Investigating Native Metal Ion Binding Sites in Mammalian Histidine-Rich Glycoprotein

Mammalian histidine-rich glycoprotein (HRG) is a highly versatile and abundant blood plasma glycoprotein with a diverse range of ligands that is involved in regulating many essential biological processes, including coagulation, cell adhesion, and angiogenesis. Despite its biomedical importance, structural information on the multi-domain protein is sparse, not least due to intrinsically disordered regions that elude high-resolution structural characterization. Binding of divalent metal ions, particularly ZnII, to multiple sites within the HRG protein is of critical functional importance and exerts a regulatory role. However, characterization of the ZnII binding sites of HRG is a challenge; their number and composition as well as their affinities and stoichiometries of binding are currently not fully understood. In this study, we explored modern electron paramagnetic resonance (EPR) spectroscopy methods supported by protein secondary and tertiary structure prediction to assemble a holistic picture of native HRG and its interaction with metal ions. To the best of our knowledge, this is the first time that this suite of EPR techniques has been applied to count and characterize endogenous metal ion binding sites in a native mammalian protein of unknown structure.


1) Experimental procedures
Protein purification HRG was purified from rabbit serum (Sigma Aldrich, Poole UK) using immobilised Ni 2+ -affinity chromatography (HisTrap column; Cytiva, Little Chalfont, UK). Prior to purification the serum was centrifuged at 4,000 × g for 30 min at 4°C, filtered through a 0.45 µm filter (Sartorius, Epsom, UK) and imidazole was added to a final concentration of 5 mM. The resultant fractions containing HRG were further subjected to anion exchange (HiTrap DEAE FF column, Cytiva) and gel filtration (HiLoad Superdex-75 column, Cytiva) chromatography. All purification steps were carried out using an ÄKTA Purifier (Cytiva). Prior to any experiment, the purified protein was dialysed in the appropriate buffer.
Prior to titration, rabbit HRG and CuCl2 (Sigma Aldrich, Poole UK) were dissolved separately in ITC buffer (140 mM NaCl, 50 mM Tris, pH 7.4) to final concentrations of 10 µM and 1.5 mM, respectively, at 25°C and pH was re-adjusted to 7.4 to avoid mismatch conditions. The titration involved a single injection of 0.4 µl of CuCl2 over 0.8 s followed by 18 injections of 2 µl CuCl2 over 5 s with a 150 s interval between injections to allow full equilibration. The stirring speed was set to 750 rpm throughout.
Additionally, a control experiment was carried out whereby the titrant was injected into buffer and the resultant heats were deducted from those obtained in the main experiment.

Heparin binding assay
Zinc ion binding to HRG is known to increase the affinity of this protein to unfractionated heparin (UFH). In order to assess whether Cu II ions elicit a similar effect, binding of HRG to immobilised UFH in the presence and absence of Zn II and Cu II was performed. For this Costar Carbohydrate Binding Plates (Sigma-Aldrich, Poole, UK) were used. UFH (Acros Organics, Loughborough, UK) was activated with 15 mM sodium periodate according to the manufacturer's protocol. Purified rabbit HRG (rHRG) was biotinylated using a Biotinylation Kit (Stratech, Newmarket, UK) and then dialysed overnight at room temperature against 50 mM HEPES, 150 mM NaCl, 0.2 % Tween 20, pH 7.4. The activated UFH was dialysed twice for 1 h at room temperature against 0.1 M sodium acetate, pH 5.5. The activated UFH (10 µg/well) was then bound to the plate overnight. Wells were washed with 50 mM HEPES, 150 mM NaCl, 0.2 % Tween 20, pH 7.4 and then blocked with 50 mM Tris buffer pH 8.2, 0.2 % fish gelatine for 1 h at room temperature. Biotinylated rHRG was added to the wells (0-3 µM) for 2 h at 37 °C in the absence and presence of 50 µM ZnCl2 or 50 µM CuCl2. After washing with 50 mM HEPES, 150 mM NaCl, 0.2 % Tween 20, pH 7.4, binding was detected with alkaline phosphatase-linked streptavidin (1:10,000 dilution, ThermoFisher Scientific, Loughborough, UK) and p-nitrophenol phosphate substrate (200 µg/well, Sigma-Aldrich). The reaction was stopped with 3 M NaOH and the absorbance was read at 405 nm using a Dynex MRX spectrophotometer (Dynex Technologies, West Sussex, UK).

