O R I G I NA L A RT I C L E doi:10.1111/j.1558-5646.2011.01313.x POPULATION DIVERGENCE ALONG LINES OF GENETIC VARIANCE AND COVARIANCE IN THE INVASIVE PLANT LYTHRUM SALICARIA IN EASTERN NORTH AMERICA Robert I. Colautti1,2,3 and Spencer C. H. Barrett1 1 Department of Ecology & Evolutionary Biology, University of Toronto, 25 Willcocks St. Toronto, Ontario M5S 3B2, Canada 2 E-mail: [email protected] Received September 28, 2010 Accepted March 28, 2011 Evolution during biological invasion may occur over contemporary timescales, but the rate of evolutionary change may be inhibited by a lack of standing genetic variation for ecologically relevant traits and by fitness trade-offs among them. The extent to which these genetic constraints limit the evolution of local adaptation during biological invasion has rarely been examined. To investigate genetic constraints on life-history traits, we measured standing genetic variance and covariance in 20 populations of the invasive plant purple loosestrife (Lythrum salicaria) sampled along a latitudinal climatic gradient in eastern North America and grown under uniform conditions in a glasshouse. Genetic variances within and among populations were significant for all traits; however, strong intercorrelations among measurements of seedling growth rate, time to reproductive maturity and adult size suggested that fitness trade-offs have constrained population divergence. Evidence to support this hypothesis was obtained from the genetic variance–covariance matrix (G) and the matrix of (co)variance among population means (D), which were 79.8% (95% C.I. 77.7–82.9%) similar. These results suggest that population divergence during invasive spread of L. salicaria in eastern North America has been constrained by strong genetic correlations among life-history traits, despite large amounts of standing genetic variation for individual traits. KEY WORDS: Fitness trade-off, G matrix, purple loosestrife, quantitative genetics. The invasive spread of introduced species usually occurs across heterogeneous landscapes and often along large-scale environmental gradients that correlate with latitude. Introduced species may be aided in their spread in heterogeneous environments by phenotypic plasticity, including general purpose genoypes (Baker 1965; Williams et al. 1995; Parker et al. 2003; Ross et al. 2009). However, if plasticity is costly or poorly tracks environmental conditions then local adaptation may evolve (Via and Lande 1985; Scheiner 1993; Tufto 2000), potentially increasing invasive spread 3 Current Address: Department of Biology, Duke University, PO Box 90338, Durham, North Carolina 27708. C 2514 (Garcı́a-Ramos and Rodriguez 2002). The extent to which local adaptation increases fitness in invasive species confronting novel environments will depend on the availability of standing genetic variation within populations for ecologically relevant traits (Fisher 1930; Lande 1979; Lee 2002; Lee et al. 2007). Evidence from neutral genetic markers indicates that bottlenecks during introduction are common in invasive species (e.g., Barrett and Shore 1989; Barrett and Husband 1990; Malacrida et al. 1998; Tsutsui et al. 2000; Novak and Mack 2005; Zhang et al. 2010). Nevertheless, genetic variation within introduced populations can sometimes be greater than that occurs in native source regions as a result of multiple introductions of C 2011 The Society for the Study of Evolution. 2011 The Author(s). Evolution Evolution 65-9: 2514–2529 G E N E T I C C O N S T R A I N T S O N P O P U L AT I O N D I V E R G E N C E differentiated populations followed by admixture among them (reviewed in Lee et al. 2004; Dlugosch and Parker 2008a; see also Keller and Taylor 2010; Verhoeven et al. 2011). However, genetic variation inferred from neutral markers poorly predicts genetic variation for quantitative traits (Lande 1988; Merilä and Crnokrak 2001; McKay and Latta 2002) and, consequently, determining the potential for adaptive evolution during the geographical spread of invasive species requires measurements of the standing genetic variation of ecologically relevant traits sampled from across the range. The occurrence of genetic variation for quantitative traits within populations may not guarantee contemporary evolution in introduced species because fitness trade-offs have the potential to impede local adaptation. Trade-offs among two or more lifehistory traits are often manifested as positive or negative genetic correlations that prevent natural selection from simultaneously improving correlated traits (Dickerson 1955; Lande 1982; Blows et al. 2004; Roff and Fairbairn 2007). For example, the classic life-history trade-off between age and size at maturity constrains individuals to reproduce early at a small size, or later at a large size. In such cases, population divergence will be constrained along the axis of covariance between age and size (Mitchell-Olds 1996; Schluter 1996). Fitness trade-offs can therefore serve to constrain the direction of contemporary adaptive evolution and may influence survival and reproduction at range limits (Etterson and Shaw 2001; Etterson 2004) with consequences for the geographical spread of invasive species. Understanding genetic constraints on population divergence requires the simultaneous comparison of correlations among multiple traits. The G matrix is a convenient mathematical representation of the genetic variances (diagonal cells) for a set of quantitative traits and the covariances (off-diagonal cells) among them (Lande 1979). However, the use of the G matrix to directly infer constraints imposed by genetic trade-offs presents two experimental challenges. First, variance and covariance components of G (hereafter “(co)variances”) are estimated from genetic lines or families, making empirical estimates imprecise without large sample sizes (Shaw 1991). Second, in addition to fitness trade-offs, the covariance structure of G is also shaped by natural selection, genetic drift, and migration (e.g., Turelli 1988; Phillips et al. 2001; Jones et al. 2003; Guillaume and Whitlock 2007). These processes are likely to be important during rapid range expansion by invasive species, which often includes founder events and changes in the selective landscape. As a result, the G matrix of a randomly chosen population is likely to be a poor predictor of constraints on adaptive evolution across the range of a species. Therefore, both methodological (e.g., sample size) and biological (e.g., stochastic forces) factors complicate the identification of genetic constraints on the divergence of natural populations. An alternative approach for identifying genetic correlations that may constrain population divergence is to investigate tradeoffs that are expected a priori to be of likely ecological importance (e.g., age vs. size at reproduction) and then estimate G for these traits averaged across a sample of populations. Here, the idiosyncratic effects of selection, migration, and drift in any one population should be “averaged out” (see Chenoweth and Blows 2008). For example, bottlenecks significantly changed the (co)variance structure of G in experimental populations of Drosophila melanogaster, yet the average G across populations remained similar to the ancestral outbred population (Phillips et al. 2001). Additional insight can also be obtained by investigating the primary eigenvectors (i.e., principal components) of G because these are likely to remain more stable over time than are particular (co)variance estimates (reviewed in Arnold et al. 2008). If the principal component eigenvectors of the “average” G are relatively stable over time, then populations should diverge in a predictable fashion, resulting in a correlation between the eigenvectors of G and those of the matrix of (co)variance among population means (D). The D matrix characterizes population divergence and is similar to G, but instead represents variances and covariances among population means, rather than genetic families within populations. If divergence among populations is completely constrained by genetic variance within and covariance among life-history traits, then G and D should be identical. Conversely, evolution that is not constrained by genetic (co)variances within populations will result in less similarity between G and D. Methods for comparing G and D among natural populations is an area of active research (Shaw 1991; Phillips and Arnold 1999; Roff et al. 1999; McGuigan et al. 2005; Chenoweth and Blows 2008; Calsbeek and Goodnight 2009; Simonsen and Stinchcombe 2010), but to our knowledge none of these approaches has yet been used to study adaptive evolution during biological invasion. Invasive species may be well-suited for studying genetic constraints on the evolution of natural populations because multiple introductions and admixture are likely to weaken phylogeographical relationships and the extent of population structure that can complicate comparisons of G and D. We investigated genetic constraints associated with lifehistory trade-offs within and among invasive populations of the wetland plant Lythrum salicaria L. (Lythraceae) in eastern North America. Previous work on L. salicaria identified genetically based latitudinal clines for days to first flower and several measurements of plant size in both native European (Olsson and Ågren 2002; Olsson 2004) and introduced North American populations (Montague et al. 2008; Colautti et al. 2010). Moreover, measurements of phenotypic selection indicate that introduced populations of L. salicaria are under strong selection for earlier flowering time and larger size (O’Neil 1997; Colautti and Barrett 2010). A study of populations from eastern North America indicates that EVOLUTION SEPTEMBER 2011 2515 R . I . C O L AU T T I A N D S . C . H . BA R R E T T evolution of earlier flowering in northern populations is associated with a correlated change in vegetative size (Colautti