Inter- and intra-tree variability of carbon and oxygen stable isotope ratios of modern pollen from nine European tree species

Stable carbon and oxygen isotope ratios of raw pollen sampled from nine abundant tree species growing in natural habitats of central and northern Europe were investigated to understand the intra- and inter-specific variability of pollen-isotope values. All species yielded specific δ13Cpollen and δ18Opollen values and patterns, which can be ascribed to their physiology and habitat preferences. Broad-leaved trees flowering early in the year before leaf proliferation (Alnus glutinosa and Corylus avellana) exhibited on average 2.6‰ lower δ13Cpollen and 3.1‰ lower δ18Opollen values than broad-leaved and coniferous trees flowering during mid and late spring (Acer pseudoplatanus, Betula pendula, Carpinus betulus, Fagus sylvatica, Picea abies, Pinus sylvestris and Quercus robur). Mean species-specific δ13Cpollen values did not change markedly over time, whereas δ18Opollen values of two consecutive years were often statistically distinct. An intra-annual analysis of B. pendula and P. sylvestris pollen revealed increasing δ18Opollen values during the final weeks of pollen development. However, the δ13Cpollen values remained consistent throughout the pollen-maturation process. Detailed intra-individual analysis yielded circumferential and height-dependent variations within carbon and oxygen pollen-isotopes and the sampling position on a tree accounted for differences of up to 3.5‰ for δ13Cpollen and 2.1‰ for δ18Opollen. A comparison of isotope ranges from different geographic settings revealed gradients between maritime and continental as well as between high and low altitudinal study sites. The results of stepwise regression analysis demonstrated, that carbon and oxygen pollen-isotopes also reflect local non-climate environmental conditions. A detailed understanding of isotope patterns and ranges in modern pollen is necessary to enhance the accuracy of palaeoclimate investigations on δ13C and δ18O of fossil pollen. Furthermore, pollen-isotope values are species-specific and the analysis of species growing during different phenophases may be valuable for palaeoweather reconstructions of different seasons.

Introduction Stable carbon and oxygen isotope ratios of plant material are generally determined to understand the relationship between plants and their surrounding environment [1]. The continuously deepening knowledge of stable isotope patterns in plants of natural habitats finds applications in, for example, plant ecology, phytochemistry, genetic research and reconstructions of past environmental changes [2][3][4][5][6]. Plant physiological reactions to environmental factors, such as temperature, moisture availability and density of the surrounding vegetation, are known to affect the stable carbon and oxygen isotope composition of plant material [1,7,8]. Carbon isotope ratios (δ 13 C) of plant material are mostly determined by the factor-dependent amount of discrimination against 13 C during CO 2 uptake and by subsequent photosynthetic processes [e.g. 9, 10], while oxygen isotope ratios (δ 18 O) are often linked with the isotopic composition of environmental source water [11][12][13].
In general, the stable carbon isotope composition of stem material, leaves and pollen of the same plant individual highly correlate with one another [13][14][15]. Research on stable carbon isotope ratios of modern pollen focused mainly on species-specific patterns and ranges [15][16][17][18] and has been applied to investigate predominant photosynthetic pathways within grasslands [19][20][21]. Loader and Hemming [22] analysed δ 13 C pollen of Pinus sylvestris from 28 sites across Europe and identified a positive linear correlation between δ 13 C values and the prevailing temperature during pollen formation. Also, Jahren [14] detected positive correlations of δ 13 C pollen values with temperature for nine out of 14 plant species. Studies focussing on the determination of influencing climate factors on the isotope values of modern pollen include Bell et al. [15], who showed that δ 13 C pollen of Cedurs atlantica (Atlas cedar) correlates with precipitation and a long-term annual and summer scPDSI (self-calibrating Palmer Drought Severity Index). Schwarz [13] suspected a relationship between δ 13 C pollen and relative humidity, but no significant correlation of δ 13 C pollen of Pinus retinosa (Red pine), Pinus sylvestris (Scots pine) and Quercus rubra (Northern red oak) could be detected at North American sampling sites. However, all correlations were highly species-dependent and several plants strongly reacted to other untested environmental factors, superimposing the climate signal archived in the pollen [14].
Little is known about the variability within modern oxygen pollen-isotope values. Nonetheless, they have already been successfully applied to determine the provenance of honey [e.g. 23,24]. Loader and Hemming [25] identified a negative relationship of δ 18 O pollen values with the δ 18 O values of precipitation during pollen formation, contrasting δ 18 O of wood and leaves that is typically positively related to the δ 18 O of precipitation [e.g. 26,27]. Even if the δ 18 O pollen is highly determined by the δ 18 O of local precipitation, the degree of dependence seems to vary with plant type and physiology [13]. Hence, δ 13 C pollen and δ 18 O pollen values are influenced by local climate conditions during pollen formation [15,22], but not all variability within pollenisotope values can be ascribed to climate-related environmental factors alone.
