Sewall Wright concocted one of those metaphors in science that lingers long past its “best by” date. In 1932, he coined the term “the fitness landscape.” He envisioned a mythical land of peaks and valleys, with the peaks indicating higher fitness, and valleys populated by evolving organisms starting out on their journeys toward progressively higher fitness levels. Impelled by the struggle for existence, organisms would climb higher till reaching a peak. One difficulty with this picture appeared soon after the metaphor gained popularity: to get to a higher peak, an organism would have to climb down, lowering its fitness on the way to a neighboring peak. Some workarounds were concocted, but the evocative metaphor lent itself to 3-D graphs and formulas of positive selection, giving evolutionary biologists hopes of empirical rigor as they measured their research organism’s progress up the landscape.

Stability of the Landscape

Unwarranted assumptions are the bugaboo of clever models like this. One is the stability of the landscape. Does the hypothetical landscape undulate over time, such that a peak at one epoch becomes a valley in another? After all, the dynamic environment is oblivious to the needs of organisms. How quickly does a given habitat change? How can evolutionists be sure that fitness for a savannah does not become a detriment if the population finds itself in a habitat undergoing desertification? For reasons like this, Mustonen and Lässig in 2009 dubbed it a fitness “seascape” instead of a landscape.

Another of Wright’s assumptions was that the fitness landscape follows Gaussian curves consisting of smooth lines without discontinuities. Even if some of those Gaussian curves rose steeply like a cliff, Darwin defenders like Richard Dawkins could get their organisms up to the summit of Mount Improbable by envisioning a gradual staircase from another direction, allowing natural selection to maintain Darwin’s narrative of the accumulation of small, incremental steps.

But what if the Gaussian assumption is wrong? What if, instead, the structure of the landscape is like a block of Swiss cheese, flat and riddled with holes that a blind watchmaker cannot foresee? The probability of a “holey” fitness landscape becomes credible when considering dependent traits. These “quantitative traits” are made up of components that must cooperate to work. Without all of them emerging simultaneously, no organism can ascend to higher levels. With dependent traits in operation, such as in the case of powered flight, a mutation or defect in one can send the organism to immediate extinction — as if, as in the old board game, a trapdoor opened underneath it.

This Is Not a New Worry

Sergey Gavrilets thought of this back in 1997 and raised it again in 2004. Now, worries about a “holey landscape” have been given new emphasis in a paper in PNAS — “Drift on holey landscapes as a dominant evolutionary process.” The four authors, from universities in North Dakota, California, and Paris, complain that this worry has been largely neglected.

Our understanding of selection has been strongly shaped by Sewall Wright’s conceptualization of an evolutionary landscape, with populations moving from areas of low fitness to areas of higher fitness. While the one- and two-trait landscapes Wright originally described have been criticized as unrealistic, including by Wright himself, the general metaphor has nonetheless guided much of evolutionary thought. [Emphasis added.]

What if the metaphor has “guided much of evolutionary thought” astray? Then, the “understanding of selection” has been shaped awry.

An important conclusion from this research is that evolutionary dynamics on simple landscapes often fail to properly predict evolution on higher dimensional landscapes. Empirical research into quantitative traits has been slow to incorporate this need for a higher-dimensional perspective.

Perhaps most conceptually unfamiliar and unintuitive to researchers focused on quantitative traits are holey landscapes (Fig. 1C; ref. 16). Holey landscapes are high-dimensional evolutionary landscapes that consist of trait combinations that are either of average fitness or that are inviable. This results in flat landscapes with holes at inviable or low fitness phenotypes (Fig. 1C). 

Quantitative traits comprise “many aspects of physiology, behavior, and morphology,” they say, illustrating them with things like “most behaviors, physiological processes, and life-history traits.”

To investigate whether such traits tend to be distributed on a Gaussian landscape or a holey landscape, they looked at genetic variations in sixty species, including animals and plants. They found the results to be consistent with “high-dimensional, holey landscapes” instead of simplistic single-peak depictions or “badlands” landscapes consisting of rolling hills and gentle valleys. This suggested to them that

the leading conceptualizations and modeling of the evolution of trait integration fail to capture how phenotypes are shaped and that traits are integrated in a manner contrary to predictions of dominant evolutionary theory.Our results demonstrate that our understanding of how evolution has shaped phenotypes remains incompleteand these results provide a starting point for reassessing the relevance of existing evolutionary models.

