The effects of climate change on floral anthocyanin polymorphisms

Abstract

Pigmentation affords resistance to abiotic stressors, and thus can respond adaptively or plastically to drought and extreme temperatures associated with global change. Plants frequently display variability in flower coloration that is underlain by anthocyanin pigmentation. While anthocyanin polymorphisms impact plant–animal interactions, they also impact reproductive performance under abiotic stress. We used descriptions of flower colour from over 1900 herbarium records representing 12 North American species spanning 124 years to test whether anthocyanin-based flower colour has responded to global change. Based on demonstrated abiotic associations with performance of anthocyanin colour morphs, we predicted pigmentation would increase in species experiencing increased aridity, but decline in those experiencing larger increases in temperature. We found that the frequency of reports of pigmented morphs increased temporally in some taxa but displayed subtle declines in others. Pigmentation was negatively associated with temperature and positively associated with vapour pressure deficit (a metric of aridity) across taxa. Species experiencing larger temperature increases over time displayed reductions in pigmentation, while those experiencing increases in aridity displayed increases in pigmentation. Change in anthocyanin-based floral colour was thus linked with climatic change. Altered flower coloration has the strong potential to impact plant–animal interactions and overall plant reproductive performance.

1. Introduction

Plastic and adaptive phenotypic change are crucial for the persistence of wild populations in the face of rapid climate change [1]. Changing climatic conditions should favour traits and physiological processes that reduce stress under altered temperature and precipitation regimes [2]. However, among populations, and among species, responses to global change often depend on a variety of factors including habitat type [3] or the magnitude of abiotic change experienced by a given species or population [4,5]. Determining the factors that contribute to variability in temporal phenotypic change among taxa affords predictive power for understanding the evolutionary trajectories of traits.

Across organisms, pigmentation phenotypes can strongly impact thermoregulation [6], photoprotection [7] and desiccation [8]. Thus, pigmentation is a likely target of selection by altered abiotic environments associated with global change [9]. In flowering plants, pigmentation can vary dramatically within taxa [10], and this variability is, in many cases, linked with differences in overall plant performance and reproductive success. Anthocyanins are phenolic pigments that provide colours to flowers like blue, pink and purple. Many species exhibit variability in the intensity of anthocyanin pigmentation, ranging from white (unpigmented) to deeply pigmented (figure 1). Floral colour variation often has profound repercussions for interactions between plants and visually oriented pollinators and herbivores, and thus plant fitness [10,11].

Figure 1.

Examples of floral anthocyanin polymorphisms in three focal species investigated in this study. Figures on the left show unpigmented morphs while those on the right display pigmented morphs. Allium cernuum (a,b), Raphanus sativus (c,d) and Hydrophyllum virginianum (e,f). (Online version in colour.)

While anthocyanin-based floral colour mediates plant–animal interactions, anthocyanins strongly impact plant performance in the face of abiotic stress as well [10]. Anthocyanins function as anti-oxidants, osmoregulators and photoprotectants, providing tolerance under a wide variety of abiotic stressors [12]. Thus, anthocyanin-less morphs that produce unpigmented white flowers may be subject to elevated abiotic stress compared to pigmented morphs. Water availability in particular has been linked with anthocyanin-based flower colour polymorphisms. Assessing results from multiple systems, darker morphs appear to be favoured in drier conditions. Pigmented morphs of species from a variety of plant families outperformed unpigmented morphs under drought conditions [13], and in one species, Boechera stricta, non-pigmented morphs outperformed pigmented morphs in well-watered treatments [14]. Moreover, drier populations of Clarkia xantiana have elevated floral anthocyanin concentrations [15], and precipitation patterns are predicted to drive temporal turnover in pigmented morphs of Linanthus parryae [16].

Differential performance of pigmented and unpigmented morphs is also detected under varied temperature regimes. Pigmented morphs have higher reproductive success than unpigmented morphs under high heat stress in Ipomoea purpurea [17]. However, darker flowers have the potential to absorb more solar radiative energy [18], potentially elevating floral temperatures and damaging pollen or ovule function [4]. Accordingly, increased floral pigmentation is associated with reduced temperatures while lighter morphs are associated with higher temperatures in a number of taxa [19,20].

