Work was carried out in the Andes east of Santiago, in the Farellones-La Parva-Valle Nevado area (~33° S). This area of the Andes has a semi-arid climate with a Mediterranean influence. Snow begins to accumulate from April-June and remains on the ground until late August-September to late November depending on elevation, with considerable variation among years. The climate during the spring and summer months is mainly sunny. However, cloudy days, intermittent afternoon cloudiness and windiness increase with elevation (Viale and Garreaud, 2014) and short-duration summer snow and hail may be received. January and February are warm months (Valle Olivares weather station and El Yeso Embalse, 33.68°S, 70.09°W, 2475 m.a.s.l. - http://explorador.cr2.cl/). Because of the altitudinally depressed treeline caused by aridity (Piper et al., 2016), above treeline gradients in the central Chile Andes extend for over 1000 m elevation providing an ideal system for investigating the questions of interest.

Above treeline vegetation in the central Chilean Andes comprises two physiognomically-recognizable vegetation belts. The first is a subalpine scrub dominated by small, rounded shrubs, accompanied by subshrubs, and perennial and annual herbs (Cavieres et al., 2000). The subalpine scrub belt is superseded by the high alpine belt comprised of flat cushion species, accompanied by subshrubs and perennial herbs, giving way to scattered perennial herbs in its upper limit. In terms of their floristic composition, the distributions of many of the species in these two vegetation belts are limited to the mediterranean-type climate area of central Chile. Nevertheless, the high alpine belt contains a number of species that are distributed widely along the Andean corridor (e.g. Caltha appendiculata, Colobanthus spp., Perezia pilifera, Azorella monantha). Northern-hemisphere clades with important presence in the Southern Andes typically involved initial colonization of lowland areas followed by subsequent ascent into alpine habitats (e.g. Valeriana, Bell et al., 2012; and probably Astragalus, Scherson et al., 2008).

Two sites were selected in the zonal vegetation in the subalpine belt (Sites I and II) and two in the high alpine belt (Sites III and IV) (Figure 1; see Table 1 for site characteristics and species numbers). Sites I and III, and Sites II and IV, respectively, pertain to two adjacent but distinct valley systems, with the former predominantly west-facing and the latter mainly east-facing. In the study area, plant species composition changes gradually with increasing elevation (Tamburrino et al., 2025) with only three species shared between our subalpine and high alpine sites. On each site, 10 individuals were marked per plant species. The total numbers of species per site ranged from 20 to 36, with fewer species on the two upper alpine sites due to the lower species richness there. Many species were shared among the two upper and two lower sites, respectively (Table 1, see Supplementary Figure 1 for species composition on each site). A few rare species, where 10 individuals could not be found, were not included. The total number of species considered in the study was 86. Because species in the upper alpine are mostly perennial and tend to begin to flower shortly after snowmelt, all plants (except for a single annual species) were marked in late autumn prior to winter snowfall. Many species in the subalpine sites were likewise marked in the previous autumn. On Site II, individuals in a dense patch of grasses could not be reliably marked as to their species identity and thus were excluded. Where needed, plants were protected with large wire netting enclosures to prevent damage by wandering cattle. As of the date of snowmelt, the sites were visited at five-day intervals until the end of the flowering season, from September 11^th, 2018, to May 15^th, 2019, spanning a continuous period of 35 weeks. On each observation date, all individuals of a species that were in flower were registered and the number of open flowers per plant counted. For plants with a very large number of flowers, the number of open flower-bearing branches was counted. For large flat cushion plants, flowering in 10 cm x 10 cm fixed quadrats running across the diameter of the cushion was recorded. For annual species where the density of individuals was very high, quadrats were used to quantify flowering. Snowmelt dates and the duration of phenological observations for each site can be found in Table 1.

