Abstract
Morphospace representations of phenotypic characteristics are often populated unevenly, leaving large parts unoccupied. Such patterns are typically ascribed to contingency, or else to natural selection disfavouring certain parts of the morphospace. The extent to which developmental bias, the tendency of certain phenotypes to preferentially appear as potential variation, also explains these patterns is hotly debated. Here we demonstrate quantitatively that developmental bias is the primary explanation for the occupation of the morphospace of RNA secondary structure (SS) shapes. Upon random mutations, some RNA SS shapes (the frequent ones) are much more likely to appear than others. By using the RNAshapes method to define coarse-grained SS classes, we can directly compare the frequencies that non-coding RNA SS shapes appear in the RNAcentral database to frequencies obtained upon random sampling of sequences. We show that: a) Only the most frequent structures appear in nature; the vast majority of possible structures in the morphospace have not yet been explored. b) Remarkably small numbers of random sequences are needed to produce all the RNA SS shapes found in nature so far. c) Perhaps most surprisingly, the natural frequencies are accurately predicted, over several orders of magnitude in variation, by the likelihood that structures appear upon uniform random sampling of sequences. The ultimate cause of these patterns is not natural selection, but rather strong phenotype bias in the RNA genotype-phenotype map, a type of developmental bias or “findability constraint”, which limits evolutionary dynamics to a hugely reduced subset of structures that are easy to “find”.
Introduction: The Mystery of the Empty Morphospace
In evolutionary biology, the "morphospace" represents the total set of all possible forms a biological structure can take. When we observe the natural world, we find that this space is remarkably empty; only a tiny fraction of theoretically possible shapes actually exist. Traditionally, scientists have explained these gaps through two primary lenses: natural selection (the missing shapes were unfit and eliminated) or contingency (evolution is a historical accident and simply hasn't reached those regions yet).
This paper explores a third, often-overlooked possibility: developmental bias. This concept suggests that the "map" from genotype (the genetic sequence) to phenotype (the physical form) is inherently skewed. If certain shapes are mathematically much easier to "find" via random mutation than others, evolution may be trapped on a path of "low-hanging fruit," regardless of the potential fitness of more obscure shapes. This creates a "findability constraint" that dictates the menu from which natural selection can choose.
Methodology: Mapping the RNA Genotype-Phenotype Space
To test this hypothesis, the researchers focused on RNA secondary structures (SS). RNA is often called the "hydrogen atom of developmental biology" because the rules governing how a sequence of nucleotides (A, U, C, G) folds into a shape are well-understood and computationally predictable.
The study utilized two main approaches:
- G-sampling (Genotype Sampling): The team generated millions of random RNA sequences and used the standard Vienna package algorithm to predict their folded shapes, determining which structures appear most frequently by chance.
- Coarse-Graining: To compare these results to nature, they used the RNAshapes method to group structures into levels of abstraction. For example, a "Level 5" abstraction ignores individual base pairs and only looks at the high-level arrangement of helices and loops.
These sampled frequencies ($f_p^G$) were then compared against natural data from RNAcentral, a massive database of non-coding RNA sequences, and Rfam, which provides structures determined by consensus alignment.
Results: The Arrival of the Frequent
The results reveal a staggering correlation between mathematical probability and natural occurrence. The researchers found that the frequency of shapes in nature correlates remarkably well (Pearson $r \approx 0.9$) with the frequency of shapes produced by uniform random sampling.
As shown in the rank-frequency plots, the likelihood of a structure appearing in a random sequence predicts its prevalence in the natural world over several orders of magnitude. Remarkably, over 99% of the abstract shapes found in natural databases were discovered through relatively small-scale random sampling. This suggests that nature is not exploring the full breadth of the morphospace, but is instead restricted to the most "findable" structures. This phenomenon is termed the "Arrival of the Frequent": before a structure can be selected for its fitness, it must first appear frequently enough for evolution to "see" it.
Discussion and Implications
This "findability constraint" acts as a powerful filter on evolutionary dynamics. For an RNA of length 126, there are $10^{75}$ possible sequences—a hyper-astronomical number. Yet, a sample of just a few million random sequences is enough to recover nearly every shape found in nature. This implies that the vast majority of possible biological forms have never been tested by selection because they are simply too difficult to find through random mutation.
Interestingly, the authors note a parallel in artificial intelligence. The Stochastic Gradient Descent (SGD) used to train Deep Neural Networks (DNNs) also tends to find "frequent" functions—those that have a large volume of parameter space mapping to them. This suggests that "simplicity bias" or phenotype bias might be a universal property of complex mapping systems, whether biological or computational.
Conclusion
By quantitatively comparing natural RNA databases to random sequence sampling, this study demonstrates that phenotype bias is a primary driver of biodiversity patterns. The "missing" shapes in the RNA morphospace aren't necessarily unfit; they are simply absent because the genotype-phenotype map makes them nearly impossible to encounter. Evolution, it seems, is less a journey through all possibilities and more a selection from a highly biased menu of the mathematically probable.
Summary updated