30 years of advances in functionalization of carbon nanomaterials for biomedical applications: a practical review

Carbon-based nanomaterials (CANOMATs), including fullerenes, carbon nanotubes, graphene, and their derivatives, are widely considered to be the next-generation materials for a broad range of biomedical applications, owing to their unique opto-electronic, chemical, and mechanical properties. However, for bio-applications, CANOMATs need to be surface-functionalized, to render them passive, non-toxic, and water-soluble. Here, we review the current state-of-the-art in the methods of functionalization of CANOMATs. In contrast to other Reviews, we present an objective analysis of the various approaches reported in the literature, using metrics such as the agent of functionalization, number of steps, and time required, the need for special instruments, effect on properties, scalability, reproducibility, and applications. Our Review offers a way for researchers to make a rational selection of the process of functionalization to best suit their desired application. This opens up new opportunities for developing targeted functionalization strategies, based on the need to excel at the above metrics.


I. INTRODUCTION
In the past 30 years, there has been explosive growth in interest in the use of carbon-based nanomaterials, ever since the introduction of 0D (C 60 buckminsterfullerene), 1 1D (carbon nanotubes), 2 and 2D (graphene) 3 forms (Fig. 1) of CArbon-based NanO MATerials (CANOMATs), in that chronological order, where D refers to the dimensionality. These low-D CANOMATs are exceptionally advantageous for biomedical applications, because of their unique optoelectronic, chemical, mechanical, and physical properties and their large surface area, [4][5][6][7] coupled with dimensions that are comparable to those of biomolecules such as DNA or proteins. 8 For a brief overview of the properties of pristine CANOMATs, the reader is referred to the Supplementary Material. What makes these materials challenging to use is that the protocols for synthesis of carbon nanomaterials result in large dispersion in the purity, yield, and consequently the physico-chemical properties of these materials. Since physiological responses in biological conditions are highly sensitive to uniformity and reproducibility of the material properties, some purification and surface functionalization steps are necessary. Further, to achieve biocompatibility and water solubility, CANOMATs need to be surface-functionalized.
While there have been many reviews [9][10][11][12][13][14][15][16][17][18][19][20][21][22][23] over the years on the use of CANOMATs for biological applications (focusing on aspects such as biological sensing, imaging, drug delivery, and therapy), to the best of our knowledge, there have been no comprehensive reviews comparing the various strategies for the functionalization of these materials. This is an especially important consideration for early-career scholars and researchers in the materials science community. Although there is great variety [24][25][26][27][28][29][30][31] in the availability of functionalization schemes for CANOMATs, young investigators are often faced with the challenging optimization problem of deciding on a technique which is suitable for their particular application, while satisfying the following constraints: (i) agent of functionalization; (ii) number of steps and time required to develop; (iii) the availability of necessary instruments; (iv) the desirability of material properties affected by the functionalization; (v) the scalability of the approach; (vi) reproducibility (both by the originators of the technique, and by subsequent researchers); and (vii) the application space which has already been explored. The details about the parameters used for this Review are listed in the Supplementary Material. This Review aims to aggregate the various approaches used for functionalization of carbon-based nanomaterials, including chemical and biological functionalization approaches. The parameters used in this Review are discussed in the Supplementary Material.
We discuss examples of the Belcher group's work [Figs. 4(a)-4(c), 7(a), 7(b) and 8(g)] on the use of a biological template (M13 bacteriophage), as a scaffold for noncovalent functionalization of carbon nanotubes for detection of cancers 32,33 and infectious diseases, 34 and also the use of physical or chemical modification to functionalize graphene oxide (GO) for biosensing applications 35 involving cell capture from whole blood; using the same metrics described above. The goal of this Review is to provide an objective, quantitative (wherever possible) comparison of these parameters, to help researchers make an informed decision while choosing a functionalization technique for their desired application.

II. FUNCTIONALIZATION OF FULLERENES
A summary of the various techniques of functionalization used for fullerenes is presented in Table 1 in the Supplementary Material.
A. Non-covalent functionalization of fullerenes Fullerenes are notoriously hard to solubilize in most solvents 36 (let alone aqueous dispersion). Variations in temperature and pressure are required to solubilize fullerenes. The solubility of C 60 shows a maximum in solvents such as CS 2 , toluene, and hexane at ;280 K. Given these challenges, it would appear reasonable that most approaches to functionalizing fullerenes have relied on chemical modification through covalent bonding. Nevertheless, there are a few reports of non-covalent functionalization on fullerene molecules.
Manna, A.K. et al. 37 have performed first-principle density functional theory (DFT) calculations on the effect of non-covalent interactions of fullerenes (C 60 , C 70 , C 80 and B 80 ) with single-layer graphene. Their calculations suggest that larger fullerenes result in greater binding strength, facilitated by van der Waals interactions between the adsorbed fullerene and graphene [ Fig. 2(a)]. All graphene-fullerene composites displayed perfectly metallic behavior. Besides graphene, several groups of researchers have attempted to form stable polymer composites with fullerenes through non-covalent interactions. For example, Jung, S. et al. 38 studied the interaction between protonated porphyrin and fullerenes (C 60 and C 70 ) in the gas phase, synthesized by electrospray ionization of the porphyrin-fullerene mixture. C 70 was shown to bind more strongly than C 60 , in agreement with the DFT results discussed above. It was rationalized that the protonated porphyrin-fullerene complex is stabilized by p-p interactions. In another manifestation of electron donor-acceptor interactions, Teh, S-L. et al. 39 have demonstrated the dispersion of C 60 in a polymer matrix, Fig. 2(c). Subsequently, Li, F. et al. 40 developed polymer/fullerene nanofibers through cooperative noncovalent interaction, by reacting a diblock copolymer derivative of poly(3-hexylthiophene) (P3HT) and phenyl-C 61 -butyric acid methyl ester (PCBM), with improved performance in organic photovoltaic (OPV) devices. Recently, Reddy, B.K. et al. 41 studied the crystal structure of fullerenes complexed with antiaromatic isophlorin molecules (a tetra-pyrrolic macrocyclic compound, with a structure similar to porphyrin). They concluded that the complex is stabilized owing to the favored van der Waals interaction of the curved p-surface of the C 60 molecule with the nearly planar surface of the anti-aromatic p-surface of the isophlorin molecules [ Fig. 2(b)]. Basiuk, V.A. et al. 42 reported that upon calculating the binding energies of 20 L-amino acids with C 60 , they found poor correlation between the binding energies and the scale of FIG. 1. The beauty of carbon. Schematic showing the hexagonal lattice structure of the most common forms of CANOMATs: a semi-infinite sheet of graphene (2D), a single-walled carbon nanotube of (6, 5) chirality (1D), and a C 60 bucky ball (0D). The red spheres represent C atoms. Rendered using VESTA. 155 hydrophobicity proposed by other research groups. More work needs to be done to understand the mechanism (or possibly a combination of mechanisms) of interaction of the bucky ball with adsorbed molecules.

