Experiences with archived raw diffraction images data: capturing cisplatin after chemical conversion of carboplatin in high salt conditions for a protein crystal

Archiving of raw diffraction images data has led to new structural chemistry information being obtained for previously published results, which leads to the conclusion that carboplatin has partially converted to cisplatin in the high NaCl concentration conditions used in the crystallization procedure.


Introduction
The archiving of raw diffraction images data is the focus of an IUCr Diffraction Data Deposition Working Group (see http:// forums.iucr.org/). Experience in archiving and sharing of diffraction data in collaboration between Manchester and Utrecht Universities studying the binding of the important anti-cancer agents, cisplatin and carboplatin to histidine in a protein has recently been described in detail (Tanley et al., 2103a). Subsequently, these studies have been expanded when one of us (KD) joined in the detailed further analyses of each set of raw diffraction images with XDS (Kabsch, 1988). The motivation for this collaboration was due to the title of the previously published article (Tanley et al., 2013a) which suggested that a rather comprehensive approach was taken to compare software packages, with the implicit intention to correlate some metrics of data quality [R merge , I/, CC 1/2 , completeness and diffraction precision index (DPI) values, all as a function of resolution] with the amount of biological insight that these data allow one to obtain. Thus, a comprehensive investigation becomes even more useful with every addition that broadens its basis, which suggested contributing the processing with yet another widely used software package (XDS) on the existing raw diffraction data images. All of the raw diffraction images data can be found at http://rawdata. chem.uu.nl/#0001, except for 4g4c, which are at http://rawdata. chem.uu.nl/#0002, and now also mirrored at the Tardis Raw Data Archive in Australia (http://vera183.its.monash.edu.au/ experiment/view/40/). The raw data sets, being a part of the supplementary materials available with the previously published article (Tanley et al., 2013a), have also led to a Supplementary Materials button being added to the Journal of Applied Crystallography contents page for our article (doi:10.1107/S0021889812044172), the first of its kind. The raw diffraction data were measured on home X-ray sources at Manchester University but our study has implications for synchrotron radiation facilities, where a large fraction of macromolecular X-ray data these days are measured. That said, the lead investigator for any given synchrotron radiation beam time proposal presumably would hold the data management responsibilities of their relevant funding organization and would thus finally place the data for any given publication with their preferred raw data archive in addition to the formally required Protein Data Bank (PDB) depositions of processed and derived data. It is noted that various Universities in the UK are setting up research data archives for use by its research staff, and to serve all research fields, so as to satisfy the existing, and increasing, mandates of funding organizations. Specialist needs such as at CERN are being assumed by Universities as to be met centrally, i.e at CERN itself. Neutron facilities are also in parallel setting up or have set up raw data archives hosted by the neutron facility itself. Synchrotron radiation facilities have a variety of 'raw data retention policies' (see the IUCr Forum for details).
This study focuses on a direct comparison of processed diffraction and derived protein model data from XDS (Kabsch, 1988) with the results published previously (Tanley et al., 2013a). The possible issue of partial conversion of carboplatin to cisplatin under a high chloride salt concentration (Tanley et al., 2012a;Gust & Schnurr, 1999) has been taken up based on new evidence coming from the re-processing of our diffraction images as well as re-visiting the already published data and a detailed crystallographic assessment is provided.
Overall we give a mix of research results in a particular study (anti-cancer agents in their chemical behaviours and their binding to histidine in a protein) and also a short summary of developing raw data archives for diffraction data images in terms of policies and practicalities. We will also document the not insignificant challenge of making appropriate and detailed recording of the metadata for crystallography raw diffraction data.

Anti-cancer agents in their chemical behaviours and their binding to histidine in a protein
From previous studies (Tanley et al., 2012a,b) it is shown that carboplatin bound to the N and N" atoms of His-15 of HEWL in DMSO crystallization conditions. It was also shown that cisplatin did likewise but in addition it proved possible to show that its binding also took place over prolonged chemical exposure of around a year; carboplatin-based lysozyme crystals under aqueous conditions produced crystals too fragile after a year's storage to study successfully. Through the Utrecht archive, one of us (KD) downloaded the data sets diffraction images reanalysed here. Thus, using XDS to process these archived raw diffraction images (PDB ID 4dd7/ 4dd9/4g4c), clearly significant anomalous difference density peaks in the carboplatin binding site were seen; appropriately placed for where a Cl atom in cisplatin would be expected to be for the N binding site and for two Cl atoms in the N" binding site. Comparing these anomalous difference density peak heights with the peak heights from the EVAL (Schreurs et al., 2010), MOSFLM (Leslie, 1999) and PROTEUM/ D*Trek (Bruker, 2006;Pflugrath, 1999) processing programs (Tanley et al., 2013a) (Table 1), it was noted that for 4dd7, these three Cl atoms were mis-interpreted first of all by us (ST and JRH), with evidence of their presence coming from all four processing programs. Whereas, for 4dd9 and 4g4c, there is no evidence for all of these Cl atoms above our 3 cut-off level, with only one Cl in the N" binding site being present above the 3 cut-off level. This new finding, along with remodelling the EVAL deposited data, leads to the conclusion that the carboplatin could in fact have been partially or possibly even fully converted to cisplatin. This chemical conversion and the respective percentages were then evaluated for each binding site (Fig. 1). The partial transformation of carboplatin to cisplatin has been seen previously reported on in solution (Gust & Schnurr, 1999). In a 0.9% NaCl solution, 10% of the carboplatin had converted to cisplatin after 28 d storage at room temperature and the concentration of NaCl used in our crystallization conditions (10%) and the storage of our crystals between days and months at room temperature (Tanley et al., 2012a,b) Table 1 Anomalous difference density peak heights ( cut-off level) in the N and N" binding sites for Cl atoms for three different datasets using a number of different processing programs. LAPMAP (CCP4 package) was used to give the statistical correlation coefficient of which ligand could be bound at each binding site based on the 2F o À F c maps as well as SHELX (Sheldrick, 2008) being used to obtain the refined occupancies of the Pt atom centre at each binding site, and the occupancy values of each bound ligand remaining atoms at each binding site. This information, using both EVAL and XDS processing programs for each dataset (4dd7, 4dd9 and 4g4c), is given in Tables 2 and 3.

