Deconvoluting hepatic processing of carbon nanotubes

Single-wall carbon nanotubes present unique opportunities for drug delivery, but have not advanced into the clinic. Differential nanotube accretion and clearance from critical organs have been observed, but the mechanism not fully elucidated. The liver has a complex cellular composition that regulates a range of metabolic functions and coincidently accumulates most particulate drugs. Here we provide the unexpected details of hepatic processing of covalently functionalized nanotubes including receptor-mediated endocytosis, cellular trafficking and biliary elimination. Ammonium-functionalized fibrillar nanocarbon is found to preferentially localize in the fenestrated sinusoidal endothelium of the liver but not resident macrophages. Stabilin receptors mediate the endocytic clearance of nanotubes. Biocompatibility is evidenced by the absence of cell death and no immune cell infiltration. Towards clinical application of this platform, nanotubes were evaluated for the first time in non-human primates. The pharmacologic profile in cynomolgus monkeys is equivalent to what was reported in mice and suggests that nanotubes should behave similarly in humans.

S ingle-wall carbon nanotubes (SWCNT) have attracted immense interest as a platform for pharmaceutical interventions due to their unique physicochemical properties and remarkable fibrillar pharmacology 1,2 . These high aspect ratio macromolecules can be readily derivatized via covalent chemical or non-covalent physical means, which advocates for their consideration in the development of multi-functional scaffolds capable of targeting disease for diagnostic imaging or therapeutic drug delivery 1,3,4 . Despite the potential to assume a novel position in medical applications, carbon nanotubes have not entered the clinic due to a limited understanding of their fate in vivo as well as a lack of consideration for the role that functionalization confers to pharmacology 3 . Sidewallfunctionalized SWCNT, covalently appended with multiple primary amines 5 , have been shown to undergo very rapid renal elimination in animal models via glomerular filtration [6][7][8][9][10][11][12] . Thus, these water-soluble macromolecules exhibited a pharmacokinetic profile similar to that of a small molecule one-thousandth their size because they possess a high aspect ratio 6,8,11 .
Most particulate drugs are scavenged and processed by the liver 1 . Hepatic accumulation and subsequent hepatobiliary elimination of functionalized nanotubes has been reported in animal models, however, the fraction of the administered dose that localized in the liver varied tremendously in correlation with the methods used to functionalize the nanotube [6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24] . Furthermore, the hepatic location of SWCNT has generally been described as liver accumulation with only limited identification of the cell populations involved in accretion and no information on the processes of receptor-mediated endocytosis 25,26 or explanation of hepatobiliary elimination. The liver is an essential organ with a complex cellular architecture that regulates an extensive range of metabolic functions. The liver is comprised primarily of hepatocytes, Kupffer cells, liver sinusoidal endothelium (LSEC), stellate cells, bile duct epithelium and vascular endothelium. Mapping the cytotopographic distribution and hepatic processing of functionalized nanotubes is required for understanding their fate in vivo. These data will, in turn, inform on the optimal applications and possible toxicities of these unique materials with wider import for nanomedicine as a whole.
This paper reports several novel and unexpected details of hepatic processing of covalently functionalized nanotubes including receptor-mediated endocytosis, cellular trafficking and biliary elimination. First, water-soluble ammonium-functionalized fibrillar nanocarbon is found to preferentially localize in the fenestrated sinusoidal endothelium of the liver but not hepatocytes or resident macrophages. Second, both scavenger Stabilin receptors mediate the endocytic clearance of nanotubes. This particular receptor specificity for nanocarbon leads us to find that lymph tissue also accumulates carbon nanotubes. Third, we propose a mechanism for the hepatobiliary elimination of nanocarbon. Fourth, biocompatibility is evidenced by the absence of cell death and no immune cell infiltration. Towards clinical application of this platform, nanotubes were evaluated for the first time in nonhuman primates. The pharmacologic profile in cynomolgus monkeys is equivalent to what was reported in mice and suggests that nanotubes should behave similarly in humans.