B. Covalent functionalization of fullerenes
Among the first of the CANOMATs to be discovered, fullerene chemistry has been thoroughly investigated and studied. 43 39 Chaudhuri et al., 46 Mikawa et al., 56 Zhang et al., 61 © 2011, 2009, 2001, 2010  schemes subsequently applied to carbon nanotubes and graphene trace their roots back to the methods used for functionalization of fullerenes.

Oxidation and hydroxylation
Fullerenes can also be oxidized with reagents such as OsO 4 , ozone, oxone monopersulfate, and 3-chloroperoxy benzoic acid, to name a few. In fact, the reaction of OsO 4 causes an adduct, C 60 (OsO 4 ) (4-tert-butylpyridine) 2 , which breaks the cage structure, and can be detected in x-ray crystallography, which was one of the original confirmations 44 used in the determination of the buckyball framework. Similarly, poly-hydroxylated derivatives, called "fullerenols" or "fullerols", C 60 (OH)  have been synthesized using fullerene reactions with dilute H 2 SO 4 and KNO 3 , or with dilute NaOH in the presence of H 2 O 2 or other peroxides.
In terms of applications, functionalized fullerenes have been studied extensively for their potential use in drug delivery, photodynamic therapy (reactive oxygen species generation), and in targeted imaging. For example, fullerenol, C 60 (OH) 24 has been shown to protect liver and heart tissue, from the chronic toxicity effects of administering doxorubicin, a chemotherapy agent, in male Wistar rats being treated for colorectal cancer, by Injac, R. et al. 45 Chaudhuri, P. et al. 46 have shown that fullerenol complexed with doxorubicin can suppress the proliferation of cancer [ Fig. 2(d)] by a G2-M cell cycle block, resulting in apoptosis, with in vivo efficacy comparable to that of the free drug, without the systemic toxicity side effects. Partha, R. et al. 47 have formulated "buckysomes", analogous to liposomes used for the delivery of hydrophobic drugs and other molecules. Amphiphilic fullerene structures embedded with the hydrophobic anticancer drug, paclitaxel (PTX), resulted in in vitro therapeutic efficacy (toward MCF-7 breast cancer cells) comparable to that of a commercial formulation of the drug. Similarly, Liu, J-H. et al. 48 have characterized a fullerene-DOX conjugate to be stoichiometrically well-defined, and studied its uptake in MCF-7 cells. They determined that the lipophilic fullerene-DOX conjugate accumulates predominantly in the cytoplasm (unlike free DOX, which proliferates into the nucleus), and may be contributing to cell toxicity through other mechanisms (structural damage to mitochondria, non protein-associated DNA strand break, or the generation of cytotoxic free radical species). Several groups of researchers have attempted to use functionalized fullerenes as gene delivery vehicles. For example, Nakamura, E. et al. 49 synthesized two-handed tetramineconjugated fullerene, which binds to duplex DNA in a reversible manner, which is internalized into cells by phagocytosis. The efficiency of transfection of a GFP-expression vector [ Fig. 2(e)] was determined to be comparable to that of commercial reagents. Building on this approach based on tetramino fullerene, Maeda-Mamiya, R. et al. 50 developed an in vivo gene delivery system to deliver the insulin-2 gene, with consequent reduction in blood glucose levels in a mouse model.
Photodynamic therapy is another active area of research where fullerene derivatives play an important role. For example, Tokayuma, H. et al. 51 showed that fullerene carboxylic acid, a water-miscible derivative, is biochemically active under low-energy light. As a photosensitive molecule, it was shown to have enhanced cytotoxicity in vitro against HeLa S3 cancer cells, as well as an ability to cleave DNA selectively at the G-sites, in a 182 base-pair fragment. In another approach, Liu, J. et al. 52 have fabricated a photosensitizer with magnetic resonance activity, to enable photodynamic therapy of tumors, by chelating Gd 31 to PEG-functionalized C 60 . The PDT efficacy of this construct was compared to that of Magnevist, a commercial MRI formulation of gadopentetic acid, in a murine tumor model, and was simultaneously shown to maintain an enhanced MRI signal for a longer duration in tumor tissue. A review of the work done in the area of PDT using fullerene chemistries [ Fig. 2(f)] is done by Mroz, P. et al. 53 One of the main challenges that need to be overcome (in addition to obvious concerns of cytotoxicity, which are discussed in the Supplementary Material) is the absorption maxima of fullerenes being in the UV-blue part of the wavelength spectrum. This limits their use to shallow or sub-surface diseases (most commercial formulations of photosensitizers are activated in the red-IR wavelengths), 54 since tissue penetration depth is limited in these wavelength ranges due to very high attenuation 55 caused by a combination of autofluorescence, absorption, and scattering mechanisms.
Fullerenes can also be designed to act as MRI contrast agents, in the form of endohedral fullerenes encapsulating the Gd 31 ion inside the fullerene cage (also known as "gadofullerenes"), as pH-responsive imaging probes. The first demonstration of metallofullerenes as in vivo contrast agents was by Mikawa, M. et al., 56 who synthesized water-soluble, polyhydroxyl forms, Gd@C 82 (OH) 40 [ Fig. 2(g)]. Upon i.v. administration into mice at a typical dose of ;100 lmol Gd/kg, they studied the biodistribution at 24 h, which indicated entrapment of the Gd 31 complex in the reticuloendothelial system (RES) organs, i.e., the lungs, liver, and spleen. This has the benefit of ;20-100% signal enhancement in these organs in the T 1 -weighted mode of MRI, at 30 min post i.v. injection. In subsequent development, Bolskar, R.D. et al. 57 reported the first water-soluble contrast agent, Gd@C 60 [C(COOH) 2 ] 10 with non-localizing behavior to the RES organs. The carboxylated metallofullerene showed less aggregation compared to the fullerenols, while providing relaxivity similar to commercially available Gd 31 -based contrast agents. A comparative study of hydroxyl-and carboxylfunctionalized fullerenes done by Sitharaman, B. et al. 58 revealed pH-dependent agglomeration, with sizes ;30-90 nm at pH 9, to hydrodynamic diameters as large as 600-1200 nm at pH 5. In another approach, it was shown possible to encapsulate Gd 31 ions endohedrally by using the "trimetallic nitride template" (TNT) process proposed by Stevenson, S. et al. 59 Building on this TNT-fullerene approach, Fatouros, P.P. et al., 60 and Zhang, J. et al. 61 have shown PEG-functionalized Gd 3 N@C 80 [DiPEG5000(OH) x ] to be effective proton relaxation agents for in vivo tumor imaging in the rat brain [ Fig. 2(h)]. As a brief footnote, fullerene derivatives have also been explored as contrast agents for X-ray imaging, such as the 2,4,6-triiodinated benzene ring substructure conjugated on to the C 60 structure by Wharton, T. and Wilson, L.J. 62

