Covalent Functionalization of Multi‐walled Carbon Nanotubes with a Gadolinium Chelate for Efficient T1‐Weighted Magnetic Resonance Imaging

Given the promise of carbon nanotubes (CNTs) for photothermal therapy, drug delivery, tissue engineering, and gene therapy, there is a need for non‐invasive imaging methods to monitor CNT distribution and fate in the body. In this study, non‐ionizing whole‐body high field magnetic resonance imaging (MRI) is used to follow the distribution of water‐dispersible non‐toxic functionalized CNTs administrated intravenously to mice. Oxidized CNTs are endowed with positive MRI contrast properties by covalent functionalization with the chelating ligand diethylenetriaminepentaacetic dianhydride (DTPA), followed by chelation to Gd3+. The structural and magnetic properties, MR relaxivities, cellular uptake, and application for MRI cell imaging of Gd‐CNTs in comparison to the precursor oxidized CNTs are evaluated. Despite the intrinsic T2 contrast of oxidized CNTs internalized in macrophages, the anchoring of paramagnetic gadolinium onto the nanotube sidewall allows efficient T1 contrast and MR signal enhancement, which is preserved after CNT internalization by cells. Hence, due to their high dispersibility, Gd‐CNTs have the potential to produce positive contrast in vivo following injection into the bloodstream. The uptake of Gd‐CNTs in the liver and spleen is assessed using MRI, while rapid renal clearance of extracellular Gd‐CNTs is observed, confirming the evidences of other studies using different imaging modalities.


Introduction
Carbon nanotubes (CNTs) have great promise for numerous technological as well as biomedical applications, which arise from their unique architecture and their outstanding mechanical and electronic properties. [ 1 ] However, their toxicity and fate in the organism is still a subject of debate. The lack of consensus on a range of toxicology and biodistribution issues regarding CNTs is due, on one part, to the wide heterogeneity of CNTs used in terms of length, diameter, number of walls, purity (e.g., the presence of catalysts), morphology, functionalization, and aggregation state. [ 2 ] An additional confounding factor is the limited range of methods available to study CNTs, and more generally carbon-based materials, in biological environments. [ 3 ] As pristine CNTs are water insoluble, a variety of chemical strategies have been explored to functionalize the nanotube surface and render them dispersible in aqueous Covalent Functionalization of Multi-walled Carbon Nanotubes with a Gadolinium Chelate for Effi cient T 1 -Weighted Magnetic Resonance Imaging excellent spatial resolution and soft-tissue contrast providing simultaneously anatomic, functional and molecular information. The main advantage of MRI over other modalities, such as fl uorescence or nuclear imaging, is its ability to penetrate deep into tissue without employing ionizing radiation. In addition to the enormous, and ongoing, growth in its application in routine clinical practice, MRI has also proven to be a method of choice for molecular imaging, providing cell tracking and direct information on molecular events through the use of smart contrast agents. To date there have been few reports on the use of MRI for visualization of CNTs due to the poor MR contrast generated by CNTs. Recently the magnetic properties of metal impurities present in raw commercially available single-walled CNTs (mainly iron and nickel catalysts) were exploited to investigate CNT distribution with hyperpolarized 3 He and conventional 1 H MRI. [ 18 ] However, the intrinsic T 2 and T 2 * contrast that are exploited in this approach are highly dependent on the quality of purifi cation, on the type of CNTs, and indeed on the aggregation propensity in physiological media; [ 19 ] all factors which are not easily controllable. Alternatively, CNTs may be coupled to exogenous contrast agents. These agents may allow both, the T 2 contrast and signal reduction, as for example is the case of CNT conjugates with iron oxide superparamagnetic nanoparticles [ 16b , 20 ] or provide T 1 -based contrast and signal enhancement when the tubes are conjugated to paramagnetic gadolinium. [ 21 ] The CNT-contrast agent complexes might be either made by loading the tube cavity or by functionalization of their carbon backbone. For the goal of monitoring the behavior of CNTs in vivo, it is important that any exogenous agent does not infl uence either the CNT biodistribution, or their ability to transport and deliver a therapeutic agent to an identifi ed target. In this regards both external coupling to or internal loading of CNTs with iron oxide nanoparticles may partially modify their fate in vivo, as the organism must degrade inorganic nanoparticles in addition to carbon materials, which are both slow processes. [ 22 ] Alternative strategies such as nanoscale confi nement of large quantities of Gd within ultra-short SWCNTs through sidewall defects or at the opened ends (so called gadonanotubes), [ 23 ] synthesis of SWCNT-Gd oxide complexes by use of Gd 2 O 3 catalytic nanoparticles [ 24 ] or non-covalent adsorption of Gd 3+ chelates onto MWCNT surface [ 25 ] have been shown to produce agents with very high r 1 relaxivity (relaxation enhancement per millimolar concentration of agent). However, noncovalent approaches have the disadvantages of saturating the loading capacity of CNTs or strongly reducing the possibility of further functionalization. In addition, most of these Gd-CNT complexes have unknown behavior in vivo, especially after intravenous injection due their poor dispersibility. Only one recent study described intravenous administration of a low dose of SWCNTs, synthesized using gadolinium nanoparticles as catalysts, however an in vivo MRI follow-up was not reported and the fate of gadolinium nanoparticles is not known. [ 26 ] In this context, our goal was to produce water-dispersible nontoxic functionalized short MWCNTs suitable for targeting and delivery of drugs [ 7a ] with fully characterized MRI contrast properties. The critical point was to endow carbon nanotubes with positive contrast properties without interfering with the nanotube structure and fate. We proposed a covalent grafting of molecular paramagnetic species on the nanotubes.
