EPR assessment of protein sites for incorporation of Gd(III) MRI contrast labels

Abstract

We have engineered Apolipoprotein A-I (apoA-I), a major protein constituent of high-density lipoprotein (HDL), to contain DOTA-chelated Gd(III) as an MRI contrast agent for the purpose of imaging reconstituted HDL (rHDL) biodistribution, metabolism, and regulation in vivo. This protein contrast agent was obtained by reacting the thiol-reactive Gd[MTS-ADO3A] label with Cys engineered at four distinct positions (52, 55, 76 and 80) in apoA-I. MRI of infused mice previously showed that the Gd-labeled apoA-I migrates to both the liver and the kidney, the organs responsible for HDL catabolism; however, the contrast properties of apoA-I are superior when the ADO3A moiety is located at position 55, compared to the protein labeled at positions 52, 76 or 80. It is shown here that continuous wave X-band (9 GHz) EPR spectroscopy is capable of detecting differences in the Gd(III) signal when comparing the labeled protein in the lipid-free to the rHDL state. Furthermore, the values of NMR relaxivity obtained for labeled variants in both the lipid-free and rHDL states correlate to the product of the X-band Gd(III) spectral width and the collision frequency between a nitroxide spin label and a polar relaxation agent. Consistent with its superior relaxivity measured by NMR, the rHDL-associated apoA-I containing the Gd[MTS-ADO3A] probe attached to position 55 displays favorable dynamic and water accessibility properties as determined by X-band EPR. While room temperature EPR requires >1 mM Gd(III)-labeled and only >10 μM nitroxide-labeled protein to resolve the spectrum, the volume requirement is exceptionally low (~5μL). Thus, X-band EPR provides a practical assessment for the suitability of imaging candidates containing the site-directed ADO3A contrast probe.

Introduction

Apolipoprotein A-I (apoA-I) is the major protein among high-density lipoproteins (HDL). Because HDL provides the circulation efflux pathway for cholesterol and displays anti-inflammatory properties (1), it plays a central role in cardiovascular health. The beneficial effects can largely be attributed to the function of apoA-I in the reverse cholesterol transport pathway where excess of phospholipids and cholesterol is removed from peripheral tissues, including the vascular wall, for transfer to the liver, adrenal gland, and steroidogenic organs.

In the absence of lipids, apoA-I is proposed to organize into a helical bundle that is stabilized by helix-helix interactions, which are largely replaced by helix-lipid interactions in the lipid bound state (2,3). The transition of lipid-free apoA-I to HDL and the HDL maturation process rely on a structural plasticity of the apoA-I protein that involves several structural intermediates (4–6). Common for these structures is the ability of apoA-I/HDL to maintain water solubility in either the lipid-free (LF) or the lipid-bound (LB) states. In addition to native apoA-I in HDL, several amyloidogenic variants of apoA-I that accumulate in harmful plaques in vivo have been described (7). Although information on the molecular transition leading to amyloidogenesis is emerging (8–11), it appears that multiple organs serve as targets for amyloidogenic apoA-I, depending on the nature of the modification. Understanding the beneficial as well as the potential pathological influences of apoA-I in vivo demands the capability to chronicle complex changes in apoA-I/HDL localization.

To facilitate the imaging of apoA-I/HDL targeting and metabolism in vivo, we have modified the protein with a Gd(III) contrast species such to provide contrast enhancement in magnetic resonance imaging (MRI) measurements (12). The contrast provided by Gd(III) arises from an accelerated relaxation rate of the protons surrounding the contrast agent. This proton relaxivity is defined as the increase of the longitudinal water proton relaxation rate per millimolar concentration of Gd(III). Although the theory for designing optimal contrast agents is under consideration, the principle factors regarding the level of contrast enhancement have been identified. Those include the chemical environment (the number of the vacant coordination sites of the paramagnetic ion accessible to water molecules and the rate of water exchange) and the physical properties of the contrast agent (its electronic spin relaxation times and the rate of rotational diffusion) (13). With the ability to engineer proteins for carrying out site-specific contrast probes, one wishes to conceive a rational strategy that would optimally position the attached contrast species with favorable solvation and dynamical properties.

The specific NMR signal intensity provided by the relative relaxivity of a tethered MRI contrast agent depends on both the intrinsic properties of the paramagnetic metal ion and also the nano-milieu of the attached label, including its sterical environment and solvent exposure. Experimentally, the relaxivity of contrast agents can be evaluated through in vitro NMR measurements. However, NMR requires a substantial amount of labeled recombinant protein, typically, in excess of 100 nmoles. A simplified systematic approach for predicting and evaluating the efficiency of biomolecular contrast agents (such as engineered HDL for in vivo MRI biodistribution analysis) is, therefore, of great practical and technical value.

We have designed human apoA-I with single cysteine mutations at positions 52, 55, 76 and 80, and used these mutants for specific incorporation of the contrast probe Gd[MTS-ADO3A] (14) within the apoA-I protein (12). These targeted positions were selected according their likelihood (2) of being exposed towards aqueous phase and, therefore, suitable for modification by a contrast agent (Fig. 1). Our previous evaluation of the contrast properties for Gd-labeled apoA-I infused into mice provided a time-dependent imaging of the species in liver and kidney, the organs central to the metabolism of apoA-I/HDL. NMR evaluations of these Gd-labeled apoA-I agents showed significant relaxivity differences depending on the site of the probe attachment, and also whether the protein was lipid-bound or lipid-free (12). Here, we report that X-band (9 GHz) electron paramagnetic resonance (EPR) of the attached Gd[MTS-ADO3A] probe is useful for predicting these differences, where the flexibility of the attached probe affects the central line shape of the chelated Gd(III). These differences in line shape and their predictive value for protein imaging agent design are discussed according to (i) dynamic modulation of the static electronic zero-field splitting (ZFS) of chelated Gd(III) that is predicted to contribute at low magnetic fields (13,15,16), (ii) the dynamic reorientation of the labeled apoA-I side chain, and (iii) the water exchange rate for the attached Gd(III) label.

Fig. 1.

Structure of the nitroxide (MTSL, top) and the macrocyclic DOTA ligand complexed with Gd(III) (Gd[MTS-ADO3A]). Both probes react with engineered Cys residues in proteins via the methanthiosulfonate (MTS) attachment group, which provides a specific conjugation to the targeted Cys residue via a disulphide bridge.

