The Hox code

Blogging on Peer-Reviewed Research

The Hox genes are a set of transcription factors that exhibit an unusual property: they provide a glimpse of one way that gene expression is translated into metazoan morphology. For the most part, the genome seems to be a welter of various genes scattered about almost randomly, with no order present in their arrangement on a chromosome — the order only becomes apparent in their expression through the process of development. The Hox genes, on the other hand, seem like an island of comprehensible structure. These are all genes that specify segment identity — whether a segment of the embryo should form part of the head, thorax, or abdomen, for instance — and they’re all clustered together in one (usually) tidy spot.

Within that cluster, we see further evidence of order. Look at just the Drosophila part of the diagram below: there are 8 Hox genes in a row, and their order within that row reflects the order of expression in the fly body. On the left or 3′ end of the DNA strand, lab (labial) is expressed in the head, while Abd-B (Abdominal-B) is expressed at the end of the abdomen.

Schematic of relationship between Drosophila and mouse Hox genes. Hox genes are shown as colored boxes in their order on the chromosome. Orthologous genes between Drosophila and mouse, and paralogous mouse genes are shown color-coded.

Knocking out individual Hox genes in the fly causes homeotic transformations — one body part develops into another. These genes are early actors in the cascade of interactions that enable the development of morphologically distinct regions in a segmented animal — the activation of a Hox gene from the 3′ end is one of the earliest triggers that leads the segment to develop into part of the head.

Now look at the mouse part of the diagram above. We vertebrates have Hox genes that are homologous to the fly Hox genes, and they’re also clustered in discrete locations with 3′→5′ order reflecting anterior→posterior order of expression. There are differences — the two most obvious that we have more Hox genes on the 5′ side (these correspond to expression in the tail—flies do not have anything homologous to the chordate tail), and vertebrates also have four banks of Hox genes, HoxA, HoxB, HoxC, and HoxD. This complicates matters. Vertebrates have these parallel, overlapping sets of Hox genes, which suggests that morphology could be a product of a combinatorial expression of the genes in the four Hox clusters: there could be a Hox code, where identity can be defined with more gradations by mixing up the bounds of expression of each of the genes.

In the fly, we have a relatively easy situation. Since each segment more or less expresses only one Hox gene, mutating or knocking out a single Hox gene will have an effect on a corresponding segment. In the chordate, though, each segment has at least two and in some cases four Hox genes that may be involved in its development. There is the possibility of redundancy here.

For instance, the HoxA3 gene is expressed in the anterior cervical vertebrae, in the neighborhood of the region where the first neck vertebra articulates with the skull. Deleting HoxA3 had no detectable effects on that joint, however; either its influence is too subtle to be measure, or it has effects on some other aspect of cervical specification, or it has a partner gene that takes over its job in its absence. Notice in the diagram above that HoxA3 has a paralog, or copy, called HoxD3 which is expressed in a very similar place. When HoxD3 is mutated all by itself, there are serious abnormalities: the first neck vertebra has a partial fusion with the base of the skull.

Knock out both HoxA3 and HoxD3, though, and we see evidence that HoxA3 is important after all: the first neck vertebra doesn’t form at all. In fact, it’s thought that the initial mesodermal tissue for the bone has been so thoroughly respecified that it instead fuses completely with the skull, becoming part of the base of the skull.

These results tell us that a combination of Hox genes are required for the proper development of the first cervical vertebra. They also complicate analyses. They say that if you try to knock out the Hox genes one at a time in the mouse, there will be cases where you will see no phenotype or only a partial phenotype, even when the gene does have an important role to play in that segment. What needs to be done is knock out all of the paralogous genes. That is, in order to see what the third Hox genes in the clusters do, we need to carry out a paralogous deletion that destroys the function of HoxA3, HoxB3, and HoxD3 (there is no HoxC3) to assess the phenotype.

This phenomenon is also one reason why we so rarely see homeotic mutations in vertebrates. In flies, you can mutate one gene and you get a haltere transformed into a wing or an antenna turned into a leg; in the mouse, you need to simultaneously zap 2 to 4 genes to get a similar complete transformation.

Now the technology has progressed to the point where we’re starting to see published descriptions of mice with complete paralogous sets knocked out. Look below!

Changes in specific vertebral elements for the Hox5, Hox6, Hox9, Hox10, and Hox11 paralogous mutants. On the left side of the panel, a diagram of the axial skeleton is shown, with specific vertebral elements shown in the right panel marked (C, cervical; T, thoracic; L, lumbar, S, sacral). Wild-type, control elements from specific vertebral positions are denoted by letter and number. The analogous segment from the paralo- gous mutants are shown on the right and left, with colored boxes for each paralogous mutant group.

