Geographic and climatic drivers of reproductive assurance in Clarkia pulchella

Climate can affect plant populations through direct effects on physiology and fitness, and through indirect effects on their relationships with pollinating mutualists. We therefore expect that geographic variation in climate might lead to variation in plant mating systems. Biogeographic processes, such as range expansion, can also contribute to geographic patterns in mating system traits. We manipulated pollinator access to plants in eight sites spanning the geographic range of Clarkia pulchella to investigate geographic and climatic drivers of fruits production and seed set in the absence of pollinators (reproductive assurance). We examined how reproductive assurance and fruit production varied with the position of sites within the range of the species and with temperature and precipitation. We found that reproductive assurance in C. pulchella was greatest in populations in the northern part of the species’ range, and was not well-explained by any of the climate variables that we considered. In the absence of pollinators, some populations of C. pulchella have the capacity to increase fruit production, perhaps through resource reallocation, but this response is climate-dependent. Pollinators are important for reproduction in this species, and recruitment is sensitive to seed input. The degree of autonomous self-pollination that is possible in populations of this mixed-mating species may be shaped by historic biogeographic processes or variation in plant and pollinator community composition rather than variation in climate.

increases in reproductive assurance, higher water availability likely increases plant longevity and productivity during the flowering season. which is expected to experience warmer temperatures and redistributed rainfall in the coming decades 106 ( Figure 2). Temperature increases are expected to be especially prominent in the summer months (Wang 107 et al., 2012;Meyer et al., 2014). Anticipated changes in precipitation are variable and uncertain across the 108 range of our focal species, with many sites expected to experience decreases in summer precipitation, but 109 central sites projected to experience slight increases in annual precipitation (Wang et al., 2012;Meyer et al., 110 2014).  Table S1). Our original intention was to 116 treat the southern and western edges of the range separately and establish three sites at each edge. However, 117 due to difficulty finding populations of sufficient size in sites where we could also obtain permits, we used 118 just two populations in the west and one in the south. Because the climatic similarity among these sites 119 is nearly comparable to that among sites in other regions (Figure 2), we decided to treat them as a single 120 region, the southwest. At each site, 5-8 blocks containing four plots each were marked with 6-inch steel nails, 121 this resulted in a total of 50 blocks and 200 plots in the experiment. Each plot consisted of a 0.8 m 2 area.

122
Plots were intentionally placed with the goal of obtaining 5-20 individuals per plot, therefore the density 123 in plots was typically higher than the overall site density. Plots were placed closer to other plots in their 124 block than to those in other blocks (with exceptions in two circumstances where low plant density meant 125 very few suitable plot locations were available). Blocks were placed to capture variation in microhabitat 126 characteristics across the site, and their spacing varied depending on the population size and density. Each 127 plot was randomly assigned to one of four factorial treatment groups: control, water addition, pollinator 128 exclusion, or both water addition and pollinator exclusion. Plots receiving water additions were at least 0.5 129 m away from unwatered plots, except when they were downslope from unwatered plots, in which case they 130 were sometimes closer. Plots receiving pollinator exclusion treatments were tented in bridal-veil mesh with 131 bamboo stakes in each corner and nails tacking the mesh to the ground. Some pollinator exclusion plots had 132 their nets partially removed by wind or cows during the flowering season (n = 13 out of 100 total tented 133 plots), so all analyses were performed without these plots.

134
The majority of the summer precipitation in these sites falls in summer storms. Plots receiving supple-135 mental water were watered 1-2 times during the summer (when plants were flowering) to simulate additional 136 rainfall events. During each watering event, 15 mm of water was added to each plot (9.6 L per plot). This 137 approximated the typical precipitation of a summer rainfall event based on data from Wang et al. (2012), and 138 in an average year, would have increased the total summer precipitation in these plots by 30-70%. However, 139 our experiment was conducted during a drought year (Figure 2), therefore, in the central sites, plots receiving 140 water additions still fell short of average summer precipitation levels. In southwestern and northern sites, 141 the water addition likely raised the summer precipitation amount slightly above the historic average. In all 142 sites, we consider the water additions to represent a drought relief treatment, because unwatered plots were 143 already experiencing natural drought. The first watering was performed when the experiment was set up.

