- Email: [email protected]
Descriptive and predictive approaches to biodiversity measurement
contribute to the observed phenOtypiC variance In clutch size. The herrtability of avian clutch Size Is relatively low (0.2-0.37. Ref. 51, as expected for a life-history trait closely related to fltneS+. Environmental factors play an important role in &&rr,ining differences in clutch size within and between individuals. The observed phenotypic variation may be much more Important than just noise around an optimal genotypic value. We disagree only with Dewitt’s suggestion that ‘instead of seeking optimal phenotypic values, we mud seek optimal genotypic values’. Both approaches are required. To understand fully the evolutiirn of complex, phenotypically plastic traits in the wild we need a sound knowledge of the proxtlllaie iacloij _:,~,:----:ninr( ur,~,,llillll16 ,,&=iinn IUIIY.I.ll. in 111c!utch size in heterogeneous environments as well as the ultimate factors determining the form of the fitness functions in these various environments”. In the goldeneye experiments feferred to by Pdys;i and Milonoff. one group of birds had thiee eggs removed, another group had three eggs added and there was an unmanipuiated control group. The mean clutch size of the ‘egg removal’ group was 11.9, almost exactly three eggs greater than the mean of the control group (9.0). We assumed (wrongly it turns out) that the three removed eggs must have been returned to the nest after clutch completion. Since, as Pbysa and Milonoff stress in their letter, this was not the case, it is then rather puzzling that the ‘egg removal group’ should have a clutch size significantly greater than the control group, and not significantly different from the ‘egg addition’ group, if all that the ‘egg removal’ group did was simply to compensate for the removed eggs. The ‘egg removal’ birds must in fact have laid an average of 14.9 eggs. Either the experimental groups contained birds of very different ql;a!ity (particularly high in the egg removal group, low In the control group?), which makes the whok? experiment very difficult to interpret, or the ‘egg removal’ birds somehow overcompensated in their replacement of the removed eggs. One (night speclrlate that the latter could occur in response to perceived high predation risk, or be 3 consequence of a slow shutdown in the egg Production process, though neither of these seems very convincing. In any event, if the ‘egg removal’ birds are not higher quality birds than tne control group, thep they still got an enlarged brood by having been tricked into laying more eggs-that is, they are actually a full-cost group. Despite the confusion over the expel imental design. our interpretation of the results would then stand. TO clarify the two other points raised by Pijysl and Milcnoff. the data in Fig. 1 are standardized differences in fledging producbon between parents whose brood rvas increased and unmanipulated Control birds, as stated in the text of Box 1. This was done to facilitate comparison of the effect across species with very clrfferent brood sizes. In saying that the origmal clutch size of females was not COntrOlled, we were drawmg attention to the need to standardize the Initial clutch size of birds across treatment groups as much as possible. This would standardize both the quality of pairs to some extent. and the proportional mampulation theyexperience. The problems in the interpretation of the goldeneye experime!lt highlighted in this correspondence emphasize how important it is tjo have comparable groups in these kinds of e).perirnents.
answering Roy and Foote’s three questions concerning how to measure diversity. as though for the phylogenetic approach. First. regarding which taxonomic levels should be sampled, the phylogenetlc approach necessarily includes all levels within the chosen group, down to the selected umts. The choice of group depends on which values are agreed in a particular siluation. For measuring overall biooiversity value, Ideally the group would be all life and the units would be individual organisms. Monaghan. P. and Nager. R.G. (1997) Trends In practice, it is usually necessary to work with Ecol. Evol. 12,270-274 surrogates from higher levels of organization. such 2 Gillesple.J.H.(1977)Am. Nat111.1010-1014 3 Joyce.MS. and Perrins.C.M.(1987) Ecology68, as species or familiesg,lO. Second, regarding which trails to score. the 142-153 phylogenetic approach was developed to assess a 4 Dhondt.A.A.etaL.1990) Naiure 348, 723-725 value of biological variety that people need to 5 Welgensberg, I. and Roff. D.A.(19961EvOlutfon conserve. A iialiow view of biological variety. 50.2149-2157 accepting only morphological disparity as valuable, 6 Falconer. D.S. (19891 Introductfon to Quant/tabre may be acceptable. However. in a changing and Gene&. 