Functionalized carbon nanotubes : from intracellular uptake and cell-related toxicity to systemic brain delivery

Citing this paper Please note that where the full-text provided on King's Research Portal is the Author Accepted Manuscript or Post-Print version this may differ from the final Published version. If citing, it is advised that you check and use the publisher's definitive version for pagination, volume/issue, and date of publication details. And where the final published version is provided on the Research Portal, if citing you are again advised to check the publisher's website for any subsequent corrections.


Introduction
Carbon nanotubes (CNTs) are defined as cylindrical nanomaterials composed of a continuous, unbroken hexagonal mesh of carbon atoms. The first observation of CNTs by electron microscopy, credited to Iijima in 1991, opened a plethora of applications for this material [1]. This included high-strength composites, energy storage, field emission device, but also the use of CNTs for biomedical applications [2]. In particular, CNT ability to cross efficiently cell membranes and carry large amount of molecules has encouraged the design of nanotube-based delivery systems [3,4].
The concept of drug delivery was probably introduced by Paul Ehrlich, in 1897, when he theoreticized the use of "zauberkugeln" (in English "magic bullets") intending to improve the efficacy of available therapeutics [5]. Long after this statement, delivery of therapeutic and imaging agents into specific organs or tissues has remained a promising approach to modulate the pharmacokinetics and bioavailability of therapeutics, and provide controlled release kinetics at a target site.
Numerous materials with sizes between 10 to 1000 nm have been investigated, including liposomes, dendrimers, nanoemulsions, nanoparticles, quantum dots and CNTs. With their needle-like shape, CNTs display singular physico-chemical properties. Their large surface area, ranging from 50 to 1315 m 2 /g, allows the conjugation with extensive amount of therapeutic and imaging molecules [6][7][8].
Moreover, the high CNT length-to-diameter ratio enables them to efficiently penetrate biological membranes and accumulate into intracellular compartments [9].
Consequently, attachment of molecules to CNTs helps overcoming several administration problems, including insolubility, poor biodistribution and inability of therapeutic or diagnostic molecules to cross cellular barriers [3].

A C C E P T E D M A N U S C R I P T
ACCEPTED MANUSCRIPT 7 Despite their undeniable potential, concerns have emerged regarding the toxicity of CNTs, as various reports showed that pristine nanotubes could induce biological damage [10]. Excessive nanotube length, the presence of impurities from the synthesis process and the introduction of carboxylic groups at the CNT surface could trigger unattended and detrimental cellular responses [11]. Such parameters must therefore be thoroughly controlled and characterised to design safe and biocompatible nanotubes applicable as delivery systems. The post-synthesis surface modification of nanotubes with hydrophilic molecules, named functionalisation, has been reported as an efficient approach to enhance their water dispersibility and reduce their toxicity [12,13]. This can be performed by covalently attaching moieties at the surface of CNTs or by non-covalent interactions between nanotube surface and hydrophobic/aromatic regions of amphiphilic molecules [14].
To tailor nanotube function, therapeutic molecules or imaging probes can be added to functionalised CNTs (ƒ-CNTs) side-walls [4]. By taking advantage of their inner cavity, ƒ-CNTs can also be filled to keep the surface available for further modifications [15]. Contrast agents can be combined to nanotubes to generate CNTbased hybrids with clinical imaging capabilities [16]. If such hybrids display desirable targeting capabilities, they become versatile imaging tools for diagnostic applications [17,18]. CNT hybrids can also help tracking administrated nanocarriers to assess in real-time their spatial distribution and therefore measure their biodistribution profile [19]. The major medical imaging techniques, namely ultrasound, nuclear and magnetic resonance imaging (MRI), display limitation in terms of sensitivity or image resolution. To improve this, the combination of synergistic imaging modalities in a single carrier, such as CNTs, could be particularly valuable [20]. Beyond the promising properties of CNT-based hybrids for multi-imaging capabilities, their dimensions need to be optimised in order to control their intrinsic imaging properties, improve their accumulation in target cells and enhance their biocompatibility profile. This dimension refinement is essential to demonstrate the potential of CNT-based hybrids and confirm their safety before conducting clinical studies.
A wide range of studies have also reported on the development of carbon nanotubes for brain delivery, with results showing that adequate functionalisation is essential to produce biocompatible CNTs capable of local or systemic delivery of therapeutics to brain cells [21].
In this review, a description of the physico-chemical properties and surface modification of CNTs needed for delivery will be presented. Moreover, the interaction between CNTs and mammalian cells will then be described, followed by a summary of their toxicity. Finally, we will look into the most recent advances involving CNT-mediated systemic brain delivery and in situ CNT biodegradation.

Synthesis, classification and properties
Carbon nanotubes can be generated by electric arc discharge and laser ablation using vaporisation of graphite target [22,23]. Alternatively, they are synthesised by chemical vapour deposition which rely on the passage of carbon-containing vapours in a furnace containing metal catalysts [24]. CNTs can be classified as single-walled (SWNT) or multi-walled (MWNT) nanotubes, in accordance with the number layers that compose a single nanotube (Figure 1).

