Pollen and spores from molar folds reflect food choice of late Pleistocene and Early Holocene herbivores in The Netherlands and the adjacent North Sea area

Molars of eight large herbivore species (Megaloceros giganteus, Cervus elaphus, Rangifer tarandus, Alces alces, Bison priscus, Ovibos moschatus, Coelodonta antiquitatis and Stephanorhinus kirchbergensis) were collected ex situ from Pleistocene and Holocene sands dredged in the North Sea, and from Dutch inland sites. Folds in many molars contained compacted masticated plant remains, and also microfossils. We identified pollen, spores, and non-pollen palynomorphs and discuss and interpret food preferences, represented flowering seasons, or parts of flowering seasons, and we discuss effects of changing vegetation composition in relation to climate and age of the molars, based on the pollen spectra. Various confounding factors have contributed to the recorded pollen composition, but nevertheless the pollen spectra show valuable aspects of vegetation composition, food choice, age, and landscapes, from subarctic open areas to interglacial forest. Ecological and statistical analysis of the results shows dietary differences between the mammal species analyzed. © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
The mammoth steppe ecosystem contained a combination of plant and animal species that does not occur anywhere today. It was characterized by diverse megafauna of mostly herbivorous species (Guthrie, 1990). Food choice is an interesting and crucial part of a species' ecology, and understanding the diet of animals within an ecosystem helps to better understand the system (Owen-Smith, 2002). Hofmann (1989) classified extant ruminants, based on extensive dietary work on fresh specimens into concentrate selectors (selecting high quality food, rich in plant cell contents), grass/roughage eaters (selecting low quality food, rich in plant cell wall, cellulose) and intermediate, or mixed feeder types. There is a link with body size: the bigger the ruminant, the more suited is its digestive tract to process large volumes of low quality diet. Most large ruminants (e.g. bison) are classified as grass/roughage eaters, while most medium sized species of deer (including reindeer and red deer) are intermediate types. The moose is an exception, as it can be clearly classified as a concentrate selector, despite its large body size (Hofmann, 1989). Herbivores actively select certain plants from a given area based on their physiological needs and past experiences. Even large grazing animals, classified as roughage eaters, do not indiscriminately graze all plants (e.g. Villalba and Provenza, 2009;Kleynhans et al., 2011). Here, we use observations on fossil molar contents of ruminants and non-ruminants to assess fossil mammal diets in the mammoth steppe ecosystem and during the Early Holocene.
In 2011 an artificial peninsula called 'Zandmotor' was created along the shoreline of The Netherlands south of The Hague. For that purpose late Pleistocene and Early Holocene sediments were dredged from the relatively shallow North Sea bed. The new land functions as a coastal reinforcement. 'Maasvlakte 2' is a westward extension of the port of Rotterdam and it was also built from dredged sandy North Sea deposits. Three sediment units were dredged: from marine Holocene/recent sandbars, from Early Holocene deposits, and from underlying (predominantly) Late Pleistocene sandy sediments. The dredged sediments were originally deposited in ancient Rhine and Meuse deltas (Laban and Rijsdijk, 2002;van der Valk et al., 2011;Langeveld, 2013;Kuitems et al., 2015) when, worldwide, sea levels were much lower than today and therefore the southern part of the North Sea area was dry land.
Maasvlakte 2 and Zandmotor became localities for collecting fossils by citizen scientists, including mammal bones from Pleistocene (Weichselian) and Early Holocene sediments (Mol et al., , 2008van der Valk et al., 2011). The fauna was dominated by large herbivores, such as woolly mammoth, woolly rhinoceros, steppe bison, giant deer and wild horse. Radiocarbon dated material shows that representatives of this fauna lived between c. 50,000 and 25,000 BP when North West Europe was part of the mammoth steppe ecosystem (Mol et al., , 2008. A study of the botanical infill of folds (infundibula) in a molar of a mid-Weichselian giant deer from Zandmotor yielded wellpreserved pollen and shed light on the diet of giant deer (van Geel et al., 2018). For the present study we selected a number of species to study the pollen contents of molar infundibula. The selected species were the ruminants Megaloceros giganteus (giant deer), Cervus elaphus (red deer), Rangifer tarandus (reindeer), Alces alces (moose), Bison priscus (steppe bison) and Ovibos moschatus (muskox), and the non-ruminants Coelodonta antiquitatis (woolly rhinoceros) and Stephanorhinus kirchbergensis (forest rhinoceros; the fossil record of this species ended shortly after the Eemian interglacial). We borrowed molars of these species from various private collections and museums in The Netherlands. Most molars were from North Sea deposits but some of the molars were recovered from dredged inland river deposits (Table 1). The molars were identified to species level through comparison with molars of identified skeletons in the reference collection of the Natural History Museum Rotterdam (NMR). Botanical material present in infundibula was sampled for the analysis of pollen and non-pollen palynomorphs (NPP). Quantities of organic material were not sufficient for macrofossil analysis and radiocarbon dating.
Pollen and spore records from cored lake sediments and peat deposits show regional trends of vegetation changes. The vegetation composition in The Netherlands during mid-Weichselian interstadials is recorded in detailed pollen records (Brinkkemper et al., 1987;Ran, 1990). For the vegetation history of the Lateglacial period in The Netherlands we refer to Hoek (1997) and for the Holocene to van Geel et al. (1981).
Palynologists are accustomed to the fact that wind-pollinated taxa produce high amounts of pollen, and that such taxa therefore are over-represented compared to the relative abundance of the source plants in the vegetation. In contrast, insect-pollinated taxa produce relatively low amounts of pollen and are thus underrepresented. A sediment sample or a peat sample normally represents a multi-annual mixture of pollen deposition. The interpretation of the presence and frequencies of pollen and spore taxa from molar infundibula in terms of past vegetation cover is different, and the pollen records will be biased for several reasons. For the interpretation of our results we have to keep in mind that there remain some basal, still unsolved questions about the indicative value of such samples (chronology, vegetation composition, phenology, food choice, age of the material).

