Heredity 74 (1995) 28—38 Received 9 February 1994 Genetical Society of Great Britain

The spatial genetic structure in natural populations of the Australian tree Atherosperma moschatum (Labill.) (Monimiaceae)

ALISON SHAPCOU* Department of Science, University of , GPO Box 252C, Hobart 7001, and Parks and Wildlife Service, Department of Environment and Land Management, GPO Box 44A, Hobart,

Thespatial distribution of genotypes at six enzyme loci was investigated in 17 natural populations of the temperate rainforest tree Atherosperina moschatum using spatial autocorrelation. The results from the different enzymes were consistent at each population as well as among populations, particularly at short distances. There were stronger associations overall between like homozygous trees than like heterozygous trees. Unlike trees were generally significantly negatively autocorre- lated at short distances. Two consistent patterns of associations were found with increasing distances between trees. A highly consistent average patch length was found in most populations for the enzymes studied. The results suggest that there are small neighbourhood sizes m this species and therefore small stands are worth conserving. In addition, larger populations are unlikely to be homogenous. Therefore, conserving one part of a population may not adequately sample its genetic variability.

Keywords: Athospermamoschatum, genetics,natural populations, spatial autocorrelation, temperate rainforest, trees.

Introduction conservation, or sampling them for breeding programmes. It should be taken into account in order The spatial distribution of genetic variability within to maximize diversity, and not to misrepresent species natural plant populations may significantly influence or population diversity (Epperson, 1989). If the evolutionary and ecological processes (Lewontin, presence of spatial structure is ignored, it may lead to 1974; Endler, 1977; Brown, 1978; Wright, 1978; misinterpretations of the significance of breeding Epperson, 1989). The spatial distribution of genetic systems or the selective significance of particular geno- variation in populations forms an integral part of neigh- types (Epperson, 1989). Genotypic structuring over bourhood concepts (Wright, 1978), as well as theories short distances within populations may be caused by of allopatric and parapatric speciation (Mayr, 1970). limited dispersal of pollen or seed, spread of vegetative Even when the scale of structure is fine, localized clones or by selection within a patchy environment genotypic combinations may arise, and small demes (Dewey & Heywood, 1988; Epperson, 1989). may become differentiated, despite being part of a Spatial autocorrelation analysis has been used to large ensemble (Wright, 1978). Survival may depend study genetic structure within plant populations (e.g. on tolerance of locally patchy environmental factors Epperson & Clegg, 1986; Dewey & Heywood, 1988; such as soil type (Bradshaw, 1984). Spatial structure of Fortin et a!., 1989; Schoen & Latta, 1989; Argyres & genetic variation within populations has immediate Schmitt, 1991; Schnabel et a!., 1991; Coates, 1992; ecological—genetic consequences for , since it Knowles et a!., 1992). Spatial autocorrelation analysis may result in aggregates of particular genotypes. makes no assumptions about the scale of the structure Knowledge of population spatial structure is (Dewey & Heywood, 1988), and since individuals can important when selecting natural populations for be used for the analysis, any scale of pattern can be analysed. The results of studies to date have revealed *Current address: Department of Botany, University of Queensland, localized structure over short distances in some St. Lucia, Brisbane 4072, Queensland, Australia. species' populations (e.g. Schoen & Latta, 1989; 28 GENETICS OF A THEROSPERMA MOSCHA TUM POPULATIONS 29