Figure 1. Schematic representation of single-walled (SWNT) and multi-walled (MWNT) CNTs.
Single and multi-walled CNTs have similar structures but different diameter. The figure was redrawn and modified from [25] and [26].
SWNT and MWNT exhibit a diameter of 0.4-2 nm and 10-100 nm, respectively [26]. Both types are utilised as delivery systems and display large aspect ratios with lengths ranging from 50 nm to several microns. The length and diameter can be tuned by controlling the production conditions, but the design of CNT-based delivery systems require further post-synthesis shortening procedures to increase their biocompatibility and bioavailability [10,19]. A reduction in the CNT length to diameter ratio can be achieved by strong acid treatment, ultrasonication, steampurification and mechanical methods [27][28][29].
The unique physicochemical properties of CNTs, namely high surface area and length-to-diameter ratio, optimal electrical conductivity, and thermo-chemical stability, make them particularly attractive for biomedical applications [30].
However, pristine CNTs must be functionalised to improve their hydrophilicity and biocompatibility.
A C C E P T E D M A N U S C R I P T ACCEPTED MANUSCRIPT 10

Non-covalent functionalisation
The delocalised aromatic system of nanotubes layers makes them aggregate in bundles and results in their poor dispersibility in physiological aqueous environment [31]. The large surface area of CNTs enables non-covalent or covalent conjugation of hydrophilic molecules to enhance their dispersibility [13]. The non-covalent modification consists in the physical adsorption of amphiphilic surfactant molecules onto the surface of the CNTs by Van der Waals interaction, π-π stacking or electrostatic interaction [32][33][34]. The main advantages of this approach are the preservation of the intrinsic optical properties of CNTs and the simplicity of the functionalisation procedure [30]. However, the interaction after coating should be limitedly affected by presence of salt to maintain the stability of the complex in a physiological environment. Biocompatible polymers (e.g. Pluronic ® F-127, polyethylene glycol (PEG)) [35,36], gum arabic [37], single-stranded DNA (ssDNA) [38] and proteins (e.g. bovine serum albumin, BSA) [39,40] were reported to increase the dispersibility of CNTs.

Covalent functionalisation
Covalent CNT functionalisation relies on the chemical bound of functional groups to the wall of CNTs, and is also referred as chemical functionalisation [41]. In contrast to the non-covalent approach, chemical functionalisation leads to strong and stable chemical bonds grafted onto the sp 2 carbon framework of the tips and sidewall of CNTs [42]. The functionalisation of CNT can be carried out by oxidation under strong acidic conditions, which produces carboxylic acid groups and shortening of CNTs [43]. However, the introduction of carboxylic groups at the surface of CNTs has been associated with cellular toxicity [28,44]. Further reactions can conjugate A C C E P T E D M A N U S C R I P T ACCEPTED MANUSCRIPT 11 carboxylic acid groups to an amine or alcohol groups to obtain amide or ester linkage, respectively [13]. In addition to oxidation, another common chemical reaction is the 1,3-dipolar cyclo-addition using the condensation of an amino acid and an aldehyde [45].
Both covalent and non-covalent functionalisations have shown their ability to increase the dispersibility of nanotubes in a physiological environment, making them available to cross cell membranes and accumulate into intracellular compartments.

Uptake and cellular fate of functionalised CNTs (ƒ-CNTs) in cancer cells and macrophages
Efficient uptake properties of CNTs have encouraged their use as drug delivery systems. After crossing the plasma membrane, the intracellular pathways of CNTs can lead to organelle accumulation and/or nanotube elimination [46][47][48][49][50]. The characterisation of their mechanisms of uptake and elimination remains of great interest to shape the delivery properties of CNTs and ensure their bio-clearance.

Passive versus active mechanism
A leading study by Pantarroto and collaborators demonstrated that the high aspect ratio of CNTs allowed them to efficiently cross cellular membranes [9].
Subsequently, a dual mechanism of uptake into mammalian cells has been described [51]: CNTs were shown to use either an endocytic pathway or the passive diffusion to penetrate through cellular membranes (Figure 2). In the endocytic mechanism, CNTs are internalised inside vesicles, named endosomes, before being directed to the lysosomes localised in the perinuclear compartment [52]. The energy-dependent uptake of nanotubes was described as predominantly clathrin-dependent for both SWNT and MWNT [53,54]. Kang and colleagues showed that SWNT / doxorubicin complexes, conjugated via hydrophobic π stacking, were internalised using the endocytic pathway and accumulated into the perinuclear lysosomal compartment of endothelial progenitor cells [49]. Whilst SWNT remained entrapped into lysosomes, DOX detached in the acidic lysosomal environment due to the pH-dependent π-π stacking interaction, diffused into the cytoplasm and reached the nuclear compartment to induce cell killing.
In contrast, the passive diffusion of CNTs, also called needle-like penetration, results in the simple diffusion of CNTs through the cellular membrane without need of energy consumption [9,55]. Following computational and electron microscopy studies, the passive diffusion of ƒ-CNTs through the phospholipid bilayer membrane, has been broken down in three steps: i) landing and floating of the ƒ-CNTs on the membrane surface; ii) penetration of the lipid head groups; and iii) sliding through the lipid tails ( Figure 2) [56,57]. individualised CNTs (i) land on the surface of the plasma membrane, (ii) penetrate the lipid head groups and finally (iii) slide through the lipid tails to passively diffuse through the cell membrane. Both residual bundled MWNT in endosomes and free MWNT in the cytosol are recruited into lysosomes. CNTs can be excreted by exocytosis (not shown) or in autophagic microvesicles in case of cellular stress.
Another exit mechanism has been reported in polynuclear neutrophils and macrophages where nanotubes are digested enzymatically. CNTs are also able to enter organelles and the nucleus. The figure was redrawn and modified from [51].
A model explaining the differential mechanism of uptake between active and passive mechanisms has been proposed by Yan and co-workers [51]. Accordingly, CNT clusters would be internalised by an endocytic mechanism, whereas individualised nanotubes would enter the cell by membrane diffusion (Figure 2).
Interestingly, despite the significant effort dedicated to the characterisation of CNT internalisation pathways, there is no report suggesting that passive transport is preferred to endocytosis for drug delivery applications. However, one could indicate that both pathways support the versatile capabilities of CNTs to cross biological membranes.