For this purpose, oxidized MWCNTs were covalently functionalized with the chelating molecule diethylenetriaminepentaacetic dianhydride (DTPA), followed by chelation to Gd 3+ . As our approach is based on covalent bonding between CNTs and DTPA, which has a high association constant for Gd 3+ , the chelation of Gd 3+ by DTPA/CNTs is strong, resulting in stable Gd-CNT conjugates and thus reducing the risk of Gd 3+ release.
The magnetic properties and magnetic resonance relaxivities of the Gd-CNTs in comparison to oxidized CNTs showed the paramagnetic contribution of Gd-chelates which led to positive MR contrast. Remarkably these nanotubes were internalized by cells without toxicity and the positive contrast effect was preserved following cell internalization. Finally, the potential for in vivo monitoring of nanotube fate following intravenous administration of Gd-CNTs in mice was assessed demonstrating that MRI can be used as suitable alternative imaging technique to track CNTs in vivo.

CNT Functionalization
MWCNTs with a diameter of 20-30 nm functionalized with a DTPA derivative were prepared using a four step procedure ( Figure 1 ). Pristine purifi ed MWCNTs were fi rst oxidized in acidic conditions to shorten the tubes to an average length of ≈400 nm and to introduce COOH groups (Supporting Information Figure S1). [ 27 ] The carboxylic acids were activated using oxalyl chloride. The corresponding acyl chloride groups were then coupled to the DTPA derivative 2 bearing a free amine function (Supporting Information Figure S2). [ 25,28 ] The tert -butyl ester moieties were subsequently hydrolyzed wileyonlinelibrary.com © 2014 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim using trifl uoroacetic acid (TFA). Finally, chelation of Gd 3+ was achieved by dispersing the DTPA/CNTs 4 in an aqueous solution of gadolinium(III) perchlorate, followed by dialysis against water to remove free gadolinium. The usual approach to conjugation of molecules with DTPA involves the conversion of one of the four carboxylic acids to an amide functionality by the use of DTPA dianhydride. In addition to the disadvantage of being non-selective, due to the concomitant formation of diamide derivatives, this method results in conjugates with lower chelating capacity for gadolinium. [ 29 ] In our case the DTPA derivative 2 has the advantage of bearing fi ve free COOH groups available for chelation of gadolinium.
The Gd-CNTs were characterized by thermogravimetric analysis (TGA), inductively coupled plasma atomic emission spectrometry (ICP-AES), high resolution transmission electron microscopy (HR-TEM) and scanning transmission electron microscopy (STEM) coupled to energy-dispersive X-ray spectroscopy (EDX). TGA in N 2 of pristine MWCNTs, oxidized MWCNTs (ox-CNTs) 1 and DTPA/CNTs 4 showed an increasing weight loss, confi rming the effi ciency of the functionalization ( Figure 2 A). The weight loss difference at 500 °C between ox-CNTs 1 and DTPA/CNTs 4 provides an assessment of the level of functionalization; the DTPA loading was estimated 540 µmol per gram of CNTs. The level of functionalization was also determined spectrophotometrically using xylenol orange. [ 30 ] For this purpose, after chelation of DTPA/CNTs 4 with gadolinium(III) perchlorate and before fi ltration and dialysis, the free Gd (i.e., not chelated to DTPA/CNTs) was titrated. The dye xylenol orange is used as complexometric indicator of gadolinium as it shows a color change in the presence of free Gd. This colorimetric test allowed us to determine that the amount of bound gadolinium was 485 µmol per gram of CNTs 5 , which is consistent with the value obtained from TGA. The xylenol orange test was also performed to ensure that no free Gd 3+ was present in the Gd-CNT samples. Control reactions were performed using the precursor CNTs (i.e., ox-CNTs 1 ). In this case, we did not observe a signifi cant complexation of Gd 3+ ions by the carboxylic acid functions or the nanotube sidewall. Finally, ICP/AES of Gd-CNTs 5 gave a more precise value of the functionalization degree; 460 ± 4 µmol of gadolinium per gram of nanotubes. Overall, characterization of the CNT conjugates by ICP/AES, TGA, and xylenol orange test is consistent and demonstrate full chelation of Gd 3+ by DTPA. TEM analysis of Gd-CNTs 5 shows that conditions used for functionalization did not alter the morphology and dispersion of the nanotubes (Supporting Information Figure S1). The ultrastructure of Gd-CNTs 5 , detailed by HR-TEM, revealed structural defects and dislocations within the external graphene layers on which it is likely that the functional groups and Gd-DTPA are anchored (Figure 2 B). Gd-CNTs were also characterized using high angle annular dark fi eld (HAADF)-STEM where the intensity in the image is proportional to the atomic number of the elements present in the sample. HAADF-STEM has been shown as a powerful technique to allow the direct visualization of heavy-element bearing organic molecules down to the atomic scale. [ 31 ] In the HAADF-STEM images of Gd-CNTs 5, bright dots correspond to the Gd atoms (Figure 2 C), while they did not appear in the HAADF-STEM images of DTPA/CNTs 4 . The presence of gadolinium in Gd-CNTs 5 was confi rmed by EDX (Figure 2 D), while no Gd was found in the case of DTPA/CNTs 4 .

