Online ion-exchange chromatography for small-angle X-ray scattering

SAXS coupled with online ion-exchange chromatography allows the collection of high-quality BioSAXS data.


Introduction
Biological small-angle X-ray scattering (BioSAXS) can reveal solution structures in terms of the average particle size and shape of biological macromolecules, as well as information on the surface-to-volume ratio (Graewert & Svergun, 2013;Kikhney & Svergun, 2015;Putnam et al., 2007;Jacques & Trewhella, 2010). This method is an accurate, mostly nondestructive approach, which requires little sample preparation compared with that required for other structural biology techniques, such as crystallography. However, for accurate interpretation the sample is required to be monodisperse; this requirement is often a problem as biological macromolecules can be susceptible to aggregation. Recently, the combination of online size-exclusion chromatography (SEC) and SAXS has been implemented directly on beamlines in order to overcome this obstacle, to ensure data quality and to make this technique more accessible for increasingly difficult samples (Lambright et al., 2013;Round et al., 2013;David & Pé rez, 2009;Graewert et al., 2015;Mathew et al., 2004;Watanabe & Inoko, 2009;Acerbo et al., 2015;Grant et al., 2011). Additional techniques, such as time-resolved (TR) SAXS experiments using desalting columns for quick buffer exchanges (Jensen et al., 2010) and differential ultracentrifugation, have been successfully coupled with SAXS (Hynson et al., 2015). However, even when using SEC-SAXS, samples can still be difficult to measure, for example when different species do not separate. This can be owing to their size (the difference in molecular mass should be at least 10%), owing to the limited resolution range of the SEC column or owing to the physical properties of the sample such as hydrophobic surfaces, flexibility or lack of stability, e.g. separation of complexes into individual protomers. In these cases, data collection, analysis and interpretation can be difficult.
Another common problem is the quantity of sample needed. Many proteins are difficult to express or purify and the required quantity (about 100 ml at 3 mg ml À1 ) of monodisperse sample cannot be obtained. For SEC-SAXS measurements, the high degree of dilution on the column [up to a factor of ten depending on the column type and sample (Watanabe & Inoko, 2009;Kirby et al., 2013)] requires even more concentrated samples (5 mg ml À1 or above), although depending on the column less volume (down to 10 ml) might be sufficient. Even if the samples are soluble at these concentrations, they are not always stable and can often aggregate.
An alternative purification method is ion-exchange chromatography (IEC). IEC separates ionized molecules based on their total net surface charge, which changes gradually with the pH and/or the salt concentration of the buffer (Karlsson & Hirsh, 2011). This technique enables the separation of molecules of similar size, which can be difficult to separate by other techniques. In general, if the net surface charge of a protein is higher (positive or negative) than that of the IEC resin (anion or cation exchange, respectively), the protein will bind to it. When using a linear salt gradient to increase the charge in the mobile phase, a specific protein will be eluted at a specific salt concentration. Given that different oligomeric states and aggregates differ in their surface area, and hence their surfacecharge distribution, it is also possible to separate oligomeric species (Kluters et al., 2015).
IEC has the advantage of providing moderate resolution and, when working with large volumes of dilute samples, concentrating the sample, since the eluate concentration is determined by the capacity of the column and the binding affinity of protein to the column, and not by the initial concentration. Therefore, it is possible to store and transport samples at low concentrations prior to the IEC experiment. Additionally, many IEC columns perform well at flow rates in the millilitre per minute range, which limits the risk of collecting data from radiation-damaged material. However, IEC does require some optimization regarding the pH and the salt gradient (Yigzaw et al., 2009), which can and should be performed prior to any SAXS-coupled data collection.
Here, we present a proof of principle for an alternative to SEC-SAXS measurements using ion-exchange chromatography (Selkirk, 2004) online with the SAXS experiment. This has been implemented and tested on the BioSAXS beamline BM29 at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France (Pernot et al., 2013).

