Activesite

Group-2neuraminidase

a

b

c

150-loop

Newcavity

Group-1neuraminidase

Neuraminidase–inhibitor complex

Inhibitor

STRUCTURAL BIOLOGY

Antiviral drugs fit for a purposeMing Luo

Did drug researchers have a lucky break when they developed antiviral drugs for influenza? Crystal structures of enzymes from the H5N1 virus suggest that they did, and provide avenues for further exploration.

Oseltamivir (Tamiflu) and zanamivir (Relenza) have been stockpiled by several nations to counter the threat of a flu pandemic, should the highly pathogenic avian influenza virus H5N1 develop into a human strain. These drugs are inhibitors of an enzyme known as neurami-nidase, which is found on the surface of the flu virus. But would these medications hold their own against a pandemic? Some answers might be found from the crystal structure of H5N1 neuraminidases bound to the antiviral drugs, as reported by Russell et al.1 on page 45 of this issue*. These structures also reveal a cavity in the active site of certain neura-minidase subtypes that could be exploited to

make more effective antiviral drugs.Avian flu belongs to the genus of influenza

virus known as type A, which is divided into nine subtypes based on the variety of neura-minidase expressed. The subtypes fall into two groups2: group-1 contains the subtypes N1, N4, N5 and N8, whereas group-2 con-tains the subtypes N2, N3, N6, N7 and N9. The enzyme facilitates the spread of virus dur-ing an infection; oseltamivir and zanamivir strongly inhibit neuraminidase activity, so limiting the disease. These inhibitors were originally developed using crystal structures of neuraminidase subtypes N9 and N2 and another neuraminidase from the type B genus of influenza viruses3–5. The crystal structures, along with those of neuraminidase–inhibitor

complexes, showed that the three-dimensional structure of the enzyme active site is the same for all of these subtypes, and that the inhibitors bind to the active sites in the same way.

The inhibitors don’t just suppress the activ-ity of the enzyme subtypes that were used to direct drug development — they have similar activity against N1 neuraminidase, and are effective against flu viruses that contain an N1 neuraminidase6. It was always assumed that the inhibitors bind to the active sites of the group-1 neuraminidases in the same way as for group-2 enzymes, as the amino-acid sequences of the active sites are essentially the same for all the subtypes.

However, the crystal structures of the N1, N4 and N8 neuraminidases, reported by Russell et al.1, surprisingly reveal that the active sites of these group-1 enzymes have a very differ-ent three-dimensional structure from that of group-2 enzymes. The differences lie in a loop of amino acids known as the 150-loop, which has an unexpected conformation that opens up an adjacent cavity. The 150-loop contains an amino acid designated Asp 151; the side chain of this amino acid has a carboxylic acid that, in group-1 enzymes, points away from the active site as a result of the ‘open’ conformation of the 150-loop (Fig. 1). The side chain of another active-site amino acid, Glu 119, also has a different conformation in group-1 enzymes compared with group-2 enzymes.

The Asp 151 and Glu 119 amino-acid side

as black holes5. Typically, Reynolds numbers in astrophysics are much larger than the values of a few thousand dealt with by Hof et al.1, indi-cating a much greater propensity for turbulent flow. If any turbulence in these astrophysical flows were transient over long timescales, this might explain the enhanced accretion and in-fall rates frequently observed, without the need to invoke additional instabilities. Such insta-bilities are generally postulated to be caused by magnetic fields6,7. Because of the significant interaction of rotation and turbulence in astro-physical objects, however, we cannot be sure that Hof and colleagues’ results1 would apply.

From the applied mathematician’s perspec-tive, the implications of these investigations are fascinating. The motions are best understood by thinking about a ‘phase space’ of configu-rations of velocities present in a fluid. In this depiction, a system evolves in time by mov-ing around the phase space. The laminar state has a constant velocity field that is smooth and steady in time. In velocity phase space, this is represented by a single point, with trajectories pointing inward from every direction indicat-ing its stability.

For turbulent flow, however, a great tangle of trajectories would show up some distance from the laminar point. The turbulent state amounts to wandering on that tangle, and getting lost on it for considerable periods of time. If Hof et al. are correct, however, the system always — however long it takes — finds a connecting trajectory that causes it to fall back onto the laminar attracting point. The exponentially long times indicate a change in the character

of the turbulent tangle that begs for a deeper understanding.