Five calendar-based floral traits and six thermal sum-based traits were defined, giving a total of 11 flowering traits (see Box 1 for full detail of abbreviations). The calendar-based traits were:

Thermal sum-based traits were expressed in Growing Degree Days (GDD) which give the accumulated heat required for achieving reproductive milestones (FF, PPF and PFA). The biofix day was the date of snowmelt (Table 1). Snowmelt dates were established by direct observation over the last days the snow lay on the ground. Following other alpine studies, GDDs were calculated for two values of base temperature (T_BASE), T_BASE = 5 ^°C and T_BASE = 0 ^°C (GDD₀ and GDD₅) using hourly temperatures recorded at 1.5 m a.g.l HOBO U23 Pro v2 battery-powered sensors (Onset Computer Corp., Cape Cod, MA, USA). Sensors were protected with custom-made solar radiation shields. The two measures of GDD were calculated with HOBOware using the double sine method with no upper threshold. Calculation of GDD for the aforementioned reproductive milestones resulted in six GDD-based traits (GDD₀FF; GDD₅FF; GDD₀PFP; GDD₅PFP; GDD₀PFA and GDD₅PFA).

Flowering overlap among species on each site was estimated using Augspurger’s flowering synchrony index (Augspurger, 1983) with the flower R Package (Michalski and Durka, 2007). This index depicts the degree of overlap between all pairs of plants species on a site based on presence or absence of flowering on each observation date. All calendar-based and GDD-based traits, as well as Augspurger’s index, were compared among sites with ANOVA or the Kruskal-Wallis test, upon normality check. Post hoc Tukey test and pairwise Wilcoxon test were used, respectively, to detect differences among pairs of sites. All statistical analyses were carried out in R v.4.1.2 (R Core Team, 2021; RStudio Team, 2023).

Functional structure, i.e. the distribution of traits among species within a community, was estimated using functional diversity indices, calculated for each community from a trait matrix containing locally measured phenological traits for each species. To avoid collinearity among traits and to ensure that diversity metrics reflect orthogonal functional axes, a Principal Component Analysis (PCA) was performed on the standardized trait matrix. The first three principal components were retained, which together accounted for over 98% of the total variance. This dimensionality reduction allows for a more robust and interpretable representation of functional space, while minimizing numerical issues associated with calculating multidimensional convex hulls (Laliberte and Legendre, 2010). The functional structure indices calculated were:

To assess whether the observed values of the three metrics were greater or lower than expected by chance, functional diversity indices were compared to 9,999 randomized communities generated by randomly selecting species from the species pool of all four sites combined, maintaining species richness. The standardized effect size of the functional structure indices (χ_SES) was calculated as χ_obs- χ_null/SD_null, where χ_obs was the observed functional diversity index, χ_null was the mean value of the 9,999 randomizations and SD_null was the standard deviation of these simulated values. Additionally, empirical two-tailed p-values were computed as the proportion of null values with equal or greater deviation from the null mean than the observed value. This allowed us to determine whether each community exhibited significantly higher or lower functional structure than expected under random assembly. All calculations were carried out using the FD R package (Laliberté et al., 2014).

To obtain real branch lengths resolved at the species level, a phylogenetic tree based on three genetic markers (ITS, rbcL and matK) was constructed. For species where DNA sequences were not already available in GenBank, DNA was extracted in the laboratory from leaf material collected in the field and from herbarium material stored at CONC (Herbarium of the Department of Botany, University of Concepcion) and at SGO (Herbarium of the National Museum of Natural History). Samples collected in the field were stored in silica gel. Vouchers for field-collected material are deposited in the herbaria CONC. Ginkgo biloba (Ginkgoaceae) was chosen as outgroup. Downloaded and original sequences are detailed Supplementary Table 1.

A concatenation of nuclear marker ITS and plastidial regions rbcL and matK (Supplementary Data, Supplementary Table 2 for laboratory protocols and primers details) was used in a Bayesian inference analysis performed in the CIPRES Science Gateway V. 3.3 Portal (www.phylo.org). Three partitions were used corresponding to each gene, in which evolutionary models for each genetic region were: ITS GTR+I+G, rbcL GTR+I+G y matK GTR+G were applied based on the Akaike Information Criterion. Parameters were sampled 20 x 10⁶ generations and 25% of the first samples were discarded. Nodes with ≥ 0.95 were considered robust for posterior probabilities (Ronquist et al., 2012). The Tracer program v1.7 (Rambaut et al., 2018) was used to visualize output parameters to prove stationarity and assess convergence of duplicated runs on the same mean likelihood.