Argyres & Schmitt, 1991; Perry & Knowles, 1991, (1994). Stands were sampled along belt transects. All Schnabel et al., 1991; Wagner et a!., 1991). However, A. moschatum plants within transect boundaries were random genetic distributions have been found in other recorded and given a unique number, then sampled, species (Waser, 1987; Dewey & Heywood, 1988; and their height and diameter measured. Their loca- Epperson & Ailard, 1989; Coates, 1992). Differences tions were recorded as co-ordinates (X, Y), relative to in life histories, dispersal characteristics, breeding the transect line. In many cases transects included all or systems and stand histories may account for some of most of the trees in the stand. Transects varied in length the differences in results among species and popula- and breadth depending on the size, shape and density tions (Schoen & Latta, 1989; Xie & Knowles, 1991; of the stand sampled. They varied in length by multi- Coates, 1992). Fine-scale substructuring seems to be ples of the minimum length of 30 m up to 360 m. the most common kind of genetic structure within Width ranged from 5 to 40 m. Sometimes transects ran populations (Epperson, 1989). However, in some in parallel rather than end to end. Thus the dimensions species and populations (e.g. Delphinium nelsonii) gene of the belt transect and the shape of the stands deter- flow appears to be greater than expected from their mined the distances over which the analysis was biology and environment, leading to randomization of undertaken. their genetic structure (Waser, 1987). The actual co-ordinates of all A. moschatum trees Atherosperma moschatumis an Australian within the sample sites were used for the analysis. temperate rainforest tree. It is the most widespread of These formed an irregular lattice of sample points. the rainforest trees found in Tasmania (Australia) and Spatial autocorrelation analysis requires the construc- is typically one of the dominant canopy species. A. tion of a connection matrix (Sokal & Oden, 1978a) moschatum is likely to have restricted pollen dispersal defining which trees are considered 'neighbours' or since it is insect-pollinated, and though its plumed seed 'joined'. This allows flexibility in determining is capable of long-distance dispersal, most seed in relationships between trees, and allows customizing of closed canopy situations falls beneath the parent tree their connections. However the requirement to specify (Read & Hill, 1988). A. moschatum occurs in mixed- these parameters means that patterns may be missed if aged stands but vegetative reproduction is thought to the scale used does not detect them (Fortin et a!., be prolific and to account for much regeneration (Read 1989). The variation in autocorrelation and hence & Hill, 1988). F statistics (Wright, 1965) based on genetic relationships between trees that occurs with isozyme variation indicated that most variation was increasing distances between trees was analysed by within populations and that they were mostly inbred assessing autocorrelation with systematically increasing and not in Hardy-Weinberg equilibrium (Shapcott, distances and by constructing correlograms. These 1994). It was postulated that inbreeding could be the were used to investigate any pattern in relationships result of two factors: vegetative reproduction, and the between trees with different distances between them. development of family clusters due to poor pollen and Two scales of incremental distance increase between local seed dispersal. These attributes suggest that A. trees were used. moschatum stands are likely to exhibit genetic sub- In this study pairs of trees were considered to be structuring. Therefore this study used spatial auto- 'neighbours' within a radius of 10 m. Ten metres was correlationanalysis to investigate the spatial estimated as the approximate area of influence of an distribution of allozyme genotypes at six enzyme loci average A. inoschatum tree based on field observations within 17 natural stands of A. moschatum. of canopy area under which related seed were likely to fall and vegetative spread. This distance also took into account the density of trees and the average distance Methods between trees in the populations (Table 1), so that there Spatialautocorrelation analysis tests whether the would be enough 'neighbours' to give statistically observed value of a variable at one point or locality is meaningful results. Pairs of 'neighbour' trees were independent of values of the variable at neighbouring given a binary score of 1, if 'joined', by the pair defini- points or localities. If dependence exists the variable is tion, or 0, if not 'joined' within each annulus distance. said to exhibit spatial autocorrelation (Sokal & Oden, Thus all pairs of genotypes were assessed for the 1978a). For the purposes of this type of analysis the number of times they co-occurred within a distance distribution of points is considered as given (Sokal & annulus. This number was compared to the number of Oden, 1978a; Legendre & Fortin, 1989). joins expected and the variance for each type of join if Spatial autocorrelation of trees within stands was tree genotypes were randomly distributed (assuming calculated separately for six enzyme loci ineachof 17 sampling without replacement) (Sokal & Oden, 1978a). sites and the results compared. The genetic analysis If there were significant excesses of 'joins' between and stand structure of these sites is given in Shapcott genotype pairs, those pairs of genotypes were said to 30 A. SHAPCOTT

Table 1 Summary of the sample size, average distance between trees, tree density and the area sampled by the belt transect at each Atherosperma moschatum site

Average distance Area of Sample between trees transect Density Site size (m) (m2) (stems/rn2)

Mt. Field 40 3.2 0.0306 1470 Mt. Wellington 55 4.7 0.0145 3982 Bun Hill 62 3.5 0.0258 2400 Little Florentine 25 4.2 0.0178 1400 Creepy Crawly 60 4.4 0.0165 3640 Frodshams 44 2.2 0.0667 660 Bruny Island 25 4.6 0.0148 1690 Meetus Falls 45 3.2 0.032 1 1560 \Vyefield Rivulet 61 1.9 0.0920 750 Tahune 42 5.3 0.0115 3640 Newail Creek 46 3.5 0.0260 2160 AnthonyRoad 34 3.6 0.0250 1350 Murchison Highway 55 1.8 0.0950 600 Milkshake Reserve 51 3.8 0.0225 2400 Liffey Falls 60 5.9 0.0093 6450 Errinundra 30 3.1 0.0333 900 TarraBulga 28 7.3 0.0059 5025 be positively autocorrelated. A deficiency of joins indi- reliable due to the edge effects of the transects cated negative autocorrelation (Epperson & Clegg, (Epperson, 1989). Therefore, correlograms were not 1986). plotted at distances greater than 30 m .Toassist with In the case of nominal data such as these, the signifi- interpretation of results the SND values were averaged cance is assessed by the estimation of the standard for all like-genotype pairs or all unlike-genotype pairs normal deviate (SND), which is compared to a t- at each site, and these average SND values were plot- distribution (Sokal & Oden, 1 978a). Corrections for ted. All enzymes at each site were then compared on small sample sizes and small degrees of freedom are the same graph for these genotype-pair groupings. The given by Sokal & Oden (1978a). The observed and significance of correlograms was checked using the expected join counts and the variances were calculated Bonferroni method of correcting for multiple tests using a FORTRAN program based on the formulae (Oden, 1984), such that at least one SND value was given by Sokal & Oden (1 978a) as used by Zalucki et significant by this method (Legendre & Fortin, 1989) a!.(1987). where x'= xlv (x = 0.05significance level; Correlograms were used to investigate the patterns v =numberof tests). Since some values in all correlo- of association of genotypes with increasing distance at grams were significant at Bonferroni levels the correlo- each site. A correlogram plots the standard normal grams were generally considered significant (Legendre deviate (SND) of each pair-type at increasing distance & Fortin, 1989). However, the power and significance classes or annuli (Legendre & Fortin, 1989). Thus the at each distance class will be different due to the vary- scale of pattern can be investigated. Correlograms with ing number of connected pairs at each distance. There- 10 m distance intervals were initially constructed to fore, correlograms are most powerful at the shortest define broad patterns in the data. Correlograms with 5 distance classes (Epperson, 1989). Scatter plots map- m annulus increments were then constructed to clarify ping the location, diameter and genotypes of each tree and resolve more accurately the initial patterns were used as an aid in interpretation of the spatial auto- observed. Distance classes with increments smaller correlation results (e.g. Fig. 5). than 5 m was considered inappropriate given that the average distance between trees came close to this distance. Systematic investigation of autocorrelation with distances of multiples of 5 m was therefore an Results appropriate scale for the analysis. In this study patterns Theresults from the different enzymes were consistent at distances greater than 30 m are unlikely to be at each site as well as between sites, particularly at GENETICS OF A THEROSPERMA MQSCHA TUM POPULATIONS 31