144
The second watering was performed 22-25 June 2015, except at two sites (SW3, C1), which had completed 145 flowering and fruiting at that time. Efficacy of the water addition treatment was checked by measuring the 146 soil water content with a probe (Hydrosense, Campbell Scientific Inc.) before and after water additions.

147
Prior to water additions, there were no significant differences between plots receiving a water addition treat-148 ment and those not receiving this treatment (linear mixed effects model with a random effect of site and 149 a fixed effect of water addition treatment; first watering: P = 0.839 (7 of 8 sites were measured); second 150 watering: P = 0.277 (5 of 8 sites were measured)). Shortly after watering (within one hour), plots receiving 151 a water addition treatment had higher soil moisture than those not receiving treatment (first watering: P 152 < 0.0001, average soil moisture of unwatered plots = 11.0% , watered plots 22.2%; second watering: P = 153 0.0001, average soil moisture of unwatered plots 3.7%, watered plots 11.5%). plot ranged from 1-43 (mean = 7.9, median = 7). We counted the number of fruits per plant on every plant in each plot, as a proxy for the number of flowers per plant (aborted fruits were rare overall). Plants that had died before producing any flowers were not included in our analyses. Some plants (n = 14, 0.7% of all 159 plants counted) had experienced major damage prior to our final census making fruit counting impossible, 160 so they were assigned the average number of fruits per plant in that plot type at that site for estimation of 161 plot-level seed input, but we excluded them from analyses of fruit counts. Other plants (n = 25, 1.4% of 162 all plants counted) still had flowers at the time of the final census. It was assumed that these flowers would 163 ripen into fruits, so they were included in the fruit counts. When possible, up to four fruits per plot (average 164 number of fruits per plot = 3.67) were collected for seed counting. After counting, seeds were returned to 165 the plots that they were collected from by sprinkling them haphazardly over the plot from a 10 cm height.

166
In 3 of 200 plots, no intact fruits were available for seed counting (all had dehisced), so these plots were 167 excluded from analyses of seed set and plot-level seed input, but included in analyses of fruit counts. To  Climate variable selection 172 We expect long-term climatic conditions, particularly those that might contribute to drought stress, to 173 influence selection for autonomous selfing. Concurrent work with C. pulchella (M. Bontrager, unpublished 174 data) has indicated that fall, winter, and spring growing conditions play a large role in overall plant growth 175 and reproductive output, therefore we considered not only flowering season (June-July) climate variables but 176 also annual temperature and precipitation for inclusion as predictors. We obtained 50-year climate normals 177 (1963-2012) from ClimateWNA (Wang et al., 2012) and climate data during the study from PRISM (PRISM  Table S2.

183
Statistical analyses 184 We used generalized linear mixed effects models (GLMMs) to evaluate the effects of pollinator exclusion, 185 region, and each of the selected climate variables on reproductive assurance and fruits per plant. Initial 186 data exploration indicated that our watering treatment did not have a strong or consistent biological effect 187 ( Figure S1, Figure S2), so we omitted this factor from our analyses to keep models simple and facilitate interpretation of interactions between the other factors. For each predictor variable of interest (the four 189 climate variables and region), we built a model with a two-way interaction between this variable and pol-190 linator exclusion on both seed counts and fruit counts. We used negative binomial GLMMs for both seeds 191 and fruits, and we included a zero-inflation parameter when modeling seed counts. In all models we included 192 random effects of blocks nested within sites. Because our data do not contain true zero fruit counts (i.e., 193 we did not include plants that did not survive to produce fruits, so all plants in our dataset produced at 194 least one fruit), we subtracted one from all counts of fruits per plant prior to analysis in order to better In all regions, Clarkia pulchella produced fewer seeds in the absence of pollinators (Table 1). Climatic or 204 geographic drivers of variation in reproductive assurance were indicated by our models of seeds per fruit 205 when there was a significant interaction between pollinator exclusion and region or pollinator exclusion and 206 a given climate variable. We found that reproductive assurance varied by region, with greater rates of 207 reproductive assurance in northern populations ( Figure 3, Table 1). We did not find any strong effects of 208 climate on seed production or reproductive assurance (Table 1). However, there was a marginally significant 209 interaction between mean annual precipitation (MAP) and pollinator exclusion: populations in historically 210 wetter sites tended to be more negatively affected by pollinator exclusion (i.e., populations in drier sites 211 had slightly higher rates of reproductive assurance) (Table 1). This could be a causal relationship, or the 212 correlation could have been driven by the high degree of reproductive assurance in the northern part of the 213 range, which has low MAP. If low MAP was really a driver of reproductive assurance, we might expect to 214 have seen a greater degree of reproductive assurance in the southwestern sites, which also have low MAP.