Phylogenetic signal for the 11 flowering traits was evaluated using Blomberg’s K (Blomberg et al., 2003). Blomberg’s K quantifies the observed similarity among phylogenetically related species relative to expectations under a white noise model (i.e., no phylogenetic signal). A value of K = 1 corresponds to Brownian motion expectation, values less than 1 indicate weaker phylogenetic signal (traits less similar among relatives than expected), whereas values greater than 1 suggest stronger-than-expected trait conservatism within clades. Phylogenetic signal was calculated individually for each site. Regardless of whether traits exhibited a phylogenetic signal, Brownian motion (BM), Ornstein–Uhlenbeck (OU), and early-burst (EB) evolutionary models were fitted to the trait data, and the best model was selected based on the Akaike Information Criterion (AIC). All analyses were performed using the phytools and geiger package in R (Pennell et al., 2014; Revell, 2024). Given the high consistency between GDD₀ and GDD₅ traits, only GDD₀ results are presented in the main text.

Species richness generally declines monotonically along elevational gradients in high mountains (Körner, 2021), signifying that local communities at higher elevations will naturally tend to contain smaller numbers of species. Münkemüller et al. (2012) concluded that the uncertainty of detecting phylogenetic signal decreases as sample size increases. To rigorously determine whether high elevation communities have been shaped by environmental filtering or trait convergence, the effect of differences in species richness must be dealt with. In this paper, a novel procedure was employed to detect any effect of species richness on the detection of phylogenetic signal and take concrete measures to overcome the large differences in species numbers between our subalpine and high alpine sites. To test for the effect of sample size on phylogenetic signal, 999 random subsamples of species for each site were run, ranging from the original species richness values down to 20 in the subalpine sites and from the original species richness down to 10 in the high alpine sites. Blomberg’s K and associated p-values were calculated and graphed against simulated richness. To overcome the spurious effect of the smaller sample sizes on the upper sites, an artificial single community was constructed by merging the phenology data for the two upper alpine sites. As not all species on the two sites are the same, this provided us with a community with a larger number of species (N = 30) which came close to sample sizes in the two subalpine communities. For the enlarged community, phylogenetic signal was recalculated for all eleven floral traits using mean dates for the shared species. The same procedure of constructing a new single community was carried out also for the subalpine sites to assure that the original results on phylogenetic signal continued to be detected after the merging procedure.

To investigate possible phylogenetic structure that would be suggestive of the existence of environmental filtering, mean pairwise phylogenetic distance (MPD) and mean nearest taxon phylogenetic distance (MNTD) among species was calculated (Webb et al., 2002) for each site. To test whether the phylogenetic structure of each site differed from random expectation, standardized effect size of MPD and MNTD was calculated using ses.mpd and ses.mntd functions of picante R package (Kembel et al., 2010). Positive values of MPD_SES or MNTD_SES are found when a community has higher observed phylogenetic distances than expected by chance, indicating phylogenetic evenness or overdispersion, whereas negative values are associated with lower phylogenetic distances among taxa, suggesting phylogenetic clustering. MPD is generally thought to be more sensitive to tree-wide patterns of phylogenetic clustering and evenness, while MNTD is more sensitive to patterns of evenness and clustering closer to the tips of the phylogeny (Webb et al., 2002).

The total evidence matrix for the 86 species included 5009 nucleotide characters (1103 ITS, 1669 rbcL and 2237 matK). The topology of the Bayesian Inference tree shows the relationships between the species considered in the study. Most taxa are well supported (>0.95; Supplementary Figure 2) and fall into their respective orders and families. The effective sample size (ESS) value was greater than 200 in a range between 2546 and 30945, indicating that the three genetic regions could be legitimately combined.

Snowmelt occurred in September on the lowest sites I and II and in November in the highest sites III and IV (see Table 1 for details). Whereas the snowmelt dates between the subalpine and high alpine sites differed by almost two months, mean date of first flowering, despite the much colder conditions, differed only by one month (November 30^thvs December 29^th). A few species (Barneoudia chilensis in Site I and Viola atropurpurea in Sites III and IV) began flowering immediately after snowmelt, whereas in Site II flowering began one week later.