Morphology-based analysis of fossil pollen is frequently used to reconstruct palaeovegetation, since fossil pollen are often well preserved, widespread and abundant in various Cenozoic archives [28][29][30][31][32][33][34]. A combination of traditional pollen analysis and stable isotope analysis of fossil pollen might enhance environmental reconstructions in a high spatio-temporal resolution. Due to plant-specific timings in pollen production and pollen shedding, even intraannual weather signals may be recorded in the δ 13 C pollen and δ 18 O pollen [13]. Some studies have already applied δ 13 C analysis to fossil pollen in an attempt to reconstruct past environmental changes [35][36][37]. A fossil δ 13 C pollen record of C. atlantica from Morocco revealed the feasibility of reconstructing a long-term trend of increasing aridity by analysing species-specific pollenisotopes [38]. Also, fossil Nothofagus (Southern beech) δ 13 C pollen indicated different moisture availability in Antarctica during the early and middle Eocene [39].
However, interpretation of fossil pollen-isotopes are based on observations of modern δ 13 C pollen and δ 18 O pollen patterns and ranges. In addition to climate conditions during pollen formation, non-climate impact factors may need to be considered when interpreting the δ 13 C pollen and δ 18 O pollen values. These factors include site-specific environmental parameters such as type of soil, plant associations, position on slopes, and the proximity of the individual tree to perennial waterbodies. A comparison of several abundant tree species growing at different sites under the same environmental conditions, intra-tree differences and fluctuations over several vegetation periods helps to assess the impact of non-climate environmental factors on the pollen-isotope values.
In the present study, we address the species-specific natural variability of pollen-isotopes of nine abundant tree species across seven European sites ranging from Belgium to Poland and Finland to Italy, sampled during the years 2015 and 2016. The species have been chosen based on their widespread abundance in natural European forests and the frequency of their pollen in fossil records.
Inter-and intra-tree δ 13 C pollen and δ 18 O pollen isotope variabilities were assessed and tested for relationships with non-climate environmental factors by stepwise regression analysis. In doing so, we aimed to advance our understanding of pollen stable isotope signals for future palaeoclimate reconstructions. In detail we investigated: 1. Species-specific δ 13 C pollen and δ 18 O pollen patterns of selected tree taxa and the variability of their pollen-isotopes between two consecutive vegetation periods (2015/2016).
2. δ 13 C pollen and δ 18 O pollen at different stages of the pollen maturation process.
3. Variability of δ 13 C pollen and δ 18 O pollen at different heights and cardinal directions of individual trees. 4. δ 13 C pollen and δ 18 O pollen along a gradient of continentality (W-E transect) and along a gradient in day length (N-S transect).

Sampling locations, sample collection and preparation
We sampled modern pollen from 658 individual trees of nine selected tree species growing in seven national and nature parks, which we consider natural habitats (Fig 1; Tables 1 and 2). None of the species sampled in this study is endangered or protected and sampling followed generally a non-invasive scheme of few inflorescences per individual tree. Therefore, after contacting and consulting with the national park authorities of each sampling site, no specific permissions were required for these locations and activities. Pollen were collected during two consecutive vegetation periods (February to June of 2015 and 2016) within the species-specific flowering periods (Fig 2; Table 2). The schedule for sampling followed individual phenology and thus roughly the geographic distribution and climate conditions of west-east and southnorth gradients in Europe [40]. All selected tree species use the C3 photosynthetic pathway. In principle, 20 flowering individuals were sampled per site and per species (Table 2). In case of dense forests, trees close to hiking trails, forestry roads or glades were sampled because sunlight illuminating the full height of tree crowns allows the development of lower branches with inflorescences. Samples were taken with a pruning device and an extendable stick from branches at positions of one metre up to seven metres above ground. Male inflorescences were cut off and placed in plastic bags. Bulk -samples of an individual tree were composed of several inflorescences of different branches from various heights. In the field, the samples were kept in a cooling box. Later, they were stored in a refrigerator at 6˚C to prevent mould infestation. In the laboratory, the samples were dried in a drying oven at a maximum temperature of 45˚C for seven to nine days. Dry samples were kept in a freezer at -16˚C until further processing. The separation of pollen from other flower tissue was achieved by thorough rinsing with deionised water using sieves with mesh sizes from 10 μm to 200 μm. Following rinsing and sieving the pollen were freeze-dried for 48-72 hours until fully dehydrated, transferred into Eppendorf (2 ml) safe lock tubes and frozen at -16˚C for preservation. To investigate intra-tree isotope patterns, an additional 152 intra-tree sub-samples were taken at different heights and cardinal directions from 22 trees of eight species (Fig 3). In

PLOS ONE
Variability of carbon and oxygen stable isotope ratios of modern pollen addition, individual inflorescences were collected separately in plastic bags to allow a high-resolution intra-tree variance analysis. Intra-annual analyses at different stages of the pollen maturation process were carried out for B. pendula and P. sylvestris, which were sampled twice at the same location within one vegetation period. Betula pendula was collected at Forêt d'Anlier on 10 March and 5 May 2015, whereas P. sylvestris was sampled twice in Gorczański National Park on the 20 May and 1 June 2015. The individual trees sampled for the intra-annual analysis grew within a small assessable area with similar habitat conditions. Due to minimal individual offsets in flower development and senescence, only some tree individuals could be sampled and analysed twice. Seven of the P. sylvestris individuals were identical (total number of samples at first sampling: 12; total number of samples at second sampling: 16) and eight B. pendula trees were samples twice (total number of samples at first sampling: 11; total number of samples at second sampling: 22). An elevation transect in the Tatrzański Mountains National Park extends from 1053 m a.s.l. to 1345 m a.s.l. on a north-facing slope. Along the gradient of roughly 300 m, 15 individual trees of Picea abies were sampled at five different elevations in 2015 and 2016.