Newall Creek GDH Correlograms for like and unlike pairs for each enzyme Genotypes of tree pairs were plotted together for each site. From these correlo- 12.00 AB-AB grams consistent patterns of results emerged. When a 10 m annulus was used half the sites (8 of the 17) dis- 0.00 AB-BB played a pattern where there were significant positive associations at short distances of around 10 m, a cross- z BB-BB Cl) ing of the X-axis (SND =0) at around 20 m, then -12.00 AB-AA becoming significantlynegatively autocorrelated (Murchison Highway, Creepy Crawly, Tarra Bulga, Mt. -24.00 BB-AA Wellington, Newall Creek, Tahune, Liffey Falls, Mt. 10 20 30 _ AAAA Field). Unlike-pair combinations were negatively auto- correlated at 10 m and continued to be significantly Distance (M) negatively associated at greater distances as at Tarra Fig.I Correlogram showing the change in spatial autocorre- Bulga (Fig. 2a). When tested with 5 m increasing lation with increasing distance between pairs of Athero- annuli, the pattern for like associations remained but sperma moschatum trees with like or different GDH became compressed with SND values generally (glutamate dehydrogenase) genotypes. becoming nonsignificant by 15 m and negative by 20 m as shown in the example of Tarra Bulga (Fig. 2b). Unlike pairs of trees were significantly negatively auto- correlated at all distances. short distances (10 m). Generally, pairs of trees of the When correlograms with 10 m annuli were used, a same genotype (like trees) were significantly positively second pattern was observed in seven sites (Meetus autocorrelated at short distances (up to 10 m). A Falls, Little Florentine, Errinundra, Bruny Is., Anthony comparison of all like-pair correlations in all enzymes Road, Bun Hill, Frodshams). Again it consisted of at all populations, found that overall 73 per cent were significant positive associations at 10 m for like geno- significantly positively autocorrelated at 10 m with types. Like autocorrelations dipped to nonsignificant at proportions ranging from 67 per cent to 83 per cent 20 m, but continued to be positive with greater between enzyme systems (Table 2). The remainder distance, though mostly nonsignificant (e.g. Meetus gave nonsignificant results (Table 2) which were Falls, Fig. 3a). Unlike pairs were initially significantly generally due to small numbers of trees of a particular negatively autocorrelated, but approached zero by 30 genotype. m (e.g. Fig. 3a). When correlograms with 5 m incre- There were stronger associations overall between ments in the annuli were used the pattern of like like-homozygous trees than like-heterozygous trees. associations became consistent with those of the first An average of 83 per cent of all homozygous like-tree populations mentioned above (e.g. Fig. 3b). That is pairs were significantly positively autocorrelated at 10 there was significant positive autocorrelation at m. This value ranged between 76 per cent and 92 per approximately 10 m, nonsignificant at 15 m and cent for different enzymes. Like-heterozygote tree- negative by 20 m. Unlike trees were significantly nega- pairs were on average significantly positively asso- tively autocorrelated at all distances. This pattern was ciated in 52 per cent of cases with a range from 46 per consistent among the enzymes studied. cent to 61 per cent for different enzymes (Table 2) but Positive autocorrelation at short distances represents there were fewer heterozygote trees in the populations. clusters of trees with like genotypes (Legendre & By 20 m the results between pairs of like trees were Fortin, 1989). Negative autocorrelation for short nonsignificant in most cases, 78 per cent overall (Table distances can reflect either an avoidance phenomenon 3). Unlike pairs of trees were generally significantly such as found among regularly spaced plants, or the negatively autocorrelated at both 10 (72 per cent) and sampling interval may be too large to detect associa- 20 m (84 per cent) distance annuli, or else were not tions (Legendre & Fortin, 1989). significantly autocorrelated (Tables 2 and 3). The first correlogram pattern described above (e.g. Correlograms were used to visualize the patterns of Figs 1 and 2), is the most common autocorrelation associations with increasing distances between trees. pattern found in plants (Epperson, 1989; Legendre & The resulting patterns were consistent for either like or Fortin, 1989). It is typical of a pattern found in popula- unlike genotype-pair combinations at each site, as can tions with relatively greater proportions of homo- be seen in the example at Newall Creek (Fig. 1). There- zygotes and poor dispersal (Epperson, 1989). The fore, the standard normal deviate (SND) was averaged pattern usually represents aggregates of homozygotes, between all like or unlike pairs for each enzyme. with zones of heterozygotes intergrading between 32 A. SHAPCOTT

