Evaluation of polyol‐made Gd3+‐substituted Co0.6Zn0.4Fe2O4 nanoparticles as high magnetization MRI negative contrast agents

The structural, microstructural, and magnetic properties of ~5‐nm‐sized Co0.6Zn0.4Fe2 − xGdxO4 nanoparticles were investigated in order to evaluate their capability to enhance the magnetic resonance imaging contrast as high magnetization agents. A focus was made on the solubility of Gd3+ cations within the spinel lattice. By coupling X‐ray diffraction to X‐ray fluorescence spectroscopy, we demonstrated that only a limited fraction of Gd3+ can substitute Fe3+ ions into the whole crystal structure and does not exceed 6 at.‐%. At this concentration, the room temperature (27°C) saturation magnetizations of the prepared superparamagnetic nanocrystals were found to be close to 80 emu g−1. Coating these nanoparticles with hydrophilic dopamine ligands leads to the formation of ~50‐nm‐sized clusters in water. As a consequence, relatively high r2/r1 ratios of transverse to longitudinal proton relaxivities and high r2 values were measured in the resulting colloids at physiological temperature (37°C) for an applied magnetic field of 1.41 T: 33 and 188 mM−1 sec−1, respectively, for the richest system in gadolinium. Moreover, after incubation with healthy human model cells (fibroblasts) at doses as high as 10 μg mL−1, they induce neither cellular death nor acute cellular damage making the engineered probes particularly valuable for negative magnetic resonance imaging contrasting.


Introduction
Superparamagnetic iron oxide nanoparticles (SPIONs) are largely used as negative contrast agents for magnetic resonance imaging (MRI), because of their ability to shorten the longitudinal T 1 and transversal T 2 relaxation times of the nuclear magnetic moment of water protons. According to the model of Gillis et al. (1999) named "Outer Sphere" mechanism, the increase in the relaxation rate 1/T 2 compared with pure water originates from fluctuating dipolar interactions between nuclear spins of water protons and the electronic magnetic moment of SPIONs. This increase is as important as the transversal relaxivity, r 2 , of SPIONs, defined as the slope of the variation of 1/T 2 as a function of the iron atom concentration, is high, and then as the contrasting efficiency is improved. Vuong et al. (2012) have shown that r 2 follows a universal scaling law that is quadratic both with the volume magnetization and with the radius of the outer sphere, d, also called the relaxometric size: d is defined as the minimum approach distance between H 2 O molecules and the SPIONs and very often approximated with the hydrodynamic diameter of the particles. They defined two limit water diffusion regimes, depending on the magnetic contrast agent characteristics. The first one is called the "motional averaging regime" (MAR), for which the protons of freely diffusing water molecules surrounding the particle explore all the possible values of magnetic dipolar field created by the SPIONs. The second one is called the "static dephasing regime" (SDR), for which, during one echo time of the applied radiofrequency sequence, water molecules explore only a small space compared with the hydrodynamic particle volume. In these regimes, the r 2 relaxivity can be expressed as follows, where equation 1a is related to the MAR and equation 1b to the SDR: ; (1a) where D measures the water translational diffusion constant (D = 3·10 À9 m 2 sec À1 at 37°C; Solomon, 1955;Bloembergen and Morgan, 1961), v mat is the iron oxide molar volume defined by the ratio of the molar mass divided by the number of magnetic ions in the formula unit and by the mass density (v mat = 1.56·10 À5 and 1.49·10 À5 m 3 mol À1 for maghemite and magnetite, respectively), μ 0 corresponds to the magnetic permeability of vacuum (μ 0 = 4π·10 À7 T m A À1 ), and γ is the gyromagnetic factor of proton (γ = 2.67513·10 8 rad sec À1 T À1 ). The transition between the two regimes is determined by the so-called Redfield condition (Vuong et al., 2012), in which Δω = (1/3) (γμ 0 M v ) is the angular frequency shift experienced by a water proton, at the closest distance to the particle surface, and τ D = d 2 /4D measures the translational diffusion time of a water molecule around the SPION sphere: Δω·τ D < 1 means the predominance of the MAR regime, while Δω·τ D > 1 means the predominance of the SDR regime.
The first regime (MAR) is the most common for small single nanoparticles (NPs) of pure magnetic materials or with a thin fully hydrated shell and for hybrid entities whose overall magnetization M v (the total magnetic moment divided by the particle volume) remains small compared with the specific magnetization of their inorganic part (m S ). In that case, the magnetic cores are either dispersed individually or clustered yet wrapped by a rather thick coating (silica, polymer) impermeable to water (thus, the magnetization is averaged over the whole outer sphere diameter instead of just the magnetic core). According to this model, the design of powerful negative contrast agents for MRI is dictated by three parameters only: the intrinsic magnetization of their constituting magnetic core, their equivalent outer sphere diameter, and the volume fraction of magnetic materials contained in it. The three parameters must be as large as possible, which is favored by compact structures (having a high volume magnetization) rather than loose aggregates (having a small magnetic volume fraction). Increasing these three parameters will increase the r 2 relaxivity up to a maximum, corresponding to the MAR-SDR transition. Besides, the hydrodynamic diameter of MRI contrast agents must not exceed an upper limit classically estimated around 300 nm to enable sufficiently long half-life inside the blood vessels (Okuhata, 1999), when intravenously administrated, otherwise, they are rapidly cleared by the reticuloendothelial system.