Differential uptake between non-phagocytic and phagocytic cells
The use of CNTs for drug delivery entails their capture by macrophages, which participate in homeostasis and physiological defence mechanisms [58][59][60]. These cells are especially involved in the removal of external materials by phagocytosis, a cell endocytic pathway similar to endocytosis but involving the uptake of large particles (∼1 μm). Both endocytosis and phagocytosis are energy-dependent mechanisms that are impeded at low temperature [53]. Phagocytic cells take up CNTs mainly through phagocytosis but the blockade of this energy-dependent pathway still allows the uptake of CNTs by passive diffusion [61]. SWNT coated with phospholipid-polyethylene glycol (PL-PEG) were found to cross the cell CNTs accumulated in lysosomes following a phagocytic mechanism [62]. These results suggest that the internalisation mechanism of nanotubes was not only dependent on the properties of CNTs but also on the phagocytic nature of cells.
Within the same population of phagocytic cells (human monocyte-derived macrophages), a study intended to establish a differential uptake of fluorescently labelled nanotubes as function of their length [63]. Constructs above 400 nm in length were mainly localised in endocytic vesicular structures, while the fluorescent signal from shorter CNTs was more diffuse, supporting their extra-vesicular localisation in cell cytoplasm.
Overall, these results suggest that shortening of CNTs enhances their passive diffusion uptake mechanism, even in phagocytic cells.

Cellular fate of CNTs
Following their passage through cellular membranes, CNTs were reported to accumulate in various subcellular compartments, such as the cell cytosol [61], endosomes [54,64], the perinuclear region [65], mitochondria [62,66] or the nucleus [51], according to their physicochemical properties and functionalisation. Exocytosis and biodegradation of CNTs have also been reported as possible cell elimination mechanisms for this material [46,47,50]. The description of such outcomes is crucial to confirm the potential of CNT bio-elimination, ultimately lowering the risk of CNT toxicity.

Exocytosis
Taking advantage of the intrinsic CNT photoluminescence properties, SWNTs were tracked by Jin in real-time using single particle tracking, and showed similar endocytosis and exocytosis rate (Figure 2). [46,47]. Using Raman spectroscopy mapping, Neves and colleagues found that oxidised and RNA-wrapped doublewalled CNTs (DWNTs) accumulated in cells over 3 h before being progressively released out of the cells over a 24 h period [67]. It recently emerged that the process of exocytosis could also be induced under stress conditions. Naive human monocyte macrophages and endothelial cells exposed to stress were able to release microvesicles containing CNTs [68]. This mechanism could eliminate exogenous and toxic carbon material by inducing the formation of autophagic microvesicles [69].

Enzymatic degradation
The enzymatic-based degradation of CNTs was reported as a possible mechanism by which cells eliminate this material (Figure 2). Studies by Allen and collaborators provided evidence of the degradation of oxidised SWNTs through enzymatic catalysis in abiotic conditions [70,71]. Following the incubation of carboxylated SWNT with horseradish peroxidase in low H 2 O 2 concentrations (40-80 μM), a combination of techniques including transmission electron microscopy (TEM), Raman and ultraviolet-visible-near-infrared (UV-NIR) spectroscopy, and showed that digested CNTs displayed reduced absorbance, dramatic length shortening and disappearance of their discriminating G-and D-bands. It was later proposed that the presence of carboxylic group and defects at the surface of CNTs are a prerequisite to trigger the interaction with the oxidative agent and that nanotube degradation was function of the defect density [72]. Using similar techniques, it was demonstrated 16 that in vitro exposure of SWNTs to neutrophils followed by enzymatic digestion with myeloperoxidase (MPO) promoted alterations in CNT structure [73,74]. These findings were supported by in vivo studies using TEM, Raman spectroscopy and photoacoustic imaging, which showed degradation of SWNTs in lung tissue following pharyngeal administration in mice [48,75]. Using TEM and Raman spectroscopy, our group also found evidence of SWNT degradation in mouse brain after stereotactic injection [76]. In a recent study, the amount of CO 2 released from the enzymatic digestion of nanotubes in contact with horseradish peroxidase was measured, with a CNT degradation rate of ~0.002 % per day being reported, which corresponds to a half-life of ~80 years [77].
While studies confirm that CNTs can be enzymatically degraded, the notion of degradation is still broadly employed and future reports must provide qualitative and quantitative data regarding the formation of by-products induced by CNT enzymatic degradation, especially for in vivo investigations.

Biocompatibility of CNTs
The toxicity of CNTs has been widely reported and is of major concern for human use [78]. It is accepted that CNTs are heterogeneous material with certain physicochemical properties that can promote deleterious biological responses [11].
Therefore, the use of CNTs, applied to the drug delivery field, requires the design of materials with enhanced biocompatibility properties to ensure the safe translation of this material into clinical use.

