far a meaningless theory. It is possible, though, that these various theories, along with all that we have learned in physics and other scientific disciplines, will yet merge into the best science can do: a theory of almost everything. P.-M. Binder is in the Department of Physics and Astronomy, University of Hawaii, Hilo, Hawaii 96720, USA, and at the Center for Nonlinear Dynamics, University of Texas, Austin. e-mail: pbinder@hawaii.edu

1. Barrow, J. D. Impossibility: The Limits of Science and the Science of Limits (Oxford Univ. Press, 1998).

2. Wolpert, D. H. Physica D 237, 12571281 (2008). 3. Wolfram, S. Phys. Rev. Lett. 54, 735738 (1985).4. Frieden, B. R. Science from Fisher Information: A Unification

(Cambridge Univ. Press, 2004).5. Kantz, H. & Schreiber, T. Nonlinear Time Series Analysis

(Cambridge Univ. Press, 2004).6. Binder, P.-M., Crosson, I. J. & Cadmus, R. R. Jr Astrophys. J.

685, L145L148 (2008).7. Crutchfield, J. P. & McNamara, B. S. Complex Systems

1, 417452 (1987).8. Ellis, J. Phys. World 12 (12), 4348 (1999).9. Wilczek, F. The Lightness of Being (Basic Books, 2008).

instant in time can infer all others. The two statements together imply that, at best, there can be only a theory of almost everything. Hence, they slam the door on Pierre-Simon Laplaces demon, introduced in 1814. This hypothetical being has a vast intellect, such that, with full knowledge of the state of the Universe at one time, it can completely predict the future and recall the past with no uncer-tainty whatsoever. Wolperts results are par-ticularly compelling because they are totally independent of both the details of the laws of physics and the computational characteristics of the machines.

What are the practical consequences of these findings for science? A prescription for deriving the laws of nature was proposed by Roy Frieden4 a few years ago. He asserted that any attempt to measure a physical quan-tity elicits a transfer of information from the source physical phenomenon to the observer. As a consequence, the observation of a physical phenomenon cannot be entirely accurate. By manipulating Fisher information, a measure of the quality of data, one can obtain informa-tion about the equations that govern the phe-nomenon in addition to the numerical value one seeks. Most fundamental equations of physics can be derived in this way (thus mak-ing physics a minor branch of statistics!). But Wolperts work warns us that Friedens recipe is bound to fail at least once, perhaps by produc-ing multiple solutions in certain cases.

Another example of the relevance of Wolp-erts work to science is in predicting the behav-iour of chaotic systems. Through the attractor reconstruction method, in which a time series is converted into a geometrical trajectory in higher-dimensional spaces5, one can forecast the evolution of fairly complex systems up to a specified prediction horizon. This method works quite well, and does not need the explicit knowledge of the systems governing equa-tions (see ref. 6 for an astrophysical example). However, when one tries to infer the equa-tions themselves, the results are often unclear or ambiguous7, possibly as a manifestation of Wolperts theorems.

Finally, Wolperts findings have a bearing on the possible limitations of two theories in physics. One set of limitations concerns the standard model of particle physics, which has accumulated a long list of shortcomings8: these include an exceedingly high predicted cosmo-logical constant, failure to predict the mass of the Higgs particle, and failure to account for the dark matter in the Universe. The other limitation is our inability to bring quantum mechanics and gravity into a single theory, although several viable alternative theories are being studied9. Quantum electrodynam-ics, a refinement of quantum mechanics, is defined by just two parameters (the charge and mass of the electron), whereas quan-tum gravity would require infinitely many parameters, and hence infinite experiments to determine those parameters, making it so

Complex eukaryotes, such as animals, have extensive RNA splicing to remove sequences that dont encode proteins (introns) and to connect those that do (exons). Often, several messenger RNAs can be generated by a single gene, because different patterns of splicing place different exons into the final mRNAs. As an example, the Dscam gene of the fruitfly Drosophila contains 24 exons, and is thought to encode 38,016 different protein isoforms by alternative splicing (although were not sure anyone has counted)1. In this particular case, docking and selector sequences within the primary transcript help regulate splicing. But in general, understanding exactly how a cell picks and chooses among the many possible combinations of splices has been a long-stand-ing problem2. Reporting on page 997 of this issue, Moldn et al.3 investigate how a genes promoter region the sequence that regulates

gene expression can regulate splicing.As ever, one approach is to work with a

simpler eukaryote. The fission yeast Schizo-saccharomyces pombe has a sophisticated splic-ing apparatus, and regulates splicing, although this regulation is somewhat different from that in more complex eukaryotes. In the lat-ter, introns are humongous, and the splicing system may therefore focus on recognizing the relatively small exons. When splicing is altered, it may fail to recognize an exon, leaving that exon out of the final product, and giving alter-native splicing. In yeast, introns are tiny, and the splicing system seems to focus on recognizing the introns. Thus, when splicing is altered, the system may fail to recognize an intron, retain-ing the intron in the final product. Whatever the reason, the usual nature of regulated splic-ing in animals is alternative splicing (an altered selection of exons), whereas the usual nature of

