lection clearly can favor sperm quality (e.g., length) at the expense of sperm quantity, even when males have limited resources for ga- mete production (26). Although we now understand what drives sperm length evolution, we do not know what is driving the evolution of SR length. None- theless, this trait offers an exceptionally trac- table system for studying the evolution of a female preference and of male-female inter- actions. The functional relationship between the female preference and the corresponding male ornament is unambiguous, the prefer- ence and ornament are both easy to quantify, the macroevolutionary pattern of coevolution between the preference and ornament has been established (17), costs of relative ex- pression of each have been quantified (19, 23, 26), and each is amenable to genetic analysis and artificial selection (27). Our results are consistent with several mod- els developed to explain the evolution of female mate preferences. Linkage disequilibrium be- tween the female preference and male ornament is consistent with the Fisherian runaway pro- cess and ���good genes��� models (28). Also con- sistent with good genes models, recent studies have suggested a link between male condition and sperm quality (29), including sperm length (30). Next, interactions between the sexes are rife with conflict in D. melanogaster (31) and the coevolution of sperm and SR length may be sexually antagonistic, as has been suggested for sperm length and sperm-storage tubule length in birds (15). Finally, data reported here refute predictions of two sexual selection models as applied to this system. First, the ���direct bene- fits��� model (28) cannot apply, as the long sperm tails are not absorbed by females and have not evolved to serve a post-fertilization function (32). Second, the ���sensory exploitation��� model (28) is not applicable, as phylogenetic analysis reveals a pattern of correlated evolution be- tween the female preference and male trait (17) rather than a pattern of the male trait evolving in response to a preexisting female bias. The sperm-female coevolution demon- strated here has important implications for diversification and speciation. Rapid mor- phological divergence of sperm has been re- ported for numerous taxa, including primates (33). Such divergence has been shown to drive correlated divergence of important life history traits (19, 26). Further, as sperm mor- phology and sperm usage by females are central to successful reproduction, their di- vergence will likely contribute to reproduc- tive isolation between populations and the formation of new species (1, 7, 34). References and Notes 1. W. J. Swanson, V. D. Vacquier, Nature Rev. Genet. 3, 137 (2002). 2. G. J. Wyckoff, W. Wang, C.-I. Wu, Nature 403, 304 (2000). 3. K. A. Sutton, M. F. Wilkinson, J. Mol. Evol. 45, 579 (1997). 4. V. D. Vacquier, Science 281, 1995 (1998). 5. W. J. Swanson, A. G. Clark, H. M. Waldrip-Dail, M. F. Wolfner, C. F. Aquadro, Proc. Natl. Acad. Sci. U.S.A. 98, 7375 (2001). 6. W. R. Rice, Nature 381, 232 (1996). 7. G. A. Parker, L. Partridge, Philos. Trans. R. Soc. London Ser. B 353, 261 (1998). 8. A. G. Clark, D. J. Begun, T. Prout, Science 283, 217 (1999). 9. S. Gavrilets, Nature 403, 886 (2000). 10. W. J. Swanson, Z. Yang, M. F. Wolfner, C. F. Aquadro, Proc. Natl. Acad. Sci. U.S.A. 98, 2509 (2001). 11. G. Arnqvist, L. Rowe, Evolution 56, 936 (2002). 12. B. G. M. Jamieson, The Ultrastructure and Phylogeny of Insect Spermatozoa (Cambridge Univ. Press, Cam- bridge, 1987). 13. L. W. Simmons, Sperm Competition and Its Evolution- ary Consequences in the Insects (Princeton Univ. Press, Princeton, NJ, 2001). 14. C. W. LaMunyon, S. Ward, Proc. R. Soc. London Ser. B 269, 1125 (2002). 15. J. V. Briskie, R. Montgomerie, T. R. Birkhead, Evolution 51, 937 (1997). 16. M. J. G. Gage, Proc. R. Soc. London Ser. B 258, 247 (1994). 17. S. Pitnick, T. A. Markow, G. S. Spicer, Evolution 53, 1804 (1999). 18. T. R. Birkhead, Evolution 54, 1057 (2000). 19. S. Pitnick, T. A. Markow, G. S. Spicer, Proc. Natl. Acad. Sci. U.S.A. 92, 10614 (1995). 20. G. T. Miller, S. Pitnick, data not shown. 21. Female and male body sizes were measured. Initial analyses of P2 by analysis of covariance (ANCOVA), with female and male lines as main factors and male and female sizes as covariates, revealed that size and its interaction with other variables were never signif- icant and inclusion of body sizes never improved model fit. Entering the variable ���remating interval��� as a covariate was only justified in the analysis for the third replicate experiment. 22. E. H. Morrow, M. J. G. Gage, Proc. R. Soc. London Ser. B 268, 2281 (2001). 