et al. 2010). However, the extent to which multitrait genetic constraints may limit population divergence was not previously investigated and this is one of the main goals of this study. Here, we evaluate constraints on population divergence during biological invasion in L. salicaria by measuring the standing genetic variance and covariance of 12 ecologically relevant lifehistory traits. This was undertaken by sampling 20 populations along a latitudinal gradient of 10 degrees latitude from southern Maryland (U.S.A.) to Timmins, Ontario (Canada) and conducting a quantitative genetic experiment under uniform growth conditions in a glasshouse. The specific questions we addressed in our study were: (1) Is there evidence for quantitative genetic variation within and among populations for life-history traits associated with growth and reproduction? The rapid reestablishment of latitudinal clines in North America (Montague et al. 2008) implicates the occurrence of significant standing genetic variation within populations; however, genetic drift and founder events are also known to play an important role in eroding diversity in introduced populations of this species (Eckert and Barrett 1992; Eckert et al 1996). (2) Are life-history traits correlated among populations, and among families within populations? Within populations, fitness trade-offs among traits such as seedling growth, time to maturity, and plant size should result in significant genetic correlations. Correlations among population means could arise from constraints on population divergence and/or selection for particular trait combinations. We predicted that selection for early flowering and larger size at all latitudes should result in weaker correlations between these traits relative to within-population genetic correlations. (3) To what extent is any correlated divergence among population means (i.e., the D matrix) constrained by the structure of genetic correlations among families within populations (i.e., the G matrix)? Constraints on adaptive evolution resulting from trade-offs among life-history traits should result in a strong correlation between D and G, whereas divergence resulting from selection for larger plants that flower sooner should not. Methods STUDY SYSTEM AND SAMPLING Lythrum salicaria (purple loosestrife) is an insect-pollinated, outcrossing, autotetraploid, perennial herb native to Eurasia. During the last century it has become a successful invader of North American wetland habitats, roadside ditches and other moist disturbed sites, particularly in eastern North America (Thompson et al. 1987; Mal et al. 1992; Blossey et al. 2001; USDA 2009). Herbarium records indicate initial introduction to ports along the eastern seaboard at the end of the 18th century, with more recent 2516 EVOLUTION SEPTEMBER 2011 (< 70 years) spread into central and northern Ontario, Canada (Thompson et al. 1987). Both herbarium records and studies using molecular markers suggest multiple introductions to North America with admixture among introduced populations (Thompson et al. 1987; Houghton-Thompson et al. 2005; Chun et al. 2009). Lythrum salicaria is tristylous and self-incompatible with successful mating largely between plants of different style morphs. All shoot growth is localized within a few centimeters of the primary stem and therefore population growth occurs exclusively through seed production (Yakimowski et al. 2005). We used populations previously studied by Montague et al. (2008) in which latitudinal clines for several life-history traits were detected; details of population sampling and seed collections used in this experiment are given therein. Briefly, in autumn 2003, open-pollinated infructescences were collected from 25 populations in eastern North America. Populations were chosen to represent a latitudinal cline from Timmins, Ontario (48◦ 48 N, 81o 30 W) to Easton, Maryland (38o 75 N, 75o 99 W). The latitude of these populations is a strong predictor of season length (see Montague et al. 2008). From the initial 25 populations, we chose a subsample of 20 representing the entire latitudinal gradient. EXPERIMENTAL DESIGN AND MEASUREMENTS In June 2004, we planted 20 seeds from 20 families from each of 20 populations (8000 total) into 2 cm × 2 cm plug trays filled with Pro-MixTM “BX” peat containing vermiculite and perlite. The position of each seed family was randomized across trays and trays were placed on a glasshouse bench at the University of Toronto and rotated thrice weekly to reduce position effects. We monitored seeds daily for germination and after two weeks we randomly selected eight seedlings from each of 17 families, which we transplanted into 10 cm pots filled with Pro-MixTM . The experiment involved a randomized block design with two individuals per family per population per block (N = 4 blocks). The final dataset contained 2623 (of 2720) individuals from 339 (of 340) seed families due to mortality. All blocks received 1.5 g/L of fertilizer (ratio 20N:20P:20K) every three weeks and were kept partially flooded so that the lower one-fourth of each pot was in water. To prevent aphid outbreaks, we treated plants with Dursban 2E (chlorpyrifos) pesticide at a rate of 4 mL/L, once in July and again in August. Following transplant, we measured seedling height and estimated total seedling leaf area as the diameter of the largest leaf multiplied by the largest total width of two-leaved seedlings. Two and four weeks after transplant we measured seedling height. Onset of flowering of plants occurred from July to September 2004 and during this period we monitored plants daily. On the first day of anthesis for a given plant, we measured stem width of the primary stem and the height of the primary stem, divided into two segments. The first, “vegetative size” was measured from G E N E T I C C O N S T R A I N T S O N P O P U L AT I O N D I V E R G E N C E the soil surface to the base of the inflorescence. The second, “inflorescence length” was measured from the base of the inflorescence to its tip. Division of the primary stem height into these two measures is a convenient division between resourceaccumulation structures (i.e., leaves) and resource sinks (i.e., flowers and fruits). Field surveys of natural populations, and a common garden experiment, confirm that the height of the vegetative portion of the primary stem is a strong predictor of total vegetative growth (see Colautti and Barrett 2010; Colautti et al. 2010). In October, we again measured vegetative height and inflorescence length; we then harvested and oven-dried plants to a constant weight to measure biomass. Similar to measurements of stem length, we divided shoot biomass into vegetative and inflorescence structures. We recorded seed germination day but found it had a negligible effect on seedling growth and time to maturity because germination date was highly skewed with no seed germinating before day 3 and 90% of seeds germinating between days 3 to 6. In total, our analysis included 12 traits: four involving seedling growth and development (seedling leaf area, seedling height, Height-2wk and Height-4wk), hereafter referred to as “seedling traits”; as well as four “adult traits” (days to first flower, stem width, vegetative size at flowering, and inflorescence length at flowering); and four “harvest traits” (vegetative size at harvest, inflorescence length at harvest, vegetative biomass, and reproductive biomass). separate 12 × 12 matrices using ReML: one from 339 seed families (G) and one from 2623 individuals (R). To investigate genetic variation within and among populations, we compared the fraction of total phenotypic variance manifest as (1) variance among population means (Vpop ), (2) variance among seed families within populations (Vfam ), and (3) variance among individuals within seed families (Vres ). These variance components represent: (1) genetic differentiation among populations (Vpop ), (2) ∼ 12 to 14 of the average standing genetic variation within populations (Vfam ), and (3) the residual phenotypic variance (Vres ) among individuals within a seed family. Variance among seed families represent 12 to 14 of standing genetic variation within populations because full- and half-siblings share on average 12 or 14 of their genes, respectively. Note that Vpop and 4 × Vfam are maximum estimates of the additive genetic variance because they also include di-, tri-, and tetragenic interactions, which are analogous to dominance variance in diploids but are much smaller in magnitude (Kempthorne 1955). Under random mating, with no epistasis or linkage disequilibrium, genetic covariance (σG ) between tetraploid half sibs x and y is: STATISTICAL ANALYSIS OF QUANTITATIVE TRAITS which are functions of the additive genetic variance (A) as well as interactions among two (D), three (T), or four (Q) alleles at each locus (Kempthorne 1955; Lynch and Walsh 1998). Differences in maternally inherited genes or in maternal provisioning to seeds can also inflate Vfam and Vpop . However, maternal provisioning is minimal as seeds of L. salicaria are very small (200 × 400 μm) and lack endosperm (Thompson et al. 1987). Moreover, we detected no significant maternal effects on adult size and flowering time in experimental crosses (see supplementary material in Colautti et al. 2010). To investigate whether there was significant genetic variation among and within populations, we ran a separate mixed model for each trait with populations and seed families, both as random effects. We tested whether the variance component for Vpop or Vfam were significantly different from zero, using a likelihood ratio test (LRT) with one degree of freedom. We treated population as a random effect in this analysis because it is a more conservative test of the null hypothesis that there is no significant variation among population means. We tested for latitudinal trends in each phenotypic trait using generalized linear models in