Variation in nectar quality across 34 grassland plant species

Variation in nectar quality across 34 grassland plant species C. Venjakob, F. A. Ruedenauer, A.-M. Klein & S. D. Leonhardt 1 Institute of Ecology, Ecosystem Functions, Leuphana University of L€ uneburg, L€ uneburg, Germany 2 Agroecology, DNPW, University of G€ ottingen, G€ ottingen, Germany 3 Plant–Insect Interactions, TUM School of Life Science Systems, Technical University of Munich, Freising, Germany 4 Faculty of Environment and Natural Resources, Nature Conservation and Landscape Ecology, University of Freiburg, Freiburg, Germany 5 Department of Animal Ecology and Tropical Biology, University of W€ urzburg, W€ urzburg, Germany


INTRODUCTION
Many flowering plant species need flower-visiting insects for pollination. They often provide rewards (mostly nectar and occasionally pollen) to attract pollinators (Agthe 1951;Baker 1963;Kearns et al. 1998;Klein et al. 2007;Ollerton et al. 2011). Although most macronutrients, which are essential for organisms, are located in pollen (in particular protein and fat) (De Groot 1953;Roulston & Cane 2000;Keller et al. 2005;Weiner et al. 2010), nectar represents the main source of carbohydrates and additionally contains variable concentrations of amino acids (Heinrich 1981;Wcislo & Cane 1996;Carter et al. 2006;Venjakob et al. 2020). In fact, some pollinators, e.g. butterflies, even entirely depend on nectar as sole nutrient source (Erhardt & Rusterholz 1998) and rely on its amino acid content to compensate for nitrogen deprivation during larval development (Mevi-Sch€ utz & Erhardt 2005). In other groups, e.g. parasitoid wasps, nectar can increase longevity, fecundity and mobility (Winkler et al. 2006). Nectar was therefore considered the most important floral reward for attracting pollinators (Simpson & Neff 1983;Somme et al. 2015;Parachnowitsch et al. 2019).
From the pollinators' perspective, nectar needs to meet nutritional requirements, outbalance foraging costs (Pyke et al. 1977;Waddington 1982) and provide sufficient energy for mobility and thermoregulation (herein McCallum et al. 2013;Hendriksma et al. 2014). From the plants' perspective, nectar needs to attract pollinators at minimum production costs (Pyke 1991;Nepi & Stpiczy nska 2008), deter non-mutualists, e.g. nectar robbers and micro-organisms, from stealing or degrading nectar (Adler 2000;Gonz alez-Teuber & Heil 2009;Escalante-P erez & Heil 2012) and ideally manipulate pollinator behaviour to their advantage (Pyke 2016). These different requirements render the nutritional composition of nectar (henceforth referred to as nectar quality) a multifunctional trait, which can be subject to conflicting interests (Parachnowitsch et al. 2019;van der Kooi et al. 2021). For example, different flower-visiting and pollinating species (henceforth all referred to as pollinators) prefer different sugar concentrations, i.e. bees prefer nectar with 50-60% sugar concentration and butterflies and birds prefer nectar with 35% sugar concentration, likely due to different ways of consumption (i.e. dipping versus suction) (Kim et al. 2011). Also, nectar sugar concentration negatively correlates with yeast abundance (Herrera et al. 2009) which may in turn affect pollinator preferences Schaeffer et al. 2017). Consequently, nectar quality can directly and indirectly affect visitation patterns of pollinators.
Several biotic and abiotic factors were found to be related to variation in nectar quality, including age and damage to flowers (Gottsberger et al. 1990), soil conditions (Baude et al. 2011;Becklin et al. 2011), genetic relatedness (i.e. phylogeny) (Baker & Baker 1976;Nicolson & Van Wyk 1998;Perret et al. 2001), flower morphology (Gusman & Gottsberger 1996;Torres & Galetto 2002), microbial communities (Vannette et al. 2013), selection by specific pollinators (Petanidou et al. 2006;Willmer 2011;Chalcoff et al. 2017;Tiedge & Lohaus 2017;Silva et al. 2020) and various interactive effects [see Parachnowitsch et al. (2019) for a complete review of factors affecting nectar quality]. For example, phylogenetically related plant species, i.e. belonging to the same tribe or family, can show similar chemical characteristics of nectar, such as a clear dominance of hexose or sucrose (Percival, 1961;Bernardello et al. 1994;Wolff 2006) and similar amino acid profiles (Baker & Baker 1986). However, so far, only a few studies have investigated heritability and evidence for selection in nectar traits, such as sugar concentration and content (Parachnowitsch et al. 2019). Similarities in nectar quality may alternatively be due to similar morphological traits, e.g. shape of flower tubes, and thus similar requirements for protection against evaporation (Witt et al. 2013). For example, plant species with long tubular flowers typically contain sucrose-rich nectar (high sucrose proportion), whereas plant species with more open flowers typically contain hexose-rich nectar (high hexose proportion) to reduce evaporation via increased osmolality (Percival 1961;Bernardello 2007;Witt et al. 2013), because the loss of water in nectar leads to increased viscosity, which decelerates nectar uptake or even prevents pollinators from consumption (K€ ohler et al. 2010;Kim et al. 2011). In fact, several studies found a strong preference of long-tongued pollinators, such as butterflies, moths and long-tongued bees, for sucrose-rich nectar, and of short-tongued bees and flies for hexose-rich nectar (Baker & Baker 1983;Petanidou et al. 2006;Gonz alez-Teuber & Heil 2009), indicating that floral morphology and nectar quality interact with pollinator preferences. Despite its importance as floral reward and source of nutrients for pollinators, the precise factors driving variation in nectar quality, such as the partial roles of phylogenetic relatedness, pollinator preferences and floral morphology, are still largely unclear (Parachnowitsch et al. 2019).
To contribute to a better understanding of the parameters which affect variation in nectar quality, we analysed potential correlations between the phylogenetic relatedness and morphology of plants and the amino acid and carbohydrate composition of their nectar for 34 grassland species. We hypothesized that: (i) plant species generally differed both qualitatively and quantitatively in their amino acid and carbohydrate composition, but that (ii) related plant species (i.e. species within the same families) would show a more similar nectar composition (with regard to concentrations and proportions of sugars and amino acids as well as their ratios) than plant species from different families due to phylogenetic and/ or morphological constraints.