3rd edn, Longman uncertain world, many people see biodiversity 7 Price. T.D., Kirkpatnck. M. and Arnold, S.J. (1988) value as lying more broadly in the continued Science 240, 798-799 possibihties for adaptation and for sustainable 8 Alatalo. R.V.. Gustafsson. L. and Lundberg. A. exploitationll. This variety value is associated with (1990) Am. Nat. 135,464-471 richness in the different expressed genes or 9 Via. S. et al. 11995) irerids Ecol. f/o/. IO. characters of organisms, including not only 212-217 morphological characters, but also chemical, behavloural. ecologlcal and functional characters3,8. Third, regarding how to calculate diversity (value), the common problem is that only small samples of the broad range of characters are available. In an attempt to predict the distribution of richness for the majority of unsampled characters, the phylogenetic approach adopts the In their recent Tf?EEarbcle. Roy and Foote’ general model that valued characters are heritable, and show descent with modification. In suggest that ‘morphologlcal’ and ‘phyiogenetlc’ measures can be used to emphasize different practice, this requires a special pattern model, an aspects OFbiodiversity, but they do not explain estimate of the phylogenetic relationships among fully the Imost important difference. wllictl IS organisms. It alsfl requires a special process between the purposes that shape these model. an estimate of the way in which characters measures. Clanficatlon of these purposes may change with the pattern of relationship. There may help to highlight the different merits of Roy and be alternatives for both of these models Foote’s descripttve approach io morphology. and (competing estimates of phylogeny; and competrng the more predictive approach that uses phylogeny. empirical, ‘clock’-like and punctuated equilibrium Roy and Foote state that ‘&antltative data models of character change), so appropriate explicit!y designed to assess the geographic choices must be made for each situation”. distribution of morphological diversity should be What is at issue is not the universal superionty useful for understanding the distribution of of one approach over the other, but the question biodiversity in modern ecosystems.’ Their of when to adopt one or the other. Descriptive approach is an empirical, phenetic descrlptlon of a measures of morphological diversity are needed by sample of characters. They admit that ‘in practice those exploring patterns in space and time. We only a small number of features can be studied.’ believe that predictive measures are needed by We would argue that while a phenetic approach those with more applied questions, such as how may be well suited to tine description of variation to choose areas to represent as much potentially within these samples, it is not so well suited to useful biological variety as possible. Predicting patterns among the great majority of unsampled characters because of iiomoplasy2~~. In contrast, diversity measures based on genealogical relationship+, or ‘phylogenetic’ messure@, have been developed specifically to predict the distribution of an entire ‘popu!atton’ of characters using data from small samples”. 1’1 addition, this approach can cover a broader range of possible values than just morphological disparity. and consequently it is of broader (including applied) interest7,8. In particular, phylogenetic measures have been considered as a Dept of Animal and Plant Sciences, Part of a scale of surrogate methods for comparing University of Sheffidd, Sheffield, the distribution of biodiversity value among biotas ilK SIO UN for use in assessments of conservation priorities9.i”. Some of the distinctions between the two 1 Roy, K. and Foote, M. (1997) Trends Em/. ho/. approaches may be better appreciated by 12,277-281
(:qwighrO 1997,Elsevirr Science Ltd. All rights reserved.l~169-5347/97!$17.00
Farris. J.S. (1977) in Major Patterns in Vertebrate Evolution (Hecht. M.K.. Goody, PC and Hecht. B.M., edsj. pp. 823-850. Plenum Wrlliams. P.H.. Gaston. K.J. and Humphnes. C.J. (1994) Biodivers. LeR. 2. 67-78 Vane-Wright. R.I.. Humphries. C.J. and Williams, P.H. (1991) 5rol. Conserv. 55. 235-254 Nixon. K.C. and Wheeler. Q.D. (1992) in EHinction and Phylogeny (Novacek. M.J. and Wheeler, Q.D.. eds). pp. 216-234, Columbra Llniversity Press Faith, Q.P. (1992) Broj. Conserv. 61, l-10 Vane-Wright. R.I. (1996) Ann. MO Bat. Gard. 83. 47-57 Williams. PH. et al. (1997) Trends Ecol. Eva!. 12. 66-67 Wiliiams, PH. and Humphries. C.J. (1996) in Biodiversity: A Bology of Numbers and D#ferexe (Gaston. K.J., ed.). pp. 54-76. Blackweii Williams, P.H. (1996) 11//161.Cower.. 1. 12-14 Reio, W.V.(1994)inSystematics and Consewation Evaluation IFor’~y. P.I.. , Humphnes. C.J. and Vane-Wright. RI.. edsl. pp. l-13, Oxford Unrversity Press
which is how ‘to iestimate] rhe distributrorr of richness for the majority of unsampled characters.’ Tine crucial questions here are: !l) How well, for an assumed model of character evolution, do true and esttmated character richness correlate? (2) How does the correiatron vary with the true model of cf;aracter evoiutlon and wrth the deviation between the true and assumed models? Whether one method is better at estimating total character diversity from sampled character diversity is an open question to be settled empirically. A detailed exploration of which approaches ,)erform better under which evolutionary circumstances would indeed be worthwhile. The problem is very similar to one that exists in the estimation of genealogicai relationshipsi. Monte Cario simuiatton WIII kkeiy be useful here just as it IS rn assessing the accuracy of phylogenehc methods, since, with this approach, we knGw what we are trying to estimate:. There is also a clear need for jackknife-style analyses comparing obsenjed richness of one subset of characters with richness estimated on the basis of a different subset.