Main properties of CNTs influencing toxicity
CNTs exhibit heterogeneous purity, length, type of functionalisation and surface interaction with plasma proteins that can affect their cellular toxicity. CNT toxicity mechanisms have been mostly explored by measuring cell viability, cell inflammation and the production of reactive oxygen species (ROS).
The concept of cellular toxicity can be described using the Hierarchical Oxidative Stress Model associating the cell toxicity mechanisms with the intracellular levels of ROS [79] (Figure 3). At low concentrations, ROS can be neutralised by antioxidants -e.g. glutathione (GSH), and detoxification enzymes. When the antioxidant defence is overwhelmed, further damages occur such as lipid peroxidation, change in cell morphology and genotoxicity [80,81]. Excessive ROS production initiates an inflammatory response through the release of cytokines and chemokines [80,82].
Finally, further ROS production induces the release of apoptotic factors leading to cell death [83]. The measurement of such end points relies more frequently on the quantification of cell metabolism, DNA content, membrane disruption or cellular apoptosis induction [28,84]. In vitro toxicity studies comparing physico-chemical properties of CNTs are summarised and classified in Table 1.   21 The main factors involved in CNT toxicity are reported hereafter:

Impurities
Catalyst remnants from the CNT synthesis, such as nickel (Ni), cobalt (Co), iron (Fe) and molybdenum (Mo), and amorphous carbon, localised at the surface of nanotubes or entrapped within the CNTs, can lead to oxidative stress, anti-oxidant depletion and a reduction in cell viability [86,101,102]. Several methods can be used to reduce the presence of impurities including high-temperature annealing, acidic treatment by reflux or steam-purification [101,103,104]. It has been demonstrated that CNTs free of catalyst metals and graphitic contaminants are unlikely to result in any inflammatory response or impairment of phagocytosis [105].

Dimensions
CNT length has been shown to greatly influence CNT toxicity. Extremely long CNTs (10-20 μm) displayed asbestos-like behaviour and long bioretention in peritoneal mesothelium [11]. When macrophages attempted to engulf long CNTs displaying larger dimension than the actual cell, it resulted in frustrated phagocytosis leading to formation of granuloma [106]. Such lengths are impractical for drug delivery and the shortening of CNTs has emerged as a logical requirement for biomedical applications. Following the same mechanism, MWNT of 2.4-10 μm length, coated with Pluronic ® F-127, exerted higher toxicity than shorter materials (0.4-1.4 μm) in a murine macrophage cell line, while shorter CNTs led to higher inflammatory response [93]. This opposite effect of CNT length on cell viability and inflammation suggested that the Hierarchical Oxidative Stress Model could not always be applied to characterise the toxicity of CNTs. Other studies reported disparities between the impact of CNTs on cell viability and inflammation/oxidative responses [97,101,107,108]. The effect of nanotube diameter on cellular toxicity was described only in few studies and contradictory results were reported, thus limiting any statement about the influence of such parameter [109,110].

Defects
Defects at the surface of nanotubes may be topological (e.g. ring shapes other than hexagon), sp 3 hybridised carbon atoms, incomplete bonding defects, doping with elements other than carbon, as well as various functionalities at the surface [111]. Studies by Fenoglio and Muller established that the presence of defects at the surface of CNTs triggered acute pulmonary toxicity and genotoxicity [84,112]. It was later reported that the shortening of CNTs by sample ultra-sonication using concentrated acid solution produced defects at the surface of the nanotubes, which were correlated with enhanced pro-inflammatory and pro-oxidative response [28].
Such report demonstrated that the shortening process could be associated with the formation of defects leading to inflammation and cytotoxicity.
Due to the heterogeneity of CNTs and the lack of data about CNT surface characterisation, the association between CNT toxicity and their physico-chemical properties can be puzzling. For example, Cheng and collaborators found that short MWNT (0.2 ± 0.1 μm) altered the development of zebrafish embryo, while long MWNT (0.8 ± 0.5 μm) showed reduced embryo-toxicity [113]. Although the authors concluded that nanotube length influenced the toxicity on embryos, the formation of defects at the surface of nanotubes induced by prolonged ultra-sonication in concentrated acid could just as well be responsible for the toxicity measured [113].

Functionalisation
Efficient dispersion resulting in individualised CNTs must be achieved for drug delivery applications in order to reduce the formation of CNT aggregates and increase their biodisponibility [19]. The type and degree of functionalisation tends to influence CNT toxicity. Surface functionalisation of nanotubes with carboxylic groups, using acid treatment, showed increased defect formation at the surface of CNTs adsorbing catalyst particles and generating free radicals [114]. Human neuroblastoma cells exposed to different concentration of oxidised MWNT showed dose-dependent decrease in cell viability [114]. In contrast, T-lymphocytes incubated with CNTs functionalised by 1,3-dipolar cyclo-additionforming aminated nanotubesdid not increase the apoptotic proportion of immune cells and preserved their inflammatory functionality by maintaining their interleukin secretion following the activation by lipopolysaccharides (LPS) [12].

Protein binding
Protein binding to the surface of nanotubes occurs in a physiological environment, such as the blood circulation system, but can also be used as a strategy for CNT dispersion. The compact and multi-layer form of bovine fibrinogen (BFG) proteins was shown to reduce the toxicity of CNTs proportionally to the degree of adsorption onto the surface of nanotubes [115]. In contrast, Dutta and colleagues showed that MWNT dispersed in serum bovine albumin induced pulmonary fibrosis [116]. By coating the same material with Pluronic ® F-108, the amphiphilic polymer was able to protect the lysosomal membrane from CNT damage and abolish the formation of pulmonary fibrosis [116].