Purification of BSA on an ion-exchange column
For the preparative offline IEC experiment, 100 ml of a 77 mg ml À1 BSA (lyophilized powder, essentially globulinfree; Sigma) solution in buffer A [20 mM Tris pH 7, 25 mM NaCl, 5% glycerol, 1 mM dithiothreitol (DTT)] was prepared and injected onto a Uno Q-1R (Bio-Rad) column on a Biologic system (Bio-Rad) equilibrated with buffer A. The protein concentration was determined by measuring the absorption at 280 nm with a NanoDrop (NanoDrop 1000, Thermo Fisher) using a mass extinction coefficient of 6.7 for a 1% (10 mg ml À1 ) BSA solution. The high concentration enables easy injection onto the column using injection loops. A salt gradient was made by mixing buffer A with buffer B (20 mM Tris-HCl pH 7, 1 M NaCl, 5% glycerol, 1 mM DTT). The flow rate for the offline experiment was 2 ml min À1 .
For IEC-SAXS experiments, only 50 ml was injected using the autosampler of the HPLC system (SIL-20ACXR) and the flow rate was 1 ml min À1 .
Peak fractions from the preparative run (20 ml per 1500 ml fraction size) were analysed on a 12% SDS-PAGE stained with InstantBlue (Expedeon). PageRuler Plus Prestained Protein Ladder (Thermo Fisher) was used as a molecularweight marker.

D5 323-785 cloning, expression and purification
A fragment of the helicase-primase D5R representing the D5N and helicase domains, D5 323-785 , was cloned, expressed and purified as described in Hutin et al. (2016). The construct was cloned into the pProEx HTb vector (Life Technologies) using the primers 5 0 -GCGCCATGGGTAATAAACTGTTT-AATATTGCAC-3 0 and 5 0 -ATGCAAGCTTTTACGGAGA TGAAATATCCTCTATGA-3 0 and expressed in Escherichia coli BL21 (DE3) Star cells (Novagen). The bacterial pellet was resuspended in lysis buffer (50 mM Tris-HCl pH 7, 150 mM NaCl, 5 mM MgCl 2 , 10 mM -mercaptoethanol, 10% glycerol) with cOmplete protease-inhibitor cocktail (Roche) and 1 ml benzonase per 10 ml and lysed by sonication. The supernatant was loaded onto a nickel-affinity column (HIS-Select; Sigma), which was washed with 10 column volumes (CV) of lysis buffer, 10 CV washing buffer (50 mM Tris-HCl pH 7, 1 M NaCl, 10 mM -mercaptoethanol, 10% glycerol) and 10 CV imidazole wash (50 mM Tris-HCl pH 7, 150 mM NaCl, 10 mM -mercaptoethanol, 10% glycerol, 30 mM imidazole). D5 323-785 was eluted in 20 mM Tris-HCl pH 7, 150 mM NaCl, 10 mM -mercaptoethanol, 10% glycerol, 200 mM imidazole and the buffer was exchanged back to the lysis buffer on an Econo-Pac 10DG Desalting column. His-TEV cleavage was performed at room temperature overnight and the cleaved protein was passed over a second nickel column before injection onto a Superose 6 column (GE Healthcare) equilibrated with gel-filtration buffer (20 mM Tris-HCl pH 7, 150 mM NaCl, 10% glycerol, 1 mM DTT). The eluted peak fractions of D5 323-785 were then combined and diluted to 25 mM NaCl and 5% glycerol by keeping the buffer concentration equal. 30 ml of the sample, containing about 5 mg of protein, were then loaded onto a Uno Q-1R column (Bio-Rad; buffer A, 20 mM Tris pH 7, 25 mM NaCl, 5% glycerol, 1 mM DTT; buffer B, 20 mM Tris pH 7, 1 M NaCl, 5% glycerol, 1 mM DTT). The protein concentration was estimated by measuring the absorption at 280 nm with a NanoDrop (NanoDrop 1000, Thermo Fisher) using a mass extinction coefficient of 8.38 for a 1% (10 mg ml À1 ) D5 323-785 solution. For offline IEC the flow rate used was 2 ml min À1 , which was reduced to 1 ml min À1 in the online experiments. Glycerol was added to all buffers as a co-solvent to enhance the stability of D5 323-785 in aqueous solution in order to prevent protein aggregation.
As for BSA, fractions from the preparative run were analysed by SDS-PAGE.