Dept of Geophysical Sciences, University of Chicago, Ckicago, IL 60637. USA We welcome the perspective Willrams el al. present on measurmg biodiversity. As paleontologists, we emphasize fossrlizable traits, but certainly agree tha? biodiversity measures should consider Ither traits when posstble. Although we agree that different approaches are suited to different questions, the distincbon between ‘morphulogrcal’ (in thus context. ‘phenetrc’) and ‘phylogenetic’ measures ;s not equtvalent to that between descriptive and predictive approaches. (This raisst a necessary semantic point. Wrlliams et al. use obsemed character data to ‘predict’ unobserved character diversity. If they are truly foretelling what someone will potentially observe in the future, we can accept this as prediction, However, if, as we suspect, the observed data are used as a proxy for a quanbty that is unlikely ever to be observed, the operation in question is estimation.) In fact, either family of methods can be used to describe or to estimate. One could, tor example, use a genealogy as nothing more than a description of evolutionary history. Likewise, one could use phenetic-distance data for more than description, For example, from an observed temporal acceleration in morphological diversity, we may infer that an evolutionary mode: of variable evolutionary :ates is more likely than a constant-rate mode?. Observed phenetic distances corJld also be used to estrmate unobserved character drstances if we dared that ail characters followed the same evolutionary model. Such esbmabon, as Wrlliams et al. are aware, naturally requires more burdensome assumptions. Although we thmk that drversrty exists in Its own right, and are uncomfortab!e m equatrng it with ‘value’ (a subjective. human construct). we will accept, ior the sake of agument, the importance of Williams et a/.‘~ ‘applied’ (i.e. conservatron-related) issue, one aspect of
Y Dept of Biology, University of Califoma, San Diego. La JoHa. CA 92093-0116, US.4 e&mSMxs % Foote. M. (1997) Aanu. Rev. Ecol. Syst. 28. 129-152 2 Huelsenbeck. J.P. (1995) Cyst. BKJI. 44, IT-48
--
in a recent TREEnews 4
high relatedness between parents tramiatesIntO lower offspnng vrabilityj. Thus effect also appears, but does COt attaln StatiStiCal srgnificance. rn the smaller present data se:. As expected, the number Of unhatched eggs in a clutch tended to he higher when parents were more closely related. both iri absolute terms and the proportion of these eggs in a clutch. Although neither of these ;or~eiat~ons was statistically significan: (P=Q.12; p=O.ZO. respectively), both were in the predicted direction and close to significant with a one-taited tes?. Thus, more closely related parents may have higher offspring monaiity also in the present data set. In some species, paternity may be scored soon after fertilization by using PCR-derived markers which require little DPIA. However. due to decomposition of tissue, such analyses are problematic when a proportion of the incubation pertod takes place in utero (as rn sand lizard+). Thus, since this avenue was closed to us. we adopted the most conservative approach by assigning the paternity of all unhatched eggs to the male with the highest genetrc similarity wcrth the female, and reanalystng :he daia in our first contribution2. Even after this highly conservative ‘correction’. band-shering was negatively correlated with the proporhon of the clutch tha, a male sired (p=O.O24). In nme out of 11 clutches, the male with the lowest genetic similarity with the female still sired most of the offspring (p = 0.033). Other plausible COntributGrS to fertilization rates, such as time for sperm replenishment and male mass were not correlated with male proportional paternlty (r=0.26, p=O.63. respectively), or male relatedness wrth the female ([email protected]. p=O.76. respectively]. Thus, our data provide strong evidence that female sand lizards are able to control the paternrty of their offspring by selectwe use of sperm. Few studres m natural populations allow researchers t0 IOGk for RegatiVe effeCtS Of pare:::dl consanguinrty on offspring vrability: when they do. this relationship has been repeatedly confrrmed6-r. This scenano may have imtoortant rmpircarrons for the evolutron of female sperm choice and the question ‘Why do not the first sperm tG arrive at the fertilization site always fertilize the egg?