Experimental field site
The study was conducted within the framework of the Jena Experiment, which is a grassland biodiversity experiment in Thuringia, Germany (50°55 0 N, 11°35 0 E; 130 m a.s.l.) established in 2002, comprising 60 native plant species of the plant association Arrhenatherion, with 16 grass species and 44 flowering herb species (Roscher et al. 2004). The study field is situated near the river Saale and encompasses 82 plots containing different plant species mixtures, forming a plant species richness gradient. We collected nectar of 34 flowering plant species from 22 of those plots, preferentially from species' monocultures, but also from plots with higher levels of plant species richness, when monocultures comprised insufficient numbers of flowering individuals. We selected all flowering plant species for which we could access a minimum of 1 µl, i.e. 34 out of 41 species. For more details on the design of the Jena Experiment, see Roscher et al. (2004).

Nectar sampling
In 2011, we sampled nectar of a total of 34 flowering plant species on sunny to cloudy days. Flowers were bagged in cotton gauze 1 day before nectar sampling (mesh size 0.80-1.00 mm) to exclude pollinators (Klein et al. 2003) and ensure that nectar was not depleted by insects prior to sampling. Accumulated nectar was collected in the morning using a microcapillary pipette (1 µl holding volume) equipped with a pipetting aid (for more details see Corbet 2003;Venjakob et al. 2020). We aimed to collect a minimum of 1 µl nectar per sample. When 1 µl could not be obtained from one flower, additional flowers were sampled. We collected samples from at least seven plant individuals per species. Exceptions with less samples for carbohydrate analysis were Geranium pratense L. (6 samples), Trifolium fragiferum L. (6), and Trifolium pratense L. (3). Six plant species (Anthriscus sylvestris Hoffm., Campanula patula L., Glechoma hederacea L., Primula veris L., Ranunculus repens L. and Veronica chamaedrys L.) were re-sampled in 2012 because nectar was too dilute due to heavy rain during the flowering period or because of insufficient amounts of nectar. Microcapillaries for nectar sampling were stored in Eppendorf tubes (see Venjakob et al. 2020) and kept at À20°C until analyses.

