Let us now turn our attention to the hypothetical addition of a filament to a protein export system:

Let us first consider filament formation with sickle cell hemoglobin, as this example highlights problems with such a cooption scenario rather than supporting it.

Hemoglobin is normally a globular, soluble protein. In sickle cell anemia, a mutation has occurred in the sixth position of beta globin that substitutes a hydrophobic valine residue for a charged glutamate. This results in a conformational change when the protein is not bound to oxygen that exposes the newly created oily patch (thanks to valine) on the surface. Thus, individual hemoglobin proteins can now stick together in a chain-like fashion because every protein has two oily patches (due to there being two copies of beta globin per hemoglobin molecule). Yet there are some characteristic features of HbS (sickle cell hemoglobin) filaments that make them much different from any of the surface filaments the EFM hypothesis may allude to, including the flagellum.

• HbS filaments are homogenous, being composed only of chains of hemoglobin. Bacterial surface structure filaments are heterogeneous and ordered, as we shall explore some examples.
• HbS filaments form inside the membrane of the red blood cell (RBC). In fact, something like 30-40% of the RBC's volume is filled with hemoglobin. This crowding ensures that the oily patches of the HbS proteins are likely to find each other and form stable interactions. None of this applies to the surface structure filaments of bacterial, which form on and outside the cell's membrane.
• HbS filaments are solid, while the surface structure filaments are hollow. This is especially relevant in the case of the bacterial flagellum, as its hollow core is essential to its construction.

When we consider the differences outlined above, it should be clear that any filament formation associated with the bacterial structures is going to be quite different from the process of HbS filament formation, indicating that any claim of filament formation being easy for a protein is misleading in this context.

It is understandable that the EFM hypothesis keeps the cooption event so vague, as attempting to be a little more specific would only highlight the inherent problems associated with such musings. For example, it proposes a single mutation that does two things, causing the exported proteins to stick to themselves and causing the exported proteins to stick to the outer components of the export machinery, is necessary to bring about some change that is even vaguely selectable. That is, we assume such simultaneous changes because if they are separated, export proteins that simply stuck to themselves upon secretion would simply float away as multimers and export proteins that stuck only to the export machinery would merely decorate it, perhaps hindering transport of other material, and not produce anything that resembles a filament.

But in order to have these simultaneous changes, one might expect the mutation to be rather non-specific. And this poses a huge problem for the EFM hypothesis. What is to prevent the exported proteins from aggregating before they are transported? What's to prevent them from sticking to something else not in the basic story? What's to prevent them from being degraded? And what's to prevent them from clogging up the transport channel during any stage of the transport process? And what if this mutant export-protein polymerized like HbS filaments and formed a solid core? Then, we'd seal off the export machinery, which presumably is important for other reasons. Thus, it would seem the EFM hypothesis is relying on deleterious mutations to evolve his filament, unless of course, it is relying on a special mutation that just happens to perfectly fit the story.

And then I can't help but wonder just what type of advantage there is to having some protein filaments globbed onto the export machinery. One might argue that there are many uses for nonmotile filaments as seen in extant bacteria, such as adhension and conjugation. Yet we can rule out things like conjugation as such functions are unlikely to have been carried out in the very simple filament imagined by EFM hypothesis (conjugation, for example, requires another story about IC machines). I suppose adhering to a surface could be useful, although in most cases, the surface is usually another cell. But that brings up another point about this special mutation. Not only is it supposed to allow for the exported protein to stick to themselves, without aggregating into some clump prior, during, or just after transfer, and it is supposed to allow for the exported protein to stick to the export machinery without clogging it, but now the end of this filament must also stick to something else. A mutation that does three good simultaneous things without causing any harm. Like I said, it's special.

Nevertheless, it would seem even if we had a special mutation, the export machinery is now attached to some filament, which would seem to cause some form of interference with the other material the system was previously selected to transport. Thus, regardless of the mysterious advantage associated with attaching a proto-filament to the export apparatus, we ought not ignore the disadvantages associated with gunking it up. Yet the EFM hypothesis does exactly this.

