Evolution of median fins

Often, as I’ve looked at my embryonic zebrafish, I’ve noticed their prominent median fins. You can see them in this image, although it really doesn’t do them justice—they’re thin, membranous folds that make the tail paddle-shaped.


These midline fins are everywhere in fish—lampreys have them, sharks have them, teleosts have them, and we’ve got traces of them in the fossil record. Midline fins are more common and more primitive, yet usually its the paired fins, the pelvic and pectoral fins, that get all the attention, because they are cousins to our paired limbs…and of course, we completely lack any midline fins. A story is beginning to emerge, though, that shows that midline fin development and evolution is a wonderful example of a general principle: modularity and the reuse of hierarchies of genes.

First, I have to explain a little bit about the organization of embryonic mesoderm. Recall that there are three germ layers: ectoderm, which forms skin and nervous tissue; endoderm, which contributes to the gut; and mesoderm, which forms connective tissue and muscle. Mesoderm is also organized from medial to lateral in very specific ways, diagrammed below (taken from this more detailed overview of mesoderm development; I’ll also recommend these animations for those still struggling to sort the bits out.)


In the center is the neural tube, which isn’t mesodermal at all—it’s ectodermal. There is also another important ectodermal derivative, the neural crest, which separates away from the neural tube and is going to migrate into various places in the embryo, and is going to make essential contributions to lots of tissues.

At the midline is a special mesodermal tissue, the axial mesoderm, which forms the notochord. Next that is paraxial or somitic mesoderm, which forms the segmented blocks of muscle running the length of the animal. Next out is a band of intermediate mesoderm, which contributes to the urogenital system. Finally, and lateralmost is the lateral plate mesoderm, which is split into a somatic and splanchnic component. This sheet of tissues is going to fold over to form a tube (those animations will help!), and the splanchnic layer will line the endoderm of the gut, while the somatic is going to form various connective tissues and muscle of the body—in particular, somatic lateral plate mesoderm is going to form the connective tissue of the limbs. The paraxial/somitic mesoderm is going to form the body wall musculature, and also some cells are going to peel off and migrate into the limbs to build limb musculature. Complicated, I know…so here’s a another level of complication: the somites are also broken up into subunits.

This diagram (click on it for a larger image) shows how the paraxial mesoderm has a subdivision into a dermomyotome component in orange that will form the dermis of the skin and various muscles, and a sclerotome component in blue that will form, for instance, the vertebrae. Basically, the message here is that the embryo partitions the mesoderm into zones from the midline to further lateral positions, and these subsets of the mesoderm have specific roles to play.


There is also a regionalization of the mesoderm from anterior to posterior. The Hox genes are important players in this process, specifying positional identity along the long axis. Another class of transcription factors, the Tbx genes, are also crucial in defining boundaries and identities—expression of the Tbx5 gene, for instance, determines that a forelimb will form, while Tbx4 determines the formation of a hindlimb.

One way to think of this is that developmental processes take this sheet of mesodermal tissue and mark it off with a grid: there’s an organization from axial to lateral plate mesoderm, and there’s a demarcation from anterior to posterior defined by Hox and Tbx genes. Then, at specific coordinates in this grid, other genetic cascades are activated to start limb formation. I’ve described them before. An apical ectodermal ridge (AER) forms laterally for each of the fore and hind limbs, a zone of polarizing activity (ZPA) develops along the posterior edge and confers polarity on the limb, and a series of Hox genes define position within the limb. So, at a particular coordinate, a whole cascade of genes are turned on the generate a thickening, and a bump, and lead to a well-organized limb forming.

Now, how does a medial fin form? If you review this article on the formation of the genitals, you’ll have a clue. That article describes a series of genes activated to form the ‘bump’ of the embryonic penis/clitoris, and shows that it is a reactivation of the same genes used in forming the ‘bump’ of the limb. As you might guess, Freitas et al. find the same genes playing a role in making fins along the midline in sharks and lampreys as the ones that set up paired limbs.