Table 2 Summary over all sites studied of spatial autocorrelation results for Atherosperma moschatum. The summation of pairwise associations between trees of different genotype combinations within a 10 m radius is given. The results of the significance test of each genotypic-pair combination are pooled over all study sites and given as a percentage of the total number of each type of pair combination

SND SND SND Enzyme Pair-type % sig.+ye % nonsig. % sig.—ye

PER Total like 67 33 Homozygote like 77 23 Heterozygote like 47 53 Total unlike 23 77 PGI-2 Total like 83 17 Homozygote like 92 8 Heterozygote like 61 39 Total unlike 20 80 PGI-3 Total like 72 28 Homozygote like 82 18 Heterozygote like 47 53 Total unlike 31 69 SDH Total like 74 26 Homozygote like 78 22 Heterozygote like 60 40 Total unlike 1 37 62 GDH Total like 75 25 Homozygote like 85 15 Heterozygote like 53 47 Total unlike 20 80 AAT Total like 70 30 Honiozygote like 85 15 Heterozygote like 46 54 Total unjike 35 65 Mean Total like 73 27 Homozygote like 83 17 Heterozygote like 52 48 Total unlike 28 72

SND: standard normal deviate.

them, so that unlike trees are negatively associated, ing structure and dispersal (Epperson & Clegg, 1986), especially unlike homozygotes, as they do not occur which may be affected by density (Antonovics & Levin, together (Epperson, 1989). The increasing negativity 1980). with distance reflects the reduced likelihood of Generally for the populations and enzymes studied encountering the same genotype with increasing there was a high level of consistency for the average distance (Epperson, 1989). patch length (approximately 15 m). However, one site, The point at which a correlogram crosses the X-axis Liffey Falls, had a slightly larger average patch length of (SND =0),or switches sign, is an operational estimate approximately 25 m. of the length of one side of the 'average patch'. In Two sites, Milkshake Reserve and Wyefield Rivulet irregular shaped or sized patches, it equals the average (e.g. Fig. 4), deviated from all others in their cone- length of the shortest side (Epperson & Clegg, 1986). logram patterns, in that like trees continued to be either Patch dimensions are primarily determined by breed- significantly positively autocorrelated at all distance GENETICS OF A THEROSPERMA MOSCHA TUM POPULATIONS 33

Table 2 Summary over all sites studied of spatial autocorrelation results for Table 3Summary of the significance of associations, tested (a) PER Like --SDH Like Atherosperma moschatum. The summation of pairwise associations between trees using spatial autocorrelation, between Atherosperma PER Unlike SDH Unlike of different genotype combinations within a 10 m radius is given. The results of the moschatum tree genotype pairs over all sites within a 10 to - PG 1-2 Like -GDH Like 20 m radius. The results of each genotypic-pair combination significance test of each genotypic-pair combination are pooled over all study sites 20 -•- P01-2 Unlike -'-GDH Unlike and given as a percentage of the total number of each type of pair combination are pooled over all study sites and given as a percentage of N. -. AAT Like -PG 1-3 Like the total number of each type of pair combination 16 Unlike -* 12 PGI-3 Unlike SND SND SND Enzyme Pair-type % sig.+ye% nonsig.% sig.—ye 4. 0• PER Total like 18 82 -4. — Total unlike 22 78 PGI-2 Total like 21 79 -12 PER Total unlike 5 95 -16 PGI-3 Total like 31 69 -20 Totalunlike 25 75 0 -24 z 10 SDH Total like 23 77 C/) C) Total unlike 16 84 D) (b) -u• PGI-3 Like CC ' SDH Like GDH Total like 27 73 C) 15 ' GDH Like Total unlike 3 97 -- PER Like AAT Total like 8 92 10 —R PGI-2 Like Total unlike 25 75 Mean Total like 22 78 Total unlike 16 84

SND: standard normal deviate.

classes or randomly distributed (nonsignificant auto- correlation) at distances greater than 10 m. This 10 15 20 25 30 pattern was consistent at both scales of distance incre- ment investigated. Positive associations at larger Distance (m) distances are unusual (Epperson, 1989), and usually associated with repeated or regular patterns in the Fig.2 Correlograms showing the relationships between pairs of Atherosperma moschatum trees of either like or environment (Legendre & Fortin, 1989). They repre- unlike genotypes with increasing distance apart at Tarra sent the distance between aggregates of the same geno- Bulga. Like-genotype pair and unlike-genotype pair signifi- type. A significant autocorrelation at 10 m followed by cant normal deviate (SND) scores were averaged for each nonsignificant results at greater distances (e.g. Fig. 3a), enzyme locus and used for these plots. (a) Ten metre annulus would be consistent with short-distance clustering of distances. (b) Five metre annulus distances; only like-geno- like genotypes in an otherwise random spatial arrange- type average SND values are plotted. ing structure and dispersal (Epperson & Clegg, 1986), ment. The unusual pattern found at these two sites may especially unlike homozygotes,which may be affectedas they by density do (Antonovics not occur & Levin, be most easily explained by the existence of smaller together (Epperson, 1989). The increasing negativity patches which are repeated within the distance classes with distance reflects theGenerally reduced for the populations likelihood and enzymes studied of scored, or patches randomly distributed in populations tion between stand density or stand size structure with there was a high level of consistency for the average with some genotypes' patches occurring more the size of the patches measured at these sites. frequently. At these sites like-heterozygous genotypes Liffey Falls, had a slightly larger average patch length of were strongly positively correlated, so that the small Discussion clumps may suggest small clusters of vegetative clones, of the length of one side Twoof sites,the Milkshake 'average Reserve and patch'. Wyefield Rivulet In or offspring under parent trees. Distances between Positiveautocorrelation of plants with the same geno- (e.g. Fig. 4), deviated from all others in their cone- such clumps would be of the order of 5 to 10 m. At type at short distances is predicted where reproduction logram patterns, in that like trees continued to be either these sites the average distances between trees were 3.8 occurs vegetatively (Sokal & Oden, 1978b; Oden, significantly positively autocorrelated at all distance m and 1.9 m (Table 1). There was no apparent associa- 1984). In such cases clusters of clones would result. 34 A. SHAPCOTT

