Introduction

What can be more curious than that the hand of a man, formed for grasping, that of a mole for digging, the leg of the horse, the paddle of the porpoise, and the wing of the bat, should all be constructed on the same pattern, and should include the same bones, in the same relative positions? (Darwin 1859, p. 434)

Darwin’s answer, and ours today, is that these curious anatomical correspondences are instances of homology—the same characters in organisms of different species. Most would say that homologies are simply similarities of character that can be traced to common ancestry. But the concept of homology predates the theory of evolution and, perhaps because of this, although it is central to biology, its material definition and determination remain controversial (Hall 1994; Wake 1994, 1999; Wagner 2014). Molecular phylogeneticists sometimes talk as if their discipline has resolved these controversies: it has not.

Homology Before Molecular Phylogenetics

Homology as Archetype

Although correspondences between characters of different species were noted by Aristotle, and frequently cited throughout the early modern period, the most explicit, precise early formulation of homology was given by nineteenth-century morphologist Richard Owen (Belon 1551; Owen 1843; Panchen 1994, 1999). Drawing on a long German Naturphilosophie tradition—in particular, the work of Johann Wolfgang von Goethe, Carl Gustav Carus, and Lorenz Oken—and the work of French anatomist Etienne Geoffroy Saint-Hilaire, Owen argued that the way to make sense of the limb structures referred to in our epigraph, for example, was to think of them as instances of a more general theme or, what Goethe called an “archetype”: in this case, the vertebrate limb (Owen 1848, 1849; Richards 1992). According to Owen, themes like this should be thought of as God’s plan for the production of different vertebrates. God intended that variants of these themes would be differently adapted to different environmental circumstances—the mole’s forearm for digging, the porpoise’s for swimming (Beatty 2006). So, according to Owen, homologous structures were those based on the same theme. And we explain the forms of organisms we find in nature through the investigation of these common themes. They are powerful explanatory devices.

Owen defined homologs as “The same organ in different animals under every variety of form and function” (Owen 1843, p. 379). And he contrasted this with an “analog”—“A part or organ in one animal which has the same function as another part or organ in a different animal” (Ibid., emphasis ours). While the human hand and porpoise flipper are homologs, the wing of a bee and the wing of a bat are analogs. Both function in flight, but there is no common pattern in underlying structural components. It was the job of the comparative anatomist to distinguish between structures that were instances of an archetype, or homologous, and structures that were merely analogous, a more superficial form of similarity.

In later work, Owen (1848) described three kinds of homology: “special homology” (the relation between the same part or organ in different animals), “general homology” (the relation between a part or organ of an animal and its archetype), and “serial homology” (the relation between two structures that are equivalent in the same organism, for example, vertebrae in vertebrates).

Darwinian Homology

A major shift in the meaning of “homology” came with the publication of Charles Darwin’s Origin of Species (1859). Giving an evolutionary spin to Owen’s archetype, Darwin writes, “If we suppose that the ancient progenitor, the archetype as it may be called, of all mammals, had its limbs constructed on the existing general pattern […] we can at once perceive the plain significance of the homologous construction of the limbs throughout the whole class” (Darwin 1859, p. 435). In other words, we account for homologies—the reason why moles and humans share the same bones in their forearms—not by referencing that they are instances of a general vertebrate theme, but by showing that they share a common ancestor which also contained the same bones in its forearm. We no longer need mysterious archetypes.

For the Darwinian biologist, it was thus no longer the job of the morphologist to distinguish between similar traits that were instances of a common archetype and those that were merely analogous, but instead to distinguish between those that could be traced to a common ancestor and those that arose independently. Comparative morphology and the finding of homologies thus became phylogenetic research. Homologies, understood as evidence of common ancestry, could be used to construct genealogies. As Darwin writes, “classifications will come to be, as far as they can be so made, genealogies” (Darwin 1859, p. 486).

Although common ancestry initially provided an explanation for homology, it soon became part of the definition: the same structure in two species that can be traced to common ancestry (Donoghue 1992). In a sense, this merged two meanings of “related,” of the same type or of the same genealogical family.

