Shifts in vernalization and phenology at the rear edge hold insight into adaptation of temperate plants to future milder winters

Campanula americana, climate change, clines, flowering time, phenology, rear edge, 46 reproductive cues, vernalization. 47 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 3

48 Organisms exposed to cyclic al environmental changes often evolve mechanisms to sense these 49 fluctuations and time developmental shifts to occur under favorable conditions (Preston & Sandve, 50 2013). In temperate plants, vernalization, the prolonged exposure to non -lethal winter cold 51 (Chouard, 1960) , serves as an important cue so that key life -cycle transitions will occur after 52 winter, e.g. vegetative growth and reproduction (Amasino, 2005). However, relying on such cues 53 may be detrimental under rapidly changing environments . This is of particular concern in the 54 context of ongoing global warming, as it is expected to affect the strength and timing of seasonal 55 temperatures experienced by natural and agricultural species (Willis et al., 2008; Blackman, 2017). 56 Species requiring vernalization will likely be negatively affected by climate warming as shorter 57 and milder winters become more common (Luedeling et al. , 2011; Anderson, 2023) . Such 58 conditions are expected to lead to a reduction in reproduction (Padhye & Cameron, 2009; Liu et 59 al., 2012; Satake et al., 2013) or shifts in reproductive phenology outside of optimal time-windows 60 (Fitter & Fitter, 2002; Parmesan & Yohe, 2003; Faidiga et al., 2023; Geissler et al., 2023). It is 61 therefore critical t o understand response to seasonal cuing , and variation in this response , to 62 determine possible consequences of expected warmer climates for temperate plant species. 63 Vernalization requirements and cue response s often vary within and between species in 64 response to differences in winter condition (Andrés & Coupland, 2012; Blackman, 2017; Preston 65 & Fjellheim, 2022) . This suggests that while these mechanisms are crucial to complete the life 66 cycle, they vary enough to allow persistence across heterogeneous environments . Studies have 67 explored variation in cueing reproduction across environments by testing for clines in vernalization 68 requirements and phenological response to temperature gradients (Blackman, 2017). This body of 69 work provides evidence of a reduction in vernalization requirements with increases in temperature 70 (Wesselingh et al., 1994; Dijk et al., 1997; Boudry et al., 2002; Stinchcombe et al., 2005; Jokela 71 et al., 2015). Reproductive phenology also varies over temperature gradients though a general 72 pattern is less clear. For example, flowering is later with the later arrival of spring-like conditions 73 at high latitudes than lower ones in Arabidopsis thaliana (Stinchcombe et al., 2004; Lempe et al., 74 2005). In contrast, earlier flowering occurs at higher latitudes in other species (e.g. Paccard et al. 75 2014; Vest and Sobel 2021) . These studies reveal the potential for both the mechanism of 76 environmental cuing as well as the pattern of reproductive phenology to evolve to track warming 77 climates. 78 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 4 Populations at the warmer limits of temperate species ranges are particularly important for 79 understanding adaptation to shorter winters. This part of the range, colloquially known as the “rear 80 edge” (Hampe & Petit, 2005), often consists of relict populations that have persisted more or less 81 in place at least since the last glacial maximum (LGM, c. 23 000–19 000 years ago; Hughes et al. 82 2013). Long-term persistence at the warmer range limit over multiple glacial cycles and under 83 continuously warming climates after the LGM has likely resulted in adaptation to warmer climates. 84 With this history, the rear edge may provide insight into how species requiring vernalization adapt 85 to mild winters, and which adaptations may be best suited to future warming climates. However, 86 these populations are typically not sampled in studies investigating clines and therefore our 87 knowledge of phenology and vernalization requirements in populations from the warmest habitats 88 is limited (but see Vest and Sobel 2021). 89 Here we investigate whether rear-edge populations show patterns of differentiation 90 consistent with adaptation to milder winters compared to populations from elsewhere in the range 91 in the North American herb Campanula americana . This species requires vernalization , but 92 occupies a wide latitudinal and temperature gradient, with ancestral rear -edge populations 93 occurring in distinctively warmer climates in the southern parts of the range . To test for 94 differentiation between the rear edge and the rest of the range , we first characterized differences 95 in flowering phenology across latitudes using range-wide observations in natural population s 96 gathered from citizen science data, and then tested whether this variation reflects differences in 97 winter or growing season climate (Experiment 1). We also assessed genetic differen ces in 98 phenology by testing for variation in flowering among populations from across latitudes raised 99 under common greenhouse conditions (Experiment 2) . We then evaluated plasticity in 100 reproductive phenology in response to a range of natural growing season cues by raising 101 populations sampled across latitudes in common gardens along a latitudinal gradient (Experiment 102 3). Finally, we tested differences in vernalization requirement and phenological plasticity in 103 response to these winter cues by raising populations under experimentally manipulated 104 vernalization length (Experiment 4). In total, these four experiments allow us to comprehensively 105 test for differences in phenology and its regulation across the range of this species. Results of these 106 experiment will allow us to understand not only the mechanisms that underly how rear-edge 107 populations have adapted to their distinct habitats, but will also shed light on how populations 108 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 5 across the range vary in their response to milder winter, and ultimately the mechanisms that may 109 evolve to allow persistence under future conditions. 110 111

112 Study system 113 Campanula americana is a monocarpic herb found in open forests, clearings and forests margins 114 across the eastern USA (Fig. 1a). Seeds germinate in spring or fall, vernalize over winter as rosettes 115 and then bolt, elongating into a reproductive stalk. Following several months of growth, flower 116 buds develop and the plants transition to flowering in mid -summer, with flowers open ing 117 progressively over several weeks. Most populations require at least six weeks at 4° C for successful 118 vernalization that cues reproduction (Baskin & Baskin, 1984; Etterson & Galloway, 2002) , but 119 southernmost populations may not always require vernalization to bolt (Kalisz and Wardle 1994). 120 This species occupies a large range , spanning ~15 ° latitude over ~1800km with a 121 difference of ~22 °C in minimum annual temperature (Fig. 1a; bio6, inferred from WorldClim 2.0, 122 Fick and Hijmans 2017) . The species comprises three geographically distinct genetic clades 123 (Barnard-Kubow et al., 2015). Here we focus on the largest clade, found west of the Appalachian 124 Mountains, that covers most of the range . We limit ed sampling to low elevation s (<600m) to 125 reduce confounding climate gradients across elevation and latitude. This “Western” clade is 126 characterized by a history of persistence in glacial refugia in the southern part of the current range 127 during LGM, followed by postglacial range expansion (Barnard-Kubow et al., 2015; Koski et al., 128 2019; Prior et al., 2020). We define the “ rear edge” as the lower latitudinal third of the range 129 (below 35° N, Fig. 1 a). This part of the range largely overlaps with putative glacial refugia 130 (Barnard-Kubow et al., 2015). It also has a climate defined as subtropical , characterized by mild 131 winters and rare freezing events (American meteorological society, 132 https://glossary.ametsoc.org/wiki/Subtropics), distinct from the continental temperate climate in 133 more northern parts of the range. 