Nectar preparation
Prior to analysis, microcapillaries (32 mm) with nectar were rinsed with 100 µl 99.8% ethanol and centrifuged for 5 min to remove nectar from capillaries into 2-ml Eppendorf tubes (see Venjakob et al. 2020). Eppendorf tubes with nectar were kept in a DURAN â desiccator to evaporate ethanol, and nectar was subsequently re-dissolved in 50 µl ultra-pure water and centrifuged for 3 min (Venjakob et al. 2020). Finally, 48 µl of the supernatant were transferred into 1-ml glass vials equipped with 250 ll pulled-point glass inserts and stored at À20°C (for details see Venjakob et al. 2020).

Amino acid and carbohydrate analysis
Analyses of amino acids and carbohydrates were performed with high performance liquid chromatography (HPLC: Agilent Technologies 1260 Series; Agilent, B€ oblingen, Germany). The LC was equipped with an Agilent 1260 Infinity Quaternary Pump (G1311C, Agilent), an Agilent 1260 Infinity Standard Autosampler (G1329B) and an Agilent 1260 Infinity Thermostatted Column Compartment (G1316A). Oven temperature was 40°C for amino acids and 30°C for carbohydrates.
After every five samples, we ran an external standard containing 17 amino acids (Amino Acid Standard solution; Sigma-Aldrich) or three carbohydrates (sucrose, fructose and glucose, HPLC grade; Sigma-Aldrich) in four different concentrations. Using the Agilent ChemStation software for LC 3D systems (Agilent), amino acids and carbohydrates were identified and quantified by matching retention times and areas of calibrated standard compounds with retention times and areas of compounds found in nectar samples. Note that we cannot measure glutamine, arginine and tryptophan with our method due to either their destruction during acid hydrolysis (tryptophan) or a lack of standards (asparagine, glutamine).

Statistical analysis
We investigated whether nectar from the 34 plant species and the four most abundant plant families differed in the composition of carbohydrates and/or amino acids using permutation tests (PerMANOVA, using 10,000 permutations), which were based on Bray-Curtis distances between substances (R package: vegan; Adonis command). Separate permutation tests were performed for concentrations (in mg ml À1 ) and proportions of all carbohydrates and amino acids. We obtained proportions of individual compounds by dividing the concentration of each individual carbohydrate/amino acid by the total concentration of all carbohydrates/amino acids analysed.
We tested for plant species-specific differences in total carbohydrate and amino acid concentrations (i.e. the sum of all individual compounds), the total concentration and proportion of all essential amino acids and the total concentrations of all non-essential amino acids, as well as the ratio of all carbohydrates to all amino acids. Due to the nested plot design from which the samples were taken, we always tested first whether sample plot influenced the explained variance by composing both a generalized linear model (GLM) and generalized linear mixed effect models (GLMMs), with plant species entered as fixed factor and the plot from which the sample was taken as random factor. Due to the lack of a phylogenetic signal for all nutrient groups (see Table 1) we did not use GLMMs corrected for phylogenetic relatedness. Models were compared using the Akaike information criterion (AIC) and likelihood ratio tests (R package: lme4; anova command; following Zuur et al. 2009). We always provide results for the GLM if it did not have a significantly higher AIC value than the GLMM. For tests on plant species-specific differences, GLMs were always better than GLMMs, which renders the application of permutation tests for plant species-specific differences in compound compositions valid despite the nested plot design.
All analyses were repeated for plant family-specific differences between Asteraceae, Apiaceae, Fabaceae and Lamiaceae, as well as for different morphological flower traits. We confined family analyses to these four plant families, because they were the only families with sufficient numbers of plant species sampled per family (N ≥ 3). Information on floral traits composed for the same plots was obtained from Fornoff et al. (2017) and included flower symmetry (binomial: actinomorphic or bilateral), nectar access (binomial: open or hidden), inflorescence area (mm 2 ) and flower height (mm). All morphological traits were obtained as mean per plant species. We therefore also calculated mean per plant species for amino acid and carbohydrate concentrations and proportions when analysing the effect of family and traits on nectar chemistry and for the equivalent permutation tests. We again used generalized linear (mixed effect) models (GL(M)Ms) and compared GLMs with GLMMs, with plot included as random factor. GLMMs often had significantly higher AIC values than GLMs, indicating that a high proportion of the observed variance was explained by plot identity. We additionally extracted marginal R 2 values for the different models to compare the variance explained by morphological traits and by plant family.
Following Junker et al. (2017) and Ruedenauer et al. (2019), we tested for a phylogenetic signal in the total, essential and non-essential content of amino acids, as well as the content of sugars and the carbohydrate (C) to total, essential and nonessential amino acid (AA) ratios (i.e. C:AA ratios) using Blomberg's K. The underlying phylogenetic tree was based on the phylogeny of Zanne et al. (2014). Response variables were always tested for normality and homogeneity of variances using graphical tools, as suggested by Zuur et al. (2009) and either log-or square root-(concentrations, ratios) or arcsine square root (proportions)-transformed when these requirements were not met. Due to multiple use of the same dataset, we only considered P < 0.01 as being significant. We finally used the mean carbohydrate and amino acid proportions and concentrations of each plant species to visually assess similarities within plant families using cluster dendrograms (R package vegan), also based on Bray-Curtis distances between substances. All analyses were performed in R version 3.1.3 (R Development Core Team 2015).