While the length of the flowering period per species did not differ significantly among sites (Kruskal-Wallis, χ² = 1.79, p-value=0.62, Figure 2A), calendar day to first flowering was significantly earlier in Site I than in the high alpine sites (Figure 2B). First flowering (FF), peak flowering plants (PFP) and peak floral abundance (PFA) were significantly shorter in at least one of the high alpine sites than in at least one of the subalpine sites (Figures 2C–E). When translated to growing degree days (GDD), high alpine sites always showed significantly lower GDD for FF, PPF and PFA than the subalpine sites (Figure 3). Site IV showed significantly greater flowering synchronicity than all other sites (Figure 2F).

The first three principal components from the PCA analysis were retained to use as orthogonal proxies of trait variation for the functional structure analysis. These components explained 78.6%, 10% and 9.6% of the total trait variance (98.2% overall). PC1 loaded strongly on all GDD-based and most of the calendar-based traits, PC2 was associated with calendar day of first flowering (FF_DOY) and flowering season length (FSL), and PC3 was dominated by FLS (Supplementary Table 3).

While subalpine sites show higher raw values of functional richness (FRic), all four communities exhibited significantly lower FRic than expected under the null model (Table 2). Both functional diversity (FDis) and Rao’s quadratic entropy (RaoQ) were markedly lower for the two high alpine compared to the subalpine sites, particularly at Site IV (Table 2), indicating that species on the high alpine sites are more functionally similar than on the subalpine sites. In contrast, Site II displayed higher FDis and RaoQ values which do not significantly differ from random expectations, suggesting a lack of trait convergence on that subalpine site.

When sites were analyzed individually, phylogenetic signal was predominantly detected in flowering times and thermal requirements at the lower elevation sites. At Site I, significant phylogenetic signal was found for traits related to the timing of peak flowering (PFP and PFA), as well as for the growing degree day requirements (all GDD₀-based traits), whereas at Site II, significant phylogenetic signal also extended to the onset of flowering (FF_DOY, FF, PFP, PFA and GDD₀FF; Table 3). In contrast, significant phylogenetic signal was largely absent in the high alpine sites, with the exception of flowering season length (FSL) at Site IV (Table 3). For all traits and sites, the Ornstein–Uhlenbeck was the best evolution model (Supplementary Table 4).

The simulated models of decreasing richness on each site detected a very clear effect of species richness on phylogenetic signal for the GDD metrics; i.e. significant phylogenetic signal is less likely to be detected in the same community as species richness decreases (Supplementary Figure 3). As mentioned earlier, to discard that the lack of significant phylogenetic signal in the high alpine sites was related to the lower number of species there, the data from sites belonging to the same altitudinal belt were combined to obtain higher richness levels. The newly calculated flowering traits continued to show significant phylogenetic significance only in the subalpine belt (Supplementary Table 5). Therefore, despite the possibility of an effect of differences in species numbers for detecting phylogenetic signal, the lower number of species on the high alpine sites can be confidently dismissed as the reason for a lack of phylogenetic signal in our study system.

When compared to null models for the same richness, the two high alpine sites showed significantly negative standardizes mean nearest taxon distance (MNTD_SES) values while for Sites I and II there were no significant effects (Table 4). The results for the high alpine sites indicate phylogenetic clustering (i.e. less phylogenetic distance among co-occurring species than expected by chance). However, for standardized mean pairwise distance (MPD_SES), the values for the two high sites were not significant. Overall, results indicate that phylogenetic clustering is restricted to the high alpine sites.

The lower growing degree days values (GDD) on the higher elevation sites support the notion of strong flowering phenological adjustment over the ~1000 m altitudinal gradient in the central Chilean Andes representing less than seven kilometers of lineal distance. Lower GDD values accompanying flowering onset that was retarded by one month, indicate that, in general, flowering was accelerated at higher elevations. These results agree with earlier work in the alpine (Ahmad et al., 2021; Arroyo et al., 1981; Bucher and Römermann, 2020; Meng et al., 2016). A decrease in thermal sums required for flowering over elevation or snow melt gradients has been reported both for populations of the same species and at the community level (Kawai and Kudo, 2011; Bucher et al., 2018; Arroyo et al., 2021). The effect of this speeding up is significant; if plant species on the upper sites had required the same amount of heat to produce open flowers as those in the lower sites (mean GDD₀PFA=1038.38), peak flowering in the community would have occurred in early April, well past the warmer period of the year. Huelber et al. (2006) found no difference in cumulative energy input to reach flowering in a 1-month snowmelt gradient, but their research took place on a single site covering a 50m elevational gradient, which is much smaller than the elevational gradient considered in our study.