Results and Discussion

Factors affecting Gd(III) proton relaxivity

There exists a reliable description for the relaxation of water protons in a vicinity of paramagnetic species provided by the Solomon – Bloembergen theory that accounts for dipolar interaction between the water protons and the electronic S=7/2 spin of the Gd(III) ion (17–19). According to this theory, the dipolar magnetic field between the proton and paramagnetic metal ion such as Gd(III) is modulated by the electronic relaxation of the ion itself as well as by dynamics of the ion-to-proton interspin vector, including changes in the interspin distance (vector length) and vector orientation in the external magnetic field. Predicting and understanding the overall proton relaxivity of MRI contrast agents is not an easy task, as several competing factors may be contributing to the overall effect observed in the experiment (reviewed in (20)). However, one of the most important parameters contributing to the relaxivity is the shortest correlation time among the processes that modulate the dipolar field originating at the paramagnetic metal ion. Those processes mainly include chemical exchange of water protons between the bulk and the inner coordination sphere, rotational diffusion of the complex, and the ion electronic spin relaxation. In addition, it has been shown that continuous wave (CW) EPR spectra of small Gd(III) chelates report on the zero-field splitting parameter of this S=7/2 spin system as well as on the electronic T₂ relaxation time, which can then be related to the electronic T₁ (21–23).

Here we explore the usefulness and sensitivity of EPR spectroscopy in evaluating macromolecular MRI contrast agents. Specifically, we compare EPR data of Gd[MTS-ADO3A]-labeled apoA-I and rHDL at X- and W-band (9.0 and 94 GHz respective resonance frequencies) from aqueous solutions with the corresponding proton relaxivity values obtained from NMR analysis in vitro. We also consider the information available from EPR of nitroxide spin labels attached to the same cysteine sites. The latter labels provide direct access to local dynamics and water accessibility with high-sensitivity and exceptionally small quantity of protein required. We also demonstrate the utility of nitroxide spin-labeling in predicting MRI contrast when the same sites are labeled with the Gd[MTS-ADO3A] probe.

Selection of protein sites for covalent modification with Gd[MTS-ADO3A] reagent

We have recently shown that engineered human apoA-I MRI contrast agents provide means for visualizing the biodistribution of lipoprotein particles in vivo (12). For this approach single-cysteine apoA-I proteins were specifically and covalently modified at the cysteine thiol-group with the Gd[MTS-ADO3A] contrast agent (14) (Figure 1). The biodistribution of the synthesized apoA-I contrast agents were then followed by real-time MRI analysis of a mouse model system in vivo that revealed significantly enhanced signals in both the liver and kidney.

The preparations of the lipid-bound apoA-I resemble the nascent HDL disc, where two apoA-I molecules surround the periphery of the discoid phospholipid particle (Figure 2) (4,24,25). For initial targeting of the Gd[MTS-ADO3A] label in apoA-I, the site selection was based on the results of site-directed EPR studies employing nitroxide spin labels (2). Specifically, EPR experiments on side chain mobility and nitroxide accessibility to hydrophilic relaxation agents identified a number of surface-exposed residues in the lipid-bound state of apoA-I. Such positions when modified with Gd[MTS-ADO3A] label are expected to offer a high water exchange between the bulk and the inner Gd(III) coordination sphere, as well as effective proton relaxation in the outer sphere. We also sought positions that would not be in a close proximity to one another in the assembled disc (4,24). The following mutation sites Ser52Cys, Ser55Cys, Glu76Cys and Glu80Cys met these criteria and were, therefore, examined for relaxation properties following labeling with Gd[MTS-ADO3A].

Fig. 2.

Single-Cys substitutions were made at the positions 52, 55, 76 and 80 of human apoA-I. (A) Side view of the predicted orientation of positions 52, 55, 76 and 80 on lipid-bound apoA-I. Cross-sections of the two anti-parallel apoA-I monomers (brown and grey circles, respectively) are shown at the edge of the discoidal particle. The extension of the two apoA-I monomers around the entire perimeter of the discoidal phospholipid bilayer is omitted in the figure. (B) Helical wheel representations of the residues 50 to 86 showing the relative locations of the positions 52, 55, 76, and 80 (arrows) along the amphipatic sequence of apoAI. For clarity, the helical projection is rotated 90 degrees clockwise relative to the orientation shown in (A). The color code is dark grey for hydrophobic, white for polar and uncharged, and pink for charged residues. The apolar and polar solvation space is represented by yellow and blue backgrounds, respectively.

X-band EPR analysis of lipid-free apoA-I with the site-directed Gd[MTS-ADO3A] label

Recombinant human apoA-I, engineered to contain single-cysteine replacements at the position 52, 55, 76 or 80, was produced in a bacterial system, affinity-purified and covalently modified (Figure 1) with the Gd[MTS-ADO3A] or MTSL molecules as described earlier (12). To provide an initial evaluation of the feasibility of our approach we first compared X-band EPR spectra of the lipid-free apoA-I protein covalently modified at residue with the Gd(III)-DOTA label (i.e, apoAI-S55C-Gd[ADO3A]) with free Gd[MTS-ADO3A] contrast agent; both at concentrations of 1.5 mM). Although EPR spectra of both samples appear as broad single lines, detailed least-squares simulation analysis revealed significant differences. Specifically, EPR spectrum from free Gd[MTS-ADO3A] contrast agent can only be modeled by a superposition of at least two Lorentzian shapes with peak-to-peak width of 104.1±0.5 and 399±6 G respectively (Figure 3A) while the spectrum of apoAI-S55C-Gd[ADO3A] is well fitted by a single Lorentzian component and a linear baseline term (Fig. 3B). Parameters of the fits including predicted errors obtained by a covariance matrix method (26,27) are summarized in Table 1.

Fig. 3.

X-band EPR spectra comparing equimolar concentrations of free Gd[MTS-ADO3A] label (black, Voight fit in dashed line) and apoAI-S55C-Gd[ADO3A] contrast agent (red, Voight fit in dashed line), which contains the Gd[MTS-ADO3A] label attached position 55 of apoA-I.

Table 1.