Megaloceros giganteus (Blumenbach, 1799) -giant deer
The extinct M. giganteus could attain a maximum body weight of 600e700 kg (Moen et al., 1999); this matches modern moose. The fossil record can be traced back to at least the early Late Pleistocene in Europe (Kahlke, 1999). The North Sea has yielded numerous specimens, some of which were radiocarbon dated. The youngest is 36,300 ± 1100 BP and the oldest is > 45,000 BP. The species is known from both glacial and interglacial (or interstadial) deposits (Lister and Stuart, 2019). Chritz et al. (2009) studied stable isotopes in dentition and concluded a grass-and forb-based diet, supplemented by browsing. Rivals and Lister (2016) and Saarinen et al. (2016) studied meso-and microwear and concluded that M. giganteus had a wide range of dietary behaviour including grazing, leaf browsing and mixed feeding, but generally tending towards mixed feeding and grazing. The diet of one specimen, based on pollen in molar folds, was reviewed by van Geel et al. (2018): it was completely dominated by Artemisia, indicating that the animal foraged in a steppe environment and may have preferred to eat Artemisia, which contain a high level of mineral nutrients such as calcium and phosphorous (Klein, 1965;Ashraf et al., 2010;van Geel et al., 2018).

Cervus elaphus Linnaeus, 1758 -red deer
C. elaphus is very flexible in its choice of habitat and diet. Hofmann (1989) classifies it as a mixed feeder. This large deer (75e340 kg; Nowak, 1991) is known from both glacial and interglacial fossil deposits since at least the Middle Pleistocene (Rivals et al., 2009). The Dutch and North Sea fossil record includes many Early Holocene occurrences, especially worked antlers from the Mesolithic (e.g. Louwe Kooijmans, 1970;Mol et al., 2008). There are no Late Pleistocene radiocarbon dates, but the state of fossilisation of some of the red deer material suggests that the species was part of the Late Pleistocene fauna as well (Mol et al., 2008). C. elaphus currently has a very extensive range, including much of Europe, Asia and North America. It occurs in various habitats, ranging from coniferous and deciduous forests to open grasslands, heathlands, moorlands and even semi-arid environments. Its diet reflects its broad habitat range and varies from grass to sedges, ferns, mosses, lichens, fungi, twigs, nuts, shoots and bark of trees. It usually varies from season to season and whenever possible includes both herbaceous and browsed material (Straus, 1981;Nowak, 1991;Chen et al., 1998;Clutton-Brock and McIntyre, 1999;Gebert and Verheyden-Tixier, 2001;Suter et al., 2004;Augustyniak, 2010;Ligi and Randveer, 2012;Berlioz et al., 2017). A similar flexibility in the diet of C. elaphus was found through isotopic and dental wear analyses for Pleistocene populations (Drucker et al., 2003;Rivals et al., 2009Rivals et al., , 2010Saarinen et al., 2016).

