# Source paper archive

- Source title: Pollen Ornamentation Morphospace: Interpretable Quantification by Persistent Homology
- Author(s): Deng, Q., Ishihara, K.
- Source landing URL: https://www.biorxiv.org/content/10.64898/2026.11.18.767721v1
- PDF URL: https://www.biorxiv.org/content/10.64898/2026.11.18.767721v1.full.pdf
- PDF SHA-256: ccb579cf396f28a510b12e5217ebd49c2e904837fc02529d3bce0b81b4154389
- Converter: pymupdf4llm 0.0.27
- Limits: Automated PDF text extraction; reading order, equations, tables and formatting may be incomplete or incorrect. Images are omitted, and no OCR was performed. Markup and links are neutralized for safe downloading. Consult the original PDF for the authoritative version.

---

&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
&#91;perpetuity. It is made available under a CC-BY-NC 4.0 International license.&#93;(http://creativecommons.org/licenses/by-nc/4.0/)

# Pollen Ornamentation Morphospace: Interpretable Quantification by Persistent Homology


A Preprint



**Qianzhi Deng (** `邓骞之` **)** &#91;∗&#93;

Department of Computational and Systems Biology

School of Medicine, University of Pittsburgh

Pittsburgh, PA, USA
```
       qid49@pitt.edu

```


**Keisuke Ishihara** &#91;∗&#93;


Department of Computational and Systems Biology

School of Medicine, University of Pittsburgh

Pittsburgh, PA, USA
```
      ishihara@pitt.edu

```


November 18, 2026


**Abstract**


Ornamentation, the sculpture of the outer pollen wall, is one of the central traits of pollen. However, it is mostly
described with qualitative categories that rely on subjective judgment, and continuous, interpretable descriptors
are still lacking. Here we used persistent homology, a fundamental method in topological data analysis, to
define three interpretable descriptors (intensity, density and polarity) for three-dimensional pollen data. These
descriptors form the three axes of a quantitative morphospace of pollen ornamentation, in which real, theoretical
and ancestral pollen can be placed. The three axes were largely independent and separated common ornamentation
types. In tests against known values, the morphospace recovered the number of ornamental elements in both
theoretical pollen and real echinate pollen. Applied to pollination, it showed that, after phylogenetic correction,
insect-pollinated pollen differed significantly from wind-pollinated pollen only in having more pronounced relief.
In an evolutionary analysis within Asteraceae, it traced the three subfamilies diverging in different directions from
their estimated common ancestor. By replacing qualitative categories with continuous, interpretable descriptors,
this morphospace provides a common quantitative framework for pollen ornamentation. As three-dimensional
pollen data accumulate, new grains can be placed in this morphospace without redefining the axes.


_**Keywords**_ morphospace · persistent homology · pollen ornamentation · pollination syndrome · topological data analysis



**1** **Introduction**


Pollen is central to sexual reproduction in seed plants
and is relevant to studies of pollinators, food security,
human health and past climates (Hornick _et_ _al._, 2022).
Angiosperm pollen is highly diverse in morphology, varying markedly in size, shape, apertures and ornamentation
(Halbritter _et_ _al._, 2018). Size can be measured at high
throughput by imaging flow cytometry (Hornick _et_ _al._,
2025). Overall shape is conventionally summarized by
the ratio of the polar axis to the equatorial diameter (P/E
ratio; Punt _et_ _al._, 2007) or quantified by elliptic Fourier
descriptors of the outline (Bonhomme _et_ _al._, 2013; Kriebel
_et_ _al._, 2017). Apertures are counted directly and represent
one of the three characters of the NPC aperture classification system (Erdtman &amp; Straka, 1961). In contrast, the
characterization of ornamentation, the sculpture of the
outer pollen wall, remains largely qualitative. Categories
such as echinate (spiny), lophate (ridged), psilate (smooth)
and reticulate (net-like) impose discrete boundaries on
continuous variation and do not capture differences within
each category. Assigning a grain to a category depends on
subjective judgment, and the same grain may be classified


∗ Co-corresponding authors.



differently (Halbritter _et_ _al._, 2018). As a result, despite
being one of the core traits of pollen, ornamentation still
lacks a set of continuous, interpretable descriptors that
would allow objective and reproducible comparisons.

Pollen ornamentation is difficult to quantify because it
consists of many local elements, such as spines and ridges,
whose form and number vary among grains. Existing
morphometric methods are poorly suited to such structure.
Fourier-based methods, whether elliptic Fourier analysis
in two dimensions or spherical harmonic analysis in three
dimensions, require the outline or surface to be closed
(Kuhl &amp; Giardina, 1982; Brechbühler _et_ _al._, 1995), but the
exine (pollen outer wall) can be interrupted by openings
such as apertures. Even when the outline or surface is
closed, a truncated Fourier or spherical harmonic series
can produce spurious oscillations where the ornamentation
changes abruptly (the Gibbs phenomenon; Gelb, 1997).
Moreover, because each harmonic extends over the whole
grain, the ornamentation is spread across many coefficients
that are difficult to interpret biologically (Bookstein _et_ _al._,
1982). Geometric morphometrics relies on landmarks that
are homologous across specimens (Mitteroecker &amp; Gunz,
2009), which are not available for pollen ornamentation.


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint



Image-based methods are widely used to identify and
compare pollen, based on features that are either handcrafted (Mander _et_ _al._, 2013; Wei _et_ _al._, 2025) or learned
by machine learning (Romero _et_ _al._, 2020). These features
are often high-dimensional, and those learned by machine
learning are usually hard to interpret (Barnes _et_ _al._, 2023).

Persistent homology, a fundamental method in topological data analysis, offers an alternative. Given a scalar field
on a surface (e.g. the height of each point), persistent homology sweeps a threshold across the range of the field and
records the thresholds at which each topological feature,
such as a connected component, appears (its birth) and
disappears (its death), with its lifetime known as its persistence (Edelsbrunner _et_ _al._, 2002). LO-analysis, a classical
light-microscopy method, examines ornamentation in a
similar way by continuously varying the focus through the
grain and recording the light patterns (Erdtman, 1956).
Persistent homology requires no homologous landmarks,
does not need the scalar field to be defined everywhere,
and the persistence it records is stable under small perturbations of the field (Cohen-Steiner _et_ _al._, 2007). While
persistent homology has been proposed for quantifying
pollen ornamentation (Li _et_ _al._, 2018; Radja, 2021), it has
yet to be applied to three-dimensional pollen data. Further,
how to summarize the output of persistent homology as
a result that is easily interpretable by a pollen biologist
remains an open question.

Here we apply persistent homology to three-dimensional
pollen data and introduce three interpretable descriptors:
intensity _𝐼_, density _𝐷_ and polarity _𝑃_, which measure how
pronounced the ornamentation is, how finely it is patterned
and whether it is dominated by protrusions or by depressions, respectively. These three descriptors define a fixed
morphospace that facilitates the comparison of theoretical,
real and ancestral pollen without any free parameters. We
first test the descriptors against known values: for theoretical pollen generated by a model of pattern formation
(Radja _et_ _al._, 2019), they recover the prescribed number of
ornamental elements, and for real echinate pollen, they predict the number of spines. We then apply the morphospace
to pollination, where, after phylogenetic correction, insectpollinated pollen differs from wind-pollinated pollen only
in having more pronounced relief, and to evolution within
Asteraceae, where the three subfamilies diverge in different directions from their estimated common ancestor.

Together, we show that the three descriptors provide an
interpretable quantification of pollen ornamentation, and
that the resulting morphospace allows the study of pollen
formation, function and evolution in a common quantitative
framework.


**2** **Materials** **and** **Methods**


**2.1** **Mapping** **pollen** **ornamentation** **to** **morphospace**
**coordinates** **by** **persistent** **homology**


Our pipeline maps each grain to its morphospace coordinates. Its input is a single-valued radial function _𝜌_ on
the sphere, the distance from the center of the grain to its
surface in each direction. In practice, _𝜌_ is sampled at the
vertices of a grid formed by subdividing an icosahedron
six times and is normalized by its maximum. Where



_𝜌_ is undefined, the corresponding vertices are excluded
together with their edges.