Stable isotope analysis
For each measurement, the amount of 220 μg ± 10% of chemically untreated pollen material was weighed directly into silver capsules using a high-precision scale (Mettler Toledo AX 26 Delta Range). δ 13 C and δ 18 O were determined using a DELTA V isotope ratio mass spectrometer (IRMS; Thermo Fisher Scientific™, Bremen) at the dendrochronological laboratory, section 4.3, GFZ Potsdam, Germany. To exclude potential water contamination from air humidity, all samples were vacuum dried at 100˚C for at least 12 hours in a Thermo Scientific Heraeus VT 6060 P prior to measurement. The pollen material was reduced to CO for simultaneous IRMS analysis of carbon and oxygen isotope ratios in a High Temperature Conversion Elemental Analyzer (TC/EA; 1400˚C; Thermo Fisher Scientific™, Bremen) coupled to the IRMS. All isotope ratios are expressed relative to VPDB for δ 13 C and VSMOW for δ 18 O. Isotope data were compared against international and lab-internal reference material (IAEA-CH3, IAEA-CH6 and IAEA 601 and 602) using two reference standards with widespread isotopic compositions for a single-point normalisation [42]. Most of the 809 individual pollen samples were weighed and measured with two or three repetitions. In total, we conducted 2132 measurements of stable isotopes. The pollen-isotope dataset is deposited at Pangea Database (https://doi.org/10.1594/PANGAEA.910977).

Statistical analysis
All calculations and graphics were done using programmes R [43] and RStudio [44]. Value distributions for each site and year with indicated average are shown as bean plots [45]. Many parametric tests assume a normal distribution. Hence, we tested whether the pollen-isotope values in a given sample of a species from each location and year were normally distributed using the Shapiro-Wilk test. The minimum sample size analysed was three. The null hypothesis of normal distribution was rejected, if the probability p was smaller than the significance level (range: 0-1; significance level: p = 0.05). As several distributions were non-normal, we used the non-parametric Mann-Whitney U test for equality of medians (p(sm)) to compare inter-annual distribution patterns of pollen-isotopes (range: 0-1; significance level: p(sm) = 0.05). A possible relationship between isotope values and elevation at Tatrańzki Park Narodowy (Poland) was investigated using linear correlation. Inter-and intra-tree variability of sub-samples taken at different tree heights and cardinal directions was characterised by calculating respective standard deviations (1σ).

Stepwise regression analysis
Variables influencing the stable isotope composition in pollen (δ 13 C, δ 18 O) were explored by means of stepwise regression as implemented by the JMP Pro 13.1.0 software. Stepwise regression reduces variance to a linear model by eliminating insignificant predictors and was performed (1) by species (to investigate the most important influencing factors for each species over several sites) and (2) by site (to evaluate possible location-dependent environmental factors for the pollen-isotopes). For the analysis by species eight potential predictors entered our models, including the categorical variables year (of sampling: 2015 and 2016), month (of sampling: February to June), maturity (of the pollen at the time of sampling: -1 = immature, 0 = mature, 1 = withered inflorescence), slope (steepness: 0 = flat, 1 = minimal incline, up to 10˚, 2 = moderate incline, more than 10˚), water (proximity to water body: 0 = none in the direct vicinity, 1 = one between 10 and 20 m away, 2 = one up to ten metres away), water classification (type of water body, e.g. river, lake, wetland), soil (type of soil; Table 1) and the predictor site as a site-specific combination of latitude, longitude and altitude. For the second analysis (by site) the continuous variable altitude was additionally included to the categorical variables year, month, maturity, slope, water, water classification, soil and species and thus, nine potential predictors entered the second model. The factors were noted for each individual tree during field work. For all analyses, we deleted singletons by having a look at column variation (< 5 values in a column), hence, several analyses were performed using a subset of the predictors mentioned above. Categorical variables were hierarchically coded by maximizing the sum of squares between groups. Therefore, the analysis also informs about how levels in categorical predictors are associated with each other. For example, a notation such as site{FAN&STE&GOR&LIE-TAT&-TRE} (SI Dataset 1 and 2) contrasts sites Forêt d'Anlier, Steigerwald, Gorczański, and Liesjärvi against Tatrzański and Tre Cime.