Magnetic Characterizations
Magnetic characterization confi rmed the paramagnetic properties of Gd-CNTs 5 in comparison with the ox-CNT precursor 1 . The fi eld-dependent magnetization curve at 5 K ( Figure 3 A) was fi t using a Brillouin function characteristic of paramagnetic materials, with a J angular momentum of 3.2. This gives an effective magnetic moment of 6.9 µ B , close to the theoretical magnetic moment of Gd 3+ ( µ eff = 7.94 µ B , J = 7/2). However, fi eld-cooled (FC) and zero-fi eld-cooled (ZFC) magnetization measurements (using magnetic fi elds of 100 and 1000 Oe for the former) revealed an additional ferromagnetic component (Figure 3 B) which is added to the paramagnetic contribution arising from gadolinium. The FC and ZFC magnetization curves were not strictly superimposable as expected for purely paramagnetic behavior (inset in Figure 3   low temperature ferromagnetic component (with a blocking temperature of about 65 K) was also displayed by the gadolinium-free control nanotubes (ox-CNTs) (Figure 3 C). This suggests that the ferromagnetic contribution arises from residual catalyst impurities (620 ppm of iron and 1283 ppm of nickel), which are probably entrapped into the inner empty cavity of CNTs and were not completely eliminated by the acid treatment, for both ox-CNTs and Gd-CNTs. The chelation of gadolinium into DTPA/CNTs gives rise to the stronger paramagnetic contribution, demonstrated by the linear dependence of magnetization with inverse temperature. MRI contrast agents (CAs) act as enhancers of longitudinal and transverse local 1 H relaxation rates R 1 = 1/ T 1 and R 2 = 1/ T 2 . The effi cacy of a gadolinium-based CA is measured by its relaxivities r 1 and r 2 defi ned as r i = ( R i -R i 0 )/[Gd], where i = 1, 2 and R i 0 is the endogenous relaxation rate (Figure 3 D). Gd-DTPA (Magnevist, Bayer Schering Pharma) is one of the most common clinically approved CAs used in the majority of clinical procedure. [ 32 ] It has r 1 value of 4.1 s −1 m M −1 and r 2 of 4.6 s −1 m M −1 at an imaging fi eld of 1.5 T (or 60 MHz 1 H Larmor frequency). A relaxivity ratio r 2 / r 1 of close to 1 ensures that the CA acts as a good local signal enhancer (a positive CA), particularly in T 1 -weighted sequences. Here, the MR properties of Gd-CNTs are compared to those of a Gd-DTPA chelate we synthesized (Supporting Information Figure S3). The longitudinal relaxivities, r 1 , were measured at 25 °C over a large range of 1 H Larmor frequencies (from 0.01 to 200 MHz). Interestingly, the resulting nuclear magnetic relaxation dispersion (NMRD) profi le of the Gd-CNT suspension is distinctly different from that of non-conjugated Gd-DTPA, with r 1 values enhanced by a factor of two at low frequency (Figure 3 E). The decrease of r 1 at clinical fi elds above 10 MHz is less marked for the Gd-CNTs in comparison to Gd-DTPA. However, the result is a threefold increase in r 1 at 200 MHz (4.7 T) due to conjugation to CNTs. This effect could be due to an enhancement of the innersphere relaxivity mechanism by slowing down the tumbling of Gd-chelates which exhibit fast exchange of the coordinated water molecules. This characteristic effect has been achieved before by binding Gd ions to objects of increasing size from macromolecules to proteins or nanoparticles. [ 33 ] In our case, the nearly one-dimensional structure of functionalized CNTs together with their high dispersibility, highly hydrated surface and high permeability to water may favor water exchange, while increasing the correlation time for tumbling of the whole conjugate. Importantly the huge increase of low fi eld relaxivity demonstrated by non-covalent confi nement of Gd-complexes or gadolinium oxide into nanosystems permeable to water, such as zeolites, apoferritin, silicon microparticles and single-walled carbon nanotubes was not observed. [ 23a , 34 ] This is consistent with an external covalent coupling of water-accessible Gd-chelates at the surface of the oxidized nanotubes. It should also be noted that we did not observe variations of relaxivity with CNT concentration, as previously reported. [ 25 ] All relaxation rates showed linear variation with CNT and Gd concentration with identical r 1 relaxivities whether CNTs have been dispersed in water or further immobilized in agarose gel (Figure 3 D). All relaxation rates showed linear variation with CNT and Gd concentration, with identical r 1 values obtained irrespective of whether the Gd-CNTs were dispersed in water or immobilized in agarose gel (Figure 3 D). The former observation confi rms both the remarkably high dispersibility of the functionalized CNTs and their colloidal stability in water suspension over the concentration range studied. The latter suggests that further increases in the overall tumbling correlation time (in agarose) have no effect, and so this is no longer the r 1 -limiting process. Once bound to the CNT it seems that the timescale for local dynamics about the chelate determines the relaxivity.