134 135 Experiment 1: Variation in flowering phenology in natural populations 136 We first tested whether rear -edge populations in C. americana differ from the rest of the range 137 based on flowering phenology observed in natural populations across latitudes, and then explore 138 associations between phenology and both winter and growing season climates. 139 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 6 140 Inference of flowering phenology in natural populations 141 We inferred flowering phenology of natural populations from the publicly available repository 142 iNaturalist (https://www.inaturalist.org. Accessed 23/02/2023). We extracted all 10,250 records of 143 Campanula americana observed between 2018 and 2022 (earlier records did not cover the 144 distribution well). This set was reduced to 335 observations following pruning to select for low 145 elevation populations, reduce duplicates and oversampling close to cities, and reduce bias of fewer 146 occurrences at the rear edge (Supporting method S1.1, Fig. S1.1). We inferred the day of first 147 flower (recorded as day of the year) based on the photograph provided with each observation 148 (Table S1.1, Fig. S1.2), resulting in a final dataset of 238 observations (Fig. 1a). 149 150 Climatic variables 151 For each observation, we obtained daily mean, maximum, and minimum temperature and daily 152 precipitation for the year of observation and the year before from the PRISM database 153 (https://prism.oregonstate.edu, accessed 01/09/2023) using the prism package (Edmund & Bell, 154 2015) in R (R Core team, 2024) . We then generated 17 variables from this data that capture the 155 climate at key life cycle stages, to test associations of phenology with climate (Table S1.2a). 156 157 Statistical analysis 158 We first explored how the day of first flower varied across latitudes. To do this, we compared three 159 mixed-effect models with latitude as a continuous fixed effect and the year of observation as 160 random effect. The first model tested for a linear relationship between the day of first flower and 161 latitude, the second tested a second-degree quadratic relationship, and the third model tested a 162 piecewise linear relationship with one breakpoint. The latter two allow clines with latitude to vary 163 across the range . All analyses were performed in R. Model parametrizations are provided in 164 Supporting method S1.2a. Model fits were compared using the corrected Akaike’s information 165 criterion (AICc, Sugiura 1978) using MuMin (Bartoń, 2023). From this we identified a breakpoint 166 in the relationship between the day of first flower and latitude at ~35 °N (see Results), that we 167 incorporated into the design of subsequent experiments. 168 We then assessed whether the observed variation in the day of first flower was associated 169 with environmental variables. We first tested which climatic variable best discriminated between 170 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 7 the part of the range below or above the breakpoint in latitude by performing a discriminant 171 analysis of principal components (DAPC) using the dapc function in adegent (Jombart, 2008; 172 Jombart et al., 2010). DAPC was performed on all 17 climate variables (transformed by PCA, 8 173 PC retained) with the grouping factor of whether observation s were located at latitudes above 174 (group 1) or below (group 2) the predicted breakpoint. We then evaluated whether identified 175 climate variables changed across the range in mixed effect models that included the latitude of 176 each observation as a continuous fixed effect, range position of each observation relative to the 177 estimated breakpoint as a categorical effect , and the ir interaction. The year of observation was 178 included as a random effect (model in Supporting method S1.2b). We also tested the effect of 179 latitude on the day of first flower using the same model structure. 180 181 Experiment 2: Variation in reproductive phenology in a common environment 182 We tested for genetic differences in flowering phenology between C. americana’s rear edge and 183 the rest of the range by raising populations from across latitudes under a common environment in 184 a greenhouse experiment. 185 186 Experimental design 187 We raised two cohorts of C. americana populations that cover a similar latitudinal breadth 188 below (~4.4° N) and above (~6.6° N) the breakpoint identified in Exp. 1. For the first cohort, we 189 collected seeds in 23 natural populations (Fig. 2a, Table S2.1) in late summer of 2020 and 2021 190 and stored them by maternal plant at 4 °C under dark and dry conditions . For the second cohort, 191 we used greenhouse-reared seeds from 25 populations (Supporting method S2.1), 20 from the first 192 cohort and five additional populations to yield better sampling (Fig. 2 a). Greenhouse produced 193 seeds have less variation due to maternal effects and collection date . We raised 544 s eedlings in 194 the first cohort (Fall 2021 ), and 535 in the second cohort (Fall 2022 ) in similar conditions 195 (Supporting method S2.1). Conditions were designed to mimic the natural growth cycle of the 196 species, starting with germination and vegetative growth in growth chambers simulating fall , 197 followed by vernalization in a cold room simulating winter, and then bolting and reproduction in 198 a greenhouse with conditions simulating summer. 199 200 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 8 Recording of traits and statistical analysis 201 We recorded the date of bolting (presence of a stem) twice a week for the first cohort and weekly 202 for the second . We recorded the date of first flower three times a week for both cohorts. Most 203 plants that bolted also flowered (>98%). 204 Based on this data, we generated four traits related to reproduction initiation and 205 phenology: flowering success (binary, production of at least 1 flower), time to flowering (days 206 from vernalization to flowering) and its components time to bolting and time between bolting and 207 flowering (Table S2.2a). Each was analyzed at the individual level in mixed effect models in R, 208 with population latitude as a continuous fixed effect, cohort a categorical effect, and their 209 interaction. Population and seed family nested within population were random effects (models in 210 Supporting method S2.2). Similar to Exp. 1, we initially tested whether latitude was best described 211 as a linear relationship or a second-degree quadratic relationship for each trait. This was done by 212 comparing model perform ance based on AICc. We report results only for the best performing 213 model (all results Table S2.3). For this model, we performed a post-hoc test to estimate the slope 214 of the relationship between latitude and reproductive traits within each cohort, cohort effects and 215 the difference in the slope between cohorts using the package emmeans (Lenth, 2019). 216 217 Experiment 3: Variation in reproductive phenology in a transplant experiment 218 We tested for differences in phenology in response to spring and summer conditions between the 219 rear edge and the rest of the range by raising populations from across the range in common gardens 220 established on a latitudinal gradient. 221 222 Experimental design 223 We selected 25 populations with similar distribution as Exp. 2 (Fig 3a, Table S2.1), and approx. 