Looking forward to next week’s 30th anniversary of the Center for Science and Culture, we’ll be reviewing the scientific evidence and how it has persistently supported intelligent design. But that’s not the only 30th anniversary this month. August 2 was also the 30th anniversary of the publication of Darwin’s Black Box. As such, it’s an ideal time to remember how the scientific evidence has repeatedly vindicated Michael Behe since then.
Irreducible Complexity and Molecular Machines
In Darwin’s Black Box, Michael Behe cited multiple examples of irreducible complexity, including the biochemistry of vision, the cilium, the bacterial flagellum, the blood clotting cascade, and aspects of the mechanisms that generate antibodies in the vertebrate adaptive immune system. This was a powerful, direct challenge to the view that Darwinian pathways existed — but the extraordinary integrated and purposeful organization of these systems also testified to their intelligent design.
Since Darwin’s Black Box, a variety of additional molecular machines and biochemical pathways have been cited as irreducibly complex, including the ribosome, ATP synthase and mechanisms involved in oxidative phosphorylation, the mechanisms of chemotaxis, and the transcription-translation system and machinery involved in eukaryotic cell division. Various macrobiological features have also been cited as irreducibly complex, including animal joints and the trap mechanisms of carnivorous plants (pictured at the top). In fact, the cell itself has been called irreducibly complex by an authority no less than Eugene Koonin, who wrote in 2014:
Given the intricacy of the molecular machinery of even the simplest cells, the lack of any observations on cell origin, except from other cells, and the lack of any apparent intermediate stages in the evolution of cells, the origin of cells engenders the ominous specter of irreducible complexity.
There are good reasons that Koonin felt the cell is difficult to explain in evolutionary terms. Other research has shown that the first, or at least the most fundamental forms of cellular life would have required large genomes and many genes. A 2016 study in Science found that simplest-known free-living form of life is a bacterium that requires a genome of over 500,000 base pairs long, encoding 438 protein-coding genes and 35 RNA-coding genes. In 2024, a study in Nature Ecology & Evolution found that the hypothetical “last universal common ancestor” or LUCA — not the first life, but a life-form ancestral to all organisms alive today, which is said to have lived very early in the history of life — would have required a genome composed of around 2.5 million base pairs, encoding around 2,600 proteins. The origin of anything resembling what we know as cellular life would have therefore required a massive amount of proteins and genomic information — evidence not just of irreducible complexity, but also that an intelligent agent was at work.
Koonin noted the lack of intermediate stages for the first cell, and he’s not the only one to acknowledge the problem posed by irreducible complexity. In the early 2000s, biochemist Franklin Harold recognized in an Oxford University Press book that, “There are presently no detailed Darwinian accounts of the evolution of any biochemical or cellular system, only a variety of wishful speculations.” In 2014, a paper in PNAS noted the lack of progress towards explaining the evolution of such cellular features: “Despite the surge of interest at the molecular, genomic, and developmental levels, much of today’s study of evolution is only moderately concerned with cellular features.” Over a decade later, some two-plus decades after Harold, and thirty years after Behe’s book, we can still ask: are there detailed, stepwise explanations for the evolutionary origin of the molecular machines and biochemical pathways that Behe cited as irreducibly complex? Even today, if you search the literature, you find much handwaving but certainly no detailed evolutionary pathways.
To appreciate this lack of detailed explanations, it’s worth focusing on what has the become most famous example of irreducible complexity, the bacterial flagellum. In 2009, a Japanese biologist observed that “the flagellum is so well designed and beautifully constructed by an ordered assembly pathway, even I, who am not a creationist, get an awe-inspiring feeling from its ‘divine’ beauty.” But where do things stand thirty years after Behe’s book? In 2023, my colleagues and I evaluated a recent academic treatment of the evolution of the bacterial flagellum which had claimed evolutionary explanations of the flagellum are forthcoming since 90 percent of flagellar parts have homologues that perform functional roles outside the flagellum. Our review showed it was far from explaining how the flagellum evolved.
One of the main sources for the claim, Pallen and Matzke (2006), found that some 20 proteins appear “indispensable” for flagellar function, and 15 of the 42 flagellar proteins studied did not have known homologues. Of the remaining 27 proteins for which they reported homology, we found that only 4 were actually homologous to parts in other systems which could have potentially served as “precursors” to flagellar components.
Eleven of those homologues were similar to proteins in the type three secretory system (T3SS), a different molecular machine which is often called a potential evolutionary precursor to the flagellum. However, the evidence strongly points against the T3SS being a precursor to the flagellum.