IEC-SAXS data collection
SAXS data were collected on BM29 at ESRF (Pernot et al., 2013) using a PILATUS 1M detector (Dectris) at a distance of 2.864 m from the 1.8 mm diameter flowthrough capillary. The scattering of pure water was used to calibrate the intensity to absolute units (Orthaber et al., 2000). The intensities were scaled such that the forward scattering corresponds directly to the concentration (in mg ml À1 ) times the molar mass (in kDa) of idealized proteins, i.e. 1 a.u. = 8.03 Â 10 À4 cm À1 , unless explicitly stated otherwise. Data collection was performed continuously throughout the chromatography run at a frame rate of 1 Hz. The X-ray energy was 12.5 keV and the accessible q-range was 0.032-4.9 nm À1 . The incoming flux at the sample position was of the order of 5 Â 10 11 photons s À1 in 700 Â 700 mm. A summary of the acquisition parameters is given in Table 1. All images were automatically azimuthally averaged with pyFAI (Ashiotis et al., 2015;Kieffer & Wright, 2013;Kieffer & Karkoulis, 2013).
Online purification was performed with a high-pressure liquid-chromatography (HPLC) system (Shimadzu, France) consisting of an inline degasser (DGU-20A5R), a binary pump (LC-20ADXR), a valve for buffer selection and gradients, an auto-sampler (SIL-20ACXR), a UV-Vis array spectrophotometer (SPD-M20A) and a conductimeter (CDD-10AVP). The HPLC system was directly coupled to the flowthrough capillary of the SAXS exposure unit . The flow rate for all online experiments was 1 ml min À1 , resulting in a mean passage time of material through the X-ray beam of 0.1 s.
Initial evaluation of the data quality was performed using the automatic SEC-SAXS processing pipeline available at the BM29 beamline (Brennich et al., 2016;De Maria Antolinos et al., 2015).

Background subtraction for BSA on a linear gradient
In order to subtract an appropriate background, the method developed by Hynson and coworkers for differential ultracentrifugation (Hynson et al., 2015) was applied. Similarly to the salt gradient in IEC-SAXS used here, the background signal in differential ultracentrifugation-coupled SAXS changes owing to a sucrose gradient. The backgroundcorrection method of Hynson and coworkers identifies the appropriate background subtraction as that which provides a stable SAXS signal throughout the peak. The stability of the signal can be assessed by comparing the ratio of scattering in a low-q region to that in a mid-q region: for a stable signal this ratio is constant, whereas for an unstable signal it changes throughout the peak. Incorrect background correction results in a systematic change in the scattering signal depending on the protein concentration. The choice of the regions depends to some extent on the protein of interest and in particular on its size. However, we found that the regions used by Hynson and coworkers (0.11-0.5 and 1.5-2.5 nm À1 for low-q and mid-q regions, respectively) suited well: the low-q region reflects the correct overall size of the protein, whereas most of the characteristic features of BSA are present in the mid-q region (see, for example, Fig. 5d).
In Hynson et al. (2015) the offset between the individual sample and buffer runs is first determined by matching the high-q region of the scattering profile. This shift is then fine tuned by assessing the variability of the low-q to mid-q ratio for a fine grid of interpolated buffers.
In IEC-SAXS experiments, the overall change as well as the difference between individual frames in the background signal throughout the gradient and the offset between the buffer and sample measurements are much smaller than in the sucrose gradient used in the ultracentrifugation experiment [compare the bottom part of Fig. 1(b) in this paper with Fig. 2(a) in Hynson et al. (2015)]. As a consequence, the approach can be slightly simplified: the optimal shift between the recorded frames can be determined directly without the  Table 1 Parameters of SAXS data acquisition and analysis.
68.2 Calculated monomeric M r from sequence (kDa) 66.5 Structural parameters for BSA, step gradient I 0 (from Guinier) (arbitrary units) 68.2 Calculated monomeric M r from sequence (kDa) 66.5 Structural parameters for D5 323-785 , step gradient 338 Calculated monomeric M r from sequence (kDa) 321 need for interpolation. To achieve this, we shift the buffer run with respect to the sample run in steps of five frames and subtract frame-wise. The low-q to mid-q ratio is then calculated for each frame and the variability in the region of interest is assessed. The perfect shift would result in a parallel line in the ratio versus frame-number plot throughout the region, whereas systematic under-subtraction results in a concave shape and over-subtraction in a convex shape.