Relaxivity values for Gd(III)-labeled apoA-I derived by NMR and line width values of the same samples examined by X-band EPR. The lipid-bound agents are designated “rHDL”.

Contrast Agent	Relaxivity, (mM−1s−1) a	X-band EPR Lorentzian peak-to-peak line for width Component 1, Gd	X-band EPR Lorentzian peak-to-peak line width for Component 2, Gd
free Gd[MTS-ADO3A] b	7.7	104.1±0.5	399±6
apoAI-S55C-Gd[ADO3A] b	2.0	141.7±1.0	n/a
apoAI-S52C-Gd[ADO3A] c	9.7	143.6±1.0	n/a
apoAI-S55C-Gd[ADO3A] c	2.0	140.8±1.0	n/a
apoAI-E76C-Gd[ADO3A] c	4.2	152.6±1.0	n/a
apoAI-E80C-Gd[ADO3A] c	3.9	126.8±1.0
apoAI-S52C-Gd[ADO3A]-rHDLc	7.9	130.9±1.0	n/a
apoAI-S55C-Gd[ADO3A]-rHDLc	11.7	148.2±1.0	n/a
apoAI-E76C-Gd[ADO3A]-rHDLc	5.1	161.4±1.0	n/a
apoAI-E80C-Gd[ADO3A]-rHDLc	4.0	123.9 ± 1.5	n/a

^aWater Relaxivity at 9.4 T magnetic field (25 °C) from previous study (12).

^b[contrast agent] = 1.5 mM. Relaxivity values from previous study (12).

^c[contrast agent] = 0.7 mM

^dLine width data were obtained using EWVoigt program [36, 37] as described in the text. Line width errors correspond to 68% confidence intervals predicted from the fit [ibid.]

At X-band the EPR line shape of Gd(III) complexes is modulated by a partial averaging of the static ZFS due to molecular tumbling. Under conditions of fast tumbling the X-band EPR spectrum of these complexes is dominated by the ±1/2 electronic spin transitions although other transitions are also known to contribute (17,18). Thus, for the Gd[MTS-ADO3A] spectrum we attribute the Component 1 (Figure 3A) to the ±1/2 transition while the broader Component 2 accounts for all the remaining terms. For the apoAI-S55C-Gd[ADO3A] addition of the second component does not improve the fit and satisfactory results could be obtained by modeling the spectrum with just single Lorentzian function and a linear baseline term. The peak-to-peak width increases significantly to 141.7±1.0 G and is indicative of an immobilization of the Gd(III)-DOTA complex through a covalent attachment to the protein. A decrease in rotational tumbling would result in broadening of Gd(III) EPR spectra for all the spin-transitions involved. The ±1/2 transition should be affected to a lesser degree, as the other transitions become too broad to be detected. The latter are likely to contribute to the spectrum only as a drifting baseline. Moreover, fast motion conditions may not be satisfied and this could explain some small but detectable deviations between the fit and the experimental spectrum (Figure 3B). It should be noted here that detailed analysis of slow motion of Gd(III) EPR spectra is not a trivial task as the conditions for the Redfield theory may not be longer satisfied (16,28). Furthermore, such analysis often requires additional assumptions and is facilitated by an extended set of EPR spectra at a number of magnetic fields and frequencies. Without such data readily in hand, we, thus, restrict the discussion to a qualitative analysis summarized above.

The data of the Figure 3 demonstrate that X-band EPR can also be useful for analyzing macromolecular Gd(III)-DOTA contrast agents. However, as this analysis will not discern between the role of the local nano-milieu of the probe and, e.g., the influence of the size-dependent rotational velocity, we next analyzed the spectral properties of the apoA-I single cysteine mutants modified at residues 52, 55, 76 or 80 with the Gd[MTS-ADO3A] reagent.

X-band EPR analysis of the site-directed Gd[MTS-ADO3A] label at positions 52, 55, 76 and 80

Despite the large line width, X-band EPR measurement of apoA-I labeled with Gd[MTS-ADO3A] resolved the differences in the peak-to-peak width depending on whether the probe is attached to the position 52, 55, 76 or 80 (Figure 4). Specifically, in the lipid-free state the peak-to-peak width increased measurably from 140.8±1.0 G for the position 55 to 161.4±1.0 G for the position 52. Such a difference could be attributed to an increase in the static ZFS (29) of the Gd(III) or a slower local tumbling of the chelated ion (21). However, as was shown by Smirnova et al. (23) an increase in static ZFS would cause a shift of the X-band downfield due to the second-order effects. The latter shift was not observed in the experiment, as the position of the lines remained virtually the same (Figs. 3 and 4). Thus, it appears that local dynamics (discussed in more detail below) is the main source of modulation for the X-band EPR line width of the attached Gd[ADO3A] probe. Furthermore, these differences in local dynamics, along with the accessibility of the Gd(III) to the bulk water (also described in more detail below) can account for the large difference in contrast relaxivity observed for the probes attached to different positions or when the protein is in a different conformational state. Moreover, when the same protein sites were labeled with nitroxide MTSL, the corresponding EPR spectra revealed different local dynamics (see below). Notably, the increase of ~8 G in the Gd(III) peak-to-peak EPR linewidth correlates with a more than two-fold increase in the proton relaxivity between the lipid-free- and rHDL state of apoA-I when the apoAI-S55C-Gd[ADO3A] sample converts from the lipid-free to the rHDL state (Table 1). This indicates that in NMR observations at 9.4 T, the rotational correlation time of the Gd[ADO3A] side chain is one of the shortest dynamic parameters modulating the dipolar relaxation process.

Fig. 4.

X-band EPR spectra of apoA-I 52, 55 and 76 with attached Gd[ADO3A] in the presence (rHDL) or absence of lipid. A) ApoA-I lipid-free (black) and lipid-bound (red) form with the attached Gd[ADO3A] at the position 52. B) ApoA-I lipid-free (black) and lipid-bound (red) form with the attached Gd[ADO3A] at the position 55. C) ApoA-I lipid-free (black) and lipid-bound (red) form with the attached Gd[ADO3A] at the position 76. D) ApoA-I lipid-free (black) and lipid-bound (red) form with the attached Gd[ADO3A] at the position 80. The concentration of protein was 10 mg/ml. Voight fits of the spectra are shown as dashed lines.