R (version 2.8.1), with the standardized (BLUE) estimates of population means of each trait as Prior to analysis we log-transformed vegetative and inflorescence biomass to meet assumptions of multivariate normality and then standardized the phenotypic distributions of each trait to a mean of zero and a standard deviation of one. We performed a statistical linear mixed model using the MIXED procedure in SAS 9.1 (SAS Institute Inc., Cary, NC) with population as a fixed effect, seed family nested within population as a random effect, and repeated measurements on each individual nested within a seed family. This model estimates the “average” (co)variance matrix among seed family means within populations (i.e., the G matrix), and the residual (co)variance among individuals within a seed family (R matrix) by restricted maximum likelihood (ReML). Satterthwaite (1946) correction was used for the degrees of freedom of the fixed effects. We used best linear unbiased estimators (BLUEs) of population means from this model to calculate the matrix of divergence among population means (D matrix) because (co)variance components are not estimated directly for fixed effects in a mixed model. We used a Linux-based processor on the University of North Carolina’s research computing cluster to run the model (20 GB RAM; runtime ∼72 h). This was necessary due to the memory and processing power required to simultaneously estimate two σG (x, y) = σ2A /4 + σ2D /216 and for tetraploid full siblings is: σG (x, y) = σ2A /2 + 2σ2D /9 + σ2T /12 + σ2Q /36 EVOLUTION SEPTEMBER 2011 2517 R . I . C O L AU T T I A N D S . C . H . BA R R E T T dependent variables and latitude as the independent variable. We fitted quadratic regression terms to allow for nonlinear relationships with latitude and used an LRT to hierarchically test the significance of the quadratic and linear regression coefficients. The analysis comparing the similarity between G and D (see below) uses the first six principal component (PC) eigenvectors of these two matrices. To determine if all six PCs should be included in the analysis, we determined whether there was significant genetic variation in the sixth eigenvector of G using a LRT and a factor analytical structure for G. This was particularly important because several of the life-history traits we measured are likely to be correlated (e.g., height measured at different stages), which could result in nonsignificant genetic variation for higher PCs of G. This procedure allowed us to test the significance of a statistical model with six versus five principal components (see also McGuigan and Blows 2010). COMPARING G AND D The classic test for evolution along lines of genetic variance compares gmax and dmax , which are the first principal component eigenvectors of G and D, respectively (Mitchell-Olds 1996; Schluter 1996). The angle between gmax and dmax is calculated as θ = cos−1 [gmax ] [dmax ] where indicates a transpose of the gmax vector so that the two vectors can be multiplied (Schluter 1996). The angle θ ranges from 0 (complete similarity) to 90 (orthogonal, no similarity) and therefore provides a useful and intuitive measure of similarity (see Schluter 1996 for details). Although gmax by definition contains the most variation of any single eigenvector, the interpretation of θ as a measure of similarity between G and D ignores variance in other eigenvectors of G, which may be considerable and together may account for more phenotypic variance than gmax alone. In such cases, a more biologically informative approach is to compare multiple eigenvectors of G and D. Therefore, we used the Krzanowski (1979) method described in Blows et al. (2004; see also McGoey and Stinchcombe 2009), with the exception that we were interested in potential constraints on D (the matrix of variance–covariance among population means) instead of γ (the matrix of nonlinear selection gradients). If population divergence is constrained by G, then population divergence will occur along the major eigenvectors of G, resulting in an overall similarity between G and D. In contrast, population divergence that is not constrained by G will not constrain eigenvectors of D, resulting in little similarity between G and D. The Krzanowski method for comparing the similarity between two matrices begins with a separate principal components analysis (PCA) for each matrix. The variance–covariance matrices of G and D were standardized to the same scale (μ = 0, σ = 1) prior to the mixed model analysis and thus traits had the same total phenotypic variance but could differ in the relative amount 2518 EVOLUTION SEPTEMBER 2011 of variation among populations, among seed families, or within seed families. We then used the principal components of G and D to generate a “matrix of similarity” (S). S is calculated as A BB A, where A and B are matrices containing principal component eigenvectors of G and D, respectively, with each column representing a separate eigenvector. Here again the symbol indicates matrix transposition. Technically S describes the similarity of “subspaces” of G and D because only six of 12 eigenvectors (i.e., principal components) from each of G and D were included. However, this is a necessary limitation of the Krzanowski method because including more than half of the eigenvectors of G and D in A and B will constrain the comparison to recover angles of 0o (see Blows et al. 2004 for details). Therefore, we included only the first six eigenvectors of each matrix. Interpreting the matrix of similarity (S) in biological terms is difficult, but the principal component eigenvectors of S, denoted ai , have two useful properties. First, the sum of their eigenvalues represents the overall similarity between G and D, in this case ranging from 0 to 6. For ease of discussion, we refer to this sum as λi, or the “index of similarity.” Second, the eigenvalues (λi ) of each ai range from 0 (orthogonal, no similarity) to 1 (complete similarity), which can be translated into angles using √ the equation cos−1 λi (Blows et al. 2004). These angles can be compared directly with the angle between gmax and dmax (θ), as described above. Furthermore, Blows et al. (2004) show how each ai can be “projected” back into the subspace of G or D, to identify the phenotypic traits responsible for the similarity (or dissimilarity) between G and D. Here again “subspace” refers to the fact that this analysis uses the first six eigenvectors of G and D represented by each ai . As in Blows et al. (2004), we chose to project ai into the subspace G because we wanted to identify trait variances and covariances within populations that might constrain divergence among populations. The projection of ai into the subspace of G results in vectors (bi ), which we refer to as “vectors of similarity,” and are calculated as bi = Aai . Factor coefficients of the vectors of similarity (bi ) are biologically meaningful as they can be interpreted as coefficients describing the loadings of each phenotypic trait on bi . This is analogous to the coefficients of a PCA, which describe the relative “loadings” of each phenotypic trait on each principal component eigenvector. In this way, we could identify individual traits and trait combinations in G with variance–covariance structure most similar to D. We used a bootstrap method to generate 95% confidence intervals for the angle of similarity (θ) and the index of similarity (λi ). For each iteration of the bootstrap model (10,000 iterations total), 17 standardized BLUPs of seed family means were resampled (with replacement) within each of the 20 populations. These data were used to generate new G and D matrices and to recalculate both λi and θ. A second bootstrap model was used G E N E T I C C O N S T R A I N T S O N P O P U L AT I O N D I V E R G E N C E Table 1. The proportion of phenotypic variation of 12 traits associated with growth and phenology within and among 20 pop- ulations of Lythrum salicaria grown under uniform glasshouse conditions. Trait Vpop Vfam Vres H2 Leaf area at transplant Height at transplant Height at week 2 Height at week 4 Days to first flower Stem width at maturity Vegetative size at maturity Inflorescence length at maturity Vegetative size at harvest Inflorescence length at harvest Final vegetative biomass (ln) Final inflorescence biomass (ln) 0.178 0.242 0.257 0.184 0.383 0.427 0.542 0.103 0.513 0.253 0.377 0.197 0.088 0.109 0.069 0.068 0.129 0.100 0.111 0.044 0.104 0.060 0.070 0.057 0.734 0.649 0.674 0.747 0.488 0.473 0.347 0.853 0.383 0.686 0.553 0.745 0.428 0.575 0.371 0.334 0.836 0.698 0.969 0.196 0.854 0.322 0.449 0.284 Variance components are standardized to sum to one and were calculated from a mixed model; they describe divergence among populations (Vpop ), variation among seed families within populations (Vfam ), and residual variation (Vres ). Average within-population broad-sense heritabilities (H2 ) are estimated as 4×Vfam /(Vfam +Vres ). following Blows et al. (2004, electronic supplement), which is more appropriate for testing the null hypothesis because it uses an orthogonal factor rotation to force G and D into coincident subspaces (for details see Cohn 1999 as described in Blows et al. 2004). Both bootstrap models were written and implemented in R (version 2.8.1). In summary, the key measurements of the similarity between G and D are as follows. The first is θ, the angle of similarity between gmax and dmax (Schluter 1996), which ranges from 0 (identical vectors) to 1 (orthogonal, no similarity). Second, angles calculated from the eigenvalues of ai are analogous to θ, but allow for multiple dimensions of G and D to be compared. Third, the coefficients of bi show the “loadings” of each phenotypic trait for the corresponding ai . Finally, the sum of eigenvalues of ai represents an overall metric of similarity for the first six principal components of G and D and ranges from 0 (no similarity) to 6 (completely identical). Results