scientific correspondence term, ǁk[B][P(r)], corresponds to an a induced screening charge in a conducting Scelidosaurus medium. For a uniform concentration of Robust dinosaur protease, the long-range 1/r potential is Nodosauridae reduced exponentially: phylogeny? → Gargoyleosaurus Ƞ(r)ǃ1/r Ƞscr(r)ǃ(1/r)exp(ǁr/ȕ) and ȕ2ǃD/k[B]. The Ankylosauria comprises two families In simple terms, degradation by the pro- of armoured dinosaurs (Nodosauridae and Ankylosauridae tease limits the size of the pheromone cloud Ankylosauridae) that are best known from , 8 around each Ȋ-cell to a radius ȕ the average well-preserved specimens from the Cret- Scelidosaurus distance that a pheromone molecule diffus- aceous period. In their report on the skull b es before being degraded. This screening of a new Jurassic ankylosaur, Gargoyle- Ankylosauridae increases the alignment between the gradi- osaurus, Carpenter et al.1 presented a phy- ent and the direction to the nearest mating logenetic analysis of four terminal taxa, partner, although the gradient strength is which yielded a tree with Gargoyleosaurus Nodosauridae reduced (Fig. 1b,c). Consequently, the pro- as the sister taxon of the Ankylosauridae. tease is not helpful when there is only a sin- But the authors’ claim that their tree is Gargoyleosaurus gle source of pheromone, as was observed robust is undermined when their data and 3 by Segall . Effective screening is obtained their tree are evaluated using numerical c Scelidosaurus when ȕ is of the order of the mean distance techniques. between cells (about 10 Ȗm), which for a Although Carpenter et al.’s preferred Gargoyleosaurus diffusion-limited process indicates a pro- tree (Fig. 1a) is the most parsimonious tree tease concentration of about 1 nM. for their data, the alternatives in which Gar- Nodosauridae For simplicity, we assumed uniform goyleosaurus is the sister taxon of either the protease concentration when deriving the Nodosauridae (Fig. 1b) or Nodosauridae + last equation. However, our results general- Ankylosauridae (Fig. 1c) are only slightly Ankylosauridae ize to non-uniform conditions provided less parsimonious. Of their 26 characters, Figure 1 The three possible trees of ankylosaur rela- that the protease is present over a range cor- just six support their preferred tree, four tionships (rooted on Scelidosaurus with the basal responding to several intercell distances support Gargoyleosaurus + Nodosauridae, polytomy unresolved). a, Tree based on the data of around the a-cell. This is likely to occur, as and two support Nodosauridae and Anky- Carpenter et al.1, which is the most parsimonious protease-secreting a-cells are distributed losauridae. The remaining 14 characters are tree (tree length of 32 steps); b, the next most parsi- throughout the population and the protease phylogenetically uninformative. monious tree (tree length of 34 steps); c, the least 4,8 2 is a stable, widely diffusing molecule . The ‘winning sites’ test shows that the parsimonious tree (tree length of 36 steps). Mechanisms that screen concentration differences between Carpenter et al.’s pre- gradients may be more generally applicable. ferred tree and the slightly less parsimo- one of the six characters used to support Cells of mating type Ȋ may have an activity nious alternatives shown in Fig. 1b and c Gargoyleosaurus + Anklyosauridae, is pre- that inactivates a-factor9. Screening may are not significant (P values of 0.75 and sent in several ankylosaurids (for example, also be used in chemotaxis; for example, the 0.29, respectively), so they do not allow us Talarurus, Tarchia and Maleevus)6. Recod- slime mould Dictyostelium secretes factors to identify any tree as being significantly ing this character as polymorphic for the that antagonize the attractants cyclic AMP better supported by the data than any other. aggregate Ankylosauridae reduces the and folate. The effective range of signalling Similarly, bootstrap support for the associa- number of potential synapomorphies for by diffusible growth factors may also be reg- tion of Gargoyleosaurus and the Anky- Gargoyleosaurus + Ankylosauridae to just ulated by a similar mechanism. For exam- losauridae is not compelling (71%, 2,000 five, and highlights the possibility that ple, the growth factor Spitz is inhibited in replicates). Matrix randomization tests also Gargoyleosaurus might nest within the the developing Drosophila eye10. Studies of indicate the limitations of the data. Permu- Ankylosauridae. model systems, such as that discussed here, tation tail probabilities for the data based Carpenter et al.1 noted that strong brain may provide a better understanding of the on parsimony tree length3 and pairwise flexure in Gargoyleosaurus is shared with at quantitative effects of mechanisms that character compatibilities4,5 are 0.51 and least some nodosaurids, but they interpret- control the range of signalling. 0.43, respectively, and do not allow us to ed this similarity as primitive because the Naama Barkai*†, Mark D. Rose†, reject the null hypothesis that congruence ornithopod Hypsilophodon has a similar Ned S. Wingreen‡ within the data is no greater than expected condition. In contrast, comparison with Departments of *Physics and †Molecular Biology, by chance alone. Stegosauria, a more proximate outgroup, Princeton University, Taken together, these results indicate suggests that this similarity is derived with- Princeton, New Jersey 08544, USA that Carpenter et al.’s claim that their pre- in Ankylosauria, increasing the evidence for ‡NEC Research Institute, ferred tree is robust is not justified. Their Gargoyleosaurus + Nodosauridae to five 4 Independence Way, conclusions may also be undermined by potential synapomorphies. That these Princeton, New Jersey 08540, USA their use of aggregate in-group terminal minor revisions to the character data can e-mail: [email protected] taxa (nodosaurids and ankylosaurids), render the hypotheses represented in Fig. 1a