23. G. T. Miller, S. Pitnick, J. Evol. Biol., in press. 24. S. Pitnick, unpublished data. 25. T. C. M. Bakker, A. Pomiankowski, J. Evol. Biol. 8, 129 (1995). 26. S. Pitnick, Am. Nat. 148, 57 (1996). 27. G. T. Miller, W. T. Starmer, S. Pitnick, Heredity 87, 25 (2001). 28. M. Andersson, Sexual Selection (Princeton Univ. Press, Princeton, NJ, 1994). 29. A. Rakitin, M. M. Ferguson, E. A. Trippel, Can. J. Fish. Aquat. Sci. 56, 2315 (1999). 30. L. W. Simmons, J. S. Kotiaho, Evolution 56, 1622 (2002). 31. S. Pitnick, F. Garc�� ��a-Gonza ��lez, Proc. R. Soc. London Ser. B, 269, 1821 (2002). 32. T. L. Karr, S. Pitnick, Nature 379, 405 (1996). 33. M. J. G. Gage, Proc. R. Soc. London Ser. B 265, 97 (1998). 34. P. E. Eady, J. Zool. (London) 253, 47 (2001). 35. S. Pitnick, G. T. Miller, J. Reagan, B. Holland, Proc. R. Soc. London Ser. B 268, 1071 (2001). 36. We thank J. Reagan and D. Trinkaus for technical assistance and J. Alcock, T. R. Birkhead, A. Bjork, W. D. Brown, T. L. Karr, L. A. McGraw, M. Polak, R. R. Snook, W. T. Starmer, and L. L. Wolf for comments on an earlier draft of the manuscript. Supported by NSF grants DEB-9806649 and DEB-0075307 (S.P.). 5 August 2002 accepted 9 September 2002 Ecological Predictions and Risk Assessment for Alien Fishes in North America Cynthia S. Kolar*��� and David M. Lodge Methods of risk assessment for alien species, especially for nonagricultural systems, are largely qualitative. Using a generalizable risk assessment approach and statis- tical models of fish introductions into the Great Lakes, North America, we devel- oped a quantitative approach to target prevention efforts on species most likely to cause damage. Models correctly categorized established, quickly spreading, and nuisance fishes with 87 to 94% accuracy. We then identified fishes that pose a high risk to the Great Lakes if introduced from unintentional (ballast water) or intentional pathways (sport, pet, bait, and aquaculture industries). Increased trade and tourism associated with globalization have facilitated one of the least reversible human-induced global changes now under way: the homogenization of Earth���s biota through the establishment and spread of alien species (1, 2). Given the myriad detrimental impacts attributed to alien species in invaded ecosystems (3, 4) and the limited possibilities for eradication, predict- ing potential alien species and preventing their establishment are important policy goals (5). Invasion biology has, however, been plagued by a paradox that has hindered pre- vention. On the one hand, there is a wide- spread perception that diagnostic characteris- tics of weedy species have long since been identified (6). Current risk-screening proto- cols, such as the Weed Risk Assessment of Australia (7) and the Ecological Risk Assess- ment Framework of the U.S. Government (8), are based on largely qualitative categoriza- tions of such putative diagnostic characteris- tics. On the other hand, there is a widespread perception that predictions about which species will invade are impossible (9). This perception has emerged from searching for Department of Biological Sciences, University of Notre Dame, Notre Dame, IN 46556, USA. *Present address: U.S. Geological Survey Upper Mid- west Environmental Sciences Center, 2630 Fanta Reed Road, LaCrosse, WI 54603, USA. ���To whom correspondence should be addressed. E- mail: ckolar@usgs.gov R E P O R T S www.sciencemag.org SCIENCE VOL 298 8 NOVEMBER 2002 1233

characteristics that apply generally to all tax- onomic groups and in all ecosystems (6). It should not be surprising, however, that such overarching characteristics do not exist. Fur- ther, growing evidence suggests that charac- teristics important at the spread stage differ from those important to other stages of the invasion sequence (10). Recognition of the incommensurability of these two perceptions, therefore, is the key to the paradox and is the basis on which we build an approach to quan- titative, predictive risk assessments of alien species consistent with recommendations of the U.S. National Research Council (11). Here, we develop quantitative models using species characteristics to predict po- tential alien species and their impact. A similar approach has been used to predict invasiveness of terrestrial plants (12, 13) and provides the foundation of our ap- proach. We control for factors not usually considered explicitly in previous investiga- tions: We examine one ecosystem (the Great Lakes of North America) and one taxon (fishes) and consider invasion stages independently (establishment, spread, and impact) (Fig. 1). We develop and