Evolutionary Vibes:

When it comes to the study of evolution, something is appealing about evolutionary adaptations. Understanding how we, or any species for that matter, adapted not only to survive but thrive in an environment speaks to the resilience of life (“life finds a way”). This makes it possible to craft exciting narratives around the what, how, and why of evolutionary adaptations, particularly human adaptations. Pick any popular science book on the topic of evolution: you will likely find numerous examples of how life adapted to new environments. A good narrative is a powerful tool for communicating scientific results to the general public and other researchers. However, even the best scientist can let a narrative go beyond the evidence. The proverb “never let the truth get in the way of a good story,” commonly attributed to Mark Twain, implies a power of emotion over logic: a powerful narrative will almost always outpace a convincing piece of evidence. 

Although natural selection is an important, if not the most important, evolutionary process, it is also not an all-powerful force driving species towards perfection. However, many biologists across all disciplines (including evolutionary biology) often treat adaptation by natural selection as the de facto null: if we observe a trait and that trait has a function, then it must have been an adaptation for something. From there, you run into the problem with “adaptive storytelling,” or “just-so stories,” the latter in reference to Rudyard Kipling's fanciful fictional children’s tales of how certain animal traits came to be (e.g., “How the leopard got its spots”). Again, adaptation by natural selection has a strong narrative appeal, and it is possible to construct plausible-sounding explanations for why a trait is an adaptation if you start from the assumption that it is an adaptation. 

To this day, the most (in)famous paper tackling the issue of adaptive storytelling in evolutionary biology is Dr. Stephen J. Gould and Dr. Richard Lewontin’s 1979 article “The spandrels of San Marco and the Panglossian paradigm: a critique of the adaptationist programme,” (Gould & Lewontin, 1979). (Please, take a moment to appreciate the pedanticism of that title.) The term “Panglossian” alludes to the character of Dr. Pangloss from the French philosopher Voltaire's 18th-century satirical novel Candide. Pangloss believes the current world to be the best of all possible worlds and that “things cannot be otherwise than they are, for since everything is made to serve an end, everything necessarily serves the best end.” For example, Pangloss argues that our noses were made to support glasses; thus, we wear glasses. Clearly, this inverts the cause and the effect: our glasses are designed to fit our noses. Gould and Lewontin argued that many evolutionary biologists were Panglossian in their approach to studying the evolution of traits because they a priori assume all traits must be an adaptation created by natural selection for (i.e., they are tuned to their environment). They deemed this the “adaptationist programme” with 5 key practices (nicely summarized by Pigliucci and Kaplan (2000)):

• Organisms are composed of independent traits or characteristics, which can be considered evolving independently of one another.

• Natural selection is so strong relative to other evolutionary processes that it can be assumed most traits are adaptations.

• A plausible explanatory story that proposes a role for natural selection and is consistent with the observed trait is sufficient for the preliminary acceptance of the adaptive hypothesis. To put another way, the adaptive story needs to have the right vibes. 

• Should one adaptive story fail, search for another adaptive story.

• The failure of a trait to be non-optimal is due to trade-offs with other adaptive traits.

Gould and Lewontin argue that, given these practices, plausible-sounding adaptive stories are easy to come up with, but hard to refute (i.e., falsify), which goes against the standards of good science. Instead, evolutionary biologists must be more rigorous in identifying adaptations and consider non-adaptive evolutionary processes such as genetic drift (the random fluctuation of alleles, or versions of a gene, from generation to generation due to finite population sizes) and constraints on natural selection, such as constraints on organism development.