Despite its simplicity, this definition caused considerable conceptual confusion. What did it mean to find the “same” structure in two species and in their common ancestor? As zoologist Ray Lankester lamented in 1870, there were problematic cases, such as structures in different species which are not very similar, but that can be traced to a common ancestor (e.g., the bones of the middle ears of mammals and of the lower jaws of reptiles), or similar structures in different species that can be traced to a common ancestor, but to one that had a dissimilar structure (e.g., the four cavities of the bird heart and the mammalian heart can be traced to a common ancestor having a heart with only three cavities) (Lankester 1870). Attempting to avoid ambiguity, Lankester introduced a new concept, “homogeny,” to replace homology. It would apply only when the same structure in different species could be traced to a common ancestor with that same structure, thus excluding cases like the mammalian and bird heart.

On Lankester’s account, common ancestry was not sufficient for homogeny, and one must also make a judgement about structural similarity, between the species compared and their shared common ancestor. This leads to a tension in his analysis, and one which recurs throughout the subsequent history of homology. Two structures must be either homologous or not, Lankester holds, because two structures are either related by common ancestry or they are not. And yet this decision is in part based on a further decision about what counts as a meaningful similarity and, we would add, on how we define or delimit comparable structures to begin with. Lankester slips into such tensions when he writes that the bones of the middle ears of mammals and the bones of the lower jaws of reptiles represent “a much less detailed kind” of homogeny than that which holds between their forelimbs (Ibid., p. 37). But, we are left wondering, are they homogens or not, or does the concept come in degrees after all?

Lankester’s discussion was influential, but his concept was not, and ironically, homology simply acquired the meaning of homogeny. In the years after his intervention, the concept of homology broke up and, as Otto Hass and G. G. Simpson put it in 1946, underwent a “considerable amount of amplification on the one hand and subdivision on the other” (1946, p. 320). Amplification because it was commonly understood to be a concept of principal importance in biology, and subdivision in the sense that the concept was defined in many, heterogeneous ways. While its early history was tied to morphology, homology broadened as it worked its way throughout an early twentieth century biology that was fracturing into distinct sub-disciplines (Tait 1928; Haas and Simpson 1946; Brigandt 2003; Kleisner 2007). And in each sub-discipline, it continued to be a topic of discussion and controversy (Hubbs 1944; Haas and Simpson 1946; Boyden 1947).

Regardless of sub-discipline, two questions continued to plague discussions of homology: first homologs may be structurally dissimilar, and nonhomologs may be similar. But if similarity and common ancestry are not coextensive, which matters more for homology? Some went with common ancestry, while others chose morphological similarity (Hertwig 1906; Boyden 1947; Hubbs 1944; Donoghue 1992). Second, when can similarity be trusted as an indication of homology, rather than merely analogy (similar form from dissimilar ancestry)? Both questions continue to be controversial, as we will see below.

Homology and Development

One way to answer the second question was pioneered by a post-Darwin, initially German, school of comparative anatomy, known as evolutionary morphology (Laubichler 2000; Wagner 1989a; Di Gregorio 1995; Amundson 2005). These biologists used homologs as tools to derive phylogenetic relationships between organisms. But deriving common ancestry from morphological similarity was uncertain without independent criteria for determining if a similarity was trustworthy. Ernst Haeckel and Karl Gegenbaur found such criteria in embryology and ontogeny (Haeckel 1866; Gegenbaur 1878). According to Haeckel’s “biogenetic law,” ontogeny recapitulates phylogeny, and thus if two organisms exhibit similar structures in the adult stage, one way to determine whether such structures are homologous is to look at their ontogenies, since these also reflect their phylogenetic evolution. Haeckel thus stated that a “true homology can only exist between two parts which have arisen from the same primitive ‘Anlage’ [rudimentary basis of an organ, etc. in an embryo] and have deviated from one another by differentiation only after the lapse of time” (quoted in Tait 1928, p. 157).