Rangifer tarandus (Linnaeus, 1758) -reindeer
R. tarandus is one of the large mammals that lived on the mammoth steppe and it survived into the Holocene. Hofmann (1989) classifies it as a mixed feeder. This moderately sized deer (body weight 60e318 kg; Nowak, 1991) has a fossil record that starts at least in the Early Pleistocene. It was common in the European Weichselian, especially during the Magdalenian period. Its range extended from northern Spain into North America (Kahlke, 1999). Radiocarbon dates from The Netherlands and the North Sea range from >45,000 BP till 39,000 þ 700, À600 BP and an isolated date at 29,460 ± 250 BP . Today, reindeer populations may migrate for up to 1000 km between summer grounds in the tundra and winter grounds in forested areas. They feed on a variety of plants, including lichens and leaves of shrubs and fine twigs (Nowak, 1991;Augustyniak, 2010). Using dental mesowear analysis, Saarinen et al. (2016) found that Pleistocene reindeer browsed for soft forage even in more open vegetation of the mammoth steppe. Using carbon and nitrogen isotopes, Fox-Dobbs et al. (2008) reconstructed Pleistocene reindeer diet in eastern Beringia as consisting mostly of lichens, mosses and fungi. Utilizing both meso-and microwear, Rivals et al. (2010) and Rivals and Semprebon (2017) concluded that reindeer from the Brown Bank in the North Sea were mixed feeders, possibly with a more abrasive diet than modern populations.