Transverse relaxivities r 2 were measured in the frequency range between 20 MHz (0.47 T) and 200 MHz MHz, respectively. This change corresponds to an evolution of the r 2 / r 1 ratio from 1.8 to 0.7 for Gd-CNTs and from 1.42 to 1. 40 for Gd-DTPA when the imaging fi eld is increased from 0.47 to 4.7 T. Hence, in terms of r 2 / r 1 ratio, the effi cacy as a positive CA is enhanced for Gd-CNTs on increasing the fi eld, in comparison to Gd-DTPA. Therefore, for the in vitro and in vivo evaluation of Gd-CNTs, an imaging fi eld of 4.7 T was used.

Cellular Uptake and MRI Cell Imaging
As intrinsic MR contrast highly depends on tissue structure and water dynamics, the ability of a CA to modify the 1 H signal is strongly affected by the local environment and by CA organization in this environment. During their odyssey through the organism, CNTs are likely internalized by cells, and most probably by cells of the reticulo-endothelial system (resident macrophages or circulating phagocytic cells). However, the magnetic properties of MRI contrast agents are deeply affected by cellular internalization and intracellular compartmentalization. For instance it is well-known that there is a drastic reduction of the longitudinal relaxivity of iron oxide nanoparticles following cellular uptake. [ 35 ] Similar effects were observed for gadolinium oxide nanoparticles depending on the internalized dose [ 36 ] and for a macromolecular Gd(III)-based probe [ 37 ] depending on its intracellular localization. [ 38 ] This so-called intracellular "quenching" effect tends to enhance r 2 / r 1 ratio, in such manner  that the generation of positive contrast by labeled cells becomes challenging. Regarding CNTs, it is not clear how r 1 could be modifi ed by cell confi nement and how intrinsic properties of precursor ox-CNTs could affect transverse and longitudinal relaxation in that environment. It was reported that, despite a very high r 1 value in suspension, gadolinium-loaded SWCNTs failed to induce positive contrast when internalized by murine macrophages and were subsequently detected with a negative signal on T 2 or T 2 *-weighted sequence. [ 21c ] Likewise, CNTs that were free of any gadolinium label [ 18b ] were found to induce 1 H signal suppression in mouse liver. These confusing results call for a thorough examination of MR properties depending on cellular uptake and on the intrinsic magnetic properties of CNTs. In this work, RAW 264.7 mouse macrophages were used as a cell model to study Gd-CNT uptake in vitro. The cells were incubated for 24 h with Gd-CNTs in the concentration range of 0-100 µg/mL. Cell morphology and adherence on culture fl asks was not affected by CNT labeling as shown by optical microscopy ( Figure 4 A). Cellular metabolic activity, assessed by the Alamar blue test, revealed a slight decrease in comparison to non-labelled cells at the highest concentration tested (100 µg/mL) (see Supporting Information Figure S4). TEM confi rmed the intracellular uptake of CNTs; indeed elongated hollow structures were easily identifi able within the intracellular organelles (Figure 4 B). Most nanotubes were assembled as large bundles within vesicular membranes, only a small fraction could be observed isolated within the cytoplasm. This suggests an active process of cellular internalization, involving membrane traffi cking and dynamic compartimentalization within lysosomal storage organelles. The dose dependent CNT uptake by macrophages was assessed by imaging fl ow cytometry (ImageStream). [ 11b ] This technique combines the statistical power of high throughput fl ow cytometry with imaging of each analyzed cell. [ 11b , 39 ] As illustrated in Figure 5 A, Gd-CNT-labeled cells exhibit dark spots on bright fi eld images that were not observed in non-labeled cells. These dark spots colocalized with high intensity spots on the dark fi eld image, corresponding to areas of CNT accumulation that strongly absorb and scatter the light. The biparametric dot plot representing the mean intensity on dark fi eld versus the mean pixel intensity on bright fi eld (Figure 5 B) clearly shows that both parameters increase (in absolute value) with increasing concentration of CNTs in the incubation medium. This indicates a dose dependent CNT uptake for the entire cell population, which tends to saturate at the highest concentration. Similar results were obtained for both Gd-CNTs and ox-CNTs.