Analysis of lipid-bound apoA-I with the site-directed Gd[MTS-ADO3A] label

As infused lipid-free apoA-I is expected to rapidly exchange with endogenous lipoprotein particles in vivo in the form of HDL structures (30,31), we also examined EPR spectra of the Gd[MTS-ADO3A] and nitroxide-labeled apoA-I proteins in their lipid-bound discoidal particle state (reconstituted HDL; rHDL). Specifically, for the 55 and 76 sites we have observed a further increase in the Gd(III) EPR peak-to-peak width for the protein in the lipid bound states (Figure 4; Table 1), whereas the 52 site (and also the 80 site but to a lower extent) exhibits a decrease in the Gd(III) EPR peak-to-peak width in the lipid bound state. Thus, as the X-band EPR linewidth of the apoAI-Gd[ADO3A] is dependent on the local environment of the four unique positions within the protein, the spectra are likewise sensitive to the conformational rearrangement of the protein that occurs upon lipidation.

Analysis of Gd(III) EPR spectra by high frequency EPR at W-band (94 GHz)

As has been shown previously, an increase in the magnetic field decreases the relative contribution to the EPR line width arising from ZFS, thus, causing a substantial line narrowing (23). This improves concentration sensitivity and simplifies the spectral analysis as the line becomes Lorentzian in shape [ibid.]. While the static contributions to the electronic ZFS are suppressed at the magnetic fields above ca. 3 T, modulation of the transient ZFS by anharmonic vibrations and solvent collisions may start to play a role (21).

When the W-band EPR spectra from label positions and the two protein states were compared, they were found all to be essentially identical (Figure 5): neither apparent g-factor shifts nor changes in the peak-to-peak width of 13.3±0.1 G were observed. Notably, previous studies have demonstrated that due to significant line width narrowing and remaining second-order line shifts, the W-band frequency is near the optimal to resolve changes in EPR spectra of Gd(III) chelates with different static ZFS (23). Thus, the absence of any field shifts in W-band EPR spectra for all 4 protein samples studied indicates that the static ZFS is unaffected by either changes in the protein attachment sites or binding to the lipid particles. Furthermore, the peak-to-peak width of the W-band spectra was virtually the same indicating the same magnitude of the transient ZFS effects. To conclude, the close identity of the experimental W-band spectra indicates that both static and dynamic ZFS of the Gd(III) complex remain the same. Thus, the only other parameters that could affect the proton relaxivity of the studied paramagnetic contrast agents would be rotational tumbling and the water exchange rates. The latter are known to be of the order of 200–300 ns and are even longer when one acetate arm is replaced by amide functions as in ADO3A (14).

Fig. 5.

High frequency (W-band) EPR scan of lipid-bound and lipid-free apoA-I containing the Gd[ADO3A] label at position 55 (A) or 76 (B). No differences are observed among any of the spectra, indicating an absence of g-factor perturbation.

Local dynamics at positions 52, 55, 76 and 80 of apoA-I

Typically, the water exchange rate for Gd(III) complexes is evaluated through variable temperature transverse ¹⁷O relaxation rate measurements. Unfortunately, such measurements require water labeled with a rare and expensive ¹⁷O isotope in relatively large (~500 μl) quantities suitable for NMR. Concentration of Gd(III) complexes should also be raised to ca. 5 mM but such concentrations are often unattainable for Gd(III)-labeled proteins. Here we analyze the observed differences in the contrast agent relaxivity without the expensive ¹⁷O relaxation rate measurements and at protein concentrations of only ca. 1 mM. We note that the inner ligands of Gd(III) are the same for all the compound studied here. Thus, we hypothesized that the main difference in relaxivity may arise from solvent exposure of the labeled side chain rather than the structure and dynamics of the Gd(III) inner sphere.

In order to more closely evaluate the local dynamics at positions 52, 55, 76 and 80 in lipid-free and lipid-bound apoA-I, we attached the MTSL nitroxide label at each position and examined the proteins by X-band CW EPR spectroscopy. Previously, the utility of this nitroxide label for resolving both backbone and side chain dynamics (32,33), as well as providing direct access to solvation parameters (34) has been demonstrated for several protein systems. With apoA-I, our extensive application of nitroxide spin labels has been used to identify conformational switching, secondary structure and tertiary and quaternary alignments (2–4,8,24,35). Here the EPR spectra of MTSL nitroxide labels at positions 52, 55, 76 and 80 in lipid-free apoA-I reveal a relatively high degree of motional freedom for all four nitroxide-labeled side-chains (Figure 6). This mobility appears to be largely preserved upon lipidation of the proteins. However, the spectra of nitroxide-labeled position 52, 55 and 80 depict an overall higher mobility than at position 76 (Figure 6). Specifically, the position 76 displays a significantly immobilized population, evident in the broad shoulder of its low-field peak (Figure 6). This is also reflected in the peak-to-peak line width values (ΔH_pp) of the central, m_I=0 nitrogen hyperfine transition, with the lipid-free apoA-I spin-labeled at position 52 displaying the highest mobility (Table 2). ΔH_pp has been shown to provide a simple indicator of the rates and amplitudes of MTSL motion on protein surfaces (33), and constraints on its motion (or lack thereof) may prove informative in selecting more ordered locations for Gd[MTS-ADO3A] targeting. Figure 7 shows a reasonable correlation between the nitroxide and Gd(III) linewidths for the unique samples, with the exception of protein labeled at position 52. Because the solvent accessibility of this sample is relatively high (see below), the large ΔH_pp for this sample likely arises from a more ordered backbone or stronger interaction therewith.

Fig. 6.

EPR scans of apoA-I (lipid-free and lipid-bound) labeled with the nitroxide spin probe at position 52 (A), 55 (B), 76 (C), or 80 (D). Spectral parameters are given in Tables 1 and 2. The concentration of protein was 1 mg/ml.

Table 2.

X-band EPR mobility and accessibility parameters for nitroxide-labeled apoA-I in the lipid free and lipid-bound (rHDL) states.