224 15 seed families per population ( Table S2.1), to be raised in five common gardens . We used 225 greenhouse-reared seeds for 20 of the populations and field-collected seeds for five . For each 226 garden, we sowed two seeds in each of two pots per seed family (three seeds for field -collected 227 seeds, one if few seeds) , thinning to one plant per pot after germination . This gave 2 228 individuals/family x 15 families/population x 25 populations = 750 individuals per garden, 3750 229 total. Seeds were planted in fall 2022, germinated for six weeks in growth chambers and vernalized 230 for seven weeks in a cold room (Supporting method S2.1). 231 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 9 Seedlings were acclimated to outdoor conditions for one week (Charlottesville, VA, USA), 232 and then transplanted to one of five common garden sites (CG, Fig. 3a, Tables S2.4): Lexington 233 KY (CG1, 38.1° N), Blaine TN (CG2 , 36.2° N), Clemson SC (CG3, 34.7° N), Akron AL (CG4 , 234 32.9° N), and Tallahassee FL (CG5 , 30.5° N). These sites span an 850 km latitudinal gradient in 235 roughly 2° steps and reflect the temperature gradient within the sampled range. Transplant for each 236 site occurred when mean daily temperature was consistently above 10°C in spring, estimated using 237 PRISM daily mean temperature 2018 to 2022. Sowing was timed for each site such that 238 vernalization ended one week before the projected transplant date. 239 Garden locations were chosen to mimic the natural habitat of the species. In each site, 240 individuals were transplanted in 30 spatial blocks of 25 individuals, with one individual per 241 population per block . Sites were raked to reduce initial competition. Individuals were then 242 transplanted into the exposed ground with the substrate they were raised in and watered once. Each 243 site was surrounded by a 2m high fence to exclude large herbivores. No further intervention was 244 performed. One data logger (iButton®, Maxim Integrated Products, Inc) was placed in each site 245 in the shade, 1.5m above ground, to monitor air temperature every hour for the length of the 246 experiment. This data revea led that plants experienced additional days of vernalization after 247 transplant (Table S2.4a), but less so in the southernmost garden. The Experiment was stopped 153 248 to 170 days after transplant (Table S2.4b), when most plants had reached fruit maturation but seeds 249 were not yet dispersing. Unfortunately, plants in CG4 experienced very low survival (<10%), so 250 this site was removed from analysis. 251 252 Recording of trait and statistical analysis 253 Data was collected three times , about a month apart , in the three northern sites (GC1-GC3), and 254 only two times for the southern site (GC5). In each round of data collection, we recorded survival, 255 transition to bolting and to flowering (here production of at least one bud) and inferred the 256 flowering date as in Exp. 1 (Table S2.5). 257 Based on this data, we generated three binary traits at the plant level: transition to bolting 258 and transition from bolting to flowering estimated at each time point, flowering success based on 259 the first time point to provide a “snapshot” of flowering phenology. We estimated day of first 260 flower (day of the year), as well as time to flowering since transplant (days from transplant) (Table 261 S2.2b). The effects of latitude (of population origin), common garden (categorical, four sites) and 262 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 10 their interaction on each of these traits was analyzed in mixed effect models ( Supporting method 263 S2.2). Preliminarily analysis tested whether latitude was best described by a linear or a quadratic 264 relationship, and results are only reported for the best model (Table S2.3). If models performed 265 equally well, the quadratic one was retained for ease of comparison among traits. Post-hoc tests 266 were performed for the best models. 267 268 Experiment 4: Variation in vernalization requirements and response across the range 269 We tested for differentiation in vernalization requirement and in phenological response to 270 vernalization length between the rear edge and the rest of the range by subjecting populations from 271 across latitudes to experimentally manipulated vernalization length. 272 273 Experimental design 274 We selected 12 populations with greenhouse-reared seeds, representing the same latitudinal range 275 as Exp.2 (Fig. 4a, Table S2.1). We selected approx. 15 families per population (Table S2.1), and 276 sowed two seeds in each of nine pots per seed family, resulting in 45 individuals per population 277 after thinning to one seedling per pot. Pots were then randomly assigned to one of three 278 vernalization treatments (see below), resulting in 3 individuals/family x 15 families/population x 279 12 populations = 540 individuals per treatment, 1620 total. Seedlings were raised from germination 280 in fall 2022 to flowering in 2023 (Supporting method S2.1), but with the difference that treatments 281 differed in vernalization length, lasting either six, four or two weeks. Sowing was timed such that 282 all three treatments finished vernalization at the same time (Fig. S2.1). 283 284 Recording of trait and statistical analysis 285 We recorded survival, bolting and flowering three times a week during the greenhouse phase. 286 Bolting was recorded for 52 days after vernalization and flowering for 141 . We then estimated 287 flowering success and its two components transition to bolting and transition from bolting to 288 flowering, as well as time to flowering and its two components time to bolting, and time between 289 bolting and flowering (Table S2. 2a). Similar to Exp. 2, the effects of latitude, vernalization 290 treatment (categorical with three levels) and their interaction on each of these traits was analyzed 291 in mixed effect models (Supporting method S2.2). For the three phenology traits, the two -week 292 treatment was excluded as few individuals bolt ed. Preliminarily analysis tested whether latitude 293 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 11 was best described by a linear or a quadratic relationship (Table S2.3), results are reported for the 294 best model and post-hoc tests were performed. 295 296

297 Experiment 1: Variation in flowering phenology in natural populations 298 In natural populations, the relationship between the day of first flower and latitude was best 299 described as a quadratic or piecewise relationship (Table S1. 3, Fig S1. 3), with both models 300 performing equally well (ΔAICc < 2). The earliest f lowering was at 36.64 ° N and 34.98 ° N 301 respectively, with flowering at increasingly later dates towards low er and high er latitudes. 302 Subsequent analyzes were parameterized with the breakpoint from the piecewise model because it 303 matched best our a priori delimitation of the rear edge (~35 °N), though use of the minimum from 304 the quadratic model yielded similar results (not shown). Climates of populations north and south 305 of this breakpoint were best discriminated by the length of vernalization as well as the time 306 between the start of the growing season and flowering (Table S1.2b). The other climate variables 307 had negligeable contributions. 308 The day of first flower and the two climate variables (length of vernalization, time between 309 the start of the growing season and flowering) showed strong clines across latitudes, with the slope 310 depending on their position relative to the breakpoint ( L*BP, Table 1). Flowering was earliest at 311 the breakpoint, with much later flowering per unit of latitude toward low latitudes (26 days over 312 ~4.5 °) than towards high latitudes (18 days over ~9.9 °, Table 1, Fig. 1 b). The length of 313 vernalization increased with latitude (Table 1, Fig. 1c), but days of vernalization accumulated more 314 rapidly north of the breakpoint than to the south , where vernalization was limited. The time 315 between the start of the growing season and flowering was substantially shorter and relatively 316 insensitive to latitude to the north (65 to 51 days predicted), whereas it increased sharply to the 317 south of the breakpoint (62 to 130 days predicted; Table 1, Fig. 1d). 