© 1999 Macmillan Magazines Ltd

partner whose low masculine appearancesuggests cooperation in parental care(‘long-term’ preferences are unchangedacross the menstrual cycle) but occasionallycopulate with a male with a more masculineappearance (indicating good immunocom-petence) when conception is most likely.Sexual behaviour arising from cyclic prefer-ences might allow individuals to accruebenefits from polyandry while maintainingthe advantage of ostensive monandry.I. S. Penton-Voak*, D. I. Perrett*, D. L. Castles†‡, T. Kobayashi†, D. M. Burt*, L. K. Murray*, R. Minamisawa†*School of Psychology, University of St Andrews, Fife KY16 9JU, UKe-mail: ip@st-and.ac.uk†Hasegawa Laboratory, Department of Life Sciences, University of Tokyo, Komaba, Tokyo 153, Japan‡School of Life Sciences, Roehampton Institute London, Whitelands College, London SW15 3SN, UK

1. Perrett, D. I. et al. Nature 394, 884–887 (1998).

2. Berry, D. S. & Wero, J. L. F. J. Pers. 61, 497–523 (1993).

3. Folstad, I. & Karter, A. Am. Nat. 139, 603–622 (1992).

4. Regan, P. C. Can. J. Hum. Sexual. 5, 145–156 (1996).

5. Gangestad, S. W. & Thornhill, R. Proc. R. Soc. Lond. B 265,

927–933 (1998).

6. Gangestad, S. W. & Thornhill, R. Evol. Hum. Biol. 18, 69–88

(1997).

7. Baker, R. R. & Bellis, M.A. Human Sperm Competition:

Copulation, Masturbation and Infidelity (Chapman & Hall,

London, 1995).

8. Christenfeld, N. & Hill, E. A. Nature 378, 669 (1995).

9. Frost, P. Percept. Motor Skills 79, 507–514 (1994).

10.Grammer, K. Ethol. Sociobiol. 14, 201–208 (1993).

11.Graves, J., Ortegaruano, J. & Slater, P. J. B. Proc. R. Soc. Lond. B

253, 3–7 (1993).

both indicated that the juxtaposition of cellsthat did and did not express Rfng is suffi-cient for the formation of the apical ecto-dermal ridge (AER), an organizing centrethat forms at the boundary between dorsaland ventral cells at the distal edge of thelimb bud4,5. We have used homologousrecombination in mouse embryonic stemcells to investigate Rfng expression in themouse limb bud and to test whether it isnecessary for the formation of the AER.

We inserted the Escherichia coli lacZgene into the Rfng genomic locus to createthe RfnglacZKI ‘knock-in’ allele. Analysis of b-galactosidase activity in RfnglacZKI limb budsrevealed expression in the developing andmature AER (Fig. 1a–c). To create a nullallele, Rfng∆1-8neo, we deleted 2.4 kilobases ofthe Rfng genomic locus containing theentire coding region (exons 1–8) andreplaced it with a ‘PGKneo-selection’ cas-sette (which confers a resistance to the

antibiotic neomycin and acts as a means forselecting these clones) in the opposite tran-scriptional orientation.

No homozygous Rfng∆1-8neo-deficientmouse survived past birth (n4244 progenyfrom heterozygote intercrosses on a mixed129P22C57BL/6J hybrid background;n470 progeny from heterozygote inter-crosses on a mixed 129P22129S6 hybridbackground). The mutant mice displayedseveral developmental defects, including acleft palate and reduced mandible (64%),exencephaly (24%), and forelimb defects(12%) consisting of soft-tissue syndactylyor oligodactyly (total n4225 on a mixed129P2/OlaHsd2C57BL/6J hybrid back-ground, examined between 18 days post-coitum (d.p.c.) and birth; Fig. 1).