Starting from iron oxide spinel system, and particularly the most used maghemite one, the magnetization increase can be achieved by replacing it by cobalt ferrite. The magnetization of bulk CoFe 2 O 4 is about 85-90 emu g À1 while that of bulk γ-Fe 2 O 3 does not exceed 75 emu g À1 (Valenzuela, 1995). CoFe 2 O 4 nanoferrites have been widely considered for various nanomedicine applications (Davies et al., 1995;Veverka et al., 2007;Joshi et al., 2009;Bregar et al., 2013;Salunkhe et al., 2013), and the replacement of maghemite by cobalt ferrite may be considered for biological applications if the absence of cobalt cation release, suspected to induce oxidative stress and cell toxicity, is demonstrated. Then, the substitution, in these ferrites, of a controlled amount of Co 2+ by diamagnetic Zn 2+ and Fe 3+ by paramagnetic high spin lanthanide Ln 3+ would significantly increase the total magnetization (Kolekar et al., 1994;Tahar et al., 2008;Pilati et al., 2018). The rationale is that, according to the collinear ferrimagnetic Néel model, a decrease of the magnetic moment of the tetrahedral spinel sub-lattice and an increase of the magnetic moment of the octahedral spinel sub-lattice both lead to an increase of the total magnetic moment of the spinel crystal. It may be noticed that, because of their ionic respective radii, lanthanide ions prefer the octahedral spinel sites (Kolekar et al., 1994;Yan et al., 1999;Tahar et al., 2008;Pilati et al., 2018), while zinc ions prefer the tetrahedral sites (Petitt and Forester, 1971;Ammar et al., 2006).
To date, wet chemistry has allowed the production of such mixed Co ferrite NPs, yet only substituted by either Zn or Gd, but not simultaneously. A rapid overview of the relevant literature shows that chemically made (Co 1 À x Zn x )Fe 2 O 4 nanocrystalline solid solutions exhibit a maximal magnetization at x ranging between 0.4 and 0.5 (Sharifi and Shokrollahi, 2012;Tahar et al., 2012;Ghasemian et al., 2015;Tatarchuk et al., 2017). In the case of Co(Fe 1 À x Ln x ) O 4 nanocrystalline solid solutions, the highest magnetization was obtained for Ln = Gd and x~0.1, the solubility limit of Ln cations being of the same order (Tahar et al., 2007;Kumar et al., 2011).
These substitutions may affect the Curie temperature T C of the spinel phase. T C is indeed closely related to the number of linkages between 3d transition paramagnetic ion cations through oxygen anions per formula unit (Gilleo, 1960). This number decreases when diamagnetic or weakly interacting 4f cations are introduced into the spinel lattice, reducing the T C value. This feature must be controlled because a T C decrease lowered down to the body temperature makes the produced particles paramagnetic and less valuable for negative MRI (Mertz et al., 2017;Shlapa et al., 2016).
In this context, we prepared a series of nanometersized Co 0.6 Zn 0.4 Fe 2 À x Gd x O 4 NPs, varying the Gd 3+ concentration up to its solubility limit and characterizing their structural and magnetic properties. We then evaluated their colloidal stability in water when they were coated by a hydrophilic ligand, typically positively charged dopamine (DA) species, and we measured their longitudinal and transversal relaxivities in water at an applied static field of 1.41 T that is close to the value of 1.5 T used in most clinical magnetic resonance imagers. Finally, we evaluated their cytotoxicity on model healthy human cells, with a special emphasis on their capability to be used as efficient magnetic probes for biomedical imaging.

Results and Discussion
Gd 3+ solubility limit The X-ray diffraction (XRD) patterns of all the prepared Gd-substituted particles were recorded. All were indexed within the cubic spinel structure without evidence of any trace of impurities ( Fig. SI-1). In addition, all the diffraction lines are broadened in agreement with the formation of ultrafine crystals. A close look at the most intense (311) diffraction peak revealed a slight shift toward lower diffraction angles, particularly for the lowest nominal Gd compositions (Fig. 1), indicating a progressive increase of the unit spinel cell constant. This increase is compatible with cationic radii change by substituting the Fe 3+ ions by larger Gd 3+ (0.940 vs. 0.645 Å; Shannon, 1976). The Rietveld refined unit cell parameters were found to be slightly higher than that reported for similarly prepared Co 0.6 Zn 0.4 Fe 2 O 4 particles. Noteworthy, the variation of the unit cell parameters as a function of the nominal Gd composition, x nom , was not trivial. In particular, no Vegard law was evidenced.
The chemical composition of all the prepared ferrites was accurately determined by X-ray fluorescence (XRF) spectroscopy analysis (selected spectra are given in Fig. SI-2). All the metallic Co, Zn, Fe, and Gd elements were detected, and their measured atomic contents allowed us to formulate each produced spinel phase (Table 1), evidencing an increasing elemental chemical composition departure from the nominal composition.
The most noticeable change relatively to the nominal composition is revealed for Gd, indicating the existence of a solubility limit of this element in the host spinel lattice. In order to gain further information about the solubility limit upon Fe 3+ substitution by Gd 3+ , we plotted the experimental atomic ratio of gadolinium to total gadolinium and iron elements, [Gd] exp / [Fe + Gd] exp , versus its associated nominal content, x nom ( Fig.SI-3). The experimental ratio initially increased monotonically with increasing nominal ratio and then reached a plateau at about 3% for a nominal value of Gd of x nom = 0.10 associated to the sample having the experimental chemical composition of Co 0.56 Zn 0.33 Fe 2.03 Gd 0.06 O 4 .