EPR sample preparation
Pseudo-titration samples were prepared with a protein concentration of 250 µM in buffer containing 50 mM Tris and 140 mM NaCl at pH 7.4 and Cu II from 1 to 20 molar equivalents (from 250 µM to 5.0 mM). The control sample contained no protein and 250 µM Cu II . Two batches of protein were prepared, the first one yielded the control sample and pseudo-titration samples from 1 to 10 molar equivalents of Cu II ; the second batch yielded pseudo-titration samples at 5 and 10 molar equivalents Cu II (repeat samples for reproducibility), and the 12, 15, and 20 molar equivalent samples. Samples were mixed with equal amounts of ethylene glycol for cryoprotection, resulting in a final protein concentration of 125 µM and Cu II concentration of 125 µM to 2.5 mM, transferred to 3 mm (70 µL sample volume) quartz EPR tubes, and immediately frozen in liquid nitrogen.
To determine the maximum binding capacity of HRG four further samples were prepared at a final Cu II concentration of 2.5 mM, thus corresponding to the Cu II concentration in the 20 molar equivalents sample of the pseudo-titration. Samples had final protein concentrations of 50, 25, 12.5, and 6.25 M, corresponding to 50, 100, 200, and 400 molar equivalents of added Cu II , respectively.
Note that Tris buffer shows weak complexation of free copper ions, leading to 14 N hyperfine interactions. However, other commonly used buffers such as PBS buffer, lead to precipitation resulting in reduced Cu II -loading. 1 Indeed, there are very few (if any) options for buffers that do not complex free copper ions. This is highlighted by a review on the use of pH buffers and their interaction with metal ions, where all of the buffers examined (out of the 31 buffers listed) exhibited the capacity to undergo complexation with copper ions. 2 Therefore, for consistency and comparability ITC buffer (140 mM NaCl, 50 mM Tris, pH 7.4, see above) was used for EPR sample preparation.

Continuous wave (CW) EPR
CW EPR spectra were obtained at 120 K with a Bruker EMX 10/12 spectrometer running Xenon software and equipped with an ELEXSYS Super Hi-Q resonator at an operating frequency of ~9.5 GHz with 100 kHz modulation. Temperature was controlled with an ER4141 VTM Nitrogen VT unit (Bruker) operated with liquid nitrogen. CW spectra were recorded using a 160 mT field sweep centred at 310 mT, a time constant of 40.96 ms, a conversion time of 6.67 ms, and 8000 points resolution. An attenuation of 10.0 dB (20 mW power) and a modulation amplitude of 0.2 mT were used. CW spectra were phase-and background-corrected and the double integral was obtained using the Xenon software. Spectra were field-corrected using DPPH as a standard.