Subsequently the MR contrast properties of cells labeled with increasing concentration of Gd-CNTs or ox-CNTs and dispersed in agarose gels mimicking the cell density of tissue (7 × 10 6 cells in 70 µL of gel) were examined. MR images were acquired using a spin echo sequence, while varying the echo time T E or repetition time T R in order to determine T 1 and T 2 values and the optimal conditions for T 1 -or T 2 -weighting. Representative T 1 -weighted images in Figure 6 A show a dose-dependent signal increase in the macrophages labeled with Gd-CNTs in comparison to the non-labeled cells or to the cells labeled with ox-CNTs. The signal increase was nearly twofold for the highest Gd-CNT concentration with respect to non-labeled cells. Conversely, in T 2 -weighted images, the macrophages labeled with Gd-CNTs display decreased MR signal intensity with increasing CNT concentration (Figure 6 A). Hence, the Gd-CNTs act as a bimodal contrast agent for cell imaging, generating positive contrast under T 1 -weighted and negative contrast under T 2 -weighted imaging conditions. By comparison, no signal change was Adv. Funct. Mater. 2014, 24, 7173-7186 www.afm-journal.de www.MaterialsViews.com observed for macrophages labeled with ox-CNTs. The quantitative determination of relaxation rates (1/ T i expressed in s −1 , i = 1,2) reveals a linear increase of 1/ T 1 with Gd-CNT concentration in macrophages, whereas no variation was observed after labeling with control ox-CNTs (Figure 6 B). This shows that Gd-DTPA chelates covalently coupled to CNTs can still be effective as T 1 -contrast agents when internalized in cells. Regarding the transverse relaxation, the analysis is more complex. In comparison to cell-free agarose, the high density of cells induces a 3-fold increase in transverse relaxation time 1/ T 2 (Figure 6 B). This observation is consistent with relaxation dynamics in tissue, which is accelerated in comparison to water. [ 40 ] More surprisingly, 1/ T 2 also increased with the concentration of CNTs in cells, regardless of whether they were labeled with Gd-CNTs or ox-CNTs. The T 2 effect appears as an intrinsic characteristic of CNTs, not of the complexed gadolinium probe. In summary covalent grafting of Gd-DTPA induces an effi cient intracellular T 1 reduction that depends on CNT cellular uptake. Despite the associated T 2 reduction in labeled cells arising from the nanotubes, positive cell contrast due to the paramagnetic label can be observed when appropriate T 1 -weighting is applied.

In Vivo Imaging
The next step was to evaluate the use of Gd-CNTs in vivo to monitor the organ biodistribution of functionalized nanotubes. Gd-CNTs were injected intravenously to mice at two different doses (50 µg and 250 µg corresponding to 2.5 and 12.5 mg/kg, respectively) and monitored over 5 hours following injection.
A T 1 -weighted sequence (fast spin echo) was applied since it offered the best contrast-to-noise ratio after Gd-CNT administration. Following the injection of the low dose of Gd-CNTs, signifi cant signal changes in liver, spleen, kidneys and bladder were not observed. However a fi ve-fold increase in Gd-CNT dose (250 µg corresponding to a gadolinium dose of 5.75 µmol/ kg body weight) yielded a clear signal enhancement in liver, spleen and to a much larger extent in the bladder, where signal enhancement in excess of +250% was obtained ( Figure 7 ). While the contrast enhancement decreased after 2 h in the bladder (Figure 7 C), it was slightly increased over time in liver reaching a value of +27% after 320 min (Figure 7 AB). In spleen the signal reached a maximum after two hours (+16%) and then decreased to close to its initial value (+3%). Histological specimens of hepatic and splenic tissues confi rmed the presence of nanotubes with a spotty homogenous distribution in liver and in the red pulp periphery of the spleen ( Figure 8 and Supporting Information Figures S5,S6).
Unambiguous identifi cation of carbon nanotubes in the intracellular compartments of Kupffer cells in liver was provided by electron energy loss spectroscopy spectrum (EELS). This technique can differentiate graphitic sp 2 signal from CNTs and amorphous carbon signal either from the endogenous cellular background or from carbon membranes present on TEM grids ( Figure 9 ). [ 41 ] Moreover, EDX spectroscopy also confi rmed the presence of gadolinium co-localizing with carbon nanotubes within membrane-bound intracellular compartments that are presumably lysosomes, in both liver and spleen ( Figure 9 ). Thus, the increased MR signal in these organs is unambiguously related to the rapid uptake of CNTs by the Adv. Funct. Mater. 2014, 24, 7173-7186 www.afm-journal.de www.MaterialsViews.com Figure 5. Imaging fl ow cytometry analysis of Gd-CNT uptake by RAW 264.7 macrophages. A) Representative bright fi eld and corresponding dark fi eld images of cells (among 10 000) acquired with ImageStream for different Gd-CNT incubation conditions. Black spots on bright fi eld images corresponding to bright spots on dark fi eld images are indicative of CNT uptake. B) Relative quantifi cation of Gd-CNT uptake on biparametric dot plot of dark fi eld intensity versus mean pixel intensity on bright fi eld image.