Sublevel-set filtrations of _𝜌_ and − _𝜌_ over this icosahedral grid give the depression and protrusion channels,
respectively. We computed the zero-dimensional persistent homology of each channel (Edelsbrunner _et_ _al._,
2002), discarded infinite or zero persistence, and pooled
the remaining persistence of both channels into a set of
signed persistence &#123;( _𝑝_ _𝑖_ _, 𝜎_ _𝑖_ )&#125; _𝑖_ _&#91;𝑁&#93;_ =1 &#91;,&#93; &#91;where&#93; _&#91;𝑁&#93;_ &#91;is&#93; &#91;the&#93; &#91;number&#93;

of features, _𝑝_ _𝑖_ is the persistence of the _𝑖_ th feature, and _𝜎_ _𝑖_
is its channel, +1 for protrusion and −1 for depression.
Persistence was computed with GUDHI 3.13.0 (Maria
_et_ _al._, 2014).

We define three scalars, intensity _𝐼_, density _𝐷_ and
polarity _𝑃_, from the persistence fraction of each feature,
_𝜋_ _𝑖_ = _𝑝_ _𝑖_ / &#91;�&#93; _𝑗_ _&#91;𝑝&#93;_ _𝑗_ &#91;,&#93; &#91;as&#93;



These three scalars form the axes of the morphospace,
and their values for each grain are its morphospace coordi
nates.


**2.2** **Processing** **real** **pollen** **data** **from** **the** **3D** **Pollen**
**Library**


We used three-dimensional meshes of real pollen from
the 3D Pollen Library (Perry _et_ _al._, 2017; Hayes, 2024),
which were reconstructed from confocal images of exine
autofluorescence and are openly available through NIH 3D.
Ornamentation types of the specimens were taken from
PalDat (Weber &amp; Ulrich, 2017).

We restricted the analysis to near-spherical grains.
Sphericity was defined as _𝑠_ = ~~√~~ _𝜇_ 3 / _𝜇_ 1, where _𝜇_ 1 ≥ _𝜇_ 2 ≥

_𝜇_ 3 are the eigenvalues of the covariance matrix of the
mesh vertices. Of the 286 angiosperm pollen meshes, we
retained the 162 whose sphericity exceeded 0.7. For each
retained mesh, rays were cast from its center of mass along
the directions given by the vertices of the icosahedral grid.
The radial function _𝜌_ was taken as the distance to the
farthest intersection, and directions without an intersection were omitted. The resulting _𝜌_ was then passed to
the pipeline. Meshes were processed with trimesh 5.0.0
(Dawson-Haggerty et al., 2019) and rendered with PyVista
0.48.4 (Sullivan &amp; Kaszynski, 2019).
Of the 162 grains retained for analysis, 20 were echinate
Asteraceae. We excluded _Artemisia_ _absinthium_, whose
microechinate surface is dominated by three deep colpi
rather than by spines, and manually counted the spines on
the three-dimensional meshes of the remaining 19 grains
to test the predicted number of spines.


**2.3** **Generating** **theoretical** **pollen** **from** **a** **model** **of**
**pattern** **formation**


Theoretical pollen grains are equilibrium states of the
model of Radja _et_ _al._ (2019), computed with the program
released by the authors. In dimensionless form, the model
is controlled by the wavenumber _ℓ_ 0, the cubic coefficient



_𝑝_ _𝑖_

_𝑁_


_𝜋_ _𝑖_ ln _𝜋_ _𝑖_



_𝐼_ = ln


_𝐷_ = −



∑︁

_𝑖_

∑︁

_𝑖_



_𝑃_ = tanh &#91;−&#93;&#91;1&#93; &#91; ∑︁&#93;


_𝑖_



_𝜎_ _𝑖_ _𝜋_ _𝑖_



2


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint



_𝜆_ 3 and the reduced temperature _𝜏_ . We varied _ℓ_ 0 from 12.0
to 20.0 in steps of 0.5, set _𝜆_ 3 to −0 _._ 35, −0 _._ 25 and −0 _._ 15,
and fixed _𝜏_ at −0 _._ 25, giving 51 parameter sets. For each
set, the program minimizes the free energy of the model
and returns the equilibrium pattern as spherical harmonic
coefficients. We reconstructed this pattern at the vertices
of the icosahedral grid and passed the resulting values to
the pipeline as _𝜌_ .


**2.4** **Testing** **the** **pollination** **syndrome** **and** **estimating**
**ancestral** **ornamentation**


Species names of the specimens were standardized to
accepted names in WCVP (Govaerts _et_ _al._, 2021) with
rWCVP 1.3.0 (Brown _et_ _al._, 2023). A dated phylogeny
was built on the backbone of the GBOTB.extended.TPL

megatree (Smith &amp; Brown, 2018) with V.PhyloMaker2
0.1.0 (Jin &amp; Qian, 2022). Species absent from the backbone
were grafted onto their genus, and those whose genus was
also absent were excluded. The morphospace coordinates
of each species were averaged over its pollen grains, and all
subsequent analyses were performed at the species level.

Pollination modes were taken from the pollen vector
records of the Ecological Flora of the British Isles (Fitter
&amp; Peat, 1994), in which a species can have more than one
vector. Species names in this database were standardized
as above and matched to the specimens by accepted name.
Species recorded as insect but not wind were classed as
insect-pollinated, and those recorded as wind but not insect
as wind-pollinated; all other species were excluded. This
left 116 species from 41 families, 95 insect-pollinated and
21 wind-pollinated.
Analyses of pollination mode used the dated phylogeny
pruned to these 116 species. On each axis, the difference
between insect- and wind-pollinated species was calculated as insect-pollinated minus wind-pollinated values.
Without phylogenetic correction, the difference was expressed as Cohen’s _𝑑_ . With phylogenetic correction, it
was estimated by phylogenetic generalized least squares
(PGLS; Grafen, 1989), with Pagel’s _𝜆_ (Pagel, 1999) fitted
by maximum likelihood. Each PGLS difference was reported with its 95% confidence interval and tested with a
_𝑡_ -test, and the _𝑝_ -values of the three axes were adjusted for
multiple testing with the Benjamini–Hochberg procedure
(Benjamini &amp; Hochberg, 1995).

Analyses within Asteraceae used the dated phylogeny
pruned to the 28 Asteraceae species, regardless of pollination records. On each axis, phylogenetic signal was
measured by Pagel’s _𝜆_, estimated by maximum likelihood
with a 95% confidence interval from its likelihood profile.
Each _𝜆_ was tested against _𝜆_ = 0 with a likelihood-ratio
test, and the _𝑝_ -values of the three axes were adjusted in
the same way as above. Ancestral values on each axis
were estimated by maximum likelihood under a Brownian
motion model (Schluter _et_ _al._, 1997).

All phylogenetic analyses were run in R 4.4.3 (R Core
Team, 2025). Trees were handled with ape 5.8.1 (Paradis &amp;
Schliep, 2019). PGLS of pollination mode was fitted with
the gls function of nlme 3.1-170 (Pinheiro _et_ _al._, 2026).
Within Asteraceae, Pagel’s _𝜆_ was fitted with caper 1.0.4
(Orme _et_ _al._, 2025), and ancestral states were estimated
with the fastAnc function of phytools 2.5.2 (Revell, 2024).



