Computational model of gene network evolution reveals hidden buffer cells are crucial for stable petal pattern development
A collaboration between computational modellers and experimental biologists at the Sainsbury Laboratory Cambridge University (SLCU) has revealed how a hidden ‘boundary’ cell type spontaneously emerges during flower development, ensuring stable petal patterning.
More broadly, the research shows that the evolution of physical traits does not always result directly from natural selection, but can arise as a by-product of the selection for other traits.
A two dimensional, cell-based model representing the petal epidermis: Gene regulation dynamics, governed by a gene regulatory network are simulated within each epidermal cell to drive the patterned expression of a proximal identity gene (PROX) that is expressed in the proximal region and a distal identity gene (DIST) in the distal region. The gene regulatory network dynamics are then simulated on the petal tissue to produce a phenotype. This is a proxy for the actual, more complex selection pressures on floral pattern arising from pollinator preference and abiotic factors. Graphic by first author Steven Oud.
Published in Development, teams led by Dr Renske Vroomans (computational modelling) and Dr Edwige Moyroud (experimental biology) present new evidence that these boundary cells may represent an ancient, conserved biological patterning mechanism present in both leaves and petals, which share an evolutionary origin dating back over 125 million years.
These boundary cell types offer a clear example of what palaeontologists Stephen Jay Gould and Richard Lewontin in 1979 termed ‘spandrels’ – traits that arise as by-products of evolutionary processes rather than through natural selection.
The findings also provide additional support for the idea that there is a bias in evolution towards specific novel patterns due to constraints of developmental systems.
Similar adaptations for diverse patterns
Flower petal patterns like spots, stripes and bullseye are diverse in form and fulfil crucial functions: They attract and guide pollinators to nectar and pollen rewards but also protect the reproductive part of the flowers from desiccation and UV damage. Hence, these colourful motifs are important for the reproductive success of flowering plants.
These petal patterns are not only formed by pigments, but also by specialised cell shapes and surface structures that reflect ultraviolet light, which is invisible to humans but visible to many pollinators such as bees. However, we still don’t fully understand how flowering plants build these patterns.
Hibiscus trionum petal developmental: The flower of H. trionum features a bullseye pattern on its corolla. During the early stages of petal development there is no pattern visible to the naked eye. To investigate how the pattern forms, a computational model was developed with a gene regulatory network within each epidermal cell to simulate the evolution of pattern formation.
Simulating millennia in minutes
To investigate how these petal patterns arise, Dr Vrooman’s team created a two-dimensional cell-based model of the petal epidermis.
They simulated more than 30,000 generations of flowers, allowing gene regulatory networks (GRN) to naturally evolve and mutate.
The simulations were programmed with a single selection pressure: create a stable, two-tone bullseye pattern consisting of a central proximal zone and an outer distal zone.
Boundary cell type plays a key role in correct bullseye patterning: (A) Pruned GRN of an individual from a representative simulation at generation 30,000. Vertex labels indicate gene types. Multiple arrows between nodes reflect multiple TFBs, resulting in a stronger regulatory effect on transcriptional activity (see Methods). (B) Functional analysis of gene 5 in the GRN. (i) Wild-type phenotype showing normal bullseye pattern formation. (ii) Phenotypic effects of gene 5 knockout and overexpression, both leading to bullseye loss. (iii) Gene 5 restricts gene 10 expression to the distal region via cell-cell signalling, establishing a boundary inhibition zone essential for pattern formation. (C) Mutagenesis experiment converting gene 5 from a cell-cell communication gene, which influences neighbouring cells’ transcription, to one coding for a transcription factor (TF) acting cell-autonomously, only influencing transcription in Development • Accepted manuscriptcells in which it is expressed. (i) Resulting mutant phenotype showing loss of boundary cell type and bullseye symmetry. (ii) Temporal gene expression of genes 5 and 10 in the wild type and mutant, shown for three cells within the proximal bullseye region along the mediolateral axis. In the mutant, expression of gene 5 is both delayed (ΔT), reduced in magnitude (ΔC) eventually lost with increasing distance from the signal origin.
Individuals are selected for a bullseye pattern, characterised by mutually exclusive expression domains of PROX and DIST. Examples of both high-fitness (left) and low-fitness (right) phenotypes are shown. In the low fitness example, unwanted PROX and DIST expression is marked in red. The proportion of the bullseye pattern is flexible and can evolve, ranging between 20% and 80% (proximal to total petal height).
Not all traits evolve through direct selection
“Remarkably, in over 70% of the evolutionary lineages, a third, unselected cell type spontaneously emerged precisely at the boundary where the two zones met,” said first author and PhD student, Steven Oud.
“We explicitly selected for just two cell types arranged in a bullseye, yet a third cell type appeared repeatedly at the boundary.”
All biological systems, from bacteria and animals to plants, exhibit natural fluctuations in gene expression, often called molecular noise. This is caused by natural variation in the number of proteins produced by identical genes in identical cells in the same environment.