This is a typical developmental paper, as I mentioned the other day, that means it is full of in situ stains for expression patterns of RNA. It’s a lot of data—there are matrices of the region of expression for various Hox genes in a series of midline fins, there are markers for sclerotome and neural crest, there are genes, genes, genes turned on all over the place. I’m not going to show any of the data for a change, since it’s either show the whole pattern or show a fragment out of context, which is unhelpful, so instead I’ll just summarize and you’ll have to trust me. Or look it up in Nature yourself.

The story is straightforward. At the midline, an apical ectodermal fold (AEF) forms that leads to the growth of the dorsal and ventral fins. Where the limbs are going to develop out in the lateral plate mesoderm, the median fins are going to draw on the cells of the paraxial mesoderm to form their substance—in particular, cells of the sclerotome are going to migrate dorsally to form the tissue, with some contribution from the dermamyotome. In addition, neural crest cells will migrate into the fin fold. Within the fin itself, anterior-posterior regionalization is defined by the same genes that do likewise in the developing limb, Hoxd9, Hoxd10, Hoxd12, and Hoxd13. A Tbx gene, Tbx18, is also expressed at the anterior border of each midline fin. Here’s a simple diagram of the mesodermal sources for the dorsal fin.

Schematic summary of the cellular contributions to the dorsal median fins.

A dorsal fin looks an awful lot like a grossly simplified pectoral fin, which is a simpler version of a tetrapod limb (with differences, of course!) A dorsal fin, though, is an older structure, and those paired lateral fins are later evolutionary innovations. Both sharks and lampreys use homologous genes to develop these midline fins, which implies that we’re looking at a primitive, conserved mechanism. What all this suggests is that Cambrian fishes evolved patterning mechanisms to form a simple midline structure, the dorsal and ventral fins, with a small suite of genes to impose spatial organization on it, and sculpting paraxial mesoderm to build that thin flap. With that module in place, the easy way to build a lateral fin was to simply reactivate that very same cascade of genes in the lateral plate mesoderm, and presto…a flap forms in a novel location. Later elaborations on that basic genetic cascade led to the details of the teleost paired limbs, and other elaborations would have led to the tetrapod limb, but the core of the developmental process involved seems to reside right there, in a simple midline fin.

The authors propose that the next step is to look at the midline fins of cephalochordates, predicting that the same genes will be used there. I expect that they’ll be right. I would predict that we’d also find the same genes used in another novel structure, the re-evolved dorsal fins of whales. The redeployment of previously refined genetic modules is going to turn out to be a universal property of evolved systems, I expect.

Freitas R, Zhang G-J, Cohn MJ (2006) Evidence that mechanisms of fin development evolved in the midline of early vertebrates. Nature advance online publication, (doi:10.1038/nature04984).


  1. caynazzo says

    Excellent summary. I’m working toward a zebrafish model for T-cell Leukemia. I’m interested in endogenous oncogenes such as scl and lmo1 as a positive control to test my riboprobe in full mount in situ hybridization experiments. These endogenous oncogenes are switched on at around 48 hours post fertilization and staining shows up in hematopoietic tissue, which means I need to brush up on my developmental biology to determine where this occurs in 48hpf zebrafish embryos.
    Thanks for the insights.

  2. vandalhooch says

    If your prediction about whale fins holds, would that mean that whale and shark dorsal fins would not be an appropriate example of analogous structures as they are often depicted in High School textbooks?

    Do you know of any plans to follow that line of investigation?

  3. says

    You raise a very interesting point, vandalhooch. As I understand it, they would still be considered analogous, because they function in the same way. “Analogous” is often used as shorthand for “analogous although not homologous”, which I think your textbook example would be referring to. So if it is borne out, they would be considered analogous AND homologous, rather than analogous AND NOT homologous, as the textbook probably means.

    But the implications of redeployment for homology are most interesting–at what level would redeployed structures be considered homologous? At the level of fin itself, or at a more abstract level? And often, in the absence of a perfect fossil record, how would we detect redeployment versus retainment of soft structures?