**3** **Results**


**3.1** **Persistent** **homology** **decomposes** **pollen**
**ornamentation** **into** **three** **interpretable** **axes**



We illustrate how persistent homology captures ornamentation using a schematic grain surface _𝜌_ bearing three
protrusions and three depressions (Fig. 1a). Consider a
spherical water surface of radius _𝑟_ centered on the grain’s
center of mass, and let _𝑟_ sweep across the range of _𝜌_ . The
sweep defines a filtration, with _𝑟_ as the filtration value.
The part of the pollen surface under water, &#123; _𝜌_ ≤ _𝑟_ &#125;, is the
sublevel set of _𝜌_ at _𝑟_, and forms the depression channel;
the part above water, &#123; _𝜌_ ≥ _𝑟_ &#125;, is equivalently the sublevel
set of − _𝜌_ at − _𝑟_, and forms the protrusion channel. In the
depression channel, rising water forms a pool in each depression; in the protrusion channel, falling water exposes
an island at each protrusion. Each pool or island disappears
when it merges with an older one at a saddle. At each
filtration value, the reduced zero-dimensional homology
counts the separate pools ( _𝐻_ &#91;�&#93; 0 &#91;−&#93; &#91;)&#93; &#91;or&#93; &#91;islands&#93; &#91;(&#93; &#91; �&#93; _𝐻_ 0 &#91;+&#93; &#91;)&#93; &#91;except&#93;

the oldest pool or island, which never disappears. For
example, three separate pools at _𝑟_ = 0 _._ 5 give _𝐻_ &#91;�&#93; 0 &#91;−&#93; &#91;=&#93; &#91;2&#93; &#91;,&#93; &#91;and&#93;

three separate islands at _𝑟_ = 0 _._ 75 give _𝐻_ &#91;�&#93; 0 &#91;+&#93; &#91;=&#93; &#91;2&#93; &#91;.&#93; &#91;Persistent&#93;

homology then follows each finite component from its
birth to death, giving a persistence bar with persistence _𝑝_ _𝑖_
(the depth of a depression or the height of a protrusion)
and channel _𝜎_ _𝑖_ ( +1 for protrusion and −1 for depression).
Together the bars form a barcode of signed persistence,
and the number of finite bars crossing any filtration value
equals _𝐻_ &#91;�&#93; 0 &#91;±&#93; &#91;at&#93; &#91;that&#93; &#91;value&#93; &#91;(Fig.&#93; &#91;1b,&#93; &#91;top).&#93; &#91;Plotting&#93; &#91;the&#93; &#91;death&#93;

of each component against its birth gives the persistence
diagram (Fig. 1b, bottom).

The three descriptors defined from signed persistence
(see Materials and Methods) can be interpreted both in the
persistence diagram and on the pollen surface (Fig. 1c). Intensity _𝐼_, the logarithm of the mean persistence, measures
how far the points lie from the diagonal on average and
hence the mean amplitude of the ornamental relief. Among
the four grains, _𝐼_ is highest in the echinate pollen of _Bellis_
_perennis_ ( _𝐼_ = −2 _._ 78 ), whose tall spines give high persistence, slightly lower in the lophate pollen of _Hypochaeris_
_radicata_ ( _𝐼_ = −3 _._ 18 ), whose spines are shorter than those
of _B._ _perennis_, and lowest in the psilate pollen of _Trisetum_
_flavescens_ ( _𝐼_ = −5 _._ 71 ), whose shallow relief gives low
persistence. Density _𝐷_, the Shannon entropy of the persistence fractions, reflects how evenly the persistence is
spread, that is, how finely the ornamentation is patterned.
_𝐷_ is highest in the reticulate pollen of _Cardamine_ _praten-_
_sis_ ( _𝐷_ = 6 _._ 50 ), whose numerous lumina form a large
cloud of points in the depression channel, and lowest in
_T._ _flavescens_ ( _𝐷_ = 3 _._ 71 ), whose smooth surface is barely

patterned. Polarity _𝑃_ compares the fractions of persistence
carried by the two channels and thereby indicates whether
the ornamentation is dominated by protrusions or by depressions. _𝑃_ is high in _B._ _perennis_ ( _𝑃_ = 1 _._ 05 ), whose
spines carry most of the persistence, lower in _H._ _radicata_
( _𝑃_ = 0 _._ 32 ), where the lacunae between ridges add persistence to the depression channel, and strongly negative in
_C._ _pratensis_ ( _𝑃_ = −1 _._ 62 ), whose lumina carry most of
the persistence in the depression channel. Each descriptor



3


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint



therefore corresponds to a feature of ornamentation that
can be recognized by eye.

We placed all 162 near-spherical angiosperm grains in
the morphospace (Fig. 2). They spread across all three axes,
with intensity _𝐼_ ranging from −7 _._ 03 to −2 _._ 68, density _𝐷_
from 1.11 to 7.46 and polarity _𝑃_ from −1 _._ 74 to 1.41. The
axes were only weakly correlated, with Pearson correlation
coefficients of −0 _._ 15 between _𝐼_ and _𝐷_, 0.18 between _𝐼_
and _𝑃_ and 0.09 between _𝐷_ and _𝑃_ . Five example grains of
each common ornamentation type illustrate where these
types lie in the morphospace: the echinate grains at high
_𝐼_ and high _𝑃_, the lophate grains at high _𝐼_ and lower _𝑃_,
the psilate grains at low _𝐼_ and low _𝐷_ and the reticulate
grains at negative _𝑃_ and high _𝐷_ . Overall, persistent
homology decomposes pollen ornamentation into three
largely independent, interpretable axes, and the common
ornamentation types fall in different regions of the space
they span.


**3.2** **Morphospace** **coordinates** **recover** **the** **number** **of**
**ornamental** **elements**


To quantitatively assess the validity of the morphospace,
we analyzed pollen whose number of ornamental elements is known. Radja _et_ _al._ (2019) developed a physical
model where pollen ornamentation forms as the primexine
(a polysaccharide layer that templates the exine) phaseseparates into regularly spaced elements. In this model,
the wavenumber _ℓ_ 0 specifies how many times the elements repeat around the grain and thus sets the number
of ornamental elements. Using this model, we generated
theoretical pollen of varying _ℓ_ 0 and projected the resulting patterns onto the morphospace (Fig. 3a). We asked
whether _ℓ_ 0 can be recovered from _𝐷_ and _𝑃_ .


We derived that _𝐷_ and _𝑃_ are related to the wavenumber

_ℓ_ 0 as



_ℓ_ 0 =



√︁



_𝜋𝑒_ _&#91;𝐷&#93;_ &#91;−&#93; _&#91;ℎ&#93;_ &#91;(&#93; _&#91;𝑃&#93;_ &#91;)&#93;



where _ℎ_ ( _𝑃_ ) is the binary entropy of the persistence fractions of the two channels (Supporting Information A).
Since the elements tile the whole surface, a pattern with _ℓ_ 0
elements around the grain has about _ℓ_ 0 &#91;2&#93; &#91;/&#93; _&#91;𝜋&#93;_ &#91;elements&#93; &#91;in&#93; &#91;each&#93;

channel, so the same relation also predicts the number of
spines _𝑛_ on real echinate pollen as


_𝑛_ = _𝑒_ _&#91;𝐷&#93;_ &#91;−&#93; _&#91;ℎ&#93;_ &#91;(&#93; _&#91;𝑃&#93;_ &#91;)&#93;


For theoretical pollen, the predicted _ℓ_ 0 differed from
the true value by 6.1% on average (Fig. 3b). We further
tested our prediction for real pollen, where we manually
counted the spines on the three-dimensional meshes of 19
echinate Asteraceae grains (see Materials and Methods).
These grains bore 50 to 203 spines each, and the predicted
number differed from the count by 11.9% on average
(Fig. 3c; Supporting Information B). _𝐷_ and _𝑃_ thus recover
the number of ornamental elements in both theoretical and

real pollen.