The “spandrels of San Marco” also introduced the term “spandrel” into the evolutionary biologists’ vocabulary. “The spandrels of San Marco” opens with a description of the Christian artwork adorning the dome of St. Mark’s Cathedral in Venice, Italy. The dome sits atop 4 arches, creating a triangular space. Given the beauty of the artwork in the spandrels, Gould and Lewontin argue that one might assume the structure was designed to fit the artist’s vision. Much like Dr. Pangloss, this logic inverts the cause and effect: the spandrels are a natural result of putting a dome structure on top of 4 arches, and the artwork was created around this constraint. Analogous to this, an evolutionary spandrel is a trait that was assumed to be an adaptation but is actually the result of other evolutionary processes or constraints. Notably, an evolutionary spandrel need not imply non-functional. According to Gould and Lewontin, functional spandrels could arise by co-opting a trait that emerged for another purpose. For example, feathers are now thought to originally function in heat regulation for dinosaurs, but have since been co-opted for flight by birds.

Ultimately, the general importance of “The Spandrels of San Marco” remains contested. Some argue they provided no real solution, while others claim there was no problem to begin with (Pigliucci & Kaplan, 2000; Queller, 1995). At a minimum, it likely made evolutionary biologists more careful in their analysis before claiming a trait is an adaptation created through natural selection (Nielsen, 2009). I can say it significantly influenced my thinking on evolution and approach to my research, and I think it is more important now than was written.

“The Spandrels of San Marco” was written well before the omics era in biology, which is when it became possible to sequence large amounts of DNA (relatively) quickly. The omics era presents its own challenges regarding evolutionary spandrels. In my opinion, genomics and molecular biology research are often implicitly adaptationist. I am not the only one to feel this way (Koonin, 2016; Linquist et al., 2020; Palazzo & Kejiou, 2022; Sarkar, 2015). Although genomics and molecular biology may not explicitly deal with evolutionary fitness, research questions and functional results are often developed and interpreted through the lens of natural selection. As noted earlier, a biochemical function need not imply that it emerged as an adaptation driven by natural selection. 

Compounding this adaptationist problem is the large amounts of data we are now able to generate and analyze. Large datasets make it possible to detect statistically significant small effects. In this sense, large datasets can be a double-edged sword: you get the power to detect lots of patterns, but that also means you’re more likely to detect spurious noise if you stick with a p-value cutoff of 0.05. (If you are unfamiliar with p-values, when the p-value from a statistical test comparing 2 or more things is less than a pre-defined threshold – usually 0.05 – the result is considered statistically significant. In the lives of many scientists, the difference between 0.049 and 0.05 is the difference between a good and a bad day.) Statisticians have been ringing this bell for years (Wasserstein & Lazar, 2016). As a result, there are many opportunities for just-so stories to emerge.

I have encountered examples of spandrels in my own research. A major focus of mine for the last decade has been the evolution of codon usage bias, or the non-uniform usage of codons that code for the same amino acid within an organism’s DNA. (If you are unfamiliar with these terms, please read the Appendix before continuing). Mutations in a gene that change a codon but not the translated amino acid are known as “synonymous mutations.” It was once thought that synonymous mutations must be neutral because they do not alter the protein sequence. As we began to sequence genes, it became clear that codons were not used at equal frequencies within an organism’s genome. This is largely due to mutation biases (e.g., the frequency of GC nucleotides vs. AT nucleotides), but various patterns were suggestive of natural selection on codon usage. For example, the codons used most frequently in highly expressed genes tend to be those with highly abundant transfer RNA (tRNA), which recognize specific codons and carry the amino acid to the elongating peptide chain in the ribosome. More abundant tRNAs are expected to lead to shorter waiting times between peptide elongations, leading to more efficient synthesis of the protein. Population genetics models accounting for natural selection, mutation bias, and genetic drift showed that codon usage is often at an equilibrium between these evolutionary forces (selection-mutation-drift equilibrium), with estimates of natural selection (called selection coefficients) correlating with tRNA abundances (Shah & Gilchrist, 2011). 