‘8 lndrviduars do not carry the same (defect) genes/alleles and. hence, for several reasons the same male may be a good partner for one female and a bad one for anotherg-12. How females discrimmate anlOng sperm 15 not yet resolved but two nossrble ,,:“h?riiSrnb have been descnbedi: selection by ~“ctive tract. and choice of :ne ier, ;;_ ,.r?,nru... sperm by ova. In sand lizards, more closely related males also have more Similar proPorhonat paternity (p=O.O33). Thus. the more similar In chemical structure the sperm membraires. the greater may the dnfrculty be for maternal ;mmunorecognitron systems to dtstrngulsh among them.
The University of S#IIeY. School of Biological Sciences. Zoology Building AC% NSW 2OOC, Australia
All rights reserved.
0169-5343/9i!$li.OU
answering Roy and Foote’s three questions concerning how to measure diversity. as though for the phylogenetic approach. First. regarding which taxonomic levels should be sampled, the phylogenetlc approach necessarily includes all levels within the chosen group, down to the selected umts. The choice of group depends on which values are agreed in a particular siluation. For measuring overall biooiversity value, Ideally the group would be all life and the units would be individual organisms. Monaghan. P. and Nager. R.G. (1997) Trends In practice, it is usually necessary to work with Ecol. Evol. 12,270-274 surrogates from higher levels of organization. such 2 Gillesple.J.H.(1977)Am. Nat111.1010-1014 3 Joyce.MS. and Perrins.C.M.(1987) Ecology68, as species or familiesg,lO. Second, regarding which trails to score. the 142-153 phylogenetic approach was developed to assess a 4 Dhondt.A.A.etaL.1990) Naiure 348, 723-725 value of biological variety that people need to 5 Welgensberg, I. and Roff. D.A.(19961EvOlutfon conserve. A iialiow view of biological variety. 50.2149-2157 accepting only morphological disparity as valuable, 6 Falconer. D.S. (19891 Introductfon to Quant/tabre may be acceptable. However. in a changing and Gene&. 3rd edn, Longman uncertain world, many people see biodiversity 7 Price. T.D., Kirkpatnck. M. and Arnold, S.J. (1988) value as lying more broadly in the continued Science 240, 798-799 possibihties for adaptation and for sustainable 8 Alatalo. R.V.. Gustafsson. L. and Lundberg. A. exploitationll. This variety value is associated with (1990) Am. Nat. 135,464-471 richness in the different expressed genes or 9 Via. S. et al. 11995) irerids Ecol. f/o/. IO. characters of organisms, including not only 212-217 morphological characters, but also chemical, behavloural. ecologlcal and functional characters3,8. Third, regarding how to calculate diversity (value), the common problem is that only small samples of the broad range of characters are available. In an attempt to predict the distribution of richness for the majority of unsampled characters, the phylogenetic approach adopts the In their recent Tf?EEarbcle. Roy and Foote’ general model that valued characters are heritable, and show descent with modification. In suggest that ‘morphologlcal’ and ‘phyiogenetlc’ measures can be used to emphasize different practice, this requires a special pattern model, an aspects OFbiodiversity, but they do not explain estimate of the phylogenetic relationships among fully the Imost important difference. wllictl IS organisms. It alsfl requires a special process between the purposes that shape these model. an estimate of the way in which characters measures. Clanficatlon of these purposes may change with the pattern of relationship. There may help to highlight the different merits of Roy and be alternatives for both of these models Foote’s descripttve approach io morphology. and (competing estimates of phylogeny; and competrng the more predictive approach that uses phylogeny. empirical, ‘clock’-like and punctuated equilibrium Roy and Foote state that ‘&antltative data models of character change), so appropriate explicit!y designed to assess the geographic choices must be made for each situation”. distribution of morphological diversity should be What is at issue is not the universal superionty useful for understanding the distribution of of one approach over the other, but the question biodiversity in modern ecosystems.’ Their of when to adopt one or the other. Descriptive approach is an empirical, phenetic descrlptlon of a measures of morphological diversity are needed by sample of characters. They admit that ‘in practice those exploring patterns in space and time. We only a small number of features can be studied.’ believe that predictive measures are needed by We would argue that while a phenetic approach those with more applied questions, such as how may be well suited to tine description of variation to choose areas to represent as much potentially within these samples, it is not so well suited to useful biological variety as possible. Predicting patterns among the great majority of unsampled characters because of iiomoplasy2~~. In contrast, diversity measures based on genealogical relationship+, or ‘phylogenetic’ messure@, have been developed specifically to predict the distribution of an entire ‘popu!atton’ of characters using data from small samples”. 1’1 addition, this approach can cover a broader range of possible values than just morphological disparity. and consequently it is of broader (including applied) interest7,8. In particular, phylogenetic measures have been considered as a Dept of Animal and Plant Sciences, Part of a scale of surrogate methods for comparing University of Sheffidd, Sheffield, the distribution of biodiversity value among biotas ilK SIO UN for use in assessments of conservation priorities9.i”. Some of the distinctions between the two 1 Roy, K. and Foote, M. (1997) Trends Em/. ho/. approaches may be better appreciated by 12,277-281
(:qwighrO 1997,Elsevirr Science Ltd. All rights reserved.l~169-5347/97!$17.00
Farris. J.S. (1977) in Major Patterns in Vertebrate Evolution (Hecht. M.K.. Goody, PC and Hecht. B.M., edsj. pp. 823-850. Plenum Wrlliams. P.H.. Gaston. K.J. and Humphnes. C.J. (1994) Biodivers. LeR. 2. 67-78 Vane-Wright. R.I.. Humphries. C.J. and Williams, P.H. (1991) 5rol. Conserv. 55. 235-254 Nixon. K.C. and Wheeler. Q.D. (1992) in EHinction and Phylogeny (Novacek. M.J. and Wheeler, Q.D.. eds). pp. 216-234, Columbra Llniversity Press Faith, Q.P. (1992) Broj. Conserv. 61, l-10 Vane-Wright. R.I. (1996) Ann. MO Bat. Gard. 83. 47-57 Williams. PH. et al. (1997) Trends Ecol. Eva!. 12. 66-67 Wiliiams, PH. and Humphries. C.J. (1996) in Biodiversity: A Bology of Numbers and D#ferexe (Gaston. K.J., ed.). pp. 54-76. Blackweii Williams, P.H. (1996) 11//161.Cower.. 1. 12-14 Reio, W.V.(1994)inSystematics and Consewation Evaluation IFor’~y. P.I.. , Humphnes. C.J. and Vane-Wright. RI.. edsl. pp. l-13, Oxford Unrversity Press
which is how ‘to iestimate] rhe distributrorr of richness for the majority of unsampled characters.’ Tine crucial questions here are: !l) How well, for an assumed model of character evolution, do true and esttmated character richness correlate? (2) How does the correiatron vary with the true model of cf;aracter evoiutlon and wrth the deviation between the true and assumed models? Whether one method is better at estimating total character diversity from sampled character diversity is an open question to be settled empirically. A detailed exploration of which approaches ,)erform better under which evolutionary circumstances would indeed be worthwhile. The problem is very similar to one that exists in the estimation of genealogicai relationshipsi. Monte Cario simuiatton WIII kkeiy be useful here just as it IS rn assessing the accuracy of phylogenehc methods, since, with this approach, we knGw what we are trying to estimate:. There is also a clear need for jackknife-style analyses comparing obsenjed richness of one subset of characters with richness estimated on the basis of a different subset.