This solubility limit of 0.06 is in the same range than values reported in the literature on differently prepared lanthanide substituted spinel ferrites. Rezlescu et al. (1997) and Ahmed et al. (2005) evidenced a limit value of x s < 0.01 and x s < 0.05 in solid state made Ni 0.7 Zn 0.3 Fe 1 À x Ln x O 4 (Ln = Yb, Er, Dy, Tb, Gd, Sm, and Ce) and Ni 1 À y Zn y Fe 2 À x Ln x O 4 (Ln = La) particles, respectively. Kolekar et al. (1994), Tahar et al. (2008), and Kahn and Zhang (2001) determined a limit value of x s = 0.10 and x s = 0.06, respectively, in wet-chemically made CoFe 2 À x Ln x O 4 (Ln = Gd, Sm, Eu, Nd, Ho) and CoFe 2 À x Ln x O 4 (Ln = Ce, Ln = Ce, Sm, Eu, Gd, Dy, Er) particles, respectively. The low Gd solubility limit is mainly attributed to the high ionic radius of Gd 3+ . It is a reasonable approximation indeed that a (Co 1 À x Zn x )(Fe 2 À y Gd y )O 4 solid solution crystallizes in the inverse spinel structure, in which almost all the Co 2+ and Zn 2+ cations are respectively octahedrally and tetrahedrally coordinated. As a CoFe 2 O 4 -based NPs for MRI contrasting Walid Mnasri et al.
consequence, the inter-ionic cation-oxygen distances R B-O , defined as the average cation-oxygen bond length on octahedral spinel sites, can be expressed as (Salah, 2006;Beji et al., 2015): Using the values of ionic radii determined by Shannon (1976) and refined by O'Neil and Navrotsky (1984) ( . This size contraction in octahedral sites seems to be sufficient to notably reduce the solubility limit of Gd 3+ from x s = 0.10 to x s = 0.06 in Co 0.6 Zn 0.4 Fe 2 À y Gd y O 4 mixed ferrite NPs by comparison with CoFe 2 À y Gd y O 4 ones (Kolekar et al., 1994;Tahar et al., 2008). Even a small decrease in R B-O can lead to much more energy required for Gd 3+ to enlarge the B sites and enter in them.
To try establishing experimentally this feature, the infrared spectra of representative samples (mainly those corresponding to a nominal Gd composition smaller or equal to 0.1) were recorded (Fig. 2). Focusing on the spinel oxide skeleton vibrations, four infrared active modes, called ν 1 , ν 2 , ν 3 , and ν 4 , are expected at 650-550, 530-390, 380-340, and 300-200 cm À1 , respectively (Waldron, 1955;White and DeAngelis, 1967). The former bands (ν 1 and ν 2 ) are ascribed to the intrinsic vibrations of tetrahedral and octahedral metal-oxygen complexes, while the latter (ν 3 and ν 4 ) are related to more complex vibrations. Besides, their energy positions and intensities are strongly dependent on the nature and oxidation state of the metal cations. Therefore, valuable information can be obtained on the local structure of the studied phases by the analysis of their infrared spectra in these wavelength ranges (Waldron, 1955;Nasrazadani and Raman, 1993;Salah, 2006). Careful observation of the recorded spectra allowed us to distinguish two main regions: beyond~800 cm À1 and below 700 cm À1 . In the first region, the spectra show bands characteristic of chemisorbed species including water, polyol, and acetate ions (see Supporting Information) (Ammar et al., 2001). In the second region, they show the spinel lattice vibration bands (Table 2).
Focusing on the ν 2 band, its energy position varies when the composition of the investigated ferrites changes (see the dotted line in the right part of Fig. 2). This variation traduces changes in the metal-oxygen distances in the octahedral spinel sites and in the metal atomic masses. A decrease of the wavenumber means a decrease of the bonding length, and an increase of the metal atomic mass: ν 2 value is found to be 409 cm À1 in Co 0.60 Zn 0.40 Fe 2.00 O 4 particles and   . Surprisingly, by increasing the nominal Gd content, ν 2 increased up to 400 cm À1 , as ascribed to subsequent decrease of the average metal-oxygen distance, which can in turn be explained by the departure of a certain quantity of Co 2+ ions from B to A sites (with the transfer in the opposite direction of the same number of Fe 3+ ions of lower ionic radius) to induce a certain relaxation of the spinel lattice upon insertion of increasing quantities of Gd 3+ . In other words, whereas Co 0.60 Zn 0.40 Fe 2.00 0 4 particles exhibit a spinel structure close to the thermodynamically stable one, in which almost all the Co 2+ and Zn 2+ cations are octahedrally and tetrahedrally coordinated, respectively, Gd 3+ -substituted particles depart from this structure with a nonzero amount of tetrahedrally Co 2+ coordinated cations.
To complete these investigations, Raman spectroscopy was performed on the same samples and on bulk CoFe 2 O 4 , which serves as reference. All the recorded spectra were characteristic of the cubic spinel structure (Fig. 3). Usually five active modes, one of E g symmetry, two of T 2g , and two of A 1g are observed. The modes, which mainly involve the motion of oxygen in both tetrahedral and octahedral sites, appeared in the low frequency range~200-750 cm À1 (White and DeAngelis, 1967;Nasrazadani and Raman, 1993;Kreisel Wang et al., 2003). Among these modes, those which mostly correspond to the motion of oxygen in metal-oxygen tetrahedral complexes appear above 600 cm À1 , while those mainly reflecting the stretching vibrations of metal-oxygen octahedral complexes appear at higher frequencies, as summarized in Table 2. At first sight, the Raman spectra of our particles exhibit clearly broadened bands compared with bulk CoFe 2 O 4 . This feature is quite common in nanostructures (Yu et al., 2002;Tahar et al., 2007;Kumar et al., 2011). Moreover, the bands above 600 cm À1 , related to the two A 1g modes, are found to overlap, in accordance with previous works on nanocrystalline spinel ferrites (da Silva et al., 2010;Kumar et al., 2016). Finally, the band at around 200 cm À1 (not shown), usually related to one of the T 2g modes, is very weak in intensity and hard to detect at ambient temperature (Shebanova and Lazor, 2003;Kumar et al., 2016). Now, comparing the band energy position of our samples ( Table 2) with those of bulk CoFe 2 O 4 (a pure inverse spinel) on the one hand, and with those of each other on the other hand, relative shifts were observed with Gd content increase. These shifts are ascribed to the changes in the metal-oxygen bond lengths within both the tetrahedral and octahedral spinel sub-lattices with the introduction of large Gd 3+ cations. In other words, Gd 3+ incorporation affects the average length of the metal-oxygen bonds in both octahedral and tetrahedral complexes. As suggested by infrared spectroscopy, Raman spectroscopy confirmed the cation migration between the spinel sub-lattices. Although they are weak, these structural changes explain why we did not observe a Vegard law for the variation of the unit cell parameter of the studied solid solution as a function of its Gd composition. They also suggest nontrivial variations of their magnetic properties with their composition, particularly their magnetocrystalline anisotropy and their total magnetization. Whereas a simple increase of the magnetization was expected by replacing Fe 3+ cations by Gd 3+ , namely, by replacing cations leaving a magnetic moment of 5 μ B by ones leaving a magnetic moment of 7 μ B , the displacement of Co 2+ cations from the octahedral to the tetrahedral sites and reversely, the displacement of an equal amount of Fe 3+ cations from the tetrahedral to the octahedral sites should decrease the total magnetocrystalline anisotropy constant (lower spinorbit coupling in the B sub-lattice) and increase the total crystal magnetization (smaller magnetic moment in the A sub-lattice).