The cartoon on the left illustrates the skeletal morphology that was assessed. At the top are the cervical vertebrae, C1-C7, which have no ribs. Next are the ribbed thoracic vertebrae, T1-T13.T1-T7 also wrap around to connect to the sternum, which is part of the abaxial skeleton (the vertebrae are part of the primaxial skeleton). Then come the lumbar vertebrae, L1-L6, the sacral vertebrae S1-S4 (which articulate with the pelvis), and the many small tail vertebrae. Each has a discrete, recognizable morphology.

On the right, we see cross-sections of these vertebrae. The middle column is the normal control — that’s what the vertebrae are supposed to look like in a non-mutant mouse. On either side are the mutant forms for each of the paralogous mutants.

For example, look at T1 in the control. In addition to the oval profile of the vertebra, it’s supposed to have a stout pair of ribs. To the left, bordered in green, is the effect of a complete knockout of all the Hox5 genes — HoxA5, HoxB5, and HoxC5. The ribs have started to form, but are incomplete. This is a partial transformation towards a more cervical morphology. To the right, bordered in purple, is what happens to T1 when all of the Hox6 genes (HoxA6, HoxB6, and HoxC6) are taken out: it looks almost exactly like the control C7 vertebra. This is a complete homeotic transformation of T1 to C7.

Now here’s a dorsal view illustrating the effects of these paralogous knockouts.

(click for larger image)

Schematic representation of regions of reported phenotypes in Hox paralogous mutants. Different vertebral elements are denoted by unique shapes, shown in the bottom panel . Aqua-shaded areas demonstrate the regions of anterior homeotic transformations of the somite-derived primaxial phenotypes. Purple-shaded areas show the lateral plate-derived, abaxial phenotypes for each group. The orange background highlights the regions of phenotypic overlap between adjacent paralogous mutants.

Each knockout affects a region. When all of the Hox9 genes are mutated, for example, we see an anterior shift in the midtrunk. The posterior thoracic segments now have ribs that meet the sternum — it’s as if T8-T13 are trying to be T1-T7. In addition, some of the lumbar vertebrae are confused and have acquired thoracic characters, sprouting ribs where there should be none. With these more complete knock-outs of paralogs, we see homeotic transformations all over the place!

Now we can pull the whole story together and map out the morphological domains over which each of these Hox paralogs hold sway.

Schematic of overlaps in and differences between the somite-derived primaxial phenotypes and the lateral plate-derived, abaxial phenotypes of Hox paralogous mutants. The regions for both primaxial and abaxial defects are shown as color-coded bars adjacent to the segments affected in paralogous mutants. Note the differences in AP position as well as the overlap differences in the primaxial versus the abaxial phenotypes.

Notice that not only do we have combinatorial arrangements within a bank of paralogs (those subtleties are not illustrated in the diagram above), but also combinations of sets of paralogs. The sacral segments, for instance, are defined by the expression of both Hox10 and Hox11 genes — one can imagine a kind of logical AND gate in the regulatory circuitry that switches on the downstream genes that signal the specific morphology required for joining to the pelvis only in the presence of both sets of Hox genes. Other experiments suggest that the ground state for a segment is to be thoracic-like and develop limbs; Hox10 and Hox11 may also have functions to suppress rib formation.

What the Hox code represents is a somewhat digital mechanism for regulating axial patterning. By mixing and matching combinations of the expression of a small number of Hox genes, the organism generates a greater range of morphological possibilities. The experiments described in this summary by Wellik are at a rather coarse level, revealing broad chunks of the Hox regulatory scheme, but future work should distill out the details and the specific and finer aspects of morphological regulation. Getting shape from genes is a difficult process to comprehend — the Hox system is one place where we’re getting closer.

Wellik DM (2007) Hox patterning of the vertebrate axial skeleton. Dev Dyn 236:2454-2463.


  1. #1 Torbjörn Larsson, OM
    September 18, 2007

    As far as I can tell, Manx cats do not grow their tail (or just a stub). In our case, we actually have a “tail” (the coccyx).

    While I now appreciate the difficulties with transforming or deleting segments, I was thinking the same. But according to the infallible Wikipedia, Manx cats fall all along the scale between normal to “tailless”, and is classified accordingly.

    The references seems to bear most of that out (but the tail length of a “Longy” isn’t specified more than “visible short”), and also suggests that different spinal defects that can be observed stems from when the gene action happen to affect the spine above the tail.

    But cats can be stably “tailless” as well, Lynx comes to mind. Also, I can think of several groups (or whatever) that have some at least superficially “tailless” species. (Turtles, rodents, et cetera.)