In 1874, Georg Cantor published a proof of the existence of uncountable infinities. He started by labelling points in the interval [0, 1) with the countable infinite natural numbers (1, 2, 3, ) as follows:

a1 = 0.d11 d12 d13 d14 d15 (for example, a1 = 0.31415926 with d11 = 3, d12=1, etc.)a2 = 0.d21 d22 d23 d24 d25 a3 = 0.d31 d32 d33 d34 d35

All ds are digits between 0 and 9. There is at least one number ax = 0.dx1dx2dx3 in the unit interval [0, 1) such that dx1 d11, dx2 d22, and so

on, guaranteeing that it differs from each number in the list by at least one digit, and hence it cannot be in the list. This proves that the number of points in the unit interval is not countable, a proof known as Cantor diagonalization.

A related result, the diagonal lemma, has played an important part in the proof of several incompleteness theorems. In 1931, Kurt Gdel proved that any mathematical system that includes enough of the theory of natural numbers contains statements that cannot be proved to be either true or false, and is thus incomplete. The general argument for the proof is

based on Epimenides liars paradox is This sentence is false a true or a false statement? but it replaces false with unprovable. The construction of a number that represents a concrete, undecidable statement requires diagonalization techniques.

In the mid-1930s, Alan Turing used similar methods to prove that no general algorithm can determine whether a given Turing machine a computer halts for a given input. It is thus not surprising that Wolperts proofs also rely on diagonalization arguments. P.-M.B.

Box 1 | Cantor, Gdel , Turing and the uncountable

MOLECULAR BIOLOGY

Bound to spliceBruce Futcher and Janet K. Leatherwood

Messenger RNAs dont usually correspond exactly to DNA portions of the primary transcript, known as introns, are removed by splicing. A study reveals new ways in which splicing can be regulated.

885

NATURE|Vol 455|16 October 2008 NEWS & VIEWS

50 &

100

YEA

RS A

GO

regulated splicing in yeast is intron retention4. Nevertheless, the mechanisms underlying these two kinds of regulated splicing may be similar.

In S. pombe, regulated splicing occurs in meiosis5. Many meiosis-specific proteins are toxic to vegetative (that is, mitotic) cells, and their expression is kept turned off by multiple mechanisms6. For many meiosis-specific genes, transcription is repressed in vegetative cells, and furthermore, if and when small amounts of transcript are made, their splicing is repressed; that is, introns are retained, so that no active protein is made. When the cells enter meiosis, transcription is induced, and so is splicing.

Moldn et al.3 have studied the splicing of rem1, a meiosis-specific gene of S. pombe. Transcription of rem1 is repressed in vegetative growth, and any transcript that does get made does not get spliced. On entry into meiosis, the rem1 transcript is induced and spliced5,7. The major finding of Moldn et al.3 is both simple and remarkable: the information that specifies meiosis-specific splicing lies entirely inside the promoter, and not in the transcribed region. For example, when the rem1 transcript is expressed from some other promoter, it is spliced in both vegetative and meiotic cells. Conversely, when the authors used the rem1 promoter to drive transcription of a normal vegetative gene (cdc2, which has four introns, and which is usually spliced in both vegetative and meiotic cells), then splicing occurred only in meiosis, and in the same temporal pattern as for the wild-type rem1 gene.

In meiosis, the rem1 promoter is bound by Mei4, a meiosis-specific protein belonging to the forkhead family of transcription factors. S. pombe has three other forkhead transcrip-tion factors, and Moldn et al. suggest that, in vegetative cells, the Mei4-binding sites in the rem1 promoter are probably occupied by one of these other factors, Fkh2. When the authors deleted the fkh2 gene from S. pombe, vegetative transcription of rem1 was slightly increased, and some of this transcript was spliced. This suggests that Fkh2 represses both transcription and splicing in vegetative cells. The authors show that Mei4, which is made only in meiosis, binds to rem1 and turns on both transcription and splicing.

Why would one forkhead transcription factor induce splicing, but not the other? On the basis of co-immunoprecipitation experi-ments, Moldn et al. found that Mei4, but not Fkh2, forms complexes with the spliceo-some. The authors therefore suggest that the Mei4 transcription factor actively recruits splicing factors to the rem1 gene and tran-script, whereas Fkh2 does not. They suggest that it is this recruitment of the spliceosome by a meiosis-specific transcription factor that is responsible for meiosis-specific splic-ing. However, less direct explanations are also possible: transcription factors can affect the conformation of chromatin, the rate of mRNA elongation, and the 5 capping and 3 poly-adenylation processing of transcripts, and all

these are interrelated with splicing8.The model proposed for rem1 is exciting,

but still leaves us with a major puzzle. How can Fkh2 at a promoter inhibit splicing of rem1 and even an unrelated gene such as cdc2 whose transcript has good splicing signals? Maybe control of RNA processing (5 capping and 3 polyadenylation as well as splicing) will be a more general feature of promoters. It now seems that most of the genome is transcribed to at least some degree, so turning gene expres-sion off completely may depend on regulating steps of RNA processing in addition to control-ling efficiency of transcription.