Pulse experiments
Pulse experiments were performed at X- (9.5 GHz) and at Q-band (34 GHz) both operating on a Bruker ELEXSYS E580 spectrometer, with probe-heads supporting a split ring resonator (4118X-MS3) for Xband and a 3 mm cylindrical resonator (ER 5106QT-2w in TE012 mode) for Q-band, respectively. Pulses were amplified by pulse travelling wave tube (TWT) amplifiers (Applied Systems Engineering) with nominal output of 1 kW and 150 W at X-and Q-band, respectively. Temperature was controlled via cryogen free variable temperature cryostats (Cryogenic Ltd) operating in the 3.5-300 K temperature range.
Temperature optimisation for PELDOR experiments were performed between 10 K and 50 K using the HRG sample with 2 equivalents of Cu II . Here, T2 (or Tm) was determined from a 2-pulse decay experiment with stretched exponential decay for fitting, and T1 was the determined longer  from an inversion recovery experiment. Determined relaxation times were then used to assess the relative sensitivity per temperature as described previously. 3 T2 experiments were further performed on selected samples to investigate the change in relaxation behaviour between 1 and 400 molar equivalents of Cu II .
PELDOR experiments were performed using the four-pulse DEER 4-6 pulse sequence π/2(A) -τ1 -π(A) -(τ1 + t) -π(B) -(τ2 -t) -π(A) -τ2 -echo at 30 K as described previously, 7 with a frequency offset (pump -detection frequency) of +80 MHz (~3 mT). Shot repetition times were set to 300 s; 1 was set to 380 ns, and 2 to 3300 ns. Pulse lengths used were 16 and 32 ns for π/2 and π detection, and 14 ns for the ELDOR π pump pulse. The pump pulse was placed on the resonance frequency of the resonator and ~1.7 mT away from the maximum of the spectrum to lower field to allow placing pump and detection positions symmetrically around the maximum. PELDOR data were analysed using DeerAnalysis2015. 8 Raw experimental PELDOR data were cut at 2500 ns to remove artefacts at the end of the time trace, and background-corrected using a monoexponential decay function with the background start point set to 1250 ns before subjecting the trace to Tikhonov regularisation. A regularization parameter α of 100,000 was chosen by visual inspection for all traces. Further statistical analysis of PELDOR data was performed as described previously, 9 using the validation tool of DeerAnalysis 2015. 8 Briefly, the background start time was varied from 5% to 80% of the dipolar evolution time in 16 trials and 50% random noise was added (level 1.50) with 50 trials, yielding 800 trials per trace. Trials were pruned, keeping only those where the root mean square deviation (rmsd) was within 15% of the rmsd of the best fit; these trials were further used for determination of the mean modulation depth and error.

Hyperfine spectroscopy
3-pulse electron spin echo envelope modulation (ESEEM) 10-12 spectroscopy was performed at 30 K at X-band frequencies (~9.5 GHz) and on the maximum of the field-swept spectrum with a pulse length of 16 ns for π/2 detection and inter-pulse delay  set at the blind spot of the proton (~216 ns). The delay T, set at 280 -300 ns, was incremented with a dwell time T of 8 ns and a 4-step phase cycle was used. Four dimensions were recorded, whereby the inter-pulse delay was incremented by 0.5 1(H); the first  was selected for further processing and setting up the HYSCORE experiment (see below). Data were analysed by fitting an exponential decay background function to the (phasecorrected) raw data, subtracting the raw data by this background function, and then dividing the difference by the background function, thus retaining amplitude information after fast Fourier transformation (FFT) similar as described previously. 13 The resulting trace was further subjected to a Hamming window, zero-filling and FFT, before obtaining the absolute (or magnitude) spectrum.
Hyperfine sublevel correlation (HYSCORE) 14 spectroscopy was performed at 15 K at X-band frequencies (~9.5 GHz) and on the maximum of the field-swept spectrum with a pulse length of 16 and 32 ns for π/2 and π, respectively,  set at the blind spot of the proton (~216 ns) as chosen from the 3-pulse ESEEM, t1 = t2 = 56 ns, and a 4-step phase cycle. Data were processed and analysed using the Matlab plugin HYSCOREAN, 15 employing Hamming apodization, zero-filling, 3 rd order polynomial background correction, and diagonal and anti-diagonal spectral symmetrisation, keeping a similar amount of noise for each spectrum by adjusting the minimum contour level percentage accordingly.
EDNMR 16-17 measurements were conducted by using an additional second frequency option (E580-400U) jointly with the Q-band setup described above at 30 K. Measurements were taken at low field (1.0240 -1.0630 T) and high field (1.1674 -1.1833 T) on a Hahn echo with an ELDOR HTA (high turning angle) pulse length of 8 µs and an ELDOR attenuation of 30 dB. Only low-field spectra are shown as they offer increased resolution. Note that directly coordinated nitrogen atoms visible in EDNMR would not contribute to the ESEEM signal under our experimental conditions (pulse lengths of π/2 = 16 ns at X-band).