Non-telic transition stories typically have to keep things simple and sloppy like this. But in doing do, one wonders if they also have to abandon biology. Let's consider what is known to be involved in the simplest and most common form of filament formation in bacteria. Let's consider the P pilus as a model system to see how close it is to the EFM hypothesis.

The P Pilus

The P pilus is a very thin filament, whose outer diameter is only about 7 nm with a hollow core about 2 nm in diameter. The rod is thicker near the membrane and thins as it nears the tip. It functions as an attachment organelle, that is, it can reach out and anchor bacterial cells to other cells. The end of the filament has a protein that specifically binds to certain sugar molecules found on kidney cells.

Although the P pilus is among the simplest of attachment filaments, it is encoded by 11 genes. The filament itself is a heterogeneous structure. The primary subunit is PapA (it forms the thicker rod near the membrane). But as we get near the tip, we find another protein, PapE, forms the thinner filament.. At the very end, is PapG, the specific adhesin that binds to sugars on other cells. PaPG binds to PaPE through an adaptor protein, PapF. And PapE binds to PapA through another adaptor protein, PapK. Thus, the pilus itself is composed of five different proteins that are assembled in a fixed order (PapA - PapK - PapE-PapK-PapG, proximal to distal).

How is this pilus synthesized in such an orderly fashion? Like most other pili and adhesive organelles, it starts with the highly conserved usher/chaperone pathway. And here is where things not only get interesting, but also begin to look very different from the simplistic account of filament formation assumed by the EFM hypothesis.

It all begins with the Sec export machinery found in the cytoplasmic membrane:

Protein translocation across the bacterial cytoplasmic membrane has been studied extensively in Escherichia coli. The identification of the components involved and subsequent reconstitution of the purified translocation reaction have defined the minimal constituents that allowed extensive biochemical characterization of the so-called translocase. This functional enzyme complex consists of the SecYEG integral membrane protein complex and a peripherally bound ATPase, SecA. Under translocation conditions, four SecYEG heterotrimers assemble into one large protein complex, forming a putative protein-conducting channel. This tetrameric arrangement of SecYEG complexes and the highly dynamic SecA dimer together form a proton-motive force- and ATP-driven molecular machine that drives the stepwise translocation of targeted polypeptides across the cytoplasmic membrane. Recent findings concerning the translocase structure and mechanism of protein translocation are discussed and shine new light on controversies in the field. [1]

However, since this four-part machine is ubiquitous and has many uses other than helping form filaments, let's grant its existence and deal with it by itself as a separate topic some other day.

Also, at this point, I should mention that we're discussing gram-negative bacteria, which have two membranes. The inner membrane is a typical cytoplasmic membrane and the outer membrane is more porous (due to many barrel-shaped protein pores that filter out large material but allow smaller things like sugars and amino acids inside). The space between the two membranes is called the periplasm. Transport via the Sec pathway dumps material into the periplasm. The trick for the bacteria is to grow this into a filament that penetrates the outer membrane in a coordinated manner. So how do cells make P pili?

First, you export all the pilus subunits into the periplasm using the sec-machinery. The proteins are threaded through the sec-machinery in an unfolded state and most refold in the periplasm. And therein lies the problem, as the pilus subunits easily form insoluble aggregates (or clumps) in the periplasm through hydrophobic interactions. To prevent this, we need to invoke another component, a special chaperone encoded by PapD. PapD does two things - it binds to the pilus subunits after they are pumped into the perisplasm and prevents them from clumping with each other and also helps the pilus subunits to fold into their proper conformation. In fact, the pilus subunits are not stable as monomers and exist either as bound to the chaperone or as bound to each other as part of the filament. The manner in which the chaperones carry out their function is far more elegant than anyone assumed, employing something that is now called "donor strand complementation" (DSC).