Given the strong associations between floral colour and plant performance under differing abiotic conditions, it follows that global change has the potential to drive temporal change in flower coloration [4]. As global temperature increases, declines in floral pigmentation may be expected if reduced pigment production modifies the operative temperature of heat-sensitive pollen and ovules [18]. However, increases in drought-like conditions may favour pigmented morphs [13]. Thus, responses of flower colour to global change may depend on the relative changes in temperature, precipitation and humidity. Temporal change in floral pigmentation can be assessed by phenotyping historical specimens housed in herbaria. While anthocyanins degrade over time, obscuring floral colour phenotypes of herbarium specimens (CNS and MHK 2019, personal observation), plant collectors often describe floral colour as unpigmented (white), or pigmented for species with anthocyanin polymorphism. Thus, the frequency of collections of pigmented and unpigmented morphs can be tracked over time. Phenotypes can subsequently be linked with climatic conditions at the time of plant collection using climate databases, making herbarium records instrumental for examining the effects of global change on plants [21]. While plant–pollinator and plant–herbivore interactions could be important for shaping any observed phenotypic changes in flower colour over time, the nature of such interactions are more difficult to ascertain from historical specimens than plant–abiotic associations inferred from climate databases.

Here, we used historical data from herbaria and climate databases of 12 species known to be polymorphic in their expression of flower colour to address the following questions: (i) does the incidence of pigmented morphs change over the course of 124 years (1895–2019)? (ii) Does the incidence of pigmented morphs covary with climate at the time of plant collection? (iii) Does the magnitude of climate change experienced by a given species predict the degree of temporal change in the incidence of pigmented morphs? Based on previous research linking flower pigmentation with temperature and water availability, we predicted that species experiencing larger increases in temperature over time will decline in pigmentation, and that those experiencing larger decreases in precipitation or humidity over time will display an increase in the frequency of pigmentation.

2. Methods

(a). Species selection and scoring flower colour

We selected 12 species with reported floral colour polymorphisms (figure 1) that occur in North America, representing eight families and 10 genera (table 1). We obtained herbarium specimen data from the SouthEast Regional Network of Expertise and Collections (SERNEC), Consortium of Pacific Northwest Herbaria, Consortium of California Herbaria and the Consortium of Northeastern Herbaria. We additionally searched the physical collection at Clemson University Herbarium for specimens not already represented in SERNEC. Scoring colour by eye on specimens did not provide reliable phenotypes. For example, some specimens that appeared to have white flowers were described by collectors as purple-flowered, indicating that pigments likely degraded over time. Thus, we relied on collectors' descriptions of flower colour.

Table 1.

List of species used to test the effects of global change on floral anthocyanin pigmentation. The number of herbarium accessions recorded as pigmented, mixed (collector described both pigmented and unpigmented floral morphs) and unpigmented, and the temporal range of specimens are provided. For analyses, mixed and pigmented were grouped and represent accessions for which pigmentation was documented.

family	genus	species	pigmented accessions	mixed accessions	unpigmented accessions	year range
Amaryllidaceae
	Allium	cernuum	161	15	11	117
Asteraceae
	Symphiotrichum	lanceolatum	199	28	225	123
	Symphiotrichum	racemosum	6	2	26	99
Boraginaceae
	Hydrophyllym	virginianum	13	8	17	104
Brassicaceae
	Boechera	stricta	35	10	89	83
	Hesperis	matronalis	98	23	7	121
	Raphanus	sativus	89	50	40	122
Convolvulaceae
	Calystegia	sepium	88	7	91	120
Polemoniaceae
	Linanthus	dianthiflorus	92	6	6	92
	Linanthus	parryae	16	44	55	96
Polygonaceae
	Polygonum	persicaria	166	24	29	122
Solonaceae
	Solanum	carolinense	74	18	76	118

For all downloaded records of each species, we used the conditional highlighting feature in Microsoft Excel to identify all specimens described as having either unpigmented (white/cream) flowers or pigmented (red, pink, blue and purple) flowers. Variants of these colours were also included in the highlighting rules, e.g. ‘mauve’, ‘rose’ and ‘pinkish’.

We scored each herbarium accession as ‘unpigmented’, ‘pigmented’ or ‘mixed’ based on colour descriptions. We scored accessions as unpigmented when collectors only noted unpigmented flowers, such as ‘Frequent, flowers white’. We scored accessions as pigmented when collectors only noted only pigmented flowers, such as ‘Common, flowers pink’, or ‘flowers mauve’. When a collector noted colour variation, we scored an accession as mixed. While in some cases it was clear that colour variation was among plants (e.g. ‘Flowers white, and reddish purple on separate plants'), in most cases, colour variability was recorded in a manner that could also be interpreted as colour variation within flowers, or among flowers within an individual (e.g. ‘flowers white to blue’). Both accessions scored as mixed and pigmented thus represent collections in which pigmented flowers were recorded, whereas those scored as unpigmented represent accessions in which pigmented flowers were not recorded. We included species in our study for which we collected at least 20 reports of flower colour in separate localities (i.e. those with unique latitude/longitude) across all four herbaria consortia. This resulted in 2371 records across all 12 species.