**3.3** **The** **intensity** **axis** **reflects** **the** **classical** **pollination**
**syndrome**


A quantitative morphospace makes it possible to test qualitative theories in biology (Chartier _et_ _al._, 2014). We test
the pollination syndrome, which posits that plants pollinated by the same vector, such as wind or insects, converge



on similar floral traits. Flowers of wind-pollinated plants,
for example, tend to be small and inconspicuous and lack
nectaries (Friedman &amp; Barrett, 2009). For pollen morphology, the syndrome predicts that wind-pollinated pollen
is smoother, which lowers its settling velocity and lets it
travel farther on the wind, whereas insect-pollinated pollen
bears more conspicuous ornamentation, which helps it
attach to pollinators (Rosbakh _et_ _al._, 2026). Testing these
predictions across the Central European flora, the same
study scored ornamentation types as present or absent and
found that only reticulate ornamentation differed between
wind- and insect-pollinated pollen. Using our quantitative morphospace, we instead measure how much windand insect-pollinated pollen differ along the three axes of
intensity, density and polarity.

We compared 95 insect-pollinated and 21 windpollinated species on their dated phylogeny (Fig. 4a).
Without phylogenetic correction, insect-pollinated pollen
had higher values than wind-pollinated pollen on all three
axes (Fig. 4b), with Cohen’s _𝑑_ of 1.10 for _𝐼_, 0.72 for _𝐷_
and 0.42 for _𝑃_ . After phylogenetic correction by PGLS
(Fig. 4c), the difference remained significant only for _𝐼_
(0.87, 95% CI 0.44 to 1.30, _𝑝_ FDR _&lt;_ 0 _._ 001 ), not for _𝐷_
(0.49, 95% CI −0 _._ 15 to 1.13, _𝑝_ FDR = 0 _._ 129 ) or _𝑃_ (0.23,
95% CI −0 _._ 04 to 0.51, _𝑝_ FDR = 0 _._ 129 ). Wind-pollinated
pollen therefore has lower relief than insect-pollinated
pollen, as the pollination syndrome predicts, but does not
differ significantly in the fineness of the ornamentation or
in whether it is dominated by protrusions or by depressions.


**3.4** **Asteraceae** **subfamilies** **show** **evolutionary**
**divergence** **in** **the** **morphospace**


The evolution of pollen morphology is increasingly studied
in phylomorphospaces, in which species and their ancestors
are placed together in a morphospace (Jardine _et_ _al._, 2022;
Luo _et_ _al._, 2026). In these studies, ornamentation is scored
as discrete characters and the morphospace is obtained by
ordination of these characters. A change in ornamentation
therefore appears as a switch between character states. Our
morphospace measures ornamentation along continuous
axes, so it can also show in which directions, and how
far, ornamentation has evolved. As an example, we traced
these changes in Asteraceae, the family best represented
in our dataset.


We placed the 28 Asteraceae species in the morphospace
together with their ancestral states, estimated on each axis
under Brownian motion (Fig. 5a,b). From the estimated
root, the three subfamilies shifted in different directions.
Cichorioideae shifted toward higher density and lower polarity, that is, toward finer and more porous ornamentation,
as expected from their lophate pollen (Jardine _et_ _al._, 2022).
Asteroideae shifted the opposite way, toward coarser ornamentation dominated by protrusions, and Carduoideae
shifted toward lower intensity, that is, toward lower relief.
_Artemisia_ _absinthium_, the only wind-pollinated species

among the 23 with pollination records, lay apart from the
rest at the low end of all three axes. This is consistent

with the smaller spines of _Artemisia_ pollen (Jardine _et_ _al._,
2022) and with the characters of wind-pollinated pollen
in the previous section. In this way, the morphospace



4


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint



Figure 1: Signed persistence of pollen ornamentation. (a) A schematic grain _𝜌_ with three protrusions and three depressions at filtration
values _𝑟_ = 0, 0.5, 0.75 and 1, shown in three dimensions (top) and as an unrolled equatorial band (bottom), with _𝐻_ &#91;�&#93; 0 &#91;+&#93; &#91;and&#93; _𝐻_ &#91;�&#93; 0 &#91;−&#93; &#91;at&#93; &#91;each&#93;

value. Teal, water surface; orange, shoreline ( _𝜌_ = _𝑟_ ). (b) Persistence barcode (top) and persistence diagrams (bottom) of the grain in (a).
Red, protrusion channel; blue, depression channel; dashed bars and open circles, essential components excluded from the analysis. (c)
Four real grains, one per ornamentation type: _Bellis_ _perennis_ &#91;(3DPX-017818,&#93;(https://3d.nih.gov/entries/3DPX-017818) echinate), _Hypochaeris_ _radicata_ &#91;(3DPX-021466,&#93;(https://3d.nih.gov/entries/3DPX-021466) lophate),
_Trisetum_ _flavescens_ &#91;(3DPX-020938,&#93;(https://3d.nih.gov/entries/3DPX-020938) psilate) and _Cardamine_ _pratensis_ &#91;(3DPX-020611,&#93;(https://3d.nih.gov/entries/3DPX-020611) reticulate). Left, the exine and water surface at the

filtration values in (a), with the intensity _𝐼_, density _𝐷_ and polarity _𝑃_ of each grain; right, persistence diagrams. Colors and symbols as in
(a) and (b).


5


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint


Figure 2: Real pollen in the morphospace. (a) The 162 near-spherical angiosperm grains in the morphospace of intensity _𝐼_, density _𝐷_ and
polarity _𝑃_ . Gray, all grains; colored and numbered, the representative grains in (c). (b) The same grains on the _𝐷_ – _𝐼_, _𝐷_ – _𝑃_ and _𝑃_ – _𝐼_ faces.
Colors as in (a). (c) Twenty representative grains, five per common ornamentation type. Echinate: 1, _Tripleurospermum_ _maritimum_
&#91;(3DPX-021473);&#93;(https://3d.nih.gov/entries/3DPX-021473) 2, _Leucanthemum_ _vulgare_ &#91;(3DPX-017803);&#93;(https://3d.nih.gov/entries/3DPX-017803) 3, _Solidago_ _virgaurea_ &#91;(3DPX-021264);&#93;(https://3d.nih.gov/entries/3DPX-021264) 4, _Bellis_ _perennis_ &#91;(3DPX-017818);&#93;(https://3d.nih.gov/entries/3DPX-017818)
5, _Senecio_ _vulgaris_ &#91;(3DPX-020862).&#93;(https://3d.nih.gov/entries/3DPX-020862) Lophate: 6, _Lapsana_ _communis_ &#91;(3DPX-022934);&#93;(https://3d.nih.gov/entries/3DPX-022934) 7, _Crepis_ _vesicaria_ subsp. _taraxacifolia_ &#91;(3DPX-&#93;(https://3d.nih.gov/entries/3DPX-023360)
&#91;023360);&#93;(https://3d.nih.gov/entries/3DPX-023360) 8, _Hypochaeris_ _radicata_ &#91;(3DPX-021466);&#93;(https://3d.nih.gov/entries/3DPX-021466) 9, _Taraxacum_ _laevigatum_ &#91;(3DPX-021472);&#93;(https://3d.nih.gov/entries/3DPX-021472) 10, _Picris_ _hieracioides_ &#91;(3DPX-021105).&#93;(https://3d.nih.gov/entries/3DPX-021105)
Psilate: 11, _Phalaris_ _arundinacea_ &#91;(3DPX-021112);&#93;(https://3d.nih.gov/entries/3DPX-021112) 12, _Dactylis_ _glomerata_ &#91;(3DPX-022933);&#93;(https://3d.nih.gov/entries/3DPX-022933) 13, _Parietaria_ _judaica_ &#91;(3DPX-021111);&#93;(https://3d.nih.gov/entries/3DPX-021111)
14, _Trisetum_ _flavescens_ &#91;(3DPX-020938);&#93;(https://3d.nih.gov/entries/3DPX-020938) 15, _Poa_ _trivialis_ &#91;(3DPX-022883).&#93;(https://3d.nih.gov/entries/3DPX-022883) Reticulate: 16, _Nasturtium_ _microphyllum_ &#91;(3DPX-021959);&#93;(https://3d.nih.gov/entries/3DPX-021959)
17, _Cardamine_ _pratensis_ &#91;(3DPX-020611);&#93;(https://3d.nih.gov/entries/3DPX-020611) 18, _Brassica_ _napus_ &#91;(3DPX-017819);&#93;(https://3d.nih.gov/entries/3DPX-017819) 19, _Raphanus_ _raphanistrum_ &#91;(3DPX-023493);&#93;(https://3d.nih.gov/entries/3DPX-023493) 20,
_Potamogeton_ _berchtoldii_ &#91;(3DPX-022903).&#93;(https://3d.nih.gov/entries/3DPX-022903)