reticuloendothelial system, as confi rmed by histology and TEM. The massive and transient signal enhancement in bladder also suggests that a signifi cant fraction of Gd-CNTs is eliminated by renal clearance. Both histology and TEM confi rmed MRI results showing that Gd-CNTs do not accumulate in the kidneys (as only one isolated aggregate was observed in histological sections), which is consistent with effective elimination in the urine (Supporting Information Figure S7). Overall, our fi ndings are consistent with the recent SPECT/CT imaging study on functionalized CNTs radiolabeled with 111 Indium-DTPA. Different distribution profi les have been reported with a balance of renal clearance versus RES accumulation depending on the degree of chemical functionalization of the nanotubes. [ 16a ] Individual functionalized CNTs with outer diameter lower than the glomerular fi ltration cut-off have been shown to escape the glomerular kidney fi lter to be excreted in urines in several studies. [ 42 ] In contrast, most of the largest CNTs or CNTs forming bundles in vivo are uptaken in liver, spleen and lung. [ 5 ] Therefore, full dispersion (or individualization) of CNTs which is favored by chemical functionalization appears to be a pivotal factor determining their excretion profi le. In the present study, a single bolus injection of a relatively high dose of Gd-CNTs (250 µg versus 50 µg and 10 µg in two recent studies) [ 16a , 26 ] was used with no acute or sub-acute effect on the animal behavior apparent (until day 8 post-injection) and with no signs of tissue damage, infl ammatory response and cellular toxicity that could be noticed on histological and TEM images in liver, spleen, kidney and lung. This clearly indicates that the dispersion of the functionalized nanotubes in physiological saline medium at the concentration of 5 mg/mL was appropriate for intravenous administration. Moreover, unlike previous studies reporting accumulation of CNTs in the lungs after their systemic administration, [ 5 ] the lung histological specimens show very few nanotubes despite the high dose injected (Supporting Information Figure S7). The lack of lung accumulation is a further indication of the good dispersibility of Gd-CNTs which persists in blood circulation. The ultrastructural examination of Gd-CNTs in liver and spleen also revealed that well individualized CNTs Adv. Funct. Mater. 2014, 24, 7173-7186 www.afm-journal.de www.MaterialsViews.com

Conclusion
In conclusion, we covalently functionalized oxidized MWCNTs with DTPA for chelation of Gd 3+ and the Gd-CNTs were administrated intravenously to mice. The CNT uptake in liver and spleen could be successfully assessed using in vivo MRI owing to the preserved capacity of Gd-CNTs to generate signal enhancement after cellular internalization. In addition, the rapid clearance into the urines of extracellular functionalized CNTs confi rmed previous studies using other different imaging modalities. [ 43 ] Overall, this study represents the fi rst success in monitoring the CNT distribution with functionalized tubes used to generate a T 1 positive MR contrast after intravenous injection. This work shows that combining the properties of Gd 3+ and CNTs allows performing MRI in vivo. MRI has a high resolution compared to other imaging techniques and can thus be used as an alternative imaging method to track CNTs. Despite the fact that the amount of gadolinium chelates injected with nanotubes is far lower the gadolinium dose in clinical practice (5.75 µmol/kg versus 100 µmol/kg body weight), the high longitudinal relaxivity of Gd-CNT conjugates enabled their detection with T 1 -weighting in mice using a 4.7 T scanner. In contrast to other strategies using nanoparticles to enhance MR contrast of carbon materials, the present approach exploits paramagnetic molecular chelates that are widely used in the clinics and minimally interfere with the structure, dispersibility, functionalization, loading capacity and degradability of CNTs. The approach to material design described here could open new routes to highly dispersible nanomaterials functionalizable with signifi cant payloads for drug delivery, tumor targeting and photothermal therapy or vaccination. The second aspect is the retention of T 1 -weighting properties for MRI following cellular uptake, which may provide possibilities for study the biodistribution of functionalized CNTs and for image guided delivery.

General Information
MWCNTs (20-30 nm diameter, 0.5-2 µm length, 95% purity; batch 1240XH) were purchased from Nanostructured and Amorphous Materials. When stated, suspensions of CNTs were sonicated in a water bath (Transsonics Digitals Elma, 20 W, 40 kHz). All reagents and solvents were purchased from different commercial suppliers and used Adv. Funct. Mater. 2014, 24, 7173-7186 www.afm-journal.de www.MaterialsViews.com as received. In particular, xylenol orange tetrasodium salt was purchased from Alfa Aesar. Thermogravimetric analyses were performed by using a TGA Q500 TA instrument with a ramp of 10 °C min −1 under N 2 using a fl ow rate of 60 mL min −1 . 1 H and 13 C NMR spectra were recorded on Bruker DPX 300 instrument. The peak values were obtained as ppm (δ) and referenced to the solvent. The resonance multiplicity is indicated as s (singlet), br s (broad singlet), and m (multiplet). LC/MS analyses were performed on ThermoFisher Finnigan 6 LCQ Advantage Max. TLC was performed on aluminum sheets coated with silica gel 60 F254 (Merck, Darmstadt). Chromatographic purifi cation was done with Merck silica gel (Kiesegel 60, 40-60 µm, 230-400 mesh ASTM). Dialysis was performed using Spectra/Por dialysis membrane, MWCO: 12-14 000 Da.

Functionalization of MWCNTs
ox-CNTs 1 : A suspension of pristine MWCNTs (1 g) in a mixture of sulfuric acid 98% (108 mL) and nitric acid 65% (36 mL) was sonicated in a water bath for 24 h. The temperature was kept below 35 °C. Deionized water was then carefully added and the CNTs were fi ltered (Omnipore TM membrane fi ltration, 0.45 µm), re-suspended in water, fi ltered again until pH became neutral and dried under vacuum.
Functionalized CNTs 3 : A suspension of ox-CNTs 1 (21 mg) in oxalyl chloride (8.5 mL) was sonicated in a water bath for 30 min and then heated at refl ux for 24 h. After evaporation under vacuum, the acyl chloride-CNTs were dispersed in a solution of DTPA ligand 2 (237 mg) in dry THF (10 mL). The mixture was sonicated for 2 min and then heated at refl ux for 60 h. The suspension was fi ltered over a PTFE membrane (0.1 µm, Omnipore, Millipore). The solid recovered on the fi lter was dispersed in DMF (100 mL), sonicated for 3 min in a water bath, and fi ltered over a PTFE membrane (0.1 µm). This sequence was repeated with DMF, methanol (twice), and dichloromethane. The resulting solid was dried under vacuum.