A C C E P T E D M A N U S C R I P T
ACCEPTED MANUSCRIPT 24

Influence of cell model on CNT toxicity
Recent studies have highlighted that the toxicity of nanotubes is not only dependent on their physico-chemical properties but also on the cellular model used for toxicity assessment. In a study by Foldbjerg and colleagues, epithelial cancer cells (A549), human monocyte-derived macrophages (THP-1) and mouse macrophages (J774) were exposed to SWNT dispersed in BSA. Incubation at 10 µg/mL led to reduction in cell viability, necrosis, reduction of phagocytic ability, increased ROS levels and cytokine release in J774 cells, while A549 and THP-1 cells treated in the same conditions were not affected [117]. concentrations resulted in a significant increase in the production of reactive superoxide species [118].
The different response of epithelial and macrophage-like primary cells to CNTs highlights the need to select appropriate and relevant models to test in vitro the toxicity of CNTs. In our opinion, epithelial cells, macrophages and cells from the reticuloendothelial system should be included in any studies aimed at assessing toxicity of CNT-based nanocarriers.

Mechanisms of cellular toxicity triggered by CNTs
The mechanisms of cellular toxicity induced by nanotubes have been essentially described using pristine CNTs and were mostly related to the production of oxygen radicals [50] (Figure 4). The induction of oxidative stress activates signalling pathways mediating inflammation and, ultimately, to cellular toxicity. Activation of NF-κB or AP-1

A C C E P T E D M A N U S C R I P T
ACCEPTED MANUSCRIPT 26 transcription factors by CNTs has been directly involved in upregulating genes involved for the release of cytokines such as IL-1, IL-6 and TNF-α [50,119]. CNTrelated oxidative stress could result in oxidation of mitochondrial phospholipids and nicotinamide adenine dinucleotide phosphate (NADPH) oxidation leading to inflammation and apoptosis, but also positive outcomes such as CNT biodegradation in neutrophils [71,120].
Several non-oxidative stress-dependent effects were found related to the cellular accumulation of CNTs, such as the blocking of ion channels leading to loss of enzyme function, the interference with the cytoskeleton impacting proliferation, migration and phagocytosis, and the potential induction of a tumourigenic response [121][122][123]. It was reported that SWNT and DWNT could be responsible for activating the classical and alternate pathways of the complement system [124]. This was further associated with increased cellular infiltration and phagocytosis, as well as reduced pro-inflammatory cytokine secretion, thus supporting the beneficial effect of complement activation triggered by CNTs [125].

In vivo toxicity studies in non-brain tissues after intravenous administration of CNTs
A majority of pre-clinical toxicological studies assessed the toxicity of CNTs after pulmonary inhalation/exposure [10,106]. However, drug delivery mediated by CNT carriers requires the study of their toxicity following intravenous (i.v.) administration. In most reports, this evaluation was done using histological analysis [58,126,127], blood cell counting [128] and inflammation detection [126,127] in organs and blood. A summary of the studies reporting the toxicity of ƒ-CNTs in murine models is presented in Table 2.

Table 2: Representative in vivo studies of ƒ-MWNT and ƒ-SWNT injected IV in murine models
List of abbreviations in Table 1 29 Dispersion was shown to be a key factor influencing CNT toxicological effect.

ACCEPTED MANUSCRIPT
Indeed, purified pristine MWNT injected intravenously accumulated mainly in the lungs as large aggregates, causing short-term respiratory distress [58]. The intravenous injection of pristine MWNT was also shown to form large CNT aggregates in liver and lungs and induce inflammatory cell infiltration around the airways and blood vessels in lung tissue [127]. Nevertheless, the results available to date suggest that i.v. injection of properly functionalized CNTs is well-tolerated and their use as carriers for therapeutic and imaging applications can be justified.

CNTs as carriers for therapeutic brain delivery
The delivery of therapeutic molecules to the brain is severely restricted by the presence of the blood-brain barrier (BBB), a complex network comprising brain endothelial cells, astrocytes and other support cells, that control the influx and efflux of nutrients and other molecules to the brain parenchyma. For this reason, treatment of complex neurodegenerative disorders and brain gliomas remains a challenge.
When particle size is not small enough to overcome the BBB size restriction (i.e. <100 nm), nanoparticle-based brain-targeted therapeutic approaches have, thus far, taken advantage of the existing physiological mechanisms of transport to improve brain delivery. Polymer-and lipid-based nanoparticles are usually decorated with targeting moieties to support receptor-mediated transcytosis across the BBB. The capacity of CNTs to cross the BBB by both receptor-or adsorptive-mediated energydependent pathways (transcytosis) and passive energy-independent mechanisms (needle-like crossing) constitutes a major advantage compared to other nanocarriers.
In addition to the intrinsic ability to cross biological membranes, CNTs possess high surface area, which enables the loading and delivery of high doses of drugs to the therapeutic site, as well as intrinsic optical and thermal properties, with potential multimodal real-time tracking and photo-thermal applications. Although CNTs display optimal characteristics for use as nanocarriers in brain delivery, the intrinsic barrier-crossing capacity constitutes a limitation, due to unspecific bioaccumulation, and therefore specific brain target is important to increase brain accumulation and reduce systemic side effects.

Intracranial administration of CNT-based therapeutics
Local administration of therapeutics could bypass the BBB restriction and allow effective therapeutic doses to be achieved in the diseased tissues. This approach could be relevant in certain brain disorders, such as tumours and stroke, in which initial surgical approach is often necessary. Scheme 1 highlights the main strategies involving local CNT-mediated therapeutic brain delivery.