III. FUNCTIONALIZATION OF CARBON NANOTUBES
A summary of the various techniques of functionalization used for carbon nanotubes is presented in Table 2 in the Supplementary Material.

A. Non-covalent functionalization of carbon nanotubes
The non-covalent functionalization process has the advantage of improving the solubility of carbon nanotubes significantly, while avoiding modification of the sidewall composition. Therefore, the electronic structure of the carbon nanotubes is not perturbed to a great extent, which preserves the opto-electronic properties of the material, and makes it useful for applications such as sensing and imaging.

Small molecule surfactants
This is the most common class of surfactants, or stabilizing agents, used to disperse carbon nanotubes in an aqueous medium. Small-molecule surfactants, such as sodium dodecyl sulfate (SDS) or sodium cholate (SC) are amphiphilic in nature, comprising of a hydrophobic core, and a long hydrophilic tail. When the surfactant molecules are present in the dispersing medium of the CNTs at a concentration higher than the "critical micelle concentration" (CMC), they spontaneously self-assemble into micellar structures, with the hydrophobic core attached to the hydrophobic surface of the carbon nanotubes, and the hydrophilic ends sticking outward, into the aqueous medium, thereby making the wrapped CNTs form a stable suspension in water. Upon sonication, the aggregated bundles of CNTs break apart, which can then further be wrapped with the surfactant molecules, or are eliminated from the suspension during subsequent centrifugation.
The first reported use of small molecule surfactants [ Fig. 3(a)] was by O'Connell, M.J. et al., 63 who used 1% SDS to disperse single-walled carbon nanotubes (SWNTs) in H 2 O. Using this method, they were the first to report bright fluorescence from aqueous-dispersed SWNTs in the second near-infrared window (NIR-II: 900-1600 nm), which makes SWNTs ideally suited for in vivo biomedical imaging applications. Upon measurement of the photoluminescence spectra, they deduced the lifetime to be ,2 ns, and a quantum yield ;10 À3 , which conclusively suggested a fluorescence emission mechanism from singlet excitons. Following this exciting development, there has been a wide range of other small molecule surfactants reported to have been used for the suspension of carbon nanotubes in hydrophilic environments. Along with SDS, dodecyl-benzene sodium sulfonate (NaDDBS) is also commonly used, due to the high dispersive efficiency of the latter owing to the presence of the benzene ring stacking on to the CNT surface. In general, it is observed that the dispersive power of a surfactant is improved with the magnitude of the charge present on the surfactant molecule, the presence of benzene (or other saturated rings, such as naphthenic groups), or the presence of unsaturated carbon bonds (C5C) in the case of non-ionic surfactants such as Tween-20 or Tween-80. Moore, C.V. et al. 64 were among the first to perform a comparative study of the dispersion of SWNTs in various surfactants [ Fig. 3