Microstructural properties
From the acquired XRD data, the average crystal size, <L XRD >, and the micro-deformation, <ε>, were determined for all the produced nanoferrites. They were found to be consistent with 6 nm sized almost isotropic in shape strain free crystals. The incorporation of Gd 3+ within its solubility limit did not induce apparent distortions in the spinel lattice: <ε> ranged between 1.5% and 2.0%. These results were confronted to those obtained by transmission electron microscopy. Representative micrographs showed roughly spherical particles, slightly agglomerated, with an average size ranging between 5 and 6 nm (Fig. 4), suggesting that the produced NPs are consistent with single crystals, interacting mutually magnetically. They appeared as almost uniform in size, and the statistical analysis of hundreds of particles considered as spheres agreed with a standard size deviation of less than 15%. Such low size dispersity can be underlined because it suggests a rather uniform magnetic response of these NPs under magnetic stimuli.
High-resolution transmission electron microscopy observations confirmed that all the produced NPs were single crystals. All images show fringes corresponding to crystallographic planes of the cubic spinel lattice. There was no evidence of structural defects such as dislocations, stacking faults, nor surface amorphous region, establishing their high crystalline quality despite their reduced size and their local structural changes induced by Gd incorporation. A zoomed picture of one representative particle among the studied nanocrystalline Co 0.6 Zn 0.4 Fe 2 À x Gd x O 4 (0 ≤ x ≤ 0.06) solid solutions is shown in Figure 5. From all these observations, one may expect relevant magnetic properties for the desired application, because a high crystalline quality usually means less broken magnetic pathways and then higher specific magnetization.

Magnetic properties
Direct current (DC) magnetic measurements were performed on three selected samples: Co 0.60 Zn 0.40 Fe 2.00 0 4 , Co 0.60 Zn 0.40 Fe 1.98 Gd 0.02 O 4 , and Co 0.60 Zn 0.40 Fe 1.94 Gd 0.06 O 4 , corresponding to nominal Gd compositions of x nom = 0.0, 0.04, and 0.10, respectively, and assumed to be representative of the general magnetic features of the produced nanocrystalline solid solutions. In practice, the zero field cooling (ZFC) and field cooling (FC) thermal variation of the DC magnetic susceptibility, χ(T), were recorded on the samples on their powder form and depicted in Figure 6. A net irreversibility was thus evidenced between the ZFC-and FC-χ(T) branches in agreement with a superparamagnetic behavior. ZFC-χ(T) exhibits a narrow peak at a critical temperature defined as the average blocking temperature T B , while the FC-χ(T) curve reaches a plateau below T B . T B represents the threshold temperature above which the magnetic anisotropy barrier is overcome by thermal activation energy alone, causing the transition of the magnetic NPs from their ferrimagnetic state to the relaxed superparamagnetic one and reversely. The T B value is directly related to the volume of the magnetic particles. It increases when the size increases, and it becomes widely distributed when the distribution of the size is broadened. For a given average size, T B also increases with the effective magnetic anisotropy constant, K eff . Finally, T B is also affected by the strength of dipolar interactions. It is shifted toward higher temperatures when these interactions between magnetic moments are stronger. Therefore, comparing now the T B values measured on our samples, supposed to consist in strongly and similarly interacting almost uniformly and similarly sized magnetic single nanocrystals, we found that they increased slightly from about 150 K in Gd-free NPs to about 165 K in the Gd-richest NPs (Table 3).
Meanwhile, the magnetic susceptibility values at the ZFC-χ(T) maximum increased too. This T B shift toward higher temperature can be theoretically explained by three raisons: (1) an average particle size increase from one sample to another, (2) an increase of anisotropy constant K eff when x increases in the Co 0.60 Zn 0.40 Fe 2 À x Gd x O 4 nanocrystalline solid solution, and/or (3) an increase of dipolar interaction strength when x increases due to an increase of the total magnetic moment with x. Because the size of all the produced particles is supposed to be constant, and the magnetocrystalline anisotropy is expected to be quite unchanged by introducing Gd in the spinel lattice, only dipolar interactions would be incriminated. Dipolar interactions refer to the direct interaction between two magnetic domains bearing a given magnetic moment and separated by a given distance. The dipole interaction energy is proportional to the square of the magnetic moments and inversely proportional to the cube of the distance (Mørup et al., 2010). This distance is assumed to be constant in all the samples, while the average magnetic moment is expected to increase from one sample to another by increasing the Gd content.
