Hagenia from the early Miocene of Ethiopia: Evidence for possible niche evolution?

Abstract Fossil pollen believed to be related to extant Hagenia abyssinica were discovered in the early Miocene (21.73 Ma) Mush Valley paleoflora, Ethiopia, Africa. Both the fossil and extant pollen grains of H. abyssinica were examined with combined light microscopy, scanning electron microscopy, and transmission electron microscopy to compare the pollen and establish their relationships. Based on this, the fossil pollen grains were attributed to Hagenia. The presence of Hagenia in the fossil assemblage raises the questions if its habitat has changed over time, and if the plants are/were wind pollinated. To shed light on these questions, the morphology of extant anthers was also studied, revealing specialized hairs inside the anthers, believed to aid in insect pollination. Pollen and anther morphology are discussed in relation to the age and origin of the genus within a molecular dated phylogenetic framework, the establishment of complex topography in East Africa, other evidence regarding pollination modes, and the palynological record. The evidence presented herein, and compiled from the literature, suggests that Hagenia was an insect‐pollinated lowland rainforest element during the early Miocene of the Mush Valley. The current Afromontane habitat and ambophilous (insect and wind) pollination must have evolved in post‐mid‐Miocene times.

from the Missouri Botanical Garden (MO) was prepared according to the protocol outlined in Grímsson et al. (2017), Grímsson et al. (2018) and Halbritter et al. (2018, pp. 103-105, Acetolysis the Fast Way). Fifty pollen grains from each sample were measured and studied with LM and SEM, and five pollen grains with TEM.  . The fossil pollen grains were investigated both by LM and SEM using the single grain method as described by Zetter (1989) and Halbritter et al. (2018, pp. 121-123).
Five to ten fossil Hagenia pollen from each sedimentary sample were measured and studied with LM and SEM, and two to four grains with TEM. Additional fossil Sanguisorbeae pollen (see Appendix) originate from the early to middle Miocene of Assoyo de los Mineros, South America, and the middle Miocene of Botn, Iceland (Table 2).
SEM stubs with fossil pollen produced under this study are stored in the collection of the Department of Paleontology, University of Vienna, Austria.
For ultrastructural study of fossil and extant pollen, grains were prepared using the advanced TEM protocol by Ulrich and Grímsson (2020). Subsequent to SEM investigation, fossil pollen grains were transferred directly from SEM stubs, with a micromanipulator, into a final embedding mold filled with a mixture of Agar low-viscosity resin (LV-resin) and acetone for infiltration and final embedding. Pollen of Hagenia abyssinica (from Ethiopia, coll. J.J.F.E. de Wilde,5,836 [MO]; from Burundi, coll. J. Lewalle,5,739 [MO], from Tanzania, coll. Y.S. Abeid,3,376 [MO], and from Malawi,coll. I.F. LaCroix,4,822 [MO]) was first acetolyzed and then transferred directly with an acetone filled pipette into the final embedding mold. Following polymerization, ultrathin sections were made with a diamond knife on a Reichert Ultracut microtome and collected onto formvar film-coated copper grids. For contrast, sections were stained with uranyl acetate (U, for 40 min), followed by lead citrate (Pb, for 3 min) (Hayat, 2000).
Technical note: The heat of the electron beam in TEM sometimes leads to problems when the pollen has been sputter coated and the formvar coated grids are thin. The formvar film adjacent to the gold cover can expand and rupture, resulting in a gap between the pollen wall and the gold cover, which can increase until the film completely ruptures (Ulrich & Grímsson, 2020).

| Anthers
Anthers from extant flowers (Table 1) were prepared for SEM to investigate the pollination mode of H. abyssinica by observing anther dehiscence and the presence/absence of pollen coatings and Ubisch bodies. The characteristic extent of dehiscence is well preserved in dried specimens (Castellanos et al., 2005). Open and closed anthers from different flowers were dissected, mounted on SEM stubs with double-sided adhesive tape, and sputter coated with gold (Halbritter et al., 2018).

| Phylogeny
The concatenated alignment and the consensus tree from Xiang et al. (2017) were downloaded from TreeBase (Piel et al., 2009) (matrix S22054). The single orthologous alignments were extracted using the python script split_concat_nexus.py (from https://gist. github.com/brant fairc loth/2999578), obtaining a total of 172 alignments. Taxa without any data were removed from the alignments using the program pxclsq from the phyx collection (Brown et al., 2017). Maximum Likelihood (ML) trees were generated for each ortholog group using RAxML ver. 8.0.0 (Stamatakis, 2014) run using the script raxml_wapper.pl (https://sco.h-its.org/exelixis/web/ software/raxml/index.html). A GTR model with Gamma-distributed rate variation and a proportion of invariable site model was employed for all tree searches. Five loci were then selected for the clock analysis using the approach of Smith et al. (2018) as implemented in the SortaDate script repository (Smith, 2019). First, the variance in the path between root and tips was calculated and then the proportion of splits present in the consensus tree of Xiang et al. (2017). Then, these metrics were combined and the five best loci were assessed.  (a, b, f, g, k, l, p, q, u, v, z, a′) Polar view, optical cross section. (c, h, i, m, n, r, w, x, b′, c′) Equatorial view, optical cross section. (d, e, j, o, s, t, y, d′) Equatorial view, high focus. Scale bars -10 µm

TA B L E 1 Herbarium material used for this study
The five selected loci (ORTHOMCL46200, ORTHOMCL58800, ORTHOMCL97220, ORTHOMCL116510, ORTHOMCL122180) were then concatenated in a single alignment. The matrix was then trimmed to only include Rosaceae. Thirteen fossil taxa were added as tips, and the ages of these taxa were implemented as uniform distribution, taking into consideration the uncertainty in the dating of the strata where the fossils were recovered (see Tables 2 and 3). The placement of the fossil taxa was constrained using expert-based assignation, while the backbone topology (of the extant taxa) was fixed to the consensus topology from Xiang et al. (2017). A dated analysis using the fossilized birth-death prior was run using MrBayes ver.