We went on to remove the PGKneo-selection cassette to create the Rfng∆1-8 nullallele by two independent methods: matingRfng∆1-8neo mice to mice that expressed Cre

scientific correspondence

742 NATURE | VOL 399 | 24 JUNE 1999 | www.nature.com

Limbs move beyond theRadical fringe

The fringe genes and the Notch signallingpathway are important in limb develop-ment in vertebrates and in Drosophila, help-ing to establish the ‘organizing centres’ thatare required for the proximal–distal out-growth of appendages1–5. Three vertebrateFringe genes have been identified: Manic(Mfng), Lunatic (Lfng) and Radical fringe(Rfng)4–8. Here we show that the mouseRfng gene is not required for limb develop-ment, even though it is expressed in thedeveloping limb bud. But we have foundthat several developmental defects, which atfirst we attributed to a loss of Rfng functionafter mutating the gene, in fact arose as aresult of the insertion of a selection cassetteinto the Rfng locus that affects the expres-sion of a neighbouring gene or genes. Ourfindings highlight the need for optimizationin designing mutant alleles to probe devel-opmental processes9,10.

Studies of Rfng expression in chick limbmutants and gain-of-function experimentsin limb buds using retroviral misexpression

p m

a b c

d e f

g h i

j k l

Rfng lacZKI

Rfng ∆1-8neo –/– Rfng ∆1-8 –/–

9.5 10.5 12.5

Rfng ∆1-8neo +/+

D V

p pm m

FFiigguurree 11 Rfng is expressed in the mouse limb bud and is not required for limb development. Rfng expressionin RfnglacZKI heterozygote limbs, as shown by X-gal staining, at: a, 9.5 d.p.c.; b, 10.5 d.p.c.; and c, 12.5 d.p.c.Dorsal (D), left; ventral (V), right. Replacement of the Rfng locus with a PGKneo-selection cassette results inpleiotropic developmental defects. d, g, j, Wild-type fetuses; e, h, k, Rfng∆1-8neo homozygotes; and f, i, l, Rfng∆1-8

homozygotes at 18.5 d.p.c. d–f, Skeletal staining of the secondary palate. Anterior is to the right. d, Thepalatal (p; indicated by arrows) and maxillary (m) shelves are nearly fused. e, Rfng∆1-8neo homozygote palatalshelves (p; white arrows) have not elevated towards the midline, and the maxillary shelves (m) are widerthan in the wild type. f, Rfng∆1-8 homozygote, showing normal palatal (p) and maxillary (m) shelf positioning.g, Wild-type fetus with closed cranium. h, Rfng∆1-8neo homozygote, showing exencephaly. i, Rfng∆1-8 homo-zygotes are indistinguishable from wild type (g). j–l, Skeletal staining of the forelimbs. Anterior is up, posterioris down. j, Rfng∆1-8neo wild-type limb has five digits. k, Rfng∆1-8neo homozygote, showing soft-tissue syn-dactylism of digits 4 and 5 (arrow). l, Rfng∆1-8 homozygotes are indistinguishable from wild type (j).

© 1999 Macmillan Magazines Ltd

recombinase (FVB/N genetic background)11

and microinjection of a circularized Crerecombinase-expressing plasmid intoRfng∆1-8neo heterozygote fertilized eggs(129P2/OlaHsd2C57BL/6J hybrid back-ground). In contrast to Rfng∆1-8neo homo-zygotes, Rfng∆1-8 homozygotes on bothgenetic backgrounds were viable and weremorphologically wild type (total n4172;Fig. 1f, i, l). We conclude that Rfng functionis not essential for AER formation or forproximal–distal limb outgrowth12. Theabsence of a limb phenotype in the Rfngnull allele may be due to functional overlapbetween Fringe family members.

We suggest that the phenotypes associ-ated with the introduction of a PGKneo-selection cassette into the Rfng locus(Rfng∆1-8neoallele) are caused by interferencewith the proper expression of a neighbour-ing gene or genes whose expression overlapswith that of Rfng and are required for neuraltube, craniofacial and limb development.There have been previous examples inwhich PGKneo effects have been examined.In such cases, the target genes have beenpart of a related gene complex that sharesboth regulatory control and functions (forexample, the myogenic regulatory factors,HoxD and globin gene clusters13).