Poly(ethylene glycol), PEG and its derivatives
PEG and its derivatives have long been used as a molecule of choice for functionalizing nanomaterials 66 for intracellular drug delivery 67 and targeted imaging. 68 This polymer is useful to render the nanomaterials biocompatible, increase their circulation lifetime in vivo, and to improve uptake and minimize clearance by the RES system, 69 i.e., "stealth" properties, because of its similarity to naturally-occurring carriers such as lipoproteins or viruses. Work done by Dai, H. and co-workers has adapted this method to be applicable to the dispersion of carbon nanotubes. One such approach uses a phospholipid-polyethylene glycol (PL-PEG) as a surface-coating agent for the non-covalent functionalization of SWNTs. PL-PEG is an amphiphilic polymer, in which the hydrophobic phospholipid chain sticks to the nanotube sidewall, while the hydrophilic PEG end faces outward, making a strong supramolecular complex in water. One advantage of this functionalization technique is the ability to separate excess PL-PEG molecules from the solution by repeated centrifugation, as the method does not rely on micellar stabilization of SWNTs. In addition, the terminal -NH 2 or -COOH groups at the end of the PEG chain allow for the conjugation of targeting agents for specific applications. A drawback of this approach is that the extensive sonication used during the functionalization process leads to significant sidewall damage, resulting in cutting of SWNTs into short segments, ;50-300 nm. This increases the defect density in the sidewall surface, and makes them less desirable for imaging applications. As an alternative, dialysis of the polymer from the reaction medium (post surface functionalization) can serve as an effective method to remove excess, unreacted PL-PEG.
In the first demonstration of the use of SWNTs as image contrast agents in mice, Welsher, K. et al. 70 64 Robinson et al., 72 Dhar et al. 78 70 Hong et al., 73 Hong et al., 76  a surfactant-exchange method, to displace the organic surfactant, sodium cholate, from the surface of SWNTs, with the PL-PEG polymer (1,2-distearoyl-sn-glycero-3phosphoethanolamine-N-[methoxy(polyethyleneglycol)], 5 kDa mol wt). At an injected dose of 300 lL of ;170 lg/mL, they were able to perform high-resolution intravital microscopy imaging of tumor vasculature [ Fig. 3(d)] beneath the skin. In further development, Welsher, K. et al. 71 have applied an image-processing technique called Principal Component Analysis (eigenvector decomposition) to track the whole-body blood circulation [ Fig. 3(f)] of injected PL-PEG coated SWNTs. Using this time-resolved technique, coupled with the benefits of NIR-II imaging using SWNTs, they were able to distinguish organ-level anatomical features, by taking into consideration the hemodynamics of injected nanoparticles. Subsequently, Robinson, J.T. et al. 72 were able to obtain ultrahigh passive tumor uptake [ Fig. 3(g)], ;30% injected dose/g (using enhanced permeability and retention, or the EPR effect) in a xenograft breast cancer model in mice, through long circulation lifetime (;20 h) of SWNTs solubilized using poly(maleic anhydridealt-1-octadecene)-methoxy PEG, C18-PMH-mPEG. In parallel, Hong, G. et al. 73 have used DSPE-mPEG to stably suspend SWNTs in aqueous medium. Using these PEGylated SWNTs, they were able to visualize peripheral arterial disease in a mouse model [ Fig. 3(h)], with high spatial (;30 lm) and temporal (,200 ms per frame) resolution, up to 3 mm depth, which offers ;3Â improvement in spatial resolution compared to micro-CT imaging, without the drawbacks of long exposure (hours) and high radiation doses associated with the latter. It is worth noting here that these PEGylated SWNTs are useful for therapeutic applications, too, beyond the realm of non-invasive imaging. For example, Antaris, A.L. et al. 74 have demonstrated simultaneous tumor imaging and photothermal therapy (PTT), using chirality-separated (6, 5) SWNTs. Upon extension of the vascular imaging technique, Hong, G. et al. 75 have developed NIR-II fluorescence angiography, to track the regeneration of hind limb non-invasively, post ischemic surgery. Similarly, Hong, G. et al. 76 have demonstrated imaging at depths .2 mm in the mouse brain, through the skull, with sub-10 lm resolution. This method has the ability to dynamically observe blood perfusion in the cerebral vessels in real-time, and monitor blood flow during arterial occlusion in a model of cerebral stroke [ Fig. 3 PEGylated carbon nanotubes have also been proven versatile as targeted drug-delivery systems. For example, Liu, Z. et al. 77 have shown ultrahigh loading capacity of doxorubicin, by p-stacking onto PEG-coated SWNTs, with pH-controlled release. Similarly, Dhar, S. et al. 78 have used amine-functionalized PEG-wrapped SWNTs to deliver folate-conjugated Pt(IV) complex, an anticancer drug [ Fig. 3(k)], to selectively destroy FR 1 cells. In an in vivo mouse model, Liu, Z. et al. 79 have shown ;10Â tumor uptake of the water-insoluble drug, PTX, through the application of a SWNT-PTX conjugate, by loading the drug onto the branched PEG chains of the functionalized SWNT through a cleavable ester linkage. There is a wealth of available literature using different variants of this construct for targeted drug-delivery systems. For a thorough primer on the use of PL-PEG conjugation to prepare delivery vehicles (for radiolabels, drugs, image contrast agents, gene or siRNA delivery) based on SWNTs [ Fig. 3(j)], the reader is referred to Liu, Z. et al. 80