Alces alces (Linnaeus, 1758) -moose
A. alces is the largest extant deer species with a body weight of 200e825 kg. Hofmann (1989) classifies it as a concentrate selector. It occurs in the northern parts of Eurasia and North America (Nowak, 1991). Fossil moose are known from The Netherlands and the North Sea deposits. They date from the Early Holocene until late medieval time (Walch, 2000;van Geel and van Wijngaarden-Bakker, 2002;Mol et al., 2006;Spinney, 2018), but there is one date of 44,560 þ 1840/-1490 BP (Mol et al., 2008). Moose are rather specialized feeders that live in forested areas. They mainly feed on willow (Salix spp.), but they also forage on water plants and can swallow with their head submerged (Dungan and Wright, 2005;Shipley, 2010). Rivals et al. (2010) studied meso-and microwear of A. alces from the North Sea (Brown Bank) and reconstructed a strictly non-abrasive browsing diet, comparable to the modern diet. Saarinen et al. (2016) reached the same conclusion.
2.5. Bison priscus Bojanus, 1827 -steppe bison B. priscus was a characteristic component of the Mammuthus-Coelodonta Faunal Complex (Kahlke, 1999). Recent genetic evidence showed that there were two distinct types of Bison in the Late Pleistocene of Europe, including the North Sea plain (Soubrier et al., 2016;Grange et al., 2018), but they appeared to be morphologically indistinguishable and therefore we here use B. priscus. They were large animals, with a body weight of around 1000 kg (Boeskorov et al., 2014). A radiocarbon date of 45,350 þ 2400/-1850 BP (Mol et al., 2008) places B. priscus in the Late Pleistocene of the North Sea area. Bison has a very rich fossil record across all of its geographical range. It appears to be the most common taxon from the North Sea, but paleoecological data on the diet are scarce. Guthrie (1990) provided data on North American B. priscus. Using cuticle analysis of plant material obtained from molar infills and mesowear analysis, he concluded that these animals were grazers, with grasses making up 80e90% of their diet (followed by other herbs). Rivals et al. (2010) studied microwear of B. priscus from the North Sea (Brown Bank) and concluded that they were primarily grazers, or mixed feeders with a grass-dominated diet. Saarinen et al. (2016) studied mesowear of fossils from the United Kingdom and Germany and reconstructed mixed feeding diets with a significant component of grass.
2.6. Ovibos moschatus (Zimmermann, 1780) -muskox O. moschatus weigh 200e410 kg. Hofmann (1989) classifies it as a mixed feeder. Their fossil record in Europe starts in the late Middle Pleistocene (Kahlke, 1999). Together with reindeer, muskox is one of the few large mammals of the mammoth steppe that survived the end of the Pleistocene in Eurasia with the youngest radiocarbon date from the Taimyr Peninsula at 2900 ± 60 BP (MacPhee et al., 2002). Today they occur only on arctic tundra, with a preference for moist habitats in summer and windy higher areas (thus with little snow cover) in winter. Currently they occur in parts of the far north of Eurasia, where they were reintroduced in the 20th century, e.g. in Norway, the Taimyr Peninsula, Yakutia, North America and Greenland (Nowak, 1991). Remains of O. moschatus are rare in The Netherlands and the North Sea area; they have been dated to the Late Pleistocene . Modern O. moschatus feed primarily on grasses, sedges and dwarf willows. They are not very selective feeders and in winter they do not travel extensively to suitable foraging grounds (Klein, 1991;Oakes et al., 1992;Klein and Bay, 1994;Kristensen et al., 2011). Carbon and nitrogen isotopic analysis of fossil O. moschatus from the Holarctic suggests a diet flexibility that followed changes in precipitation (Raghavan et al., 2014).
2.7. Coelodonta antiquitatis (Blumenbach, 1799) -woolly rhinoceros C. antiquitatis was a characteristic component of the Mammuthus-Coelodonta Faunal Complex (Kahlke, 1999). It was a large animal (over 2000 kg) and it occurred over a large range, stretching from Great Britain and Spain across Europe to northeast Siberia, northern China and Mongolia; it never reached North America (Stuart and Lister, 2012). It is a common species in Late Pleistocene fossil assemblages, though never as common as woolly mammoth, steppe bison and wild horse. Its fossil record extends at least back to the late Middle Pleistocene (Kahlke, 1999). Its range contracted eastwards from 35 ka and it went extinct ca. 14 ka (Stuart and Lister, 2012). A radiocarbon date from the North Sea places the species in the Late Pleistocene at 39,910 þ 1070/-950 BP . Much of its ecology has been studied from permafrost and natural paraffin wax preserved carcasses, including stomach contents and molar infills. Both skeletal and soft tissue morphology (Stuart and Lister, 2012) as well as these food remains (Boeskorov et al., 2011) characterize C. antiquitatis as a true grazer, well adapted to the tough and dry mammoth steppe vegetation. While mesowear analysis from the North Sea (Brown Bank) and adjacent areas supports this, microwear analysis of North Sea material suggests periodical inclusion of some woody components in its diet (Rivals et al., 2010;Saarinen et al., 2016).

2.8.
Stephanorhinus kirchbergensis (J€ ager, 1839) -forest (or Merck's) rhinoceros S. kirchbergensis was a large animal with relatively long legs (Fortelius et al., 1993). It is considered an interglacial species with a fossil record in Europe starting in the Middle Pleistocene at 0.6 Ma and ending shortly after the Eemian interglacial (van der Made, 2010;Mol and van der Plicht, 2012). Analysis of plant macroremains trapped in molar folds and mesowear from specimens recovered from Germany and Russia indicates that the species was a mixed feeder that tended to browse on foliage of trees or shrubs, but also included grasses and herbs in its diet (van der Made and Grube, 2010;van Asperen and Kahlke, 2015;Kirillova et al., 2017). Based on mesowear analysis, Saarinen et al. (2016) reconstructed a browsing diet for this species.