The results1 do need to be verified by other means and in other shear geometries. Previous studies of the turbulent–laminar transition2,3, including a study from one of the paper’s authors2, indicated that the turbulent lifetime becomes infinite at Reynolds numbers above around 2,200. This stark contrast between pre-vious and the present results will surely gener-ate considerable activity. The biggest difficulty here will be that the long lifetimes involved limit the range of Reynolds numbers that are practicably testable.

Understanding the idealization of very smooth walls and a quiet environment will also require further study. Beyond that, the possi-bility that these results could be extended to other shear flows, and the potential for new turbulence controls, will continue to generate excitement in this area of research. ■

Daniel Perry Lathrop is in the Departments of Physics and Geology, and at the Institutes for Physical Sciences and Technology, and for Research in Electronics and Applied Physics, University of Maryland, College Park, Maryland 20742, USA.e-mail: lathrop@umd.edu

1. Hof, B., Westerweel, J., Schneider, T. M. & Eckhardt, B. Nature 443, 59–62 (2006).

2. Faisst, H. & Eckhardt, B. J. Fluid Mech. 504, 343–352 (2004).3. Peixinho, J. & Mullin, T. Phys. Rev. Lett. 96, 094501 (2006).4. Ott, E., Grebogi, C. & Yorke, J. A. Phys. Rev. Lett. 64,

1196–1199 (1990).5. Balbus, S. A. & Hawley, J. F. Rev. Mod. Phys. 70, 1–53 (1998).6. Velikhov, E. P. J. Exp. Theor. Phys. 9, 995 (1959).7. Sisan, D. R. et al. Phys. Rev. Lett. 93, 114502 (2004).

Figure 1 | Induced fit of an enzyme inhibitor. The influenza neuraminidase enzyme is blocked by antiviral drugs. Subtypes of the enzyme exist and are divided into two groups, which differ in the conformation of their active sites. a, A loop of amino acids (known as the 150-loop) in the active sites of group-2 enzyme subtypes is arranged such that no conformational changes occur in the active site on binding of an inhibitor. b, For group-1 subtypes, the 150-loop has a different conformation from group-2 subtypes, exposing a large cavity next to it. c, When an inhibitor binds to group-1 subtypes, the 150-loop adopts a conformation similar to that of group-2 neuraminidases.

*This article and the paper concerned1 were published online on 16 August 2006.

37

NATURE|Vol 443|7 September 2006 NEWS & VIEWS

7.9 n&v MH 377.9 n&v MH 37 1/9/06 6:02:30 pm1/9/06 6:02:30 pm

Nature Publishing Group ©2006

–1 –0.75 – 0.5 –0.25 0.25Scaled error

Relativeprobability

0.5 0.75 1

0.25

0.5

0.75

1

chains form critical interactions with neurami-nidase inhibitors. For neuraminidase subtypes with the open conformation of the 150-loop, the side chains of these amino acids might not have the precise alignment required to bind inhibi-tors tightly. If so, influenza viruses that contain these subtypes would be resistant to oseltamivir and zanamivir. However, the crystal structure of N1 neuraminidase in a complex with osel-tamivir shows that the open polypeptide loop adopts the conformation observed in the N2 neuraminidase–inhibitor complex. In other words, when an inhibitor binds, the active site adopts the conformation seen in the crystal structures of group-2 enzymes3–5. This ‘induced fit’ restores the contacts between the amino-acid side chains of the N1 active site and the inhibitor, which probably explains why the inhibitors are effective against N1 influenza viruses.

The difference in the active-site conforma-tion between the two groups of neuraminidases must be caused by changes to amino acids that lie outside the active site. This means that an enzyme inhibitor for one target will not neces-sarily have the same activity against another with the same active-site amino acids and the

same overall three-dimensional structure. We are fortunate that the group-1 influenza virus neuraminidases assume a tight interaction with their inhibitors. Otherwise, these anti-viral drugs might be ineffective against influ-enza viruses such as H5N1.

Russell and colleagues’ discovery1 cautions us to pay more attention to changes in amino-acid sequences outside the active site of neuramidinases. Such changes might result in a mutant virus that is resistant to current drugs if the conformation of the active site is altered enough to prevent an induced fit of the inhibi-tors. We must therefore determine the binding strength of our neuraminidase inhibitors for each new strain of influenza virus, even if the amino-acid sequence of the neuraminidase active site is unchanged compared with that of existing viruses.