QUANTITATIVE GENETIC VARIATION AND LATITUDINAL CLINES Variance–covariance matrices among populations (D), among seed families within populations (G), and among individuals within seed families (R) were estimated from a single mixed model and standardized to sum to one (Table 1). The LRT of a factor analytic model comparing six versus five principal compo- nents was highly significant (χ2 = 47.4, df = 7, P < 0.001), indicating significant standing genetic variation in all six eigenvectors of G. Separate LRTs for each trait confirmed highly significant effects of population (across all 12 traits: χ2 > 114.9, df = 1, P < 0.001) and seed family within population (across all 12 traits: χ2 > 15.6, df = 1, P < 0.001). Adult vegetative size and time to first flower exhibited the greatest level of divergence among populations (i.e., Vpop ), with variance among populations explaining 38.3% (days to first flower) to 54.2% (vegetative size at maturity) of the total phenotypic variance (Table 1). In contrast, “populations” explained only 17.8% (leaf area at transplant) to 25.7% (height-2wk) of the phenotypic variance in seedling traits, and 10.3% to 25.3% of the variance in inflorescence length and biomass (Table 1), indicating less divergence among populations for these traits relative to vegetative size and time to first flower. Broad-sense heritability (H 2 ), an estimate of standing genetic variation within populations (4 × Vfam /[Vfam +Vres ]), was highest for days to flower (83.6%) and vegetative size measured at harvest (85.4%) and at maturity (96.9%). In contrast, the lowest H 2 estimates were for inflorescence length measured at flowering (19.6%) and at final harvest (32.2%), as well as inflorescence biomass (28.4%). Seedling traits had intermediate heritabilities ranging from height-4wk (33.4%) to height at transplant (57.5%). Nine of the 12 traits we investigated were significantly correlated with latitude (Fig. 1). Consistent with patterns of population divergence and the relative amounts of standing genetic variation within populations, the four strongest clines involved days to flower, stem width, vegetative size, and vegetative biomass (R2 = 0.506 to 0.709). In contrast, the weakest clines (R2 = 0.031 to 0.279) were among seedling traits, inflorescence length, and inflorescence biomass, which generally showed quadratic or nonsignificant correlations with latitude. In general, these patterns indicate that northern plants flowered earlier at a smaller size and also remained small in stature until the end of the experiment. CORRELATIONS AMONG TRAITS—G AND D To visualize potential trade-offs among life-history traits and their influence on population divergence, we present for each pair of traits the bivariate plots of BLUPs for family means (Fig. 2, above diagonal) and BLUEs for population means (Fig. 2, below diagonal), which were estimated from the large mixed model. Correlations estimated from G may indicate genetic constraints among life-history traits, whereas intercorrelations in D may arise as a correlated response to selection or through stochastic processes acting on genetically correlated traits. The G matrix estimated directly from the mixed model is presented as Supporting information (Table S1, above diagonal), as well as genetic correlation coefficients calculated from G (Table S2, above diagonal). The variance and covariance components of the D matrix are also EVOLUTION SEPTEMBER 2011 2519 R . I . C O L AU T T I A N D S . C . H . BA R R E T T Figure 1. Bivariate plots illustrating latitudinal clines in 12 phenotypic traits related to growth and phenology measured in 20 populations of Lythrum salicaria grown under uniform glasshouse conditions. Measurements were made on developing seedlings (panels 1–4), on mature individuals on the day of first flowering (panels 4–8), and on all individuals at the end of the experiment (panels 9–12). Linear and quadratic approximations and R2 values are estimated by least-squared means regression. Phenotypic traits showing significant quadratic relationships with latitude are indicated by curved lines; traits with significant linear but nonsignificant quadratic relationships are indicated with straight lines. Significance of linear and quadratic terms was estimated by likelihood ratio tests using general linear models. P-values correspond to the significance of the best-fit model based on likelihood ratio tests of a single intercept (i.e., mean) and zero slope. provided (Table S1, below diagonal) along with correlation coefficients (Table S2, below diagonal). The strongest correlations in G were between time to first flower and stem width and between vegetative size and biomass (+0.549 < R < +0.994) (Fig. 2 and Table S2, above diago- 2520 EVOLUTION SEPTEMBER 2011 nal). These traits all loaded positively on gmax , the first principal component of the G matrix, which accounted for 45.7% of the total phenotypic variation among seed families within populations (Table 2). Genetic correlations among seedling traits were generally weaker (+0.173 < R < +0.827) (Fig. 2 and Table S2, above G E N E T I C C O N S T R A I N T S O N P O P U L AT I O N D I V E R G E N C E Figure 2. Visualization of the genetic variance–covariance matrix (G) and the variance–covariance matrix of divergence (D) among 20 populations of Lythrum salicaria grown under uniform glasshouse conditions. Above Diagonal: G estimated from standardized means of 339 seed families, estimated as best linear unbiased predictors (BLUPs) by restricted maximum likelihood (ReML), with seed family, nested within population as a random effect. Below Diagonal: D estimated from standardized means of 20 populations, estimated best linear unbiased estimators (BLUEs) by ReML, with population as a fixed effect. diagonal). The second principal component of G (g2 ), which accounted for 23.0% of the total variance among seed families, was positively correlated with seedling traits (Table 2). Correlations between size at transplant (i.e., transplant height and leaf area) and adult traits were considerably weaker (−0.306 < R < + 0.279) (Fig. 2 and Table S2, above diagonal). Combinations of these traits define other eigenvectors of G (g3 - g6 ), each of which accounts for less than 13% of the variation among seed families (Table 2). Thus, genetic correlations were strongest among days to flower and adult vegetative traits, intermediate among seedling traits, and weakest between pairs of adult and seedling traits. Population means were highly intercorrelated for most traits and correlations were generally stronger among population means (i.e., D matrix) than among seed families within populations (Fig. 2 and Table S2, below diagonal). Similar to the correlations evident in the G matrix: (1) flowering time and vegetative size measurements exhibited the highest trait correlations among populations (+0.698 < R < +0.999) (Fig. 2 and Table S2, below diagonal), (2) these traits loaded primarily on dmax , which explained 67.8% of the total phenotypic variation among population means (Table 3), and (3) seedling traits were not as strongly correlated (+0.532 < R < +0.893; below diagonal). However, unlike G, pairwise combinations of seedling and adult traits in D varied markedly (−0.201 < R < +0.847) (Fig. 2 and Table S2, below diagonal) and days to first flower loaded more heavily on d2 than dmax (Table 3). Thus, in common with genetic correlations within populations, population divergence was highly correlated among adult traits and less correlated for seedling traits. However, unlike genetic correlations, population divergence was strongly correlated for a number of pairs of adult and juvenile traits, and days to first flower was not as highly correlated with traits associated with adult vegetative size. EVOLUTION SEPTEMBER 2011 2521 R . I . C O L AU T T I A N D S . C . H . BA R R E T T Table 2. Factor loadings, eigenvalues, and percent variation explained by the first six eigenvectors of the average matrix of broad-sense genetic variance–covariance (G) within 20 populations of Lythrum salicaria grown under uniform glasshouse conditions. Trait gmax g2 g3 g4 g5 g6 Leaf area at transplant Height at transplant Height at week 2 Height at week 4 Days to first flower Stem width at maturity Vegetative size at maturity Inflorescence length at maturity Vegetative size at harvest Inflorescence length at harvest Final vegetative biomass (ln) Final inflorescence biomass (ln) Eigenvalue Percentage of variation Cumulative percentage of variation −0.082 −0.189 −0.147 −0.192 0.515 0.374 0.426 0.115 0.423 −0.210 0.252 −0.110 0.474 45.7% 45.7% 0.422 0.494 0.420 0.335 −0.049 0.174 0.237 0.019 0.231 0.153 0.274 0.220 0.238 23.0% 68.7% 0.118 0.366 0.214 −0.087 0.151 −0.272 0.110 −0.299 0.083 −0.445 −0.328 −0.537 0.133 12.8% 81.6% 0.560 0.302 −0.319 −0.490 0.190 0.079 −0.259 0.267 −0.254 0.018 0.093 −0.018 0.100 9.6% 91.2% 0.339 −0.294 −0.024 0.042 −0.165 0.584 −0.122 −0.616 −0.123 −0.054 −0.026 −0.124 0.035 3.4% 94.5% 0.446 −0.403 0.002 −0.004 0.053 −0.302 0.310 0.055 0.154 0.541 −0.274 −0.231 0.026 2.5% 97.0% Table 3. Factor loadings, eigenvalues, and percent variation explained by the first six eigenvectors of the matrix of population divergence (D) calculated among 20 population means of traits associated with growth and phenology in Lythrum salicaria grown under uniform glasshouse conditions. Trait dmax d2 d3 d4 d5 d6 Leaf area at transplant Height at transplant Height at week 2 Height at week 4 Days to first flower Stem width at maturity Vegetative size at maturity Inflorescence length at maturity Vegetative size at harvest Inflorescence length at harvest Final vegetative biomass (ln) Final inflorescence biomass (ln) Eigenvalue Percentage of variation Cumulative percentage of variation 0.202 0.255 0.246 0.109 0.311 0.386 0.453 0.092 0.440 −0.024 0.376 0.164 2.624 67.8% 67.8% 0.191 0.224 0.281 0.372 −0.420 −0.170 −0.125 0.239 −0.132 0.497 0.089 0.380 0.845 