Material and methods
The molar folds of about half of the screened specimens appeared to be empty or too poor in pollen to get sufficient microfossil spectra, while some other molar samples contained badly preserved microfossils. Such samples had to be discarded. Molar folds of Megaloceros giganteus (Mg), Cervus elaphus (Ce), Rangifer tarandus (Rt), Alces alces (Aa), Bison priscus (Bp), Ovibos moschatus (Om), Coelodonta antiquitatis (Ca) and Stephanorhinus kirchbergensis (Sk) contained plant remains and microfossils were successfully retrieved. Fifty-two samples were analyzed for pollen, spores and non-pollen palynomorphs (NPP; van Geel, 2001); the microscope slides are stored in the collection of the Natural History Museum Rotterdam, accession number 19e112. For practical reasons we abbreviate the zoological species names in Table 1, in the plotted DCA, and in the text below. Table 1 shows the numbering of samples, the collection codes, the location of the sites, and the molar positions in the jaws.
Sampling of the botanical material occurred by using clean preparation needles in order to separate the plant remains from the molar folds. The obtained palynological 'mini-samples' were collected in small beakers and a standard pollen preparation method was applied (Faegri and Iversen, 1989). Microscopic analysis was applied using Beug (2004) and Punt et al. (1976e2009) for difficult pollen identifications. Detrended Correspondence Analysis (DCA; Hill and Gauch, 1980) was applied in PAST 3.16 (Hammer et al., 2001) as an addition to a conventional ecological interpretation.

Results
In most cases more than three hundred pollen grains and spores were counted per sample, but for some molars that sum could not be reached. Calculation of percentages, as shown in Fig. 1AeD, was based on sums of pollen grains and spores of vascular cryptogams. NPP frequencies were also calculated as percentages of those sums. The diagrams show the percentages of pollen grains, spores and NPP. Observations of monospecific aggregates of pollen grains (each aggregate was recorded as 1 pollen grain) are shown in the diagram with a black square. Such aggregates (Fig. 2AeD) may indicate that flowering plants with unripe pollen grains were consumed, thus providing extra information about the herbivore diets.

Megaloceros giganteus
Mg03, Mg08, Mg10 and Mg11 show relatively high Betula percentages. In Mg08 and Mg11 aggregates of Betula pollen show that flowering birch was consumed. Alnus and Corylus with some Tilia and Ulmus pollen in Mg11 suggest that the sample may represent post-depositional Holocene contamination, but an upper Eemian or an early Weichselian interstadial age cannot be excluded. Artemisia and Asteraceae sf. Asteroideae are common in most of the Mgsamples. Other light-demanding taxa (Poaceae and a variety of herbaceous taxa, like Caryophyllaceae, Chenopodiaceae, Filipendula, Linum, Plantago, Thalictrum and Helianthemum) are of regular occurrence. Humulus was also consumed, considering the presence of aggregated Humulus pollen in Mg04. Equisetum shows high values in Mg07 and Osmunda was very common in Mg03, but the absence in other Megaloceros samples shows that they were not regular components of the giant deer food choice. Pollen of Poaceae is present in all samples analyzed, but always in low percentages; only in Mg05 a slightly higher percentage and an aggregate of unripe grass pollen was observed. This matches work based on isotopes and tooth wear that suggests mixed feeding for M. giganteus. The dominance of Artemisia, that was observed by van Geel et al. (2018) for one giant deer molar, was found here in other molars as well, also with aggregates of unripe Artemisia pollen. This provides further evidence of M. giganteus selecting Artemisia as a food plant. Saarinen et al. (2016) found that M. giganteus was commonly associated with rather open vegetation, but also occurred in localities relatively rich in arboreal pollen.

Cervus elaphus
Based on the relatively high numbers of Corylus and Alnus and ferns, samples Ce07, Ce11, Ce 12, Ce13, and Ce14 may point to a Holocene (but pre-inundation) age. The other Ce-samples are probably of Pleistocene age. Remarkable are Ce04 and Ce10 with high Betula values. A Betula aggregate in Ce04 points to consumption of flowering birch. Ce05 contained a very large number of Helianthemum and aggregates show that Helianthemum was consumed. The quantity of Humulus and Impatiens in Ce06 probably points to consumption, while the pollen record of Ce06, Ce15 and Ce16 may point to Urtica as a food plant. We suppose that Ce08 and Ce12 point to Filipendula as a food plant, while Ce11 may indicate that Equisetum was eaten. Ce08 and Ce14 point to consumption of Symphytum. Asteroideae, Apiaceae, Poaceae and Cyperaceae were common in many samples. Saarinen et al. (2016) found that Cervus elaphus was clearly associated with both open and closed forest environments.