Nitroxide-labeled apoA-Ia	Δ Hpp (G)b	ΠCrxc	Φd
apo AI-S52C	2.44	1.69	−2.02
apo AI-S55C	2.89	1.44	−2.42
apo AI-E76C	3.14	1.30	−2.34
apo AI-E80C	2.73	1.52	−1.87
apo AI-S52C-rHDL	3.22	1.97	−2.48
apo AI-S55C-rHDL	2.94	1.77	−2.67
apo AI-E76C-rHDL	3.23	0.71	−1.61
apo AI-E80C-rHDL	2.83	0.91	−1.54

^adata from previous studies (2, 4)

^bpeak-to-peak line width of the central nitroxide (m_I=0 nitrogen hyperfine transition) line

^cpolar accessibility parameter

^dcontrast function

Fig. 7.

Plot of line width parameters of X-band EPR spectra from apoA-I labeled with the Gd[MTS-ADO3A] (Table 1) or MTSL (Table 2).

Although all the EPR spectra shown in Figure 6 could be characterized as relatively mobile when compared to other apoA-I sites studied previously (2–4,8,24,35), from a perspective of simulation analysis these spectra correspond to an intermediate or slow motion regime. The latter is defined by condition τ_RΔω≥1, where τ_R is a rotational correlation time and Δω is a measure of the magnitude of the orientation-dependent part of the spin Hamiltonian. These slow-motional line shapes were further analyzed using a theoretical approach based on a numerical solution of the stochastic Liouville equation (36,37). Recently, slow motional MTSL EPR spectra of spin-labeled T4 lysozyme were extensively examined by Freed and coworkers by this approach through simultaneous simulations of experimental spectra measured over an exceptionally broad range of magnetic fields and frequencies, i.e., at 9, 95, 170, and 240 GHz using slow relaxing local structure (SRLS) model (27). Here we analyze EPR spectra of apoA-I measured at a single frequency using slightly varied magnetic parameters reported by Freed and coworkers for MTSL-labeled proteins (27). With EPR spectra measured at a single frequency, we chose a somewhat simpler dynamic model of an axially symmetric rotational diffusion in a local orienting potential known as MOMD (microscopic order with macroscopic disorder), which provides the quantitative rotational diffusion tensors R_II and R_⊥(26) that have been previously employed for modeling EPR spectra of T4 lysozyme labeled with MTSL (38). We also neglect effects of the overall protein tumbling because for a protein of molecular weight of ~29 kDa as apoA-I the main source of motional averaging of nitroxide spectra is expected to originate from local segmental motion.

MOMD simulation for lipid-free and lipid-bound apoA-I labeled with MTSL at positions 55 and 80 produces a satisfactory fit using a single component model (Figure 8) with the best fit parameters summarized in the Table 3. In contrast, least squares simulations of EPR spectra from the positions 52 and 76 reveal the presence of two or more components (Figure 9 and Table 3). However, the predominant component in these samples is similar to the single component provided by the spectra of position 55, and corresponds to an intermediate motion regime. Regarding the additional components in the 52 and 76 spectra, an immobile component, corresponding to a slowly reorienting nitroxide side chain, is present in the lipid-bound state of the position 52 and both states of the position 76. Both positions 52 and 76 show a minor population of mobile spins tumbling on a fast (R_⊥=8–20 × 10⁸ s⁻¹) time scale. Notably, our three component simulations are in agreement with previous SRLS analysis of T4 lysozyme spectra that were explained by protein local conformations and protein-nitroxide ring interactions (27). Similarly to the SRLS simulations, we find perpendicular component of rotational diffusion, R, greater than the parallel one, R_II, for all the mutants, samples and components. Finding R_⊥>R_II is in agreement with the geometry of the spin label, which has R_⊥ diffusion axis approximately parallel to the last χ₅ tethering bond (see Figure 1). The immobile component has R_⊥ that is nearly an order smaller than the intermediate component but exhibits a similarly strong ordering: the first coefficient representing the ordering potential in the series of Wigner rotational matrices is c₂₀≈0.7. The mobile component corresponds to an unrestricted axial diffusion with very small or even negligible ordering effects. Diffusion tilt is of about 56° and this is somewhat larger than the 20–30° angles found by others (26,27).

Fig. 8.

Least-squares simulation analysis of X-band EPR spectra of the indicated MTSL-labeled apoA-I samples. Experimental spectra are shown in black while the simulations assuming one spectral component and MOMD model are given as red lines. See text for further details and Table 3 for simulation parameters.

Table 3.

MOMD simulation parameters for nitroxide-labeled apoA-I in the lipid free and lipid-bound (rHDL) states.

Nitroxide-labeled apoA-Ia	R⊥(107s−1)b	RII (107s−1)b	Component Fraction (%)c
apo AI-S52C	6.3, 25, 79	0.3, 0.1, 2.0	35, 60, 5
apo AI-S55C	x, 20, x	x, 0.6, x	0, 100, 0
apo AI-E76C	4.0, 20, 100	0.3, 0.1, 2.0	36, 59, 5
apo AI-E80C	x, 16, x	x, 0.4, x	0, 100, 0
apo AI-S52C-rHDL	x, 10, 200	x, 2.5, 0.6	0, 98.5, 1.5
apo AI-S55C-rHDL	x, 20, x	x, 0.1, x	0, 100, 0
apo AI- E76C-rHDL	3.2, 16, 79	0.3, 0.2, 3.2	25, 68, 7
apo AI- E80C-rHDL	x, 12, x	x, 0.35, x	0, 100, 0

^aThe following nitroxide magnetic parameters were used for simulations: g= (2.00839, 2.00616, 2.00232), A=(6.06, 4.22, 38.80 G), diffusion tilt parameter β_d = 54–56°, orienting potential coefficient c₂₀ = 0.7.

^bRate of motion along the principal nitroxide axis in either the perpendicular or parallel directions for the immobile, intermediate, and mobile spectral components, respectively. Note that for Brownian axially anisotropic diffusion the rotational correlation time is calculated as (τ_R)⁻¹=6(R_⊥R_II)^1/2.

^cPopulations of immobile, intermediate, and mobile spectral components, respectively.

Fig. 9.

Decomposition of X-band EPR spectra of MTSL-labeled apoA-I samples showing more than one motional component. Experimental spectra are shown in black while the simulations assuming three spectral component and MOMD model are shown as colored dashed lines. See text for further details and Table 3 for simulation parameters.