318 319 Experiment 2: Variation in reproductive phenology in a common environment 320 Rear-edge populations showed differentiation in reproductive phenology from the rest of the range 321 when raised in a common controlled environment. Populations from low latitudes had greater 322 flowering success but flowered later relative to those in the rest of the range (Table 2, Fig. 2, Table 323 S2.6). Populations from the highest latitudes also flowered later than central populations, but this 324 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 12 was weaker than in low latitude populations. Later flowering was due to a delay in time to bolting 325 in northern populations, but a longer developmental period between bolting and the opening of the 326 first flower in southern populations (Table 2, Table S2.6, Fig. S2.2). Patterns across latitudes were 327 consistent between cohorts, though a greater proportion of plants flowered in the second cohort. 328 329 Experiment 3: Variation in reproductive phenology in a transplant experiment 330 Rear-edge populations showed strong differentiation from the rest of the range in the common 331 gardens, indicated by delayed r eproductive phenology relative to those from more central sites. 332 Southern populations experienced smaller flowering success (except CG5), due to a reduction in 333 both transitions to bolting and from bolting to flowering (Table 2, Fig. 3b, Tables S2.7, Fig S2.3a, 334 b). These populations also flowered later, both in day of first flower and time to flowering since 335 transplanting (Table 2, Fig. 3 c, Tables S2. 7, Fig S 2.3c). Populations at high latitudes also had 336 smaller flowering success and a moderate delay in phenology compared to central populations in 337 all gardens, but differences were weaker than in low latitude populations (except CG5). 338 Patterns of trait variation among southern populations were generally consistent across 339 garden sites despite differences in spring and summer climate (Fig. 3) . In contrast, northern 340 populations exhibited plasticity in flowering phenology across gardens. For example, in the most 341 southern garden (CG5), northern populations showed a stronger delay in phenology relative to 342 central populations than in any of the other sites (Table 2, Fig. 3, Tables S2.7, Fig S2.3). 343 Variation in phenology across gardens was generally linked to transplant time. The 344 phenology “snapshot”, flowering success, was significantly smaller in northern sites compared to 345 southern sites (except for CG1 vs CG2; Fig. 3b, Table S2.7b), indicating a delay in flowering in 346 northern sites. In line, the day of first flower generally occurred later in northern sites compared 347 to southern sites ( except for CG1 vs CG2; Table 2, Table S2.7b, Fig. S 2.3c). However, these 348 differences were almost eliminated when flowering was measured relative to transplant date (Fig. 349 3c, Table S2.7b). 350 351 Experiment 4: Variation in vernalization requirements and response across the range 352 Vernalization requirements of rear-edge populations were strongly differentiated from the rest of 353 the range , characterized by a reduced need for vernalization and smaller plastic response to 354 variation in vernalization length . In keeping with this pattern, there was a decrease in flowering 355 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 13 success in populations from higher latitudes that was exacerbated when vernalization was reduced 356 (Table 2, Fig. 4b, Tables S2.8). In northern populations, a reduction in vernalization from six 357 weeks to two weeks yielded a large reduction in flowering success. In contrast, southernmost 358 populations had high flowering success regardless of vernalization treatment. Most of this 359 variation was due to variation in transition to bolting among populations (Table 2, Tables S2.8, 360 Fig. S2.4a). 361 Differentiation in response to vernalization was also observed for phenology, with rear -362 edge populations again showing smaller plastic response to variation in vernalization. In the six-363 week treatment, populations from low latitudes flowered later compared to more central 364 populations (Fig. 4c). Reducing vernalization led to a general delay in time to flowering except for 365 in the southernmost populations which changed little (Fig. 4c, Tables S2.8). Differences in time to 366 flowering across treatments were mainly due to changes in the time between bolting and flowering, 367 rather than changes in time to bolting (Table 2, Tables S2.8, Fig. S2.4). 368 369

370 Understanding how relict rear-edge populations at the warmer range limits of temperate species 371 differ from younger populations at higher latitudes can help us predict how populations across the 372 range may respond and adapt to future climates . We found that in the North American herb 373 Campanula americana, rear-edge populations occurring in the lower latitudinal part of the range 374 have evolved to rely less on winter cues to regulate phenology compared to the rest of the range. 375 This suggests that critical life cycle mechanisms, such as the regulation of phenology, may evolve 376 rapidly to adapt to changing environments. In addition, contrary to classic expectations, we found 377 the warmer range limit may be less affected by climate change’s mild winters compared to higher 378 latitude regions of the range. Finally, our results suggest that northern populations will suffer losses 379 in reproduction with a reduction in winter cues although, over time, they may adapt to these 380 changes similarly to the rear edge. 381 382 Natural rear-edge populations differ in phenology from the rest of the range 383 In natural C. americana populations, timing of flowering had opposing clines across latitudes 384 (Exp. 1, Fig. 1b). The breakpoint of the clines was ~35 °N, coinciding with our a priori definition 385 of the rear-edge (below) relative to the rest of the range (above). The day of first flower was the 386 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 14 earliest at the breakpoint , becoming progressively later towards high er and low er latitudes. 387 Previous work has found l atitudinal clines in f lowering are most often continuous in temperate 388 plants (e.g. Stinchcombe et al. 2004). These clines are thought to reflect continuous differences in 389 cues such as photoperiod or vernalization used to signal the onset of favorable growing seasons 390 (Preston & Sandve, 2013; Blackman, 2017; Preston & Fjellheim, 2022) . However, r ear-edge 391 populations have typically been neglected in such studies , so we don’t know how common the 392 opposing clines seen here are. Opposing clines suggest a divergence in the response to seasonal 393 cues, or a change in the cues used to regulate phenology between the rear edge and the rest of the 394 range. 395 396 Most of the range shows adaption to winter conditions typical for temperate plants 397 We found that differentiation in the day of first flower between rear-edge populations and the rest 398 of the range was explained by differences in the length of vernalization (Exp. 1) , suggesting 399 vernalization may serve as environmental cue for flowering. In nature, populations occurring 400 above 35 °N experience increasing winter length towards higher latitudes (Fig. 1c). These longer 401 winters were associated with an increase in vernalization required for reproduction in both 402 greenhouse experiments (Exp. 2, 4). Fewer plants flowered from high-latitude populations exposed 403 to short vernalization periods, and this response was strongest in the northernmost populations 404 which had a complete loss in reproduction under limited vernalization . A longer vernalization 405 requirement in high latitudes is common for temperate plant taxa (Wesselingh et al., 1994; Boudry 406 et al., 2002; Jokela et al., 2015), and may reflect an adaptation to prevent premature flowering in 407 areas with long winters and detrimental early spring frost events (Inouye, 2000). 