Scheme 1. Different stereotactically-delivered CNT-based strategies for brain therapy.
The conjugation of immunostimulatory CpG oligonucleotides to SWNTs was shown by Zhao and collaborators to be beneficial for the treatment of glioma-bearing mice [137]. Increased uptake of SWNT-CpG by tumour-associated inflammatory cells was found after a single intracranial injection (compared to the free CpG),

A C C E P T E D M A N U S C R I P T
ACCEPTED MANUSCRIPT 32 which resulted in a strong anti-tumoural response, decreased tumour size and an increase in median animal survival time [137].
Taking advantage of the capacity of CNTs to absorb NIR radiation and convert it to heat, Santos and colleagues used CNTs for photothermal-mediated brain tumour therapy. The authors showed that a combination of intratumoural SWNTs injection and NIR exposure in athymic GBM-bearing mice not only suppressed tumour growth compared to SWNTs or NIR per se, but also inhibited tumour recurrence for up to 80 days [138]. The electrical properties of CNTs prompted several research groups to use these carriers as scaffold for stem-cell mediated neuronal repair.
Studies by Moon and colleagues in rats with stroke-induced brain injury revealed that focal injection of neural progenitor cell (NPC)-impregnated CNTs improved rat behaviour and reduced infarct cyst volume and area [139]. Similarly, pre-injection of amine-functionalised SWNTs onto the right lateral ventricle was shown to reduce the infarction area and improve behavioural functions following focal ischemic injury in rats [140]. The authors showed that the neuroprotective effect was achieved by reducing apoptosis, inflammation and glia activation and proposed that, even without NPCs, the high surface energy of the positively-charged SWNTs provided a favourable environment for neuronal regeneration. In a very recent study by Xue and colleagues, intraventricularly-injected aggregated SWNTs (aSWNTs) were shown to reduce methamphetamine (METH) addiction symptoms in mice by causing oxidation of METH-enhanced extracellular dopamine, which induced inhibition of the rewarding and psychomotor-stimulating effects of METH [141].
A previous study from our group revealed that the local f-MWNT-mediated delivery of siRNAs targeting caspase-3 (siCas3) successfully prevented neuronal death in an endothelin-1 (ET-1) rat stroke model (Figure 5) [142]. This was supported by observation of a decrease in apoptosis in the penumbra lesion ( Figure   5A-C) and a statistically significant improvement in the "skilled reaching" behaviour test ( Figure 5D) for f-CNT:siCas3 treated rats, compared to animals receiving complexes of non-specific siRNA (siNEG) and f-CNT (f-CNT:siNEG) or siCas3 alone.  -1 injection. (G) The f-CNT: siCaspase3 group showed the least apoptosis quantitatively indicating effective and specific siRNA delivery in vivo compared to siRNA alone. (H) Behavioural analysis of rats after stroke induction in the all the treated groups using the skilled reaching test (adapted from [142]).
While the intracranial delivery of drugs could be of clinical utility, its wide application is nevertheless limited by the invasiveness of the technique and patient compliance. The versatility of nanotubes and their unique mechanism of interaction with cells paved the way for researchers to explore the use of CNTs as carriers for systemic drug delivery to the brain.

Systemic administration of CNT-based therapeutics
The ability of CNTs to penetrate biological membranes without perturbing the membrane integrity prompted researchers to test CNTs as vehicles for systemic therapeutic delivery across the BBB. Early evidence of efficient BBB translocation by CNTs was provided by studies in primary brain endothelial cells [143][144][145] and other commonly used in vitro BBB models [146].

In vitro BBB translocation
The interaction of f-MWNTs with the BBB has been previously investigated by our group using a BBB co-culture model comprised of primary porcine brain endothelial cells (PBEC) and primary rat astrocytes. This model replicates the physiological and biochemical features of the human BBB, including high trans-endothelial electric resistance (TEER), expression of membrane transporters and tight junction proteins [143]. TEM analysis revealed that f-MWNTs were quickly internalized (within 4h) by PBEC cells via endocytosis, with f-MWNTs being released from endocytic vesicles near the abluminal side of PBEC within 24-48 h (Figure 6) [144].
Moreover, Scanning TEM (STEM) showed that this process did not cause any damage to the cell membrane (Figure 6). Since no involvement of the tight junctions was detected during translocation, it was suggested that f-MWNTs use a transcellular route to cross the BBB.
In a subsequent study, the effect of CNT diameter on BBB translocation was investigated. Gamma counting was used to quantify the rate of translocation of "wide" (~35.9 nm diameter) (w-MWNTs) or "thin" (~9.2 nm) (t-MWNTs) f-MWNTs across the PBEC co-culture model over a 72 h period. In general, higher percentage transport across PBEC was achieved for w-MWNTs compared with t-MWNTs (~15.6% and 7.6% of total dose after 72 h, respectively) [145]. Targeting of f-MWNTs was also tested in this study, in order to assess whether BBB translocation could be further improved. For this purpose, CNT synthesis was modified to incorporate a targeting peptide, angiopep-1 (ANG), to the carrier surface. This small peptide binds LRP1, a lipoprotein receptor that is overexpressed on brain endothelial cells of the BBB and several human tumours [147]. Indeed, higher values were obtained for ANG-functionalized w-MWNT and t-MWNTs (~20.3% and 11.6%, respectively) compared to the non-targeted carrier (~15.6% and 7.6%) [145], using the in vitro PBEC co-culture model. This confirmed that conjugation of this small LRP1-targeting peptide to f-CNT, enhances BBB translocation and therefore could be of beneficial use for brain-targeted therapies.
The capacity of CNTs to cross the BBB in vitro without compromising this barrier was also demonstrated by Shityakov and colleagues. The authors used phasecontrast and fluorescence microscopy in combination with molecular dynamics simulation to demonstrate that amine-functionalized FITC-labelled MWNTs (MWCNT-NH 3 + -FITC) were able to penetrate murine microvascular cerebral endothelial (cEND) monolayers layers over a 48 h period, without compromising cell integrity [146].