Even though the EPR spectra from positions 52 and 76 have additional components present; the most dominant one (59–98%) has about the same dynamic parameters (within the errors of simulations) as the single component spectra provided by apoA-I bearing a spin label at positions 55 and 80. On top of that we see only moderate effects of lipid binding on the overall local label dynamics. As shown in Figure 10, one could relate a higher NMR relaxivity of the lipid-bound 55 protein by it being dominated by the slowest R_II, but the X-band EPR spectra are not sufficiently sensitive to changes in the diffusion rate in this range and the simulated parameters are less significant. Nevertheless, it is useful to consider the significance of potential relationship between R_II and the Gd[ADO3A] on the surface of proteins as this parameter does identify the two samples with the highest relaxivity 55 LB (apoAI-S55C-Gd[ADO3A]-rHDL) and 52LF (apoAI-S52C-Gd[ADO3A]). The dynamics of spin labels on the surface of a protein have been investigated in detail experimentally and the modes and rates of motion of the MTSL adequately described to generate simulations of observed spectra (see (27) and references therein). The results of these efforts established that the principal modes of diffusion for the spin-labeled nitroxide MTSL side chain involve rotations about the χ₄ and χ₅ bonds (see Figure 1) as well as backbone fluctuations. For surface sites R_⊥ is one- to two-orders of magnitude faster than R_II, reflecting rapid backbone motions and rotations about the χ₅ bond, which are relatively unrestricted on the protein surface regardless of the volume of the neighboring residues (27,39). R_II, which is approximately parallel to the z-axis of diffusion, is modulated by rotations about the χ₄ bond. Thus, rates of motion about the nitroxide χ₄ bond for a given sample may provide a predictor for the reorientations of a much bulkier Gd[ADO3A] probe at the same site.

Fig. 10.

Plot of the R_II diffusion tensor of nitroxide labeled protein versus NMR relaxivity observed for the Gd[ADO3A] probe located in the same sample.

Solvent accessibility of positions 52, 55, 76 and 80

Collision frequency of nitroxide-labeled side chains (as measured by power saturation EPR) with hydrophilic and non-polar relaxation agents reports on the local polarity (40). The collision frequency is expressed as the normalized parameter Π, and was determined for each nitroxide labeled protein in the presence of non-polar O₂ (aqueous solution equilibrated with air) and 20 mM chromium oxalate (crx) that will access nitroxide-labeled side-chains localized to non-polar and polar environments, respectively. As shown by the polar accessibility parameter values (Π_crx; Table 2) residue 52, 55, 76 and 80 of lipid-free apoA-I are localized to polar environments (Π_crx values of 1,69, 1.44, 1.30 and 1.52, respectively). Upon lipidation and formation of rHDL, the local environment of the nitroxide-label at residues 52 and 55 become even more polar (Π_crx increases to 1.97 and 1.77, respectively, for positions 52 and 55) and, thus, retains its high accessibility from the aqueous phase. However, for residues 76 and 80 the polar accessibility value is significantly decreased (from 1.30 to 0.71, and 1.52 to 0.91, respectively)) after the conversion to rHDL. The apparent decrease in polarity can possibly be explained by the quaternary organization of the apoA-I proteins in the lipid-bound HDL discs where the side chain of residue 76 (and possibly also residue 80) is expected to position at the protein-protein interface between the two apoA-I proteins ((4); Figure 2). Such a shielding effect would thus make the labeled side-chain less accessible for the polar signal quencher.

The contrast function value (Φ), which in addition to the hydrophilic quencher also takes into account the accessibility of an apolar quencher (O₂) as well as steric effects, was then used to further evaluate the local chemical environment. Similar to the polar accessibility value, a significant increase in the contrast function value (inversely related to the polarity of the experienced local environment) for the nitroxide-labeled residue 76 (and to a lower extent for nitroxide-labeled residue 80) upon lipidation and rHDL formation was observed whereas the environmental changes for the residues 52 and 55 were less pronounced and in the opposite direction, i.e., in a more polar milieu after lipidation.

Correlation between Gd(III) line widths and local solvation with proton relaxivity

If the rotational correlation time of the Gd(III) chelates is the only factor determining proton relaxivity at 9.4T and other contributing parameters including electronic ZFS do not change, then the experimental peak-to-peak EPR line width observed at X-band should approximately correlate with the observed relaxivity. Indeed, assuming that the Redfield theory and the fast motion conditions are satisfied for both EPR and NMR experiments, the EPR line width should be proportional to the rotational correlation time, as influenced by the rate of dipolar relaxation. Although consideration of the X-band Gd(III) EPR line width alone is not sufficient to predict the magnitude of relaxivity expected from a specific position in the protein, we find a consistent correlation between the changes in EPR line width and relaxivity for each of the three positions when apoA-I transitions from the lipid-free to the rHDL state (Table 1). When the Gd[ADO3A] label is located at positions 55 and 76, lipidation results in an increase in the relaxivity that accompanies the increase in EPR linewidth. Likewise, as apoAI-S52C-Gd[ADO3A] experiences a narrowing of its X-band Gd(III) spectrum upon lipidation, a lower relaxivity from this site is generated in the rHDL state, whereas both the relaxivity value and the X-band Gd(III) line width for apoAI-E80C-Gd[ADO3A] are essentially unchanged upon lipidation.

With no evidence for differences in Gd(III) ZFS or electronic relaxation in this sample, the exceptional relaxivity for apoAI-S55C-Gd[ADO3A]-rHDL is likely attributed to increases in the water exchange rate and/or the water accessibility of the chelated ion. Although the label is different in size and chemistry, the Gd[ADO3A] water exchange/accessibility should be related to the collisional frequency of a hydrophilic relaxer measured for the nitroxide at the same position. Of the eight samples examined, the nitroxide-labeled apoAI-S52C displays the highest collision frequency with chromium oxalate (Π_crx; Table 2). A similarly high Π_crx value for nitroxide-labeled 55C in rHDL also predicts an increased solvation of the Gd[ADO3A] probe, and together with the X-band Gd(III) linewidth, correlates with the NMR relaxivity results (see below).