Using this approach, we can determine the best shift with an error of ten frames. This might seem less accurate in comparison with the subframe precision of Hynson and coworkers, but the difference between individual buffer frames in an IEC gradient is much lower than in differential ultracentrifugation [a CORMAP test  on 20 frames in the region of interest showed no significant deviation]. This implies that the effect of these ten frames on the subtraction is small.

SAXS data analysis
To compare the scattering from different frames, the ratio of scattering in the low-q region (0.11-0.5 nm À1 ) to the high-q region (1.5-2.5 nm À1 ) was calculated in the same manner as in Hynson et al. (2015). For easier comparison between different buffer subtractions, the result was normalized to give the same mean over the region of interest. Radii of gyration were calculated using AUTORG from the ATSAS package (Petoukhov et al., 2007), P(r) functions were calculated using GNOM (Svergun, 1992) and Porod volumes were estimated using the Volume interface of SCÅ TTER (available at http:// www.bioisis.net/tutorial/9). The estimation of molecular mass based on the Porod analysis was performed by dividing the Porod volume by 1.7 nm 3 kDa À1 (Petoukhov et al., 2007). All other manipulations were performed in the Spyder interface for Python 3.4 using the NumPy and SciPy packages (Jones et al., 2001). The scripts used are available at https://github.com/ maaeli/IEC. As the background scattering is dependent on the buffer composition, the gradient or stepwise buffer changes during elution must be accounted for in the subtraction process. We present three methods to find the optimal background subtractions, which are explained for the individual cases in x3.
Known BSA crystal structures were compared with the experimental data using CRYSOL (Svergun et al., 1995). Default fitting parameters were used, which allow the hydration shell to be adjusted to optimize the fit.

Results and discussion
3.1. IEC-SAXS of bovine serum albumin using a linear salt gradient The first test case was bovine serum albumin (BSA), using a standard linear salt gradient for elution and a blank run of the same gradient for buffer subtraction. We chose BSA as it is known to form higher oligomers and aggregates in solution (Folta-Stogniew & Williams, 1999), allowing us to assess the ability of the method to reduce sample polydispersity.
We performed a standard BSA run on the ion-exchange column using an offline FLPC system, evaluated the chromatogram and observed several peaks in the first third of the chromatogram (Fig. 1a). Their approximate buffer B percentages were determined to be 10, 12 and 16%, corresponding to NaCl concentrations of 122.5, 142 and 181 mM, respectively. SDS-PAGE of different fractions throughout the peak confirmed that the first two peaks correspond to BSA, while the third peak is a higher molecular-mass contaminant (Fig. 2a).  (d) Comparison of the scattering ratio (0.11-0.5 nm À1 versus 1.5-2.5 nm À1 ) for background correction using a shift of 100 frames (cyan), 130 frames (blue) and 160 frames (green) between the sample and buffer runs. The dotted black line serves as a visual aid for a constant line.
We then ran the same sample online on the BM29 beamline at ESRF, as described in x2, and continuously acquired SAXS data from the eluent at a rate of 1 Hz.