In vivo BBB crossing
In addition to several in vitro studies, pre-clinical studies involving i.v. injection of CNTs provided unequivocal evidence of the capacity of CNTs to reach the tissues beyond the BBB.
In one of our very first studies, aiming at determining the impact of f-MWNT diameter on systemic organ biodistribution, it was revealed that radiolabelled t-MWNT conjugated to humanised IgG or fragment antigen binding region (Fab′) showed higher tissue affinity (including higher brain affinity) compared to w-MWNTs [148]. In a subsequent study, designed to investigate in detail the brain uptake, accumulation and elimination of t-and w-MWNTs (with or without ANG conjugation), following systemic administration in healthy mice [145], higher brain uptake was also found for t-MWNT (~2.6% ID/g tissue at 5 min) compared to w-MWNT (~1.1%) (Figure 7). Importantly, ANG brain targeting effect was significant for w-MWNT-ANG (~2.0% vs 1.1.%) but not for t-MWNT-ANG (~3.0% vs 2.6%).
The capillary depletion method (which removes the vascular fraction of the brain), was then used to evaluate the f-MWNT content in the parenchyma and vascular brain fractions. Greater parenchyma uptake/accumulation was found for t-MWNTs suggesting elimination from parenchyma.
Since LRP1 overexpression has been described not only on brain endothelial cells but also for malignant brain tumours [147], experiments were also performed to investigate the capacity of i.v. injected ANG-coupled w-MWNTs, which show higher brain parenchyma retention, to target this type of tumour. Interestingly, w-MWNT-ANG not only showed significantly higher uptake in glioma when compared to normal brain, but also enhanced accumulation in glioma compared to the passively targeted w-MWNT [145], which suggests that ANG-coupling can provide a double-targeting effect, with improved brain and tumour uptake.
Overall, although in vitro data suggested that wider f-MWNTs are more efficient in crossing the BBB, the available in vivo data suggests that uptake in healthy brain tissues after systemic injection is favoured by f-MWNTs with smaller diameter, while wider MWNT exhibits better brain retention. ANG conjugation enhances brain uptake of wider MWNTs but offers no advantage to brain uptake of thinner ones.
ANG-modified f-MWNT, of wider diameter, seems to be the most suitable candidate among the ones studied, for BBB and brain tumour double-targeting. Since the BBBcrossing studies were performed with "empty" CNTs, nevertheless, it is not known whether the % ID achieved is sufficient to generate a disease-specific response. . After perfusion with heparinecontaining saline, brain accumulation was quantified by gamma scintigraphy, which was followed by capillary depletion at 5 min, 30 min, 1 h, 4 h and 24 h. The overall brain uptake of nontargeted and targeted (ANG-conjugated) (A) t-MWNTs or (B) w-MWNTs, expressed as % injected dose per gram of brain (%ID/g), revealed higher uptake for t-MWNTs. The radioactivity of (D) t-MWNTs or (D) w-MWNTs in brain parenchyma and capillaries, measured after capillary depletion, was used to calculate the brain parenchyma/blood ratios. * p < 0.05; ** p < 0.01. Adapted from [145].

In vivo brain distribution
Following the encouraging results where f-MWNT achieved reasonably high brain parenchyma accumulation when delivered systemically, we used multi-modal

A C C E P T E D M A N U S C R I P T
ACCEPTED MANUSCRIPT 40 imaging techniques to study in more detail the kinetics and spatial distribution of i.v.
injected f-MWNTs in the brain [149]. SPECT/CT imaging clearly showed accumulation of radiolabelled t-MWNTs over the entire brain at the early time points after injection (5 min and 30 min), with higher radioactivity being detected in the mid-brain region ( Figure 8A). Autoradiography of brain slices (2 mm thick), which allows greater spatial resolution via increased exposure times and support of semiquantitative analysis, provided further evidence of the preferential accumulation in the mid-brain region. Higher intensity was detected in sections of mid-brain (sections c3-5), while relatively lower radioactivity was detected in brain cortex ( Figure 8B).
Importantly, gamma counting and TEM ( Figure 9A) showed preferential location of t-MWNTs in the brain capillaries up to 24 h after injection, which indicates that the MWNTs enter the brain via its endothelium. Moreover, TEM images showed that brain capillaries remained intact and circular, thus indicating that t-MWNT uptake into the brain was not caused by inflammation or the damage to the BBB.
Taking advantage of its inherent optical properties, t-MWNTs were also directly imaged in brain sections and whole brain using state-of-the-art techniques, namely Raman and multi-photon luminescence microscopy ( Figure 9B). The typical D (1309 cm − 1 ) and G (1599 cm − 1 ) MWNT Raman bands were seen in the spectra of capillaries (as well as in control bulk material), while no bands could be detected in whole brain sections. It is worth noting that t-MWNT accumulation in capillaries was ~ 4-fold higher than brain parenchyma. This highlights the limitation of this technique in terms of poor sensitivity.