DTPA/CNTs 4 : A suspension of the functionalized CNTs 3 (21 mg) in TFA (1.8 mL) and water (200 µL) was sonicated in a water bath for 3 min and stirred at room temperature for 2 h. A solution of TFA (1.8 mL) Adv. Funct. Mater. 2014, 24, 7173-7186 www.afm-journal.de www.MaterialsViews.com Figure 9. A) TEM of liver 6 h after injection of Gd-CNTs. B) EELS analysis on carbon K-edge can differentiate between Gd-CNTs and the carbon rich cellular background. The strength of the fi rst sharp peak (285 eV) due to the 1s to 2π* can be related to the predominance of sp 2 hybridizations. Therefore, the strong initial peak unambiguously reveals the presence of carbon nanotubes in liver. This characteristic signal is easily differentiated from the carbon rich cellular background which shows an amorphous carbon-like K-edge. C) EDX spectrum confi rms the presence of Gd on CNTs internalized by liver Kupffer cells.  24 h. An aliquot of the suspension was centrifuged. The supernatant was titrated with xylenol orange to assess the amount of free Gd. The reaction mixture was then dialyzed against water for 2 days. The suspension was lyophilized to obtain Gd-CNTs 5 .

Aberration-corrected High Resolution Transmission Electron Microscopy (HR-TEM), Energy-dispersive X-Ray Spectroscopy (EDX), and High Angle Annular Dark Field Scanning Transmission Electron Microscopy (HAADF)-STEM
: HR-TEM investigations were performed with the newly developed JEOL ARM 200 F microscope operating at 80 kV. This microscope is equipped together with a CEOS aberration corrector, a cold fi eld emission gun, a Gatan GIF quantum ER and a JEOL EDX Diode. [ 45 ] HAADF-STEM images and EDX spectra were acquired on a FEI Tecnai G2 F20 HRTEM operated at 200 kV and equipped with an EDAX super ultra-thin window (SUTW) X-ray detector. Samples were deposited on lacey carbon Cu TEM grids (Agar).
Inductively Coupled Plasma Atomic Emission Spectrometry (ICP-AES) Analysis : Determination of gadolinium, iron and nickel content in CNTs was performed by ICP-AES analysis (Agilent 7500ce). The samples were digested in concentrated HNO 3 (3:1, v/v) solution and diluted with ultrapure water for the analysis.
Magnetization Measurements : Magnetic measurements were carried out on a Quantum Design MPMS-5S SQUID Magnetometer. Fielddependent magnetization curves were measured at 5K as function of the external fi eld in the range 0-5 Tesla. Field-cooled (FC) and zerofi eld-cooled (ZFC) temperature-dependant magnetization was measured using a magnetic fi eld of 100 Oe or 1000 Oe and the sample was cooled with the same magnetic fi eld.
NMRD and Relaxivity Measurements : The frequency dependence of the 1 H relaxation for the aqueous nanotube suspensions was recorded over the frequency range 0.01-20 MHz using a Spinmaster FFC-2000 Fast Field Cycling NMR Relaxometer (Stelar SRL, Mede, Italy). T 1 measurements were performed as a function of external fi eld, B 0 , with standard pulse sequences incorporating B 0 fi eld excursions. The temperature of the samples was maintained at 25 ± 1 °C using a thermostated airfl ow system. All of the 1 H magnetisation recovery curves were singly exponential within experimental error and the random errors in fi tting T 1 were always less than 1%. T 2 was measured using the CPMG pulse sequence at 20, 40, and 60 MHz using the Stelar spectrometer and a reconditioned Bruker WP80 electromagnet. T 2 was measured using the CPMG pulse sequence at 20, 40 and 60 MHz. Images of CNT liquid suspensions, agarose gel (0.3%) and cell samples (7 × 10 6 cells in 70 µL agarose) were obtained on a Bruker Bio-Spec 47/40 USR scanner operating at 4.7 T (200MHz) using fast spin echo sequences with variable echo time T E and repetition time T R . T 1 and T 2 were deduced from monoexponential fi t of MR signal as function of T R and T E , respectively, using the ParaVision software.