In order to improve the resolution of the peaks and to ensure a sufficiently high sampling rate, the flow rate for the online experiments was reduced to 1 ml min À1 . The UV absorption signal at 280 nm and the total scattering elution profiles show the expected three peaks (Fig. 1b, top), whereas the scattering at higher angles increases linearly (Fig. 1b,  bottom). The matching background was found to be slightly shifted with respect to the buffer run (compare the bottom graph in Fig. 1b with Fig. 1c). It is therefore necessary to subtract not the directly corresponding frames but slightly shifted ones. Possible reasons for this shift are a nonperfect synchronization between the ion-exchange run and the SAXS data acquisition or the co-elution of ions bound to the column. To identify the optimal shift, as described in x2, we applied the approach developed by Hynson et al. (2015) for differential ultracentrifugation coupled to SAXS. For each shift the ratio of scattering at low q (0.11-0.5 nm À1 ) versus mid-to-high q (1.5-2.5 nm À1 ) is calculated for each frame. If the sample scatters in the same way throughout the peak, this value should not change and a 'constant' value over the peak indicates a suitable background subtraction. Fig. 1(d) shows how the ratio changes in the region of interest for different values of the shift. For the first two peaks the ideal shift was found to be 130 AE 10 frames, which corresponds to about 2.2 ml. In addition, the ratio was constant over both peaks, indicating that one species elutes in a double peak. However, for the third peak no shift giving a flat line was found, indicating that the scattering from the sample is not constant throughout the peak and that the peak therefore represents more than one species. The SDS-PAGE from the offline run shows that in this peak BSA is contaminated by a larger protein species (Fig.  2a). The forward scattering and the radius of gyration obtained using AUTORG (Petoukhov et al., 2007) were calculated from subtracted curves and the mass was estimated via the correlated volume approach (Rambo & Tainer, 2013). Fig. 3(a) confirms that, as expected for a single species, both the mass and the radius of gyration are constant throughout the initial double peak. Based on the forward scattering intensity (Mylonas & Svergun, 2007), and assuming monomeric BSA (66.5 kDa), the maximum protein concentration throughout the peak was estimated at 1.75 mg ml À1 , which is well below the regime in which significant structure-factor contributions to the signal would become relevant (Skou et al., 2014;Zhang et al., 2007). An additional comparison of individual frames throughout the double peak shows that the SAXS signal is identical throughout the peak, with the lowest     Open symbols represent points that were not used for fitting. The lower panel shows the residuals. ( f ) Differences in scattering between the curve found by shifting the buffer run by 130 frames and a clearly under-subtracted curve (120 frames shift, orange) and over-subtracted curve (160 frames shift, blue). adjusted p-value in a CORMAP test  being 0.16 (Figs. 3b and 3c). This suggests that the two subpeaks do not represent different conformations of BSA. 169 frames were averaged in the region of interest shown (grey) in Fig.  3(a). The resulting curve (Fig. 3d) gives a radius of gyration of 2.7 AE 0.1 nm (Fig. 3e) and a Porod volume of 116 AE 5 nm 3 , corresponding to a molecular weight of about 68 kDa (Petoukhov et al., 2007; see Table 1 for further details). Both of these values correspond well to the expected size (2.77 nm; PDB entry 3v03; Majorek et al., 2012) and molecular weight (66.5 kDa) of monomeric BSA. Additional comparison to the monomeric crystal structure (PDB entry 3v03; Majorek et al., 2012) shows a similarly good match as the SEC-SAXS data for BSA ( 2 = 1.82; Fig. 5d and Supplementary Fig. S1b). To estimate to what extent an incorrect shift affects these results, the corresponding averages for shifts of 120 and 140 frames, respectively, were calculated. The absolute differences from the 130-frame shift are shown in Fig. 3( f ). In both cases, above 1 nm À1 the curve is shifted by a small constant, which many modelling algorithms take into account (Knight & Hub, 2015;Petoukhov et al., 2007;Svergun et al., 1995). Below 1 nm À1 differences in the relative contribution of capillary scattering result in a q-dependent difference, which could in principle affect modelling. However, these differences contribute less than 0.2% to the scattering signal and thus their influence is negligible.

IEC-SAXS of bovine serum albumin using a stepwise salt gradient
The linear gradient IEC works well for many proteins, but for samples where the elution peak is broad or is not well separated from other peaks, manual selection of appropriate steps in the gradient can ensure a purer and higher peak on faster timescales. For these cases, we describe an elution system in which the salt concentration is increased in predefined steps. The advantage of this approach is that by choosing a sufficiently long step length, it is possible to use buffer measurements from the same chromatography run for background correction, reducing the risk of nonmatching buffers owing to slow drifts. On the downside, even if the change in the buffer mixing ratio at the pumps is instantaneous, various effects, such as Taylor dispersion of flow in the capillary (see, for example, Wunderlich et al., 2014), co-elution of small molecules from the columns and the creation of new interaction sites for salt on the column and the eluted protein, result in non-instantaneous gradients at the measurement position. To show the validity of this approach, we again used BSA with the salt concentrations determined above from the offline IEC run.