Pursuing our magnetic investigations, we decided to perform further magnetic measurements on weakly interacting particles, to avoid any dipolar interaction interferences. In practice, the produced particles were mixed, by soft milling, with a diamagnetic alumina matrix (4.2 wt.%), and their first magnetization curves (Fig. 7) as well as their hysteresis loops (Fig. 8) were recorded at different temperatures. The collected magnetic data were then corrected from the diamagnetic alumina contribution and expressed per gram of magnetic Co 0.6 Zn 0.4 Fe 2 À x Gd x O 4 NPs.
As expected, M(H) curves did not evidence any hysteresis feature when cycling the applied magnetic field between 50 and À50 kOe at 300 K (data not shown). On the contrary below T B , typically at 5 K, they did show a hysteresis, with a coercivity of around 4 kOe and a remanence of more than 50% (Fig. 8).
Using these data, the saturation magnetization, M sat , was determined by extrapolating M versus 1/H curve to 1/H = 0. The inferred values were found to be relatively  high, close to 80 and 130 emu g À1 at 300 and 5 K, respectively, for all the Gd-substituted samples, increasing slightly by increasing the Gd content (Table 3). The temperature dependence of the spontaneous magnetization, M sp , measured at high field, typically 50 kOe, was also plotted to be tentatively used to estimate the Curie temperature of the produced particles (Fig. 9). Within the Bloch (1930) model, the thermal variation of M sp , in a temperature range far below T C , can be used to estimate T C (Hendriksen et al., 1993;Chen et al., 1996;Ngo et al., 2001;Demortière et al., 2011;Basti et al., 2014) according the following equation: where β is the so-called Bloch's exponent and M sp (0) is the spontaneous magnetization at 0 K. β is usually set at 1.5 for bulk ferromagnets and ferrimagnets (Bloch, 1930). It is also equal (Ammar et al., 2006;Basti et al., 2014;Hanini et al., 2016) or slightly smaller (Hendriksen et al., 1993;Chen et al., 1996;   CoFe 2 O 4 -based NPs for MRI contrasting Walid Mnasri et al. Zhang et al., 1998;Ngo et al., 2001;Demortière et al., 2011) than 1.5 for nanocrystals, depending on their size and their crystallinity. In the present case, based on the high-resolution transmission electron microscopy results, β was set at 1.5, and the experimental data were fitted to estimate T C . Values ranging between 543 and 566 K were obtained (Table 3), without evidencing a clear dependence on the Gd content. These values are far from the body temperature, making the engineered nanoferrites valuable negative contrast agents for MRI.

Colloidal stability and magnetic resonance imaging relaxivity
To evaluate the MRI contrasting ability of the produced NPs, it is important to disperse them in an aqueous medium as stable colloids. A catechol ligand bearing an amino group, also known as dopamine (DA), was grafted onto the surface of the particles to make them repulsive and slightly positively charged in physiological pH conditions (Figure 10). A specific protocol was developed, adapted from an already established one (Basti et al., 2010;Fouineau et al., 2013) to attach the required amount of DA at the surface of the NPs, favoring weak yet nonzero clustering. The idea was to form small controlled aggregates of less than 100 nm in hydrodynamic diameter to improve the total relaxivity of the particles without furtherly limiting their circulation time inside the blood vessels, when administrated intravenously.
At first, the grafting yield was checked by both X-ray photoelectron spectroscopy (XPS) and thermogravimetric analysis (TGA). In particular, the replacement of the residual polyol (and/or acetate) molecules adsorbed at the surface of the as-produced particles by DA ligands was confirmed, mainly by the appearance of nitrogen XPS signature only in the functionalized NPs ( Fig. SI-4). Moreover, the weight content of DA was quantified by TGA as 45 wt.% in the studied hybrids ( Fig. SI-5).
Secondly, the zeta potential was measured on the aqueous suspensions of bare and DA grafted NPs, by varying the pH from 2 to 12 ( Fig. SI-6). Positive charges were evidenced up to isoelectric points at pH = 7 and 9 (close to the tabulated pKa value of 8.93 for DA) on bare and functionalized NPs, respectively. These surface charges render the functionalized particles hydrophilic at physiological pH and should avoid their flocculation in water, when also combined with the steric repulsions due to the organic shell.
Thirdly, dynamic light scattering (DLS) measurements were carried out on the two suspensions at neutral pH. While a hydrodynamic diameter of several hundreds of nanometer was measured for bare particles, indicating their large aggregation, a value close to 50 nm (monomodal size distribution) was measured for the functionalized particles, suggesting their limited clustering (Fig. 11). In particular, the effect of a pulsed ultrasound treatment (repeated cycles 2″ on at full power, 1″ off, during 10 0 ) was tested to tentatively decrease the hydrodynamic size by breaking the clusters, yet it did not succeed. This moderate clustering, because of the balance between electro-steric repulsions between the weakly positively charged DA-grafted particles and their mutual magnetic and van der Waals attractions, was also confirmed by transmission electron microscopy observations (Fig. 11) of drops from the resulting suspensions deposited on the carbon grids.
As for the proton relaxation properties, the prepared aqueous colloidal suspensions diluted to total metal concentrations ranging from~0.05 to~0.8 mM depending on the samples were supposed to verify the Redfield condition defining the MAR model (Vuong et al., 2012). In other words, freely diffusing water molecules, surrounding the previously described magnetic clusters, explore a space larger than the NP volume during one echo time of the applied radiofrequency sequence. The theoretical estimation of the transversal relaxivity of the studied systems becomes possible even if their composition is far from that of maghemite or magnetite. Replacing maghemite or magnetite spinel phase by a Zn-substituted and Gd-substituted cobalt ferrite should not affect significantly the value of the molar volume v mat in equation 1a. A variation of only ±5% is expected on the 1.5·10 À5 m 3 mol À1 maghemite v mat value (Vuong et al., 2012). Using equation 1 with the experimental values of M v and approximating d by the experimental <D DLS >, an r 2 value of less than 10 mM À1 sec À1 was estimated theoretically on the studied sample series, far below our expectations.