1. Jackson, C. & Hartwell, L. H. Cell 63, 1039–1051 (1990). which precludes the placement of Gar- and b equally parsimonious underscores 2. Jackson, C. & Hartwell, L. H. Mol. Cell. Biol. 10, 787–796 (1990). goyleosaurus within either of these clades. the frailty of the data and the phylogenetic 3. Segall, J. E. Proc. Natl Acad. Sci. USA 90, 8332–8336 (1993). Consequently, a phylogenetically important conclusions. 4. Hicks, J. B. & Herskowitz, I. Nature 260, 246–248 (1976). position for Gargoyleosaurus as sister taxon Gargoyleosaurus may well be an anky- 5. Ciejek, E. & Thorner, J. Cell 18, 623–635 (1979). 6. Sprague, G. F. & Herskowitz, I. J. Mol. Biol. 153, 305–321 (1981). to a major clade is inevitable. losaurid, but this inference has not yet been 7. Chan, R. K. & Otte, C. A. Mol. Cell. Biol. 2, 11–20 (1982). Choice of taxa also has an effect on the demonstrated. Robust resolution of the 8. Manney, T. R. J. Bacteriol. 155, 291–301 (1983). character evidence. The assumed ple- phylogenetic placement of Gargoyleosaurus 9. Steden, M., Betz, R. & Duntze, W. Mol. Gen. Genet. 219, 439–444 (1989). siomorphic condition of character 22 (a and assessment of its implications for the 10. Wasserman, D. & Freeman, M. Trends Cell Biol. 7, 431–436 (1997). ‘neck’ at the base of the occipital condyle), evolution of Ankylosauria will require a NATURE | VOL 396 | 3 DECEMBER 1998 | www.nature.com Nature © Macmillan Publishers Ltd 1998 423 scientific correspondence

thorough analysis of a more comprehensive explanation, invoking a longer delay for the that of the strobed segment (ds) remains data set. processing of a flashing stimulus, was based constant. The latency-difference hypothesis Mark Wilkinson*†, Paul Upchurch*, on attentional mechanisms10. therefore predicts that the observed spatial Paul M. Barrett‡, David J. Gower§, According to the hypothesis based on lead of the moving central segment should Michael J. Benton§ differential visual latencies, the observed increase. *School of Biological Sciences, spatial lead of the moving central segment To test this prediction, we measured the University of Bristol, in Fig. 1a is directly proportional to the dif- spatial lead of the moving central segment Bristol BS8 1UG, UK ference between the latencies of the strobed as a function of the detectability of the cen- †Department of Zoology, and the moving central segments. For a tral segment while keeping the detectabilit8y Natural History Museum, given stimulus, the visual latency varies of the strobed segments constant. Here we London SW7 5BD, UK inversely with its luminance5–9, so the use detectability to refer to the number of ‡Department of Earth Sciences, observed spatial lead in the flash–lag para- log units of luminance (Lu) above the University of Cambridge, digm should vary according to the lumi- detection threshold; detectability of the Cambridge CB2 3EQ, UK nance of the strobed and moving central strobed segments was 0.3 Lu for subjects §Department of Earth Sciences, segments. Increasing the luminance of the S.S.P. and G.P., and 0.5 Lu for T.L.N. The University of Bristol, moving central segments but not that of the temporal lead of the moving central seg- Bristol BS8 1RJ, UK strobed segments should decrease the delay ment averaged across subjects increases sys-