Differences between plant species
The 34 plant species sampled showed species-specific compositions for both carbohydrates (PerMANOVA: concentrations: R 2 = 0.69, P < 0.001; proportions: R 2 = 0.75, P < 0.001) and amino acids (concentrations: R 2 = 0.65, P < 0.001; proportions: R 2 = 0.82, P < 0.001) with significant differences between species in the concentrations and proportions of all individual carbohydrates and amino acids (GLMs: P-values always ≤0.001; Tables S1, S2, Fig. 1). Intraspecific variation was also pronounced, particularly for amino acids (see standard variations in Table S1). Interestingly, related species (i.e. species within the same family) could be either chemically similar (e.g. Apiaceae species with regard to amino acid compositions, except Pimpinella major) or chemically distinct (e.g. Lamiaceae species with regard to amino acid compositions) or both (Asteraceae species with regard to amino acid compositions) as assessed by dendrograms and chemical distances between species (see Table S2a-d, Fig. S1). Also, some plant species (e.g. Crepis biennis or Ajuga reptans with regard to amino acid compositions, Ranunculus acris and Sanguisorba officinalis with regard to carbohydrate compositions) showed very distinct nutritional profiles which strongly deviated from most other plant species (Table S2a-d, Fig. S1). None of the nutrient contents, proportions and ratios showed a phylogenetic signal (Table 1).
The C:AA ratios in nectar of the 34 plant species were generally carbohydrate-biased, but ranged from equal mean ratios of C:AA = 1:1 [for A. sylvestris (Apiaceae) and L. hispidus (Asteraceae)] to mean ratios largely dominated by carbohydrates, C: AA = >20:1 [for Geranium pratense L. (Geraniaceae), A. reptans and P. vulgaris] and differed between plant species (GLM: F = 19.17, P < 0.001; Fig. S1, Fig. 2). The same was true for the ratio of all carbohydrates to all essential amino acids (henceforth referred to as C:EAA) (GLM: F = 17.83, P < 0.001).

