Annual populations exhibit clinal variation in phenological, physiological, and vegetative traits that reflect climatic variation in the length of the growing season , whereas perennial populations from coastal, inland, and montane habitats comprise separate ecotypes with consistent phenotypic differences . Coastal and inland perennials overwinter as vegetative rosettes produced from above-ground stolons , whereas montane perennials primarily overwinter as below-ground rhizomes . Given extensive phenotypic variation in these and other traits and incomplete reproductive barriers, taxonomists have variously recognized between 4-20 species in the M. guttatus complex . Recent genomic evidence suggests extensive introgression among ecotypes, reflecting the recent origin and /or on-going gene flow among taxa in this complex .Direct selection – The strength and pattern of direct selection varied among ecotypes, years, and fitness components . In general, fitness components related to fecundity showed consistently positive relationships with flowering stem diameter and flower size across ecotypes and years . Conversely, selection on flowering time via fecundity contributions varied from positive to negative among years and ecotypes,though selection through ovule number was always positive whereas selection through fruit number was always negative . Within perennial ecotypes, direct selection through rosette production was only detected in 2012 . In contrast to the pattern for fecundity selection, selection on flowering stem diameter through rosette production was negative in montane perennials, though this was only marginally significant . In inland perennials, wholesale plant containers later flowering and larger seedlings increased rosette production, though selection on seedling size was only marginally significant .

In 2012, inland perennials only experienced selection through rosette production, whereas in 2013 all selection in this ecotype occurred through fecundity . There was no significant relationship between adult overwinter survival and any of the four traits examined in either year. Integrated selection – Integrating selection by estimating trait elasticities altered the strength of selection relative to estimates through individual fitness components. For most traits, integrated selection was weaker than direct selection because individual fitness components are weighted by their elasticities, which are necessarily less than one. This effect was most pronounced in perennial ecotypes because fitness component elasticities were decreased in perennials relative to annuals, which exhibit high elasticities for fecundity . Large selection gradients in perennial ecotypes occurred through fitness components with smaller elasticities . Correspondingly, integrated selection on flowering time was generally weaker in perennials relative to annuals , even though direct selection gradients through individual fitness components were larger in perennials . In one case, integrated selection was stronger than direct selection because selection occurred in the same direction through multiple fitness components. Robust annuals experienced selection for larger flowering stem diameter through both fruit number and ovule number per fruit in 2013 , resulting in strong integrated selection for larger stems . A similar pattern occurred in 2012 and for larger flowers in inland perennials in 2013 , though in both cases selection through fruit number was only marginally significant. Total integrated selection – Correlations among traits had diverse effects on the strength of selection . Positive correlations among traits increased the strength of selection in some cases. In fast-cycling annuals, for example, selection for increased flowering stem diameter caused correlated selection for larger flowers and later flowering in both years .

Similarly, positive correlations between flowering stem diameter and flowering time increased the strength of selection on both traits in robust annuals in 2012 . In other cases, correlations among traits opposed the direction of selection on individual traits, weakening the strength of selection overall. For example, selection for earlier flowering in robust annuals in 2013 was opposed by a positive phenotypic correlation between flowering time and flowering stem diameter combined with selection for larger flowering stems . Similarly, selection for larger seedlings in fastcycling annuals in 2013 was opposed by a negative correlation between seedling size and flowering time combined with selection for later flowering . In one case, multiple trait correlations opposed each other, effectively canceling out any effect of correlated selection on stem diameter in robust annuals in 2013. In this example, a positive correlation between stem diameter and flower size combined with selection for larger flowers was opposed by a positive correlation between stem diameter and flowering time combined with selection for earlier flowering . These opposite and nearly equal effects resulted in an estimate of total integrated selection very similar to the strength of integrated selection on stem diameter alone .Although divergent life history strategies within M. guttatus have become a model system for ecological genomics and speciation, the selective forces maintaining life history variation have not been well characterized in perennial populations. The adaptive significance of rapidly developing genotypes in seasonally drying habitats is intuitive, yet evidence that a perennial life history strategy is adaptive in habitats with greater water availability remains scarce. Although coastal perennials are locally adapted to coastal habitats relative to annuals, this appears to be largely mediated by increased tolerance to salt spray, and thus greater seedling survival and fruit set, rather than an advantage of perenniality per se . In this study, I directly compared the demographic performance of annual and perennial ecotypes within a montane perennial habitat and tested how specific transitions in the life-cycle contribute to differences in performance using LTREs.