Not only was flowering speeded up after snowmelt, but flowering overlap among species at higher elevation was greater indicating a tendency for convergence in flowering times. This is also supported by functional structure indices measured across elevations. All sites showed negative values of functional richness (FRic_SES), indicating that communities represent discrete subsets of the functional space. This suggests that the set of phenological traits present on each site is shaped by local conditions. Moreover, functional dispersion and quadratic entropy (FDis_SES and RaoQ_SES) show significantly more negative values at the higher-elevation sites compared to the lower ones, indicating exceptionally low functional dispersion and pairwise dissimilarity among coexisting species, reinforcing the described pattern of convergence. In contrast, the lower-elevation sites have less extreme FDis_SES and RaoQ_SES values, suggesting that species in these communities occupy a broader range of functional strategies. Taken together, these results highlight the adaptive and evolutionary lability of flowering phenology along the elevational gradient. Schroeder et al. (2024) found that phylogenetic dispersion related to vegetative traits was also lower at 3500 m a.s.l. than at 2400 m a.s.l. in the central Chilean Andes. It is worth noting that the flowering patterns documented in our study were recorded during a single flowering season and therefore reflect the climatic condition of that particular year. Nevertheless, the large temperature differences between the two elevational belts considered are expected to be relatively consistent despite interannual variation. Moreover, previous studies indicate that the study period fell within the normal range of interannual temperature variation for the region (Arroyo et al., 2020).

The two components of phenological adjustment over elevational gradient (faster flowering and greater flowering time overlap) found in this study suggests a strong selective pressure of colder temperatures and perhaps of the length of the flowering season. If phenological adjustment is driven by an evolutionary process governed by temperature favoring trait convergence, a weakening of phylogenetic signal would be observed. Accordingly, phylogenetic signal on floral metrics was almost absent in the high alpine sites yet clearly present on the lower sites. Low Blomberg’s K values may be interpreted as evidence of trait convergence operating at all sites but with much greater strength in the upper sites. Although only slightly, Site I was the warmest of the two subalpine sites, and consequently, temperature would exert less selective pressure. Thus, it is perhaps not surprising that higher phylogenetic signal was found in all GDD-based traits on this site in comparison with the cooler subalpine site II.

Similar weakening in phylogenetic signal at higher elevations for flowering has been reported on other alpine gradients. Although no significant phylogenetic signal was detected by Basnett et al. (2019) in several elevational bands, an elevational decrease in the values of Blomberg’s K and Pagel’s λ was observed across a 1200 m elevation gradient. In the same way, Lessard-Therrien et al. (2014) showed less frequent significant phylogenetic signal at higher elevation over a 200 m elevational gradient. It should be pointed out, however, that these results could be influenced by the differences in species richness over elevation in those systems. Nonetheless, stronger phylogenetic signal at higher elevations was reported by D. Li et al. (2021) in a sea-level to high mountain gradient. These authors invoked the tropical niche conservatism hypothesis to explain that colder regions communities (like high alpine communities) are younger and shaped by environmental filtering, thus, showing similar phenological traits in closely related species.

It is worth mentioning that our phylogenetic signal analyses were conducted on an ecological community rather than on a well-sampled clade. Any community phylogeny represents only a subset of taxa drawn from multiple clades, and, therefore, misses trait information from taxa that are absent in that environment. While this approach may not be ideal for investigating the evolutionary mechanisms that shape traits, it certainly offers valuable insight into eco-evolutionary patterns, such as those underlying community assembly (Caradonna and Inouye, 2015).

In our study, we found significant phylogenetic clustering at the higher sites which is commonly interpreted as evidence of environmental filtering (Webb et al., 2002). However, significant values of standardized mean nearest taxon distance (MNTD_SES) but not for standardized mean pairwise distance (MPD_SES) indicate that the clustering on these sites is concentrated near the tips of the phylogeny rather than at its base, reflecting the coexistence of many clusters of closely related, relatively recently-evolved taxa spread over several lineages rather than a concentration of closely related species representing a limited number of lineages (Figure 4).