M13 bacteriophage
Work done in our research group, led by Belcher, A.M. and co-workers, has led to the development of M13 bacteriophage as a multifunctional platform, for synthesizing aqueous-dispersed, actively-targeted SWNTs [ Fig. 4(a)], for non-invasively detecting, imaging and monitoring of cancers and infectious diseases in living hosts. M13 is a filamentous, non-lytic bacteriophage, with dimensions ;880 nm in length and ;6 nm in diameter. The five capsid proteins (p3, p5, p7, p8, and p9) of the virus can be genetically engineered to display peptides with multiple functionalities, such as targeting motifs against tumor cells (or against bacteria), drug molecules for therapy, or fluorescent probes for imaging. Specifically, our group has functionalized SWNTs longitudinally along M13, through a multivalent, high-copy p-p interaction. This M13-SWNT conjugate maintains sufficiently high fluorescence activity, in aqueous medium and upon in vivo injection into serum, to allow for high-quality NIR-II imaging of living subjects.
Yi, H. et al. 32 have demonstrated that the M13-functionalized SWNT probes can be detected in deep tissues, up to 2.5 cm in tissue-like phantoms, at very low dosage ;2 lg/mL. Further, using specific targeting of the prostate-specific membrane antigen (PSMA), they were able to show 4Â improvement in uptake of the anti-PSMA-M13-SWNT probe, in a xenograft model of PSMA-positive prostate cancer [ Fig. 4(a)]. In further development, Ghosh, D. et al. 33 have engineered the p3 protein to express a peptide against secreted protein, acidic and rich in cysteine (SPARC), which is overexpressed in aggressive subtypes of ovarian cancer. Using this targeting agent, the SPARC-binding peptide (SBP), the SBP-M13-SWNT probe was used in a pre-surgical planning of cytoreductive surgery in an orthotopic mouse model of human ovarian cancer. With a high targetto-background ratio ;112, they reported the detection of millimeter-scale tumors [ Fig. 4(c)], which should improve "optimal debulking" upon clinical translation of this technology. Simultaneously, Bardhan, N.M. et al. 34 have shown the application of M13-SWNTs for fluorescence imaging of pathogenic infections; at an order-of-magnitude lower dosage compared to other small molecule organic dyes reported in the literature. In an intramuscular model of infection, the authors were able to obtain a ;3.4Â enhancement in the fluorescence signal intensity over background [ Fig. 4(b)]. This was made possible by the use of NIR-II imaging with SWNTs, resulting in a lower signal spread, and higher signal amplification compared to shorter-wavelength  34 Zheng et al., 81 Iverson et al., 89 Kim et al., 88 Giraldo et al. 92  organic dyes. Further, in a mouse model of endocarditis (infection of the heart valves), they were able to image a ;5.7Â enhancement in the signal intensity. This is an important milestone toward the non-invasive detection of deep-tissue infections, and could go a long way toward management of the problem of the spread of antibiotic-resistant infections in the clinic.

DNA wrapping
Several groups have reported the use of single-stranded DNA (ssDNA) for the functionalization of carbon nanotubes. The p-p interaction between the nanotube and the aromatic bases of ssDNA forms a functional complex, with the helical DNA structure extending along the length of the carbon nanotube. In this structure, the nitrogenous bases are attracted to the hydrophobic CNT structure, while the hydrophilic phosphate-sugar moieties face outward, making the complex water soluble, with low toxicity effects. In the first reported use of DNA wrapping, Zheng, M. et al. 81 showed helical wrapping of ssDNA on the surface of SWNTs, supported by molecular modeling [Fig. 4(d)]. Further, it was shown that it is possible to separate fractions of SWNTs based on their electronic structure [ Fig. 4(e)], through DNAwrapping, by chromatographic techniques. A significant advantage of using this method, compared to PEGylation or virus-templated functionalization, is the ability to select 82 and separate 83 the length 84 and chirality 85,86 of the SWNTs based on the sequence of DNA used. It has been shown to be systematically possible to individually sort all the major species of singlechirality SWNTs from a heterogeneous synthesized mixture, by using recognition sequences obtained from employing a randomized DNA library 87 with a diversity ;10 60 .
Finally, it is worth mentioning a few other non-covalent approaches to SWNT functionalization, although they do not strictly fall into the above categories. Work done by Strano, M.S. and co-workers has focused on the detection of nitric oxide (NO) as a signaling molecule for inflammation. Kim, J-H. et al. 88 have designed an optical sensor [ Fig. 4(g)] based on the detection of fluorescence modulation of SWNTs in the presence of NO, by wrapping the former with 3,4-diaminophenyl-functionalized dextran (DAP-dex), which only responds to NO but not to other reactive nitrogen or oxygen species. Subsequently, Iverson, N.M. et al. 89 have designed a double-stranded DNA oligonucleotide, ds(AAAT) 7 , which allows for selectivity toward NO [ Fig. 4(f)] and is used for wrapping the SWNT, which maintains its reactivity even after conjugation to 5 kDa PEG (for improving in vivo circulation). With a half-life of ;4 h, the sensor enabled detection of NO with a detection limit of 1 lM, with no immune response to the implanted inflammation sensor observed over 400 days. This construct can also be used for the detection of intracellular signaling pathways through NO generation. 90 Using a similar approach, Iverson, N.M. et al. 91 have shown the detection of NIR-II fluorescence signal at ;5 mm depth in tissue, for alginateor PEG-hydrogel-encapsulated ssDNA-wrapped SWNT nanosensors, in response to riboflavin. In yet another interesting manifestation, Giraldo, J.P. et al. 92 have come up with a "nanobionics" approach to augment photosynthesis [ Fig. 4(h)], by suppressing the concentration of reactive oxygen species, and thereby increasing photosynthetic activity by ;3Â, via the delivery of ss(AT) 15 DNA-coated SWNT-nanoceria composites in isolated chloroplasts through leaf infiltration.

B. Covalent functionalization of carbon nanotubes
The route to covalent functionalization involves a number of different chemical reaction approaches, each with their own benefits and drawbacks. While this Review aims to cover the main groups of reactions, it does not cover every possible reaction mechanism, and the reader is referred to an excellent review by Tasis, D. et al. 93

Oxidation and hydrogenation reactions
The ability to perform "oxidative purification" of carbon nanotubes by liquid-or gas-phase oxidation is considered to be one of the milestones in nanotube chemistry. 94 This reaction leads to 95 the formation of defects in the sidewalls, cutting of the nanotube lengths, introduction of oxygen functional groups, and opening up of the end-caps 96 for closed nanotubes, along with the removal of metal catalysts that are present as impurities due to the nanotube synthesis process. 97 Oxidation of carbon nanotubes can be done using boiling HNO 3 , H 2 SO 4 , "piranha" solutions (sulfuric acid with hydrogen peroxide), 98 or using combinations of ozone, air, or gaseous O 2 as oxidants at high-temperature, HF, OsO 4 , H 2 O 2 , K 2 Cr 2 O 7 , or KMnO 4 , to name a few. 99 In terms of applications, reaction-functionalized carbon nanotubes have long been eyed as efficient multifunctional vectors 100,101 for the delivery of therapeutics, gene delivery, as well as immunomodulatory probes. Chen, J. et al. 102 have designed biotin-functionalized oxidized SWNTs for receptor-mediated endocytosis and efficient delivery of taxoid molecules (a chemotherapeutic drug) into cancer cells [ Fig. 5(d)]. Zhang, X. et al. 103 have used sodium alginate/chitosan-wrapped SWNTs after oxidative cutting, for loading and controlled release of doxorubicin to cancer cells [ Fig. 5(c)], using folate targeting. Pescatori, M. et al. 104 have demonstrated that oxidized nanotubes act as cell-specific immunostimulatory systems, as they were observed to activate immune-related pathways in monocytes but not in T-cells, associated with inflammatory processes occurring during immunemediated tumor rejection and pathogen clearance. de Faria, P.C.B. et al. 105 have used oxidized multiwall nanotubes as an anticancer vaccine formulation, with rapid internalization into dendritic cells, both in vitro and in vivo. This