Thus, evaluating Gd(III) relaxivity on the basis of its X-band line width and the nitroxide accessibility to a hydrophilic relaxer at the same position may prove to be exceptionally useful in predicting the contrast potential for the engineered agent. As shown in Figure 11 there is an apparent correlation between the relaxivity and an increase in Gd(III) line width relative to the free label when scaled by the chromium oxalate accessibility (Π_crx) of a nitroxide located at the same position. On this basis, the lower than expected relaxivity for lipid-free apoAI-S55C-Gd[ADO3A] and apoAI-E76C-Gd[ADO3A] could perhaps be attributed to a proportionally higher water accessibility for the nitroxide at this location compared to the Gd[ADO3A]. In general, the environment of the rHDL disc may be more favorable for predicting relaxivity given the overall increase in backbone order that occurs upon lipidation (4,35), which in the case of apoA-I contrast agents for in vivo MRI would be advantageous as apoA-I rapidly exchanges with blood plasma HDL (30, 31). By combining Gd(III) line widths with nitroxide accessibility to the aqueous solution, one should be able to predict the positions that will achieve greater contrast.

Fig. 11.

Plot of relaxivity versus X-band line width scaled by the nitroxide accessibility parameter Π_crx for each engineered contrast agent. (52 LF) apoAI-S52C-Gd[ADO3A]; (55 LF) apoAI-S55C-Gd[ADO3A]; (76 LF) apoAI-E76C-Gd[ADO3A]; (80 LF) apoAI-E80C-Gd[ADO3A]; (52 LB) apoAI-S52C-Gd[ADO3A]-rHDL; (55 LB) apoAI-S55C-Gd[ADO3A]-rHDL; (76 LB) apoAI-E76C-Gd[ADO3A]-rHDL; (80 LB) apoAI-E80C-Gd[ADO3A]-rHDL. The scaled parameter Δlw × Π_crx was calculated from the change in the X-band Gd(III) line width relative to the free probe in solution (Table 1) multiplied by the hydrophilic accessibility parameter Π_crx (Table 2).

Conclusions

We conclude that putative site-directed protein contrast agents can be screened by X-band EPR spectroscopy employing both the nitroxide and Gd(III)-DOTA probes. Specifically, it was shown that the scaled EPR parameter Δlw × Π_crx correlates with the degree of water proton relaxivity for these MRI contrast agents. These studies demonstrate the utility of EPR technique for screening protein candidates as vehicles for carrying out MRI contrast agents. The advantage of EPR spectroscopy is that the concentration of the protein is significantly lower (~10 μM spin for nitroxide-labeled and ~1 mM for Gd(III)-labeled sample) than that it is necessary in NMR spectroscopy (>100 μM protein). Moreover, the sample volume for EPR experiments (~5 μl) is about two orders of magnitude smaller than that used for NMR measurements (~500 μl). New studies to investigate the relaxivity mechanism at different magnetic fields are warranted, especially lower fields used in clinical settings, and studies will explore double-Cys mutants that will target MRI contrast probes to other candidate sites that will be initially identified by nitroxide spin labels (2–4,35) to display high solvation and ordered secondary structure. Along these lines, efforts to more tightly constrained chelators via two functional groups have demonstrated the benefit of having the probe’s rotation more closely coupled to the rate of the macromolecule (41), and in cases where two Cys residues can be engineered in close proximity, bis-MTS derivatives can be employed (42).

Material and Methods

Materials

Thiol-specific nitroxide spin label (MTSL; (1-oxyl-2,2,5,5-tetramethylpyrroline-3-methyl) methanethiosulfonate) (43) was received as a kind gift from Dr. K. Hideg (University of Pecs, Hungary). 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) was purchased from Avanti Polar Lipids, Inc. (Alabaster, AL). Tris (2-carboxyethylen) phosphine hydrochloride (TCEP), isopropylthiogalactoside (IPTG) and ampicillin were purchased from Sigma-Aldrich (St. Louis, MO).

Preparation of modified apoA-I

Recombinant hApoA-I was produced in a bacterial expression system comprised of the human apoA-I gene inserted into the pET20b plasmid as described previously (44). Single Cys mutations of residues 52, 55 76 and 80 in the apoA-I cDNA were created by oligonucleotide-directed PCR mutagenesis. The mutations were confirmed by dideoxy automated fluorescent sequencing. The plasmid containing the specified Cys substitution was transferred into the Escherichia coli strain BL21 (DE3) cells (Invitrogen, Carlsbad, CA) and cultivated in the Luria-Bertani broth medium (Miller, Fair Lawn, NJ) supplemented with 50 μg/ml of ampicillin. IPTG was added at mid-exponential log-phase to induce the expression of the proteins. Mutant human apoA-I proteins were extracted and purified by immobilized metal affinity chromatography (HiTrap columns, GE Healthcare, Waukesha, WI) under denaturing conditions (3 M guanidine-HCl), extensively washed with phosphate-buffered-saline (PBS; 20 mM phosphate, 500 mM NaCl), pH 7.4, and then eluted by increasing concentrations of imidazole. The yield was 20 mg of apoA-I per 1 L of cell culture.

Gd[MTS-ADO3A] was synthesized as reported previously (14), and conjugated to apoA-I to form the labeled proteins Gd[MTS-ADO3A] apoAI-S52C, Gd[MTS-ADO3A] apoAI-S55C, Gd[MTS-ADO3A] apoAI-E76C or Gd[MTS-ADO3A] apoAI-E80C. For modifications of the engineered cysteines, purified apoA-I proteins were concentrated to ~5 mg/ml with a 20 kDa molecular weight cut-off Vivaspin 2 CTA centrifugal filter devices (Vivascience AG, Hannover, Germany). For labeling, 1 mM Gd[MTS-ADO3A] was incubated with the purified apoAI-S52C, apoAI-S55C, apoAI-E76C or apoAI-E80C for 30 min at room temperature, followed by running the sample through a Bio-spin 6 Tris Column (BioRad, Hercules, CA) equilibrated in PBS at pH 7.4 to remove any unreacted label as previously described (12). The protein was further concentrated using the Vivaspin centrifugal device to a final concentration of 20 mg/ml (or 0.7 mM). Labeling with MTSL was performed on-column after pre-treatment with 100 μM TCEP (reducing agent) for 10 min as previously reported (2). To generate the final lipidated apoA-I contrast agent, the labeled apoA-I was combined in 1:100 (mole protein:mole lipid) with small unilamellar liposomes of DMPC prepared by extrusion through 100 nm filters according to manufacturer’s instructions (Avestin, Inc., Canada).