Differences between plant families and related to morphological traits
Asteraceae, Apiaceae, Fabaceae and Lamiaceae differed in the composition of all amino acids when concentrations were considered (PerMANOVA: P 2 = 0.27, P = 0.007; Fig. S2a), and marginally when proportions were considered (R 2 = 0.25, P = 0.02; Fig. S2b). Essential amino acids also tended to show family-specific profiles (concentrations: R 2 = 0.27, P = 0.01; Fig. S2c; proportions: R 2 = 0.27, P = 0.02; Fig. S2d). However, visual inspection of chemical similarities showed that separation by family was not strict and mostly driven by differences between Fabaceae and Asteraceae (Fig. S2). Moreover, some species of different families (e.g. Centaurea and Anthriscus) were more similar to each other than to other species within their respective plant families (Fig. S2, Table S2a-d). The proportion of all essential amino acids (proportion: GLMM: F = 4.090, P = 0.003; Fig. 3c) also differed between the four plant families, whereas the overall amino acid concentrations did not (GLMM: F = 1.425, P = 0.241; Fig. 3a). Nectar of Asteraceae and Lamiaceae generally contained the highest concentrations and proportions of essential amino acids (Fig. 3b,  c). Concentrations and proportions of 11 out of 17 single amino acids (e.g. glutamic acid, histidine, arginine. . .) also differed or tended to differ between the four plant families (Table S3).
When examining the effect of morphological traits of flowers, we found the concentration and proportion of lysine differed between symmetric and non-symmetric flowers (concentration: GLMM: F = 11.329, P = 0.007; proportion: F = 38.488, P < 0.001). The proportions of three other amino acids also tended to differ between the two flower types, while the remaining species were not affected by any floral traits (Table S3) Table S3).