21.8% 89.6% 0.083 0.266 0.453 0.394 −0.152 −0.280 0.029 −0.234 0.019 −0.538 −0.118 −0.318 0.176 4.6% 94.2% 0.642 0.487 −0.191 −0.393 0.051 −0.006 −0.253 −0.132 −0.220 −0.127 0.100 0.036 0.101 2.6% 96.8% −0.245 0.054 0.018 −0.128 0.048 −0.238 −0.147 0.665 −0.133 −0.439 0.418 0.112 0.051 1.3% 98.1% −0.310 0.365 −0.085 −0.320 −0.030 −0.607 0.265 −0.213 0.289 0.263 0.147 −0.046 0.029 0.8% 98.9% POPULATION DIVERGENCE ALONG GENETIC LINES OF LEAST RESISTANCE Quantitative estimates of G and D indicate significant similarity between the two matrices (Fig. 3). Schluter’s (1996) θ, which measures the angle of similarity between gmax and dmax and ranges from 0o to 90o , suggested a moderate level of similarity of ∼46o (mean = 45.78; bootstrap mean = 45.72o ; 95% CI: 39.50–51.93o ). In contrast, the index of similarity (λi ) of the S-matrix, which compares the first six eigenvectors of G and D, was 4.8 of 6.0 (mean = 4.79; bootstrap mean = 4.80; 95% CI: 4.66–4.97), indicating a higher level 2522 EVOLUTION SEPTEMBER 2011 of similarity. Thus, while the first principal component eigenvectors of G and D were similar (θ = 45.8o /90o = 50.9%), the first six eigenvectors of S were more so (λi = 4.8/6.0 = 80.0%), and λi was less than any of the 10,000 iterations of the orthogonal rotation bootstrap, indicating highly significant similarity (P < 0.0001). The 10,000 bootstrap iterations of θ and λi indicated different levels of similarity when each was scaled from 0% (orthogonal) to 100% (identical) (Fig. 3). To investigate the trait correlations responsible for the similarity between G and D, we projected the eigenvectors of S into the subspace defined by the first six eigenvectors of G (Table 4). G E N E T I C C O N S T R A I N T S O N P O P U L AT I O N D I V E R G E N C E Table 4. Summary of the eigenvectors of the matrix of similarity (S) projected onto the eigenvectors of the genetic variance–covariance matrix (G) of 12 life-history traits in 20 populations of Lythrum salicaria grown under uniform glasshouse conditions. Trait b1 b2 b3 b4 b5 b6 Leaf area at transplant Height at transplant Height at week 2 Height at week 4 Days to first flower Stem width at maturity Vegetative size at maturity Inflorescence length at maturity Vegetative size at harvest Inflorescence length at harvest Final vegetative biomass (ln) Final inflorescence biomass (ln) Eigenvalue Angle 0.292 0.412 0.491 0.440 −0.094 0.070 0.305 −0.007 0.287 0.186 0.211 0.204 1.00 0.04 −0.307 −0.147 0.089 0.011 0.369 0.182 0.445 −0.202 0.440 −0.418 0.011 −0.316 1.00 1.25 0.195 0.075 −0.300 −0.308 0.353 0.425 0.059 0.419 0.097 0.034 0.457 0.261 0.989 5.99 0.594 0.358 0.019 −0.233 0.064 0.230 −0.156 −0.376 −0.156 −0.303 −0.106 −0.339 0.961 11.40 0.282 −0.640 −0.067 0.137 −0.147 0.448 0.090 −0.358 0.030 0.357 0.002 0.019 0.755 29.65 0.420 −0.132 −0.051 −0.192 0.248 −0.459 0.318 0.266 0.177 0.362 −0.258 −0.307 0.141 67.93 Angle quantifies the orientation of a principal component axis from each of D and G and ranges from 0 (aligned) to 90 (orthogonal). Factor loadings show the contribution of each trait to the similarity between G and D. For example, the closest principal components of G and D define b1 and are largely due to the intercorrelation of stem width, biomass, and vegetative size measured at maturity and at harvest. The first five of six eigenvectors of this projection (b1 to b5 ) were close to the maximum of one (eigenvalues b1 = 1.00, b2 = 1.00, b3 = 0.989, b4 = 0.969, b4 = 0.755) with corresponding angles of 0.04o to 29.65o (Table 4). Seedling height measurements loaded most heavily on b1 . Vegetative size at maturity and at harvest loaded most heavily on b2 , along with inflorescence length at harvest (Table 4). Stem width, vegetative biomass, and inflorescence length at maturity loaded primarily on b3 . Factor loadings for days to first flowering were highest for b2 and b3 along with measurements of adult size. The remaining eigenvectors of S (b4 b6 ) were defined by a combination of seedling and adult traits. Discussion The goals of this study on invasive populations of L. salicaria from eastern North America were: (1) to estimate standing genetic variation for ecologically relevant life-history traits using seed families grown under uniform conditions, (2) to identify genetic correlations for these traits within and among populations, and (3) to determine to what extent divergence among populations may be constrained by the structure of genetic (co)variance within populations. We found that despite evidence indicating that founder events and genetic drift play an important role during the invasion process in these populations (Eckert and Barrett 1992; Eckert et al 1996), they maintained high levels of genetic variation for all 12 of the quantitative traits that we examined (Table 1). We also detected strong population differentiation for these traits, most of which was manifested as geographical clines distributed along latitudinal gradients in growing season length (Fig. 1). Despite the availability of significant standing genetic variation within and among populations, genetic correlations among traits (Table 2, Fig. 2) appear to limit the “phenotypic space” available for populations to respond to natural selection, at least over the short term. Evolutionary change during invasion has occurred primarily along lines of greatest genetic (co)variance within populations resulting in a high similarity between the first five eigenvectors of G and D (Table 4; Fig. 3). Below we discuss the implications of these results and consider alternative hypotheses for the similarity observed between G and D, particularly the possible roles of migration and correlational selection. GENETIC VARIATION IN LIFE-HISTORY TRAITS There is considerable evidence for abundant genetic variation in life-history traits within and among wild populations of plants and animals (reviewed by Mousseau and Roff 1987; Houle 1992; Mazer and LeBuhn 1999; Geber and Griffen 2003) although relatively few studies have measured quantitative genetic variation in populations of invasive species (but see Rice and Mack 1991; Chen et al. 2006; Lavergne and Molofsky 2007; Dlugosch and Parker 2008b; Facon et al. 2008; Chun et al. 2009). Previous work on L. salicaria reported significant amounts of additive genetic variation for age and size at flowering in two native populations from Sweden (Olsson 2004) and in a single introduced population from eastern North America (O’Neil 1997). Consistent with a scenario of multiple introductions and admixture fostered by the outcrossed mating system of L. salicaria (Thompson et al. 1987; Barrett 2000; Houghton-Thompson et al. 2005; Chun et al. 2009), we found high levels of genetic variation for vegetative and EVOLUTION SEPTEMBER 2011 2523 R . I . C O L AU T T I A N D S . C . H . BA R R E T T northern range limit. Levels of standing genetic variation and the strength of genetic correlations may also differ among populations for the traits examined here. For example, stronger stabilizing selection on flowering time in northern populations may weaken the strength of the genetic correlation between days to first flower and vegetative size. Quantifying such changes in the variance and covariance for genetically correlated traits is analytically difficult and has only recently been attempted (see Hine et al. 2009). Such an analysis was beyond the scope of the present study and instead we examined the “average” genetic (co)variance matrix for the 20 populations we investigated. TRAIT CORRELATIONS AS GENETIC CONSTRAINTS Figure 3. Bootstrap estimates of two measures of similarity be- tween the matrix of genetic variance-covariance (G) and the matrix of covariance among population means (D) estimated from 20 populations of Lythrum salicaria grown under uniform glasshouse conditions. (A) The angle of orientation between gmax and dmax (θ) examines only the first principal components of G and D (see Schluter 1996) and ranges from 0◦ (complete similarity) to 90◦ (no similarity). (B) The index of similarity compares the first six principal components of G and D (see Blows et al. 2004) and ranges from 0 (no similarity) to 6 (complete similarity). The bootstrap models consist of 10,000 iterations of resampling, with replacement, from 339 standardized family means to generate a bootstrap sample of 10,000 G and D matrices. reproductive traits both within and among the 20 populations that we investigated (Table 1). Although hybridization with native L. alatum has been proposed as a genetic mechanism contributing towards invasion success in L. salicaria (Ellstrand and Schierenbeck 2000; Houghton-Thompson et al. 2005), AFLP data do not provide strong support for this hypothesis (see Fig. 2 in HoughtonThompson et al. 2005). Instead, it seems more likely that gene flow among introduced genotypes and possibly ornamental varieties may have contributed to the high genetic diversity of invasive populations (Houghton-Thompson et al. 2005; Chun et al. 2009). Our previous studies revealed striking variation among populations in their overall levels of quantitative genetic variation (Colautti et al. 2010). In particular, we reported that genetic variation for days to first flower and vegetative size declined toward the 2524 EVOLUTION SEPTEMBER 2011 Genetic correlations lower the “dimensionality” of available phenotypic space and can limit opportunities for local adaptation, at least over short timescales (Dickerson 1955; Lande 1982; Orr 2000). The high intercorrelation among adult traits reported here suggests that selection on earlier flowering in northern populations may be constrained by a trade-off with size at reproduction. Specifically, we detected strong positive genetic intercorrelations among days to first flower, stem width, vegetative size at flowering, vegetative size at harvest, and vegetative biomass (Fig. 2 and Table S2). Consequently, plants that flowered earlier are likely to suffer a