Sodium dodecyl sulfate (SDS)-polyacrylamide gel electrophoresis with Coomassie blue staining was employed for analyzing the protein purity. The concentration of total protein in the samples was determined by BCA Protein Assay Kit (Thermo Scientific, Waltham, MA), using bovine serum albumin (BSA) as a standard.

EPR spectroscopy

Electron paramagnetic resonance measurements were carried out at room temperature with a JEOL (JEOL USA, Inc., Peabody, MA) X-band spectrometer fitted with a loop-gap resonator to confirm the binding of the label to the apoA-I. Typically, EPR spectra were obtained by averaging two scans of 2 min each at a microwave power of 4 mW and modulation amplitude optimized to the natural line width of the attached spin probe. All spectra were recorded at room temperature. To estimate the changes in the peak-to-peak width of X-band Gd(III) spectrum, experimental data were fitted to a superposition of Lorentzian or a Gaussian-Lorentzian convolution line shape – a Voigt function - using EWVoigt software that included automatic phase correction (45,46). The same software allowed for assessing EPR spectra for any changes induced by the lipid binding. The accessibility of MTSL-labeled sites to polar (20 mM chromium oxalate) and non-polar (O₂, obtained by equilibrating protein solution with air) relaxation agents was determined from their collisional frequency (Π) with paramagnetic relaxers using power saturation EPR as previously described (40). Differences in chemical environment about the nitroxide spin label were identified using the hydrophobicity contrast parameter Φ = ln(Π_nonpolar/Π_polar) (40).

W-band (94.3 GHz) EPR spectra were acquired using a spectrometer developed at NCSU (47). In brief, the spectrometer employs a cryogen-free superconducting magnet (Cryogenic Ltd., London, UK) that includes a ±0.12 T superconducting sweep coil to provide accurate scans of the magnetic field in the vicinity of the target value. The microwave bridge was of a homodyne single-channel design similar to the one described earlier (48). Aqueous solutions of Gd(III) complexes were drawn into 0.15 mm i.d. quartz capillaries (VitroCom, Mountain Lakes, NJ) and carefully positioned exactly at the center of a cylindrical TE₀₁₂-type W-band resonator. When loaded with liquid aqueous samples, the quality factor of this resonator remained exceptionally high (Q≈3500–2500). All W-band spectra were measured at room temperature at incident power less than 0.5 mW and modulation amplitude not exceeding 4 G.

Simulation of nitroxide spectra

Least-squares simulation of slow/intermediate motion nitroxide EPR spectra were carried out using a PC version of NLSL software as well as its Matlab computational subroutines (36) that were further bundled at NCSU with customized searching routines found in the Matlab Simulation Toolbox. EPR spectra of the protein, both lipid bound (LB) and lipid free (LF), mutated at positions 52, 55, 76 and 80 with attached MTSL nitroxide label have been analyzed for rotational diffusion dynamics. The X-band spectra fall in slow motional regime, τ_RΔω≥1, where τ_R is a rotational correlation time and Δω is a measure of the magnitude of the orientation-dependent part of the spin Hamiltonian. These slow-motional line shapes can be fully analyzed using a theoretical approach based on numerical solution of the stochastic Liouville equation (36,37). In general, simulations should include an overall tumbling of the protein as well, which can be done within the slowly relaxing local structure (SRLS) model. Such slow motional MTSL EPR spectra were extensively studied by Freed et al. (27). Here we used slightly varied magnetic parameters obtained by Freed et al. As a dynamic model we used axially symmetric rotational diffusion in local orienting potential known as MOMD (microscopic order with macroscopic disorder) (26), neglecting the overall protein tumbling. For the actual simulation we used PC version of the NLSL software as well as its Matlab computational subroutine bundled with customized searching routines found in Matlab Simulation Toolbox (36). Up to three component spectra simulations are in agreement with SRLS simulations, and are explained by protein local conformations and protein-nitroxide ring interactions. The spectrum of apoA-I with the spin label located at position 52 exhibits three resolved components in the lipid free form. However, in the lipid bound state, the spectrum of apoA-I 52C is dominated by a single intermediate component. Spectra from lipid-free and lipid-bound samples with spin label placed at positions 55 and 80 have been simulated as single component spectra. Finally, spectra from position 76 (lipid-free and lipid-bound) were, like position 52 lipid-bound, assumed to have two more components in addition to the main (intermediately mobile) component, a slowly reorienting (immobile) and fast (mobile) components.

NMR-relaxivity and contrast measurements

A 9.4T Bruker (Billerica, MA) Avance NMR spectrometer equipped with a 5 mm probe was employed for measuring the longitudinal (spin-lattice) relaxation time T₁ of water protons in the presence and absence of gadodiamide and the Gd[MTS-ADO3A] apoA-I derivatives. The ¹H 90° pulse, calibrated against the H₂O signal from a 0.15 M aqueous NaCl solution or perfusate, was 10 μs. A typical spectrum required 8 scans and used the following signal acquisition parameters: 8,012-Hz spectral width, 16,384 data points, and 1 s acquisition time. Zero-filling the free induction decay (FID) and apodizing with an exponential window function have been employed to improve the spectra. The H₂O peak at 4.75 ppm served as the spectral reference at 25 °C relative to sodium-3-(trimethylsilyl)propionate-2,2,3,3-d4 at 0 ppm. A three parameter exponential fit mapped the spin-lattice relaxation time of the samples. The probe was detuned by 5 MHz to prevent radiation damping due to the high Q coil employed. The measured T₁ of water (T_1obs) in the presence of Gd based contrast agents at a concentration of 0.7 mM was then used to calculate the relaxivity (r₁) of the contrast agent using the following equation:

$$\frac{1}{{\left({\text{T}}_1\right)}_{\text{obs}}}=\frac{1}{{\left({\text{T}}_1\right)}_{\text{water}}}+{\text{r}}_1\left[\text{Gd}\right]$$

Where T_1water is the T₁ of water in the absence of contrast the agent.