Cell Culture, Cell Labeling, Cell Metabolic Activity, and Cell Preparation for TEM
The murine macrophage RAW 264.7 cells were maintained as monolayer cultures in RPMI (Roswell Park Memorial Institute) medium supplemented with 10% fetal bovine serum, 1% penicillin-streptomycin at 37 °C and 5% of CO 2 . For cell labeling, cells were incubated with complete RPMI medium supplemented with CNTs at concentration of 10, 20, 50, 100 µg/mL for 20 h at 37 °C. Cell metabolic activity was assessed by the Alamar blue test in 48-well plates seeded with cells the day before labeling. The labeled macrophages were washed twice in PBS and were incubated with 10% Alamar Blue in culture medium for 2 h. The fl uorescence in cell medium due to the reduction of resazurin (oxidized form) to resorufi n by cell activity was quantifi ed on a FLUOstar OPTIMA microplate reader (excitation 550 nm, emission 590 nm) and normalized to the fl uorescence signal of the control non-labeled cells. The conditions were run in quadruplicate. For TEM observation, cells were trypsinized, centrifuged at 1200 rpm for 5 min, washed in 0.1 M cacodylate buffer and fi xed with 2.5% glutaraldehyde in cacodylate buffer at 4 °C for 30 min. Cells were then post-fi xed with osmium tetraoxide 1% and passed through uranyl acetate. Samples were then dehydrated in an ethanol series (30-100%) and embedded in epoxy medium (EPON 812; Shell Chemical, San Francisco, California). Thin sections (70 nm) were collected onto 200 mesh cooper grids, and counterstained with lead citrate before examination with Zeiss EM902 electron microscope operated at 80 kV (MIMA2-UR1196 Génomique et Physiologie de la Lactation, INRA, Plateau de Microscopie Electronique 78352, Jouy en Josas, France).
Imaging Flow Cytometry : For imaging fl ow cytometry analysis, labeled macrophages were trypsinized, fi xed with 4% formaldehyde in PBS and centrifuged to obtain a pellet of about 10 6 cells in 50 µL. Cell images were acquired using the ImageStream X multispectral imaging fl ow cytometer (Amnis Corporation), collecting 10,000 events per sample at 40× magnifi cation. Dark fi eld images were acquired using a 785 nm laser. Cell images were analyzed using IDEAS image-analysis software (Amnis). Gating on bivariate plot of aspect ratio versus cell area was fi rst used to isolate a population of single cells. Cells within the focal plane were further selected using a two-dimensional plot of image contrast versus root mean squared (RMS) gradient. Mean pixel feature on bright fi eld images versus intensity feature on dark fi eld images was used as a biparametric dot plot to discriminate cells with respect to their CNT uptake.

Animal Experiments and MRI Protocol
Animal experiments were conducted in line with French Agriculture Ministry guidelines. During the MRI protocol the animals were anesthetized with 2% isofl uorane (Aerrane, Baxter, Maurepas, France) supplied in an air mixture, while their body temperature was kept constant by circulating thermostated warm water. Twelve eightweeks-old female C57/Bl6 mice (mean weight 20.5 ± 1 g) (Janvier, France) were used in this study. The mice were imaged by the whole body MRI 20 min before injection and 10, 75, 120, 320 min after injection of Gd-CNTs in the retro-orbital sinus at a dose of 1 or 5 mg/mL suspended in a 100 µL physiological saline medium. Three animals were analyzed at each time point. MRI was performed on a Biospec 47/40 USR (40-cm bore actively shielded 4.7 Tesla magnet) scanner interfaced to ParaVision software (both provided from Bruker Biospin GmbH, Rheinstetten, Germany). The whole body imaging protocol was performed with a volume transmission/reception RF coil for mice (Bruker), using a fast spin echo (Turbo RARE) sequence with a T E of 12 ms and a T R of 445 ms ( T 1 -weighted sequence), fl ip angle of 180°, 30 averages, in-plane resolution of 150 µm, slice thickness of 500 nm, for a total of seven slices. At the end of the experimental protocol, mice were sacrifi ced with a lethal injection of sodium pentobarbital. Livers, spleens, lungs and kidneys were excised and prepared for histological and TEM characterizations. Image processing and analysis were made with the open source software OsiriX water tube reference and SD is the standard deviation of the noise in the image. A water tube, used as a reference, was positioned near the mice to enable CNR measurement and 1 H NMR signal normalization. ROIs encompassing the liver parenchyma, the whole spleen, the kidneys and the bladder were manually selected for signal measurement.
Histology : After excision, pieces of liver, spleen, lung and kidney were fi xed with pH 7.4 phosphate-buffered 10% formalin and processed by embedding in paraffi n. Sections, ≈6 µm thick, were evaluated after hematoxylin and eosin or Nuclear Fast Red and Pearl's staining (all from Sigma-Aldrich, Steinheim, Germany).
Transmission Electron Microscopy in Organs : Organs were cut into 1 mm 3 pieces after excision and fi xed with 2% glutaraldehyde in 0.1 M sodium cacodylate buffer, postfi xed with 1% osmium tetroxide containing 1.5% potassium cyanoferrate, gradually dehydrated in ethanol and embedded in Epon. Thin sections (70 nm) of selected zones were observed with Zeiss EM902 electron microscope operated at 80 kV as described above for cells. Ultra-thin sections (30 nm) of organs were specially prepared for high resolution imaging, EDX analysis and Electron Energy Loss Spectroscopy (EELS) analysis on the carbon K-edge at 284 eV, performed on the aberration-corrected JEOL ARM 200F operating at 80 kV.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author. TEM of ox-and Gd-CNTs, HAADF-STEM images of ox-CNTs, synthesis of DTPA ligand 2 , HAADF-STEM images and EDX spectra of Gd-CNTs and DTPA/CNTs, synthesis of Gd-DTPA/ligand 8 , cell viability, histology in liver, spleen, kidney and lung, and TEM, EELS and EDX in spleen.