a, The ‘MESH-on-a-stick’ sample injector configuration (see Methods). In the three panels, the yellow X below the capillary indicates the X-ray path into the page. The middle panel shows a closer view of the injector tip; the right panel shows an on-axis view of the sample injecting during the experiment. b, Our choice of X-ray wavelength for diffraction and MIRAS phasing was a compromise between maximizing the heavy atom anomalous signals, f″, as indicated by the curves for each element, and maximizing the number of data sets collected in the time allotted for the experiment. The grey bar corresponds to the wavelength we used, 1.41 Å. c, Difference Patterson maps calculated at 2.8 Å resolution. Sharpening (−5.0 Å2) was applied to Hg and VIL maps. Coefficients for the PCMBS and VIL maps were obtained from both isomorphous and anomalous differences. The Gd difference Patterson map was calculated from anomalous differences only. Contours start at the 1.5σ level and continue at 0.5σ intervals. Peaks corresponding to vectors between heavy atoms stand out as high peaks, up to 7.5σ. d, Heavy-atom sites were located successfully for each of the three derivatives using the program SHELXD. We compared the quality of potential heavy-atom substructure solutions obtained from two sources of heavy-atom signal: single wavelength anomalous dispersion (SAD, red) and a combination of anomalous dispersion and isomorphous differences (SIRAS, blue). Ten-thousand independent trials were performed for each derivative and signal source. Each dot in the scatter plots indicates the quality of an individual substructure solution. The vertical axis, labelled CCall, indicates the consistency between the potential solution and the diffraction data as the correlation coefficient between normalized structure factors, Ecalc and Eobs. The horizontal axis, labelled PATFOM (Patterson figure of merit), indicates the consistency between the observed difference Patterson map and that predicted by the potential solution. Successful substructure determination is suggested by the appearance of a sharp separation between two populations of potential solutions: a cluster with lower values of CCall and PATFOM (incorrect solutions) and a cluster with higher values (correct solutions). Such is the case for all the trials performed, except for VIL using the SAD signal, where only a single population of solutions is observed. Evidently, the SAD signal was insufficient for accurate location of iodine sites. For VIL, we relied on the accuracy of sites obtained from the SIRAS signal. In most cases, the SIRAS (blue) signal is stronger than the SAD signal (red), indicating good isomorphism between native and derivative data sets. Only in the case of GdCl3 does the SAD signal appear to be better than the SIRAS signal. The histograms in the right column indicate the number of potential substructure solutions with given values of CFOM (combined figure of merit). The histograms recapitulate the trends observed in the scatter plots. e, The correlation coefficient (CCiso) measures the agreement and Riso measures the discrepancy between the native structure factors and those of each of the derivatives. Each of our three derivative data sets shows isomorphism with the native data set up to 2.8 Å resolution.