Rangifer tarandus
Most samples show relatively low tree pollen values, pointing to a sub-arctic open landscape. In Rt08 only few individual Betula pollen grains were recorded, but an aggregate of B. nana shows consumption of dwarf birch.
Rt05 and Rt07 show rather high Salix pollen values and, considering the composition of the pollen spectra of those two samples this Salix pollen may well have been produced by dwarf willows. Artemisia is common in most samples; Rt03 and Rt08 yielded aggregates of Artemisia. Aggregates in the two samples with high values of Cyperaceae (Rt02 and Rt06) point to a bias (consumption and over-representation in the pollen spectra) of flowering Cyperaceae. Several samples show rather high records of Helianthemum, and three even show aggregates, indicating a preference for Helianthemum as a food source. According to Milotic et al. (2017) herbivores in the landscape even stimulate the growth and expansion of Helianthemum.
The frequencies of Pedicularis palustris-type in the Rt-samples (absent in almost all other samples) point to Pedicularis (P. lapponica?) as a preferred species for consumption. Some other taxa worth to be mentioned are Centaurium, Papaver rhoeas-type and Polemonium in Rt09, Triglochin in Rt11 and Silene dioica-type in Rt07. Poaceae pollen generally is slightly more common in R. tarandus than in M. giganteus, but lower than in C. elaphus, suggesting that grasses were part of the reindeer diet, but not to a very large extent. The pollen data match studies on meso-and microwear of reindeer specimens from the North Sea. Based on pollen records Saarinen et al. (2016) found that Rangifer tarandus was evidently associated with open environments.

Alces alces
Considering the amount of tree and shrub pollen (Betula, Pinus, Corylus, Alnus, Taxus, Tilia, Ulmus, Frangula alnus) and fern spores, and taking into account the geological information of the dredged deposits, Aa01 and Aa02 are of Early to mid-Holocene age. An aggregate of Alnus pollen in Aa02 probably points to consumption of Alnus. Aa03 is older than the other two samples (maybe Allerød interstadial, or a Preboreal age), because Alnus and Corylus are almost absent, while it shows high Pinus pollen percentages and even a Pinus tracheid pit.
Aa03 shows indicators of open water (Nymphaea, including trichosclereid, Myriophyllum, Potamogeton, Sparganium, Typha, and various algae (Botryococcus, Pediastrum, Spirogyra, Tetraedron, Zygnema-type, Mougeotia, HdV-128A and HdV-128B)). The Equisetum spores may well have been produced by the wet growing species E. fluviatile. Aa01 and Aa02 represent relatively 'dry' environments with, among others, Calluna, Lonicera and ferns. Aa02 contains Salix epidermis: a clear indication of consumption of Salix. Willows are very important food plants for modern moose. In concert with our results, Saarinen et al. (2016) found that Alces alces -with low mesowear values -was linked with low non-arboreal pollen percentages, which supports the status as a forest-adapted browser.

Bison priscus
The only pollen spectrum that could be obtained is completely dominated by pollen of Artemisia, Asteroideae and Poaceae. Tree pollen is absent, pointing to Weichselian subarctic steppe conditions and a grazing diet. This is in accordance with the scarce previous work done on the palaeobiology of steppe bison. Saarinen et al. (2016) found that Bison priscus was commonly associated with rather open vegetation, but also occurred in environments relatively rich in arboreal pollen.

Ovibos moschatus
Aggregates of Betula pollen, present in the only available molar point to browsing of flowering birches, but the landscape was rather open, considering the representation of Asteraceae, Ericaceae and Poaceae. Polygonum viviparum is an arctic-alpine species and its presence fits in the open landscape under subarctic conditions.