The T3SS is a molecular pump which serves as part of the basal body that embeds the flagellum in the cell membrane. It is involved in flagellar assembly, pumping proteins from inside the cell to outside the cell where they self-assemble into the flagellum. Yet the T3SS’s pump function is also used by certain predatory bacteria to inject toxins into eukaryotic cells to kill them. But it is doubtful that the injectisome and its T3SS are useful in explaining the origin of the flagellum.
First, historical, chronological, and phylogenetic considerations strongly imply the flagellum long predates the injectisome (and its T3SS) and thus could not have evolved from it. To be specific, the injectisome is found in a small subset of gram-negative bacteria that have a symbiotic or parasitic association with eukaryotes. Since eukaryotes evolved over a billion years after bacteria, this suggests that the injectisome arose after eukaryotes, comparatively later in the history of life. However, flagella are found across a wide range of bacteria, and the need for chemotaxis and motility (i.e., using the flagellum to find food) is thought to have arisen very early — perhaps being present as early as the last bacterial common ancestor. Most certainly, the need for chemotaxis and motility preceded the need for parasitism, which means we would expect that the flagellum long predates the injectisome. This evidence suggests that the injectisome is a very late innovation in bacterial history, whereas the flagellum arose early. Indeed, a recent paper uploaded to bioRxiv suggests the flagellum may have been present in the last bacterial common ancestor, thus placing it extremely early in the history of life. The injectisome could not have predated the flagellum. (For additional documentation, see Jones 2008; Abby and Rocha 2012a, 2012b; Deng et al. 2017; Coleman et al. 2021.)
Second, even if the T3SS could have predated the flagellum, it could not provide anything close to a viable evolutionary precursor due to the great leap in complexity required to go from a T3SS to a flagellum. William Dembski compared it to thinking “we can travel by foot from Los Angeles to Tokyo because we’ve discovered the Hawaiian Islands.” Or, as a 2020 paper in Journal of Theoretical Biology stated:
Irreducible complexity does not mean that irreducibly complex systems are logically impossible to evolve based on existing modules. One cannot definitively rule out the possibility of an indirect, circuitous route. A well-known subsystem of the bacterial flagella (called T[3]SS secretion system) performs a function distinct from the flagellum. However, finding a subsystem of a functional system that performs some other function is hardly an argument for the original system evolving from that other system. As the complexity of an interacting system increases, the likelihood of such an indirect route drops quickly.
Another study found that the T3SS and flagellum are so distinct that they may in fact have independent origins. The bottom line is that the common evolutionary view that the injectisome and its T3SS served as some kind of an evolutionary precursor to the flagellum isn’t supported by the evidence.
Pallen and Matzke (2006) acknowledged that “the flagellar research community has scarcely begun to consider how these systems have evolved.” Twenty years after that paper, and thirty after Behe’s book, no one has still even attempted to provide anything close to a stepwise evolutionary explanation of the bacterial flagellum. Indeed, the aforementioned 2020 paper in Journal of Theoretical Biology affirmed Franklin Harold’s statement about the lack of evolutionary explanations for irreducibly complex features:
In ‘‘Darwin’s Black Box” (Behe, 1996), Behe exemplified systems, like the flagellum bacteria use to swim and the blood-clotting cascade, that he called irreducibly complex, configured as a remarkable teamwork of several (often dozen or more) interacting proteins. Is it possible on an incremental model that such a system could evolve for something that does not yet exist? Many biological systems do not appear to have a functional viable predecessor from which they could have evolved stepwise, and the occurrence in one leap by chance is extremely small. … Such systems pose a serious challenge to a Darwinian account of evolution, since irreducibly complex systems have no direct series of selectable intermediates…Darwinian explanations of irreducibly complex systems are improbable. Ultimately, this is a question that must be studied both experimentally and by computer simulations. Behe’s concept of irreducible complexity has not been falsified by computer models (Ewert, 2014), and there are presently no detailed Darwinian accounts of the evolution of any such biochemical or cellular system, ‘‘only a variety of wishful speculations” (Harold, 2003, p. 205).
It seems that little has changed in 30 years since Behe’s book: despite the handwaving, evolutionary biologists still lack detailed, stepwise evolutionary explanations for how irreducibly complex features arose.
Protein Origins: A Challenge to Co-Option
After Behe proposed irreducible complexity, the main rejoinder from critics was that irreducibly complex systems evolve via co-option, where parts are borrowed from elsewhere in biology, retooled, and then incorporated into new systems. The co-option hypothesis requires that new proteins be easy to evolve. However, evidence over the past 30 year has shown the opposite — new proteins, especially new protein folds, are very difficult to evolve.
