1. The association between seed size and habitat shade within the British flora was investigated using a data set of seed masses, life histories and quantitative measures of habitat shade for 504 species; the association between seed size and seed longevity was investigated using a data set of seed masses, life histories and seed longevities for 301 species.
2. The data were analysed using the method of phylogenetically independent contrasts (PICs) calculated using the software package CAIC (Comparative Analysis by Independent Contrasts).
3. Seed mass was found to be positively correlated with habitat shade and negatively correlated with seed longevity, after variation owing to life history had been accounted for.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The correlative and experimental evidence that large seeds are associated with tolerance of shade is abundant (see discussion in Westoby, Leishman & Lord 1996). Recently, however, analyses of comparative data sets have cast some doubt on this relationship (Mazer 1990; Kelly & Purvis 1993; Kelly 1995, 1996; but see Thompson & Hodkinson 1998), at least in so far as it exists independently of phylogeny. Difficulties which have afflicted many (but not all) correlative analyses, both in support of the relationship and otherwise, include the use of relatively few species, failure to account for possible confounding variables and subjective classification of species as occupying either ‘closed’ or ‘open’ habitats.
In this paper we present new analyses of both these relationships. In these analyses we (1) employ data for large numbers of species, (2) use the best available, objective, quantitative estimates of both habitat shade and seed persistence and (3) take account of the confounding effect of life history.
We consider the seed masses of the flora of an area of 3000km2 around Sheffield in central England. The region encompasses a wide range of geological strata, altitude and land use, and has a correspondingly diverse flora. It contains a large fraction of the British flora, excluding alpine and maritime species.
Air-dried seed masses for species in the flora were measured at the Unit of Comparative Plant Ecology (UCPE) from fresh collections (Hendry & Grime 1993). Habitat data were derived from the results of extensive field surveys, which involved recording the species composition and habitat characteristics of 10425lm2 quadrats. Only species recorded in at least 10 quadrats were used in this analysis. Each quadrat was recorded as shaded or unshaded on the basis of visual inspection; those where more than a small fraction of the sky above the quadrat was obscured were classed as shaded. Further details of the survey methods and habitat definitions can be found in Grime, Hodgson & Hunt (1988) and Hodgson et al. (1995), although some of the data analysed are previously unpublished. Each species was allocated a habitat shade score as follows:
(number of shaded quadrats where species was recorded – number of unshaded quadrats where species was recorded)/total number of quadrats where species was recorded.
This gives a score from –1 (recorded only from open quadrats) to 1 (recorded only from shaded quadrats). Species with no preference for shaded or open habitats score zero. The majority of species were recorded most often from open quadrats, with only 13% of species observed more frequently in shaded than open quadrats.
Seed persistence data were extracted from Thompson, Bakker & Bekker (1997), which classifies every available species record as either type 1 (transient, persistence < 1 year), type 2 (short-term persistent, persistence > 1 year but < 5 years) or type 3 (long-term persistent, persistence > 4 years) records. For the purpose of calculating a single longevity index for each species, we recognized only two classes of record: transient (type 1) and persistent (types 2 and 3).
Our longevity index is therefore defined as:
Σ (type 2 + type 3)
Σ (type 1 + type 2 + type 3)
which can take any value from 0 (no persistent records) to 1 (all records persistent). In the analyses reported here, we use this index as a measure of the persistence of seeds of individual species in the soil. Only species with at least 10 records were included in the analysis. For a fuller discussion of the rationale underlying this method of calculating a seed longevity index, see Thompson, Bakker & Bekker (1988).
On the basis of published sources (Clapham, Tutin & Warburg 1962; Grime et al. 1988; Stace 1991), we divided our species into four life-history classes (after Rees 1993); 1, annual; 2, biennial; 3, monocarpic perennial; 5, polycarpic perennial. Most species were either annuals (23%) or polycarpic perennials (69%); about 95% were herbaceous.
Seed mass, habitat shade and adult longevity data were available for 504 species; seed mass, seed longevity were available for 301 species. There were 283 species common to both datasets.