We used a binary colour scoring system to indicate the presence or absence of pigmentation. Unpigmented accessions were scored as ‘0’. Accessions scored as pigmented and mixed were scored as a ‘1’. Mixed accessions were not scored as ‘0.5’ because this implies an equal proportion of pigmented and unpigmented morphs which could not be deciphered from collectors' descriptions. Thus, we examined the presence or absence of pigmentation, which has been shown to influence plant fitness in many systems [22].

Collection bias based on flower colour has been documented in the herbarium record [23]. Two types of biases could impact our study. First, collectors could focus collection efforts on one colour morph at sites where multiple were present. For instance, recording only unpigmented plant(s) at a site with both morphs would underestimate the presence of pigmentation. Second, collectors could shift collection practices towards targeting a certain morph in mixed-morph populations over time. To test for these biases, we used an approach similar to Meineke et al. [24]. We subset our data to collectors that made 10 or more collections and then subset to collections that were either described as only pigmented or only unpigmented. This resulted in 33 collectors across 534 records, with the number of specimens per collector ranging 7–33. We modelled pigmentation score (0 or 1) as a function of species identity, collector, and the interaction between collector and year using a generalized linear model with a binomial distribution. We assessed the significance of terms using type III sums of squares. This analysis revealed that collector identity did not impact the likelihood of documenting pigmentation (Collector effect: p = 0.464), and that none of the collectors shifted towards or away from documenting pigmentation across the course of their collecting (Collector × Year effect: p = 0.459). Full results of this model are provided in electronic supplementary material, table S1.

(b). Geospatial and climatic data

From herbarium specimen records, we retrieved the date of specimen collection (day, month and year) and latitudinal and longitudinal coordinates. For instances in which latitude and longitude were not provided, we used detailed descriptions of the locality, city and state to retrieve approximate coordinates from Google Maps. We extracted the altitude (m) of specimen collection from the WorldClim v. 2 database at a 2.5 min grid scale using R.

We then used the coordinates and the month and year of collection to extract historic bioclimatic data during the month of specimen collection (PRISM Climate Group, Oregon State University). Data were extracted using an ArcGIS toolbox provided by Brown and colleagues [25]. Measures of monthly precipitation, minimum, maximum, and mean temperature, minimum and maximum vapour pressure deficit (VPD), and dewpoint temperature were obtained. Because PRISM did not provide climatic data for all coordinate and date combinations, our final, complete dataset consisted of 1944 herbarium specimens across 12 species, collected between 1895 and 2019.

(c). Statistical analyses

To test whether the frequency of collection of pigmented morphs changed over time, we modelled flower colour (0 = unpigmented; 1 = pigmented) as a function of species and year while controlling for latitude, longitude, altitude, month and month nested within year by including them as fixed effects (Spatiotemporal Model). In the model we also evaluated the interaction of species, year, altitude, latitude and longitude using an ANOVA with Type III SS and a binomial distribution. Geospatial parameters were included in the model to account for variance in colour explained by spatial position of collected specimens because latitude and longitude can predict flower colour [26–28]. We included month nested within year as a covariate to control for potential variation in colour across months within a given flowering season. Altitude was log transformed to improve normality.