use mul- tivariate models to assess the risk to the Great Lakes from fishes introduced unin- tentionally from ballast water or intention- ally from the aquaculture, bait, sport, or pet industries. Such models and predictions could provide the basis for quantitative risk assessment and management tools essential in reducing the threat of alien species (5). This approach could also be extended to other taxonomic groups and ecosystems. We first identified the 24 established and 21 introduced but not established alien fishes in the Great Lakes (14) (table S1), and then further identified the established fishes that spread quickly through the ecosystem (tables S2 and S3) and those that are perceived as a nuisance. We then collected data from the literature on 13 life-history characteristics, 5 habitat needs, 6 aspects of invasion history, and human use (table S4). Unfortunately, we could not perform phylogenetically indepen- dent contrasts, because the systematics of fishes is not sufficiently understood. To re- duce the likelihood that significant associa- tion was due to phylogenetic similarity, we included a variable ranking fish families by degree of derived characters (15). Discriminant analysis (DA) revealed that successful fishes in the establishment stage (Fig. 1) grew relatively faster, tolerated wider ranges of temperature and salinity, and were more likely to have a history of invasiveness than were failed fishes. A discriminant func- tion using these four characteristics discrim- inated between failed and successful fishes with 87% accuracy (83% in jackknife valida- tion) (16, 17). Categorical and regression tree analysis (CART), using minimum tempera- ture threshold, diet breadth, and two mea- sures of relative growth, classified failed and successful fishes with 94% accuracy (82% upon cross-validation) (Fig. 2) (18). Quickly spreading fishes had slower relative growth rates, survived poorly in high water temperatures, and tolerated a wider temperature range than did slowly spreading fishes (dis- criminant function 94% accurate 90% in jack- knife validation) (Fig. 1, spread stage) (19) (supporting online text). Nuisance fishes had smaller eggs, had wider salinity tolerances, and survived in lower water temperatures than did nonnuisance fishes (discriminant function 89% accurate 80% in jackknife validation) (Fig. 1, impact stage) (20) (supporting online text). For each stage of the invasion sequence, only three or four characteristics were necessary to cor- rectly classify 87 to 94% of alien fishes docu- mented in the Great Lakes. Different traits were important for different stages of invasion. For example, relatively fast growth was positively associated with estab- lishment but was negatively associated with quickly spreading species. These patterns con- firmed the necessity of invasion stage���specific analyses (supporting online text). Overall, our results demonstrate that quantitative analyses that are ecosystem specific, taxon specific, and stage specific provide a firm quantitative basis for risk assessment. We next used our models to predict the risk to the Great Lakes from potential unintentional introductions through ballast water from the Ponto Caspian basin (Black Sea, Caspian Sea, and surrounding watersheds), a recent source of alien species in the Great Lakes (21). We found sufficient species characteristics for 66 (out of 110) of these fishes (22). DA predicted that 24 species could become established CART pre- dicted 36. The predictions of DA and CART were 57% similar [Jaccard���s similarity coeffi- cient (23)]. We suggest that the 22 fishes com- mon to both predictive models (Table 1) pose Fig. 1. Invasion by alien species is a process consisting of several transitions, each with an inde- pendent probability of failure, and cumulative failure rates are high. Here, we first compare characteristics of failed with suc- cessful fishes in the Great Lakes to predict fishes capable of be- coming established in the future. To further predict high-risk po- tential invaders, we then com- pare characteristics of successful fishes that spread quickly with those that spread slowly and those of fishes that are perceived as a nuisance with those that are not. Fig. 2. CART decision tree of successful and failed introduced fishes in the Great Lakes. Ovals represent decision points rectangles are terminal points in the tree resulting in classification. The numbers of known successful and failed alien species categorized into each terminus are given, illustrating that 2 of 45 species were misclassified. R E P O R T S 8 NOVEMBER 2002 VOL 298 SCIENCE www.sciencemag.org 1234