408 Differentiation in the day of first flower across latitudes was also explained by the time 409 between flowering and the start of the growing season. Above 35 °N, flowering in nature occurred 410 somewhat earlier relative to the onset of the growing season with increasing latitudes (ca. 14 days 411 earlier across ca. 10° of latitude to the north , Exp. 1, Fig. 1d). This is also common among 412 temperate plants (e.g. Paccard et al. 2014; Vest and Sobel 2021) , and is thought to be associated 413 with adaptation to complete reproduction under shorter growing seasons (Griffith & Watson, 414 2005). A similar shift in phenology was reported along an elevational gradient in C. americana, 415 where high elevation populations with shorter growing seasons had accelerated reproduction 416 compared to low elevation populations (Haggerty & Galloway, 2011). 417 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 15 In the manipulative experiments, population s from locations north of the rear edge 418 exhibited plasticity in flowering phenology in response to vernalization length . While generally 419 flowering slightly later than central populations, they flowered even later when vernalization was 420 short (Exp. 4, Fig. 4c) or when raised in southern climates (Exp. 3, Fig. 3 c), with northernmost 421 populations exhibiting the strongest delay. This plasticity of later flowering under shorter 422 vernalization may simply be a passive byproduct of inadequate cuing . Specifically, limited 423 vernalization may partially cue the onset of reproduction, therefore resulting in a longer transition 424 to reproduction. The larger plasticity in timing of flowering in northern populations may be due to 425 greater vernalization requirements, increasing the mismatch between required and experimental 426 vernalization length . Indeed, the shortfall for vernalization for northern populations i n the 427 manipulative experiments may have masked earlier flowering relative to the start of the growing 428 season as observed in nature. Alternatively, greater plasticity in the northernmost populations 429 could be that this plasticity has evolved to modulate reproductive phenology in response to thaws 430 in late winter, extending the period prior to flowering in short winters and thereby reducing 431 negative effects of these “false springs.” 432 433 Winter cue requirements and responses are shifted at the rear-edge where winters are milder 434 Rear-edge populations showed distinct patterns of reproductive phenology compared to the rest of 435 the range, suggesting differences in adaptation in response to milder , shorter winters. In nature, 436 rear-edge populations experience less vernalization, with some not experiencing any vernalization. 437 Southern C. americana populations reflect this climate difference by having reduced vernalization 438 requirements, including southernmost populations which achieve a high percentage of flowering 439 (>90%) even under very limited vernalization (Exp. 4, Fig. 4b; Kalisz & Wardle, 1994). Reduced 440 vernalization requirements at the rear edge likely evolved in response to local shorter, warmer 441 winters where vernalization was an unreliable cue. Such patterns have been found in other plants, 442 in some cases leading up to a complete loss of a vernalization requirement (Wesselingh et al., 443 1994; Dijk et al., 1997; Boudry et al., 2002; Stinchcombe et al., 2005; Jokela et al., 2015). 444 Flowering in natural rear-edge populations was later relative to the onset of the growing 445 season towards lower latitudes, with an almost ten-fold steeper cline in flowering time relative to 446 the rest of the range . Flowering was also much later in rear-edge populations under common 447 greenhouse conditions (Exp. 2, Fig. 2c), as well as in common gardens (Exp. 3, Fig. 3c). Such a 448 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 16 dramatic difference in phenology between populations along temperature gradients has rarely been 449 documented in temperate plants. The exception, wild beets , similar to C. americana has later 450 flowering in southern populations and variation in vernalization along a latitudinal gradient (Dijk 451 et al. , 1997) . While it is possible that other environmental factors such as day length cue 452 differences in phenology (Blackman, 2017), this seems unlikely for rear-edge populations in our 453 study. Phenology in southernmost populations was not affected by variation in winter length (Exp. 454 4, Fig. 3c ), growing season climate (Exp. 3 , Fig. 3c ), or planting date (Exp. 3 , Fig. S2. 3c), 455 suggesting that the time to flowering is strongly genetically determined in these populations rather 456 than relying on an alternative environmental cue to vernalization. 457 Together, our results suggest a dramatic shift in the regulation of flowering phenology at 458 the rear edge, with less reliance on vernalization to initiate and time flowering. Instead, phenology 459 in these populations seems to be regulated by an environmentally insensitive delay in flowering, 460 contrasting with the plasticity in flowering time observed in northern populations . This evolved 461 delay may represent an adaptation to compensate for the extremely early start of the growing 462 season, thus allowing reproduction to track the warmest period of the year . A previous artificial 463 selection study revealed trade-offs between early flowering and reproductive fitness in C. 464 americana (Burgess et al., 2007), supporting an evolutionary advantage to delay flowering when 465 the growing season length does not strongly limit reproductive output. 466 467 The rear edge will differ in response to climate change compared to the rest of the range 468 Our study provides insight into how temperate plants may vary in their response to climate change 469 across their range. Warming climates are often predicted to result in range shifts towards higher 470 latitudes and altitudes through colonization at the colder edge and extinctions at the warmer edge 471 (Thomas et al., 2004; Hampe & Petit, 2005; Lenoir & Svenning, 2013) . In temperate plants that 472 rely on vernalization to cue reproduction , one may expect that populations at the warmer range 473 limits will be especially sensitive to climate change as warming winters may not meet vernalization 474 requirements, thus preventing the transition to reproduction (e.g. Padhye and Cameron 2009; 475 Satake et al. 2013). However, in contrast to this prediction, we found that rear-edge populations of 476 C. americana are likely to be the least affected by a reduction in winter length. Rather, populations 477 at mid- and high latitudes may be at risk because their sensitivity in the initiation and timing of 478 reproduction in response to winter length may lead to a substantial fitness loss under milder 479 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 17 winters. Reduced sensitivity at the rear edge may be common across temperate plants . Lower 480 latitude populations often have a reduced vernalization requirement (Wesselingh et al., 1994; Dijk 481 et al., 1997; Boudry et al., 2002; Stinchcombe et al., 2005; Jokela et al., 2015) They also show 482 strong local adaptation (Bontrager et al., 2021), likely facilitated by a history of long term exposure 483 to increasingly warmer climates. This result contributes to a growing body of literature questioning 484 the expected population decline under climate change at the rear edge (Vilà-Cabrera et al., 2019). 