  4. Steviepinhead says

    Being an inveterate (closely related to invertebrate) pinhead, my best guesses often gang aft agley.

    Howsomever: my best guess would be that, however homologous the genetic modules which induce midline fish fins and midline whale dorsal fins may turn out to be, the structures themselves (da finz!) would still be considered analagous/convergent, rather than homologous sensu strictu. The killer whale’s dorsal fin is simply not a retooled version of the shark’s midline fin, the way a bird’s wing is a retooled version of the homologous structures (quadruped forelimbs, fish lateral fins) found in related lineages.

    Likewise, pteranodon wings and bird wings may both be homologous to quadruped forelimbs, but neither wing is strictly homologous to the other–they are different adaptations or “takes” on the underlying forelimb architecture, albeit utilized for an analagous function, flight.

    Thus, I would second RavenT’s intuition that any homology here would be considered to be at a deeper, more abstract, not-strictly-structural level…

  5. says

    Being an inveterate (closely related to invertebrate) pinhead, my best guesses often gang aft agley.

    Well, perhaps. On the other hand, we seem to be thinking about it along the same lines, so if I *am* going astray with this, at least I’ll have entertaining company on the journey :).

    Thus, I would second RavenT’s intuition that any homology here would be considered to be at a deeper, more abstract, not-strictly-structural level…

    And that in turn ties into some very interesting things that John Wilkins has written on the abstract, the concrete, and the reification fallacy in biology. If we’re looking for homologies at different levels of abstraction, how do we determine them (and specifically for my research, how do we represent them in a computer system)?

  6. Steviepinhead says

    Well, not yet having had a chance to digest Wilkins, I can only hazard that “levels of abstraction” is a bit too abstract to adequately convey the physical reality. On that level of physical nitty-gritty, it seems to me that things like a midline fish fin, a whale’s dorsal fin, a quadruped limb, a pterandon wing, a bird wing, a bat wing–these are phenotypic structures about which we can talk meaningfully, even if, like every other “discrete” entity in the world, they shade off into other things: is the shoulder still part of the limb?

    Of course, such “discrete” structures can be dissolved into constellations of sub-units observed at other “levels” of physicality–the developmental precursors and post-cursors (eh?) of a given structure during the life-arc of a given crittur; the sub-units (bones, muscles, nerves, tendons, blood vessels, fingernails, hairs, skin, cell types, proteins) that at various levels of reduction make up the discrete structure, many of which may well not be unique or confinable to that structure, but which may extend into, interconnect with, and communicate with other structures.

    While, again, any given conceptual entity, even one which in some sense exists in physical reality, may blur into a welter of other concepts and entities, still it seems that meaning may be more usefully assigned to some such entities–at least for the purposes of any given inquiry–than to others. Even as we recognize the tentative and interdependent nature of our division of the physical world into such “separate” entities and structures (and levels and hierarchies of such entities and structures).

    Rather like a word, which may have a more-or-less firmly-bounded socially-constructed denotation, but which is defined only through the use of other semi-slippery words, defined by still other words… That, in some sense, our web of mutually-dependent words and concepts is full of holes and is ever-susceptible to coming unraveled no more prevents us from snaring and sharing meaning via our word-web than the lace and frailty of a fish-net prevent it from seiving fish from the torrent.

    Thus, while the levels and hierarchies may all shade into one another–and will for some purposes need to be ignored or allowed to overlap and crosscut with one another–still for other purposes a cell is a cell, a nucleus is a nucleus, DNA is DNA, and a limb is a limb and NOT a tendon, a bone, a sinew, a nerve cell or a muscle cell or a white blood cell.

    And a gene-signalling-complex can thus be meaningfully be lined up against another similar module, though they may be respectively “located” in long-separate lineages; likewise, limbs, fins, and other similar “extrusions” sported by various different critturs can be defined and compared in some meaningful way to (tentatively, at least) begin to decide whether they are “extensively” homologous (the result of developmental suites that unfold in similar fashion, “realizing” themselves in similar suites of sub-units on a host of scales and timelines), as opposed to structures, the development of which may be “triggered” by homologous gene-suites, but which then follow markedly-different pathways toward the construction of physical entities which may or may not evince a functional similarity, but which bear little architectural or developmental similarity at any “higher” level.