Simulations
Representative numerical simulations of 3-pulse ESEEM data were performed with EasySpin, 18 using the implemented simulation function "saffron". Prior to simulation, data was background corrected, apodised with a Hamming window, zero-filled and Fourier transformed. The 3-pulse ESEEM simulations were conducted in the frequency domain. EDNMR simulations for the control and HRG with 1 equivalent of Cu II were performed as described previously. 19 The central blind spot was modelled by a sum of Gaussian and Lorentzian lines and subtracted from both raw data sets.
JNet version: 2.3.1; UniRef90 release: 2014_07, 09-Jul-2014. Figure S2: AF2 prediction for rabbit HRG (UniProt Q28640). Colours indicate pLDDT confidence scores between 0 and 100; very low pLDDT scores (below 50 = orange colour) are likely to be unstructured or intrinsically disordered; note that the prediction does not inform on the relative likelihood of different conformations of such regions.   Figure S6. Stacked overlays of CW EPR spectra for comparison of four molar equivalents (1, 5, 10, and 20) of Cu II to illustrate the gradual decrease of SHF splittings. The right plot is zoomed into the SHF region for better visualisation.

3) Continuous wave (CW) EPR data
Decreasing resolution of SHF upon addition of Cu II was in agreement with earlier data. 20 While this could in principle be caused either by broadening from introduction of dipolarly coupled close-by Cu II or by (un-)specific binding of Cu II to non-histidine sites, the latter is unlikely as the resolution of the signal was lost more quickly from 10 molar equivalents of Cu II onwards than a mere dilution of one type of sites would explain. Thus, either there is dipolar broadening or existing binding sites change.
Considering an additive spectrum, where an additional component was added to the existing spectrum exhibiting 14 N SHF couplings, one could not explain our observations. Here, while the amplitude of the resolved SHF couplings seemed roughly halved going from 5 to 10 molar equivalents of Cu II , instead of being halved again going from 10 to 20 equivalents they have almost completely vanished at this Cu II loading. This suggested that the situation was more complicated and that spectra could not be simulated simply by assuming 10 binding sites with SHF and 10 binding sites without.
Instead, there was either spectral broadening involved or SHF of initial binding sites were changing upon occupation of additional sites. This situation rendered simulations based on simply adding new species with added Cu II unfeasible.  Table S1. Measured values for AII and gII. DPPH was used as a reference.

EDNMR
Effect of buffer coordination: The control sample containing no protein showed peaks arising from the interaction of Cu II with nitrogen atoms of the Tris buffer ( Figure S7).  Figure S7. EDNMR simulations for the control (Cu II in Tris buffer; left) and HRG with 1 molar equivalent of Cu II (right). Only the right-hand side of the EDNMR spectra are shown and the central blind spot has been subtracted from both raw data sets. Parameters to simulate the EDNMR spectrum of Cu II in Tris buffer are given in Table S2.
A good agreement between experiment and simulation of both the control EDNMR spectrum and the spectrum in presence of HRG could be obtained, not considering the peak of the proton Larmor frequency (as indicated in Figure S7). Details of hyperfine parameters used for the simulations are given in Table S2  In the presence of HRG, EDNMR spectra showed additional peaks with one coinciding with the proton Larmor frequency contribution, as well as a broad background feature. Based on the simulations the positions of the defined new peaks confirmed direct coordination of at least two imino nitrogen nuclei of imidazole rings to Cu II with a mainly isotropic hyperfine coupling caused by significant electron spin density delocalisation onto these nuclei that could not be attributed to histidine residues forming the Cu II binding sites.
These hyperfine couplings were apparent already in presence of 1 molar equivalent of Cu II , and EDNMR spectra remained virtually unchanged with increasing Cu II concentration up until [15][16][17][18][19][20] equivalents. This indicated that, within this metal ion ratio, all populated histidine binding sites showed very similar binding geometries. Speculatively, the broad background feature may be attributed to heterogeneity in coordination of the Cu II to the high affinity binding sites, however it is important to note that no other specific couplings (which would indicate specific sites) were resolved.
Further increasing the relative amount of Cu II (50 and more molar equivalents of Cu II ) led to the loss of the histidine-associated peaks, suggesting that non-histidine binding sites became dominant. In agreement with the SHF observed in the CW EPR spectra, the histidine-associated peaks have been lost at 50 molar equivalents of Cu II and above. If one assumed a mere additive behaviour for the composition of the spectrum, then 40% of histidine-associated peaks should remain visible at 50 molar equivalents of Cu II .
These data led to the hypothesis that different metal ion binding sites have different relaxation behaviour, where Cu II bound to high-affinity sites relaxes faster and is thus contributing significantly less to echo detected experiments once lower-affinity sites are being occupied, as otherwise one would expect the hyperfine coupling to be recoverable even from the broadened spectra.  Quantitative analysis of ESEEM peak amplitudes after Fourier transformation (FFT) and peak integration provided further support for our EDNMR data:

3-pulse ESEEM
At up to 20 molar equivalents of Cu II , the FFT amplitudes were relatively constant, while at higher molar equivalents (≥50) they were substantially reduced. The intensity of the DQ peak (or the DQ peak integral) depends on the number of histidine residues involved in the binding site. 21 DQ peak integral values were fairly stable between 5 to 15 molar equivalents of Cu II but decreased at 20 molar equivalents, suggesting that less histidine residues per Cu II were available for binding from this point.
At up to 15 molar equivalents, quantitation of the DQ peak suggested coordination of the Cu II by at least two imidazole rings, which was also in good agreement with simulations (see below). 13, 21 At 1 molar equivalent of Cu II NQI and DQ peak integrals were slightly lower than between 5 to 15, again suggesting binding to less than two histidine residues; the reason for this is not clear but might be due to some competition with the Tris buffer. shown on the right for HRG with 20 molar equivalents of Cu II , with simulations (red) demonstrating a good fit to experimental data (black) could be obtained assuming two remote histidine nitrogen atoms (see Table S3 for simulation parameters).   Figure 4C in the manuscript.
In an attempt to visualise also the weak two-nitrogen double quantum transitions (visible in the ESEEM spectra at ~8 MHz) [22][23] we re-processed the 5 to 20 molar equivalents Cu II HYSCORE spectra allowing more noise. This processing revealed weak but visible peaks at the expected positions, shown below in the projection contour plots, with strongest peaks observed at 15 molar eq. Cu II , in line with our other observations. HYSCORE with 5 molar equivalents of Cu II HYSCORE with 10 molar equivalents of Cu II Normalised signal intensity 2 x  / ns control 1 eq Cu II 1 eq Cu II 10 eq Cu II 20 eq Cu II 50 eq CuI I 400 eq Cu II Figure S13. Two-pulse echo decay for selected samples. Note the marked step from 20 to 50 molar equivalents of Cu II , with an additional component appearing in the decay.  Figure S14. PELDOR temperature optimisation. All optimisation experiments were performed with HRG + 2 molar equivalents of Cu II and 50% ethylene glycol for cryoprotection. T1 is measured with a 3pulse inversion recovery experiment (top left), Tm with a 2-pulse decay experiment (top right). The optimum temperature was derived from the relative sensitivity per temperature (bottom row) taking into account the Boltzmann factor, the temperature-dependent Tm (or T2; i.e., how fast is the loss of coherence), and averaging (i.e., how fast do populations re-equilibrate) as described previously. 3 An optimum temperature of 30 K was determined for PELDOR measurements from these relaxation data. 2 eq 4 eq 5 eq 7 eq 9 eq 10 eq 15 eq t / s   Figure S17. Exemplary PELDOR data for HRG with 1, 5, 10, and 15 molar equivalents of Cu II added. Raw and background-corrected data are shown in the left and middle graphs, respectively. Corresponding y-stacked distance distributions are given to the right. Shown are the 95% confidence estimates (± 2σ) of the distance distributions as obtained by statistical analysis. Colour bars represent reliability ranges as described in the DeerAnalysis 8 manual (green: shape reliable; yellow: mean and width reliable;

Molar eq Cu II
orange: mean reliable: red: no quantification possible).
As shown above, PELDOR experiments yielded very broad distance distributions that we refrained from quantifying and that did not change significantly within confidence intervals between 1 and 20 equivalents of Cu II added.