The strict consensus tree derived by Ken Rice from the data matrix for analysis II of Chase et al. (1993) was used as a phylogeny for plant families (http:// www.cis.upenn.edu/~krice/treezilla/index.html). The phylogeny for the genera within the Asteraceae was the majority-rule consensus tree using Dollo parsimony published in Jansen et al. (1990). Below the family level the taxonomy of Stace (1991) was used as a substitute for a phylogeny where none was available.
The method of phylogenetically independent contrasts (PICs) was used to analyse the data (Felsenstein 1985; Pagel 1992, 1994). This estimates the character states at nodes within the phylogeny. Relationships between the estimated character states at nodes within the phylogeny can be examined using traditional statistical tools. The null hypothesis is that there is no correlation between changes in traits at the nodes.
Contrasts were produced using the WinCAIC computer package written in the Delphi Pascal v 1.0 programming language, based on the source code for the CAIC (Comparative Analysis by Independent Contrasts) package for the Apple Macintosh computer (Purvis & Rambaut 1995a). This new package runs under the Microsoft Windows v 3.1 Graphical User Interface for IBM-compatible personal computers and performs the same calculations as the original program. It was evaluated using the data sets provided with CAIC v 2.0.0. The package can be obtained from the UCPE web site at http://www.shef.ac.uk/ ~nuocpe/.
It is possible to estimate the branch lengths between nodes on the phylogeny but relatively few of the species in the whole phylogeny were included in the constructed phylogeny. The branch length estimation algorithm assumes that taxon ages are proportional to the number of species they contain, therefore missing data seriously reduce the accuracy of the estimates. Furthermore, analyses of simulated data sets suggest that equal branch lengths may perform better than estimated branch lengths (Purvis, Gittleman & Luh 1994). Consequently, it was assumed that all branch lengths were the same.
Seed masses were logarithmically transformed in order to produce standardized contrasts that were independent of the seed mass at the node where the contrast was taken. No transformation was necessary for shade, seed longevity index or life history. Standardized linear contrasts were calculated using the Crunch procedure designed for continuous variables (Purvis & Rambaut 1995b). Two sets of contrasts were calculated; first, shade, seed mass and life history, and second, longevity index, seed mass and life history. Contrasts within each set were calculated simultaneously; seed mass was used as the predictor variable in each case.
Contrasts were analysed using multiple regression in Statistica v 5.1 (1997 release) and partial correlations were calculated for each predictor in the equation. These represent the independent contributions of each independent variable to the prediction of the dependent variable (Zar 1974) and therefore provide a test for the presence of correlated evolution independently of other variables in the analysis.
These data sets demonstrate the cross-species relationships observed by other authors. There are significant correlations between seed mass and both habitat shade (n = 504, r = 0·324, P < 0·001) and seed longevity (n = 167, r = –0·453, P < 0·001) (1Fig. 1a).
224 contrasts were calculated for habitat shade, seed mass and life-history class. These contrasts were analysed using multiple regression of the equation. Seed mass = β1× habitat shade + β2× life-history class (r2 = 0·106, F(2, 222) = 13·197, P < 0·001). Partial correlation showed significant positive relationships between seed mass and both habitat shade (β1 = 0·264, t(222) = 4·06, P < 0·001) and life-history class (β2 = 0·141, t(222) = 2·18, P = 0·030). That is, longer-lived species tend to have greater seed masses than short-lived species and, independently of that relationship, seeds of species associated with shaded habitats tend to be larger than seeds of species associated with open habitats.
Similarly, 167 contrasts were calculated for seed longevity, seed mass and life-history class, and analysed using multiple regression of the equation. Seed mass = β1× seed longevity + β2× life-history class (r2 = 0·110, F(2, 165) = 10·196, P < 0·001). Partial correlation showed a significant relationship between seed mass and seed longevity (β1 = –0·357, t(165) = –4·23, P < 0·001) but none between seed mass and life-history class (β2 = –0·059, t(167) = –0·71, P = 0·479). In this data set there is a tendency for smaller seeds to be longer-lived, independent of any relationship between seed mass and life-history class.