Coelodonta antiquitatis
Tree pollen values are low, and the vegetation was dominated by Poaceae, Cyperaceae, Plantago, Galium-type, Gentianella campestris, Scabiosa, Helianthemum, Artemisia and other Asteraceae. Aggregates of pollen of Poaceae in three samples point to consumption of flowering grasses and therefore Poaceae are probably overrepresented in the pollen spectra. The data clearly point to a grazing diet, which matches previous work done on molar wear and stomach contents.
Ca05 shows a spectrum pointing to a moist habitat: the aquatic HdV-128A is very common and also some other algal spores point to wet conditions. Fungal spores HdV-1400 (unequally 2-celled, longitudinally ribbed; Fig. 2F) that were only known from a Weichselian interstadial deposit (van Geel et al. 2010) now appear to be linked to the habitat of C. antiquitatis (cold, dry climatic conditions). The presence of spores of the coprophilous fungi Podospora, Sordaria and Sporormiella may well point to HdV-1400 as another coprophilous fungus.

Stephanorhinus kirchbergensis
The single molar spectrum shows a dominance of tree pollen (mainly Alnus and Corylus, with some Betula, Quercus, Tilia, Ulmus). In combination with Viscum album, Hedera helix, ferns and Kretzschmaria deusta (a fungal parasite on a variety of tree species; van Geel and Aptroot, 2006;van Geel et al., 2013) the sample represents a predominantly forested environment and supposedly it has an Eemian interglacial, or an early Weichselian interstadial age. We can exclude a Holocene age as S. kirchbergensis became extinct shortly after the Eemian interglacial (van der Made, 2010;Mol et al., 2012). Fig. 3 shows the DCA plot of the plant taxa. There are three poorly defined clusters. The cluster on the right contains mostly thermophilous interglacial taxa like Alnus, Betula, Corylus, ferns, Frangula, Hedera, Lonicera, Myrica, Quercus, Rhamnus, Tilia, Typha, Ulmus and Viscum and most of the wetland taxa. Most taxa on the left represent species of tree-less landscapes. Fig. 4 shows the DCA of the zoological taxa, as based on the pollen spectra recorded from the molars. There is clustering of the various species and variation between those species. The samples from C. antiquitatis cluster strongly together (except sample Ca01). This is in sharp contrast to the C. elaphus samples that are spread through most of the plot. This may be explained by the extreme dietary flexibility of C. elaphus compared to the strictly grazing diet of C. antiquitatis. R. tarandus samples cluster close to both M. giganteus and B. priscus and some of the R. tarandus samples are also similar to C. antiquitatis. M. giganteus samples show a rather large variation along Axis 1, but not so much along Axis 2. This suggests a more specialized diet than that of C. elaphus. A. alces and S. kirchbergensis cluster close together; the abundance of arboreal (interglacial) pollen in those samples explains this. The dominance of Betula pollen in the O. moschatus sample explains its clustering close to these temperate taxa, but Betula may be strongly overrepresented (pollen aggregate).

DCA
The three outliers have unique microfossil compositions. Mg03 is the only sample rich in Osmunda regalis. Rt11 has abundant Triglochin and Pedicularis palustris-type, which may be caused by grazing of flowers. Ce05 contains a remarkably high abundance of Helianthemum, including an aggregate.
R. tarandus and C. elaphus are both large mammals that survived into the Holocene. It is worth noting that both species show a larger variation in the DCA plot than the extinct M. giganteus and C. antiquitatis. Species with low diet diversity are most vulnerable for extinction. After abrupt climate change, sudden vegetation changes could cause a population crash and possibly extinction (Augustyniak, 2010;van Geel et al., 2018).

