Beamline Commander: Crystal Hunter

A nod to Silent Service (Sid Meier, MicroProse, 1985; on the ZX Spectrum from U.S. Gold), the submarine game where you ran on batteries under water, surfaced to recharge with the diesels and gave yourself away, hunted ships through the periscope, and went deep and quiet when the destroyers came over with their depth charges. Here the submarine is I04, and you’re a structural biologist whose group has been given a shift on it. An open hutch is “surfaced”: safe, but no X-rays. Beam dumps, heat bumps and orbit wobbles are the depth charges: close the shutter and wait quietly. The dose is the depth gauge: as it nears 30 MGy the crystal creaks, and past 30 MGy it gives way. Setting the exposure instead of the dose is navigating blind. When something breaks, you phone the beamline scientist on call. There’s a library too: the right books help for the rest of the shift; the novels are lovely, but not on beamtime. And the convoy you hunt is a drop full of crystals, only some of which will give you the structure.

The science

• The X-rays. A synchrotron such as Diamond keeps electrons circling in a storage ring at nearly the speed of light. In an undulator, rows of magnets wiggle them and they give off an intense beam of X-rays; a double-crystal monochromator, two cooled silicon crystals, picks one energy from it.
• At the sample. Compound refractive lenses set the beam size, a goniometer turns the crystal, a stream of nitrogen gas at about 100 K keeps it frozen, and a robot swaps crystals from a dewar of liquid nitrogen. A photon-counting detector records the diffraction images.
• Harvesting. Protein crystals are scooped out of their drop in a tiny nylon loop and plunged into liquid nitrogen. Freeze fast and the water stays glassy; freeze slowly, or let the drop dry, and it forms crystalline ice, which puts rings on the images at about 3.9, 3.7 and 3.4 Å and hides the protein’s spots. Cracks and knocks raise the mosaicity, and clusters of crystals give two lattices at once.
• Dose. The X-rays that diffract come with many more that are absorbed and damage the crystal. The dose is measured in grays, and for frozen protein crystals about 30 MGy is the experimental limit, set by Elspeth Garman’s group (Owen et al., 2006). So you screen with a little dose first (here a 15° wedge at 1 MGy) and plan the rest. The dose that suits a dataset depends on the resolution the crystal diffracts to: at I04, about 5.5, 10 and 16 MGy for 1.3, 1.8 and 2.3 Å, and never more than 30 MGy (anomalous data stop at 10 MGy). Misjudge the resolution by 0.25 Å and the right dose shifts by about 30 %. So at I04 you set a target dose and the exposure is worked out from the beam (RADDOSE-3D does the sums): at the default beam, 10 MGy takes about 16 s of total exposure at full transmission (13 keV). And you collect a full 360°: every spot measured several times.
• Too bright to count. Photon-counting detectors can miss photons that arrive too close together. Their correction works up to a limit, the trusted range; past it, the strongest spots are under-counted, and because all their copies are under-counted alike, the merging statistics still look excellent (Donath et al., 2023). The fix, which we teach in our training at I04: check the brightest spots from the screening wedge, lower the transmission and lengthen the exposure by the same factor. Same dose, every photon counted. You can try the interactive model yourself (eiger-countrate-dose-demo).
• Rods and needles. Damage is local: on a long crystal, a helical scan moves the crystal along the beam as it turns, so each part gets only its share of the dose. The beam’s width sets how many fresh places fit along the rod.
• Symmetry. The space group says how a crystal repeats. A symmetric crystal gives copies of each spot, so a short rotation is enough; a triclinic P 1 crystal has none, needs a full turn and, because one rotation axis always leaves a blind region, a second sweep at another tilt. A huge unit cell, like a virus capsid’s, packs the spots close together.
• Solving. The data are processed automatically; molecular replacement places a known, similar model into the new crystal, and the atoms are built into the electron density. At an absorption edge (7.112 keV for iron), atoms scatter a little differently (anomalous scattering), which shows where they are. The differences are small, so at I04 anomalous data are collected gently and from several directions: 10 MGy in all, as three 360° wedges at different χ orientations of the goniometer.
• The four structures.
  • BCL-2 keeps cells alive, and some leukaemia cells make far too much of it. Venetoclax, a drug that grew out of research at WEHI in Melbourne, blocks it so the cancer cells die. Peter Czabotar’s group solved BCL-2 with venetoclax to see why the drug stops working for some patients (Birkinshaw et al., 2019).
  • The SARS-CoV-2 main protease cuts the virus’s proteins into working parts; nirmatrelvir blocks it. The structures with an empty active site (see the structure page) were the starting point for a large fragment screen (Douangamath et al., 2020).
  • Beak and feather disease virus threatens endangered parrots such as the orange-bellied parrot. With Jade Forwood’s group we solved structures of its capsid, a starting point for a vaccine (structure page).
  • The hybrid cluster protein from Desulfovibrio holds an unusual cluster of iron, sulfur and oxygen; the reduced protein from D. desulfuricans crystallised in P 1, and it was part of my PhD (structure page).

Where the game cuts corners

• Beamtime moves in jumps: a hutch search, a mount or a dataset each take a fixed number of minutes, and model building takes 40 of them. Real model building takes days, sometimes months, and a lot more than a spin and a nudge.
• Storage rings are very steady: feedback keeps the beam in place to within micrometres, and when something does go wrong the beamline closes its own shutter. Going quiet is pure submarine.
• Robot dewars are usually topped up automatically, and real pucks hold 16 crystals, not 4.
• The count-rate check really uses a high percentile of the pixel counts in the strong spots, compared with the trusted range stored in the data file; the game boils each crystal down to one brightness number.
• In the game, the dose rate simply goes up with the square of the beam getting smaller, and spots stop overlapping once the detector distance in millimetres passes the cell edge in ångströms. Real lenses, beams and spots are more complicated, which is why there’s RADDOSE-3D and a great deal of geometry.
• One sweep of a P 1 crystal often gives better than 90 % completeness; the game makes the blind region bigger, so the second sweep matters.
• Real faults are rarer than in the game, and the beamline scientist on call can’t always fix things in a quarter of an hour. But they do pick up the phone in the middle of the night.
• Reading Crystallography for Dummies in 40 minutes doesn’t make anyone’s hands steadier at the microscope. Practice does (and the books help too).
• Coffee doesn’t speed up anyone’s hands by exactly the same amount with every cup. Probably.