To test whether the frequency of pigmented morphs covaried with climatic variables we modelled colour score as a function of species, monthly average precipitation, monthly maximum VPD, monthly minimum temperature (Climatic Model). We used an ANOVA with Type III SS and a binomial distribution. Precipitation, VPD and minimum temperature were chosen as climatic predictors of floral colour for the following reasons. First, previous studies found that decreases in temperature and reduced precipitation are related to increases in flower pigmentation [16,29]. Second, VPD is an important but underappreciated metric of drought [30]. VPD is the difference between how much moisture is in the air and the amount of moisture that can be held when the air is saturated. High VPD has important physiological repercussions, such as plant water stress from increased transpiration rates and drought-like soil conditions because VPD increases with air temperature [31]. Thus, it is possible that increases in VPD could favour pigmented morphs. Since VPD was not tightly correlated with precipitation (r = −0.328; p < 0.001) or minimum temperature (r = 0.347; p < 0.001) in our dataset, its direct effect could be evaluated in a multiple regression framework. However, dewpoint temperature, the temperature at which water vapour condenses to a liquid [32], was not included in the model because it was highly correlated with minimum temperature (r = 0.826; p < 0.001). Finally, other climatic variables (mean temperature, maximum temperature, minimum VPD) could not be used in the model due to strong multicollinearity and repetitiveness with other predictors. We considered variables with a correlation coefficient greater than 0.50 to be too tightly correlated to include in the same model. The variance inflation score (VIF) of each predictor in the final model was less than 2.0 indicating minimal multicollinearity.

Because of a strong interaction between the year of specimen collection and species identity on flower colour, we tested whether the magnitude of colour change over time was predicted by the magnitude of climatic change experienced by a given taxon [4]. For each species, we separately calculated the direct effect of year of collection on pigmentation by modelling colour score as a function of longitude, latitude, log-transformed altitude, year, month and month nested within year. We used Firth's biased-reduced logistic regression (‘logistf’ package in R) to obtain the coefficients of Allium cernuum, Hesperis matronalis and Lianthus dianthiflorus because these species had very few accessions (less than 10%) with a colour score of ‘0’ (unpigmented) (table 1). Firth's logistic regression is used to avoid potentially misleading computations and interpretations when such skewedness exists in a dataset [33]. We then used the same model predictors to calculate the effect of year on each of the three climatic variables (precipitation, maximum VPD, minimum temperature) to determine the magnitude of climatic change experienced by each taxon.

To test whether temporal changes in climate predict temporal changes in the presence of pigmentation, we modelled the yearly changes in pigmentation as a function of yearly changes in precipitation, maximum VPD, and minimum temperature using a multiple linear regression.

All analyses were done in R v. 3.6.3 [34] using generalized linear models. We checked for normality and collinearity among predictors before building all models using a correlation matrix and the variance inflation factor (VIF) function (‘car’ package in R). We scaled and centred predictors for standardization and to reduce bias in the interpretation of results due to differing unit scales.

3. Results

(a). Temporal change in floral pigmentation

Species differed in the frequency of pigmented collections (table 1; table 2). Across all taxa, there was not a significant change in the frequency of pigmented records over time, however, a strong interaction between species and year indicated that in some species the presence of pigmented individuals documented increased over time, while in others the presence of pigmented individuals documented declined (table 2 and figure 2). For example, pigmented records became more common over time in Symphyotrichum racemosum but unpigmented records tended to become more common over time in L. parryae (figure 2).

Table 2.

The effect of the year of specimen collection on the presence (1) or absence (0) of pigmentation while accounting for spatial effects (latitude, longitude, altitude) for 12 North American species that are polymorphic for floral pigmentation. Results are from a generalized linear model with a binomial response.

effect	χ2	d.f.	p-value
species	166.82	11	<0.001***
latitude	0.05	1	0.819
longitude	1.22	1	0.270
altitude	0.87	1	0.352
year	1.15	1	0.284
month	0.07	1	0.789
month nested in year	1.16	1	0.282
species × latitude	244.18	11	<0.001***
species × longitude	22.8	11	0.019*
species × altitude	65.81	11	<0.001***
species × year	1209.19	11	<0.001***
latitude × longitude	0.18	1	0.675
latitude × altitude	0.18	1	0.674
longitude × altitude	0.4	1	0.528
latitude × year	2.2	1	0.138
longitude × year	2.85	1	0.091
altitude × year	2.22	1	0.136
species × latitude × longitude	31.24	11	0.001**
species × latitude × altitude	12.46	11	0.330
species × longitude × altitude	55.27	11	<0.001***
latitude × longitude × altitude	0.48	1	0.488
species × altitude × year	15.97	11	0.142
species × latitude × year	12.69	11	0.314
species × longitude × year	336.11	11	<0.001***

*p < 0.05, **p < 0.01, ***p < 0.001.

Figure 2.

The direct effects (±s.e.) of the year of herbarium specimen collection on the presence (1) or absence (0) of floral pigmentation of 12 North American species. Effects were calculated for each species while accounting for geospatial and seasonal effects on pigmentation. Full species names are provided in table 1.