Polymer conjugation
Broadly speaking, there are two main approaches 107 to covalent functionalization by polymerization: "grafting to" and "grafting from". The former approach has been very well explored, through reactions such as amidization 108 and esterification. 109 In contrast, the latter has been used to graft polymer chains onto solid substrates, starting from polymer-functionalized carbon nanotubes. For example, Viswanathan, G. et al. 110 have developed a method to in situ composite synthesis, by conjugating polystyrene (PS) chains to full-length pristine SWNTs, without disrupting the original structure, by using an anionic polymerization scheme. Qin, S. et al. 111,112 have reported the grafting of n-butyl methacrylate and PS, to form "polymer brushes" from SWNTs, by atom-transfer radical polymerization, which could theoretically be applied to a wide range of vinyl polymers. It was claimed that initiator functionalization and polymerization did not change the sidewall structures of the SWNT significantly, according to Raman and near-IR spectra measurements.
Finally, as a class of biological polymers, several groups of researchers have demonstrated SWNTs functionalized with peptides [113][114][115] [Fig. 5(e)], which can be designed to retain their biological function, such as immunogenicity, and can be used to elicit an antibody response with the right specificity. These approaches can be greatly suitable for in vivo diagnostic and therapeutic purposes, such as vaccine 116 and gene delivery 117,118 systems, as well as an exciton quenching-based optical sensing for singlemolecule detection of nitroaromatic compounds. 119

IV. FUNCTIONALIZATION OF GRAPHENE
A summary of the various techniques of functionalization used for graphene and its derivatives is presented in Table 3 in the Supplementary Material.

A. Non-covalent functionalization of graphene and its derivatives
Since pristine graphene sheets are hydrophobic in nature, they cannot be dissolved in polar solvents. This makes it important to functionalize them for biomedical applications. Similar to carbon nanotubes, the functionalization by p-interactions is a beneficial approach, since it can enable the attachment of functional groups to graphene without changing the long-range electronic network of sp 2 -bonded carbon, while introducing additional selective properties to graphene and its derivatives.

DNA functionalization
The use of DNA as a self-assembly template has proven to be an attractive route to synthesize structures into well-defined 1-or 2-dimensional arrays, or other unique geometries through techniques such as "DNA Origami". 120 For example, Mohanty, N. and Berry, V. 121 have demonstrated a suite of bioelectronics technologies [such as a single-bacterium resolution interfacial device, a label-free, reversible DNA detector, and a polarity-specific molecular transistor for DNA adsorption, Fig. 6(a)] at the microbial and molecular levels, using chemically-modified graphene. Patil, A.J. et al. 122 have reported the use of ssDNA to stably disperse single-layer graphene in aqueous suspensions, at concentrations as high as ;2.5 mg/mL [ Fig. 6(e)]. Simultaneously, Liu, J. et al. 123 developed a method to decorate GO and reduced GO (rGO) with DNA, and used it to pattern Au nanoparticles into twodimensional arrays [ Fig. 6(c)]. This approach has the benefit of good stoichiometric control over the ratio of nanoparticles to the graphene nanosheets, compared to other in situ growth methods. In parallel, Lu, C-H. et al. 124 have shown the application of GO as a platform for fast, sensitive detection of biomolecules such as DNA, by restoration of the fluorescence emission [ Fig. 6(b)] through a binding event of a complementary sequence of DNA (analyte to be detected) to the one with quenched signal adsorbed on the surface of GO. Later, Xu, Y. et al. 125 have reported a facile 3D self-assembly method to prepare GO/DNA composite hydrogels [ Fig. 6(d)], with high mechanical strength, high dye-loading capacity, good environmental stability, and self-healing capability. These mechanically flexible, electrically conductive, functionalized graphene substrates offer new opportunities for bio-nanodevices.

Poly(ethylene glycol)
The method of PEGylation of GO and rGO has long been investigated, for the purposes of cell imaging and drug delivery. A detailed analysis by Wojtoniszak, M. et al. 126 reveals that such functionalization promotes biocompatibility of GO and rGO, over a wide range of concentrations, ;25-3125 lg/mL, with low cytotoxicity effects [ Fig. 7(c)]. These considerations are important while designing graphene-based nanocarriers for in vivo theranostic applications. As an example, Robinson, J.T. et al. 127 developed nanosized rGO, ;20 nm in size, with non-covalently functionalized PEG, with an Arg-Gly-Asp (RGD) targeting motif for highly effective photoablation of cells in vitro. In the field of biosensor device applications, Yoon, H.J. et al. 128 have non-covalently functionalized GO with PL-PEG-NH 2 , and used this construct to decorate EpCAM antibodies [ Fig. 7(e)] for the capture of circulating tumor cells with a capture yield of ;73% in whole blood [ Fig. 7(f)]. Building on this approach, the same group of researchers have developed 129 functionalized GO dispersed in a matrix of thermoresponsive polymer for thermally-triggered release of the cells post-capture. In a recent report, Gupta have reported a facile, chemically clean, and environmentally green method to modify the oxygen functionalization and intercalate PEG200 within rGO, through the use of hydrogen-bonding [ Fig. 7(g)]. This method preserves the chemical and structural integrity of the carbon network in rGO, and also provides an effective dispersion of the hydrophilic polymer in the rGO, which is useful as a lubricating agent. The thin film was demonstrated to have application as a solid-state lubricant, with improved wear resistance to c-irradiation. While there have been no reports yet of biomedical applications of this technique, it may find use in providing lubrication coatings for implants and prosthetic joints.

