Skip to main content

Analytical Chemistry

More on Crystal Formation (This Time With Proteins)

I realize that I was just talking about crystal formation here the other day, but there’s yet more news in the area, and it comes in the fiendishly difficult area of protein crystallography. All you have to do to appreciate the horrors of this field is to step into a lab that does it for a living: stacks and stacks and stacks of plates, each with a slightly different set of conditions (buffers, concentrations, additives, protein construct, and son on) to try, once again, to get a protein crystal that diffracts well enough for structure determination. It’s especially tricky to get the ones with your small molecule sitting inside them at its binding site, but when you do, the payoffs can be huge.

Producing such things at the quality levels needed, though, is an absolute black art. There are trends (sometimes), better ideas about starting points (sometimes), rational modifications of the protein sequence to increase your chances (sometimes), but in the end, it still comes down to relentless empirical try-it-and-see. This latest paper (writeup by C&E News here) is looking at a well-known protein crystal (glucose isomerase) that is also well known to exist it two polymorphic forms (see that last post on the subject for more on polymorphs). What isn’t well known is why it forms these particular two types of crystals and how that happens. In virtually every case of polymorphism, we have little or no clue about such things, and I’m very glad to see clues starting to be unearthed.

What these authors have done is to start the crystallization process in ammonium sulfate solution and then freeze the solutions (liquid nitrogen) at various time points, imaging the crystals thus found by electron microscopy. And there are some surprises. At the twenty-second time point, there are already “nanorod” crystals forming, which are, on the average, two protein molecules wide and twelve molecules long. You are not going to get too much earlier in the crystal formation process than that. The concentration of the nanorods increases over time. However, their length does not: they seem stuck at this size. They do start to associate into fibrils as time goes on.

But there are other things happening as well. There are other species in there (trimer, tetramers, single-file chains), and after about thirty minutes you can also see faceted nanocrystals of the two known bulk polymorphs, one prismatic and rodlike and one rhombic. The smallest of each are in the 100 nM range along their shortest edges. Trying the experiment using PEG instead of ammonium sulfate (another known set of conditions for producing these protein crystals) shows that the useful concentration range for the additive is much narrower, but in both cases, at the low end of the range you get rhombic, and at the high end you get prismatic. With PEG, though, if you keep pushing it, you get no crystals at all, but rather an amorphous gel-like state. And in none of the PEG experiments did they ever observe the nanorods seen in ammonium sulfate.

Nature, C&E News

That gelation pathway seemed to involve fibril formation as well, which looked in the earlier experiments like it might be a precursor to formation of the rhombic crystals. Weirdly, the authors found that you can take some of that gel and use it to nucleate another protein solution, one that would otherwise form only the rhombic crystals, and form the prismatic crystals with it instead (which is a bizarre thought to me, and probably to a lot of other people who don’t crystallize things for a living). You actually see the crystals growing out of the surface of the lump of gelatinous protein gorp, as shown at right. This is order out of chaos, for sure; I’ll bet that the temperature of the solution at that interface is changing just to make the thermodynamics of all that loss of entropy work out.

Seeing the early stages of crystal formation like this allowed the authors to come up with all the plausible models of the nanorods and compare those to experimental data. One clearly matches better than the rest of them, so now we know what the earliest crystal formation contacts are, and can start to explain why. The preferred mode of contact buries a lot more negative charges in the protein surface, and the paper goes into detail about how this can explain the effects of sulfate ions in the solution versus PEG.

Overall, it appears that the prismatic crystals (from ammonium sulfate) are made up of bunches of nanorods, rather than growing molecule-by-molecule, which is also contrary to the mental pictures that many of us have. Meanwhile, the rhombic crystals grow more like the traditional picture, monomer by monomer. The nanorod form, though, can also be trapped in a metastable fibril state, starting off as individual fibrils and going on to a gel, with too high an activation energy to dissociate and form prismatic crystals. But its surface can nucleate the formation of such crystals (as in the above photo). At really high PEG concentrations you get a different gel, which is not composed of fibrils, and seems to be a complete dead end.

What they don’t see any evidence for, though, is another mode predicted by theory (and which I mentioned in the last post on this stuff): formation of phase-separated nanodroplets that then crystallize. Those nanodroplets take us back to the hot topic of protein condensates inside cells, and perhaps its a good thing that they’re not invariably precursors to forming protein crystals (!) That is very likely the reason that such condensates seem to feature intrinsically disordered proteins as well, to avoid the crystallization problem. . .