ID has contributed hugely to this debate. In 2004, pro-ID protein scientist Douglas Axe published research showing that the likelihood of a random sequence of amino acids yielding a functional beta-lactamase enzyme’s domain (that was about 150 amino acids long) is less than 1 in 10^77. This suggests that random mutation and natural selection would have great difficulty finding sequences that yield new stable, functional protein folds. A rough calculation shows just how difficult this would be. Approximately 10^40 individual organisms have lived over the entire history of the Earth (see Behe, 2007, p. 64; Whitman et al., 1998). If we were to grant the incredibly generous assumption that every single organism that has ever lived was somehow magically gifted with a brand-new sequence of about 150 amino acids, randomly arranged such that a functional gene might arise, then we would still be short by about 10^37 trials to generate a single functional protein similar to the one Axe investigated.
Axe’s results about the rarity of functional proteins are not outliers, and the view that functional protein sequences are rare is now commonplace. As a 2009 book from Springer noted:
[O]nly a small fraction of the possible combinations of amino acids will fold spontaneously into a stable structure. If you make a protein with a random sequence of amino acids, chances are that it will only form a gooey tangle when placed in water.
A 2017 paper by two non-ID scientists in Biophysical Journal reviewed the results of additional protein-studies and reported that the rarities of ten functional protein domains range from 1 in 10^24 to 1 in 10^126 sequences. In six of the ten cases, the probabilities are below our extremely generous thought experiment which would require at least 10^40 trials to generate a single functional protein sequence. This confirms that many proteins are beyond the reach of natural selection and random mutation. For that reason, a 2013 article in ASBMB Today quoted the late Israeli protein scientist Dan Tawfik saying that the origin of major families of proteins appears to be “something like close to a miracle.”
Other research directly from the ID camp has confirmed the difficulty of evolving new proteins. Gauger and Axe (2011) found that merely converting a particular metabolic enzyme to perform the function of a closely related enzyme — the kind of conversion that evolutionists claim can readily happen — would require a minimum of seven mutations. Yet this exceeds the limits of what Axe (2010) calculated that Darwinian evolution can produce over the Earth’s entire 4.5-billion-year history. A follow-up study by Gauger, Axe, and Reeves bolstered this finding by attempting to mutate additional enzymes to perform the function of a closely related protein (Reeves et al., 2014). After inducing all possible single mutations in the enzymes, and many other combinations of mutations, they found that evolving a protein to perform the function of a closely related protein would take over 10^15 years — over 100,000 times longer than the age of the Earth.
The aforementioned ID research pertained to prokaryotes, but such problems get even worse for multicellular organisms. In 2004, Michael Behe and David Snoke published a study in Protein Science which found that in multicellular organisms, a protein-protein interaction requiring more than two specific amino acids to be present would be unlikely to evolve under reasonable population sizes and generation times.
Collectively, this research indicates that the evolutionary process simply does not have the probabilistic resources (in terms of generations, population sizes, or time) to “search” and find the necessary mutations to produce the rare amino acid sequences that yield functional proteins.
The Edge of Evolution: Tired Yet of Being Vindicated?
In 2007, Behe published his second book, The Edge of Evolution: The Search for the Limits of Darwinism, arguing that some traits are beyond the “edge” of what random mutation and natural selection can produce. Behe focused on malarial resistance to the drug chloroquine, which has such a low degree of rarity that he predicted it probably required multiple specific mutations in order to provide advantageous resistance. Behe’s critics didn’t like this argument. In multiple reviews, they claimed he misunderstood how evolution worked, which would proceed in a stepwise manner where each successive mutation yielded some advantage. However, Behe’s prediction was proven correct. In 2014, a paper by Summers et al. in PNAS vindicated Behe by finding that chloroquine resistance indeed required multiple mutations before conferring any advantage. (Behe discussed this here, and I discussed this here or here.) This didn’t mean the trait could not evolve — the large number of malarial parasite cells allow it to evolve, though it is rare. But it bolstered Behe’s basic intuition that some traits are near the “edge of evolution” because they require multiple coordinated mutations in order to confer some advantage.
Then, in 2008, two Cornell mathematical biologists sought to refute Behe’s arguments that some traits would require impossibly long periods of time to evolve. Writing in the journal Genetics, they quibbled with his math but found that in a population of organisms like humans, producing a feature via Darwinian evolution that required only two mutations before providing an advantage “would take > 100 million years,” which they acknowledged would be “very unlikely to occur on a reasonable timescale” (Durrett and Schmidt, 2008). This was a powerful vindication of Behe’s fundamental argument that some traits are indeed beyond the “edge of evolution,” validating fundamental ID arguments that there is too much information in complex biological features to be produced by random mutation and unguided natural selection.
Darwin Devolves: Incoming! More Vindication!