6


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint


Figure 3: Recovery of the number of ornamental elements. (a) Generation of theoretical pollen with the model of Radja _et_ _al._ (2019):
parameter sweep over _ℓ_ 0 and _𝜆_ 3 (left), energy minimization to the equilibrium _𝜌_ eq (center) and projection onto the morphospace _𝐷_ – _𝑃_ face
(right). Red, one parameter set followed through the three steps; gray lines, iso- _ℓ_ 0 lines; the energy surface is schematic. (b) Theoretical
pollen on the _𝐷_ – _𝑃_ face (left; coincident grains drawn once) and _ℓ_ 0 predicted from _𝐷_ and _𝑃_ against the true value (right). The relief of
theoretical pollen is schematic, as the model does not specify the height of ornamentation. Gray lines as in (a). (c) The 19 echinate
Asteraceae pollen grains in Supporting Information B on the _𝐷_ – _𝑃_ face (left) and spine number _𝑛_ predicted from _𝐷_ and _𝑃_ against the
counted number (right; logarithmic axes). Lines as in (b).


7


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint


Figure 4: Insect- and wind-pollinated pollen on the three axes of the morphospace. (a) Dated phylogeny of the 116 species from 41 families,
with pollination modes at the tips and abbreviated family names on the outer arcs. Yellow, insect-pollinated; blue, wind-pollinated. Scale
bar, 50 Myr. (b) Intensity, density and polarity of insect- and wind-pollinated species with Cohen’s _𝑑_ reported; colors as in (a). (c)
Differences between the groups (insect − wind) estimated by PGLS, with values on the right. Violins, _𝑡_ distributions of the estimates
with 95% confidence intervals; asterisks, _𝑝_ FDR _&lt;_ 0 _._ 05 (*), _&lt;_ 0 _._ 01 (**) and _&lt;_ 0 _._ 001 (***).


8


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint



turns the evolution of ornamentation into movement along
interpretable axes.

To assess how closely each axis follows the phylogeny,
we estimated Pagel’s _𝜆_ within Asteraceae (Fig. 5c). Only

_𝐷_ showed a significant phylogenetic signal ( _𝜆_ = 0 _._ 74,
95% CI 0.19 to 0.95, _𝑝_ FDR = 0 _._ 007 ). The signal in
_𝐼_ ( _𝜆_ = 0 _._ 39, 95% CI 0 to 0.90, _𝑝_ FDR = 0 _._ 164 ) and
_𝑃_ ( _𝜆_ = 0 _._ 34, 95% CI 0 to 0.83, _𝑝_ FDR = 0 _._ 090 ) was
not significant. Of the three axes, density thus tracks
the phylogeny most closely, as it separates the echinate
pollen of Asteroideae from the lophate pollen of many
Cichorioideae.


**4** **Discussion**


We have used persistent homology to construct a quantitative morphospace for pollen ornamentation, whose
interpretable axes allow theoretical, real and ancestral
pollen to be placed together. Morphospaces are either theoretical, with axes defined in advance, or empirical, with
axes obtained from an ordination of specimens (McGhee,
1999). The classic example of a theoretical morphospace
is that of shell coiling by Raup (1966), which founded theoretical morphology and originated in his attempt to capture
the “spectra of gradation” between qualitative categories
such as turriculate and turbiniform (Raup, 1961). Our morphospace pursues a similar aim for pollen ornamentation,
and its axes are likewise defined in advance. Beyond this,
the axes have no free parameters and are computed fully
automatically, so the coordinates require neither arbitrary
parameter choices nor manual intervention. The axes
can also be interpreted directly as the intensity, density
and polarity of the ornamentation, whereas the axes of
empirical morphospaces have only statistical meaning and
must be interpreted post hoc (Mitteroecker &amp; Gunz, 2009).
Finally, since the three axes capture distinct aspects of
ornamentation, differences along each can be analyzed
separately. By contrast, the independence of the axes of
classical morphospaces has long been debated; even in
Raup’s space, the axes covary mathematically (Schindel,
1990).
Our approach can be viewed as a generalization of LOanalysis. LO-analysis follows changes in the brightness of
the ornamentation as the focus moves up and down. As
the focal plane descends, protrusions appear first bright
and then dark (the LO-pattern), while depressions appear
first dark and then bright (the OL-pattern; Erdtman, 1956).
In our approach, the filtration value _𝑟_ corresponds geometrically to a spherical water surface of radius _𝑟_, which
for near-spherical pollen locally approximates the focal
plane. When the water level falls, the protrusions emerge
one by one as islands, and the water breaks into pools that
retreat into the depressions. These two processes mirror
the LO- and OL-patterns. As a generalization, persistent
homology extends LO-analysis in two respects. It covers
all filtration values rather than a few focal planes, and its
results are quantitative and do not depend on the judgment
of an observer.


One limitation of this study is the size and taxonomic
coverage of our sample. Because palynological records
consist mainly of two-dimensional light and scanning electron micrographs, few three-dimensional pollen data exist



beyond the 3D Pollen Library used here, which is the
largest public collection to date. Further limitations arise
from the pipeline itself. First, the coordinates may be
sensitive to differences in mesh resolution. Resampling
all meshes to the same resolution might reduce this effect. Second, the pipeline assumes the input pollen to be
near-spherical, so for markedly non-spherical pollen, the
overall shape could be mistaken for ornamentation. For
example, the poles of an ellipsoidal pollen grain would be
read as long spines. A possible alternative would be to
measure ornamentation relative to the overall shape, such
as that given by a low-order spherical harmonic expansion,
rather than to a sphere. Third, the pipeline treats apertures
as ornamental depressions, so both enter the depression
channel. The effect of this conflation is relatively limited,
however, because the three axes describe statistical features of all persistence bars, and apertures, in most cases,
account for only a small proportion of them. Fourth, to
keep the framework simple, we based our morphospace
only on zero-dimensional persistent homology, which does
not record the shape or spatial arrangement of ornamental
elements. The morphospace therefore cannot distinguish
ornamentation that differs only in these respects. Introducing higher-dimensional homology may overcome this
limitation.


Pollen ornamentation has long been described by qualitative categories. Our three scalars based on persistent
homology offer a continuous, interpretable quantification,
and new grains can be placed directly in the morphospace
they span. As larger and more balanced samples of threedimensional pollen data accumulate, this morphospace
may provide a common framework for pollen ornamentation.


**Acknowledgements**


We thank Anthony J. Hayes and the Bioimaging Research
Hub, Cardiff University, for generously making the 3D
Pollen Library available to the scientific community, without which this study would not have been possible. We
thank members of the Ishihara lab, in particular, Yajushi
Khurana, for comments on the manuscript. Claude Opus
4.8 (Anthropic) was used to assist with code development.
All AI-assisted outputs were reviewed, edited and validated
by the authors. This research received no specific funding.


**Competing** **interests**


None declared.


**Author** **contributions**


QD was involved in conceptualization, methodology, software, formal analysis, investigation, data curation, visualization and writing – original draft, review and editing.
KI was involved in supervision and writing – review and
editing.