cost associated with their smaller size. The smaller vegetative size of early-flowering genotypes was fixed at the time of first flowering and did not increase throughout the growing season, given the strong broad-sense genetic correlation between vegetative height at first flowering and at harvest (r = 0.994). In native populations of L. salicaria, Olsson (2004) also found significant additive genetic correlations between time to first flower and the number of vegetative nodes at maturity, a proxy for vegetative size. Thus, it appears that: (1) introduced populations are subject to some of the same genetic constraints that are evident in native populations, and (2) these trade-offs may be relatively stable over longer timescales. Indeed, this may be a more general phenomenon, as the evolution of earlier reproduction at higher latitudes appears to be constrained in two other species—Xanthium strumarium (Etterson and Shaw 2001) and Chamaecrista fasciculata (Griffith and Watson 2006). Constraints on the evolution of early flowering because of shorter growing seasons may be an important determinant of range limits generally. A longer term genetic constraint involving early flowering and size in L. salicaria may be associated with the architecture of inflorescence development. In common with several other dicotyledonous species, for example, Antirrhinum majus and Arabidopsis thaliana (reviewed in Yanofsky 1995; Ma 1998; Simpson et al. 1999), organogenesis in L. salicaria proceeds in an acropetal direction at the shoot apical meristem. In A. majus and A. thaliana, maturation of the primary stem occurs in response to external cues, particularly temperature and photoperiod, which trigger a G E N E T I C C O N S T R A I N T S O N P O P U L AT I O N D I V E R G E N C E developmental phase change from resource-gathering structures (i.e., shoots and leaves) to resource sinks (i.e., flowers, nectar, seeds). This may result in a fitness trade-off because vegetative size correlates strongly with flower and fruit production (O’Neil 1997; Colautti et al. 2010). In the plants we investigated, genetic variation for vegetative growth after the onset of flowering must be negligible compared to genetic variation for size at maturity, otherwise there would be a weaker correlation between vegetative size measured at first flower and at the end of the experiment. This developmental constraint may help to explain, in part, the intercorrelations among days to first flower and the measurements of plant size (see Fig. 2). Time to first flower and adult traits associated with vegetative size fit the model of plant development noted above, but the results for inflorescence length and seedling traits (i.e., height at transplant, leaf area at transplant, height-2wk, and height-4wk) are harder to interpret. Variance–covariance components involving inflorescence length and biomass should be interpreted carefully because plants in our experiment had no opportunity to produce seeds, which require insect pollinators. Therefore, resources available for seed production, unlike natural populations, would not limit the length and biomass of inflorescences measured in the glasshouse. Indeed, inflorescence measurements correlated poorly between the glasshouse and the field, despite strong correlations between glasshouse and field measurements for flowering time and vegetative traits (Montague et al. 2008; Colautti et al. 2010). Overall, seedling traits were less genetically intercorrelated within populations than adult traits, but there were a few adult traits that were strongly correlated with seedling traits (Fig. 2 and Table S2, above diagonal). For example, seedling height measurements were negatively correlated with days to first flower (i.e., early-flowering plants grew faster), yet seedling height measurements were only weakly correlated with adult height measurements, despite a strong positive correlation between days to flower and vegetative size at flowering. Thus, early-flowering plants grow faster as seedlings but are still not able to grow as large as later-flowering plants. This result is consistent with a model of developmental constraint proposed by Colautti et al. (2010) in which plants must reach a threshold size before they begin to flower. Seedlings that grow faster are able to initiate flowering sooner, whereas slower-growing seedlings delay flowering until they reach a threshold size. GENETIC VARIANCE–COVARIANCE (G) AND POPULATION DIVERGENCE (D) If the genetic correlations illustrated in Figure 2 represent constraints on population divergence then the genetic (co)variance matrix (G) should be a reasonable predictor of trait (co)variance among populations (D). Indeed, this is the case as the index of similarity (λi ) between G and D was 4.8, which is close to the theoretical maximum (6 = identical matrices) and highly significant based on a bootstrap model with orthogonal rotation (P < 0.001). The high value of λi derives from the fact that five of the six eigenvectors of S were close to the theoretical maximum of one, with only the final eigenvector (b6 ) showing little similarity between G and D (Table 4). The similarity in five of the six dimensions of comparison between G and D is consistent with strong, multitrait genetic constraint on population divergence during the invasion of L. salicaria in eastern North America. Interpreting the similarity between G and D as a constraint on population divergence assumes that our estimate of G is a reasonable approximation of the ancestral (co)variance structure of each population (i.e., a 20-population polytomy). The genetic relationships among the 20 populations used in this study have not been characterized using neutral genetic markers. However, introduced populations of invasive plants often show little geographical structuring relative to native populations (Barrett and Husband 1990; Dlugosch and Parker 2008a) and this is also true for other populations of L. salicaria in North America (Houghton-Thompson et al. 2005; Chun et al. 2009). Human-influenced gene flow in invasive species tends to homogenize the phylogeographic relationships among populations, which would otherwise confound interpretation of similarities between G and D as genetic constraints on evolution. Future work combining neutral markers with measurements of natural selection at different points along the latitudinal gradient would clarify the relative influence of genetic constraints, natural selection, and stochastic processes on divergence of our study populations (see methods in Chenoweth and Blows 2008; Hohenlohe and Arnold 2008; Chenoweth et al. 2010). The different levels of constraint suggested by θ and S are difficult to reconcile with nonadaptive processes. The estimated average angle (θ) between gmax and dmax was about half the theoretical minimum similarity (90o = no similarity) (Fig. 3) and less similar than five of the six eigenvectors of S (0.04–29.65o ; Table 4). Stochastic processes such as bottlenecks, founder events, and genetic drift should not affect the proportional relationship between G and D and therefore the factor loadings and eigenvalues of gmax and dmax should be similar (Lande 1979; Jones et al. 2004; Hohenlohe and Arnold 2008). Instead, dmax explains 67.8% of the total (co)variance among populations, whereas gmax explains only 45.7% of total genetic (co)variance within populations, demonstrating that the first principal component eigenvectors of G and D are not proportional, despite the high degree of similarity between G and D. Therefore, stochastic processes alone would appear to be insufficient to explain the contrasting estimates of similarity measured by θ and λi . In contrast to stochastic processes, strong natural selection can result in divergence of populations away from the primary eigenvectors of G (Lande 1979; Zeng 1988; Jones et al. 2004). EVOLUTION SEPTEMBER 2011 2525 R . I . C O L AU T T I A N D S . C . H . BA R R E T T The difference in constraint suggested by θ and λi could be explained if: (1) selection on a few traits changed the orientation of dmax relative to gmax , but (2) the evolutionary response to selection was still constrained by genetic correlations with other lifehistory traits. This seems likely for two reasons. First, loadings of traits differed between gmax (Table 2) and dmax (Tables 3), suggesting a weak constraint in the first dimensions of G and D. Indeed, the large factor loadings for days to first flower and other adult size traits in gmax is consistent with a strong genetic constraint, whereas the weaker loading of days to flowering in dmax is consistent with selection that breaks apart this constraint. Second, the eigenvalues of the first five eigenvectors of S (i.e., b1 , to b5 in Table 4) were close to their theoretical maximum of one, suggesting a strong constraint in five of the six principal eigenvectors G and D. Therefore, selection appears to have played an important role in the divergence of populations for a few traits, but overall was highly constrained along genetic lines of (co)variance among life-history traits. ALTERNATIVE HYPOTHESES FOR G-D SIMILARITY We have interpreted the similarity between G and D (Fig. 3B) as a constraint on the evolution of D imposed by the genetic variance and covariance components of G. However, theory suggests other factors that can also orient G toward D (reviewed in Arnold et al. 2008). One hypothesis is that correlational selection favors combinations of traits within populations that mirror the direction of divergence among populations. For example, the genetic correlation between stem width and vegetative size at flowering (Fig. 2) may be a result of correlational selection, as larger plants may need larger stems to increase transport of more resources or as structural support. However, time to flowering and vegetative size are positively correlated in G and D (Fig. 2) yet selection coefficients measured