In 2010, Behe published a paper in Quarterly Review of Biology reviewing what happens when Darwinian evolution operates at the molecular level. He looked at many studies of molecular adaptation and found that Darwinian evolution tends to “break or blunt” functional components — breaking or diminishing function at the molecular level far faster than it builds new functional features. Here’s how he put it:
The rate of appearance of an adaptive mutation that would arise from the diminishment or elimination of the activity of a protein is expected to be 100-1000 times the rate of appearance of an adaptive mutation that requires specific changes to a gene.
This makes sense: there are far more ways to break a system then there are to improve it. If Darwinian evolution follows the path of least resistance, it’s more likely to improve efficiency by effectively throwing away dead weight than by producing some new complex feature that would require multiple mutations — and immense amounts of time — to evolve.
ID research backed up his thesis. A paper published by a team of ID scientists that same year reported results from experiments which broke a gene in E. coli required for synthesizing the amino acid tryptophan. When the bacteria’s genome was broken in just one place, random mutations could “fix” the gene. But even when only two mutations were required to restore function, Darwinian evolution got stuck, unable to restore full function (Gauger et al., 2010). Instead, the non-functional gene would often simply be deleted — again, showing that Darwinian evolution finds it much easier to discard stuff than to convert it into something usable. Another project by pro-ID biologists Dusty van Hofwegen and Scott Minnich tested a widely touted bacterial innovation of Richard Lenski’s “Long-Term Evolution Experiment,” and found it actually involved the evolution of “no new genetic information (novel gene function)” (Van Hofwegen et al., 2016). And a theoretical simulation developed by Douglas Axe, called Stylus, modeled the long-term evolution of a population of digital organisms, and found that over time they experienced “genome decay” (Axe et al., 2008; Axe et al., 2011; Axe et al., 2023). Other ID research showed that many computer simulations purporting to show how “evolutionary algorithms” could produce new information effectively cheated, and the information was always smuggled in by the programmer (see citations in footnote 11 of Luskin and Bloom, 2025). Together, these results suggest that left to their own devices, evolutionary mechanisms don’t build new features — they tend to break or destroy useful genetic information and the functions they encode. Something else must be inputting information to allow species to persist and evolve, and that something is intelligence.
Encouraged by such results, in 2019 Behe wrote another book, Darwin Devolves, where he brings to bear numerous examples of Darwinian evolution operating by breaking or diminishing molecular function (rather than building new functions). The word “bear” is an intentional pun, because one of his biggest examples in the book was polar bear genes which he argued have been broken to allow them to eat a high-cholesterol diet of seal blubber without dying of a heart attack. Behe’s critics claimed that the polar bar gene in question, called APOB, had been improved by natural selection, allowing it to better remove cholesterol from the bloodstream. Once again, though, the critics got it wrong. APOB’s purpose isn’t to remove cholesterol to the bloodstream but rather to load it into the bloodstream. And while no one has done experiments on polar bears to understand exactly how their version of APOB works (very difficult experiments to perform!), it turns out that some humans have diseases where APOB becomes broken via mutations, leading to abnormally low cholesterol (see Luskin, 2019). This proves Behe’s thesis is viable: polar bears managed to break this gene in a manner that prevented its loading as much cholesterol into the bloodstream.
Other mainstream biologists have affirmed Behe’s basic thesis. In 2020, the Harvard geneticist Andrew Murray wrote in Current Biology:
In laboratory-based experimental evolution of novel phenotypes and the human domestication of crops, the majority of the mutations that lead to adaptation are loss-of-function mutations that impair or eliminate the function of genes rather than gain-of-function mutations that increase or qualitatively alter the function of proteins. Here, I speculate that easier access to loss-of-function mutations has led them to play a major role in the adaptive radiations that occur when populations have access to many unoccupied ecological niches.
The following year, a paper in Nature Heredity found that “Discoveries during the subsequent two decades have continued to support the idea that loss of function contributes to adaptation, with cases of adaptive or beneficial loss of function being discovered across diverse organisms, genes, traits, and environments.” Yet again, Behe was shown to be right: Darwinian evolution doesn’t tend to build new features, it tends to break them down.
If you are sensing a trend, you’re onto something: Over the last 30 years, time and again Behe has used his intelligent design paradigm to make many good predictions. This ability to make predictions shows, in and of itself, the power of intelligent design to guide scientific research. Behe’s predictions were often mocked and bitterly opposed by critics (who continue to badly misrepresent him — see for example here), only to see him later proven right. That shows that intelligent design is not just science, it’s very good science. Critics can say what they like about Behe, and they can even try to misrepresent him. But in the end, the evidence wins out.