**Data** **availability**


The three-dimensional pollen data analyzed in this study
are openly available from the 3D Pollen Library on NIH 3D
( `&#91;https://3d.nih.gov/collections/33&#93;(https://3d.nih.gov/collections/33)` ). Pollination modes were obtained from the Ecological Flora of the



9


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint


Figure 5: Evolution of pollen ornamentation in Asteraceae in the morphospace. (a) Phylomorphospace of the 28 species. Extant species
are colored by subfamily, and the shaded areas enclose the species of each subfamily. Gray lines and dots show the branches and estimated
ancestral states, and the red ring marks the estimated root. (b) The phylomorphospace on its _𝐷_ – _𝑃_, _𝐷_ – _𝐼_ and _𝑃_ – _𝐼_ faces, with _Artemisia_
_absinthium_ labeled. Colors and symbols as in (a). (c) Pagel’s _𝜆_ for the three axes, with values on the right. Violins, profile likelihoods
with 95% confidence intervals; asterisks as in Fig. 4.



British Isles ( `&#91;http://ecoflora.org.uk&#93;(http://ecoflora.org.uk)` ), and ornamentation types from PalDat ( `&#91;https://www.paldat.org&#93;(https://www.paldat.org)` ).
The code and data needed to reproduce all analyses are
available on GitHub ( `&#91;https://github.com/qianzhi&#93;(https://github.com/qianzhi-deng/Pollen-Ornamentation-Morphospace)`
`&#91;-deng/Pollen-Ornamentation-Morphospace&#93;(https://github.com/qianzhi-deng/Pollen-Ornamentation-Morphospace)` ).


**References**


Barnes, C.M., Power, A.L., Barber, D.G., Tennant, R.K.,

Jones, R.T., Lee, G.R., Hatton, J., Elliott, A., ZaragozaCastells, J., Haley, S.M., Summers, H.D., Doan, M.,
Carpenter, A.E., Rees, P. &amp; Love, J. (2023) Deductive
automated pollen classification in environmental samples via exploratory deep learning and imaging flow
cytometry. _New_ _Phytologist_ **240**, 1305–1326.


Benjamini, Y. &amp; Hochberg, Y. (1995) Controlling the false

discovery rate: a practical and powerful approach to
multiple testing. _Journal_ _of_ _the_ _Royal_ _Statistical_ _Society:_
_Series_ _B_ _(Methodological)_ **57**, 289–300.


Bonhomme, V., Prasad, S. &amp; Gaucherel, C. (2013) In
traspecific variability of pollen morphology as revealed
by elliptic Fourier analysis. _Plant_ _Systematics_ _and_ _Evo-_
_lution_ **299**, 811–816.


Bookstein, F.L., Strauss, R.E., Humphries, J.M., Chernoff,

B., Elder, R.L. &amp; Smith, G.R. (1982) A comment upon
the uses of Fourier methods in systematics. _Systematic_
_Zoology_ **31**, 85–92.



Brechbühler, C., Gerig, G. &amp; Kübler, O. (1995)

Parametrization of closed surfaces for 3-D shape description. _Computer_ _Vision_ _and_ _Image_ _Understanding_
**61**, 154–170.


Brown, M.J.M., Walker, B.E., Black, N., Govaerts, R.H.A.,

Ondo, I., Turner, R. &amp; Nic Lughadha, E. (2023) rWCVP:
a companion R package for the World Checklist of
Vascular Plants. _New_ _Phytologist_ **240**, 1355–1365.


Chartier, M., Jabbour, F., Gerber, S., Mitteroecker, P.,

Sauquet, H., von Balthazar, M., Staedler, Y., Crane, P.R.
&amp; Schönenberger, J. (2014) The floral morphospace –
a modern comparative approach to study angiosperm
evolution. _New_ _Phytologist_ **204**, 841–853.


Cohen-Steiner, D., Edelsbrunner, H. &amp; Harer, J. (2007)

Stability of persistence diagrams. _Discrete_ _&amp;_ _Computa-_
_tional_ _Geometry_ **37**, 103–120.


Dawson-Haggerty et al. (2019) trimesh.


Edelsbrunner, H., Letscher, D. &amp; Zomorodian, A. (2002)

Topological persistence and simplification. _Discrete_ _&amp;_
_Computational_ _Geometry_ **28**, 511–533.


Erdtman, G. (1956) ‘LO-analysis’ and ‘Welcker’s rule’: a

centenary. _Svensk_ _Botanisk_ _Tidskrift_ **50**, 135–141.


Erdtman, G. &amp; Straka, H. (1961) Cormophyte spore clas
sification. _Geologiska_ _Föreningen_ _i_ _Stockholm_ _Förhan-_
_dlingar_ **83**, 65–78.



10


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint



Fitter, A.H. &amp; Peat, H.J. (1994) The Ecological Flora

Database. _Journal_ _of_ _Ecology_ **82**, 415–425.


Friedman, J. &amp; Barrett, S.C.H. (2009) Wind of change:

new insights on the ecology and evolution of pollination
and mating in wind-pollinated plants. _Annals_ _of_ _Botany_
**103**, 1515–1527.


Gelb, A. (1997) The resolution of the Gibbs phenomenon

for spherical harmonics. _Mathematics_ _of_ _Computation_
**66**, 699–717.


Govaerts, R., Nic Lughadha, E., Black, N., Turner, R.

&amp; Paton, A. (2021) The World Checklist of Vascular
Plants, a continuously updated resource for exploring
global plant diversity. _Scientific_ _Data_ **8**, 215.


Grafen, A. (1989) The phylogenetic regression. _Philosoph-_

_ical_ _Transactions_ _of_ _the_ _Royal_ _Society_ _of_ _London._ _Series_
_B:_ _Biological_ _Sciences_ **326**, 119–157.


Halbritter, H., Ulrich, S., Grímsson, F., Weber, M.,

Zetter, R., Hesse, M., Buchner, R., Svojtka, M. &amp;
Frosch-Radivo, A. (2018) _Illustrated_ _pollen_ _terminol-_
_ogy_ . Springer, Cham, Switzerland, 2nd edn.


Hayes, A.J. (2024) The 3D Pollen Library collection at

NIH3D: from humble beginnings to the largest opensource collection of online 3D pollen models worldwide.
_infocus_ _Magazine_ **76**, 4–17.


Hornick, T., Harpole, W.S. &amp; Dunker, S. (2025) High
throughput assessment of anemophilous pollen size and
variability using imaging cytometry. _New_ _Phytologist_
**246**, 1875–1888.


Hornick, T., Richter, A., Harpole, W.S., Bastl, M.,

Bohlmann, S., Bonn, A., Bumberger, J., Dietrich, P.,
Gemeinholzer, B., Grote, R., Heinold, B., Keller, A.,
Luttkus, M.L., Mäder, P., Motivans Švara, E., Passonneau, S., Punyasena, S.W., Rakosy, D., Richter, R.,
Sickel, W., Steffan-Dewenter, I., Theodorou, P., Treudler,
R., Werchan, B., Werchan, M., Wolke, R. &amp; Dunker,
S. (2022) An integrative environmental pollen diversity assessment and its importance for the Sustainable
Development Goals. _Plants,_ _People,_ _Planet_ **4**, 110–121.


Jardine, P.E., Palazzesi, L., Tellería, M.C. &amp; Barreda,

V.D. (2022) Why does pollen morphology vary? Evolutionary dynamics and morphospace occupation in the
largest angiosperm order (Asterales). _New_ _Phytologist_
**234**, 1075–1087.


Jin, Y. &amp; Qian, H. (2022) V.PhyloMaker2: an updated

and enlarged R package that can generate very large
phylogenies for vascular plants. _Plant_ _Diversity_ **44**, 335–
339.


Kriebel, R., Khabbazian, M. &amp; Sytsma, K.J. (2017) A con
tinuous morphological approach to study the evolution
of pollen in a phylogenetic context: an example with
the order Myrtales. _PLoS_ _ONE_ **12**, e0187228.