DISCUSSION
As expected, nectar of the 34 studied plant species and four most abundant plant families of a mesotrophic grassland community differed in carbohydrate and amino acid composition. However, despite significant differences between families, nectar composition was not significantly influenced by phylogenetic relatedness, indicating that additional factors affected variation in floral nectar quality.
Interestingly, when comparing the explanatory power of floral morphology and plant family, family continuously explained a larger proportion of the variation in nectar chemistry, e.g. up to 92% for the concentration of lysine, than the investigated morphological parameters. Differences between Apiaceae, Asteraceae, Fabaceae and Lamiaceae (all very attractive for pollinators: Ebeling et al. 2008;Venjakob et al. 2016) were particularly pronounced for carbohydrates, with higher hexose concentrations and proportions in Apiaceae (open flowers) and Asteraceae, and higher sucrose concentrations and proportions in Fabaceae and Lamiaceae (closed/tubular flowers). This finding agrees with previous studies indicating that nectar carbohydrate composition largely correlates with floral morphology (Percival 1961;Bernardello 2007;Witt et al. 2013) and thus likely with constraints imposed through osmolality. However, Asteraceae also have tubular flowers (ray or disc flowers) composed in radially symmetrical flower heads (J€ ager et al. 2013), often formed as bowl-or even parabola-shaped flower heads (Kevan 1989). They nevertheless had mostly hexose-rich nectar, except for three species, which had relatively similar proportions of sucrose, fructose and glucose (Fig. 1). The hexose-rich nectar in most Asteraceae may be explained by the stronger tendency of the flowers to heat up as a consequence of the bowl-or parabola-shaped flower heads (Percival 1961). Different combinations of family-specific morphological aspects of flowers determining overall inflorescence shape and structure therefore need to be taken into account when addressing variations in nectar chemistry. The importance of such family-specific structural characteristics and combinations of morphological traits may also explain why 'nectar access' (a parameter based on the morphology of individual flowers in our study) had overall less explanatory power than 'plant family', and why we found strong family-specific differences in nectar chemistry but no in the phylogenetic signal. In fact, the two differently related plant families, Fabaceae and Lamiaceae, showed similarities in nectar carbohydrate composition, likely due to structural similarities of the flowers. The two more closely related plant families, Asteraceae and Apiaceae, also showed similarities in nectar carbohydrate composition, despite different flower structures. Moreover, within plant families, the nectar composition of individual species still varied independent of their relatedness. Alternatively or additionally, the observed differences in nectar chemistry may be related to different pollination syndromes. In fact, morphological and/or functional traits of specific pollinators can reliably predict the spectrum of flowering plants visited and thus specific links in pollination networks (Rosas-Guerrero et al. 2014). Besides preferences for specific flower colours or shapes, pollinators can also show preferences for specific nectar profiles. For example, many flower visitors prefer sugar solutions with amino acids (i.e. nectar) over pure sugar solutions, e.g. butterflies (Alm et al. 1990;Mevi-Sch€ utz & Erhardt 2005;Beck 2007), flies (Shiraishi & Kuwabara 1970;Potter & Bertin 1988), honeybees (Inouye & Waller 1984;Alm et al. 1990;Carter et al. 2006;Bertazzini et al. 2010) and solitary bees (Petanidou et al. 2006). Honeybees further prefer sugar solutions with essential amino acids over sugar solutions with non-essential amino acids (Hendriksma et al. 2014). Moreover, essential amino acids significantly support their feeding gland and flight muscle development (Hendriksma et al. 2019). Increasing concentrations and proportions of (essential) amino acids or of amino acids known to be perceived by pollinators (Ruedenauer et al. 2020) may consequently be a useful tool to increase attractiveness to potential pollinators and/or distract them from collecting pollen. In fact, Asteraceae plants in our study tended to have the highest concentrations and proportions of all (essential) amino acids and of the essential amino acid histidine, which may (among others) explain why Asteraceae were frequently visited by many different pollinators and other flower-visiting insects (Table S5; also Ebeling et al. 2008;Venjakob et al. 2016). However, some amino acids can also deter insects (Bell et al. 1996;Toshima & Tanimura 2012), as can several plant secondary metabolites (Stevenson et al. 2017). Moreover, nectar often contains additional substances in relatively low amounts, such as minerals or fatty acids (Nicolson & Thornburg 2007), which may also affect nectar preferences (Parachnowitsch et al. 2019). Notably, nectar of the Lamiaceae species investigated in our study did not show as high concentrations of the essential amino acid phenylalanine as Lamiaceae nectar of plants from the Mediterranean phryganic community (Petanidou et al. 2006), suggesting that different selection pressures may act on the nectar amino acid composition of related plant species growing in different habitats.
Besides individual compounds and compound groups (e.g. carbohydrates, C, or amino acids, AA), ratios (e.g. C:AA) between different compound groups were recently found to correlate with pollen foraging preferences, e.g. in bumblebees (Vaudo et al. 2016). Ratios have, to our knowledge, hitherto not been related to phylogeny or other floral traits. We found C:AA and C:EAA ratios in general to be carbohydrate-biased and variable, but also to show some plant species and family specificity. The carbohydrate-biased ratios meet the nutritional needs of most flower visitors, e.g. adult honeybee and bumblebee workers, which typically prioritize carbohydrate over (essential) amino acid intake, even over-consuming amino acids to obtain sufficient carbohydrates, and perform generally better on carbohydrate-rich diets (Paoli et al. 2014;Stabler et al. 2015;Austin & Gilbert 2021). However, (in particular female) butterflies prefer nectars rich in amino acids, and thus likely having lower C:AA ratios, because amino acids increase their fecundity (Mevi-Sch€ utz & Erhardt 2005). Nutritional requirements of different pollinator groups (e.g. butterflies versus bees) might thus exert different selection pressures on the nectar chemistry of insect-pollinated plants (Jervis & Boggs 2005), and might affect C:AA ratios (e.g. towards amino acids/ proteins in flowering plant species pollinated by butterflies). In our dataset, C:AA ratios were lowest in Apiaceae and Fabaceae and high in Lamiaceae, which might explain (among others) why Lamiaceae (with their proportionally lower amino acid content) were most frequently visited by bees, while Apiaceae and Fabaceae (with their proportionally higher amino acid content) attracted all sorts of flower visitors (including butterflies; see Tables S3, S4).
However, studies investigating the importance of various plant traits in predicting niche partitioning in temperate interaction networks between plants and insect flower visitors found floral phenology, morphology (e.g. tube length), scent and colour, but not nectar chemistry, to most strongly determine choices of insect visitors (Junker et al. 2013;Rafferty & Ives Fig. 3. Differences in (a) total concentrations (mg ml À1 ) of amino acids, (b) concentrations (mg ml À1 ) of all essential amino acids (both datasets were log-transformed) and (c) proportions of all essential amino acids (data were (arcsin(sqrt(x))-transformed)). Kantsa et al. 2018). Nevertheless, interestingly, different floral traits can interact. For example, bee-pollinated flowers often have a blue or yellow colour and UV marks (Wilson et al. 2004), known to be correlated with a comparatively high sugar content (Chittka & Menzel 1992) and to match the visual capacities of bees (Chittka & Menzel 1992). In fact, local variation in nectar sugar content might even drive selection for innate colour preferences (Raine & Chittka 2007). In turn, nectar volume in the bumblebee-pollinated Aconitum gymnandrum was found to be under strong selection pressure by pollinators (Zhao et al. 2016). These studies indicate that the plant-pollinator mutualism can affect nectar chemistry, but that directional outcomes are variable and likely depend on specific interactions. Our results suggest that these additionally depend on plant family-specific morphological constraints. Finally, an increasing number of studies highlight the role of specific microbiota in nectar (Fridman et al. 2012) and even of the foraging behaviour of pollinators themselves (Bogo et al. 2021) in affecting nectar chemistry, providing additional factors that may determine variation in floral nectar chemistry.