Variables revealing the highest statistical significance were added to the model in a stepwise process. As a stopping rule for adding terms we used the minimum Bayesian Information Criterion. Subsequently, the model parameters were estimated using least squares regression.   Table 3). With the exception of the Gorczański site (2015) the δ 13 C pollen values of two consecutive years yield comparable medians at p(sm) = 0.26 and 0.17 (Fig 4). Mean δ 13 C pollen values of C. avellana pollen range from -29.1‰ to -25.3‰ (3.8‰; Table 3) and are normally  Fig 5). With one exception at Tre Cime (2015, p = 0.02), the δ 13 C pollen values of all sites are normally distributed and they yield similar medians for both vegetation periods. δ 13 C pollen values of B. pendula are also normally distributed. Their mean ranges from -25.7‰ to -22.9‰ (2.8‰; Table 3, Fig 5), but the distributions yield unequal medians for both years at the site Forêt d'Anlier (p(sm) = 0.0005). Mean δ 13 C pollen of C. betulus range between -26.1‰ and -25.6‰ (0.5‰) and the values reveal a normal distribution at each site (Table 3, Fig 5).

Results
The range of F. sylvatica mean δ 13 C pollen values is -27.9‰ to -25.5‰ (2.4‰) and the values are normally distributed with similar medians for both years at the site Gorczański (Table 3

Intra-tree variability, intra-annual variability and variability with elevation of δ 13 C pollen
Intra-tree variability of δ 13 C pollen . 64% of δ 13 C pollen values of samples taken at lower branches of the broad-leaved species A. pseudoplatanus A. glutinosa and C. avellana are more   Table 4). The isotopic values of the samples taken at different positions within the canopy are mostly ranging within one or two standard deviations from the mean isotope value of the tree. δ 13 C pollen values from pollen growing at the east side of a tree are likely to be higher, whereas values of the west tend to show lower δ 13 C pollen values in comparison to the mean isotope value of the trees. Deviations from the mean at northern and southern positions appear to be species-specific. The δ 13 C pollen value of A. pseudoplatanus is lower in the North (-0.7‰ from the mean value) and A. glutinosa yields lower values in the North in two out of three samples ( Table 4). The carbon isotope depletion in the North averages at -0.4‰ for A. glutinosa. In contrast, C. avellana exhibits higher pollen-isotope values in the North (+0.5‰) compared to the intra-tree average of this species. Twenty samples taken at different positions within the canopy of a single P. abies tree show higher δ 13 C pollen values at eastern and western positions and lower values at the southern exposition ( Table 5). The analysis of 34 individual inflorescences from low and high

PLOS ONE
Variability of carbon and oxygen stable isotope ratios of modern pollen positions demonstrates the intra-branch variability in pollen-isotopes of six neighbouring P. sylvestris trees (Table 6). δ 13 C pollen values from the same branch of one individual tree differ in a range of 0‰ to 1.2‰ ( Table 6). The average δ 13 C pollen difference between inflorescences from the same branch is 0.3‰.  (Fig 8).
Isotope variability of δ 13 C pollen with elevation. The δ 13 C pollen values of P. abies at Tatrzański Park Narodowy increase with altitude from -26.6‰ (1053 m a.s.l) up to -22.1‰ (1344 m a.s.l.) (Fig 9). The coefficient R 2 of 0.52 reveals a weak linear correlation with elevation.
Stepwise regression analysis for δ 13 C pollen By species. The most relevant factor influencing δ 13 C pollen values for several species is site with very high probability values of Prob > F = < 0001 � (Table 7; S1 Dataset). The year of  Table 3). Values of each year are normally distributed (Fig 4) Table 3, Fig 5).  Table 3) and are also normally distributed (Fig 5). The comparatively low mean δ 18 O pollen values of F. sylvatica range from 18.6‰ to 24.1‰ (5.5‰; Table 3) and the values are normally distributed with similar medians for both years from Gorczański (Fig 6). Mean Q. robur δ 18 O pollen values range between 26.2‰ and 27.4‰ (1.2‰; Table 3) and are normally distributed but statistically distinct between both years at p(sm) = 0.006 (Fig 6).  Table 3). The values are normally distributed within each sampling (Fig 7) but comparisons between 2015 and 2016 are statistically distinct for all sites.

PLOS ONE
Variability of carbon and oxygen stable isotope ratios of modern pollen comparison to the values of the higher branches from the same individual (Fig 3, Table 4). The δ 18 O pollen values were often lower at southern and western positions, whereas values from the northern position seem to be generally higher for all species (Table 4). Twenty samples taken at different positions within the canopy of a P. abies tree show higher δ 18 O pollen at eastern and western positions, and lower values at the southern exposition (Table 5). Similar to the broadleaved species, δ 18 O pollen values from lower branches of P. abies tend to be lower than those from higher branches ( Table 5). The analysis of individual inflorescences demonstrates an intra-branch variability ranging between 0‰ and 1.4‰ (  (Fig 8), but the medians are statistically distinct. The mean δ 18 O pollen values increase by 4‰ within 11 weeks (from 20.9‰ on 10 March to 24.9‰ on 5 May). δ 18 O pollen values of P. sylvestris are also normally distributed for each sampling (Fig 8).   Stepwise regression analysis of δ 18 O pollen By species. δ 18 O pollen values were found to be affected by variable factors. The most important factor was site, which impacts the δ 18 O pollen variability in six species (Table 7). The year of sampling influences the δ 18 O pollen variability in four species, whereas the maturity of the pollen determines the δ 18 O polle variability in three species and the factor month of sampling

PLOS ONE
Variability of carbon and oxygen stable isotope ratios of modern pollen occurs twice. The factors proximity to water and water classification are rarely influential and, if significant, they were species-specific. By site. δ 18 O pollen value variability within sites is mostly determined by the factor species ( Table 8). The year of sampling is important in four cases and the month of sampling and   Table 3). Catkins of both species occur before the leaf buds emerge (Fig 2), thus the pollen can only be polymerized from stored fatty and amino acids, phenols and other precursors of sporopollenin predominately accumulated during previous vegetation periods. Alnus glutinosa and C. avellana pollen are present and fully developed in shape before winter dormancy, but the size of maturing pollen grains increases in early spring until a few days before pollen shedding [53]. This indicates that some biopolymers are still added to the pollen grains and possibly exchanged right before pollination.