However, when we directly performed relaxivity measurements to assess the exact efficiency of our prepared batches to relax nuclear spins of water protons, we found much higher relaxivity values. In practice, their transverse (r 2 ) and longitudinal (r 1 ) relaxivities were measured at physiological temperature, 37°C, with a 60-MHz relaxometer (Bruker Minispec ™ mq60) based on a 1.41-T magnet (i.e., close to the 1.5-T magnetic field of most clinical MRI machines used in hospitals). More precisely, the longitudinal (T 1 ) and transverse (T 2 ) relaxation times of water protons were measured with respectively an inversion-recovery and a Carr-Purcell-Meiboom-Gill (CPMG) sequence, at decreasing equivalent [Fe] concentrations from 0.8 to 0.1 mM. The receiver gain was adjusted between 42 and 48 dB in order to obtain a proper signal-to-noise ratio without saturation. For the inversion-recovery sequence, initial delay time was adjusted around T 1 /10, while the final delay was at least 3 T 1 , and the recycling delay at least 5 T 1 , according to recommendations for such measurements (Henoumont et al., 2009). Recycling delay was kept constant for the CPMG sequence, while choosing an echo time (TE) around T 2 /50 and a number of data points of 150, the decay curve of transverse magnetization being fitted by a monoexponential law. Relaxivities were obtained, in deionized water, from the slope of the linear variation with the total metal concentration [M] total of the longitudinal (respectively transverse) decay rate of water proton spins after subtraction of respective value measured on pure water, according to 1=T i¼1 or 2 ¼ r i¼1 or 2 · M ½ total þ 1=T i¼1 or 2 ð Þ water : (4) The obtained curves were plotted on Figures 11 and SI-7, and the measured relaxivity values were summarized in Table 4 and compared with those of some references (Table 5), namely, commercial iron oxide-based negative contrast agents (Ferridex, etc.) and to those of various highly magnetized nanometer-sized spinel ferrite particles, on the other hand. Clearly, the r 2 and r 2 /r 1 values measured on our optimized system, DA coated and clustered Co 0.60 Zn 0.40 Fe 1.94 Gd 0.06 O 4 particles are very high, higher than those reported on previous systems, in almost the same operating MRI conditions. Typically, for the Gd-richest sample, an r 2 /r 1 of 33 and an r 2 of 188 mM À1 sec À1 were obtained.
The discrepancy between the measured r 2 values and those calculated within the Vuong model suggests us that the very high magnetization of our particles makes them working in an intermediate regime, corresponding to the transition between the MAR and the SDR relaxation regimes.
Finally, the potential cytotoxicity of Co 0.6 Zn 0.4 Gd x Fe 2 À x O 4 (0.0 ≤ x ≤ 0.06) NPs was investigated on human fibroblastic cells, which are common in cytotoxicity assays of magnetic NPs (Spirou et al., 2018). Typically, the cells were incubated for 24 h Figure 11. Relaxation rates R 2 NP À R 2 water = 1/T 2 À 1/T 2 water (red data points) and R 1 NP À R 1 water = 1/T 1 À 1/T 1 water (blue data points) variation as a function of the total metal cation concentration measured on dopamine (DA)-coated Co 0.6 Zn 0.4 Fe 2 À x Gd x O 4 (x = 0.0 and 0.06) nanoparticles dispersed in deionized water (T 1 water = 3997 msec and T 2 water = 1020 msec) at 37°C for an applied static magnetic field of 1.41 T (60-MHz proton resonance frequency). The corresponding r 2 and r 1 relaxivities of the nanoparticles are deduced from the slopes of the curves.
with different doses of DA-coated NPs. The doses (viz., 2.5, 5, and 10 mg L À1 ) were chosen significantly higher than the concentrations typically used for contrast agent assisted MRI experiments. The cells were then submitted to the Alamar blue (AB) assay that tests mitochondrial activity to evaluate their cell viability. Interestingly, the cell viability was not significantly hindered at all studied concentrations. The collected data revealed for each concentration of Co 0.6 Zn 0.4 Gd x Fe 2 À x O 4 NPs incubated with the cells, a toxicity linear with composition (0.0 ≤ x ≤ 0.06) although quite limited as viability remained always above 80% relatively to control (Fig. 12). Additionally, confocal microscopy images were acquired on the cells incubated with the highest dose (10 mg L À1 ) for the longest contact time (72 h). Representative immunofluorescence images (Fig. 13) recorded on cells incubated with the solid solutions and compared with those of cells not at all incubated with NPs exhibited typical fibroblast BJH morphology, The chemical composition of the polyol-made and water washed maghemite nanoparticles (dark red color) changed during dopamine (DA) grafting to become closer to that of magnetite (black color). Indeed, thanks to the Pi donor properties of DA, a non-negligible charge density is transferred from DA to the ferric surface cations, during the complexation reaction (Fouineau et al., 2013). with no change on the nuclei shape. These results are very interesting because they confirm the biosafety of the investigated NPs.

Conclusions
The influence of Gd 3+ substitution to Fe 3+ on the spinel structure, microstructure and magnetic properties of polyol-made nanosized Co 0.6 Zn 0.4 Fe 2 O 4 particles was thoroughly investigated. A solubility limit of x s = 0.06 was determined for the Co 0.6 Zn 0.4 Fe 2 À x Gd x O 4 solid solution with, very probably, local structural change.