With this approach, I found mixed evidence for local adaptation. A significant effect of ecotype on overall performance, combined with non-significant effects of population within ecotype, support that there are real differences in life history strategy and performance among these a priori ecotypes. I found some evidence that a perennial life history strategy is adaptive in a perennial habitat, since inland perennials consistently had the greatest demographic performance whereas fast-cycling annuals had the lowest. Further, transitions associated with adult survival and vegetative growth contributed positively to λ in the montane perennial ecotype, whereas fecundity transitions contributed negatively. However, there was no evidence that the montane perennial ecotype is locally adapted since inland perennials and robust annuals outperformed montane perennials in both years. One potential explanation for the relatively poor performance of themontane perennial ecotype is that the importance of below-ground rhizomes was not fully captured during the time-frame of this study. Indeed, montane perennial genotypes invest extensively in below-ground rhizomes and may be able to persist during harsh conditions when other strategies lacking below-ground structures would become extinct. Natural recruitment plots suggest that seedling recruitment is low relative to rosette recruitment. Longer-term monitoring of the relative performance of seedlings and rosettes under a range of environmental conditions, including drought, competition, and frost, would refine estimates of demographic performance. Alternatively, the low performance of the montane perennial ecotype could be due to genetic load, given extensive clonal growth and low flower production and seedling recruitment. Within the montane perennial ecotype, I did find some evidence for local adaptation as the native Eagle Meadows population outperformed both foreign populations, although this difference was not significant.As expected, life history ecotypes exhibit different fitness strategies even within a common montane environment. The performance of annual ecotypes is influenced most strongly by seed production, whereas perennial performance is also achieved through the growth and survival of clonal rosettes. Among perennials, montane populations produce few flowers and rely extensively on clonal growth through below-ground rhizomes. This divergence in the relative importance of specific fitness components influences the net pattern of selection. Interestingly, the direction of fecundity selection was similar across life history ecotypes, consistently favoring greater allocation to reproductive structures including larger flowers and larger reproductive stems. Conversely, selection through clonal growth, when detected, favored smaller reproductive stems and seedlings . These contrasting effects of individual fitness components resulted in contrasting selective landscapes for different life history ecotypes even within a common environment. In particular, integrated selection tended to be weaker in perennials relative to annuals due to the reduced elasticity of any one fitness component. A major goal of evolutionary biology has been to document the strength of selection in wild populations. In a review of mean-standardized selection estimates, Hereford found surprisingly strong directional selection through individual fitness components and suggested that these are unlikely to represent the strength of selection in general. One potential explanation is trade-offs among fitness components in the direction of selection, plastic pot manufacturers yet evidence for such trade-offs is mixed and based on relatively few studies . I only detected concurrent selection on the same trait through multiple fitness components for stem size and marginally for flower size. In these cases, selection through multiple fitness components increased the strength of integrated selection, because underlying selection gradients through fruit and ovule number were in the same direction. These traits are directly related to overall investment in reproductive biomass; such “size” traits may frequently be positively correlated and increase multiple components of fitness due to environmental covariance . I found some evidence for conflicting selection among fitness components in comparisons across ecotypes and years.

Later flowering consistently increased ovule production but decreased fruit production, whereas larger stems increased fecundity but decreased rosette production. However, I never detected these effects within the same ecotype and year, so it is unclear whether these trade-offs act simultaneously to decrease the strength of selection. Even if selection on a given trait primarily occurs through a single fitness component, life history divergence will still affect the strength of selection. Failing to integrate multiple fitness components will bias the strength of selection, since individual fitness components contribute unequally to population growth among environments and life history strategies . Integrating selection incorporates this variation by multiplying selection gradients by the elasticities of their respective fitness components. Since elasticities are necessarily less than one, integrated measures of selection will always be weaker than selection gradients, unless multiple fitness components are positively correlated. Thus, previous conclusions about the average strength of selection in wild populations are likely overestimates . The strength of selection gradients depends in part on the degree of fitness variation, however demographic reviews have found that the fitness components with the largest elasticities are often the least variable within populations . In this study, I found that strong selection gradients in perennials occurred through fitness components with smaller elasticities, weakening the strength of integrated selection relative to annuals. Large selection gradients may reflect less important fitness components, rather than strong phenotypic selection, and this potential bias should be greatest in long-lived iteroparous organisms where some fitness components will only contribute weakly to total fitness.Incorporating phenotypic correlations among traits frequently altered the strength of selection. Although phenotypic correlations determine the strength and pattern of selection, inference about the evolutionary response to selection requires an understanding of the underlying genetic variance-covariance structure. However, there are several lines of evidence to suggest that the phenotypic correlations observed here reflect genetic correlations within or among ecotypes. First, the randomized plot design used in this experiment minimizes the potential for environmental effects to generate trait correlations while the use of highly divergent ecotypes and multiple populations within each ecotype maximizes the potential for genetically based trait variation. Second, previous research has shed light on the genetic basis of many traits in this species. Below, I explore how selection through correlated traits may drive the evolution of perennial ecotypes in this species. Total integrated selection consistently favored larger flowers in this perennial habitat, potentially explaining the observation that perennial populations have larger flowers than annuals throughout the species range . Flower size has been shown to be a target of selection in several annual populations of M. guttatus . However, the direction of selection on flower size is both spatially and temporally variable due to positive genetic correlations between flower size, flower time, and ovule and pollen production per flower . Thus, in an annual habitat, large flowers decrease survival to reproduction by delaying flowering but increase seed set per flower, and the net pattern of selection on flower size depends on the timing of drought-induced mortality . Although I did not detect correlations between flower size and flowering time directly, flower size was frequently positively correlated with flowering stem diameter, which in turn was often correlated with later flowering. Further, selection on flower size only occurred through components of fecundity. This suggests that the longer growth season in perennial habitats ameliorates viability selection on flower size, resulting in strong fecundity selection for larger flowers.