Ambient air temperatures shortly before the onset of flowering affect the catkin development of both species and the onset date for trees flowering in early spring is highly variable, even between consecutive years [54]. However, the responsiveness of the plant to favorable weather conditions does not seem to have any marked effect on its δ 13 C pollen values. The measurements indicate similar medians for consecutive years for A. glutinosa (p(sm) = 0.26 and 0.17; Fig 4), although flowering started roughly two weeks earlier at all sites sampled in 2016 (personal observation). Corylus avellana δ 13 C pollen values deviate more strongly between the two years (p(sm) = 0.07 and 0.004; Fig 4). But even though both species react differently over two vegetation periods, they show similar patterns between particular sites (Fig 4): The signals of the western sampling location of Forêt d'Anlier are similar to the signals of the eastern sampling location of Gorczański (Fig 1). Especially noticeable are the low values of δ 13 C pollen for A. glutinosa from Müritz (2016) compared to those of other sites. The mean δ 13 C pollen value is -31.0‰ and thus 2.8‰ lower than the average A. glutinosa pollen-isotope value of -28.2‰ from all sampling sites. However, all values lie within the range of a normal distribution (p = 0.81; Fig 4), so we can exclude measurement errors or tree-individual outliers to have caused the observed low mean δ 13  , but they show no consistent offset or similarity to any other species examined. In general, Acer ssp. are not dependent on prevailing spring temperatures to start seasonal development but instead require a specific amount of daylight [55]. Their leaves emerge early in the season and immediately exhibit a high rate of photosynthesis [56]. However, Acer ssp. are easily affected by short-term weather events and under unfavourable conditions, their carbon fixation during photosynthesis is rather ineffective, up to 50% less compared to the genus Quercus under the same conditions [57]. Additionally, A. pseudoplatanus is very sensitive to cold air inversions in spring and autumn [58]. Inter-annual variations in leaf senescence are highly correlated with precipitation: in dry years, high respiration rates cause early senescence and even premature leaf fall [59]. Therefore, photosynthetically active periods of Acer ssp. are highly variable even between consecutive years. Hence, their pollen-isotope values may be challenging to interpret.  Table 3). Catkins of the genera Quercus and Fagus emerge during the same period in late April or early May, simultaneously with their first leaves and twigs. This happens roughly two weeks before anthesis is complete and the pollen starts to shed [60]. They both carry winter-dormant leaf and flower buds [61], but the spring ontogeny is genus-specific [50].
Quercus robur shows a high sensitivity to favourable spring temperatures regarding bud burst, whereas F. sylvatica needs twelve hours of daylight to start developing [58]. Trees of the genus Quercus are slow in leaf unfolding and it takes about two months until full performance of carbon fixation in the leaves [62]. Thus, the length of the vegetation period in which carbon can be fixated and stored out of a positive net productivity from photosynthesis is shorter for Quercus in comparison to Fagus. In general, warmer spring temperatures tend to increase carbon uptake, whereas warmer summer and autumn temperatures decrease the uptake due to larger respiration rates [63]. The impact of prevailing temperature on pollen-isotope values during different stages of a vegetation period has yet to be examined in detail.
Betula pendula and Carpinus betulus. Mean δ 13 C pollen values of B. pendula (-24.3‰) are on average 1.3‰ less negative than the mean values of C. betulus (-25.6‰), even though they belong to the same family (Betulaceae), flower during the same period from mid-April to mid-May and prefer similar habitats (Figs 2 and 5; [60]). The total phase of pollen development spans early September to early April, and the size of B. pendula pollen grains still increases two to three weeks before pollen release [53], which implies a continuous addition and exchange of biomolecules during pollen maturation. Stach et al. [64] reported a positive correlation of aerial pollen counts of B. pendula with temperature and rainfall of the year before pollination. Thus, plant physiological reactions to prevailing environmental conditions can be assumed for B. pendula. Carpinus betulus is known to be particularly sensitive to frost damage and tends to prolong catkin proliferation when temperatures in spring are low. That affects the positive net productivity of photosynthesis in early spring [65] and leads to a shift of pollen development and maturation, which may have an effect on the δ 13 C pollen values.