Evolutionary morphology assumed that embryological data would be a reliable indicator of common descent and thus would be a way to determine homologs, but this was not so. Development turned out to be a lot more flexible than Haeckel, or anyone, realized (Wagner 1989b; Laubichler 2000; Brigandt 2006). In 1915, the developmental biologist Hans Spemann published a critique of Haeckel and Gegenbaur’s evolutionary morphology (Spemann 1915). Haeckel had assumed that homologous adult structures have the same precursor or origin—the “Anlage”—in development. In fact, that was the definition of a homolog. But this assumption was false. Regenerated organs, which Spemann experimentally demonstrated using the amphibian lens, developed in a different manner and emerged from different sources than their normal counterparts. Yet the regenerated and normal lenses were clearly homologs (the “same organ,” present in a common ancestor), contra Haeckel. Spemann also showed that adult structures do not correspond to structures in the egg, leaving the relationship between adult structure and Anlage mysterious.

While Spemann’s work specifically undermined the usefulness of the Anlage and thus Haeckel’s notion of “true homology,” further embryological research only reinforced his contention that homology was puzzling from a developmental perspective (de Beer 1971; Amundson 2005). Sir Gavin de Beer famously integrated this research into one critical review: homologous characters “cannot be pressed back to similarity of position of the cells of the embryo or the parts of the egg out of which these structures are ultimately differentiated,” nor do they always owe their origin and stimulus to similar developmental processes (1971, p. 13). And looking to the gene-level did not avoid such problems either: characters controlled by identical genes may not be homologous, and homologous structures (morphologically or developmentally similar to each other and to their inferred common ancestor) need not be controlled by identical genes (Wray 1999). The reading of homology from similar adult morphology reflected assumptions about development that were not empirically valid, and so in 1971, de Beer concluded that homology remained an “unsolved problem.”

Influenced by Spemann’s and de Beer’s critiques, a number of developmental biologists in the 1980s began to pursue a causal-mechanical, as opposed to a historical, notion of homology (Roth 1984; Wagner 1989a, b, 2014). That is, rather than aiming to specify when similarity can be a trustworthy indicator of common descent, these research programs have focused on elucidating the causal developmental processes that give rise to homologs and on providing a more inclusive notion of homology that would subsume the many noted types (such as Owen’s “serial homology,” which fit poorly with many post-Darwinian notions of homology as a comparison between different species). They call this view by the more general name, the “biological homology concept.” On this view, homologies reflect not just shared common ancestry, but shared developmental processes: they reflect aspects of development that are less malleable, or what are called developmental constraints.

Morphological and Molecular Homology

Contemporaneous with its discussion in developmental biology, homology continued to be central to comparative anatomy. Following Darwin, homology was seen not just as evidence of evolution, but also as the way to build phylogenetic relationships. But the same problem arose: when could similarities be trusted as indicating homologs? Rather than go Haeckel’s route, comparative anatomists developed similarity criteria: criteria that would help biologists distinguish trustworthy similarity from untrustworthy. Adolf Remane (1952) famously provided three such criteria based on classic morphological principles: a positional criterion (homologs should be similar in relative positions), a structural criterion (homologs should share structural detail), and a transitional criterion (phylogenetically or ontogenetically intermediate forms should be identified). Although homology was originally an all-or-nothing affair, Remane’s criteria led to the development of views of “partial homology”: similar structures can meet such multidimensional criteria to various degrees, and therefore there is a sense in which they can be “partially homologous” (Donoghue 1992).

In the 1950s, Willi Hennig, in the process of developing phylogenetic systematics (cladistics), began to use homology in a specific, theory-laden way (Hennig 1966). Within his theory, Hennig defined homologies as features that originated at the stem of a species group, as shared derived characters that characterize monophyletic groups, or what he called “synapomorphies.” But just as they had in the past, homologies still needed to be assessed by independent criteria, and for this, Hennig drew on Remane’s similarity criteria above. In other words, homology meant synapomorphy, but was discoverable only through the study of morphological similarities. In the 1980s, Colin Patterson further developed Hennig’s notion of homology by providing a new set of homology criteria, what he called homology “tests” (1982, 1988). These included Remane’s original criteria as well as other, more powerful criteria, what he called congruence and conjunction. These provided a more secure base and a theoretically-expanded notion of homology.