Dept of Geophysical Sciences, University of Chicago, Ckicago, IL 60637. USA We welcome the perspective Willrams el al. present on measurmg biodiversity. As paleontologists, we emphasize fossrlizable traits, but certainly agree tha? biodiversity measures should consider Ither traits when posstble. Although we agree that different approaches are suited to different questions, the distincbon between ‘morphulogrcal’ (in thus context. ‘phenetrc’) and ‘phylogenetic’ measures ;s not equtvalent to that between descriptive and predictive approaches. (This raisst a necessary semantic point. Wrlliams et al. use obsemed character data to ‘predict’ unobserved character diversity. If they are truly foretelling what someone will potentially observe in the future, we can accept this as prediction, However, if, as we suspect, the observed data are used as a proxy for a quanbty that is unlikely ever to be observed, the operation in question is estimation.) In fact, either family of methods can be used to describe or to estimate. One could, tor example, use a genealogy as nothing more than a description of evolutionary history. Likewise, one could use phenetic-distance data for more than description, For example, from an observed temporal acceleration in morphological diversity, we may infer that an evolutionary mode: of variable evolutionary :ates is more likely than a constant-rate mode?. Observed phenetic distances corJld also be used to estrmate unobserved character drstances if we dared that ail characters followed the same evolutionary model. Such esbmabon, as Wrlliams et al. are aware, naturally requires more burdensome assumptions. Although we thmk that drversrty exists in Its own right, and are uncomfortab!e m equatrng it with ‘value’ (a subjective. human construct). we will accept, ior the sake of agument, the importance of Williams et a/.‘~ ‘applied’ (i.e. conservatron-related) issue, one aspect of
Y Dept of Biology, University of Califoma, San Diego. La JoHa. CA 92093-0116, US.4 e&mSMxs % Foote. M. (1997) Aanu. Rev. Ecol. Syst. 28. 129-152 2 Huelsenbeck. J.P. (1995) Cyst. BKJI. 44, IT-48
--
in a recent TREEnews 4
high relatedness between parents tramiatesIntO lower offspnng vrabilityj. Thus effect also appears, but does COt attaln StatiStiCal srgnificance. rn the smaller present data se:. As expected, the number Of unhatched eggs in a clutch tended to he higher when parents were more closely related. both iri absolute terms and the proportion of these eggs in a clutch. Although neither of these ;or~eiat~ons was statistically significan: (P=Q.12; p=O.ZO. respectively), both were in the predicted direction and close to significant with a one-taited tes?. Thus, more closely related parents may have higher offspring monaiity also in the present data set. In some species, paternity may be scored soon after fertilization by using PCR-derived markers which require little DPIA. However. due to decomposition of tissue, such analyses are problematic when a proportion of the incubation pertod takes place in utero (as rn sand lizard+). Thus, since this avenue was closed to us. we adopted the most conservative approach by assigning the paternity of all unhatched eggs to the male with the highest genetrc similarity wcrth the female, and reanalystng :he daia in our first contribution2. Even after this highly conservative ‘correction’. band-shering was negatively correlated with the proporhon of the clutch tha, a male sired (p=O.O24). In nme out of 11 clutches, the male with the lowest genetic similarity with the female still sired most of the offspring (p = 0.033). Other plausible COntributGrS to fertilization rates, such as time for sperm replenishment and male mass were not correlated with male proportional paternlty (r=0.26, p=O.63. respectively), or male relatedness wrth the female ([email protected]. p=O.76. respectively]. Thus, our data provide strong evidence that female sand lizards are able to control the paternrty of their offspring by selectwe use of sperm. Few studres m natural populations allow researchers t0 IOGk for RegatiVe effeCtS Of pare:::dl consanguinrty on offspring vrability: when they do. this relationship has been repeatedly confrrmed6-r. This scenano may have imtoortant rmpircarrons for the evolutron of female sperm choice and the question ‘Why do not the first sperm tG arrive at the fertilization site always fertilize the egg?‘8 lndrviduars do not carry the same (defect) genes/alleles and. hence, for several reasons the same male may be a good partner for one female and a bad one for anotherg-12. How females discrimmate anlOng sperm 15 not yet resolved but two nossrble ,,:“h?riiSrnb have been descnbedi: selection by ~“ctive tract. and choice of :ne ier, ;;_ ,.r?,nru... sperm by ova. In sand lizards, more closely related males also have more Similar proPorhonat paternity (p=O.O33). Thus. the more similar In chemical structure the sperm membraires. the greater may the dnfrculty be for maternal ;mmunorecognitron systems to dtstrngulsh among them.
The University of S#IIeY. School of Biological Sciences. Zoology Building AC% NSW 2OOC, Australia
All rights reserved.
0169-5343/9i!$li.OU