Differences in pollen-isotopes within families and subfamilies. δ 13 C pollen deviate much less within one plant family than between families [14]. However, our findings show that this result does not apply to European taxa of the Betulaceae family. Four tree taxa of this family have been examined in this study (A. glutinosa, C. avellana, B. pendula and C. betulus). Taxonomically and genetically, the family is divided into two subfamilies: Coryloideae (including Corylus and Carpinus) and Betuloideae (including Betula and Alnus). Pollen grains of the members within the subfamilies are almost similar in shape and size [66]. However, due to their different flowering periods from January to March (A. glutinosa and C. avellana) and April to May (B. pendula and C. betulus), they are isotopically very different. We found that isotopic differences within these subfamilies were higher than between-family differences ( Table 9; differences within the Betuloideae δ 13 C pollen 3.9‰ and within the Coryloideae δ 13 C pollen 1.8‰). Analysing all Betulaceae pollen together would imply a loss of information in the isotope signal. Between-year mean δ 13 C pollen values of P. abies deviate by 0.2‰ and that of P. sylvestris by 0.3‰. Schwarz [13] reported low δ 13 C pollen ranges between four consecutive years for Pinus retinosa (0.53‰) and between three years for Pinus strobus (1.03‰). Carbon molecules (mostly soluble sugars from the previous vegetation periods) can be allocated in the individuals [68] and are remobilized directly after the resumption of growth in spring [13,69]. Thus, the usage of stored carbon as basic modules for the pollen might compensate seasonal variations in the δ 13 C pollen . Hence, between-year median δ 13 C pollen values (p(sm) ; Fig 7) of most sites cannot be statistically distinguished for both species. The stored carbon differs in age between 0.7 years [70] and up to ten years [68]. Tree growth and photosynthetic activity of P. sylvestris starts approximately 40 days prior to budburst [71]. Pinus sylvestris δ 13 C pollen values are found to correlate with temperature four to six weeks prior to pollen release [25]. Hence, it can be assumed that the plant uses newly fixated carbon in substantial portions to finish pollen maturation. There is no support for a correlation between North American Pinus-species δ 13 C pollen and temperature [13]. Thus, this correlation might be species-specific to P. sylvestris [22].
Intra-site pollen-isotope variability of the δ 13 C pollen values of P. abies range between 0.6‰ and 5.5‰ and that of P. sylvestris range between 2.5‰ and 4.0‰. Factors determining the intra-site isotope variability of C. atlantica and herbaceous species are microclimate, physiological differences between the individuals, water and nutrient availability and the number of trees in the direct vicinity [15,18]. These factors may also account for the isotope variability of P. abies and P. sylvestris. Compared to P. abies, P. sylvestris has a broader physiological tolerance range to a variety of environmental conditions [67,72]. Hence, with larger plasticity, physiological reactions turn out smaller and their pollen-isotope values fluctuate less within one location. In general, the pollen-isotope ranges of both species are broader in pollen sampled in 2015 than in samples of 2016.
Intra-tree variability of δ 13 C pollen δ 13 C pollen values vary between 1.1‰ and 3.5‰ within an individual tree (Table 4). In most cases the values are higher at the eastern exposed side than at the southern and western sides. With less insolation, open stomata do not discriminate as much against the heavy aerial δ 13 C isotopes [73]. That phenomenon is also expressed by circumferential variations of 1-3‰ in leaf-tissue δ 13 C [73] and Leavitt [74] mentioned an average of 0.5-1.5‰ deviation from the mean within a circumferential tree ring cellulose analysis. In addition, the difference between mean δ 13 C pollen values can be up to 3‰ within a height difference of 6 m (Table 4). For 64% of the individuals, δ 13 C pollen values of the lower samples of a tree are lower compared to the upper ones (Table 4). These results agree with Schleser [73] who described an enrichment of 13 C in leaf-tissue of 1-4‰ from bottom to top within one tree. The intra-tree variability between several inflorescences within one branch can be as high as 1.2‰ for δ 13 C pollen (Table 6). However, average differences of pollen-isotope values in neighbouring inflorescences of P. sylvestris are as little as 0.3‰ (δ 13 C pollen , Table 6).
Due to different carbon and oxygen sources the δ 13 C pollen shows a higher intra-tree variability than δ 18 O pollen (Table 6). Carbon is fixated in the leaves as the product of photosynthesis. The rate of photosynthesis varies in relation to the position of the leaf on the tree, which leads to isotopic differences between the cardinal directions. Stable isotopes of tree ring cellulose and other plant material vary in one individual [75,76] and it is generally suggested to pool several cores/samples in order to get the average representative isotope weight for the tree/year relation [2]. and the pollen maturation continues after a winter dormancy. In spring, the catkins emerge at the same time as the first leaf buds. Photosynthesis is not yet profitable this early in the year, thus the trees use stored and pooled carbon molecules to build new plant tissue in the beginning of the vegetation period [77]. This explains why the carbon composition of the pollen remained constant.