The elution steps were chosen in such a way that the first two subpeaks of the BSA elute in the same step. In this case, BSA elutes as a single peak (Fig. 4a). As expected, the background scattering at high angles does not increase instantaneously, but saturates slowly after a steep initial increase. In order to subtract an appropriate background from each measured frame, it is necessary to interpolate between the two buffers recorded before and after the peak. As the scattering of BSA at high angles is rather low and the increase in signal above q = 4.5 nm À1 does not follow the protein concentration (Fig. 4a), one can assume that the changes in signal in this region are only owing to the difference in background. In order to limit the effect of the rather high noise level in this region, the increase is modelled with a research papers Acta Cryst. Stepwise-gradient IEC-SAXS performed on BSA. (a) IEC-SAXS chromatograms of BSA using a stepwise salt gradient. Top panel, UV absorbance (violet) and total scattering intensity (green). Middle and bottom panels, chromatograms at 1 and 4 nm À1 , respectively. (b) The buffer signal (mean scattering above 4.5 nm À1 , blue) increases slowly during the elution of the peak (total scattering intensity, green) and can be modelled by an exponential decay (red) between the buffer before the peak (I) and after the peak (II). (c) Comparison of the scattering ratio (0.11-0.5 nm À1 versus 1.5-2.5 nm À1 ) for background correction using the buffer before the peak (I, cyan), after the peak (II, green) and the modelled buffer (blue). (d) Forward scattering (black), radius of gyration (green) and mass (red) based on the correlated volume for the background-corrected curves in the peak region. The grey area indicates the frames that were used for subsequent averaging; the arrows indicate the positions of the individual curves presented in (e)  continuous function instead of using the high-q region of each frame to interpolate the buffer individually. A least-parameter model for an asymmetric, saturating increase as observed here is a single exponential decay from the buffer before (region I) to the buffer after (region II) the peak, i.e. a fit to II À (II À I)exp[À(N À N 0 )/N 0 ], where N is the frame number, N 0 is a constant offset and 1/N 0 is the rate of increase. To obtain the correct values for N 0 and N 0 , the mean of the data was fitted above 4.5 nm À1 (Fig. 4b). The best fit is obtained with N 0 = 1289 AE 2 and N 0 = 17 AE 3, giving an 2 of 0.86. These parameters match the starting point of the increase (frame 1290) and the estimated half-life of 12 AE 2 frames well and allow us to model the entire q-range.
This procedure allows the background to be subtracted individually for each frame and a constant scattering from the sample throughout the peak to be confirmed by assessing the stability of the ratio of scattering at low q versus mid/high q, as discussed above (Figs. 4c and 4d). Individual comparison of frames confirmed this constant signal (Figs. 4e and 4f ). Based on the forward scattering, the maximum concentration in this case was only 0.35 mg ml À1 . After averaging (Fig. 5a), a scattering curve which corresponds to monomeric BSA ( 2 = 3.63; Fig. 5d) was found. The radius of gyration is 2.7 AE 0.1 nm (Fig. 5b) and the Porod volume is 116 AE 5 nm 3 , matching previous results (also see Table 1). This shows that important structural parameters can be determined using either linear or stepwise gradients.
In order to estimate the mis-subtraction of the buffer and its effect on the conclusions that can be drawn from the data, one needs to determine corresponding over-subtracted and undersubtracted curves. We decided to over-subtract by directly subtracting buffer II and to under-subtract by subtracting the mean of buffer I and buffer II. This choice might be more extreme than strictly necessary, but provides an upper limit of the effect of the mis-subtraction. The differences in this case (Fig. 5c) are much more pronounced than for a linear gradient case (Fig. 3f ). This means that the background subtraction for the linear case is more accurate.
Above 1 nm À1 , buffer mismatch causes a constant shift in the signal. Below 1 nm À1 , the differences are larger but do not impact the structural parameters determined from this region, with the radius of gyration remaining at 2.7 AE 0.1 nm and the Porod volume increasing within its error to 120 AE 5 nm 3 only for the subtraction of buffer II and remaining unchanged for the other case.
This implies that while interpretations of domain arrangements might be affected by buffer mismatches, the observed overall shape (size, anisotropy, . . . ) of the protein can still be determined despite the larger inaccuracy in the background subtraction.