The determination of habitat shade for quadrats was subjective but because all quadrats were assessed by one person, it is known that the results are consistent. The large number of quadrats recorded allows a more objective assessment of the habitat shade associated with a particular species than a classification based on the habitat type as has been used in previous analyses, which may fail to reflect the conditions experienced by individuals if they are different to those of the habitat as a whole (Rees 1996). Problems still exist with the approach used. For example, the incident light in a habitat may change with season and some groups of plants are known to exploit this variability (vernal herbs grow in an unshaded habitat even though the habitat becomes shaded when the canopy closes). The treatment of shade as two classes during field recording is less satisfactory than treatment as continuous data because shade clearly forms a continuum from open to deeply shaded habitats. However, the use of proportional data goes a long way towards providing an appropriate continuous measure. Much the same arguments apply to the determination of seed persistence. Each published seed persistence record was allocated objectively to one of three persistence categories (Thompson et al. 1997) but the proportions of records in each category provides a continuous measure of persistence for those species with numerous records.
This study shows a positive correlation between seed mass and habitat shade amongst British plants, independently of the relationship between seed mass and life history. It therefore supports Salisbury’s (1942) hypothesis that plants of shaded habitats have larger seed masses than plants of unshaded habitats. In the only other comparative study of comparable size, Mazer (1990) was unable to detect any relationship between seed size and habitat shade independently of genus or family membership. Note, however, that although the correlation reported here is powerful evidence for a key role of habitat shade in the past evolution and present function of large seed size, the absence of such a correlation is not evidence for the absence of such a role. As Westoby et al. (1996) point out, there are numerous plausible models of trait evolution in which analysis of phylogenetically independent contrasts (PIC) are not appropriate. For example, PIC analyses attach little importance to the maintenance of a trait or to traits which have arisen only infrequently, in the extreme case only once.
This study also provides strong support for the hypothesized negative correlation between seed mass and seed longevity, independent of phylogeny and of any relationship between seed mass and life history. This result confirms the subjective impression that seed size and persistence are correlated at all taxonomic levels (Thompson et al. 1997). Although we are here concerned only with the Angiosperms, the Pinopsida are uniformly large seeded and transient. Within the Angiosperms, seeds in some orders are consistently large and short lived (e.g. Fagales), while others are consistently small and persistent (e.g. Juncales and Caryophyllales). At the opposite extreme, individual families with a wide range of seed sizes (e.g. Compositae, Gramineae) show a clear trend from large-seeded genera with short-lived seeds (e.g. Tragopogon, Bromus) to small-seeded genera with very persistent seeds (e.g. Gnaphalium, Agrostis). The mechanisms which are thought to link seed size and persistence are discussed in Thompson (1987) and Thompson et al. (1993).
Seed size is correlated with both persistence and habitat shade, independently of the possibly confounding effects of life history. A disturbing feature of the results, however, is that the two data sets fail to agree about the relationship between seed size and life history; the relationship is positive within the habitat shade data set but absent from the smaller seed longevity data set. Evidence that seed size is positively correlated with adult longevity is widespread (Salisbury 1942; Baker 1972; Silvertown 1981; Foster & Janson 1985; Rockwood 1985; Hogdson & Mackey 1986; Mazer 1989). Some more recent analyses, however, suggest that the situation is more complex. Using PIC on a data set similar to that analysed here, Rees (1993) found no relationship between seed size and life history, while Rees (1996) found that the existence of a relationship depends on dispersal mode, being inexplicably absent from species with animal-dispersed seeds. Leishman, Westoby & Jurado (1995) found that an apparent correlation between seed size and life history arose from independent correlations of both variables with plant height and growth form. Rees (1996) attributed much of the confusion to scale of measurement; the power of life history as a predictor of seed size is inevitably poor if it is measured simply as annual or perennial. Perhaps the safest conclusion is that the relationship between seed size and life history is not a simple one and will depend on the data set and measurement scale employed, as well as the choice of other traits included in the analysis.
We wish to thank all of those, notably Stuart Band, who contributed to measuring seed weights. Mark Westoby, Jonathan Silvertown, Susan Mazer, Mark Rees and two anonymous referees made valuable criticisms of earlier drafts. Funding was provided by NERC, partly through a research studentship to D.J.H.
Present address: ADAS Newcastle, Kenton Bar, Newcastle-upon-Tyne NE1 2YA, UK.