Pinus sylvestris. Mean δ 13 C pollen values of P. sylvestris did not change within 12 days (-26.3‰ and -26.3‰; Fig 8). Comparatively little is known about timing, exact molecular processes and chemical compositions during pollen development [78,79]. Plants may use only storage molecules to build plant tissue in spring and early summer [13,77]. However, Loader and Hemming [22] reported a correlation between δ 13 C pollen and temperature approximately six weeks prior to pollen release. Thus, at least in parts, P. sylvestris uses newly accumulated carbon molecules to build pollen. Perhaps we detected no change in the δ 13 C pollen of P. sylvestris because the sporopollenin of the grain wall had already been synthesised during the previous weeks.  Fig 9). Although the increase of 3.4‰ is statistically distinct, it cannot be exclusively attributed to an altitudinal effect. In general, the discrimination of δ 13 C is linearly related to the ratio of intercellular to ambient CO 2 partial pressures, and thus δ 13 C values increase with increasing altitude [80]. Hultine and Marshall [81] found a linear relationship between δ 13 C from Picea ssp. needle tissue and elevation where the values of the δ 13 C needle increase by approximately 0.5‰ per 300 m in altitude, whereas Warren et al. [82] reported an increase of 2.5‰ over 1000 m of δ 13 C wood from 14 different coniferous species. The P. abies individuals chosen for the elevation transect were all located on the same slope facing north, thus ensuring comparable environmental conditions of e.g. wind, insolation and precipitation. However, the inflorescences from higher elevations were still immature, whereas they were already flowering at lower altitude. Flowering of P. abies starts with a delay of three days with every additional 100 m of elevation [83]. Bell et al. [15] found that local environmental constraints mask the effect of altitude. They concluded that the source of air masses and moisture input seem to determine the pollen-isotopic weight more than the actual altitude in the mountains. In addition, Treydte et al. [84] pointed out, that an existing temperature signal of P. abies in δ 13 C tree ring was independent of elevation. Influencing factors on pollen-isotope values are very variable in mountainous areas. Which combination of factors caused the high 3.4‰ isotope offset over 300 m remains uninvestigated for now.

Stepwise regression analysis
Factors influencing δ 13 C pollen of each species. δ 13 C pollen values of all species are determined by several variable factors, and the outcome of the statistical analysis is sometimes inconclusive ( Table 7). The factor site strongly influences C. avellana (Prob > F = <0.0001), where the maritime site (Forêt d'Anlier) contrasts the continental sites (Gorczański, Steigerwald, Müritz). Picea abies δ 13 C pollen values seem to be grouped by altitude, where the mountainous sites of Tre Cime and Tatrzański are contrasting the lower-altitude sites (Forêt d'Anlier, Steigerwald, Gorczański and Liesjärvi). However, these groupings could also be caused by genetic predispositions affecting phenological responses [65,85,86]. Pinus sylvestris is known to have different haplotypes in Europe, one of which is restricted to the Southern Alps. Another haplotype of P. sylvestris spreads throughout central Europe [87] and can be found at all other sites of this study. Pinus sylvestris has distinct δ 13 C pollen values at northern and southern European sites, which seems to be caused by a different reaction to local temperatures due to their genetic background [22].
Factors influencing δ 13 C pollen at each site. δ 13 C pollen values are mostly determined by the factor species which largely overprints other local non-climatic factors (e.g. the proximity of the tree to the next water body, type of soil, slope angle; Table 8). Species groupings of the factor species at each site are variable and do not seem to follow plant family affiliation or flowering period. In addition, the factor maturity is important for the mountainous sites Tre Cime and Tatrański.

General considerations for the usage of δ 13 C pollen in palaeoclimate studies
Species-specific isotope values and patterns display susceptibility to different environmental factors due to their phenology and physiology (Table 3; Figs 4-7). Therefore, the pollen must be separated by species prior to isotope analysis. Without separation, the individual isotope signal within the pollen will be lost and the method cannot be applied in palaeoclimate studies [14,38]. To date, the separation process is done manually which makes it very time-consuming to pick a sufficient amount of pollen for stable-isotope analysis [38,88]. The amount of pollen needed to obtain representative average isotopic values is variable. Bell et al. [38] suggest to use a sample size of at least 30 μg carbon when using standard stable-isotope measuring techniques. New technology makes it feasible to decrease this amount. Nelson et al. [35] developed an isotope measuring technique using single pollen grains in order to estimate their photosynthetic pathway and Roij et al. [88] even measured as little as 0.042 μg carbon with a precision better than 0.5‰. However, we identified significantly different pollen-isotope values within one site and even within one tree and therefore suggest using a higher amount of carbon/pollen depending on the pollen type to obtain significant mean pollen-isotope values.
Furthermore, mostly sporopollenin of the pollen wall is preserved in fossil pollen [25]. The analysis of raw pollen material, as reported by this study, shows the general variability between species and sites, but patterns and ranges cannot directly be compared to fossil pollen-isotope values [14]. To accomplish that, extant pollen will have to be treated chemically prior to analysis. The method of sporopollenin-extraction with sulfuric acid has been tested for several species and a consistent offset between raw and the chemically treated pollen material was found [15,25]. The same accounts for the species of this study: treatment with sulfuric acid resulted in a stable but species-specific offset when compared to raw pollen material (unpublished data).  Table 3). The trees sampled for this study grew on saturated soils with ample water supply out of consistent precipitation throughout the winter months and snowmelt in February and March. Therefore, both species did not face water stress during times of sampling, which may explain the observed low δ 18 O pollen values. Alnus glutinosa and C. avellana generally prefer areas with aggravated or missing drainage close to a streamside [89]. In case of insufficient water supply, riparian trees are able to switch sources and especially A. glutinosa is drought resistant [47]. In general, there is no fractionation of oxygen isotopes during water uptake by the roots [90]. The xylem water remains unaltered during transport until it reaches evaporative tissue or is used to build plant tissue, such as flower primordia and pollen grains [91]. Since A. glutinosa and C. avellana flower before leaf proliferation, the water is not passing transpiring leaf tissue, where the isotopes would be altered by evaporative 18  Quercus robur is slower in leaf unfolding, water uptake and in establishing a positive net productivity out of photosynthesis [58,62]. Oxygen isotope values of precipitation are enriched by evaporation before seeping into the soil, especially during summer months. An increased activity of Q. robur in summer and the uptake of water enriched in 18 (Table 3). However, we do not have a valid explanation for this site-specific phenomenon at the moment.