Far-infrared (FIR) and Raman spectroscopies suggested partial transfer of Co 2+ cations from octahedral to tetrahedral spinel sites and reversely transfer of the same number of Fe 3+ cations from tetrahedral to octahedral sites to relax the spinel lattice strain after large Gd 3+ cation introduction. The produced NPs appeared as almost spherical in shape, highly crystalline, magnetic monodomains, of 5-6 nm in diameter. All the NPs exhibited superparamagnetic behavior at room temperature, with a very high saturation magnetization, close to 80 emu g À1 at 300 K. These high values The results are expressed as a ratio to non-stimulated serum-free cultured cells for each dose data point. allow the classification of the engineered NPs in the highly magnetized ferrite-based negative MRI contrast agents. Relaxometric measurements confirmed this feature, because r 2 and r 2 /r 1 values of 188 and 33 mM À1 sec À1 respectively where obtained on the DA decorated Gd-richest, limited NP clusters, dispersed in deionized water, at 37°C under a static magnetic field of 1.41 T. Besides, preliminary cytotoxicity studies carried out on healthy BHJ human fibroblast model cells incubated with doses as high as 10 mg L À1 of DA-coated NPs for 72 h showed a negligible cell death, likely ascribed to the Gd content, making our engineered contrast agents definitively valuable for MRI.

Experimental Section
Particle synthesis The preparation of Co 0.6 Zn 0.4 Gd x Fe 2 À x O 4 NPs, with a nominal x composition ranging between 0.0 and 0.2 by a step of 0.02, was achieved by the so-called polyol method. Analytical grade chemicals were used including iron(III) chloride hexahydrate and gadolinium(III), cobalt(II), zinc(II), and sodium(I) acetate salts. They were accurately weighed in appropriate ratios and then dissolved in 125 mL of diethyleneglycol to reach a nominal concentration of 0.2 M of total trivalent cations (Fe 3+ and Gd 3+ ). The resulting mixture was then heated up to boiling (~220°C) and maintained under reflux for 4 h under mechanical stirring. After slow cooling, the obtained black magnetic precipitates were separated from the supernatant by centrifugation, washed with distilled water, then with ethanol, and finally dried in air at 50°C. For all the syntheses, the hydrolysis ratio, the acetate ratio, and the zinc ratio (defined as the nominal water per metal cation, acetate per metal cation, and zinc cations per trivalent cations molar ratios) were fixed to 1, 3.66, and 0.5, respectively, after a series of optimizations.

Colloidal suspension preparation
Dopamine ligand was chosen to render the produced NPs hydrophilic and water dispersible. The related hybrids were prepared according to already described operating conditions (Basti et al., 2010). In practice, 50 mg of the pre-functionalized particles were added to 3 mL of deionized water. Then, to ensure the maximum coating with the catechol derivative through an exchange chemistry of the polyol entities by DA molecules, an excess of DA was slowly added to the aforementioned mixture with assistance of ultrasonication and vortexing. Excess of acetone was then added, and a magnetic sedimentation allowed collecting the desired nanohybrids, which then became readily dispersible in water, forming very stable aqueous colloidal suspensions.

Particle characterization
The chemical compositions of the produced NPs were checked by XRF focusing on the Fe, Co, Zn, and Gd elements. The chemical compositions of the produced NPs were checked by XRF using a Epsilon 3XL (Panalytical) X-ray fluorescence spectrometer, equipped with a silver X-ray tube operating at 20 kV 200 μA. The analyses were performed on a given volume of well-concentrated aqueous solution obtained by a total acid attack of a precise mass of the as-produced particles. Fe, Co, Zn, and Gd standard and certified solutions were used for calibration. The structure of all the produced NPs was investigated by XRD using a Panalytical X'pert Pro ™ instrument, equipped with a CoK α tube (λ = 1.7889 Å). The unit cell parameter a, the average crystallite size <L XRD >, and the average lattice micro-deformation <ε> were inferred from Rietveld refinements, using MAUD software. Polycrystalline strain-free silicon was used as standard to estimate the instrumental XRD line broadening. To complete these structural analyses, FIR and Raman spectroscopies were performed at ambient temperature on the as-produced NPs. FIR spectra were recorded on polyethylene pellets using a Bruker Equinox Fourier transform infrared spectrometer in a transmission scheme. Raman spectra were obtained in a backscattering geometry using a LabRam HR-800 spectrometer, equipped with a 25-mW He-Ne laser source operating at 633 nm. Finally, the microstructure was determined thanks to a JEOL-100-CX II transmission electron microscope operating at 100 kV. The magnetic properties of the as-produced NPs were determined using a Quantum Design ™ MPMS-5S SQUID magnetometer. To do so, the thermal variation of the magnetic susceptibility of each sample was measured between 5 and 330 K within both FC and ZFC conditions for an applied dc-magnetic field of 200 Oe. Then, the ZFC isothermal variation of their magnetization as a function of the magnetic field was measured between +50 and À50 kOe at 5 and 300 K, dispersing first the NPs in a diamagnetic alumina (Kaiser, 99.9%) matrix (4.8 wt.%) to limit the effect of dipolar interactions on their whole magnetic behavior.
At the end, the temperature dependence of the high field (50 kOe) spontaneous magnetization was measured at different temperatures. All the data were expressed per gram of ferrite powders and corrected from the diamagnetic contribution of alumina and the sampling plastic tube.