2013;
More thorough investigations of nutrient amounts and ratios found in floral resources, i.e. pollen and nectar, of plant species differing in phylogenetic relatedness, morphological characters, pollinator dependency and reward strategy, as well as their effect on pollinator preferences, should enable us to better disentangle the contributions of these various factors impacting floral resource chemistry. This knowledge is essential for better understanding the mechanisms underlying plant-pollinator interactions (Kantsa et al. 2018;van der Kooi et al. 2021).

ACKNOWLEDGEMENTS
We thank all members of the Jena Experiment community, who helped in maintaining the experiment throughout many years. Especially, we would like to thank our coordinator Anne Ebeling, our technical assistant Gerlinde Kratzsch and the gardeners, particularly Ute K€ ober and Silke Hengelhaupt, who also helped with flower bagging. Further thanks go to Stephanie Schr€ oder who helped with nectar sampling. M. Heil and three anonymous reviewers provided useful input on an earlier version of the manuscript. Finally, we gratefully acknowledge Angela Gronert for performing analyses with HPLC and Richard Bolek for his useful advice.

CONFLICTS OF INTEREST/COMPETING INTERESTS
The authors declare no conflict of interest.

AUTHOR CONTRIBUTIONS
CV, AMK and SDL designed the experiment. CV collected data; SDL and FAR performed statistical analyses. CV developed the first draft of the manuscript, and all authors contributed substantially to revisions.

AVAILABILITY OF DATA AND MATERIAL
All original data is submitted with this manuscript and will be deposited at the Jena Experiment upon publication.

SUPPORTING INFORMATION
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. Ratio of total amino acids (AA) in mg to total sugar in mg. Fig. S2. Cluster dendrograms of the four most abundant plant families (Apiaceae, Asteraceae, Fabaceae and Lamiaceae). Table S1. Nectar concentration and proportion of chemical components, such as amino acids and carbohydrates, measured via HPLC. Table S2. Chemical distance matrix between 34 plant species based on Bray-Curtis distances between either: the (a,c) proportions (%) or (b,d) concentrations (conc) of amino acids and carbohydrates. Table S3. Statistical results of generalized linear (mixed) models (GL(M)Ms) testing for effects of morphological factors (flower symmetry, nectar access, inflorescence area, flower height) and family for different single and total amino acid concentrations and proportions. Table S4. Statistical results of generalized linear (mixed) models (GL(M)Ms) testing for effects of morphological factors (flower symmetry, nectar access, inflorescence area, flower height) and family for different single and total carbohydrate concentrations and proportions. Table S5. Plant species studied (Roscher et al. 2004) for their nectar amino acid and carbohydrate composition (plant species in italic and families in bold letters) with flower visitors recorded at different plots of the Jena Experiment in 2011 (for further details see Venjakob et al. 2016). Table S6. Morphological traits of flowers and accessibility of pollen and/or nectar.