As Hennig and Patterson were developing homology within cladistics, the rise of protein sequence data and the development of molecular biology made possible molecular phylogenetics. Molecular phylogeneticists inherited the concept of homology from their morphological counterparts, and as such relied on similarity as well, just of a molecular rather than morphological type (Patterson 1987). At the time (early 1960s), there was a sense among many that a molecular-based homology concept would be less problematic, or at least cleaner, than a morphological one, as amply illustrated by the various contributors to Bryson and Vogel’s (1965) epochal Evolving Genes and Proteins. And this was directly linked to the problem of trustworthy similarity: it was thought that the ability “to quantify the likelihood that sharing a certain level of similarity reflects common ancestry” would be much more secure from molecular data (Donoghue and Sanderson 1994). For example, Stephen Jay Gould (1985, p. 14) wrote that…

The decoding of phylogeny requires no more than a method for recognizing homology and eliminating analogy. […] Morphology may be our first (and usually adequate) guide, but it just doesn’t provide enough independent detail for sorting homology from analogy in difficult cases. […] DNA is complex enough to preclude analogy as a cause of overall similarity […thus,] we finally have a method that can sort homology from analogy.

This caused a heated debate between morphologists and molecular biologists over the legitimacy of their respective uses of homology, as signaled by the title of such classics as Molecules and Morphology in Evolution: Conflict or Compromise? (Patterson 1987).

Homology After Molecular Phylogenetics

In the Bryson and Vogel collection, Emile Zuckerkandl, founding editor of this journal, and his postdoctoral mentor Linus Pauling argued strongly in favor of molecular data, asserting that “the largest concentration of information present in an organism, and perhaps also the largest amount of information, and the only organically transmissible information, is in its semantides [DNA, mRNA and encoded proteins]”. (Zuckerkandl and Pauling 1965, p. 98). Such a view embraced molecular biology’s nascent central dogma (Crick 1970), the reductionist ontology of many of the evolutionary Modern Synthesis’ adherents (Williams 1966), and the brute facts of DNA replication. For DNA, there is little ambiguity, in principle, in conceptualizing ancestor–descendant relationships. It was thus in semantides that evidence of common descent could be most surely found and homology best exemplified, Zuckerkandl and Pauling reckoned. And they understood “homologous” to mean only “descended from a common ancestor.” Semantide sequence similarity logically comes in degrees, but homology—a hypothesis about ancestry for which substantial similarity may be taken as evidence—does not. In principle, homologous genes remain homologous even when diverged beyond any significant similarity (Patterson 1988).

Recasting molecular homology as nothing more than common descent has problems not so different than those encountered in morphology or development, as detailed below. Nevertheless, there is a general understanding that conflating the extent of similarity and the inference of common descent is always a misuse of the concept—that there can be no such thing as “percent homology.” Hillis (1994) interprets this common misuse cynically, writing that “perhaps the term homology is thought to make the work sound more like science than would use of the simple and obvious word similarity.” A prescriptive admonition by Zuckerkandl and ten other molecular evolutionary luminaries in the late 1980s affirming that homology “should mean ‘possessing a common evolutionary origin’ and in the vast majority of reports should have no other meaning (Reeck et al. 1987, p. 667, emphasis ours)” has apparently had little effect. “Percent homology” appears in more than 7000 citations on Google Scholar since 2000!

This issue aside, homology based on genes does, at first blush, seem a less onerous or ambiguous concept than the notion of “sameness” of traits at higher levels. Genes, or at least parts of genes, or at least individual nucleotide positions, unambiguously have common ancestors (“positional homology” [Hillis 1994]). There should be no further reduction to be entertained. But both practical and conceptual concerns about gene homology remain. Here we highlight five such concerns.

Convergence

Gene sequence similarity falls off faster than protein sequence similarity (because of the redundant code), while protein sequence similarity decays faster than protein structure (Dokholyan and Shakhnovich 2001; Orengo and Thornton 2005) as genes diverge in evolution. Similarities observed at this highest level are often taken as evidence for homology, especially with proteins of similar function. But convergence driven by functional constraints is a tenable alternative explanation, and controversy is unavoidable (Doolittle 1994; Pearson and Sierk 2005).