In the same venue, a recent review of 73 high elevation communities found a global tendency for phylogenetic clustering which was ascribed to environmental filtering imposed by the cooler temperatures at higher elevations (Qian et al., 2021). But when looking at the elevational gradient, some studies show increasing clustering with elevation (Li et al., 2014; Zhang et al., 2020) while others report the opposite trend (Culmsee and Leuschner, 2013; González-Caro et al., 2014; López-Angulo et al., 2018; Luo et al., 2023) and even increasing clustering followed by increasing dispersion in different altitudinal belts (Li et al., 2022). For our study, we used both the deep-sensitive MPD and the terminal-sensitive MNTD metrics. Analyses based on only one of these metrics run the risk of drawing erroneous conclusions regarding whether the phylogenetic structure of a community is shaped by environmental filtering, limiting similarity or non-restrictive colonization.

Given our phylogenetic structure results and the demonstration of phylogenetic signal weakening, the most likely scenario for the assembly process of these communities is the colonization of lineages dispersed across the phylogeny, with environmental filtering acting over non-clustered traits or not acting at all, followed by small radiations in the newer environment and strong adaptation resulting in convergent flowering traits across distinct species. This scenario reflects the “Tropical Niche Conservatism” hypothesis, in that taxa pertaining to high alpine communities in the central Chilean Andes are the descendants of lineages that managed to shift their niches to colder conditions (Wiens and Donoghue, 2004). Nonetheless, terminal but not basal phylogenetic clustering suggests that colonization to the alpine is not a rare event, recalling the “Out of the Tropics” (OTT) hypothesis (Jablonski et al., 2006). The OTT hypothesis posits that niche conservation, although it exists, does not limit the expansion of taxa into colder environments to the extent that upward colonizing lineages would be randomly distributed in the phylogeny (Qian and Ricklefs, 2016). This pattern is consistent with diversification documented in several Andean genera such as Adesmia (Pérez et al., 2024), Senecio (Salomón et al., 2025) and Chaeanthera (Hershkovitz et al., 2006). Regardless of whether the present-day species in the high alpine belt are a product of upward migration of linages, or dispersal along the Andean corridor, selective pressures and/or environmental filtering to adjust their flowering times would have taken place in situ o prior to migration along the Andean corridor, contributing to the observed patterns of functional structure and phylogenetic signal. Overall, our results support that taxa in both elevational belts have been subjected to selective pressures and evolved in these novel environments, but these pressures appear to have been stronger at the high alpine belt, where phylogenetic signal has weakened to a greater extent.

The integration of trait data and a thorough analysis of phylogenetic structure metrics suggest that abiotic factors act more as the drivers of evolution than as gatekeepers in these environments. Finally, if we had adhered faithfully to Webb et al. (2002) and limited our analysis to one of the phylogenetic structure metrics, we would have concluded that environmental filtering has been the main force shaping high alpine community composition. The integration of different analytic approaches suggests the co-occurrence of different processes often seen as non-complementary (i.e. environmental filtering combined with trait convergence). These findings endorse the notion that community assembly is a complex and irreducible process where straightforward and discrete frameworks can be misleading.

To our mind, an important contribution of our study is the rigorous treatment of the potential effect of sample size on the detection of phylogenetic signal. Basnett et al. (2019) found phylogenetic signal in 10 species of Rhododendron distributed across a full mountain gradient but not in the 3 to 5 species per 100 m altitudinal band while Krasnov et al. (2011) found it in 218 species in a continental pool of insects but not the 10 to 39 species of insects per regional pool. These authors concluded that phylogenetic signal is geographically and taxonomically scale-dependent. However, these results could also reflect an effect of sample size on phylogenetic signal as per shown in the theoretical work of Münkemüller et al. (2012). Here we introduced a method to detect this potential problem in studies where species numbers are unbalanced, as occurs over the alpine gradient and, indeed, in community comparisons in general. Effectively, in this study, we confirmed the decreasing probability of detecting phylogenetic signal with decreasing richness. To overcome this potential problem, we then merged sites belonging to the same altitudinal belt to obtain a species richness value similar to that in the lower-elevation sites. Upon controlling for species richness, phylogenetic signal persisted in the subalpine sites while it remained absent in the high alpine community. Consequently, we feel confident that our interpretation that phylogenetic signal on flowering traits becomes lost at the higher elevations and adaptation overrides phylogenetic conservatism is not a spurious result.