Kuhl, F.P. &amp; Giardina, C.R. (1982) Elliptic Fourier features

of a closed contour. _Computer_ _Graphics_ _and_ _Image_
_Processing_ **18**, 236–258.



Li, M., An, H., Angelovici, R., Bagaza, C., Batushansky,

A., Clark, L., Coneva, V., Donoghue, M.J., Edwards,
E., Fajardo, D., Fang, H., Frank, M.H., Gallaher, T.,
Gebken, S., Hill, T., Jansky, S., Kaur, B., Klahs, P.C.,
Klein, L.L., Kuraparthy, V., Londo, J., Migicovsky, Z.,
Miller, A., Mohn, R., Myles, S., Otoni, W.C., Pires,
J.C., Rieffer, E., Schmerler, S., Spriggs, E., Topp, C.N.,
Van Deynze, A., Zhang, K., Zhu, L., Zink, B.M. &amp;
Chitwood, D.H. (2018) Topological data analysis as
a morphometric method: using persistent homology
to demarcate a leaf morphospace. _Frontiers_ _in_ _Plant_
_Science_ **9**, 553.


Luo, Y., Li, H.T., Lu, L., Davey, A.H., Zhang, M.Y., Yu,

Y., Jiang, W., He, H.J., Tao, Z.B., Yang, L.E., Zhang,
Y.J., Burgess, K.S., Blackmore, S., Jardine, P.E., Wang,

H. &amp; Li, D.Z. (2026) Pulsed evolution shaped extant
angiosperm pollen disparity. _Nature_ _Communications_
**17**, 1531.


Mander, L., Li, M., Mio, W., Fowlkes, C.C. &amp; Punyasena,

S.W. (2013) Classification of grass pollen through the
quantitative analysis of surface ornamentation and texture. _Proceedings_ _of_ _the_ _Royal_ _Society_ _B:_ _Biological_
_Sciences_ **280**, 20131905.


Maria, C., Boissonnat, J.D., Glisse, M. &amp; Yvinec, M.

(2014) The Gudhi Library: simplicial complexes and
persistent homology. _Mathematical_ _Software_ _–_ _ICMS_
_2014_ (eds. H. Hong &amp; C. Yap), vol. 8592 of _Lecture_
_Notes_ _in_ _Computer_ _Science_, pp. 167–174, Springer,
Berlin, Germany.


McGhee, G.R. (1999) _Theoretical_ _morphology:_ _the_ _con-_

_cept_ _and_ _its_ _applications_ . Columbia University Press,
New York, NY, USA.


Mitteroecker, P. &amp; Gunz, P. (2009) Advances in geometric

morphometrics. _Evolutionary_ _Biology_ **36**, 235–247.


Orme, D., Freckleton, R., Thomas, G., Petzoldt, T., Fritz,

S., Isaac, N. &amp; Pearse, W. (2025) _caper:_ _comparative_
_analyses_ _of_ _phylogenetics_ _and_ _evolution_ _in_ _R_ . R package
version 1.0.4.


Pagel, M. (1999) Inferring the historical patterns of bio
logical evolution. _Nature_ **401**, 877–884.


Paradis, E. &amp; Schliep, K. (2019) ape 5.0: an environment

for modern phylogenetics and evolutionary analyses in
R. _Bioinformatics_ **35**, 526–528.


Perry, I., Szeto, J.Y., Isaacs, M.D., Gealy, E.C., Rose,

R., Scofield, S., Watson, P.D. &amp; Hayes, A.J. (2017)
Production of 3D printed scale models from microscope volume datasets for use in STEM education. _EMS_
_Engineering_ _Science_ _Journal_ **1**, 002.


Pinheiro, J., Bates, D. &amp; R Core Team (2026) _nlme:_ _linear_

_and_ _nonlinear_ _mixed_ _effects_ _models_ . R package version
3.1-170.


Punt, W., Hoen, P.P., Blackmore, S., Nilsson, S. &amp; Le

Thomas, A. (2007) Glossary of pollen and spore terminology. _Review_ _of_ _Palaeobotany_ _and_ _Palynology_ **143**,
1–81.



11


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint


R Core Team (2025) _R:_ _a_ _language_ _and_ _environment_

_for_ _statistical_ _computing_ . R Foundation for Statistical
Computing, Vienna, Austria.


Radja, A. (2021) Pollen wall patterns as a model for bio
logical self-assembly. _Journal_ _of_ _Experimental_ _Zoology_
_Part_ _B:_ _Molecular_ _and_ _Developmental_ _Evolution_ **336**,
629–641.


Radja, A., Horsley, E.M., Lavrentovich, M.O. &amp; Sweeney,

A.M. (2019) Pollen cell wall patterns form from modulated phases. _Cell_ **176**, 856–868.e10.


Raup, D.M. (1961) The geometry of coiling in gastropods.

_Proceedings_ _of_ _the_ _National_ _Academy_ _of_ _Sciences,_ _USA_
**47**, 602–609.


Raup, D.M. (1966) Geometric analysis of shell coiling:

general problems. _Journal_ _of_ _Paleontology_ **40**, 1178–
1190.


Revell, L.J. (2024) phytools 2.0: an updated R ecosystem

for phylogenetic comparative methods (and other things).
_PeerJ_ **12**, e16505.


Romero, I.C., Kong, S., Fowlkes, C.C., Jaramillo, C.,

Urban, M.A., Oboh-Ikuenobe, F., D’Apolito, C. &amp; Punyasena, S.W. (2020) Improving the taxonomy of fossil
pollen using convolutional neural networks and superresolution microscopy. _Proceedings_ _of_ _the_ _National_
_Academy_ _of_ _Sciences,_ _USA_ **117**, 28496–28505.


Rosbakh, S., Nepi, M. &amp; Pichler, M. (2026) Does pollen

form follow function? Effects of stigma type, pollination
mode and habitat on pollen morphological traits. _Journal_
_of_ _Ecology_ **114**, e70290.


Schindel, D.E. (1990) Unoccupied morphospace and the

coiled geometry of gastropods: architectural constraint
or geometric covariation? _Causes_ _of_ _Evolution:_ _A_
_Paleontological_ _Perspective_ (eds. R.M. Ross &amp; W.D.
Allmon), pp. 270–304, University of Chicago Press,
Chicago, IL, USA.


Schluter, D., Price, T., Mooers, A.Ø. &amp; Ludwig, D. (1997)

Likelihood of ancestor states in adaptive radiation. _Evo-_
_lution_ **51**, 1699–1711.


Smith, S.A. &amp; Brown, J.W. (2018) Constructing a broadly

inclusive seed plant phylogeny. _American_ _Journal_ _of_
_Botany_ **105**, 302–314.


Sullivan, C.B. &amp; Kaszynski, A.A. (2019) PyVista: 3D

plotting and mesh analysis through a streamlined interface for the Visualization Toolkit (VTK). _Journal_ _of_
_Open_ _Source_ _Software_ **4**, 1450.


Weber, M. &amp; Ulrich, S. (2017) PalDat 3.0 – second revision

of the database, including a free online publication tool.
_Grana_ **56**, 257–262.


Wei, C., Li, M., Mao, L., Mander, L., Jardine, P.E.,

Gosling, W.D. &amp; Hoorn, C. (2025) A 23-million-year
record of morphological evolution within Neotropical
grass pollen. _New_ _Phytologist_ **246**, 365–376.


12


&#91;bioRxiv preprint doi: https://doi.org/10.64898/2026.11.18.767721; this version posted November 19, 2026. The copyright holder for this&#93;(https://doi.org/10.64898/2026.11.18.767721)

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in
perpetuity. It is made available under a Pollen Ornamentation &#91;CC-BY-NC 4.0 International license&#93;(http://creativecommons.org/licenses/by-nc/4.0/) Morphospace . A Preprint


**Supporting** **Information**


**A** **Derivation** **of** **the** **relation** **between** **density,** **polarity** **and** **wavenumber**


By definition, _𝐷_ is the Shannon entropy of the persistence fractions _𝜋_ _𝑖_ . If the ornamentation has only one channel and
consists of _𝑛_ elements of equal persistence, each element has the fraction _𝜋_ _𝑖_ = 1/ _𝑛_, so _𝐷_ = ln _𝑛_, exactly the logarithm
of the number of elements. Ornamentation, however, usually has both protrusions and depressions. To account for both
channels, we introduce the binary entropy of their fractions.