-u PGI-3 Like -W GDH Unlike PG 1-3 Like -R GDH Like (a) (a) -4- PG 1-3 Unlike -- PER Like PGI-3 Unlike U- GDH Unlike -U- SDH Like -- PER Unlike SDH Like PER Like -- SDH Unlike •* PG 1-2 Like -4- SDH Unlike PER Unlike -. GDH Like —I- PGI-2 Unlike

z 10 20 30 10 20 30 C/) ci) Cl) C) SDH Like cci (b) a) -4- PG 1-3 Like - PG 1-3 Like — PG 1-2 Like (b) 4- SDH Like 4 PER Like U- PER Like -. GDH Like 4- GDH Like

10 15 20 25 30 -3 Distance (m) 10 15 20 25 30 Distance (m) Fig.3 Correlograms showing the relationships between pairs of Atherosperma moschatum trees of either like or Fig. 4 Correlograms showing the relationships between unlike genotypes with increasing distance apart at Meetus pairs of Atherosperma moschatum trees of either like or Falls. Like-genotype pair and unlike-genotype pair signifi- unlike genotypes with increasing distance apart at Milkshake cant normal deviate (SND) scores were averaged for each Reserve. Like-genotype pair and unlike-genotype pair signi- enzyme locus and used for these plots. (a) Correlogram with ficant normal deviate (SND) scores were averaged for each a 10 m increasing annulus distance. (b) Correlogram with a 5 enzyme locus and used for these plots. (a) Correlogram with m increasing annulus distance; only like-genotype average a 10 m increasing annulus distances. (b) Correlogram with a SND values are plotted. 5 m increasing annulus distance; only like-genotype average SND values are plotted.

The distribution of genotypes among such clusters, however, could be random. Heterozygotes would be Short-distance positive autocorrelation is also found fewer, but heterozygote and homozygote clusters where family clusters have developed (Epperson, would not be expected to be structured differently, 1989; Legendre & Fortin, 1989). Modelling studies unless heterozygotes did not reproduce vegetatively. have shown that in populations with limited gene flow, GENETICS OF A THEROSPERMA MOSCHA TOM POPULATIONS 35

generally inbred (Shapcott, 1994). Therefore it may be expected that family clusters produced by localized dispersal and vegetative spread would arise (Sokal & :: Wartenberg, 1983; Epperson, 1989). Thus, the results of this study are consistent with expectations for this species. As both of these two mechanisms for pro- ducing the small-scale genetic structure described 70 above (i.e. local seed and pollen dispersal and vegeta- tive reproduction), are likely to occur in A. moschatum 0 stands, either or both may be responsible for the sub- 60 - structuring observed in these stands. This is not easily clarified in the field as it is often difficult to distinguish between stems that have arisen vegetatively from those which have arisen from seedlings. This is exacerbated 5°O by the fact that many of the sites likely to produce 4. 0 vegetative shoots are also the places most likely to 0 provide 'safe' germination sites. Similar observations 40 - have been made for other species (e.g. Schnabel et a!., 0 0 1991). Like-heterozygote genotype pairs of trees were not 0 significantly clumped as often as like-homozygote trees 30- in the shortest distance class (0—10 m) (Table 2). There is no evidence to suggest that homozygotes are more likely to reproduce vegetatively than heterozygotes. 20 However, heterozygotes were generally infrequent Qo (Shapcott, 1994), and therefore there may not have been enough like heterozygote pairs to produce signifi- cant autocorrelations. Where there were more hetero- I0 zygotes present in a population and significant positive autocorrelations were found, they may have resulted from vegetative reproduction enlarging a zone of 0, O heterozygotes between two homozygous family 0 10 20 30 clusters. Metres The size of patches (—15m diameter) as deter- Fig.5 Map of the transect of Atherosperma moschatum mined by the correlograms (Figs 2b and 3b) was gener- trees at Tahune showing the distribution of GDH genotypes ally consistent between sites and seems larger than amongst the trees. The tree symbol sizes used are propor- would be expected if such structure was solely due to tional to eight times the actual diameters. (Genotypes: 0 BB, vegetative reproduction. When genotypic site maps 3 Correlograms showing the relationships between 8'AA,.AB). were studied, it was found that dense clumps of stems pairs of Atherosperma moschatumFig. 4 Correlograms trees showing of theeither relationships like between or unlike genotypes with increasingpairs of Atherosperma distance moschatum apart trees at of Meetus either like or were often of mixed genotypes, but that scattered trees unlike genotypes with increasing distance apart at Milkshake were often grouped by genotype (e.g. Fig. 5). Similar cant normal deviate (SND) scoresReserve. Like-genotype were pairaveraged and unlike-genotype for paireach signi- such as in insect-pollinated species or where most seed results were found by Schnabel et a!. (1991) and they enzyme locus and used for theseficant normal plots. deviate (a) (SND) Correlogram scores were averaged withfor each falls beneath the parent tree, family aggregates quickly support the hypothesis that clumps of stems are as enzyme locus and used for these plots. (a) Correlogram with develop (Ennos & Clegg, 1982). This leads to clusters likely to be produced by seedling germination as by m increasing annulus distance;a 10 m only increasing like-genotype annulus distances. (b) Correlogram average with a of homozygotes separated by heterozygous interme- vegetative growth. Since the size of patches was gener- 5 m increasing annulus distance; only like-genotype average diate zones (Turner et a!., 1982; Epperson, 1989). It ally consistent between sites it is unlikely to be gener- has also been shown that uncommon long-distance ated by spatially varying selection. There was no dispersal events have little effect on this family struc- consistent observable microhabitat heterogeneity ture development (Epperson, 1989; Schnabel et a!., within sites to have caused such small-scale selection. The distribution of genotypes among such clusters, 1991). Some weak positive autocorrelation of hetero- Other authors have interpreted such consistency of however, could be random.Short-distance Heterozygotes positive autocorrelation would is also found be zygotes may result in the intermediate zones of hetero- localized structure as generated by short-distance gene fewer, but heterozygotewhere and family homozygote clusters have developed clusters (Epperson, zygotes (Epperson, 1989). dispersal (Epperson & Clegg, 1986; Schoen & Latta, 1989; Legendre & Fortin, 1989). Modelling studies A. moschatum is known to reproduce vegetatively, is 1989). In the present case it may be a combination of a have shown that in populations with limited gene flow, insect-pollinated and populations were found to be consistency in vegetative spread, the insect pollinator 36 A. SHAPCOTT

