-phenylin-

dole (DAPI) in blocking solution; this was followed by 3 washes for 5 min each

in PBS/0.02% Triton-X100. Cells were embedded in Vectastain (Vector Labs)

and the percentage of BrdU-labelled cells (FITC:DAPI ratio) was quanti®ed

using a ¯uorescence microscope. For protein analysis, cells were washed with

PBS, scraped and lysed on ice in RIPA buffer (150 mM NaCl, 1% NP40, 0.1%

SDS, 0.5% DOC, 50 mM Tris-HCl, pH 8.0, 2 mM EDTA, pH 8.0, 0.2 mM

phenylmethylsulphonyl¯uoride (PMSF), 0.5 mM dithiothreitol (DTT)).

Cleared lysates were assayed for protein concentration. Equal amounts of

protein were separated on 12.5% SDS±PAGE and transferred to nitrocellulose.

Western blot analysis was according to standard methods using enhanced

chemiluminescence (Amersham). A list of antisera used is available on request.

Expression analysis. Total RNA was extracted using guanidinium thiocya-

nate, separated on 1.2% agarose, transferred to nitrocellulose and hybridized

according to standard procedures with 32P-labelled probes speci®c for exon 1a

of mouse p16, for exon 1b of mouse p19Arf, for mouse p15 or for rat

glyceraldehyde-3-phosphate dehydrogenase (GAPDH). For semiquantitative

reverse transcription with polymerase chain reaction (RT-PCR), ®rst-strand

complementary DNA was generated from 1 mg total RNA using Superscript II

RT (Gibco) and oligo-dT primer according to the manufacturer's instructions.

Primer sequences are available upon request. PCR reactions were performed on

®ve-times serial dilutions of ®rst-strand cDNA in 50 ml containing 1´ Taq PCR

buffer, 1.5 mM MgCl2, 200 mM dNTPs, 0.5 mM of each of four primers, 1 ml of

®rst-strand cDNA template and 1.25 units of Taq DNA Polymerase (Gibco),

using 35 cycles of denaturation (94 8C, 1 min), annealing (60 8C, 45 s) and

extension (72 8C, 2 min). Products were resolved on 2% NuSieve agarose gels.

Generation of bmi-1-/- ink4a-/- mice. bmi-1+/- FVB mice4 and ink4a+/- mice

on a mixed 129/Sv; C57BL/6 genetic background24 were crossed to generate

bmi-1+/- ink4a+/- mice, which were subsequently intercrossed to generate

double-knockout offspring together with control littermates. Mice were

genotyped routinely by PCR or Southern blot analysis of tail DNA.

Cell count and ¯ow-cytometric analysis. Cell suspensions of lymphoid

organs were prepared by mincing the tissue though an open ®lter chamber. Cell

suspensions were depleted of erythrocytes and the number of nucleated cells

was determined with a Casy-1 TT automated cell counter (SchaÈfe, Reutlingen,

Germany). Flow cytometry with standard B- and T-cell differentiation was

done nearly as described10.

Histological analysis. Brains were ®xed in 4% formaldehyde in PBS, paraf®n-

embedded and cut into 4-mm serial sagittal sections. Sections at different levels

were stained with haematoxylin and eosin.

Received 10 September; accepted 12 November 1998.

1. van Lohuizen, M. et al. Identi®cation of cooperating oncogenes in Em-myc transgenic mice byprovirus tagging. Cell 65, 735±752 (1991).

2. Haupt, Y., Alexander, W. S., Barri, G., Klinken, S. P. & Adams, J. M. Novel zinc ®nger gene implicated

as myc collaborator by retrovirally accelerated lymphomagenesis in Em-myc transgenic mice. Cell 65,

753±763 (1991).

3. van Lohuizen, M., Frasch, M., Wientjens, E. & Berns, A. Sequence similarity between the mammalianBmi-1 proto-oncogene and the Drosophila regulatory genes Psc and Su(z)2. Nature 353, 353±355 (1991).

4. van der Lugt, N. M. T. et al. Posterior transformation, neurological abnormalities, and severe

hematopoietic defects in mice with a targeted deletion of the Bmi-1 proto-oncogene. Genes Dev. 8,

757±769 (1994).

5. Alkema, M. J. et al. MPc2, a new murine homologue of the Drosophila Polycomb protein is a memberof the mouse Polycomb transcriptional repressor complex. J. Mol. Biol. 273, 993±1003 (1997).

6. Alkema, M. J., van der Lugt, N. M. T., Bobeldijk, R. C., Berns, A. & van Lohuizen, M. Transformation

of axial skeleton due to overexpression of Bmi-1 in transgenic mice. Nature 374, 724±727 (1995).7. Paro, R. Propagating memory of transcriptional states. Trends Genet. 11, 295±298 (1995).

8. van Lohuizen, M. Functional analysis of mouse Polycomb-group genes. Cell. Mol. Life Sci. 54, 71±79

(1998).

9. Gould, A. Functions of mammalian Polycomb group and trithorax group related genes. Curr. Opin.

Genet. Dev. 7, 488±494 (1997).10. Alkema, M. J., Jacobs, H., van Lohuizen, M. & Berns, A. Perturbation of B and T cell development and

predisposition to lymphomagenesis in Em-Bmi1 transgenic mice require the Bmi1 RING ®nger.

Oncogene 15, 899±910 (1997).