Our findings with Rfng-targeted allelesmay be relevant for interpreting resultsfrom two Lfng-targeted alleles, LfnglacZ andLfng∆1neo (refs 9,10). In both of these Lfnggene-targeted alleles, the PGKneo-selectioncassette was not excised from the locus9,10.The Lfng-targeted mutants exhibited phe-notypic defects restricted to somitogenesis;however, there were unexplained differ-ences in somite epithelialization. This phe-notypic difference was attributed to eitherdifferences in genetic background or to dif-ferences in the design of the targeting con-structs, although both targeted alleles werethought to be null alleles9,10.

With these Lfng alleles, it might there-fore be possible that the PGKneo cassetteaffects the expression of a neighbouringgene(s) (perhaps paralogous between fringefamily members). One such candidate isUncx4.1, a paired type homeodomain genethat is physically linked to Lfng and whoseexpression in the caudal half of condensingsomites is altered in the Lfng mutants9,10,14.Our findings reinforce the importance ofunderstanding the increasingly apparentpresence of embedded and shared regulatoryelements in genomes11.Jennifer L. Moran, John M. Levorse, Thomas F. VogtDepartment of Molecular Biology, Lewis Thomas Laboratory, Princeton University,Princeton, New Jersey 08544, USAe-mail: tvogt@molbio.princeton.edu

1. Shubin, N., Tabin, C. & Carroll, S. Nature 388, 639–648 (1997).

2. Irvine, K. D. & Vogt, T. F. Curr. Opin. Cell Biol. 9, 867–876

(1997).

3. Panin, V. M., Papayannopoulus, V., Wilson, R. & Irvine, K. D.

Nature 387, 908–912 (1997).

4. Laufer, E. et al. Nature 386, 366–373 (1997).

5. Rodriguez-Esteban, C. et al. Nature 386, 360–366 (1997).

6. Wu, J. Y., Wen, L., Zhang, W.-J. & Rao, Y. Science 273, 355–358

(1996).

7. Johnston, S. H. et al. Development 124, 2245–2254 (1997).

8. Cohen, B. et al. Nature Genet. 16, 283–288 (1997).

9. Zhang, N. & Gridley, T. Nature 394, 374–377 (1998).

10.Evrard, Y. A. et al. Nature 394, 377–381 (1998).

11.Lakso, M. et al. Proc. Natl Acad. Sci. USA 93, 5860–5865 (1996).

12.Christen, B. & Slack, J. M. W. Nature 395, 230–231 (1998).

13.Olson, E. N., Arnold, H. H., Rigby, P. W. J. & Wold, B. J. Cell 85,

1–4 (1996).

14.Moran, J. L. et al. Mamm. Genome 10, 535–541 (1999).

Zhang and Gridley reply — We too have cre-ated mice that have a large deletion in theRadical fringe gene and investigated theresulting effect on their development. Ourresults agree with those of Moran et al. inshowing that Rfng is not an essential gene inmice, but they do not support the idea thatthe presence of a neo-selection cassetteinserted into the Rfng locus invariably leadsto embryonic lethality.

We constructed and analysed a targetedmutation of the Rfng gene by using a target-ing vector in which a section of the Rfnggenomic sequence, encoding the regionfrom amino acid 52 of the Rfng proteinthrough to the end of the Rfng gene, wasdeleted. The vector contained a neo-selec-tion cassette in the same transcriptional orientation as the Rfng gene. Mice homo-zygous for our Rfng mutant allele formedviable, morphologically normal adults atthe expected mendelian frequencies (34 of126 homozygotes (27%) on a mixedC57BL/6J2129/SvImJ background, and 3of 16 homozygotes (19%) on an inbred129/SvImJ background). The absence ofwild-type Rfng transcripts in our Rfnghomozygous-mutant adult mice was con-firmed by northern blotting.