Anti-cancer agents in their chemical behaviours and their binding to histidine in a protein
Archiving our raw diffraction images enabled us to re-process them, resulting in biological insight going beyond that of our previously published results (Tanley et al., 2013a; http://rawdata. chem.uu.nl/#0001; http://vera183.its. monash.edu.au/experiment/view/40/). It is now thought that instead of carboplatin binding alone to HEWL in DMSO media (Tanley et al., 2012a,b) we actually see a mixture of carboplatin and cisplatin in the binding sites due to the partial conversion of carboplatin to cisplatin in the high salt conditions (10% NaCl) used for the crystallization procedure. Table 1 shows the anomalous difference density peak heights in the binding sites, attributed to Cl atoms for different software programs. These confirm that, for 4dd7, anomalous difference density for these Cl atoms are seen and should have been noted previously and, also, this dataset shows the most conversion of carboplatin to cisplatin based on the correlation coefficients out of OVER-LAPMAP (Table 2). However, we had set up clear chemical conditions and carboplatin only naturally was what we, wrongly as it turns out, expected. Whereas for 4dd9 and 4g4c, the anomalous difference density peaks are weaker, and thus harder to determine that the Cl atoms are in fact present. From the anomalous difference density peak heights, along with correlation coefficient of cisplatin/carboplatin binding to the 2F o À F c maps in OVERLAPMAP as well as the refined occupancy values from SHELX (Tables  1-3), it can be noted that carboplatin has been partially converted to cisplatin in the high salt conditions used for the crystallization procedure. Also, the correlation coefficient values from Table 2 for each ligand as well as the occupancy values for each ligand in Table 3 generally correlate very well between the XDS and EVAL refined models. Based on the new finding that carboplatin has partially converted to cisplatin in the high NaCl salt content used for the crystallization conditions, a new study, involving crystallization of HEWL with carboplatin in non-NaCl conditions, is underway (Tanley et al.,  2013b). The two binding sites on the His-15 residue of HEWL. F o À F c density OMIT maps (green) and anomalous difference density (orange) maps at the 3 cut-off level for (a) and (b) 4dd7 processed by XDS and EVAL, (c) and (d) 4dd9 processed by XDS and EVAL and (e) and ( f ) 4g4c processed by XDS and EVAL, respectively. The N binding site is on the left-hand side and the N" binding site is on the right-hand side. The arrows show where the possible Cl atom locations are based on anomalous difference density being seen. Table 2 Correlation coefficient of cisplatin or carboplatin binding in the 2F o À F c map for both the N and N" atoms of His-15 using OVERLAPMAP in CCP4i, with the Pt centre removed, i.e the calculation is just for the remaining bound ligands.
Note that in this calculation each ligand is considered separately whereas in the SHELX calculation summarized in Table 3  Archiving of raw diffraction data images policy and practicalities for crystallography is at its early stages (see the IUCr Forum documents); very reasonably the issues involve costs versus benefits but within a fairly rapid expansion of technical options such as University archives becoming established and cloud storage commercial options becoming more commonly available (Helliwell et al., 2012). In Tanley et al. (2013a) the very pragmatic method of using a personal web link held at Utrecht University for our various raw data sets was used. From the results shown here, without this web link archiving, we would not have re-examined the previous diffraction data (Tanley et al., 2012a(Tanley et al., ,b, 2013a and thus would have missed this most interesting chemical effect of partial conversion of carboplatin to cisplatin. Providing sufficiently rich metadata to fully describe raw diffraction data is important for carrying out data processing of diffraction images with subsequent software. One very specific point seen in this study was that XDS required manual input of the rotation axis direction and the direction of the detector X/Y axes for Bruker CCD diffractometer data, whereas, with the EVAL software, they were readily interpreted from the header information of the diffraction images. Our previously published work (Tanley et al., 2013a) presented a detailed and rich description of metadata for two commercial area detector diffractometers and is thus, we hope, an exemplar of what detail is required.

Conclusions
The studies outlined here focused on the advantages of archiving the raw diffraction images for X-ray crystallography, via a relatively simple personal web link method rather than, say, a formal single University or centralized raw data archive. Due to archiving our raw diffraction images data, new findings have been achieved owing to re-processing the images with XDS as well as re-examining the previously published data. Thus, we now see carboplatin partially converting to cisplatin in the crystal structure due to the high NaCl salt concentrations used in the crystallization conditions. This has then led us to try to find new crystallization conditions without NaCl to seek to capture carboplatin binding on its own.
We are grateful for research support from the Universities of Konstanz, Manchester and Utrecht. ST is funded under an EPSRC PhD Research Studentship. EPSRC requires data archiving at a level appropriate to the relevant research community data deposition policies. These policies are in effect for our discipline set by the IUCr and who is reviewing its policies via the Diffraction Data Deposition Working Group; this is Chaired by JRH and who warmly acknowledges many discussions with Brian McMahon and Tom Terwilliger as well as LK-B, all members of the DDD WG and its consultants. The role of Professor Dr Sine Larsen in establishing the DDD WG, when she was President of the IUCr, was obviously a pivotal leadership in these important matters and who we also heartily thank.