The photochemically reactive rhodopsins are a case in point. Both microbial (Type I) and eukaryotic (Type II) rhodopsins are G-protein-coupled receptors, sharing a seven transmembrane alpha-helical “GPCR-fold” and a Schiff base linking a conserved lysine to retinal. But a countervailing list of structural and functional dis-similarities has been widely taken as evidence that rhodopsins are nevertheless merely analogous, exemplifying convergence due to similar selection pressure under similar biophysical constraints. Mackin et al. (2013) subjected this notion to test, engineering rhodopsins with a variety of radically different structures. Many are still photochemically reactive, motivating the surprising conclusion that “because the naturally occurring rhodopsin fold is unnecessary for functional competence, it is unlikely that selection would lead unrelated proteins to this particular architecture independently” (Mackin et al. 2013, p. 90)—in other words, Type I and II rhodopsins may well be homologs after all!

Such painstaking experimental investigations of structurally and functionally similar proteins without significant primary sequence similarity are rare, and cases where similar sequences or structures have clearly arisen independently in nonhomologous genes may be limited to shorter polypeptides (Graham et al. 2013; Shafee et al. 2016). More common would be situations in which unquestionably homologous genes in different lineages have become more, rather than less, similar in amino acid sequence because they have been subjected to parallel but independent selection for similar function. For example, ruminant physiology has arisen independently several times, and Irwin writes that “Recruitment of lysozyme c as a digestive enzyme has occurred at least twice within mammals, on the lineages leading to the ruminant artiodactyls and the leaf-eating monkeys, with a similar recruitment of a calcium-binding lysozyme occurring in the hoatzin, a leaf-eating bird” (Irwin 2015, p. 1). At many residues, the same amino acid substitutions have occurred: the proteins are more similar than we might predict given the antiquity of their common ancestors—convergence (analogy) is superimposed on a deeper homology.

Orthologs, Paralogs, and Xenologs

Homologous genes present always in single copy in ancestor, and descendants are the most straightforwardly informative semantides, but genome evolution is not so simple. In 1970, Walter Fitch stipulated that …

… there should be two subclasses of homology. Where the homology is the result of gene duplication so that both copies have descended side by side during the history of an organism (for example, alpha and beta hemoglobin) the genes should be called paralogous (para = in parallel). Where the homology is the result of speciation so that the history of the gene reflects the history of the species (for example alpha hemoglobin in man and mouse) the genes should be called orthologous (ortho = exact) (Fitch 1970, p. 113).

As Fitch and Margoliash (1967) pointed out even before these terms were coined, a vertebrate phylogeny based on an unrecognized mixture of alpha and beta hemoglobin sequences would most often be wrong as a tree of species, and histories of dynamic gene families in which paralogs quickly come and go might be impossible to reconstruct. If all that ‘homology’ necessarily implies is common ancestry, gene product function is no sure guide, and the frequently employed “functional orthology” is oxymoronic when applied to genes. Simple models of gene duplication as first conceived by Ohno (one duplicate retaining the initial function and the other free to evolve more rapidly by drift or selection) may often be invalid and the distinction can be very difficult to make (Ohno 1970; Davis and Petrov 2004; Gabaldon and Koonin 2013). So ortholog/paralog language is often carelessly or mis-applied and Petsko (2001) for one has opined against it, claiming that it replaces “easily understandable simple phrases”.

However, we cannot imagine what these simple phrases might be! Moreover, many genes in many genomes owe their presence neither to “vertical descent” through replication nor to duplication but to lateral (horizontal) gene transfer (LGT) from another species or another subpopulation of the same species. Gray and Fitch (1983) first suggested calling such genes xenologs, a term that has stuck. So, in any single genome, two copies of “the same” gene could be paralogs, produced by a duplication within that genomic lineage, or xenologs, one derived by LGT from a different lineage (and thus an ortholog of the other before its transfer). Distinguishing these alternatives requires independent evidence (from phylogenetics or base compositional characteristics of the presumed xenologous genes), and this will be harder to come by in the most likely sorts of LGT, between close relatives. Most interestingly, a recent survey of closely related prokaryotic genomes, by Treangen and Rocha (2011, e1001284), concluded that …