The 3-D structures of PapD complexed with PapG (the adhesin on the tip) and PapK (one of the adaptors) have been solved. PapD forms a boomerang-shaped protein with two immunoglobulin-like (Ig-like) domains (a structure composed of layers of antiparallel beta sheets). The N-terminal end of PapK is also an Ig-like domain, but it lacks a C-terminal beta sheet that normally contributes to the hydrophobic core of the domain. This produces a cleft that exposes the hydrophobic core, which is what makes it so sticky and prone to aggregation by itself. The chaperone PapD masks this exposed region in a most fascinating manner - it donates one of its beta strands to complete the Ig-domain in PapK (Fig 1). But it does so in an atypical fashion, as the beta strand it donates runs parallel, not antiparallel, with its neighboring strand. Thus, PapD provides at least two essential functions captured in one very elegant act - by donating one of its beta strands, PapD simultaneously prevents aggregation of PapK while providing the missing steric information for proper folding of PapK.

Fig 1. Modified from [2]

And what this means is the folding of pilus subunits is IC. By themselves, the subunits don't fold properly and are unstable. The steric information for proper folding is not found in a single amino acid chain or gene, but in two distinct chains/genes. And By itself, PapD has no function. Clearly, the simplest known filament is far more sophisticated than the filament imagined by the EFM hypothesis (i.e., biology is not as simple as it assumes).

What happens next? The pilus subunit-chaperone complex interacts with a protein channel on the outer membrane, PapC (also known as the usher). The channel is large enough to accommodate the tip of the filament, but not the rod. The actual mechanism of incorporation is being worked out, as the chaperone somehow hands off the pilus subunit to the usher for incorporation into the growing filament. Interaction between the usher and chaperone-pilus subunit does not result in the chaperone-subunit complex breaking apart, thus the mechanism of handoff is also probably quite complicated and sophisticated.

But there is one more feature to the story worth mentioning. The pilus subunits themselves are thought to form a filament through a donor strand complementation-like mechanism. Each pilus subunit has an N-terminal extension that does not contribute to its own folding. By itself, it is a disordered strand. However, it has been proposed that this N-terminal extension from one subunit (let's call it A) displaces the displaces the donated chaperone strand associated with another pilus subunit (B). This N-terminal strand would then form a beta strand that runs in an antiparallel direction and complete the Ig-domain of its neighbor in a typical fashion.(Fig 2) Again, the steric information for the Ig-domain of subunit B is supplied from subunit A. This mechanism is called donor strand exchange. And the result is that the filament is made by linking subunits, where each subunit contributes a strand to perfectly complete the fold of its nearest neighbor.

Fig 2. Modified from [2]

Thus, it should be clear that some ad hoc notion of an export protein sticking to itself and sticking to the export apparatus to form a filament does not reflect the biology of the simplest known pilus. Life is much more sophisticated than this. Thus, all the examples of simple, nonmotile filaments in bacteria provide no obvious support for the EFM speculation.

As if having your supporting evidence shown to be irrelevant was not bad enough, there are more problems. For example, let's imagine that with enough luck, somehow a P-pilus-like materializes. After all, such pili are the most common. And therein lies the problem, because while the P-pilus makes a great attachment organelle, it's probably a dead-end if one wants to evolve a flagellum. For one reason, the P-pilus has not been observed to secrete proteins. This could be because the channel is so small . Or it might have something to do with the energetics of the system, as P pili formation is independent of cellular energy. It's not surprising that the P-pilus looks very different from the bacterial flagellum (or even things like type IV pili).

Finally, there is yet another fact that suggests flagella did not arise in the manner that the EFM proposes. Whether we're talking about simple type I pili or more complex type IV pili, what they all share in common is being built from the bottom-up. The flagellar filament, in stark contrast, is built from the top down. And the manner in which this done is yet another amazing story in microbiology. How amazing? Robert Macnab is an expert on the flagellum and has been working on them his whole life. As such, you might expect him to be used to the complexity and sophistication of the flagellum. Yet he reacted by noting that this mechanism is " a much more sophisticated process than any of us could have envisaged."[3] In fact, consider how this was reported:

The latest technical discoveries in flagella fascinate biologists such as Robert Macnab, a professor of molecular biophysics and biochemistry at Yale University who also studies flagella. He marvels at how organisms as simple as bacteria have evolved such complex methods to develop propelling features, especially since motility in bacteria is not directly necessary for survival, like DNA replication or protein synthesis. "We think it would not be possible for the system to work with any significantly lower complexity." [4]

So let's have a look to see how well the EFM hypothesis' filament formation story anticipates the actual mechanism bacteria use to form filaments.