485 486 Postglacial colonization history drives divergence in adaptation to changing climates 487 Our study highlights the importance of studying the effects of warming climates in a 488 phylogeographic context, as the divergence in adaptation between northern and southern parts of 489 the range may also reflect different histories of exposure to climate change. The ~35°N breakpoint 490 in the day of first flower coincides with the transition between populations persisting in former 491 glacial refugia in southern parts of the range, i.e. the rear edge, and more northern populations that 492 have established after last glacial maximum (Barnard-Kubow et al. 2015; Koski et al. 2019; Prior 493 et al. 2020). Post-glacial colonization in temperate plants often results from population s 494 establishing in newly suitable habitats, tracking the gradual retreat of ice sheets as Earths’ global 495 temperatures rose during the Holocene (Hewitt, 2000, 2004) . In this context, C. americana 496 populations at higher latitudes may have retained flowering patterns closer to the ancestral state, 497 while rear-edge populations may have adapted to climates that have warmed over the last millennia 498 such that they are now subtropical. Reduced vernalization requirements and shifts in phenology at 499 the rear edge may thus be a relatively recent, post-glacial adaptation, diverging from an ancestral 500 state of relying on winter cues to regulate phenology. Despite historical and contemporary climate 501 change progressing at very different temporal scales, the evolution of reduced vernalization 502 requirements since the LGM suggests that evolution of lower vernalization requirements in 503 response to warming climate may be possible in populations at mid- and high latitude, and may be 504 facilitated by genetic variation occurring at the rear-edge. 505 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 18

506 This work was supported by the Swiss national Foundation (P2BSP3_195363) and the University 507 of Virginia College of Arts and Sciences . We are grateful to David Westneat (University of 508 Kentucky Ecological Research and Education Center Field Station, Lexington, KY), Gordon 509 Burghardt (Blaine, TN), Trevor Stamey (Clemson University Experimental Forest, Clemson, SC), 510 Jayne Lampley and John Walton (University of Alabama Tanglewood Station , Akron, AL ), 511 Theresa Jepsen (Florida state University Mission Road Research Facility, Tallahassee, FL) for 512 logistical support in establishing field sites. For their help in raising of plants, collecting data, 513 performing crosses and setting up the common garden experiments we thank A. Burricks, C. 514 Claussen, S. Cox, L. Elhady, M. Gower -Fici, S. Kelly, O. Keenan, A. López, K. Lamb, H. 515 Makowski, M. Marcich & E. Scott. We are also grateful to D. Brown, J. Collins, A. Diamond, L. 516 Elliott, E. Galloway, F. Griffith, I. Guenther, J. Hansen, H. Horne, J. Kees, M. Kohout, R. Laporte, 517 D. Reed & B. Sutherland for seed collection in natural populations. Collection permits were 518 provided by the Florida Department of Environmental Protection, the Tennessee Department of 519 Environment and Conservation, and Missouri Department of Conservation. 520 521 Conflict of Interest Statement 522 The authors declare that there is no conflict of interest. 523 524 Author Contributions 525 All authors contributed to the study design. AP collected seeds in the field and performed crosses, 526 AP and MT raised plants and analyzed the data, AP wrote the manuscript with input from LG. 527 528 Data availability statement 529 iNaturalist observations, climate data and phenotypic data for Experiments 2, 3 and 4 will be stored 530 on Dryad and made publicly available after initial acceptance. 531 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 19

532 Amasino RM. 2005. Vernalization and flowering time. Current Opinion in Biotechnology 16: 533 154–158. 534 Anderson JT. 2023. The consequences of winter climate change for plant performance. American 535 Journal of Botany 110: e16252. 536 Andrés F, Coupland G. 2012. The genetic basis of flowering responses to seasonal cues. Nature 537 Reviews. Genetics 13: 627–639. 538 Barnard-Kubow KB, Debban CL, Galloway LF. 2015. Multiple glacial refugia lead to genetic 539 structuring and the potential for reproductive isolation in a herbaceous plant. American Journal of 540 Botany 102: 1842–1853. 541 Bartoń K . 2023. MuMIn: Multi -Model Inference. URL https://CRAN.R-542 project.org/package=MuMIn. 543 Baskin JM, Baskin CC. 1984. The ecological life cycle of Campanula americana in Northcentral 544 Kentucky. Bulletin of the Torrey Botanical Club 111: 329–337. 545 Blackman BK. 2017. Changing responses to changing seasons: Natural variation in the plasticity 546 of flowering time. Plant Physiology 173: 16–26. 547 Bontrager M, Usui T, Lee -Yaw JA, Anstett DN, Branch HA, Hargreaves AL, Muir CD, 548 Angert AL . 2021. Adaptation across geographic ranges is consistent with strong selection in 549 marginal climates and legacies of range expansion. Evolution 75: 1316–1333. 550 Boudry P, Mccombie H, Van Dijk H. 2002. Vernalization requirement of wild beet Beta vulgaris 551 ssp. maritima: among population variation and its adaptive significance. Journal of Ecology 90: 552 693–703. 553 Burgess KS, Etterson JR, Galloway LF . 2007. Artificial selection shifts flowering phenology 554 and other correlated traits in an autotetraploid herb. Heredity 99: 641–648. 555 Chouard P. 1960. Vernalization and its relations to dormancy. Annual Review of Plant Physiology 556 11: 191–238. 557 Dijk HV, Boudry P, McCombre H, Vernet P. 1997. Flowering time in wild beet (Beta vulgaris 558 ssp. maritima) along a latitudinal cline. Acta Oecologica 18: 47–60. 559 Edmund H, Bell K . 2015. prism: Download data from the Oregon prism project. URL 560 https://CRAN.R-project.org/package=prism. 561 Etterson JR, Galloway LF . 2002. The influence of light on paternal plants in Campanula 562 americana (Campanulaceae): pollen characteristics and offspring traits. American Journal of 563 Botany 89: 1899–1906. 564 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 20 Faidiga AS, Oliver MG, Budke JM, Kalisz S . 2023. Shifts in flowering phenology in response 565 to spring temperatures in eastern Tennessee. American Journal of Botany 110: e16203. 566 Fick SE, Hijmans RJ. 2017. WorldClim 2: new 1-km spatial resolution climate surfaces for global 567 land areas. International Journal of Climatology 37: 4302–4315. 568 Fitter AH, Fitter RSR . 2002. Rapid changes in flowering time in British plants . Science 296: 569 1689–1691. 570 Geissler C, Davidson A, Niesenbaum RA. 2023. The influence of climate warming on flowering 571 phenology in relation to historical annual and seasonal temperatures and plant functional traits. 572 PeerJ 11: e15188. 573 Griffith TM, Watson MA . 2005. Stress avoidance in a common annual: reproductive timing is 574 important for local adaptation and geographic distribution. Journal of Evolutionary Biology 18: 575 1601–1612. 576 Haggerty BP, Galloway LF . 2011. Response of individual components of reproductive 577 phenology to growing season length in a monocarpic herb. Journal of Ecology 99: 242–253. 578 Hampe A, Petit RJ. 2005. Conserving biodiversity under climate change: the rear edge matters. 579 Ecology Letters 8: 461–467. 580 Hewitt G. 2000. The genetic legacy of the Quaternary ice ages. Nature 405: 907–913. 581 Hewitt GM. 2004. Genetic consequences of climatic oscillations in the Quaternary (KJ Willis, 582 KD Bennett, and D Walker, Eds.). Philosophical Transactions of the Royal Society of London. 583 Series B: Biological Sciences 359: 183–195. 584 Hughes PD, Gibbard PL, Ehlers J . 2013. Timing of glaciation during the last glacial cycle: 585 evaluating the concept of a global ‘Last Glacial Maximum’ (LGM). Earth-Science Reviews 125: 586 171–198. 587 Inouye D w. 2000. The ecological and evolutionary significance of frost in the context of climate 588 change. Ecology Letters 3: 457–463. 589 Jokela V, Trevaskis B, Seppänen MM . 2015. Genetic variation in the flowering and yield 590 formation of timothy (Phleum pratense L.) accessions after different photoperiod and vernalization 591 treatments. Frontiers in Plant Science 6. 592 Jombart T . 2008. adegenet: a R package for the multivariate analysis of genetic markers. 593 Bioinformatics 24: 1403–1405. 594 Jombart T, Devillard S, Balloux F. 2010. Discriminant analysis of principal components: a new 595