215
However, this was not the case in our data.

216
Response of patch density to seed production in the previous year  (Table 2). This effect was small-plants 229 in plots without pollinators produced an additional 0.4 fruits, on average. 230 We found that the effects of pollinator exclusion on fruit production depended upon the amount of summer 231 precipitation during the experiment (Table 2, Figure 5). Fruit production was higher in wetter sites, and 232 pollinator-excluded plants that were in the wettest sites showed a greater positive effect of pollinator exclusion 233 on fruit production (Table 2). However, it should be noted that while both the main effect of climate and 234 its interaction with pollinator exclusion were significant, the difference between plots with and without 235 pollinators in wetter sites did not appear to be particularly strong, and when visualized the confidence 236 intervals were largely overlapping ( Figure 5A). We also found a main effect of mean annual temperature 237 (MAT) on fruit production (Table 2). Fruit production was higher in cooler sites ( Figure 5B). Disentangling 238 these two climatic drivers of increased fruit production is not possible with this dataset, however, because 239 summer precipitation during the experiment was negatively correlated with normal MAT. Therefore, it could 240 have been either higher water resources during flowering or cooler temperatures over the growing season that 241 resulted in increased fruit production. It is worth noting, however, that summer temperature during the 242 experiment was not correlated with either of these variables, so if temperature was the driver of this pattern, 243 it was likely because of temperature effects on earlier life-history stages.

Discussion
Pollinator exclusion in eight populations of Clarkia pulchella revealed increased autonomous reproductive 246 assurance in populations in the northern part of the species' range, as compared to the center or southwest.

247
Plants in the northern part of the species' range also produced more fruits. Fruit production was higher 248 in sites that are cooler or that received higher amounts of precipitation during the experiment. Plants 249 also produced slightly more fruits in response to pollinator exclusion, however, this reallocation was not, in 250 general, large enough to offset the reduction in seed production caused by pollen limitation.

251
Reproductive assurance is driven by geography rather than climate 252 Pollinator limitation reduced reproduction across the range of C. pulchella. Contrary to our prediction, 253 we did not observe plastic responses of decreased reproductive assurance in response to our water addition 254 treatment, or in sites with high summer precipitation during the experiment. There is some indication that 255 plants in sites with lower average precipitation may have adapted to have greater reproductive assurance 256 (Table 1), perhaps due to shorter season lengths or because gradients in pollinator abundance may be 257 driven by water availability. However, increased reproductive assurance is only apparent at the northern 258 range edge (Figure 3) despite the fact that mean annual precipitation is lower at both the northern and 259 southwestern range edges. This trend persists even after accounting for regional differences in seed set in 260 control plots, i.e., when reproductive assurance is represented as a proportion of the average seed set in 261 control plots (data not shown). In light of this, we suggest that for this species, reproductive assurance is   Table S1.   confidence intervals; open triangles are raw means of the data. Despite statistically significant interactions between pollinators and water addition, as well as between region and water addition (analyses not shown), the biological effect of watering on fruit number appears negligible.