More positively, the cavity near the active site that is exposed by the open conformation of the 150-loop might be exploited by drug designers. The affinity of enzyme inhibitors depends on their fit in their respective active sites. One could design a chemical group with the right shape, size and electronic charge to fit snugly

into the newly discovered cavity. Attaching this group to the scaffolding of existing inhibitors could then produce a class of neuraminidase inhibitor that is more effective against H5N1-like flu viruses.

When combating pathogenic microorgan-isms, either bacteria or viruses, we often face an enemy that continually alters its tactics. The flu virus is no exception. The way to stay ahead in this game is to constantly track the changes of the pathogen and adapt countermeasures accordingly. The crystal structures reported by Russell et al.1 provide valuable intelligence in the war against influenza. ■

Ming Luo is in the Department of Microbiology, University of Alabama, Birmingham, 1025 18th Street South, Birmingham, Alabama 35294, USA. e-mail: mingluo@uab.edu

1. Russell, R. J. et al. Nature 443, 45–49 (2006).2. Thompson, J. D., Higgins, D. G. & Gibson, T. J. Comput.

Appl. Biosci. 10, 19–29 (1994).3. Kim, C. U. et al. J. Am. Chem. Soc. 119, 681–690

(1997).4. von Itzstein, M. et al. Nature 363, 418–423 (1993).5. Taylor, N. R. et al. J. Med. Chem. 41, 798–807 (1998).6. Warren, M. K. et al. Antimicrob. Agents Chemother. 46,

1014–1021 (2002).

Even under the best circumstances, control-ling our errors is a dicey business. But as a recent series of mathematical papers1–3 shows, significant strides are being made — in number theory at least. The new work amounts to a proof of a 40-year-old conjecture, known as the Sato–Tate conjecture, for a class of mathematical problems with applica-tions in cryptography and the high-speed factorization of large numbers. This conjecture predicts the probabil-ity distribution of the error terms that pop up in these problems.

In any empirical study, errors accu-mulate for many reasons. All an experi-mentalist can hope for is to know the sources of most errors, and to be able to estimate how much trouble they cause.

The web page of the US Bureau of Transportation, for example, lists six possible causes of systematic error in the 1993 US Census count: inability to obtain information about all cases in the sample; response errors; defi-nitional difficulties; differences in the

interpretation of questions; mistakes in cod-ing or recoding the data obtained; and other errors of collection, response, coverage and estimation4.

In pure mathematics, however, whenever we calculate what we hope is a good approxima-

MATHEMATICS

Controlling our errors Barry Mazur

The Sato–Tate conjecture holds that the error term occurring in many major problems in number theory conforms to a specific probability distribution. That conjecture has now been proved for a large group of cases.

tion to a quantity describing a phenomenon, we also have a shot at understanding — in depth, and beyond a shadow of a doubt — the nature of the ensuing error term, defined simply as: error term�exact value�our ‘good approximation’.

Of course, if our approximation is at all good, the error term should be small. For many prob-lems, the gold standard of goodness is what we might call square-root accuracy: that the error term scales as the square root of the quantity as it, and the phenomenon it describes, grows larger and larger. Great successes in controlling error terms in number theory were achieved in the past century. Specifically, through the work of Helmut Hasse5 in the 1930s, André Weil6 in the 1940s and Pierre Deligne7 in the 1970s, a large class of major approximations were

proved to have this kind of accuracy. Take a (randomly chosen) example.

For prime numbers p, define N(p) as the number of ways in which p can be written as a sum of 24 squares of whole numbers. (Squares of positive numbers, negative numbers and zero are all allowed.) The ordering of the squares of the numbers that occur in this summation also counts. Thus, the first prime number, 2, can already be written as a sum of 24 squares of whole numbers in 1,104 ways, because there are that many different ways in which two choices of either (�1)2 or (�1)2 can be arranged in a line where the 22 other numbers are zeros.

We know, then, that N(2)�1,104. What about N(p) for the other prime numbers p�3,5,7,11,…? A good

Figure 1 | Probability distribution of error terms. The Sato–Tate distribution √1 − x2π , the smooth red curve in this figure, can be compared with the probability distribution of scaled error terms (blue bars) for the number of ways N(p) in which a prime number p can be written as a sum of 24 square numbers. The data, tabulated for primes p less than a million, agree closely with the distribution, and give hope that the Sato–Tate conjecture holds for this problem. (Courtesy of W. Stein; for details of the computation, see ref. 10.)

38

NATURE|Vol 443|7 September 2006NEWS & VIEWS

7.9 n&v MH 387.9 n&v MH 38 1/9/06 6:02:34 pm1/9/06 6:02:34 pm

Nature Publishing Group ©2006