Other approaches to functionalization
While the above sections listed the most common techniques for the non-covalent functionalization of graphene and its derivatives, there are numerous examples of unconventional methods used in the literature.
Work done in our group, in collaboration with Grossman, J.C. and co-workers, has led to the development of a 1-step, mild thermal annealing process at 80°C, to improve the properties of as-synthesized GO 131 and rGO. 132 This approach is unique because it relies simply on the thermal annealing treatment, with no additional chemical modifications involved. Moreover, it is scalable, and can be applied to a wide variety of samples in the form of suspensions, powders, thin films, and free-standing GO paper. The method works by promoting a redistribution of the oxygen functional groups on the graphene basal plane, through a phase transformation process, which drives the diffusion of the oxygen functional groups from the inhomogeneous, randomly-distributed, mixed sp 2 -sp 3 state into distinct , while preserving the total oxygen content of the system. Consequently, this annealing treatment results in an improvement in the sheet resistance of as-synthesized GO by 4 orders-of-magnitude, an improvement in the visible absorption by up to 38%, and a tunable, controlled blue-shift in the photoluminescence spectrum, which can be exploited in the fabrication of biosensing devices. This method is used for enhancing the cell capture efficiency of white blood cells from whole blood, as described in Sec. IV. B. 1.

B. Covalent functionalization of graphene and its derivatives
There have been several procedures developed to attach chemical functionalities to graphene-either through direct reaction with pristine graphene sheets, or through chemical modification of GO or rGO.

Poly(ethylene glycol) and other polymers
GO has been grafted with polymer chains containing reactive functional groups such as hydroxyls and amines. Different polymers have been used for this purpose, including PEG, polylysine, polyallylamine, poly(vinyl alcohol), and PS, to name a few. For example, Liu, Z. et al. 133 have developed PEGylated nanographene oxide in water, and used it to non-covalently load a waterinsoluble anti-cancer drug molecule, SN38, through p-p stacking. This construct was found to be ;1000Â more potent than a clinically approved camptothecin prodrug molecule; almost comparable to free drug in DMSO [ Fig. 8(a)]. Using a similar approach on pristine graphene nanosheets, Yang, K. et al. 134 showed highly efficient passive targeting (via the EPR effect) in xenograft tumors in mice. Subsequent near-infrared PTT showed an improvement in survival from ;16 days (untreated control) to .40 days with the treatment [ Fig. 8(b)]. Building upon this approach, Zhang, W. et al. 135 have used NIR PTT in synergy with chemotherapy (doxorubicin) loaded onto PEGylated nano-GO, for achieving higher therapeutic efficacy, in a xenograft mouse model. Tian, B. et al. 136 have attempted to use photodynamic therapy (PDT) enhanced by PTT, by loading a photosensitizer molecule (Chlorin e6) onto PEG-GO [ Fig. 8(c)]. As a follow up to this approach, Cao, J. et al. 137 have used diffusion-weighted MRI to monitor the effectiveness of PTT/PDT [ Fig. 8(d)], with the most effective treatment resulting in tumor-free condition after 60 days. Similarly, Feng, L. et al. 138 have used dual polymer-functionalized nano-GO (PEG and polyethyleneimine, PEI) to achieve light-controlled delivery of siRNA, by increasing the cell membrane permeability using low-power NIR irradiation. In an alternative approach, Shan, C. et al. 139 have covalently functionalized water-soluble graphene using biocompatible poly(L-lysine), PLL. PLL is an attractive polymer functionalization agent, which has applications such as biosensing, promoting cell adhesion, drug delivery, and electrochemical sensing. By immobilizing horseradish peroxidase on the PLL-graphene nanocomposites, they demonstrated amplified biosensing of H 2 O 2 . Similarly, Hu, S-H. et al. 140 have attached quantum dots on PLL-coated rGO, for folate-targeted cellular imaging and in situ-monitored PTT. Hong, H. et al. 141 have used 64 Cu-labeled PEGylated GO, for in vivo PET imaging of subcutaneous xenograft tumors [ Fig. 8(e)] through active targeting of the tumor neovasculature using CD105 overexpressed in many solid tumors. For an excellent primer on the protocol for the functionalization of graphene and its derivatives with PEG, the reader is referred to Yang, K. et al. 142 Work done in our own research group, in collaboration with Ploegh, H.L. and co-workers, has made good use of the functionality afforded by PEG-conjugated GO. GO was functionalized with diamine-terminated PEG (NH 2 -(PEG) n -NH 2 ). The terminal amines of the PEG chains were then reacted with dibenzocyclooctyne-Nhydroxysuccinimyl ester (DBCO-NHS), to activate it for a "click" coupling reaction. We then conjugated VHH7 nanobodies (single-chain only fragments of conventional antibodies, which are specific to murine Class-II MHC 1 cells) to the DBCO handle. Using this chemistry, we were able to show highly efficient, very specific cell capture 35 of white blood cell populations from small volumes (;30 lL) of murine whole blood [ Fig. 8(g)], using a relatively inexpensive device (without the need for cell sorting or microfluidic apparatus), under ambient conditions.