Betula pendula and Carpinus betulus. Mean δ 18 O pollen values of species of the Betuloideae deviate by more than 3‰ where C. betulus (27.1‰) yields higher values than B. pendula (24.1‰). In general, the mean δ 18 O pollen values within the species are very similar between the investigated sites and do not deviate much from the overall mean (Fig 5, dotted line). In particular the 2016 oxygen isotope values of B. pendula are very similar (Fig 5, black bar), implying that the δ 18 O pollen values of these two species are less affected by local environmental events but rather by trans-regional trends. Isotope values of other species deviate more between sites per year (Figs 4, 6 and 7).  Fig 7). Schwarz [13] pointed out that oxygen pollen-isotopes are highly sensitive to environmental conditions. We expect the varying interaction of all local conditions (e.g. micro-climate at the sampling site and water availability) to have caused the different δ 18 (Fig 7). Both species display the same patterns, therefore weather conditions, which were not investigated further in this study, are expected to have caused the δ 18 O pollen inversions. The North Atlantic Oscillation (NAO) is known to influence phenology and temporal variability of seasonal tree development in Europe [92]. The site of Forêt d'Anlier lies closest to the Atlantic Ocean and is therefore more strongly influenced by the NAO than more inland-located sites.
In general, δ 18 O values of precipitation become more negative with increasing altitude and latitude. The δ 18 O cellulose of coniferous trees shows a clear latitudinal pattern at a transect in North America covering the area between 65˚N and 33˚N [93]. Our study covers an area from 60˚N to 46˚N, but the pollen-isotopes do not show any pattern resembling this latitudinal effect. The northernmost site (Liesjärvi) yields low δ 18 O pollen values for P. sylvestris  (Table 4). In general, lower samples yield lower isotope values than upper samples. In addition, the southern and western samples yield lower isotope values than the northern samples ( Table 4). The δ 18 O pollen intra-branch variability can be as high as 1.4‰ (Table 6), but the average difference between neighbouring inflorescences is 0.5‰, which can be considered to lie within a natural tolerance range. There is also a circumferential variability of pollen-isotopes at different cardinal directions of up to 3.5‰ for δ 13 C pollen and 2.1‰ for δ 18 O pollen . Because of that variability, an appropriate amount of pollen is needed for stable isotope analysis to enhance the precision of the measured mean δ 13 C pollen and δ 18 O pollen values. 5. δ 13 C pollen values increase with elevation by 3.4‰ over 300 m. Even though the two variates are significantly correlated, the difference is too high to be explained by altitude alone. Thus, additional unknown factors must have influenced the isotopic trend. δ 18 O pollen values do not change linearly with elevation.
6. Stepwise regression analysis reveals that the most important factor determining pollen-isotope values of any species in this study is their geographic location (factor site). In addition, the statistical analysis shows that the pollen-isotopes within one site are mostly determined by the factor species.
Stable carbon and oxygen isotope analysis of fossil pollen can improve the precision of palaeoenvironmental investigations. The applicability of different pollen types in palaeoclimate research relies on the specific plant physiological traits as well as on the pollen abundance in fossil archives. However, due to species-specific pollen-isotope ranges and patterns, fossil pollen should be separated by species as thoroughly as possible. After separation, varying vegetation periods and ecological preferences of the host-plant will allow climate reconstructions on an intra-seasonal scale.
Supporting information S1 Dataset. Detailed results of the stepwise regression analysis: By species. Eight variables influencing the stable isotope composition in pollen (δ 13 C, δ 18 O) for each species were explored by means of stepwise regression analysis. They include the categorical variables year (of sampling), month (of sampling), maturity (of the pollen at the time of sampling), slope, water (proximity to water body), water classification (type of water body), soil and site.
(DOCX) S2 Dataset. Detailed results of the stepwise regression analysis: By site. Nine variables influencing the stable isotope composition in pollen (δ 13 C, δ 18 O) at each site were explored by means of stepwise regression analysis. They include the continuous variable altitude and the categorical variables year (of sampling), month (of sampling), maturity (of the pollen at the time of sampling), slope, water (proximity to water body), water classification (type of water body), soil and species.
(DOCX) d'Anlier), Nils Altvater, Ralf Pauli (both Müritz NP) and Pawel Czarnota (Gorczański NP) for regional information about the national parks, local phenology and seasonality as well as pollinator behaviour. We also thank two anonymous reviewers for many positive comments and suggestions to improve an earlier version of the manuscript.