Chronology and phenology
How much time (days, weeks, months, years?) does a pollen sample from an infundibulum represent? In other words: how much time did it take to fill an infundibulum? If different very short periods are represented in the pollen records from infundibula then different parts of the flowering season (and therefore different plant taxa) may be represented in the pollen and spore samples even if the same area was foraged. Infundibula that were filled with compacted plant fragments when less or no flowering took place (late autumn and winter) will show low amounts, or may even be devoid of pollen. Probably the infill of infundibula directly followed the eruption of molars into the oral cavity. This eruption has a set order and a rather narrow timing in the ontogeny of many mammals (Hillson, 2005). However, variation in timing within a species of only a couple of weeks may lead to significantly different parts of the flowering season being represented in the infundibulum. Can we expect further compression and therefore additional, later infill of an infundibulum when decomposition of the plant remains, and related additional compaction in the infundibulum takes place? In other words: may such samples represent large time intervals, or even complete flowering seasons? Our questions remain unanswered by our data and require experimental investigations.

Food choice and age of the molars
What were random effects of individual food choices of herbivores on our pollen records from molars, and can we anyhow expect that we can derive preferences in the food choice based on pollen and spores? The differences between pollen records from the same species do show that there are factors that must cause these differences. The abundance of Poaceae pollen in Coelodonta, a known grazer, is a striking example of a link between our pollen records and the animal's diet. Furthermore, the clustering of our samples in the DCA does point to some structure in the dataset that may well be explained by food preferences and selectivity of the species.
Flowers often are the most sun-exposed (highest) parts of plants and therefore flowers with not yet dispersed pollen may have been consumed. We observed that not yet released, unripe pollen grains appeared as pollen aggregates in the microscope slides. Consumption of flowers could lead to a strong overrepresentation of some plant species, but an advantage of the presence of pollen aggregates is that direct information on consumption of certain plant species becomes available.
Individual herbivore specimens lived during a relatively short period, while climatic conditions were not at all stable over longer time intervals (interglacials, glacial stadials, interstadials and periods of transitional climatic conditions). Migration of plant species will have taken place after climate change and different successive climatic conditions were linked to the existence of different vegetation types. Therefore differences between pollen spectra may occur that are not only influenced by the season and the food choice of the herbivores but also by the age of the molar in relation to changing climatic conditions and related changes in the vegetation composition. We also should be aware that the pollen rain -especially pollen of wind-pollinated taxa -will have landed on plants that were consumed by the herbivores and thus various pollen taxa will be represented, independent from the individual food choice of the animal.
Taking into account the limitations in the interpretations of our pollen and spore record, we nevertheless obtained new information about the food choice of giant deer, red deer, reindeer, moose, steppe bison, muskox, woolly rhino, and forest rhino.

Conclusions
Based on molar-embedded pollen grains and spores we gained insight into the diet of eight herbivore species that lived during the Eemian interglacial, the Weichselian, and the Early to mid-Holocene (until inundation because of sea level rise), respectively. Food plants were represented as relatively high percentages of their pollen or spores, and monospecific pollen aggregates in the microfossil slides gave valuable additional information about food choice. Compared to pollen records from stratigraphic sequences, our pollen spectra were strongly biased. Some of our questions about the pollen sources (pollen rain versus active food selection) remained unanswered by our data and require experimental investigation.
By combining ecological data of the recovered plant taxa with statistical analysis, we showed variations in diet between the various sympatric large herbivores of the mammoth steppe ecosystem. Although resource partitioning through competition over food is still hotly debated in large herbivores (Owen-Smith, 2002), we conclude that the dietary differences we found may have allowed different herbivore species to coexist.

Author contributions
B.v.G. took the botanical samples, analyzed the microfossils, led the discussion and wrote part of the manuscript. B.W.L. and D.M. identified the molars, wrote the general text about the mammal species and contributed to the discussion and the interpretation of the data. B.W.L. executed the DCA of the zoological and botanical data. J.F.N.v.L. and P.W.O.v.d.K. identified problematic pollen taxa, made the microfossil diagram and were involved in discussions and interpretation of both zoological and botanical data.

Declaration of competing interest
None.  Kees van Hooijdonk (Natuurmuseum Brabant, Tilburg) and F. Laarman (Rijksdienst voor het Cultureel Erfgoed) for permission to take botanical samples from molar infundibula. Annemarie Philip kindly prepared the samples and Jan van Arkel took pictures of fungal spores and aggregates of pollen grains. Two anonymous reviewers provided helpful comments on an earlier draft of the manuscript.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.