Direct comparison of the two scattering curves of BSA obtained using the two different IEC-SAXS approaches (Fig. 5d) shows two very similar curves. Owing to the lower protein concentration at the peak, the curve resulting from the step-gradient method is noisier. However, from about 2.5 nm À1 onwards the signal determined using the linear gradient method decreases more steeply. In particular, above 3.5 nm À1 the signal determined using the step gradient turns upward. This increase also results in a clear deviation from the predicted signal and is responsible for the higher 2 .
3.3. IEC-SAXS of the helicase-primase protein D5 323-785 using a stepwise salt gradient To demonstrate the applicability of this approach to a novel sample, we selected D5 323-785 , the D5N and helicase domains of the Vaccinia virus helicase-primase D5. The protein fragment forms a hexamer (320.88 kDa) that is required for its activity . After two nickel columns and a size-exclusion chromatography step, a contaminant still persisted and required an additional purification step via ionexchange chromatography (Fig. 2b), followed by an additional size-exclusion chromatography step. Each step takes time and about 30-50% of the material is lost in total. To optimize the data-collection strategy it would be advantageous to measure the hexamer directly from the Uno Q-1R column.
In the standard offline IEC experiment using a linear gradient, three not completely separated peaks at 11, 14 and 16% high-salt buffer were observed (Fig. 6a).
When using a stepwise gradient a sharp peak is obtained (Fig. 6b) at 12% salt buffer (142 mM NaCl) and small addi-   tional peaks in the subsequent steps. This indicates that D5 323-785 elutes completely at 142 mM NaCl and the peaks are better separated than in the linear-gradient elution procedure (Fig. 2a). The scattering at high q (4 nm À1 ) slightly overshoots the scattering level expected from the buffer run (Fig. 6c), probably owing to the co-elution of small ions. To check that the sample composition does not change throughout the peak, the low-salt buffer from each frame was subtracted and their radii of gyration were compared. This operation is not affected by the small change of background throughout the peak (data not shown). Based on the forward scattering from this preliminary processing, the maximum protein concentration is estimated to be 3.4 mg ml À1 , corresponding to a 20-fold increase from the injected concentration of 0.17 mg ml À1 .
Owing to the irregular changes in the background signal that were observed in these data, the approaches to modelling the change in background signal on a frame-by-frame basis that were applied to the previous examples are not applicable. To overcome this issue, first the frames in the peak were averaged in order to improve the signal-to-noise ratio in the high-q region, and a matching buffer from a salt-concentration series measured directly before the ion-exchange run was then chosen by comparing the average scattering in the q-range between 4.25 and 4.75 nm À1 .
The resulting curve (Fig. 7a) has a radius of gyration of 4.7 nm (Fig. 7b) and a Porod volume of 577 AE 5 nm 3 , corresponding to a mass of 338 kDa (Petoukhov et al., 2007), matching the hexamer mass of 320 kDa well (see also Table 1). The P(r) function (Fig. 7c) shows one peak that is slightly shifted to larger distances. This shape, in addition to the clear minima, hints at a mostly spherical shape with a cavity and matches the expected molecular shape well .
As the steps were chosen more finely than in the previous experiment with BSA, it is easier to estimate the degree of buffer mismatch in the D5 323-785 experiment. The extent to which it influences the interpretation of the results can be estimated by selecting buffers from a higher and a lower salt step of the buffer run (Fig. 7d). In both cases the resulting curve is distinguished from the reported curve by a small constant above q = 0.8 nm À1 . Below q = 0.8 nm À1 the deviations are no longer constant but are below 0.1% of the signal, and therefore do not influence the interpretation.

Additional considerations
One of the major difficulties in online SEC-SAXS is the effect of radiation damage on the observed signal. The continuously changing signal makes assessment of radiation-  (a) Average sample (violet) and buffer (black) curves for the D5 323-785 protein, the resulting subtraction (red) and the fit for calculating the P(r) function (blue). (b) Guinier fit (black line) of the scattering curve shown in (a). Open symbols represent points that were not used for fitting. The lower panel shows the residuals. (c) Pair distance distribution function of the average curve, showing a single peak. (d) Differences in scattering between the curve based on the best-matching buffer and a clearly undersubtracted curve (preceding buffer step, orange) and over-subtracted curve (subsequent buffer step, blue). induced changes to the SAXS signal difficult. In addition, radiation-damaged material often displays a tendency to adhere to the surfaces of the sample environment (Brookes et al., 2013;Jeffries et al., 2015). IEC-SAXS can be performed at relatively high flow rates (1 ml min À1 for the studies in this publication); consequently, the average dwelling time of material in the X-ray beam is shorter than for standard SEC-SAXS experiments and the contribution of damaged material to the signal is reduced. Furthermore, the higher flow rate reduces the risk of damaged material spoiling the sample environment by adhering to the surface (Epstein, 1997).