It is possible that a difference in the con-struction of the two targeted Rfng allelescould explain the different mouse pheno-types generated by Moran et al. — the ori-entations of the neo-selection cassette werenot the same in the two targeted Rfng alle-les, for instance. Reversing the orientationof the neo sequence for transcription createsa potent splice-acceptor site1,2.

In considering the implications of theirfinding for the interpretation of Lfngmutant phenotypes3,4, Moran et al. correctlypoint out that, in some instances, the pres-ence of a neo-selection cassette in a targetedallele can affect the transcription of othergenes, and hence the mutant phenotypesthey generate. To confirm that the mutantphenotype is independent of the presenceof a neo cassette in the two Lfng mutant alle-les, a targeted Lfng allele constructed with-out a neo cassette needs to be tested.

However, we believe that it is unlikelythat the neo cassette in the two Lfng mutantalleles is responsible for the alterations inUncx4.1 transcription or for the defects in

somite formation that we detected in Lfnghomozygous-mutant embryos. The sponta-neous mouse mutant pudgy (pu) is causedby a frameshift mutation in the Dll3 gene5,which encodes a ligand for the Notch familyof proteins. This frameshift leads to prema-ture truncation of the Dll3 protein5. Ashomozygous Dll3pu/Dll3pu mutant embryosshow defects in somite formation very likethose seen in Lfng mutant embryos, wecompared Uncx4.1 transcription in bothDll3pu/Dll3pu and LfnglacZ/LfnglacZ mutantembryos and found that the Uncx4.1expression pattern was similarly disorga-nized in both mutants. Given that the pres-ence of a neo cassette does not alter Uncx4.1expression in Dll3pu/Dll3pu embryos, thesimplest explanation is that the alterationsin Uncx4.1 expression and somite forma-tion in LfnglacZ/LfnglacZ mutant embryos arecaused by perturbations of the Notch sig-nalling pathway, and not by the presence ofa neo cassette in the LfnglacZ mutant allele.Nian Zhang, Thomas GridleyThe Jackson Laboratory, Bar Harbor, Maine 04609-1500, USAe-mail: gridley@jax.org

1. Jacks, T. et al. Nature Genet. 7, 353–361 (1994).

2. Carmeliet, P. et al. Nature 380, 435–439 (1996).

3. Zhang, N. & Gridley, T. Nature 394, 374–377 (1998).

4. Evrard, Y. A. et al. Nature 394, 377–381 (1998).

5. Kusumi, K. et al. Nature Genet. 19, 274–278 (1998).

scientific correspondence

NATURE | VOL 399 | 24 JUNE 1999 | www.nature.com 743

Risk of collisions forconstellation satellites

The increasing quantity of debris orbitingthe Earth is causing concern for space agen-cies. At the average collision velocity of 10km s11, even projectiles 1 cm across candamage satellites. Impacts have a significantchance of occurring, especially for largestructures that remain in orbit for a longtime1. Particularly at risk are satellites ataltitudes of about 800 and 1,400 km, wherethere is already a high density of orbitingbodies and atmospheric drag is not effectivein removing small fragments. We find thatthe dynamical architecture of satellite con-stellations, several of which are to belaunched over the next decade, makes themparticularly vulnerable to the consequencesof an impact, which may set up a kind ofchain reaction, triggering more collisions.

There are currently about 104 objectslarger than 10–20 cm across orbiting theEarth, with 105 objects more than 1 cmacross, and 107 bodies exceeding 1 mm indiameter. These have a total mass of about32106 kg and a cross-section of 42104 m2

(refs 2,3), and the average flux (the expectedimpact rate per projectile and per unit targetcross-section) is about 3210110 m12 yr11.In the long term, debris from collisions with