Colloidal state characterization
The aqueous colloidal suspensions made from DAcoated NPs were also characterized. At first, the exact Fe, Co, Zn, and Gd weight contents were determined by XRF analysis using the same step-up than that described previously but working directly on the colloidal dispersions. DA grafting was checked by XPS analysis using a Thermo VG ESCALAB ™ 250 instrument equipped with a micro-focused, monochromatic A1 Kα X-ray source (1.486.6 eV) and a magnetic lens. The X-ray spot size was 500 μm (15 kV, 150 W). The spectra were acquired in the constant analyzer energy mode with pass energies of 150 and 40 eV for the general survey and the narrow scans, respectively. The final DA weight content on all the prepared hybrids was determined on their dried powders by TGA, using a TA instrument ™ Q50 or a Labsys-Evo ™ commercial machine and heating each sample from 25°C to 800°C in air (10°C min À1 ). The hydrodynamic diameter and the zeta potential of bare NPs, and their related DA grafted hybrids, were determined at room temperature on their aqueous suspensions (0.50 g L À1 ) using a Malvern Nano Zetasizer operating at 90°scattering angle, after vigorous sonication with an ultrasound probe for 10 min (repeated cycles 2″ on at full power, 1″ off).
Finally, relaxometric measurements were carried out on the prepared aqueous colloids by fixing the mass of the hybrid powders around a concentration of iron close to 0.8 mM Fe . The stock suspensions were diluted several times (up to height times), and 0.7 mL of each suspension was introduced in nuclear magnetic resonance tubes (7.5-mm outer diameter). The tubes were then inserted into a Bruker Minispec ™ mq60 relaxometer equipped with a 60-MHz/1.41-T magnet. The T 1 and T 2 relaxation times were recorded at 37°C, and their inverse (decay rates) were plotted as a function of the total metal ion concentrations, after subtraction of the respective decay rate in pure water. The concentrations were properly calculated by several methods and averaged together: through accurate titration of all the metallic ion content in all the samples inferred from XRF measurements; by a colorimetry assay after dissolution in HCl 5 M using the characteristic peak at 350 nm of the hexachloroferric complex (Rad et al., 2007); and finally, by TGA without any acid mineralization (leading to the total metal oxide weight). T 2 values were measured with a CPMG sequence using interecho times (TE) values ranging from 0.18 to 6 msec (depending on the concentration in metal centers) (Hemery et al., 2017).

Toxicity assays
In vitro cytotoxicity of the produced nanohybrids was checked by testing their effect on human fibroblastic cells cultured for 24 h. Forty-eight-well cell culture clusters (Corning, USA) were plated with 1 mL of 15,000 cells mL À1 cell suspension. After 24-h incubation, plates were washed with 1 mL per well phosphate buffer saline, and the cells were treated with various concentrations of particles prepared in Dulbecco's modified Eagle's medium: 2.5, 5, and 10 mg L À1 (corresponding to the exact inorganic magnetic core mass). The incubation was performed at 37°C in a 5% CO 2 humidified incubator for different contact times. Replicate wells were used for each control and to test the concentration in each plate. At first, AB viability assays (Al-Nasiry et al., 2007) were conducted. In practice, the cells were rinsed once with phosphate buffer saline, and 1 mL of AB medium (10% v/v solution of AB in Dulbecco's modified Eagle's medium) prepared in fresh medium was added to each well. After 3-h incubation, the AB absorbance of the samples was measured at 570 nm (A570) and 600 nm (A600) on a microplate reader. The absorbance of AB media in the absence and presence of particles was compared: no significant difference was observed, suggesting that the nanohybrids do not interact with AB or affect the assay readout. All toxicity experiments were conducted in at least triplicate (three independent experiments). Absorbance (AB assays) in optical density units was quantified using a microplate reader. Raw data from cytotoxicity assays were collated and analyzed using Microsoft Excel® (Microsoft Corporation, Redmond, WA). Cytotoxicity was expressed as the mean percentage inhibition relative to the unexposed control ± standard deviation. Statistical analyses from cytotoxicity assays were carried out using one-way analyses of variance followed by Dunnett's multiple comparison tests. Statistical significance was accepted according to Student's t-test at P ≤ 0.05.
To complete these analyses, confocal microscopy was performed on the incubated cells to visualize their morphology and possibly qualitatively evaluate the NP toxicological effect. Typically, human fibroblastic cells were seeded on glass coverslips 24 h before treatment with the highest particle dose (10 mg L À1 ) for 72 h. Fixed samples (4% paraformaldehyde) were mounted with mounting medium (Vector Laboratories, Burlingame, CA). The nuclei were counterstained with 4'6-diamidino-2-phénylindole (DAPI) (λ ex = 405 nm) and the cytoskeleton with Tetramethyl-Rhodamine Iso-Thio-Cyanate (TRITC) phalloidin (λ ex = 630 nm).

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article.
Data S1. Figure SI-1. Indexed XRD patterns of the asproduced particles for different nominal x compositions (from x nom = 0 at the bottom to x nom = 0.20 on the top by a Δx nom step of 0.02).  (Ahmed et al., 2005) and refined by O'Neil et al. in spinel oxides (White and DeAngelis, 1967). Figure SI-3. The experimental ratio of gadolinium composition to the total gadolinium and iron elements, exp-Gd/(Fe + Gd), plotted as a function of the nominal atomic gadolinium content, x nom . Table SI-2. Identification of the main bands, observed above 800 cm À1 , on the MIR spectra of the as-produced Co 0.60 Zn 0.40 Fe 2-x Gd x O 4 nanocrystals. Figure SI-4. Survey and high resolution C 1S, Gd 3d 5/2 and N 1 s XPS spectra recorded on of the Gd richest as-produced nanoparticles (grey line) and their DA coated ones (blue line).   -R 2 water (red data points) measured on iron oxide (supposed to be close to magnetite) and cobalt ferrite nanoparticles coated by dopamine at 37°C under a magnetic field of 1.41 T. The relaxation rates of pure water have been subtracted (T 1 water = 3997 ms and T 2 water = 1020 ms) in order to obtain the relaxivity r 2 and r 2 of the NPs from the slopes of the curves.
CoFe 2 O 4 -based NPs for MRI contrasting