    Put another way, I don’t think comparable or competing “styles” of architecture are merely subjective, unspecifiable, overlapping blurs. While I’ve used arguably-vague terms like “little,” “markedly,” and “extensively” above, the art historians and the cladists(? whatever you call those systematists)–to pick scholars working in widely-separate fields–have nonetheless managed to come up with schemes which enable them to render meaningful–and often independently verifiable–comparisons and distinctions: this mask was carved or this basket was woven by a (traditional) Haida artist, and not a Tlingit, because it exhibits these material, structural, and stylistic characteristics. Of two Haida baskets, this one was likely woven in mid-19th C. Skidegate rather than in late 19th C. Howkan. Of two Haida baskets woven in the village of Massett at roughly the same time, this one was likely woven by Isabella Edenshaw, while this one was likely woven by Mrs. “Tom Price.” And so forth.

    Again, I don’t know what Wilkins’ take may be, but it’s NOT “all in the eye of the beholder.”

    It’s a matter of drenching yourself in the specifics, then emerging long enough to hypothesize useful criteria, and a useful hierarchy within which to deploy them, then diving back into the data to test whether they usefully decompose the data, then refining, tweaking, thrashing, re-testing…

    All that hard stuff, that maybe only you–of all the scholars in all the world–can elucidate for your particular sub-field!

    Have fun, y’hear!

  7. Loren Petrich says

    There’s an odd aspect of the Hox genes involved in forming fins and limbs. Hox genes Hoxd9, Hoxd10, Hoxd12, and Hoxd13 are also involved in forming the tail.

    So does this mean that fins are ectopic tails?

    Vertebrates share with other chordates a tail extending past the anus; no other invertebrates have that feature. It would be interesting to see what Hox genes are involved in invertebrate-chordate tails; vertebrate-tail Hox genes have no counterparts in the better-studied invertebrates like Drosophila.

    But there is evidence from the expression patterns of Distal-less and some other genes that vertebrate and arthropod extremities have some shared growth mechanisms, even if they are not strictly homologous, and even if arthropod limbs do not have Hox genes involved internally.

    These are certainly remarkable times to be living in; I remember wondering if the genes-to-shapes mechanisms will ever be discovered before my old age. But in this and other research, we see some remarkable steps along the way.

  8. Torbjörn Larsson says

    The discussion on the abstract and the concrete was interesting. I agree with Stevepinhead as I read his conclusion, that Wilkins while raising an important point (obviously, if it is a problem in a concrete science project) is not allowing the reality/concrete enough. There seems to be methods used to judge this.

    For example, physicists like to reify what I think Wilkins call “secondary properties” for some classes of objects, seemingly based on methods or at least specific criteria.

    For example, some quantizations are called particles, some pseudoparticles. (There are more types, virtual, ghosts et cetera.) Pseudoparticles are as I understand it collective quantizations, for example vibrations, phonons, in materials. They may be more or less localised but no one thinks of them as a real particle. Here it is the idea that an essentially nonlocalised effect isn’t the same as a localised object. This is more of a criteria, I think.

    Another more fundamental idea seems to be that objects, whether observed or infered, are real if they persist after transformations, in the model or in various models. This is more of a method, it seems to me.

    Persistance in one model can be exemplified by QM. The wave-particle duality makes one able to see the same object as two different phenomena, but the wave/particle persists as a quantization of a field. Both exists for the realists.

    Persistance in several models can be exemplified with dual theories such as AdS/CFT. An atom is a local object in our seemingly AdS universe, while it probably persists as a local object of another kind in the holographic CFT theory. (I’m guessing here.) (And of course nowadays atoms can be directly observed in some microscope methods such as AFM. But even before that they were considered real, for a number of reasons.)