Conclusions
For many macromolecules in solution, ion-exchange chromatography is a required step and is the best-suited purification method, as it can often separate similarly sized proteins. However, the IEC step is usually followed by another purification step, or dialysis, to ensure optimal buffer subtraction. Here, we demonstrate for the first time the application of ionexchange chromatography directly prior to SAXS measurements.
The linear gradient method presented here is analogous to a standard ion-exchange protocol and does not inherently require additional offline tests. In addition, background correction can be achieved by correct alignment of a buffer run without any need for interpolation. As each frame can be corrected individually, it is possible to confirm a stable signal throughout the peak. However, owing to the continuity of the gradient, sharp and well separated peaks are beneficial.
By carefully selecting the salt concentration of the steps in the step-elution method, this method allows the separation of the peaks to be improved, as no limit exists on the number of substeps. However, background subtraction requires more caution, as the discontinuity of the elution gradient results in the discontinuous co-elution of small ions. In our first example (BSA), it was still possible to correct the background individually for each frame by interpolating the buffer signal. In our second example (D5 323-785 ) it was not possible to interpolate the background throughout the peak, and the correction by framewise comparisons was not applicable. Nevertheless, the correct background for the average signal can be found by measuring a variety of mixing ratios prior to the experiment and carefully choosing the matching one. Despite this difficulty, we are confident that it can be applied to a wide variety of biological macromolecules.
Online ion-exchange chromatography adds another important biochemical purification method to the repertoire of purification methods which can be coupled with SAXS (Round et al., 2013;David & Pé rez, 2009;Graewert et al., 2015;Jensen et al., 2010;Hynson et al., 2015;Mathew et al., 2004;Watanabe & Inoko, 2009). Proteins of similar apparent molecular weight, but different net surface charges, can now be separated online using the IEC-SAXS method. While background subtraction is slightly less straightforward than for SEC-SAXS, the variation in the background is smaller than for differential ultracentrifugation coupled with SAXS owing to the high sugar content of the latter (from 15 to 35% sucrose; Hynson et al., 2015). The background-correction method here for D5 323-785 , based on averages, can be performed using any software package for SAXS data analysis, such as PRIMUS (Konarev et al., 2003) or SCÅ TTER, assuming that the possible buffer range has been well determined experimentally.
For IEC, buffer conditions, such as salt concentrations or pH, are primarily determined by the biochemical characteristics of the protein of interest, such as its isoelectric point and solubility (Yigzaw et al., 2009). However, one should aim to minimize the amounts of all additives in order to maximize the X-ray contrast between the protein and the buffer in a SAXS experiment.
As with any SAS data, IEC-SAXS results need to be carefully validated. Special care should be paid to sample monodispersity, which cannot be directly assessed by IEC-SAXS, and correct background subtraction, especially in the case of flexible proteins (Jacques et al., 2012;Jacques & Trewhella, 2010). It is important to estimate how missubtraction would affect the conclusions drawn from the data. For the results shown in this paper, we have shown that deviations from the ideal background subtraction are small enough to allow reliable conclusions.
In conclusion, IEC-SAXS is a useful technique that allows accurate data collection with fewer preparation steps and minimizing the loss of time and sample. Proof of principle of a simple elution method is demonstrated in this paper with the option to optimize species separation by using steps in the salt gradient. Validation of three different approaches to achieve optimal background subtractions is also given. A suitable combination of elution method and background subtraction can be chosen to suit the sample of interest best and to provide necessary information for data validation (as demonstrated), so that the resulting data can be used with confidence for subsequent analysis and modelling using the standard tools available within the scientific community. The online SEC system installed at BM29 can routinely perform these experiments and is available for user access on request.