in L. salicaria confirm that selection favors both early flowering and larger size (Colautti and Barrett 2010) without strong correlational selection (O’Neil 1997). Therefore, correlational selection alone is an unlikely explanation for the strong similarity between G and D. Effects of maternal environment on offspring growth and development (i.e., maternal effects) could also result in correlations among life-history traits resulting in similarity between G and D. However, experimental evidence using the same populations in this study does not support a significant influence of maternal effects on adult traits (see Montague et al. 2008; Colautti et al. 2010). This conclusion was also supported by the weak correlations between vegetative size at harvest and seedling traits in G and D (Fig. 2 and Table S2). Maternal effects are known to influence seedling traits in many plant species as higher quality seeds germinate earlier and grow faster (Roach and Wulff 1987). However, in our experiment the opposite was true, as germination date was positively correlated with relative growth rate from 2526 EVOLUTION SEPTEMBER 2011 transplant to week 2 (ρ = +0.383, df = 2767 P < 0.001) and from week 2 to week 4 (ρ = +0.278, df = 2766, P < 0.001). Our results are therefore not consistent with the hypothesis that the similarity between G and D results from maternal influences on correlations among life-history traits. A third hypothesis that predicts similarity between G and D is that gene flow along latitudinal gradients can create latitudinal clines like those observed in our study (Fig. 1) and orient G in the direction of D (Guillaume and Whitlock 2007). This scenario would require a strong pattern of isolation-by-distance (IBD), which is unlikely for a species that has spread over 1000 km in the past 50–100 years and shows evidence of significant long-distance dispersal (see Houghton-Thompson et al. 2005; Chun et al. 2009). Moreover, strong IBD alone cannot explain the lower level of divergence of θ relative to λi (Fig. 3A) because the strength of the covariance components of G should be proportional to the same covariances of D, resulting in the greatest similarity between gmax and dmax . The difference in factor loadings for days to first flower in G (Table 2) versus D (Table 3) is contrary to a scenario of IBD. Instead, our results are more consistent with genetic constraints on an evolutionary response to selection on population divergence. CONCLUSIONS Identifying genetic constraints on local adaptation is an important step in understanding species’ range limits and for predicting the rate and extent of spread in invasive species. However, identifying constraints in natural populations is complicated by variation among populations in gene flow, natural selection, and genetic drift because these processes can lead to idiosyncratic differences in the magnitude and direction of genetic correlations. We have attempted to circumvent this problem by estimating an “average G” in 20 populations of L. salicaria from eastern North America, a method that is particularly well-suited to introduced species that are likely to have relatively weak phylogeographical structuring as a result of human-mediated dispersal. Despite considerable standing genetic variation within populations of L. salicaria for individual traits, genetic correlations among traits appear to have limited the phenotypic space available for divergence among populations, particularly for traits that display strong latitudinal clines. The similarity in several of the genetic correlations between native and introduced populations of L. salicaria suggests that genetic constraints on population divergence may play an important role in the establishment of range limits in invasive species. ACKNOWLEDGMENTS We thank M. Blows, J. Stinchcombe, K. Rice, A. Weis, and C. Eckert for comments on the manuscript; L. Flagel for assistance running SAS on University of North Carolina’s research computing cluster; the Ontario G E N E T I C C O N S T R A I N T S O N P O P U L AT I O N D I V E R G E N C E Government and the University of Toronto for scholarship support to RIC; the Canada Research Chair program and an Ontario Premier’s Discovery Award for funding to SCHB; and the Natural Sciences and Engineering Research Council of Canada (NSERC) for a graduate scholarship to RIC and a Discovery Grant to SCHB. LITERATURE CITED Arnold, S. J., R. Bürger, P. A. Hohenlohe, B. C. Ajie, and A. G. Jones. 2008. Understanding the evolution and stability of the G-matrix. Evolution 62:2451–2461. Baker, H. G. 1965. Characteristics and modes of origin of weeds. Pp. 147–168 in H. G. Baker and G. L. Stebbins, eds. The genetics of colonizing species. Academic Press, New York. Barrett, S. C. H. 2000. Microevolutionary influences of global change on plant invasions. Pp. 115–139 in H. A. Mooney and R. K. Hobbs, eds. The impact of global change on invasive species. Island Press, Covelo, CA. Barrett, S. C. H., and J. S. Shore. 1989. Isozyme variation in colonizing plants. Pp. 106–126 in D. Soltis and P. Soltis, eds. Isozymes in plant biology. Dioscorides Press, Portland, OR. Barrett, S. C. H., and B. C. Husband. 1990. Genetics of plant migration and colonization. Pp. 254–277 in A. H. D. Brown, M. T. Clegg, A. L. Kahler and B. S. Weir, eds. Plant population genetics, breeding and genetic resources. Sinauer Associates, Sunderland, MA. Blows, M. W., S. Chenoweth, and E. Hine. 2004. Orientation of the genetic variance-covariance matrix and the fitness surface for multiple male sexually-selected traits. Am. Nat. 163:329–340. Blossey, B., L. C. Skinner, and J. Taylor. 2001. Impact and management of purple loosestrife (Lythrum salicaria) in North America. Biodivers. Conserv. 10:1787–1807. Calsbeek, B., and C. J. Goodnight. 2009. Empirical comparison of G matrix test statistics: finding biologically relevant change. Evolution 63:2627– 2635. Chen, Y. H., S. B. Opp, S. H. Berlocher, and G. K. Roderick. 2006. Are bottlenecks associated with colonization? Genetic diversity and diapause variation of native and introduced Rhagoletis completa populations. Oecologia 149:656–667. Chenoweth, S. F., and M. W. Blows. 2008. Qst meets the G matrix: the dimensionality of adaptive divergence in multiple correlated quantitative traits. Evolution 62:1437–1449. Chenoweth, S. F., H. D. Rundle, and M. W. Blows. 2010. The contribution of selection and genetic constraints to phenotypic divergence. Am. Nat. 175:186–196. Chun, Y. J., J. D. Nason, and K. A. Moloney. 2009. Comparison of quantitative and molecular genetic variation of native vs. invasive populations of purple loosestrife (Lythrum salicaria L. Lythraceae). Mol. Ecol. 18:3020–3035. Cohn, R. D. 1999. Comparisons of multivariate relational structures in serially correlated data. J. Agric. Biol. Environ. Stat. 4:238–257. Colautti, R. I., and S. C. H. Barrett. 2010. Natural selection and genetic constraints on flowering phenology in an invasive plant. Int. J. Plant Sci. 171:960–971. Colautti, R. I., C. G. Eckert, and S. C. H. Barrett. 2010. Evolutionary constraints on adaptive evolution during range expansion in an invasive plant. Proc. R. Soc. Lond. B 277:1799–1806. Dickerson, G. E. 1955 Genetic slippage in response to selection for multiple objectives. Cold Spring Harb. Symp. Quant. Biol. 20:213–224. Dlugosch, K. M., and I. M. Parker. 2008a. Founding events in species invasions: genetic variation, adaptive evolution, and the role of multiple introductions. Mol. Ecol. 17:431–449. ———. 2008b. Invading populations of an ornamental shrub show rapid life history evolution despite genetic bottlenecks. Ecol. Lett. 11:701–709. Eckert C. G., and S. C. H. Barrett. 1992. Stochastic loss of style morphs from populations of tristylous Lythrum salicaria and Decodon verticillatus (Lythraceae). Evolution 46:1014–1029. Eckert, C. G., D. Manicacci, and S. C. H. Barrett. 1996. Genetic drift and founder effect in native versus introduced populations of an invading plant, Lythrum salicaria (Lythraceae). Evolution 50:1512– 1519. Ellstrand, N. C., and K. A. Schierenbeck. 2000. Hybridization as a stimulus for the evolution of invasiveness in plants? Proc. Natl. Acad. Sci. USA. 97:7043–7050. Etterson, J. R. 2004. Evolutionary potential of Chamaecrista fasciculata in relation to climate change. II. Genetic architecture of three populations reciprocally planted along an environmental gradient in the Great Plains. Evolution 58:1459–1471. Etterson, J. R., and R. G. Shaw. 2001. Constraint on adaptive evolution in response to global warming. Science 294:151–154. Facon, B., J. P. Pointier, P. Jarne, V. Sarda, and P. David. 2008. High genetic variance in life-history strategies within invasive populations by way of multiple introductions. Curr. Biol. 18:363–367. Fisher, R. A. 1930. The genetical theory of natural selection, Oxford Univ. Press, Oxford, U.K. Garcı́a-Ramos, G., and D. Rodrı́guez. 2002. Evolutionary speed of species invasions. Evolution 56:661–668. Geber, M. A., and L. R. Griffen. 2003. Inheritance and natural selection on functional traits. Int. J. Plant Sci. 164:S21–S42. Griffith, T. M., and M. A. Watson. 2006. Is evolution necessary for range expansion? Manipulating reproductive timing of a weedy annual transplanted beyond its range. Am. Nat. 167:153–164. Guillaume, F., and M. C. Whitlock. 2007. Effects of migration on the genetic covariance matrix. Evolution 61:2398–2409. Hine, E., S. F. Chenoweth, H. D. Rundle, and M. W. Blows. 2009. Characterizing the evolution of genetic variance using genetic covariance tensors. Phil. Trans. R. Soc. B 364:1567–1578. Hohenlohe, P. A., and S. J. Arnold 2008. MIPod: a hypothesis-testing framework for microevolutionary inference from patterns of divergence. Am. Nat. 171:366–385. Houghton-Thompson, J., H. H. Prince, J. J. Smith, and J. F. Hancock. 