Write the fractions of the protrusion and depression channels as



_𝜋_ ± =



∑︁

_𝜎_ _𝑖_ =±1



_𝜋_ _𝑖_



The two fractions sum to 1, and by the definition of _𝑃_ their difference is _𝜋_ + − _𝜋_ − = &#91;�&#93; _𝑖_ _&#91;𝜋&#93;_ _𝑖_ _&#91;𝜎&#93;_ _𝑖_ &#91;=&#93; &#91;tanh&#93; _&#91; 𝑃&#93;_ &#91;.&#93; &#91;Hence&#93;



_𝜋_ ± = &#91;(&#93; _&#91;𝜋&#93;_ &#91;+&#93; &#91; +&#93; _&#91; 𝜋&#93;_ &#91;−&#93; &#91;)&#93;&#91; ± &#93;&#91;(&#93; _&#91;𝜋&#93;_ &#91;+&#93; &#91; −&#93; _&#91;𝜋&#93;_ &#91;−&#93; &#91;)&#93;



2




&#91;(&#93; _&#91;𝜋&#93;_ &#91;+&#93; &#91; −&#93; _&#91;𝜋&#93;_ &#91;−&#93; &#91;)&#93; &#91;tanh&#93; _&#91; 𝑃&#93;_

= &#91;1&#93;&#91; ±&#93;
2 2



and their binary entropy is

_ℎ_ ( _𝑃_ ) = − _𝜋_ + ln _𝜋_ + − _𝜋_ − ln _𝜋_ −


If each channel consists of _𝑛_ elements of equal persistence, each element in the protrusion or depression channel has the
fraction _𝜋_ + / _𝑛_ or _𝜋_ − / _𝑛_ . Substituting into the definition of _𝐷_ gives



_𝐷_ = −



_𝜋_ ± ln _&#91;𝜋&#93;_ &#91;±&#93; &#91;ln&#93; _&#91; 𝑛&#93;_ &#91;+&#93; _&#91; ℎ&#93;_ &#91;(&#93; _&#91;𝑃&#93;_ &#91;)&#93;

∑︁ _𝑛_ &#91;=&#93;

±



Since _ℓ_ 0 counts how many times the pattern repeats around the grain, _ℓ_ 0 = 2 _𝜋𝑅_ / _𝜆_, where _𝑅_ is the radius of the grain and _𝜆_
is the characteristic wavelength of the pattern. If each period of the pattern occupies an area of _𝜆_ &#91;2&#93;, each channel contains
_𝑛_ = 4 _𝜋𝑅_ &#91;2&#93; / _𝜆_ &#91;2&#93; = _ℓ_ 0 &#91;2&#93; &#91;/&#93; _&#91;𝜋&#93;_ &#91;elements,&#93; &#91;so&#93;



_𝐷_ = ln



Rearranging gives



_ℓ_ &#91;2&#93;

0
_𝜋_ &#91;+&#93; _&#91; ℎ&#93;_ &#91;(&#93; _&#91;𝑃&#93;_ &#91;)&#93;


_𝜋𝑒_ _&#91;𝐷&#93;_ &#91;−&#93; _&#91;ℎ&#93;_ &#91;(&#93; _&#91;𝑃&#93;_ &#91;)&#93;



_ℓ_ 0 =



√︁



**B** **Counted** **and** **predicted** **spine** **numbers** **of** **19** **echinate** **Asteraceae** **pollen** **grains**


Predicted _𝑛_ Error (%)


Grain Species NIH 3D Counted _𝑛_ _𝑒_ _&#91;𝐷&#93;_ _𝑒_ _&#91;𝐷&#93;_ &#91;−&#93; _&#91;ℎ&#93;_ &#91;(&#93; _&#91;𝑃&#93;_ &#91;)&#93; _𝑒_ _&#91;𝐷&#93;_ _𝑒_ _&#91;𝐷&#93;_ &#91;−&#93; _&#91;ℎ&#93;_ &#91;(&#93; _&#91;𝑃&#93;_ &#91;)&#93;


1 _Tripleurospermum_ _maritimum_ &#91;3DPX-021473&#93;(https://3d.nih.gov/entries/3DPX-021473) 52 67 46 29.2 −10 _._ 9
2 _Leucanthemum_ _vulgare_ &#91;3DPX-017803&#93;(https://3d.nih.gov/entries/3DPX-017803) 50 68 51 36.9 2.1
3 _Senecio_ _squalidus_ &#91;3DPX-021972&#93;(https://3d.nih.gov/entries/3DPX-021972) 76 84 47 10.0 −38 _._ 6
4 _Petasites_ _hybridus_ &#91;3DPX-021106&#93;(https://3d.nih.gov/entries/3DPX-021106) 65 92 61 40.9 −6 _._ 3
5 _Tussilago_ _farfara_ &#91;3DPX-021534&#93;(https://3d.nih.gov/entries/3DPX-021534) 68 98 64 43.5 −6 _._ 0
6 _Helianthus_ _annuus_ &#91;3DPX-017880&#93;(https://3d.nih.gov/entries/3DPX-017880) 82 103 83 25.3 0.9

7 _Serratula_ _tinctoria_ &#91;3DPX-021968&#93;(https://3d.nih.gov/entries/3DPX-021968) 79 104 58 31.5 −26 _._ 1

8 _Pulicaria_ _dysenterica_ &#91;3DPX-021602&#93;(https://3d.nih.gov/entries/3DPX-021602) 79 108 80 36.9 0.7
9 _Jacobaea_ _vulgaris_ &#91;3DPX-021541&#93;(https://3d.nih.gov/entries/3DPX-021541) 84 116 71 38.6 −15 _._ 7
10 _Solidago_ _virgaurea_ &#91;3DPX-021264&#93;(https://3d.nih.gov/entries/3DPX-021264) 89 118 82 32.3 −8 _._ 3
11 _Bellis_ _perennis_ &#91;3DPX-017818&#93;(https://3d.nih.gov/entries/3DPX-017818) 92 122 86 32.6 −6 _._ 0
12 _Bellis_ _perennis_ &#91;3DPX-021089&#93;(https://3d.nih.gov/entries/3DPX-021089) 89 124 81 38.9 −8 _._ 7
13 _Cirsium_ _vulgare_ &#91;3DPX-017799&#93;(https://3d.nih.gov/entries/3DPX-017799) 85 126 85 48.4 −0 _._ 1
14 _Senecio_ _vulgaris_ &#91;3DPX-020862&#93;(https://3d.nih.gov/entries/3DPX-020862) 86 131 86 52.5 −0 _._ 1
15 _Senecio_ _integrifolius_ &#91;3DPX-021967&#93;(https://3d.nih.gov/entries/3DPX-021967) 107 136 98 26.9 −8 _._ 0
16 _Ambrosia_ sp. &#91;3DPX-018227&#93;(https://3d.nih.gov/entries/3DPX-018227) 138 171 115 24.0 −16 _._ 8
17 _Ambrosia_ sp. &#91;3DPX-020096&#93;(https://3d.nih.gov/entries/3DPX-020096) 161 180 123 12.0 −23 _._ 7
18 _Centaurea_ _nigra_ &#91;3DPX-017767&#93;(https://3d.nih.gov/entries/3DPX-017767) 199 244 142 22.4 −28 _._ 6
19 _Calendula_ _officinalis_ &#91;3DPX-017814&#93;(https://3d.nih.gov/entries/3DPX-017814) 203 245 166 20.5 −18 _._ 1


13