11. Dimri, G. P. et al. A biomarker that identi®es senescent human cells in culture and aging skin in vivo.

Proc. Natl Acad. Sci. USA 92, 9363±9367 (1995).12. Hara, E. et al. Regulation of p16 CDKN2 expression and its implications for cell immortalisation and

senescence. Mol. Cell. Biol. 16, 859±867 (1996).

13. Alcorta, D. A. et al. Involvement of the cyclin-dependent kinase inhibitor p16 (INK4a) in replicative

senescence of normal human ®broblasts. Proc. Natl Acad. Sci. USA 93, 13742±13747 (1996).

14. Zindy, F., Quelle, D. E., Roussel, M. F. & Sherr, C. J. Expression of the p16ink4a tumour suppressorversus other ink4 family members during development and aging. Oncogene 15, 203±211 (1997).

15. Serrano, M., Lin, A. W., McCurrach, M. E., Beach, D. & Lowe, S. W. Oncogenic ras provokes

premature cell senescence associated with accumulation of p53 and p16ink4a. Cell 88, 593±602

(1997).16. Haber, D. A. Splicing into senescence: the curious case of p16 and p19Arf. Cell 91, 555±558 (1997).

17. Kamijo, T. et al. Tumour suppression at the mouse ink4a locus mediated by the alternative reading

frame product p19Arf. Cell 91, 649±659 (1997).

18. Strutt, H., Cavalli, G. & Paro, R. Colocalisation of Polycomb protein and GAGA factor on regulatory

elements responsible for the maintenance of homeotic gene expression. EMBO J. 12, 3621±3632(1997).

19. Cavalli, G. & Paro, R. The Drosophila Fab-7 chromosomal element conveys epigenetic inheritance

during mitosis and meiosis. Cell 93, 505±518 (1998).

20. Pirotta, V. PcG complexes and chromatin silencing. Curr. Opin. Genet. Dev. 7, 249±258 (1997).

21. Alkema, M. J. et al. Identi®cation of Bmi1-interacting proteins as constituents of a multimericmammalian Polycomb complex. Genes Dev. 11, 226±240 (1997).

22. Gunster, M. J. et al. Identi®cation and characterisation of interactions between the vertebrate

Polycomb-group protein BMI1 and the human homologues of Polyhomeotic. Mol. Cell. Biol. 17,

2326±2335 (1997).

23. Akasaka, T. et al. A role for mel-18, a Polycomb group-related vertebrate gene, during theanteroposterior speci®cation of the axial skeleton. Development 122, 1513±1522 (1996).

24. Serrano, M. et al. Role of the ink4a locus in tumour suppression and cell mortality. Cell 85, 27±37

(1996).

25. Zindy, F. et al. Myc signalling via the ARF tumour suppressor regulates p53-dependent apoptosis and

immortalisation. Genes Dev. 12, 2424±2433 (1998).26. de Stanchina, E. et al. E1A signalling to p53 involves the p19ARF tumour suppressor. Genes Dev. 12,

2434±2442 (1998).

27. Blyth, K. et al. Synergy between a human c-myc transgene and p53 null genotype in murine thymic

lymphomas: contrasting effects of homozygous and heterozygous p53 loss. Oncogene 10, 1717±1723

(1995).28. Bodrug, S. E. et al. Cyclin D1 transgene impedes lymphocyte maturation and collaborates in

lymphomagenesis with the myc gene. EMBO J. 13, 2124±2130 (1994).

29. Mateyak, M. K., Obaya, A. J., Adachi, S. & Sedivy, J. M. Phenotypes of c-Myc-de®cient Rat ®broblasts

isolated by targeted homologous recombination. Cell Growth Differ. 8, 1039±1048 (1997).

30. Hall, M. & Peters, G. Genetic alterations of cyclins, cyclin-dependent kinases and cdk inhibitors inhuman cancer. Adv. Cancer Res. 68, 67±108 (1996).

Supplementary information is available on Nature's World-Wide Web site (http://www.nature.com) oras paper copy from the London editorial of®ce of Nature.

Acknowledgements. We thank G. Peters for p16 and p15 cDNA plasmids; H. Koseki and J. Deschamps formel-18-/- embryos; T. Ide for the TiG-3 cells; J. Sedivy for c-myc-/- TGR-1 cells; S. Lowe, B. Amati, C. Sherr,M. Ewen, D. Peeper and R. Bernards for gifts of recombinant retroviral vectors; G. Nolan for providingphoenix eco- and amphotropic packaging cell lines and LZRS-IRES-EGFP retroviral vectors; N. van derLugt for constructing the LZRS-bmi-1-IRES-EGFP retrovirus; E. de Pauw for telomere FISH analysis; andR. Bernards, A. Berns and W. Voncken for critically reading the manuscript J.J.L.J. and K.K. weresupported by a grant of the Dutch Cancer Society (K.W.F.)

Correspondence and requests for materials should be addressed to M.v.L. (e-mail: lohuizen@nki.nl).

Robustness inbacterial chemotaxisU. Alon*², M. G. Surette³, N. Barkai² & S. Leibler*²

Departments of *Molecular Biology and ²Physics, Princeton University, Princeton,

New Jersey 08544, USA³ Department of Microbiology and Infectious Diseases, Calgary, Alberta,Canada T2N 4N1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Networks of interacting proteins orchestrate the responses ofliving cells to a variety of external stimuli1, but how sensitive isthe functioning of these protein networks to variations intheir biochemical parameters? One possibility is that to achieveappropriate function, the reaction rate constants and enzymeconcentrations need to be adjusted in a precise manner, and anydeviation from these `®ne-tuned' values ruins the network'sperformance. An alternative possibility is that key properties of