1. Carpenter, K., Miles, C. & Cloward, K. Nature 393, 782–783 of the moving central segment (dm) while tematically from 20 to 70 ms when its (1998). detectability increases by 1.0 Lu (Fig. 1b). 2. Prager, E. M. & Wilson, A. C. J. Mol. Evol. 27, 326–325 (1988). a Increasing the luminance of the strobed 3. Faith, D. P. & Cranston, P. S. Cladistics 7, 1–28 (1991). segments while keeping that of the moving 4. Alroy, J. Syst. Biol. 43, 430–437 (1994). 5. Wilkinson, M. Biol. Rev. 7, 423–470 (1997). central segment constant should decrease ds, 6. Barrett, P. M., Hailu, Y., Upchurch, P. & Burton, A. C. J. Vert. while dm remains constant. The latency- Paleontol. 18, 376–384 (1998). difference hypothesis predicts that the b 100 observed spatial lead of the moving central

g segment should decrease and, if the lumi-

) 80 n i s

v nance of the strobed segments is high o m Moving ahead through ( 60 enough, the moving central segment should m t

f n be perceived to lag behind spatially. We test- o e differential visual latency 40 m d ed this prediction by measuring spatial lead a g S.S.P e e l as a function of the detectability of the s 20 l

l G.P. a a The time it takes to transmit information r strobed segments, while keeping the r o t 0 T.L.N p along the human visual pathways intro- n detectability of the moving central segment e m c duces a substantial delay in the processing e 0 0.5 1 1.5 2 constant (1.5 Lu above the detection thresh- of images that fall on the retina. This visual T old for subjects G.P. and T.L.N., and 0.8 Lu Detectability of moving latency might be expected to cause a mov- central segment (Lu) for S.S.P.). The observed temporal lead of ing object to be perceived at a position the moving central segment averaged across behind its actual one, disrupting the accu- subjects decreases systematically from 80 to c 100 racy of visually guided motor actions such ǁ30 ms as the detectability of the strobed g ) as catching or hitting, but this does not n segments increases by 1.5 to 2.0 Lu (Fig. 1c). i s

v 60

happen. It has been proposed that the per- o These results support predictions of the m (

m t ceived position of a moving object is latency-difference hypothesis and show that f n o e 20 extrapolated forwards in time to compen- the motion-extrapolation mechanism does m 1–3 d a sate for the delay in visual processing . g not compensate for stimulus-dependent e e l s

l —20 We have studied the spatial misalignment variations in latency. Indeed, theoretical l a a r r o perceived between moving and strobed t calculations show that the putative motion- p n

objects and find that it varies systematically e —60 extrapolation mechanism must be under- m c

e 0 1 2 3 4

with the luminance of the objects. Our T compensating by at least 120 ms to account results favour an explanation for these per- Detectability of strobed for the data in Fig. 1. But a motion-extrapo- ceived misalignments based on differential segments (Lu) lation mechanism that does not adequately visual latencies, rather than on motion compensate for variations in visual latency extrapolation. Thus, accurate visually guided Figure 1 Luminance-dependent misalignments would not appreciably improve the accura- motor actions are likely to depend on motor between moving and strobed targets. a, The stimu- cy of real-time visually guided behaviour. instead of perceptual compensation. lus was a continuously rotating central segment Gopathy Purushothaman*, Saumil S. Patel†, Evidence for a mechanism based on (40 rev minǁ1) and two flanking strobed segments Harold E. Bedell†, Haluk Ogmen* 1–3 motion extrapolation comes from the (5 ms). b, The observed temporal lead of the mov- *College of Engineering and †College of Optometry, 4 flash–lag phenomenon , in which a continu- ing central segment is shown as a function of its University of Houston, ously moving object is perceived to be ahead detectability (0ǃthreshold) for three subjects (S.S.P., Houston, Texas 77204, USA of a stationary strobed object when the two G.P. and T.L.N.). Positive values on the y-axis repre- e-mail: [email protected] retinal images are physically aligned. But sent a temporal lead in perceiving the moving cen- 1. Nijhawan, R. Nature 370, 256–257 (1994). because visual latency varies according to tral segment relative to the strobed segments. The 2. Nijhawan, R. Nature 386, 66–69 (1997). the properties of a stimulus, including its observed spatial lead was converted into a tempo- 3. Khurana, B. & Nijhawan, R. Nature 378, 566 (1995). 5–9 4. MacKay, D. M. Nature 181, 507–508 (1958). luminance , this mechanism would have to ral lead by dividing it by the velocity of the moving 5. Bacon, F. Novum Organum (Bills, London, 1620). compensate appropriately for a range of central segment. c, The observed temporal lead of 6. Cattell, J. M. Brain 9, 512–515 (1886). stimulus-dependent variations in latency to the moving central segment is shown as a function 7. Mansfield, R. J. W. Vis. Res. 13, 2219–2234 (1973). 8. Roufs, J. A. J. Vis. Res. 14, 853–869 (1974). ensure that real-time, visually guided resp- of the detectability of the strobed segments 9. Williams, J. M. & Lit, A. Vis. Res. 23, 171–179 (1983). onses are accurate. An alternative, previous (0ǃthreshold) for the same subjects. 10.Baldo, M. V. C. & Klein, S. A. Nature 378, 565–566 (1995). 424 Nature © Macmillan Publishers Ltd 1998 NATURE | VOL 396 | 3 DECEMBER 1998 | www.nature.com