Scrap and Start Over?

One way to reassess the relevance of an existing model is to scrap it and start over. The authors are not ready to try that, but they do point to serious shortcomings of conventional models: for instance, missing the holes.

Even more importantly, it is unknown what the topography of landscapes is for natural populations. While portions of selection surfaces and landscapes can be directly estimated, these estimates may differ from the full landscape due to several factors. These include the omission of fitness-affecting traits, incomplete estimation of fitness, and insufficient power to estimate non-linear selection coefficients.

In effect, modelers using Wright’s metaphor are building imaginary landscapes in thin air instead of working with real plants and animals on the real earth that must eat and survive. Simplistic Darwinian models of selection presuppose that beneficial mutations add up. This is not necessarily the case. A benefit to one gene can be a detriment to another — an example of negative pleiotropy. For realism, the whole animal must be considered, lest negative correlations open up a trapdoor that ends that organism’s progress, sending it down into one of those “inviable” outcomes. The authors call for better answers to the “crucial questions we have raised.”

Why would a holey landscape be flat? The authors explain what the model predicts for quantitative traits:

This topography stems from the multivariate nature of phenotypes: while there may be continuous fitness differences in two dimensions, fitness gradients will create holes in the landscape and peaks will average out when additional traits are considered. Unfortunately, predictions about quantitative trait evolution on holey landscapes are not clear and have rarely been pursued (e.g., ref. 18).

This means that traits are not isolated in one or two dimensions, but are interdependent on other traits in additional dimensions. This multi-dimensional consideration of traits on a flat landscape suggests that organisms are already at their optima; the fitness peaks have averaged out. The only way left is down, falling through a hole like a trapdoor if a trait changes that other traits depend on. 

Irreducible Complexity and Devolution

The picture fits Michael Behe’s concepts of irreducible complexity and devolution. No part of a mousetrap should be considered in isolation. It may have a great spring, but if the other parts are weak or absent, the trap will not catch mice. Neo-Darwinism’s focus on positive selection of individual genes or traits, therefore, misses the holistic multi-dimensional view of organisms as functional wholes, to borrow Douglas Axe’s phrase. To function, a multi-dimensional trait must reach a threshold of coherence among its parts. These can be considered a list of design requirements.

Can neo-Darwinism recover from the neglected view of holey landscapes? The authors do not offer any hope, other than to wish that better landscape models may be forthcoming. Even with that concession, they remain pessimistic.

Our implementation of Wright’s metaphor represents only one of many possible evolutionary models. It is possible that unmodeled alternative landscapes may produce populations for which variation is distributed in a manner similar to holey landscapes and empirical estimates…. Importantly, and as mentioned previously, much of the exploration of evolution of quantitative traits has focused on simple landscapes like we have implemented here. Thus, it also is an open question what different models of selection “look” like when implemented for higher-dimensional phenotypes. For example, rugged landscapes of high dimensionality may give rise to holey landscapes as peaks average out and valleys are inviable…. Nonetheless, the close correspondence between empirical data and populations simulated as evolving on holey landscapes suggest that our understanding of quantitative trait evolution remains incomplete.

Regardless of any other possible devices to rescue the picture of progress up fitness peaks, they assert that “our finding that observed patterns of quantitative genetic variation across taxonomic groups are not consistent with traditional evolutionary models stands.”

This disconnect between observed patterns of multivariate variation and expectations under conventional models of selection suggests that Wright’s metaphor of landscapes — and the subsequent implementation of this metaphor as Gaussian surfaces — may have contributed to an incomplete understanding of how selection has shaped phenotypes. A potential contributor to this problem has been the lack of clear alternative explanationsbesides a simple null hypothesis of drift with no selection. Moving forward, clear development of additional alternative models of the action of selection and evolution in multivariate space is needed. This will allow the comparison of simulated populations to empirical data as we have done here.

Ultimately, our findings suggest that evolutionary biologists need to better consider the effects of high dimensionality as simple standard evolutionary models are not consistent with available data for quantitative data.

Design Trumps Darwinism

Science is supposed to be about observable, quantitative data, isn’t it? If observations do not fit Wright’s convenient metaphor, the metaphor must be revised or discarded. Intelligent design is consistent with flat optima and built-in mechanisms to detect and avoid holes (e.g., DNA error correction, immune systems, blood clotting). Once again, design with its emphasis on engineering specifications trumps Darwinism in the real world.