As new ways of measuring biochemical function and more sequences became available, people began looking for new patterns of codon usage. One idea was that codon usage in the signal peptides – short N-terminal markers of protein secretion – in E. coli may be under natural selection for increased use of slow codons. The hypothesis posits that slow codons could improve the ability of the relevant chaperones involved in the secretion mechanism (Zalucki et al., 2009). Bioinformatic analysis of signal peptides in E. coli showed they do use slow codons more often than non-secreted proteins (Power et al., 2004). Functional analysis by mutating slow codons to fast codons of a few select proteins seemingly confirmed a role for slow codons. (I will note that in one study, they changed their definition of slow codon after they did not originally observe the expected effect.) (Zalucki et al., 2010; Zalucki & Jennings, 2007). To be clear, these are very nice molecular biology experiments, but was it safe to conclude that the slow codon usage in signal peptides was driven by natural selection?

When I quantified natural selection on codon usage in signal peptides, I found no evidence that slow codons were more or less disfavored in signal peptides than at the beginning of any other genes (Cope et al., 2018). This suggests the observed difference in codon usage is not driven by natural selection! Furthermore, when I simulated codon usage under the assumption of selection-mutation-drift equilibrium, I found signal peptides were expected to use the slow codons more frequently 100% of the time. This was despite my simulations assuming no differences in natural selection between signal peptides and the rest of the genome. This result was due, in part, because signal peptides tend to use certain amino acids. Even though signal peptides are highly variable in their specific sequences, they typically are constrained to specific biochemical properties that bias amino acid usage. In my opinion, this makes the usage of slow codons in signal peptides an example of a molecular spandrel: the observed pattern was not the result of natural selection, but the result of a constraint on amino acid usage. While I am very proud of that work, it was small potatoes compared to one of the major clashes between genomics/molecular biology and evolutionary biology in the early 2010s. 

Early studies on human genetics (even back to the 1960s) suggested that the majority of the human genome was likely non-coding (Palazzo & Kejiou, 2022). When the human genome was first assembled, genes encoding proteins made up only about 2% of the genome. So, what about the other 98%. With no clear function, the non-coding regions of the human genome were termed “junk DNA.” But was it really junk? This led to an international effort involving over 500 scientists to characterize the functional capacity of the ~3 billion nucleotides in the human genome. Termed the Encyclopedia of DNA Elements (ENCODE) Project Consortium, they were able to detect biochemical activity in 80% of the human genome, claiming to put an end to the concept of junk DNA (Dunham et al., 2012). 

The blowback from the 2012 ENCODE Project was swift and in some cases, brutal (Doolittle, 2013; Eddy, 2013; Graur et al., 2013). Many evolutionary biologists argued that if these DNA regions were truly functional (i.e., affected fitness), then they would exhibit signatures of natural selection at these sites; however, only about 10% of the human genome exhibits signatures of natural selection (Graur et al., 2013). Part of the problem stemmed from how the ENCODE team defined function, conflating the mere presence of reproducible biochemical activity with biochemical activity that is critical for human function and survival (Linquist et al., 2020). This does not mean that our non-coding DNA is completely deprived of any significant function. There are clearly some examples out there; however, these are the exceptions, not the rules.

The ENCODE project (Dunham et al., 2012) was essentially a map of biochemical activity across the human genome – an amazing scientific accomplishment – that was overblown to improve coverage and build hype (Eddy, 2013). But think about ENCODE’s logic and how easy it was to sell to the media: it must serve some role because we have it. If not, why wouldn’t it be removed by natural selection? To tie it all together, because natural selection is not all-powerful. There are many hypotheses about why genome size varies across the domains of life, and many do not invoke natural selection. One simple hypothesis is that genetic drift, a non-adaptive process, is more powerful than natural selection to remove non-functional DNA in species with larger genomes (Lynch & Conery, 2003). 

I do not intend to dissuade you from the importance of natural selection or adaptations in evolution. Both played a critical role in shaping the diversity of life. As with all hypotheses, we must carefully evaluate our adaptive hypotheses and consider the other possible causes for the observed patterns in biological data. We will, undoubtedly, occasionally miss something. That’s just part of scientific research. But this will hopefully lead us to being more confident in what we determine to be adaptations driven by natural selection.