(b). Climatic correlates with pigmentation

In a model testing for the direct association between climatic parameters and the presence of pigmentation, temperature and VPD affected the presence of pigmentation across all taxa (table 3). Specifically, elevated temperature was associated with a lower likelihood of pigmented individuals being reported (figure 3b), while increased VPD (higher aridity) was associated with a higher likelihood of pigmented individuals (figure 3c). Precipitation was unassociated with pigmentation across taxa (figure 3a).

Table 3.

The effects of average monthly precipitation, vapour pressure deficit and temperature on flower pigmentation of herbarium specimens at the time of collection. Results are from a generalized linear model with a binomial response.

effect	χ2	d.f.	p
species	429.3	11	<0.0001***
precipitation	0.41	1	0.700
maximum vapour pressure deficit	4.91	1	0.026*
minimum temperature	3.82	1	0.051*

*p ≤ 0.05, ***p < 0.001.

Figure 3.

The direct effects of monthly average (a) precipitation, (b) temperature and (c) vapour pressure deficit on the presence (1) or absence (0) of floral pigmentation in herbarium specimens of 12 North American species. The model output is provided in table 3.

(c). The effect of temporal climatic change on temporal phenotypic change

Together, temporal change in temperature, precipitation and VPD predicted 55% of the temporal change in pigmentation among taxa (R² = 0.55). Both changes in temperature and VPD over time significantly predicted the direction and magnitude of temporal change in flower pigmentation (figure 4 and table 4 temperature parameter estimate: −0.965 ± 0.310 s.e., p = 0.014 and VPD parameter estimate: 0.743 ± 0.289, p = 0.033). Larger temporal increases in temperature corresponded with larger declines in the frequency of pigmented collections, while larger increases in VPD corresponded with increases in pigmented collections (figure 4). Although the relationship between changes in precipitation over time and change in pigmentation was not significant (precipitation parameter estimate; −0.313 ± 0.202, p = 0.160), the relationship was negative, which was expected based on previous research. Species that experienced decreases in precipitation tended to increase in the frequency of pigmented collections over time (figure 4).

Figure 4.

The effects of temporal change in (a) precipitation, (b) temperature and (c) vapour pressure deficit on temporal change in anthocyanin pigmentation for 12 North American taxa. Effect plots for each parameter based on a multiple linear regression of pigmentation change as a function of change in each climatic factor are shown. Regression lines are best fit linear relationships with associated standard error, and each point represents a species. The model output is provided in table 4. (Online version in colour.)

Table 4.

The effects of temporal change in precipitation, temperature and vapour pressure deficit (VPD) on temporal change in floral pigmentation for 12 North American species. Parameter estimates are scaled so that the magnitude of their effects on pigmentation can be directly compared.

effect	estimate	s.e.	t	p
precipitation change	−0.313	0.202	−1.547	0.160
temperature change	−0.964	0.31	−3.113	0.014*
VPD change	0.743	0.289	2.573	0.033*

*p < 0.05.

4. Discussion

We linked historical climate data with historical floral colour phenotypes in the herbarium record to provide evidence that climatic change is associated with temporal change in floral anthocyanin pigmentation. Flower colour polymorphic species largely tended to increase in the frequency of pigmented records over time, while few decreased over the 124-year timeframe. As predicted, species that displayed larger increases in pigmentation experienced larger increases in aridity, a condition under which anthocyanin biosynthesis is favoured [19,29,35]. Taxa experiencing the largest increases in temperature, however, experienced declines in pigmentation potentially due to reduced pigmentation resulting in reduced absorption of radiative energy [18]. Such phenotypic responses could be expected to continue as thermal stress and drought resulting from climate change are predicted to act as directional selection pressures in natural populations [2].

Our results are suggestive of a positive directional trajectory in anthocyanin production because most of the species tended to increase in pigmentation (figure 1). Interestingly, however, the two species displaying subtle pigmentation declines, L. dianthiflorus and L. parryae, are endemic to California, while those displaying the largest increases in pigmentation, S. racemosum and Hydrophyllum virginianum, are eastern North American taxa. The different geographical distributions of species used in this study could explain differences in pigmentation change incurred among the species (figure 1). Western North America, the range of focal Linanthus species, has experienced particularly dramatic increases in temperature while in the Eastern US, temperature change has been more subtle and even declined [36]. Because the degree of temperature change varies longitudinally in North America, it is foreseeable that across taxa with disparate ranges, some may have experienced little pigmentation change in response to temperature while other species experienced more substantial change.