flight path characteristic, and the size of the area over bourhoods' of interbreeding, related individuals (Brad- which most seed fall from the parent tree. shaw, 1972; Levin & Kerster, 1974). However, the The populations that showed a pattern of short- scale of structure here is fine relative to the size of distance clumping and then nonsignificant or random trees, therefore indicating small neighbourhoods. dispersion of genotypes, when tested with 10 m annuli, These family groups are likely to be randomly dis- were all sampled by smaller transects. It appears, from persed through the population in this case. However, site maps, that these transects contain one or two major the negative autocorrelations suggest that clumps in clumps. Other trees may represent the edges of several most populations tend to be nonrandomly dispersed other patches, giving a random distribution of geno- rather than aggregated, i.e. clumps tend to be next to types outside one patch. Thus the smaller annulus clumps of a different type rather than the same type. In distances may have clarified the pattern. Significant contrast, the patterns at Milkshake Reserve and positive autocorrelations at greater distances occur Wyefield Rivulet stands suggest that clumps are either where another clump of the same genotype is encoun- randomly distributed or more likely to be aggregated tered and may also indicate smaller patch sizes in that with other like clumps than with unlike clumps. This site (Legendre & Fortin, 1989). Where stands are sort of small-scale structure has been found in several dominated by only a few genotypes, the likelihood of other studies (Epperson & Clegg, 1986; Perry & this occurring is increased. At Milkshake Reserve and Knowles, 1991; Schnabel et al., 1991). It is difficult to Wyefield Rivulet stands, this pattern was found using compare the scale of pattern between studies since one both a 10 m and a 5 m annulus, and thus failed to find would expect the scale of pattern to be relevant to plant an anticipated smaller patch length. Some genotypes size. However, in other studies of trees, patch sizes of appeared to be randomly dispersed at greater similar dimensions have been reported (Perry & distances, some wavered between significant positive Knowles, 1991; Schnabel eta!., 1991). Studies in which correlations and others remained positive. The results plants are sampled on a regular grid (e.g. Epperson & may suggest many very small clumps either randomly Allard, 1989), may miss such fine-scale patterns if the distributed or tending to be aggregated. As these sampling step is too large. Fortin et a!. (1989) demon- appear to be small they may simply represent aggre- strated that irregular sampling is more efficient for gates of clones or offspring from a parent tree. Both pattern detection than a regular lattice, since a range of stands were of very different physical structure, but neighbour distances is included. both had very clumped distributions of trees. However, The results of a study of Acer saccharum stands of the results also suggest that a larger scale of clustering mixed age (Perry & Knowles, 1991) were similar to this may be developing. study. They suggested that although Acer saccharum Since there was generally no difference between seed can be dispersed for great distances, seed disper- genotypes in patch size, although genotypes were sal in dense forests is restricted. A similar scenario is present in different frequencies within populations, suggested for A. moschatum, as Read & Hill (1988) there was therefore no strong evidence for selection of have suggested that in dense rainforest most seed fall one particular genotype over others within the sites. near the parent tree, even though A. moschatum seed is There was no evidence of differential selection plumed and can disperse great distances by wind. Perry between the different enzymes studied, since the results & Knowles (1991) suggest that stand regeneration, were consistent for all enzymes at each site. This which is primarily by gap filling, may have generated suggests that the proportions of alleles or genotypes genotypic structure as a result of clustered cohorts of present at a site may be largely due to chance or histor- similar genotypes with the resulting genotypes being ical factors, and the enzymes studied are effectively structured by age. Since A. moschatum stands were neutral (Epperson, 1989; Schoen & Latta, 1989; shown to exhibit a variety of stand size structures and Schnabel et al., 1991). This is consistent with findings regeneration patterns (Shapcott, 1994), yet had con- from other studies (Epperson, 1989; Knowles et al., sistent genetic spatial structures, genetically differen- 1992). tiated regenerating cohorts seem an unlikely cause of The populations studied have a diversity of popula- genetic structure in this case. tion sizes, degrees of isolation, stand size structures It takes several generations for family structure to and stand histories (Shapcott, 1994), and densities develop (Turner et a!., 1982; Sokal & Wartenberg, (Table 1). However, there was a high degree of con- 1983; Epperson, 1989). Knowles et a!. (1992) inter- sistency of both the presence and scale of genetic struc- preted the differences in presence of genetic structure ture among the stands of A. moschatum studied. These in populations of Larix laricina as reflecting differences results are consistent with hypotheses that plant in stand origin. They postulated that structure had populations are subdivided into