    There are more such methods, I’m sure. But the idea of persistance, on some level, seems to tie in with Stevepinhead’s discussion of that the observed levels and relations affects whether a homology should be said to be preserved or not. Either a homology isn’t concrete at all, or it is concrete on some levels over some relations?

  9. Torbjörn Larsson says

    “Both exists for the realists.”

    I’m unclear. I mean both the wave/particle and the field exist, since they are the persistent objects that remains over all experiments.

  10. says

    Either a homology isn’t concrete at all, or it is concrete on some levels over some relations?

    Yes, that’s right. But determining those homologies and their appropriate level of granularity is non-trivial. Once they are determined, then modeling them correctly (enough) in a knowledge base is yet another issue where

    For example, when I first started developing my comp anatomy information system for my dissertation research, I modeled the relation homologous-to between entities in the knowledge base such as Ventral prostate (rat) and Prostate (human). It soon became obvious that that was insufficient for a couple of different reasons: 1) partial homologies and 2) complex homologies (at least; possibly more reasons TBD). The Ventral prostate (rat) is homologous in some degree to the Prostate (human), yet not (as you might expect, and I sure did) to the Anterior lobe of prostate (human) [Price 1963]. So:

    1) there is a homology there,

    2) but I’m not quite sure what it is,

    3) so modeling it correctly in my information system is problematic.

    Of course, the cure for that problem is more information–just because I don’t yet know what the homology is is independent of the fact that it does exist.

    Second issue: Clearly homologous-to is insufficiently granular to correctly represent the situation. So on a first re-iteration, I added attributes and modeled the relationships homologous-to:embryologically and homologous-to:genetically to represent the evidence in the literature. That worked out for a while, until I came to: “…on the basis of an epidemiological study, Xue [6] reports that the mouse dorsolateral prostate corresponds to the peripheral zone of the human prostate, and that Roy-Burman [7] concurs on a preliminary basis, but cautions that Xue’s assertion is based on descriptive data, and that the molecular studies that would confirm the correspondence remain to be carried out.” [Travillian 2006]

    Once again, the relationships need refinement. Part of the critical feedback I got in the evaluation of my system is that the interface switches back and forth confusingly between terms such as “homologous”, “maps-to”, and “similiar”. And the evaluators are quite right–I need to tighten up the definitions of those relationships, and ensure that the interface is unambiguous on them–which means establishing the state of a lot of knowledge in the literature.

    And now PZ has posed another interesting issue, that of the homologies in redeployed structures (and, for me, how to represent them correctly in an information system). So you’re absolutely right–these homologies are concrete at some level over some relations. It’s just that a lot of work remains to be done in determining those relations, and in representing them in a useful information system, and in the process of that work, we run some risk of the problem that Wilkins has pointed out.

  11. says

    Oh, if I’m going to cite the literature, I should provide the references:

    [Price 1963] Price D. Comparative aspects of development and structure in the prostate. Natl Cancer Inst Monogr. 1963 Oct;12:1-27.

    [Roy-Burman 2004] Roy-Burman P, Wu H, Powell WC, Hagenkord J, Cohen MB. Genetically defined mouse models that mimic natural aspects of human prostate cancer development. Endocr Relat Cancer. 2004 Jun;11(2):225-54.

    [Travillian 2006] Travillian RS. Ontology Recapitulates Phylogeny: Design, Implementation and Potential for Usage of a Comparative Anatomy Information System. University of Washington doctoral dissertation, in preparation.

    [Xue 1997] Xue L, Yang K, Newmark H, Lipkin M. Induced hyperproliferation in epithelial cells of mouse prostate by a Western-style diet. Carcinogenesis. 1997 May;18(5):995-9.

  12. Torbjörn Larsson says

    Okay, it seems like you have your work cut out for you!

    Perhaps there is a method to tease out the concrete cases analogous to physics. Or you have to go through all these cases and relations. The list of conflicting uses of the gene concept that Wilkins mentioned doesn’t bode well.

    Good luck, in any case!