In vivo brain-targeted therapy
While the above-mentioned studies investigated the BBB interaction, translocation and brain distribution of f-CNTs in vivo, they involved "empty" A C C E P T E D M A N U S C R I P T ACCEPTED MANUSCRIPT 43 carriers. To date, very few studies investigated the delivery of CNT-formulated drugs to the brain following systemic administration.
Yang and colleagues used SWNTs for brain delivery of acetylcholine (ACh) to Alzheimer disease (AD)-bearing mice. In this regard, the administration of SWNTs-Ach, via oral gavage, resulted in improved learning and memory capabilities 8h after treatment. Contrasting results were obtained in animals treated with empty SWNTs or the free Ach, in which no significant improvement was observed [150]. This study also highlighted the importance of the correct dosing of CNTs in brain-targeted therapies, as the mitochondria were shown to be affected by high concentrations of SWNTs. In a study by Ren and collaborators, PEGylated and carboxylated MWNTs (oxMWNT-PEG) modified with ANG were used as carrier to deliver doxorubicin to mouse brain. In addition to increased brain uptake of Dox-oxMWNT-PEG-ANG compared to Dox-oxMWNT-PEG or Dox per se after 1 to 6 h, the authors observed a substantial increase in survival of glioma-bearing animals treated with Dox-oxMWNT-PEG-ANG, compared to the saline-treated group [151]. Since fluorescence non-quantitative imaging was used to evaluate the doxorubicin dispersion in the brain, the accumulation rate of Dox-oxMWNT-PEG-ANG could not be compared to other delivery systems. Intravenously-injected NIR-fluorescent PEG-conjugated SWNT sensors, developed by Iverson and colleagues, showed selective detection of local nitric oxide concentration in the brain, with a detection limit of 1 µM, with potential applications in sensing and therapy [152].

In vivo toxicity of CNTs in brain tissues
Cell and macromolecule traffic into the brain is strictly controlled by a tight BBB, efflux transporters and the presence of immune cells (microglia), to ensure a proper microenvironment in which neurons can function. Exposure to CNTs could disturb this fragile balance and result in cytoxicity. No studies have yet assessed brain toxicity following systemic administration of CNTs. The evidence available, obtained from studies involving intracranial injection of nanotubes, suggests that surface functionalisation contributes to the cytotoxic outcome.
In a study by Bardi and colleagues, neuron toxicity was not detected in mouse brain cortex after stereotactic micro-injection of Pluronic ® F-127 coated-MWNTs (MWNT:F127). Histological examination revealed small neuronal damaged near the injection site (due to the mechanical penetration of the brain) but no further damage was observed after 18 days [153]. In a previous manuscript from our group, it was shown that cortical stereotactic injection in mouse brain of MWNTs functionalised by 1,3-dipolar cycloaddition (MWNT-NH 3 + ) resulted in lower inflammation compared to the oxidised carrier (ox-MWNTs-NH 3 + ) [154]. Similar uptake levels of the compounds were found in astrocytes, microglia and neurons, but a significantly higher release of the pro-inflammatory cytokines TNF-α and IL-1β, as well as microglia activation, was detected after injection of ox-MWNTs-NH 3 + compared to MWNTs-NH 3 + . A recent follow-up study revealed that microglia is involved in the cytotoxic response generated by ox-MWNTs in brain tissue. Exposure of ox-MWNTs to neuronal cultures did not lead to detectable cytotoxicity (measured by the modified LDH assay), whereas significant cytotoxicity was found in mixed glial cultures isolated from striatum, but not from the frontal cortex. Striatum-derived mixed cultures contain a higher density of microglia, compared to the higher amount of astrocytes in the frontal cortex, thus suggesting that microglia is responsible for the CNTs-induced cytotoxicity in brain tissue [155].

Biodegradation of CNTs in brain tissues
The biodegradation of CNTs by brain cells has been studied in vitro and in vivo.
Strong evidence suggests that CNTs can be degraded within human brain tissue by human MPO and hydrogen peroxide (H 2 O 2 ).
Using Raman spectroscopy, Kagan [74]. In a follow-up study, the biodegradation of SWNTs-PEG was proposed to occur in a two-step process, with the initial stripping of the PEG coating, mediated by secreted proteases, followed by SWNT degradation via surface defects [156]. Interestingly, PEG removal from the surface of SWNTs-PEG after systemic administration in mice has been previously described, but no explanation was provided for the mechanism [157].
In vivo data provides further evidence of the role of MPO on the CNT degradation in brain cells. Increased inflammation and lower SWNTs clearance rate were detected in MPO mouse knockouts following pharyngeal aspiration, compared to wild type animals [75]. An elegant study by Nunes and colleagues, which used electron microscopy to assess degradation of stereotactically-administered f-MWNTs A C C E P T E D M A N U S C R I P T ACCEPTED MANUSCRIPT 47 in mouse brain cortex, detected loss of cylindrical structure of f-MWNTs following internalisation into microglia [76]. The oxidative environment of the microglia, which contains MPO, peroxidases and other degradative enzymes, could be involved in the observed degradation.
Overall, the high surface area of CNTs, intrinsic capacity to cross the BBB, controlled toxicity and degradability in the presence of MPO enzymes are encouraging properties/observations for future applications of CNTs in the brain.

Concluding remarks and future perspectives
The progress achieved on the synthesis, design and functionalization of carbon CNT-mediated therapeutic delivery to the brain has shown that biocompatible CNTs A C C E P T E D M A N U S C R I P T ACCEPTED MANUSCRIPT 48 can efficiently reach the brain, little information is available about the amount of therapeutic molecule needed to obtain a significant therapeutic benefit. Future work should therefore focus on determining effective therapeutic disease-specific dosage, improving brain delivery and clarifying the fate of CNTs with the different CNS tissues, in healthy and diseased brain. This information would be extremely helpful for the design/production of improved f-CNTs-based delivery systems that can, in the future, constitute primary options for efficient targeted brain therapy.