local demes or 'neigh- developed where a few remaining trees (bottleneck) GENETICS OF A THEROSPERMA MOSCHA TUM POPULATIONS 37 formed the basis for the present population, whereas conserving one part of a large population will reflect or the stand originating from heterogeneous outside seed adequately sample the genetic variability contained in sources had not yet developed a family structure. the whole, since local differentiation or drift has Coates (1992), however, found no substructuring in occurred in A. moschatum populations. populations of Stylidium coroniforme recovering from severe bottlenecks. As the A. moschaturn stands are likely to have different histories, the results suggest that Acknowledgements all have been stable long enough for some family sub- Ithank the following people for their assistance: M. structuring to have developed. A. moschatum popula- Zalucki, B. Potts, J. Reid, J. Kirkpatrick, M. Brown, S. tions are likely to have experienced size reduction, or Harris, M. Battaglia, M. Haseler, R. Wiltshire and G. bottlenecks, during past glacial cycles and these are Moran. This work was jointly funded under the likely to have facilitated substructuring in some National Rainforest Conservation Program by the populations. To the same extent, populations arising Commonwealth Government and the Department of from founders in more isolated populations, in the Parks Wildlife and Heritage, Tasmania. absence of extensive seed rain, may have facilitated substructuring of genetic composition in those stands. For example, some isolated east coast populations are References thought to be recolonized periodically after fire from ANTONOVICS, J.ANDLEVIN, D. A. 1980. The ecological and incoming seed rather than from a surviving seed pool genetic consequences of density dependent regulation in or surviving trees (Neyland, 1991). Given the longevity plants. Ann. Rev. Ecol. Syst., 11,411—452. of the species (trees may reach two hundred plus ARGYRES, A. Z., AND SCHMV[, .1991.Microgeographic genetic years), one could predict that all the stands studied structure of morphological and life history traits in a could be at least several hundred years old. natural population of Impatiens capensis. Evolution, 45, Therefore, these results provide strong evidence 178—189. that within A. moschatum stands family clusters of BRADSHAW, A. D. 1972. Some of the evolutionary consequences of being a plant. Evol. Biol., 5,25—47. related genotypes are formed due to vegetative repro- BRADSHAW,A. a 1984. Ecological significance of genetic varia- duction and localized pollen and seed dispersal. This tion between populations. In: Dirzo, R. & Sarukhán, J. would result in inbreeding within clusters as recorded (eds) Perspectives on Plant Population Ecology, pp. in Shapcott (1994). Within stands there may be high 213—228. Sinauer Associates, Sunderland, MA. allelic diversity, but little mixing of genotypes due to BROWN, A. H. D. 1978. Isozymes, plant population genetic substructuring, and hence leading to the fixation of structure and genetic conservation. Theor. App!. Gen., 52, genotypic proportions. One could predict that through 145—157. this mechanism localized fixation and differentiation of COATES, n. i. 1992. Genetic consequences of a bottleneck and allelic frequencies could develop within larger popula- spatial genetic structure in the triggerplant Stylidium tion ensembles without the need for selection as a coroniforme (Stylidiaceae). Heredity, 69, 512—520. DEWEY,S. E. AND HEYWOOD, .s.1988. Spatial genetic structure driving force. This was found to have occurred in a population of Psychotria nervosa. I. Distribution of between the Frodshams and Creepy Crawly sites, genotypes. Evolution, 42, 834—838. which are only approximately 2 km apart, and are ENDLER, S. A. 1977. Geographic Variation, Speciation and within a larger semicontinuous ensemble (Shapcott, Clines. Princeton University Press, Princeton, N.J. 1994). These results suggest that genotype distribu- ENNOS, R. A. AND CLEGG, M. T. 1982. Effect of population sub- tions and frequencies within stands may largely reflect structuring on estimates of outcrossing rate in plant popu- chance and historical occurrences. It is likely therefore lations. Heredity, 48, 283—292. that much variation in allelic proportions between EPPERSON, B. K. 1989. Spatial patterns of genetic variation many A. rnoschatum stands may be due to random within plant populations. In: Brown, A. H. D., Clegg, M. events and drift. T., Kahler, A. L., and Weir, B. S. (eds) Plant Population The results have implications for conservation Genetics, Breeding and Genetic Resources, pp. 229—253. Sinauer Associates, MA. management. These include the acknowledgement that EPPERSON, B. K., AND ALLARD, R. w. 1989. Spatial autocorrela- if there are small neighbourhood sizes in A. rnoscha- tion analysis of the distribution of genotypes within popu- turn stands, the stands should not be considered too lations of lodgepole pine. Genetics, 121, 369—377. small to be worth conserving, since they mostly inter- EPPERSON, B. K., AND CLEGG, M. T. 1986. Spatial-autocorrelation breed on a local scale, and even the smallest stands in analysis of flower color polymorphisms within substruc- this study have been relatively stable. Furthermore, tured populations of morning glory (Ipomoea purpurea). large semicontinuous populations are unlikely to be Am. Nat., 128, 840—858. homogeneous. Therefore, it cannot be assumed that FORTIN, M. 3.,DRAPEUA,P. ANDLEGENDRE, i.1989. Spatial auto- 38 A. SHAPCOTT

