The copper Kα wavelength (λ = 1.54 Å) that our home source produces is excellent for phasing structures by single-wavelength anomalous dispersion with the anomalous signal from sulfur (S-SAD).

If your crystal diffracts to 2.5 Å data, these is at least one sulfur per 25 residues and you collect highly redundant data, the structure can be solved easily.  Everything else being the same, more redundancy is better.  I've solved a structure from a dataset with a redundancy of 13, but failed to solve another when the redundancy was 20.

## Concept

Using sulfur for phasing obviates the need to incorporate anomalous scatterers like selenium, iodine or heavy metals.  This means less work for you.  However, sulfur is a weak scatterer.  Highly redundant data are needed for the signal to rise above the noise.  This can be achieved by blending data (see paper) from a large number of crystals (see paper 1 and paper 2) or by making use of a multi-angle goniometer head like the partial chi that we have (described below).

Traditionally, it was thought diffraction to at least 2.5 Å was required to solve a structure by S-SAD, but the method has recently been shown to work with data below 4 Å, with anomalous signal extending to no more 6.5 Å (see paper).

The same process that's described below also applies at the synchrotron, though it can be hard there to get enough data without grilling the crystal.  At the home source you're faster anyway because you can start right now.

## Process

With good data, solving a structure by experimental phasing is a straightforward process.

1. Grow some nice crystals.  I chose lysozyme, but the requirements of your project may vary.
2. Collect data overnight.  I used a simple strategy of two 90° sweeps 180° apart for each χ value (divisible by 10) between 0° and 60°.  Each of the 14 sweeps comprises 300 images (0.3° each, 1 s exposure).

sweepχω0 (φ = 0°)ω0 (φ = 180°)
a, n -40° -40°
b, m 10° -40° -40°
c, l 20° -40° -40°
d, k 30° -40° -40°
e, j 40° -40° -40°
f, i 50° -40° -40°
g, h 60° -40° -40°

The order of collection was a, b, c, etc.  It would have been better to collect a, n, b, m, etc to emulate an inverse-beam strategy, but it was good enough. (In fact, the first two sets, a and b, were sufficient to solve the structure.)
I also collected a high-resolution sweep for phase extention (600 images, d = 1.7 Å, χ = 0°, 2θ = 10°) at the end.  Overall, the entire dataset (1260°) took about 11 hours to collect.
Note that at χ = 60°, the pin can cast a shadow on the detector.  Inspect the images before processing them and exclude those that are flawed.
3. Process data. I used iMosflm for each sweep and then scaled them together with aimless.  The statistics were very good.

overallouter shell
Resolution 24.6-1.9 Å 1.94-1.9 Å
I/σ(I) 80 32
R merge 0.07 0.21
R pim 0.007 0.023
completeness 100% 100%
multiplicity 97x 80x

In theory you could also use HKL-3000 (see paper), which would make the process a bit smoother, but the version we have doesn't like to scale large datasets (beyond 3000 frames).
4. Phasing in ShelxC/D/E.  The mtz file produced by aimless must be converted to sca.  I used mtz2various for that.
The sca file is then used as input in hkl2map, a graphical interface for ShelxC/D/E.  Below are some screenshots to show you what successful phasing looks like.

### ShelxC

Anomalous signal with d"/sig > 1.0 throughout

### ShelxD

Correct solutions with higher CCs than the bulk

### ShelxE

Correct hand assigned quickly

Autotracing clear after four cycles

Notice how the default number of autotracing cycles in hkl2map (3) is not enough to build the model.

The whole Shelx procedure takes less than ten minutes. You'll get a pdb file with the Cα trace, a phs file with phases and a shell script to call f2mtz to obtain an mtz file for refinement.

## Conclusion

Solving a structure by S-SAD from data collected on the home source is simple.  It should be a routine consideration for all new crystals.

You might argue that lysozyme is not representative of real-world problems.  To this I retort:

• Lysozyme is not particularly good for S-SAD.  The solvent content of tetragonal crystals is only 39%.  Density modification, used extensively by ShelxE, has little effect.  There's also no NCS.
• I used minimal exposure.  If your crystals are weaker, expose longer.  The overall data collection time is of little concern when you collect overnight.  Radiation damage is unlikely to be a problem.
• You might think your crystal diffracts to less than 3 Å.  However, high reduncancy increases signal to noise.  You'll get higher I/σ(I) throughout, which increases the effective resolution of your data.
• I didn't put much thinking into this.  I chose a large number of sulfurs and ignored disulfide bonds.  The more you know about your protein, the better you can tune ShelxD/E and the more likely you are to get phases even from ma rginal data.