After removing the effects of transposable elements and phages, we show that the vast majority of expansions of protein families are due to transfer, even among large genomes. Transferred genes—xenologs—persist longer in prokaryotic lineages possibly due to a higher/longer adaptive role. On the other hand, duplicated genes—paralogs—are expressed more, and, when persistent, they evolve slower. This suggests that gene transfer and gene duplication have very different roles in shaping the evolution of biological systems: transfer allows the acquisition of new functions and duplication leads to higher gene dosage. Accordingly, we show that paralogs share most protein–protein interactions and genetic regulators, whereas xenologs share very few of them.

Gene Mosaicism

Detecting convergence and distinguishing orthologs, paralogs, and xenologs (and other sub-categories such as “synologs” and “gametologs” defined by Mindell and Meyer 2001) may be largely an empirical matter. Gene mosaicism poses thornier, ontological, problems. Many longer genes, especially in eukaryotes, have been cobbled together from domain-encoding parts of quite different pre-existing genes, challenging homology definition in an obvious way (Doolittle 1995). Fitch (2000) allowed that saying gene A is X% homologous to gene B—meaning that X% of the length of A is homologous to some stretch of B might be OK, but very easily misunderstood. And in fact, even homologous recombination compromises gene homology, if this is to be taken to mean descent from a single common ancestor. Different parts of a two “homologous genes” (considered as stretches of DNA) will derive from different ancestral DNA stretches. A workaround might be arguing that coalescent theory entails that both stretches will have a single common ancestor sometime earlier in their species’ history so that we can still speak of homology between species. But allelic divergences can predate speciation events. Moreover, as Haggerty et al. (2013) point out, the insertions and deletions common in gene diversification mean that “partial homology” (somewhat different from Remane’s concept [Donoghue 1992]) applies even to seemingly straightforward cases not involving gene shuffling. This starts us down the slippery slope to “positional homology”, based on nucleotide-by-nucleotide ancestral inference of common ancestry (Hillis 1994). There is no principled way to decide what fraction of two genes’ respective lengths need to have been inherited from a single common ancestral gene before we can call them homologous.

Most of Many Genomes

And yet, such an awkward way of thinking is almost forced upon us by the properties and mode of evolving of larger eukaryotic genomes. If most of most eukaryotic genomes is not protein-coding but is transcribed and thus (generously) might be said to comprise “genes” for noncoding RNAs, then how might homology be conceived for them? From a recent comparative survey of 16 vertebrate genomes (and that of one sea urchin) aimed at identifying “homologous” long noncoding RNAS (lincRNAs) by synteny, expression pattern and inferred regulatory role, Herzoni et al. (2015, p. 1110) note that …

These homologs share short, 5′-biased patches of sequence conservation nested in exonic architectures that have been extensively rewired, in part by transposable element exonization. Thus, over a thousand human lincRNAs are likely to have conserved functions in mammals, and hundreds beyond mammals, but those functions require only short patches of specific sequences and can tolerate major changes in gene architecture.

At the nucleotide level, positional homology of nucleotides does not demand identical bases: there could have been base substitutions. But it does require that for each properly aligned nucleotide there are only replicative events separating both homologs from the same position in their common ancestor. Indeed that is what “proper alignment” means (e.g., Mindell 1991). To call two molecules homologous when only a small fraction of their lengths meet, this test requires some concept of emergent identity not contemplated by Reeck et al. (1987), indeed of “sameness” not imagined by Owen or Darwin. Many lincRNA researchers seem to have settled on evidence of shared expression patterns or function, a step well beyond “possessing a common evolutionary origin” (Reeck et al. (1987).

Homology Lost in the Mists of Time

Hillis (1994, p. 341), in his masterful review, points to the (much neglected) obvious when he writes …

If we accept that most genes are evolutionarily related, and that extant genomes were derived by duplication, modification and recombination of a small number (perhaps one?) of original replicating sequences, then it is also the case that most genes are at some level homologous.