correlation and sampling design in plant ecology. and the implications for the conservation of its biodiver- Vegetatio, 83, 209—222. sity. Aust. J. Bot. (in press). KNOWLES, P., PERRY, D. J. AND FOSTER, H. A. 1992. Spatial genetic SOKAL, R. R. AND ODEN, N. L. 1978a. Spatial autocorrelation in structure in two Tamarack [Larix laricina (Du Roi) K. biology. 1. Methodology. Biol. J. Linn. Soc., 10, 199—228. Kochi populations with differing establishment histories. SOKAL, R. R. AND ODEN, N. L. 1978b. Spatial autocorrelation in Evolution, 46, 572—576. biology. 2. Some biological implications and four applica- LEGENDRE, P. AND FORTIN, M. i. 1989. Spatial pattern and tions of evolutionary and ecological interest. Biol. J. Linn. ecological analysis. Vegetatio, 80, 107—138. Soc., 10, 229—249. LEVIN,D. A. AND KERSTER, H. w. 1984. Gene flow in seed plants. SOKAL R. R. AND WARTENBERG. D. n. 1983. A test of spatial auto- Evol. BioL, 7, 139—200. correlation analysis using an isolation-by-distance model. LEWONTIN, R. c. 1974. The Genetic Basic of Evolutionary Genetics, 105, 2 19—237. Change. Columbia University Press, New York. TURNER, M. E., STEVENS, J. C. AND ANDERSON, W. w. 1982. Homo- MAYR, E. 1970. Populations, Species, and Evolution. Harvard zygosity and patch structure in plant populations as a University Press, Cambridge, MA. result of nearest-neighbor pollination. Proc. Nail. Acad. NEYLAND, M. a 1991. Relict Rainforest in Eastern Tasmania. Sci. U.S.A., 79, 203—207. Tasmanian NRCP Technical Report No.6. Department of WAGNER, D. B., SUN, Z.-X., GOVIDARJU, D. R. AND DANCIK, B. P. 1991. Parks Wildlife and Heritage, Tasmania and Department of Spatial patterns of chloroplast DNA and cone morpho- the Arts, Sport, the Environment, Tourism and Terri- logy variationwithin populations of a Pinus tories, Canberra. banksiana—Pinus contorta sympatric region. Am. Nat., ODEN, N. L. 1984. Assessing the significance of a spatial 138,156—170. correlogram. Geogr. Analysis, 16, 1—16. WASER, N. M. 1987. Spatial genetic heterogeneity in a popula- PERRY, D. J. AND KNOWLES, P. 1991. Spatial genetic structure tion of the montane perennial plant Delphinium nelsonii. within three sugar maple (Acer saccharum Marsh.) stands. Heredity, 58, 249—256. Heredity, 66, 137—142. WRIGHT, S. 1965. The interpretation of population structure READ, 1. AND HILL, R. s. 1988. The dynamics of some rainforest by F-statistics with special regard to systems of mating. associations in Tasmania. J. EcoL, 76, 558—584. Evolution, 19, 395—420. SCHNABEL, A., LAUSHMAN, R. H. AND HAMRICK, J. L. 1991. WRIGHT, 5. 1978. Evolution and the Genetics of Populations Comparative genetic structure of two co—occurring tree Vol. 4, Variability Within and Among Natural Populations. species, Maclura pomifera (Moraceae) and Gleditsia University of Chicago Press, Chicago. triacanthos (Leguminosae). Heredity, 67, 357—364. XIE, C. Y. AND KNOWLES, i'. 1991. Spatial genetic substructure SCHOEN, D. j.ANDLATFA, R. CL 1989. Spatial autocorrelation of within natural populations of jack pines (Pinus banksiana). genotypes in populations of impatiens pallida and Can. J.Bot., 69, 547—55 1. Impatiens capensis. Heredity, 63, 181—189. ZALUCKI, M. P., HUGHES, J. M. AND CARTER, P. A. 1987. Genetic sHAPcorr, A. 1994. The population genetics and ecology of variation in Danau.s plexippus L.: Habitat selection or the temperate rainforest tree Atherosperma moschatum differences in activity times? Heredity, 59, 213—221.