Higher temperature was associated with unpigmented morphs across taxa (figure 3b), and taxa experiencing larger temperature increases over time displayed larger declines in pigmentation over time (figure 4). Lighter coloration absorbs less radiative heat than darker coloration and is, therefore, predicted to be favoured in warmer environments in which flowers may overheat [18,20]. Since the severity of climate warming is regionally specific [36] its effects on the trajectory of flower colour are likely to be spatially heterogeneous, and depend on other aspects of climatic change like aridity which we discuss below.

We detected significant effects of VPD on pigmentation across taxa (figure 3c), and temporal increases in VPD (an indication of increased aridity) were strongly associated with increases in pigmentation over time across taxa (figure 4). Because pigmented morphs have been shown to have elevated reproductive success over non-pigmented morphs in drier soil conditions [13,37], it would follow that elevated VPD could also be associated with elevated pigmentation. In particular, higher VPD increases evapotranspiration and thus, water loss, from plants. To our knowledge, however, VPD and flower pigmentation, have not yet been functionally linked by any study. To date, studies on the effects of VPD tend to focus on transpiration rates and stomatal conductance [31,38,39], but not changes in anthocyanin pigmentation. Future research should test the degree to which humidity impacts anthocyanin production, and test whether pigmented morphs outperform unpigmented morphs under lower humidity.

Warmer temperature and drought-like conditions often go hand-in-hand as climate warms in many regions [40,41]. Higher temperatures result in VPD increases, which intensify the effects of drought on plants through increasing demands on stomatal conductance [30]. Based on our findings that temperature increases are correlated with pigmentation declines, while VPD increases are associated with pigmentation increases, conflicting selection on floral pigmentation may occur under warmer and drier conditions. Physiological drought stress is expected to select for increased anthocyanins at the level of the entire plant, with impacts on floral colour. Conversely, selection via heat stress could act at the flower level to reduce pigmentation. Thus, relative strength of selection on floral versus vegetative tissue in a given species are likely to be an important determinant of how floral pigmentation responds to global change.

Although anthocyanin biosynthesis in flowers can be upregulated or downregulated in response to abiotic stress [37], changes in flower pigmentation could also impact ecological interactions with pollinators and herbivores [42,43]. Flower colour has long been known to aid in pollinator attraction, particularly for species that are self-incompatible and rely on out-crossing [10,44,45]. Pollinator preference for colour morphs can lead to differential fitness within and among species populations [46,47]. Nevertheless, colour polymorphism may still be maintained in populations because pollinators may alternate between morphs when foraging despite preference for a particular morph [48]. Anthocyanin-related pigment in petals can also serve as a defense to deter herbivores [14,49]. In a study on Raphanus sativus, herbivores displayed poorer performance on pigmented morphs, which contained higher concentrations of foliar secondary compounds than their unpigmented counterparts [11]. Weevil herbivores preferred white morphs of Geranium thumbergii over pink morphs, also likely due to the absence of secondary compounds related to anthocyanin production [50]. Thus, flower colour change in response to abiotic aspects of global change has the potential to impact plant–animal interactions. While not addressed in the current study, temporal change in plant–animal interactions could shape floral colour in addition to abiotic stress. Innovative use of herbaria to examine plant–pollinator [51] or plant–herbivore [24] interactions could be used to explore their roles in shaping variation in anthocyanin-based floral colour polymorphisms.

5. Conclusion

Our findings link climate change in North America over a 124-year period with shifts in floral anthocyanin production that bolster previous research demonstrating the effects of climate change on pigmentation phenotypes across plants and animals [4,52,53]. Climate-induced changes in floral pigmentation are likely to directly impact plant reproductive and physiological performance, but its likely effect on the fate of species interactions with plants warrants further study. Our analyses revealed that an understanding of the severity of climate change experienced by a given taxon is important for predicting the magnitude of phenotypic change over time. Such continued research will help give insight to how species will respond to the various aspects of climate change and which species are the most vulnerable to future climate projections [9,54,55].

Data accessibility

Datasets and code associated with the manuscript have been deposited in the Dryad Repository: https://doi.org/10.5061/dryad.7wm37pvrq [56].

Authors' contributions

C.N.S. and M.H.K. conceived the study, C.N.S. collected and analysed the data and C.N.S. and M.H.K. wrote the manuscript.

Competing interests

The authors declare no competing interests.

Funding

This work was supported by NSF DEB 174590 to M.H.K. and by Clemson University.