for the analysis of genetically structured populations. BMC Genetics 11: 94. 596 Kalisz S, Wardle GM. 1994. Life history variation in Campanula americana (Campanulaceae): 597 population differentiation. American Journal of Botany 81: 521–527. 598 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 21 Koski MH, Layman NC, Prior CJ, Busch JW, Galloway LF . 2019. Selfing ability and drift 599 load evolve with range expansion. Evolution Letters 3: 500–512. 600 Lempe J, Balasubramanian S, Sureshkumar S, Singh A, Schmid M, Weigel D. 2005. Diversity 601 of flowering responses in wild Arabidopsis thaliana strains (J Doebley, Ed.). PLoS Genetics 1: e6. 602 Lenoir J, Svenning J -C. 2013. Latitudinal and elevational range shifts under contemporary 603 climate chang e. In: Levin SA, ed. Encyclopedia of Biodiversity (Second Edition). Waltham: 604 Academic Press, 599–611. 605 Lenth R . 2019. emmeans: estimated marginal means, aka least -squares means. URL 606 https://CRAN.R-project.org/package=emmeans. 607 Liu Y, Mu J, Niklas KJ, Li G, Sun S . 2012. Global warming reduces plant reproductive output 608 for temperate multi-inflorescence species on the Tibetan plateau. New Phytologist 195: 427–436. 609 Luedeling E, Girvetz EH, Semenov MA, Brown PH. 2011. Climate change affects winter chill 610 for temperate fruit and nut trees. PLOS ONE 6: e20155. 611 Paccard A, Fruleux A, Willi Y . 2014. Latitudinal trait variation and responses to drought in 612 Arabidopsis lyrata. Oecologia 175: 577–587. 613 Padhye SR, Cameron AC. 2009. Vernalization responses of Campanula ‘Birch Hybrid’. Journal 614 of the American Society for Horticultural Science 134: 497–504. 615 Parmesan C, Yohe G . 2003. A globally coherent fingerprint of climate change impacts across 616 natural systems. Nature 421: 37–42. 617 Preston JC, Fjellheim S. 2022. Flowering time runs hot and cold. Plant Physiology 190: 5–18. 618 Preston JC, Sandve SR. 2013. Adaptation to seasonality and the winter freeze. Frontiers in Plant 619 Science 4. 620 Prior CJ, Layman NC, Koski MH, Galloway LF, Busch JW. 2020. Westward range expansion 621 from middle latitudes explains the Mississippi River discontinuity in a forest herb of eastern North 622 America. Molecular Ecology 29: 4473–4486. 623 R Core Team. 2024. R: A language and environment for statistical computing. R Foundation for 624 Statistical Computing, Vienna, Austria.URL https://www.R-project.org/. 625 Satake A, Kawagoe T, Saburi Y, Chiba Y, Sakurai G, Kudoh H. 2013. Forecasting flowering 626 phenology under climate warming by modelling the regulatory dynamics of flowering-time genes. 627 Nature Communications 4: 2303. 628 Stinchcombe JR, Caicedo AL, Hopkins R, Mays C, Boyd EW, Purugganan MD, Schmitt J . 629 2005. Vernalization sensitivity in Arabidopsis thaliana (Brassicaceae): the effects of latitude and 630 FLC variation. American Journal of Botany 92: 1701–1707. 631 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 22 Stinchcombe JR, Weinig C, Ungerer M, Olsen KM, Mays C, Halldorsdottir SS, Purugganan 632 MD, Schmitt J. 2004. A latitudinal cline in flowering time in Arabidopsis thaliana modulated by 633 the flowering time gene FRIGIDA. Proceedings of the National Academy of Sciences 101: 4712–634 4717. 635 Sugiura N. 1978. Further analysts of the data by akaike’ s information criterion and the finite 636 corrections: Further analysts of the data by akaike’ s. Communications in Statistics - Theory and 637

7: 13–26. 638 Thomas CD, Cameron A, Green RE, Bakkenes M, Beaumont LJ, Collingham YC, Erasmus 639 BFN, de Siqueira MF, Grainger A, Hannah L, et al. 2004. Extinction risk from climate change. 640 Nature 427: 145–148. 641 Vest K, Sobel JM . 2021. Variation in seasonal timing traits and life history along a latitudinal 642 transect in Mimulus ringens. Journal of Evolutionary Biology 34: 1803–1816. 643 Vilà-Cabrera A, Premoli AC, Jump AS . 2019. Refining predictions of population decline at 644 species’ rear edges. Global Change Biology 25: 1549–1560. 645 Wesselingh RA, Klinkhamer PGL, Jong TJ, Schlatmann EGM . 1994. A latitudinal cline in 646 vernalization requirement in Cirsium vulgare. Ecography 17: 272–277. 647 Willis CG, Ruhfel B, Primack RB, Miller-Rushing AJ, Davis CC. 2008. Phylogenetic patterns 648 of species loss in Thoreau’s woods are driven by climate change. Proceedings of the National 649 Academy of Sciences of the United States of America 105: 17029–17033. 650 651 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 23 Supporting information 652 653 Supporting material 1: Flowering phenology in natural populations (Experiment 1) 654 Fig. S1.1: Pruning of iNaturalist observations used in Experiment 1 655 Fig. S1.2: Example of flowering stages attributed to each observation 656 Fig. S1.3: Variation in flowering time across latitudes for the two best models 657 Table S1.1 Estimation of first flower date in natural populations based on flowering stage 658 Table S1.2a: Description of the 17 climatic variables used in Experiment 1 659 Table S1.2b: Contribution to the discriminant function and range of the 17 climatic variables 660 used in Experiment 1 661 Table S1.3: Comparison of three models describing variation in the day of first flower across 662 latitude 663 Supporting method S1.1: Selection of observations for estimation of day of first flower 664 Supporting method S1.2: Parametrization of analyses 665 666 Supporting material 2: Flowering phenology under experimentally manipulated conditions 667 (Experiments 2, 3, 4) 668 Fig. S2.1: Design of Experiment 4 669 Fig. S2.2: Variation in time to bolting (a) and time between bolting and flowering (b) in 670 Experiment 2 671 Fig. S2.3: Patterns of transition to bolting (a), transition from bolting to flowering (b) and day of 672 first flower (c) in Experiment 3. 673 Fig. S2.4: Patterns of transition to bolting (a), transition from bolting to flowering (b), time to 674 bolting (c), and time between bolting and flowering (D) in Experiment 4 675 Table S2.1: Populations used in Experiments 2, 3 and 4 676 Table S2.2a: Dependent variables analyzed in Experiments 2 & 4 677 Table S2.2b: Dependent variables analyzed in Experiment 3 678 Table S2.3: Comparison of model fit for linear and quadratic relationship with latitude for 679 Experiments 2, 3 and 4 680 Table S2.4a: Conditions in each common garden 681 Table S2.4b: Dates of transplanting and phenology estimation in each common garden 682 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 24 Table S2.5: Estimation of first flower date in the common garden experiment based on flowering 683 stages. 