These remarkable findings3 leave some loose ends. First, Fkh2 regulates many veg-etative transcripts, and many of these are spliced. Thus, Fkh2 is not repressing vegetative splicing at most of its targets. Second, the rem1 promoter does not impose its usual temporalpattern of splicing when fused to the crs1 gene5, in contrast to the results obtained here with cdc2 (ref. 3). Third, in the model proposed, the failure of vegetative splicing for the rem1-driven cdc2 transcript seems to suggest little or no ability to splice rem1-driven transcripts post-transcriptionally, in contrast to Saccharo-myces cerevisiae, in which most splicing may be post-transcriptional9. It should also be noted that the study of intron retention is compli-cated by two subtle and nasty artefacts: unlike mRNAs generated by alternative splicing, mRNAs generated by complete intron reten-tion are perfectly co-linear with the DNA. Thus, the polymerase chain reaction following reverse transcription (RT-PCR), which is used to amplify an mRNA with retained introns, can amplify genomic DNA instead, with results that appear identical results (an artefact Moldn et al. addressed). Second, if a gene has an anti-sense transcript, this will never be spliced, but, in RT-PCR, will yield the same product as an unspliced sense transcript.

Still, these issues do not significantly under-cut the core result for rem1: splicing depends on which transcription factor is bound to the promoter. Understanding exactly how Mei4 turns splicing on, and how Fkh2 keeps splicingturned off, will be interesting investigations for the future. In addition, it is a reminder that efforts to understand splicing by searching for signals inside the transcript have limitations, as some of the information is elsewhere. Bruce Futcher and Janet K. Leatherwood are in the Department of Molecular Genetics and Microbiology, Stony Brook University, Stony Brook, New York 11794-5222, USA.e-mail: janet.leatherwood@sunysb.edu

1. Schmucker, D. et al. Cell 101, 671684 (2000).2. Wang, Z. & Burge, C. B. RNA 14, 802813 (2008).3. Moldn, A. et al. Nature 455, 9971000 (2008).4. Romfo, C. M. et al. Mol. Cell. Biol. 20, 79557970

(2000).5. Averbeck, N. et al. Mol. Cell 18, 491498 (2005).6. Harigaya, Y. & Yamamoto, M. Chromosome Res. 15,

523537 (2007).7. Malapeira, J. et al. Mol. Cell. Biol. 25, 63306337 (2005).8. Kornblihtt, A. R. Curr. Opin. Cell Biol. 17, 262268 (2005). 9. Tardiff, D. F. et al. Mol. Cell 24, 917929 (2006).

50 YEARS AGOThe American Air Force lunar probe attempt on October 11 failed to achieve the full objective of a circum-lunar orbit, but demonstrated that many of the technical problems of launching and accelerating a three-stage rocket to the necessary speed have been solved. Guidance of the rocket into the correct path and precise control of the final cut-off velocity appear to have been the chief problems not yet completely mastered In the instrument payload of some 40 lb. were included radiation detectors, which confirmed the earlier measurements on Explorer IV of the intense radiation belt surrounding the Earth. Preliminary analyses of the results have shown a significant decrease in intensity beyond several Earth radii, and seem to confirm the idea that the radiation is due to trapping of cosmic particles by the Earths magnetic field No doubt there are some who will not regret the failure to penetrate the mystery of the Moons unseen face, but it must be only a matter of time until this is done From Nature 18 October 1958

100 YEARS AGOOn October 10, in the presence of the leading aronautical experts of France, Mr. Wilbur Wright, with M. Painlev as a passenger, accomplished a flight of 1h. 9m. 45.6s in duration, the distance covered being estimated at nearly seventy kilometres. This successful flight is the last demanded of Mr. Wright by the French syndicate which has acquired the local rights in his aroplane by the payment of 10,000l. at once and 10,000l. in a months time, after three men have been trained to work the machine [O]n November 1 the Socit navale des Chantiers de France will begin at Dunkirk the construction of fifty Wright aroplanes, which are to be sold at the price of 1000l. each.From Nature 15 October 1908

886

NATURE|Vol 455|16 October 2008NEWS & VIEWS

/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 450 /GrayImageDepth -1 /GrayImageDownsampleThreshold 1.00000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages true /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 2400 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.00000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /PDFX1aCheck true /PDFX3Check false /PDFXCompliantPDFOnly true /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox false /PDFXBleedBoxToTrimBoxOffset [ 0.30000 0.30000 0.30000 0.30000 ] /PDFXOutputIntentProfile (OFCOM_PO_P1_F60) /PDFXOutputCondition (OFCOM_PO_P1_F60) /PDFXRegistryName (http://www.color.org) /PDFXTrapped /False

/Description >>> setdistillerparams> setpagedevice