Constructing a multi-site binding polynomial
We are assuming a speciation model with two different metal ion binding sites for the HRG, each with different affinities reflecting high and lower affinity sites. Simulation of empirical modulation depths observed in the HRG + Cu II PELDOR pseudo-titration series requires a mathematical description of the protein-ligand binding equilibria of each species in solution, which can be achieved via a general multisite binding polynomial. 24 The fractional population of each species, the macroscopic speciation vector, , is a function of the following parameters: total protein concentration, [ ] 0 , total ligand concentration, [ ] 0 , dissociation constants 1 and 2 , and the number of high affinity, , and low affinity, , sites. These fractional macroscopically-bound populations are significant in the simulation of PELDOR modulation depths because each species will contribute to the observed modulation depth with a weighting proportional to the product of their relative population, and the number of spins present in that species: Where f is as defined above, is the inversion efficiency of the pumping pulse and is the total number of spins in the system. The unmodulated echo contributions (1 − ) −1 are averaged for all species with = 1 to Cu II ions bound taking into account the increase in Cu II signal and normalising by their contribution to the signal at zero time. 9 PELDOR modulation depths were simulated using MATLAB, and mean square error was used as a metric for simulation quality:

Exploratory simulations of PELDOR modulation depths
Cu II -Cu II Q-band PELDOR measurements were performed on rHRG at 125 μM protein concentration, and with varying equivalents of Cu II , ranging in concentration from 125-2500 μM. Previous literature, and empirical ITC data suggested there are 12 high-affinity ligand-binding sites (n = 12), with a KD (at 235 K) of 5 × 10 -8 . The extrapolation of KD to low temperature was facilitated by application of the van't Hoff equation and is made necessary because the binding kinetics are fast with respect to the flashfreezing, and so EPR measurements reflect the binding equilibrium at the freezing-point temperature.
The inversion efficiency (λ) of a 16 ns rectangular pump pulse respective of Cu II spectra at Q-band frequency, and with an offset of ~150 MHz was demonstrated to be 1-2%. 25 For the subsequent simulations, inversion efficiency was approximated as 1.5% (λ = 0.015), unless otherwise stated. It should be noted that for the KD and number of low affinity sites, there was little literature or empirical data to corroborate accurate values.
Since approximate values are available for all other parameters, we first investigated the influence of the number, and KD of the low affinity sites. The experimental modulation depths and 95% confidence intervals are given in Table S7, with the corresponding Cu II concentrations. Figure S18 shows error surfaces for value pairs of (m, KD2) when comparing simulated values to the empirical data, and clearly demonstrates the relative insensitivity of the simulation to the number and affinity of the second class of sites.  Table S8 gives the error minima for value pairs of (m, KD2), and indicates that as the number of lowaffinity sites increases, the affinity correspondingly decreases, as would be expected for a constant modulation depth. While the global error minimum is at (m = 3, KD2 = 1.0 × 10 -5 ), it is seen that the error minimum is not particularly pronounced, and experimental data is also reasonably described by m = 5, 10 and 15.   Table S8. The minimum mean square error of each trace shown in Figure S18 (left) and the corresponding optimum value of KD.
For the first 12 equivalents of Cu II added to the HRG, the modulation depth is approximately linearly increasing and implies that binding is quantitative (i.e., n ≥ 12 and KD1 << 1.25 × 10 -4 ). Since Δ continues to increase for 15 and 20 molar equivalents of Cu II , this implies that HRG continues to bind Cu II , and that low affinity sites are not fully saturated in the regime of Cu II concentrations initially investigated. Therefore, this suggests that m ≥ 8 (m = 8 presupposes that binding is quantitative, which is unlikely given that the optimal low-affinity KD values predicted by the simulation (for m ≥ 5) are > 1.25 × 10 -4 ).
It is more probable that the true value of m is significantly higher than 8, since this would account for the continuously increasing Δ and the non-linearity of the increase for 15 and 20 equivalents of Cu II (many sites being sparsely populated), however one cannot use the simulations to make precise inferences about the number and KD of the low affinity sites.
Contrarily, it can be shown that to satisfactorily describe the observed data, particularly for the first 12 equivalents, n ≥ 12, as illustrated in Figure 5 in the main text.
This is highly consistent with the previous literature and empirical ITC data, and values greater than n = 12 lead to a simulated modulation depth profile which begins to plateau at higher concentrations of Cu II and therefore overestimate Δ at 20 equivalents. It should also be noted that regardless of the KD1 chosen for the simulation, the error function is always minimized for n = 12. In light of CW EPR measurements that indicated sustained Cu II -binding up to 100 equivalents of Cu II with respect to HRG protein, further simulation was performed using 100 low-affinity binding sites (m = 100), with a dissociation constant of ~5.0 × 10 -2 . Results are shown below.