13 comments on “More on Crystal Formation (This Time With Proteins)”

  1. Chrispy says:

    “It’s especially tricky to get the ones with your small molecule sitting inside them at its binding site, but when you do, the payoffs can be huge.”

    This is dead wrong. It is often the case that enzymes are too floppy in the absence of an inhibitor to crystallize; this is part of the reason that industry is so much better at crystallizing drug targets than academia. They have inhibitors.

    Floppy parts are a problem in crystallography, and a lot of effort goes into chopping them off or cutting the protein down to domains. One trick is also to use antibody Fabs that bind your target (Fabs, not full antibodies, as full antibodies are composed of discrete domains with a lot of relative flexibility). I can think of several targets that would not crystallize without a bound Fab.

    From a drug discovery point of view, it is of course lovely to have a crystal structure with your inhibitor bound. From a practical standpoint, however, crystallography generally takes so long that by the time you have this structure in hand the chemists have already explored the SAR that the crystal structure would suggest. It does make for a good image in the publication at the end, though.

    1. halbax says:

      I believe that Derek was referring to GPCRs and other receptors, which are indeed difficult to crystallize when the binding site is occupied.

      Crystal structures are good for more than just serving as pretty pictures in publications. For example, with receptors, SAR tells you nothing about the binding mode, which residues are involved in binding, conformational differences between agonist and antagonist states, etc.

    2. Barry says:

      And because many binding sites are induced, we’ve had a better record of soaking novel inhibitors into crystals co-grown with a (known) inhibitor than into apo-crystals

  2. Bryan Roth says:

    “I believe that Derek was referring to GPCRs and other receptors, which are indeed difficult to crystallize when the binding site is occupied.”

    Sadly this is not true. It is nearly impossible to crystalize GPCRs without bound ligand.

    1. Manel says:

      I wonder if it would be easier to crystalise some GPCRs with an irreversible ligand bound to the protein. Would a ligand with a slow off-rate do the same job?

      1. Barry says:

        All integral membrane proteins are problematic for crystallography; GPCRs are just a subset thereof. It’s not that they’re not well-folded. But they’re organized to display a lipophilic surface to the phospholipid environment. Without that environment, they may oil out, or even refold to bury hydrophobic side-chains in their interior

        1. Barry says:

          “Membrane proteins represent between 20 and 30% of the proteomes of most organisms [1] and more than 40% of drug targets [2] and yet very few structures of these molecules have been solved by X-ray crystallography or NMR”
          “Membrane protein structural biology is still a largely unconquered area, given that approximately 25% of all proteins are membrane proteins and yet less than 150 unique structures are available. Membrane proteins have proven to be difficult to study owing to their partially hydrophobic surfaces, flexibility and lack of stability.”

  3. Colm says:

    I did a crystallization screen in my first graduate school rotation in addition to some cryoEM work. The latter worked fine, the former, well, I put up a big table of conditions on my rotation poster and then figure 1 was just a frowny face emoji. I did not go into structural biology.

  4. mause says:

    Here’s a related interesting bit – CSIRO’s Collaborative Crystallization Centre has had a paper published on the novel structure of platypus lactation protein, which has antimicrobial properties…

    CCC appears to do a lot of interesting work…

  5. Thoryke says:

    “…you can take some of that gel and use it to nucleate another protein solution, one that would otherwise form only the rhombic crystals, and form the prismatic crystals with it instead…”

    Is this what prions do?

    1. Derek Lowe says:

      Prions are forming more of an ugly aggregate, but the whole nucleation-into-a-lower-energy-state idea is definitely very similar to crystallization.

  6. Uncle Al says:

    Have folks tried cheating? If the target can be consistently dimerized with a short linker, the resulting C_2 symmetry is friendly toward crystallization. P2(1)2(1)2(1) has all sorts of volumetric forgiveness.

    1. Lab rat, now an ambitious pi says:

      Haha. You need to stop thinking and learn your students a lesson. You will never make millions thinking clearly like this. Dont tell them the truth, just tell them to get back into the lab. They are lab rats, they are not on our level. Whats with you bro? My rats are starting to ask questions because of posts like this.

Comments are closed.