684 Table S2.6: Estimates of the effect of latitude, cohort and their interaction on traits in Experiment 685 2 686 Table S2.7a: Estimates of the effect of latitude on traits in Experiment 3 687 Table S2.7b: Comparison of traits between common gardens in Experiment 3 688 Table S2.7c: Comparison of the effect of latitude on traits between common gardens in 689 Experiment 3 690 Table S2.8a: Estimates of the effect of latitude across three vernalization treatments on traits in 691 Experiment 4 692 Table S2.8b: Comparison of the effect of vernalization treatments on traits in Experiment 4 693 Table S2.8c: Comparison of the effect of latitude on traits between vernalization treatments in 694 Experiment 4 695 Supporting method S2.1: Raising of plants and seed rearing in Experiment 2 696 Supporting method S2.2: Parametrization of the hierarchical mixed -effect models used in 697 Experiments 2, 3, 4 698 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 25 Tables 699 700 Table 1: Test of variation in the day of first flower, length of vernalization and time between the start of the growing season 701 and flowering across the latitudinal range of Campanula americana relative to a predicted breakpoint. 702 703 Latitude (L) Breakpoint (BP) L * BP Dependent variable β BP χ² β χ² β χ² R2m R2c Day of first flower -5.96 1.88 23.59 *** -273.10 39.52 *** 7.85 37.84 *** 0.18 0.18 Length of vernalization 2.63 10.75 13.57 *** -289.84 130.99 *** 8.12 119.59 *** 0.93 0.95 Time between flowering and the start of the growing season -15.58 -1.37 107.87 *** -493.68 85.97 *** 14.25 82.81 *** 0.46 0.54 704 All dependent variables were assumed to follow Gaussian distributions. Each model was optimized with the bobyqa optimizer to 705 improve convergence. Test statistics include the slope of each effect (β), the χ²-value, and the marginal and conditional R2 of the model. 706 For the effect of latitude, β was estimated for the parts of the range below or above the predicted breakpoint (BP) in latitude (34.98 °N). 707 *** P<0.001. Results for random effects are not shown. 708 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 26 Table 2: Test of variation in reproductive phenology across latitudes in a common 709 greenhouse environment (Exp. 2), common gardens across a latitudinal gradient (Exp. 3) 710 and in response to different length vernalization treatments (Exp. 4). 711 712 Latitude (L) Categorical effect (C) L * C Dependent variable N Χ2 Χ2 Χ2 R2m R2c Experiment 2 Cohort Flowering success † 994 5.95 ** 6.05 * 3.62 (*) 0.51 0.52 Time to flowering 881 9.40 ** 50.82 *** 36.63 *** 0.27 0.65 Time to bolting 893 25.70 ** 0.46 3.02 0.15 0.51 Time between bolting and flowering 877 40.08 *** 28.19 *** 31.25 *** 0.46 0.73 Experiment 3 Common garden Flowering success † 2231 31.60 *** 143.91 *** 58.07 *** 0.39 0.51 Transition to bolting † 100 2.72 (*) 3.23 * 3.63 ** 0.35 - Transition from bolting to flowering † 2341 7.09 * 59.31 *** 20.16 ** 0.44 0.56 Day of first flower 2136 40.09 *** 1365.27 *** 161.40 *** 0.53 0.70 Time to flowering (since transplanting) 2136 40.09 *** 87.05 *** 161.40 *** 0.34 0.58 Experiment 4 Vernalization Trt Flowering success † 1544 97.33 *** 143.24 *** 74.50 *** 0.69 0.80 Transition to bolting † 1544 71.72 *** 143.11 *** 42.26 *** 0.67 0.82 Transition from bolting to flowering † 979 20.00 *** 18.26 *** 16.80 *** 0.30 0.44 Time to flowering 711 2.92 127.80 *** 152.03 *** 0.20 0.64 Time to bolting 786 14.49 *** 16.39 *** 13.63 ** 0.18 0.45 Time between bolting and flowering 711 0.66 98.51 *** 117.23 *** 0.33 0.67 713 Binary variables (†) were analyzed by models predicting non -zeros on the logit scale. All other 714 dependent variables were assumed to follow Gaussian distributions. Each model was optimized 715 with the bobyqa optimizer to improve convergence. Test statistics include the χ²-value, and the 716 marginal and conditional R2 of the model. χ²-value with P-values < 0.05 are in bold; (*) P<0.1, * 717 P<0.05, ** P<0.01, *** P<0.001. The slope of each effect is reported in Table S2.6, S2.7 and S2.8. 718

for random effects are not shown. Note that transition to bolting in Exp. 3 was analyzed 719 at the population level in a simple linear model without random effects, we therefore report the F-720 value instead of χ², and R2 instead of marginal and conditional R2. 721 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 27 Figures 722 723 Fig. 1: Variation in flowering phenology observed in nature and associated climatic factors 724 across latitudes (Exp. 1). (a) Location of C. americana observations (black dots). The range of 725 the species is outlined with a gray line; red arrows represent the latitudinal delimitation of the rear 726 edge. Shading indicates the latitudinal temperature gradient based on minimum temperature of the 727 coldest month (bio6, WorldClim 2.0, (Fick & Hijmans, 2017). (b) Day of first flower inferred from 728 observations in 2018-2022 (a). (c) The length of vernalization and ( d) the time between the start 729 of the growing season and flowering inferred for each observation from PRSIM climate data 730 (https://prism.oregonstate.edu). These variables best explained the differences in the day of first 731 flower below (red) or above (black) the identified breakpoint (34.98° N, dashed vertical lines) in 732 the latitudinal cline in day of first flower. Solid lines represent the significant model predicted 733 slope of the relationship between each variable and latitude, with the 95% confidence interval 734 indicated as shading. Test statistics reported in Table 1. 735 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 28 736 Fig. 2: Variation in reproductive phenology in a common environment (Exp. 2). (a) Location of populations (dots). The range of 737 the species is shaded in gray; red arrows represent the latitudinal delimitation of the rear edge. F lowering success (b) and time to 738 flowering (c) were recorded for two cohorts of populations raised in a common controlled environment. Lines represent the significant 739 model-predicted relationship between traits estimated at the individual level and population latitude, with the 95% confidence inte rval 740 indicated as shading (cohort 1) or dotted lines (cohort 2). Test statistics reported in Table 2 and Table S2.6. 741 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 29 742 Fig. 3: Range -wide variation in reproductive phenology in common gardens (Exp. 3). (a) Location of populations and common 743 gardens. The range of the species is shaded in gray; red arrows representing the latitudinal delimitation of the rear edge. Populations are 744 indicated by white dots, and common gardens by colored triangles. CG4 (gray) was not included in the analyses because of low survival. 745 Flowering success in collection round 1 (b) and time to flowering since transplanting (c) were recorded for populations sampled across 746 the range and raised in five common gardens across a latitudinal gradient. Colors distinguish the different gardens as depict ed in (A), 747 and their latitude is indicated with arrows along the x-axis. Lines represent the significant model-predicted relationship for each garden 748 between traits estimated at the individual level and home latitude of the population, with the 95% confidence interval indica ted by 749 shading. Test statistics reported in Table 2 and Tables S2.7. 750 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint 30 751 752 Fig. 4: Effect of vernalization length on range -wide variation in flowering success and phenology (Exp. 4). (a) Location of 753 populations (dots). The range of the species is shaded in gray; red arrows represent the latitudinal delimitation of the rear edge. Flowering 754 success (b), and time to flowering (c) were recorded populations sampled across the range and raised in three vernalization treatments 755 (six, four or two weeks) in controlled conditions. For time to flowering, the 2 -week treatment is not shown as too few individuals 756 flowered. Lines represent the significant model-predicted relationship between each trait estimated at the individual level and population 757 latitude in each treatment, with the 95% confidence interval indicated as shading. Test statistics reported in Table 2 and Tables S2.8. 758 .CC-BY-NC-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 17, 2024. ; https://doi.org/10.1101/2024.08.16.608089doi: bioRxiv preprint