TA B L E 2 Information on sample sites
Size of most parameters reached a value higher than 200 or higher than 100. A consensus tree was summarized using the 'allcompat' options of the 'sumt' command.

| Pollen descriptions
The pollen terminology follows Punt et al. (2007;LM) grains are similar to those of H. abyssinica described above. The fossil and extant pollen grains are compared in Table 4 and below.

| Molecular dating analysis
Our molecular dating analysis retrieves an Early Cretaceous age for the crown group Rosaceae (

| Phylogenetic relevance
The

| Ecological preferences of fossil and extant Hagenia
The discovery of Hagenia pollen at Mush tempts one to use its current restricted distribution to infer the ecological setting at Mush, but begs the question, is it appropriate to use the ecological envelope of extant Hagenia as a key to the past. Today, Hagenia occurs predominantly at elevations from 2,000-3,000 m, within the F I G U R E 7 Light microscopy (a, f, i) and scanning electron microscopy (b-e, g, h, j, k)  Paleoclimate at the Mush locality has so far been documented by calculation of mean annual precipitation from leaf morphology at six stratigraphic horizons, including the specific location of deposits also preserving Hagenia pollen (Bush et al., 2017). Precipitation estimates have overlapping margins of error and range from about 1,500-1,600 mm/yr. This is wetter than mean annual precipitation at the site today (~1,100 mm/yr). A mean annual temperature has not yet been reported for the fossil site, but global paleotemperature curves based upon the marine oxygen isotopic record indicate the global mean was higher than modern (Zachos et al., 2008).
Taxonomic and morphologic assessments of leaves, phytoliths, and pollen are consistent with a closed-canopy, mixed-moist semievergreen forest with limited palm diversity, and showing botanical affinities with species found today in West, Central, and East Africa (e.g., Bush et al., 2017;Currano et al., 2020;Danehy, 2010;Grímsson et al., 2019;Pan et al., 2012Pan et al., , 2014. Both Newtonia and Tacca (Pan et al., 2012), found today in moist tropical or sec-  , that today grows in tropical lowland rainforests (swamps), were identified at Mush. Moreover, fossil leaf stable carbon isotope (δ 13 C) values match the average for modern African tropical rain forests (Bush et al., 2017), and the 1523-1647 mm/yr reconstructed mean annual precipitation at early Miocene Mush (Bush et al., 2017) is comparable to that of modern mixed-moist semi-evergreen rainforests in the Guineo-Congolian regional center of endemism, one of Africa's eighteen major phytochoria (White, 1979(White, , 1983(White, , 1993. While the Mush paleoflora is consistent with White's (1983) closed-canopy, mixed-moist semievergreen forest described from but Oleaceae pollen grains are not easily assigned to particular genera without SEM studies (e.g., Punt et al., 1991) and can also be confused with pollen from, for example, Euphorbiaceae. Also, Podocarpaceae (Podocarpus) and Cupressaceae (Juniperus) pollen that are certainly long-distance wind-dispersed taxa and typical of modern Afromontane forests have not been observed in any sample from the Mush Valley site (see also Danehy, 2010). It seems to be the general consensus that Hagenia abyssinica is a strictly windpollinated species, with the possibility of long-distance pollen dispersal (e.g., Ayele et al., 2009Ayele et al., , 2011Gichira et al., 2017;Negash, 2010).
The mode of anther dehiscence also affects the amount of pollen released. Pollen of zoophilous plants usually remains in the open anther after dehiscence and is gradually exposed over time, but, remaining pollen of mainly insect pollinated flowers may also be transported passively by wind when flowers wither and pollen dries out (Willmer, 2011). Usually, anthers open synchronously, releasing all pollen at once (Willmer, 2011). In some of the SEMinvestigated anthers (see Results), the slit seemed to split progressively, leading to gradual release of pollen, but without proper field observation, it remains unclear how the anther splits (synchronously or progressively) and also if pollen is deposited onto the hairy calyx lobes (accidental dislocation or secondary presentation). In case of secondary pollen presentation, pollen must be sticky to adhere to other floral organs and subsequent pollinators, but the Hagenia pollen is without pollenkitt. Moreover, secondary pollen presentation is rare in angiosperms, confined to 25 families (Howell et al., 1993;Yeo, 1993), and not reported for Rosaceae. It is more likely that H. abyssinica pollen is dislocated by wind, visitors, or at full anther dehiscence and is retained among the hairs of the calyx lobes where it is accessible to bees. A changing environment could cause transition from entomophily to anemophily, or even ambophily for reproductive assurance. The colored male and female H. abyssinica flowers, the occurrence of anther hairs, the restricted distribution of dispersed pollen, the occurrence of pollen in honey, and the documented visits of bees to both male and female flowers indicate that H. abyssinica is not strictly wind-pollinated.
We suggest that Hagenia was primarily insect (?bee) pollinated, as indicated by the anther hairs, and over time, in response to environmental pressure gradually changed into ambophily.
The data produced by our study of early Valley, and climate. It seems that when rift-related mountain chains in East Africa formed, and the climate in lowland East Africa became drier, Hagenia was pushed, giving way for competition, to higher elevations. There, on the island mountains, it survives in comparatively "cold" but mostly humid environments and is an example of niche evolution in response to a complex history of changes in topography related to regional tectonics, and climate.

CO N FLI C T O F I NTE R E S T
All authors declare no competing interests. Writing-original draft (supporting).