Shigella are nonmotile pathogens. Even though Shigella do not express flagella, they do possess the flagellar operons, suggesting this nonmotile state was recently acquired. Four strains were recently analyzed, showing that loss of flagella has occurred independently.[5] In two strains, the only thing missing was fliD, the gene that codes for the protein that caps the filament.

What happens if you don't have fliD is that no filament forms? As Ikeda et al. explain, "A fliD-deficient mutant becomes non-motile because it lacks flagellar filaments and leaks flagellin monomer out into the medium." [6] FliD is not merely a regulator or aid, but an essential component for filament formation. To understand why, let's consider the research results that fascinated Macnab and others.[7]

The fliD gene products form a five-member pentagon-shaped ring that caps the hollow filament formed by flagellin subunits. Each member of this pentamer has a leglike extension that points downward and interacts snuggly with the filament. However, there is a symmetry mismatch between the cap and the filament. The cap is formed from five protein subunits, but the helical end of the filament itself is formed from 5.5 flagellin subunits. Macnab explains the significance of this as follows: " When one protein of the cap pentamer is at the dislocation point (think of a split washer), it will be in a very different environment from the other four members of the pentamer." [3]

In other words, a significant crevice is associated with the cap and end of the filament. And it is proposed that the next flagellin subunit that gets added to the filament is added to this crevice. The addition of the new flagellin subunit is then coupled with the cap itself rotating along the filament axis to open up a new adjecent crevice. As Macnab suggested, think of the cap as a split washer (where the center is filled) sitting on the end of a hollow tube. Individual flagellin proteins travel down to the tube to be added at the tip. The flagellin then gets placed into the space of the split washer, the washer turns, and opens up a new space. Thus, you can envision the cap spinning around, inserting new flagellin monomers one-at-a-time. (Fig 3)

 Fig 3 (adapted and modified from [7])

Duane Salmon once estimated that the growth rate of the filament to be about 50 flagellin units/sec.[8] Since there are ca. 5 subunits per turn of the helical filament, this suggests that the fliD cap rotates about ten times every second as it incorporates about 50 flagellin subunits.

What's most relevant about this is that the C-terminal and N-terminal ends of flagellin subunits are unfolded as they travel down the hollow filament tube, as the folded protein has a significant kink in its middle that would prevent transport through the tube. As Macnab notes, "large conformational changes would be required in the monomers before they could be added to the filament tip." Thus, the fliD cap also does not simply provide a passive, mobile slot to insert flagellin subunits. It also helps flagellin fold. In other words, the cap is a chaperone. Thus, the flagellar filament is built in a way that is similar to P pili and quite different from HbS filaments; the flagellin units do not "self-assemble," they are assembled by a processive chaperone at a rather impressive rate.

Things get even more interesting when one considers that just below the cap, the filament cavity is expanded such that its cavity is about twice the size of the central channel that runs through the rest of the filament. It is suggested that this might be the site in which flagellin folds in a manner that is analogous to the folding that occurs in the GroEL chaperonin in the cytoplasm. The parallels are interesting. GroEL is capped by GroES to form a closed chamber, while FliD also functions as a cap to form a closed chamber. It is suggested the filament chamber can house one flagellin monomer at a time, which is exactly how GroEL works. Yet there are a couple of significant differences that probably stem from the fact that GroEL is a generic chaperone chamber that functions only to fold a diverse set of proteins, while the chamber at the distal end of the filament folds and incorporates only one protein, flagellin. The first difference is that GroEL requires energy in the form of ATP hydrolysis that alters the volume of the chaperonin. It is intriguing to speculate that the folding chamber at the end of the filament also undergoes cycles of volume changes associated with the rotation of the cap and insertion of a new flagellin filament. In such a case, the energy could be derived from the winding coupled to favorable protein-protein interactions associated with assembling flagellin subunits into the filament. Secondly, the filament chamber would cycle much faster that GroEL. The typical GroEL cycle lasts 15 sec. The filament, on the other hand, is incorporating 50 subunits/sec. That's folding individual monomers every 0.02 seconds, which is 750 times faster than GroEL.