“When we introduced molecular noise into the model these boundary cells persisted for significantly more generations,” Dr Vroomans said. “This suggests that they aren’t just an accidental by-product, but play a dynamic role in buffering the patterning process against molecular noise and developmental variability.”
Please allow social and marketing cookies to show embedded content.
Representative developmental mechanism which creates a boundary cell type by expression profile II: Gene 3 and 7 are expressed in proximal domains of different sizes, leading to a boundary cell type through expression profile II. The bullseye boundary cell type evolves most often when genes display overlapping bullseye expression patterns with different proportions.
Verification in the glasshouse
To test if these computer-predicted boundary cells actually existed in nature, Dr Edwige Moyroud’s experimental team analysed the petals of Hibiscus trionum (Venice mallow), which is a small herbaceous plant with large cream flowers with a distinct purple centre that the team is using as a model plant to study biological cell patterning.
The proximal zone of mature H. trionum flowers is coloured with a dark purple pigment but also have flat elongated striated cells, while the distal petal zone is cream with smooth conical shaped cells.
Recent work by Dr Moyroud’s team had shown that the very first signs of petal pattern formation is a pre-patterning band of large elongated cells appearing about one-third of the way from the petal base.
At this very early stage the petal cells look identical in colour, but the band of elongated cells precisely marks where the future purple and cream domains will divide.
Using RNA sequencing to analyse gene expression across different regions of early-stage petals, the researchers identified more than 600 genes that were significantly upregulated at this boundary zone.
Crucially, 30 genes showed a clear peak of expression in the boundary zone, matching the exact same profile as the computational model.
“Boundary cells possess a genetic signature long before the bullseye pattern becomes visible and the model predicted the biological reality we could not see with the naked eye,” said Dr Moyroud. “It is a great illustration of how collaborative work between computational and experimental scientists promote discovery and advance our knowledge of the natural world.”
More than half these 30 active boundary zone genes are known to be involved in cell wall remodelling, hormonal signalling and cell expansion.
“It is likely that these genes are contributing to the differences in cell shapes and cuticle decorations that lead to the two distinctly different cells present in mature petals but they could also be important to initiate the formation of the pattern itself,” said Dr Moyroud. “Our next challenge is to understand how the boundary cells are specified in the first place.”
Close-up examination of the petal surface in mature flowers (stage 5) confirms the existence of a visible boundary region where epidermal cell features are distinct from the proximal and distal cells. Boundary cells are flat and elongated (tabular), but with a smooth surface. Cells in the lower boundary are anthocyaninpigmented, while cells in the upper boundary are not.
Evolutionary implications
These findings indicate that boundary cells are not necessarily directly selected traits, but may also arise from the underlying architecture of developmental systems.
Because leaves and petals share an ancient evolutionary origin dating back over 125 million years, the researchers propose that such boundary cell types may be part of ancient, conserved mechanisms all flowering plants use to pattern the surface of their organs and produce motifs allowing them to manipulate animal behaviour or deal with environmental stress.
“Alternatively, we find that these boundary cells can evolve easily and repeatedly, so different species may have invented them independently. To know for sure, we will have to join forces with other research teams that study petal patterns in other species,” said Dr Vroomans.
“By demonstrating how simple gradients can be reliably converted into distinct regional domains provides us with new insights into our understanding of plant morphogenesis, and how robust biological patterns arise in a noisy environment.”
Highly transient evolution of boundary cell type: (A) Evolution of boundary cell types in deterministic Simulation #13, illustrating the diversity of boundary mechanisms that evolved within a single simulation. The presence of boundary expression profile I is depicted in pink, while boundary expression profile II is shown in blue. The gene(s) responsible for each emergent boundary cell type are displayed alongside the respective mechanism: for expression profile I, the gene preferentially expressed in the boundary; for expression profile II, the two genes with overlapping expression domains of uneven lengths. Expression patterns of these genes are shown at selected generations. (B) For each independent emergence of the boundary cell type mediated by gene 10 (expression profile I), we identify the final mutation(s) responsible. The first appearance at generation 7700 (i) results from a single transcription factor binding site (TFBS) weight inversion; the second at generation 13000 (ii) from three mutations (two TFBS deletions and one TFBS insertion); and the third at generation 27000 (iii) from a single TFBS insertion. All GRNs shown are unpruned networks with leaf and unreached nodes removed (i.e., genes that do not regulate other genes or are never activated) for visual clarity.
Reference
Steven Oud, Maciej M. Żurowski, Pjotr L. van der Jagt, May T. S. Yeo, Joseph F. Walker, Edwige Moyroud, Renske M. A. Vroomans; Computational model of flower pattern evolution predicts spontaneous emergence of boundary cell types across petal epidermis. Development 2026; dev.205745. DOI: https://doi.org/10.1242/dev.205745