Free radical addition
Organic free radicals and dienophiles with affinity toward the C5C sp 2 carbons in graphene are some of the most attractive candidates for reactive covalent functionalization of graphene. For example, the reaction of a diazonium salt, 143 or other reducing agents such as hydrazine monohydrate (N 2 144 with the unsaturated carbons results in the formation of nitrophenyls on the graphene sheet. Such covalent functionalization results in the introduction of a band gap (;0.4 eV) 145 in the normally zero-gap pristine graphene, which makes it a suitable candidate for semiconducting applications. Several other groups of researchers, such as Sharma, R. et al. 146 and Hossain, M.Z. et al. 147 have demonstrated high reactivity of graphene sheets toward nitrophenyl-contributing electron acceptor diazonium salts. Alternatively, Liu, H. et al. 148 have shown free radical addition of phenyl groups, by photo-activated initiator reaction of benzoyl peroxide in toluene, N.M. Bardhan: 30 years of advances in functionalization of carbon nanomaterials for biomedical applications: a practical review resulting in the attachment of the benzene ring through an sp 3 bond on to the graphene substrate.
In terms of applications, the availability of a rich chemistry of functional agents lends itself to creative uses in biology and medicine. For example, Zhang, L. et al. 149 used sulfonation with the aryl diazonium salt of sulfanilic acid, to introduce SO 3 H groups, and then conjugate folic acid (FA) onto nano-GO. Through controlled loading of 2 anticancer drugs, doxorubicin and camptothecin, via p-p stacking, they reported much higher cytotoxicity against MCF-7 breast cancer cells compared to the activity of each drug used individually [ Fig. 8(f)]. Zhang, X. et al. 150 have developed hybrid materials of graphene with porphyrin, namely, graphenetetraphenylporphyrin (graphene-TPP) and graphenepalladium tetraphenylporphyrin (graphene-PdTPP)  133 Yang et al., 134 Tian et al., 136 Hong et al. 141  through a one-pot cycloaddition synthesis route, with potential applications in fluorescence imaging of cells and in sensors. Wei, G. et al. 151 have used a wet-chemical diazotization method, to functionalize rGO with p-aminobenzoic acid. After the introduction of PEI and target molecule FA, they were able to load and deliver elsinochrome A (EA) and doxorubicin, which showed synergistic effect of enhanced chemotherapeutic efficacy by arresting cell cycle activity in the G2 phase.
In another work, Gollavelli, G. and Ling, Y-C. 152 have designed multi-functional graphene with magnetic, water-solubility and fluorescence properties, by covalently attaching poly(acrylic acid), PAA and fluorescein o-methacrylate on rGO, using benzoyl peroxide activated by microwave irradiation. Using this approach, they were able to perform in vitro cellular imaging in HeLa cells, and in vivo in zebrafish. For a good review on the use of chemically modified graphene in synthetic biology, the reader is referred to Servant, A. et al. 153

V. CONCLUDING REMARKS AND FUTURE DIRECTIONS
In this Review, we have undertaken a landmark survey of the progress made in the past 30 years, with respect to the functionalization of CANOMATs for biomedical applications. This investigation has been done taking into account relevant considerations such as the method of functionalization, the number of steps, and time required, need for special equipment, the phases in which the method can work, the properties affected, techniques to characterize the product, and importantly, whether the approach is robust enough to be scalable and reproducible, with a list of the applications explored till date. The rapid advances in the methods of functionalization, with consequent applications in targeting, imaging, and drug delivery (in pre-clinical models), have been nothing short of breathtaking.
That being said, it seems a bit disappointing, though hardly surprising, to note that very few (if any?) of these novel materials have actually made it past the research laboratory to clinical practice (i.e., from the "bench" to the "bedside"). Why is this so? Part of the reason is the regulatory burden of proving safety and efficacy of these materials prior to being approved for diagnostic or therapeutic use (for a brief discussion on the safety of CANOMATs in living organisms, refer to Sec. 5 of the Supplementary Material). However, the author opines that a bigger factor is the lack of a systematic plan to identify the areas where the applications of CANOMATs would have the most impact in human medicine. Toward this end, here are a few recommendations to direct our collective efforts in materials research, for making them more suitable for clinical translation: (1) Test safety and efficacy in larger animal models: almost all the proof-of-principle studies of targeting, as well as long-term biodistribution studies of the safety of CANOMATs, have relied on rodent models (mice or rats). As a step toward first-in-human testing, these materials need to be tested for safety in larger animals; perhaps in porcine models, and then in primates.
(2) Build an adaptive, hierarchical nano-scale architecture: there is a need for adaptive materials design, to be reversibly scalable, on-demand, in vivo, from the nano-scale to the bulk: through processes such as selfassembly or bond cleavage, mediated by environmental factors such as pH-, oxygen content-, temperature-, or light-activation for smart applications.
(3) Implement composites of C 60 -SWNT-GO: composite CANOMATs can provide a synergistic advantage by augmenting their complementary properties. For example, it may be possible to design large-area biosensors based on functionalized GO, which use SWNTs for an optical readout, and C 60 derivatives to encapsulate and deliver drug molecules, while avoiding the limitations posed by each of these individual nanomaterials.
(4) Integrate bio-nano-composites: there is a pressing need to undertake research based on the idea of "convergence", which aims to refine our nanotechnology toolkits by understanding the behavior of materials at the interface of engineering and biology. This is increasingly important as we aim to transition toward a "personalized medicine" approach in cancer therapy.
(5) Design molecular probes to penetrate the bloodbrain barrier: the efficacy of nanotechnology-based approaches toward the treatment of brain tumors, cerebral stroke, and other neurological disorders such as Alzheimer's or Parkinson's has been limited, 154 due to problems with overcoming the BBB. More rational design choices need to be considered, to make CANOMATs the preferred mode of drug delivery by overcoming the physiological barriers of the brain.
The question that medical professionals in the clinic should be asking, is: "Can we CANOMAT?" the answer, with cautious optimism, is "Yes, we can."