2005. Evidence of hybridization between Lythrum salicaria (purple loosestrife) and L. alatum (winged loosestrife) in North America. Ann. Bot. 96:877–885. Houle, D. 1992. Comparing evolvability and variability of quantitative traits. Genetics 130:195–204. Jones, A. G., S. J. Arnold, and R. Bürger. 2003. Stability of the G-matrix in a population experiencing pleiotropic mutation, stabilizing selection, and genetic drift. Evolution 57:1747–1760. ———. 2004. Evolution and stability of the G-matrix on a landscape with a moving optimum. Evolution 58:1639–1654. Keller, S. R., and D. R. Taylor. 2010. Genomic admixture increases fitness during a biological invasion. J. Evol. Biol. 23:1720–1731. Kempthorne, O. 1955. The correlations between relatives in a simple auto tetraploid population. Genetics 40:168–174. Krzanowski, W. J. 1979. Between-group comparisons of principal components. J. Am. Stat. Assoc. 74:703–707. Lande, R. 1979. Quantitative genetic analysis of multivariate evolution, applied to brain: body size allometry. Evolution 33:402–416. ———. 1982. A quantitative genetic theory of life-history evolution. Ecology 63:607–615. ———. 1988. Genetics and demography in biological conservation. Science 241:1455–1460. EVOLUTION SEPTEMBER 2011 2527 R . I . C O L AU T T I A N D S . C . H . BA R R E T T Lavergne, S., and J. Molofsky. 2007. Increased genetic variation and evolutionary potential drive the success of an invasive grass. Proc. Natl. Acad. Sci. USA 104:3883–3888. Lee, C. E. 2002. Evolutionary genetics of invasive species. Trends Ecol. Evol. 17:386–391. Lee, C. E., J. L. Remfert, and Y. M. Chang. 2007. Response to selection and evolvability of invasive populations. Genetica 129:179–192. Lee, P. L. M., R. M. Patel, R. S. Conlan, S. J. Wainwright, and C. R. Hipkin. 2004. Comparison of genetic diversities in native and alien populations of hoary mustard (Hirschfeldia incana [L.] Lagreze-Fossat). Int. J. Plant Sci. 165:833–843. Lynch, M. and B. Walsh. 1998. Genetics and analysis of quantitative traits. Sinauer, Sunderland, MA. Ma, H. 1998. To be, or not to be, a flower – control of floral meristem identity. Trends Genet. 14:26–32. Mal, T. K., J. Lovett-Doust, L. Lovett-Doust, and G. A. Mulligan. 1992. The biology of Canadian weeds. 100. Lythrum salicaria. Can. J. Plant Sci. 72:1305–1330. Malacrida, A. R., F. Marinoni, C. Torti, L. M. Gomulski, F. Sebastiani, C. Bonvicini, G. Gasperi, and C. R. Guglielmino. 1998. Genetic aspects of the worldwide colonization process of Ceratitis capitata. J. Hered. 89:501–507. Mazer, S. J., and G. LeBuhn. 1999. Genetic variation in life-history traits: heritability estimates within and genetic differentiation among populations. Pp. 85–171 in T. O. Vuorisalos and P. K. Mutikainen, eds. Life history evolution in plants. Kluwer Academic, Hague, The Netherlands. McGoey, B. V., and J. R. Stinchcombe. 2009. Interspecific competition alters natural selection on shade avoidance phenotypes in Impatiens capensis. New Phytol. 183:880–891. McGuigan, K., and M. W. Blows. 2010. Evolvability of individual traits in a multivariate context: partitioning the additive genetic variance into common and specific components. Evolution 64:1899–1911 McGuigan, K., S. F. Chenoweth, and M. W. Blows. 2005. Phenotypic divergence along lines of genetic variance. Am. Nat. 165:32–43. McKay, J. K., and R. G. Latta. 2002. Adaptive population divergence: markers, QTL and traits. Trends Ecol. Evol. 17:285–291. Merilä, J., and P. Crnokrak. 2001. Comparison of genetic differentiation at marker loci and quantitative traits. J. Evol. Biol. 14:892– 903. Mitchell-Olds, T. 1996. Pleiotropy causes long-term genetic constraints on life-history evolution in Brassica rapa. Evolution 50:1849–1858. Montague, J. L., S. C. H. Barrett, and C. G. Eckert. 2008. Re-establishment of clinal variation in flowering time among introduced populations of purple loosestrife (Lythrum salicaria, Lythraceae). J. Evol. Biol. 21:234– 245. Mousseau, T. A., and D. A. Roff. 1987. Natural selection and the heritability of fitness components. Heredity 59:181–197. Novak, S. J., and R. N. Mack. 2005. Genetic bottlenecks in alien plant species. Influence of mating systems and introduction dynamics. Pp. 201–228 in D. F. Sax, J. J. Stachowicz and S. D. Gains, eds. Species invasions: insights into ecology, evolution and biogeography. Sinauer Associates, Sunderland, MA. Olsson, K. 2004. Population differentiation in Lythrum salicaria along a latitudinal gradient. Ph.D. diss. Department of Ecology and Environmental Science, Umeå University, Umeå, Sweden. Olsson, K., and J. Ågren. 2002. Latitudinal population differentiation in phenology, life history and flower morphology in the perennial herb Lythrum salicaria. J. Evol. Biol. 15:983–996. O’Neil, P. 1997. Natural selection on genetically correlated phenological characters in Lythrum salicaria L. (Lythraceae). Evolution 51:267– 274. 2528 EVOLUTION SEPTEMBER 2011 Orr, H. A. 2000. Adaptation and the cost of complexity. Evolution 54:13–20. Parker, I. M., J. Rodriguez, and M. E. Loik. 2003. An evolutionary approach to understanding the biology of invasions: local adaptation and general-purpose genotypes in the weed Verbascum thapsus. Conserv. Biol. 17:59–72. Phillips, P. C., and S. J. Arnold. 1999. Hierarchical comparison of genetic variance-covariance matrices. I. Using the Flury hierarchy. Evolution 53:1506–1515. Phillips, P. C., M. C. Whitlock, and K. Fowler. 2001. Inbreeding changes the shape of the genetic covariance matrix in Drosophila melanogaster. Evolution 158:1137–1145. Rice, K. J., and R. N. Mack. 1991. Ecological genetics of Bromus tectorum. I. A hierarchical analysis of phenotypic variation. Oecologia 88:77–83. Roach, D. A., and R. D. Wulff. 1987. Maternal effects in plants. Ann. Rev. Ecol. Syst. 18:209–235. Roff, D. A., and D. J. Fairbairn. 2007. The evolution of trade-offs: where are we? J. Evol. Biol. 20:433–447. Roff, D. A., R. A. Mousseau, and D. J. Howard. 1999. Variation in genetic architecture of calling song among populations of Allonemobius socius, A. fasciatus, and a hybrid population: drift or selection? Evolution 53:216–224. Ross, C. A., D. Fauset, and H. Auge. 2009. Mahonia invasions in different habitats: local adaptation or general-purpose genotypes? Biol. Invasions 11:441–452. Satterthwaite, F. E. 1946. An approximate distribution of estimates of variance components. Biometrics Bull. 2:110–114. Scheiner, S. M. 1993. Genetics and evolution of phenotypic plasticity. Ann. Rev. Ecol. Evol. Syst. 24:35–68. Schluter, D. 1996. Adaptive radiation along genetic lines of least resistance. Evolution 50:1766–1774. Shaw, R. G. 1991. The comparison of quantitative genetic parameters between populations. Evolution 45:143–151. Simonsen, A. K., and J. R. Stinchcombe. 2010. Quantifying evolutionary genetic constraints in the Ivyleaf morning glory, Ipomoea hederacea. Int. J. Plant. Sci. 171:972–986. Simpson, G. G., A. R. Gendall, and C. Dean. 1999. When to switch to flowering. Ann. Rev. Cell Dev. Biol. 99:519–550. Thompson, D. Q., R. L. Stuckey, and E. B. Thompson. 1987. Spread, impact, and control of purple loosestrife (Lythrum salicaria) in North American Wetlands. U. S. Fish and Wildlife Service, Washington, DC. Tsutsui N. D., A. V. Suarez, D. A. Holway, and T. J. Case. 2000. Reduced genetic variation and the success of an invasive species. Proc. Natl. Acad. Sci. USA 97:5948–5953. Tufto, J. 2000. The evolution of plasticity and nonplastic spatial and temporal adaptations in the presence of imperfect environmental cues. Am. Nat. 156:121–130. Turelli, M. 1988. Phenotypic evolution, constant covariances and the maintenance of additive variance. Evolution 42:1342–1347. USDA, NRCS. 2009. The PLANTS Database Available at http://plants.usda.gov, (accessed July 15, 2009). National Plant Data Center, Baton Rouge, LA; 70874–4490. Verhoeven, K. J. F., M. Macel, L. M. Wolfe, and A. Biere. 2011. Population admixture, biological invasions and the balance between local adaptation and inbreeding depression. Proc. R. Soc. Lond. B 278:2–8. Via, S., and R. Lande. 1985. Genotype-environment interaction and the evolution of phenotypic plasticity. Evolution 39:505–522. Williams, D. G., R. N. Mack, and R. A. Black 1995. Ecophysiology of introduced Pennisetum setaceum on Hawaii: the role of phenotypic plasticity. Ecology 76:1569–1580. G E N E T I C C O N S T R A I N T S O N P O P U L AT I O N D I V E R G E N C E Yakimowski, S. B., H. A. Hager, and C. G. Eckert. 2005. Limits and effects of invasion by the nonindigenous wetland plant Lythrum salicaria (purple loosestrife): a seed bank analysis. Biol. Invasions 7: 687–698. Yanofsky, M. F. 1995. Floral meristems to floral organs: genes controlling early events in Arabidopsis flower development. Annu. Rev. Plant Physiol. Plant Mol. Biol. 46:167–188. Zeng, Z.-B. 1988. Long-term correlated response, interpopulation covariation, and interspecific allometry. Evolution 42:363–374. Zhang, Y.-Y., D.-Y. Zhang, and S. C. H. Barrett. 2010. Genetic uniformity characterizes the invasive spread of water hyacinth (Eichhornia crassipes), a clonal aquatic plant. Mol. Ecol. 19:1774–1786. Associate Editor: M. Blows Supporting Information The following supporting information is available for this article: Table S1. Variances (diagonal) and covariances (off-diagonal) among 12 life-history traits for the G matrix (above diagonal)—the (co)variance matrix among seed families within populations estimated by restricted maximum likelihood, and for the D matrix (below diagonal)—the (co)variance matrix among standardized population means. Table S2. Correlation coefficients for 12 life-history traits calculated from the G matrix (above diagonal)—the (co)variance matrix among seed families within populations estimated by restricted maximum likelihood, and for the D matrix (below diagonal)—the (co)variance matrix among standardized population means. Supporting Information may be found in the online version of this article. Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article. EVOLUTION SEPTEMBER 2011 2529
© Copyright 2025 Paperzz