There may not be a consensus on the origin of life, but many molecular biologists endorse some form of RNA-world theory in which the first biological “information” was cobbled together abiotically in some RNA or small collection of RNAs endowed with the ability to self-replicate and thus undergo Darwinian natural selection. However complex an RNA world then evolved and however coding RNAs became protein-coding DNAs after that, this RNA (or small team of RNAs) would be the common ancestor of all contemporary biological information. To avoid the conclusion that all genes are not “at some level homologous” requires an alternative theory entailing independent origins of biological information, or some general agreement as to when and how (and how many) genes (defined in some agreed-upon way) emerged independently in the world before LUCA (the last universal common ancestor). There are many theories (e.g., Darnell and Doolittle 1986; Poole et al. 1998; Koonin and Martin 2005) but little agreement.

Homology Deconstructed

In other words, Petsko is wrong, perhaps not in his dismissal of “percent homology,” but in his confidence that genes can be said to be either homologous or not.

There is no such thing as percent homology. The meaning of homologous is ‘related by divergent evolution from a common ancestor’. That’s the only thing it means. You can’t be partially homologous: that would be like being partially dead, or partially pregnant. You’re either homologous or you’re not (Petsko 2001).

If nothing ever happened to genes except replication and occasional nucleotide substitution, and if all genomes comprised only such genes, Petsko’s hardnosed certainty would be justified, and in-the-know molecular phylogeneticists could continue to congratulate themselves for having outgrown “percent homology.” But neither is the case, and the problem is more than merely semantic. Because of recombination and other shuffling processes, single common ancestors will be unidentifiable (and undefinable) for much of most genomes and indeed even for the most adamantine protein-coding genes, when allelic recombination is taken on board. Pushing common ancestry back to some coalescent point cannot save the concept in any final or fundamental way because we do not in fact believe that the millions of genes and thousands of gene families now known have all independently arisen from some genetic informationless abiotic state. To save homology, we must contextualize the word, and admit that homology is not something we can discover, rather it is something about which we must negotiate or decide. Like “species” or “the Tree of Life,” “homology” is an invaluable heuristic, a concept without which we could not do biology. But its referent is an ideal that is seldom if ever fully realized.

If indeed most (possibly all) genes are “at some level homologous” and yet we persist in wanting to decide of any two genes whether they are “homologous or not”, then we must impose some further criteria. Possibilities are degree of similarity in gene sequence and/or structure of gene product, inferred number and ancientness of common ancestors, and inferred or demonstrated function. All come in degrees, even the last when the extent of pleiotropy and epistasis are admitted (Storz 2016). “Percent homology” is not an altogether stupid locution, although after so much water has gone over that dam, we do not recommend it.

These issues are not new, or unique to molecular phylogeny; rather, they have coexisted with the concept of homology—a concept which, it should be remembered, predates the theory of evolution—at least since Lankester’s late-nineteenth-century intervention in comparative morphology. The initial switch to a molecular underpinning for homology may have been based on the dream that historical relationships are easier and cleaner to read from molecular rather than morphological data. But molecular data are much more equivocal than we thought in the mid 1960s. In molecular phylogenetics, just as in others areas of biology, and at other levels of the organizational hierarchy, we must contextualize homology if we are to talk about it at all.

In 1987, David Patterson mused: “I intended to float the idea that molecular homologies are no more secure, and are possibly more precarious, than morphological ones” (1987, p. 18). We are not sure that molecular data are more precarious than other data, nor that that their uncertainties are always similar in form, but faith that molecular phylogenetics can solve the homology problem once and for all is unfounded. Mindell and Meyer (2001), in a review with goals similar to ours, suggest that similarity criteria be abandoned: they “can be helpful in diagnosing, but not defining homology.” But we do need such criteria, because without them we are unable (in both conceptual and empirical senses) to identify or delimit traits (at any level) whose homology is to be assessed. That is, we cannot begin to ask if two features are homologs unless we have some way to single them out as features to be compared, that is some measure of similarity (minimally, for instance, that both meet some definition of “gene”). Lankester’s concerns have not gone away.