There are several clues that point to design here.

1. Flagellin/fliD and GroEL/GroES are not homologous. Yet if the flagellin/fliD chamber functions as I suggest, we have another system whose sophisticated mechanism is related in a logical fashion (another example would be in the similar proofreading mechanisms of DNA replication and attaching amino acids to tRNA).

2. FliD and flagellin form an IC relationship. FliD has no other basic cellular function apart from forming the filament. Flagellin too has no other basic cellular function apart from forming the filament. And both are needed to form the filament.

3. As suggested, there seems to be only enough room for one flagellin monomer to fit into the chamber and fold. If this is essential, we have another IC-like interaction. Flagellin must be first unfolded to transport through the channel. But it must also be folded again to be incorporated into the filament. If this second folding event depends on the distal chamber, then two independent events must be carefully coordinated to construct the filament.

And there is one more interesting twist on all of this. There is suggestive evidence that the hook-associated proteins, those that attach the filament to the basal body and the fliD cap itself, may be chaperoned through donor-strand complementation. Specifically, there are two chaperone proteins that specifically interact with the C-terminal ends of the hook-associated proteins and cap and prevent their premature aggregation. Thus, just as there is a mini-IC relationship with flagellin and the cap, the cap and hook proteins may also share an IC relationship with their specific chaperones. Again, we would see the basic conceptual strategy in protein folding and assembly as seen independently in the P pilus. And the "self-assembly" is highly regulated - a chaperone helps assemble the hook, another chaperone helps assemble the cap, and the cap assembles the filament. In other words, and here is the interesting point, we will soon begin to make a strong argument that assembly of the flagellum itself is IC.

To sum this section up, let's consider more problems inherent in the EFM hypothesis

• The EFM hypothesis is divorced from biological reality, as the formation of the simplest filaments (the p pili) is far more involved (at its core) than a protein simply sticking to itself.
• The EFM points to other filaments that employ bottom-up construction to explain the top-down construction of the bacterial filament.
• It is not clear that a transport system, by itself, is "preadapted to form a filament."
• Even if it is true that secretion systems are preadapted to form a filament, such "preadaptation" may very well steer a forming structure away from the fitness peak associated with a flagellum-like structure. For example, the most common filaments do not transport proteins, probably because they are too small and lack sufficient energy sources: "Thus, the chaperone/usher system might not be able to adapt for secretion of soluble proteins." And there is no reason, according to the EFM hypothesis, that the filament must be hollow. One might claim there are lots of uses for nonmotile filaments currently in use by living bacteria. Yet how many have gone on to become rotary propulsion units?
• The bacterial filament itself, along with its assembly process, is IC. It is fundamentally more sophisticated and complex than anything foreshadowed by the EFM hypothesis, indicating again that this hypothesis is divorced from biological reality.

NEXT: In the next essay, I will consider the cooptional addition of the motor to determine just how grounded this just-so story is in biology

1. Manting EH, Driessen AJ. 2000. Escherichia coli translocase: the unravelling of a molecular machine. Mol Microbiol Jul;37(2):226-38.
2. Donor Strand Exchange
3. Science 290, p. 2087
4. Bacteria create natural nanomachines
5. Al Mamun AA, Tominaga A, Enomoto M. 1997. Cloning and characterization of the region III flagellar operons of the four Shigella subgroups: genetic defects that cause loss of flagella of Shigella boydii and Shigella sonnei. J Bacteriol 179(14):4493-500
6. Ikeda T, Oosawa K, Hotani H. 1996. Self-assembly of the filament capping protein, FliD, of bacterial flagella into an annular structure. J Mol Biol 259(4):679-86.
7. Yonekura K, Maki S, Morgan DG, DeRosier DJ, Vonderviszt F, Imada K, Namba K. 2000. The bacterial flagellar cap as the rotary promoter of flagellin self-assembly. Science 290(5499):2148-52.
8. Duane