Results
Nanomaterial synthesis and characterization. An ammoniumfunctionalized carbon nanotube (fCNT) was used to probe hepatic processing (Fig. 1a). This nanomaterial was prepared and characterized as previously described 8 . Transmission electron microscopy analysis of solid fCNT confirmed a mean length of 195±69 nm (Fig. 1b). These fCNT were water soluble (10 g l À 1 ) and chromatographically characterized (Fig. 1c); in addition, they were rapidly renally filtered in both murine 6,8,11 and primate models. The Raman radial breathing mode feature (234 cm À 1 ) yielded a nanotube diameter of 0.96 nm (ref. 6). These amino-functionalized nanotubes were further appended with fluorescent moieties and metal ion chelates to report location via multiple imaging modalities and investigate whole body, organ, cell and intracellular distribution.
Murine pharmacokinetic profile of fCNT. Intravenously administered fCNT rapidly cleared the blood and a fraction accreted in the liver or was eliminated intact via the bile. Dynamic positron emission tomographic (PET) imaging and tissue collection were performed to assess the whole body pharmacokinetic profile in a naive animal model. High contrast images of a representative mouse (Fig. 1d) showed that the activity cleared the vascular compartment by 1 h and was predominantly accreted in liver, spleen and kidney (earlier images at 3 and 20 min are shown in Supplementary Fig. 1). The balance of the injected dose (80-85%) was rapidly eliminated to the urine 8 as shown by the bladder image at 60 min (Fig. 1d). Time-activity curves for blood and liver graphically show rapid clearance from the blood (Fig. 1e) and swift accumulation of a fraction of the injected dose (ID) in the liver (Fig. 1e,f). Detailed tissue harvest experiments ( Fig. 1f and Supplementary Fig. 2) measured the %ID that partitioned into the blood, tissue, and bile as a function of time. Activity was measured in the liver (1.39 ± 0.39%ID g À 1 ), bile (0.35±0.15%ID g À 1 ), spleen (0.42±0.11%ID g À 1 ), lymph node (0.47 ± 0.42%ID g À 1 ) and small intestines (the duodenum, jejunum, and ileum contained 2.17±2.31%ID g À 1 ) at 1h. Radiochromatographic analysis of bile harvested from the gall bladder indicated intact fCNT ( Supplementary Fig. 3).
LSEC accumulation of fCNT. Liver accumulated fCNT was localized exclusively in the hepatic sinusoids, associated with small-nucleated cells lining the sinusoids (Fig. 2, Supplementary  Figs 4 and 5). fCNT was imaged using anti-AF488 immunofluorescence (IF) staining that was directed at AF488 moieties covalently appended onto the nanotube sidewall. This anti-AF488 probe was multiplexed with an array of co-stains selected to distinguish distinct cell types and organelles. The absence of fCNT in hepatocytes was corroborated with N-cadherin staining that delineates the hepatocyte plasma membrane. Controls included mice that were injected with AF488-alone ( Supplementary Fig. 4) or vehicle ( Supplementary  Fig. 5) and both were negative for AF488 signal.
The fenestrated LSE that lines the hepatic sinusoids was the structure that localized fCNT as confirmed by multiplex IF detection using CD31, Lyve1 (lymphatic vessel endothelial hyaluronan receptor), anti-AF488, and DAPI stained liver sections ( Fig. 2c-e, Supplementary Fig. 6). CD31 is a pan-endothelial marker while Lyve1 is a marker for lymphatic vascular endothelial cells; LSEC express Lyve1 and are a notable exception. The CD31 and Lyve1 membrane markers entirely circumscribed the anti-AF488 signal, substantiating fCNT clearance into the LSEC population (Fig. 2d). Vascular endothelium (VE) within the liver did not accumulate fCNT (Fig. 2e) and called attention to the differential capacities of specialized fenestrated LSEC and VE to accumulate and internalize fCNT. Liver sections stained with MECA32, another VE cell marker, confirmed this finding. Resident Kupffer cells (Supplementary Fig. 7) and bile duct epithelium ( Supplementary Fig. 8) did not accumulate fCNT.
in vivo and in vitro. Competitive inhibition of these two fasciclinlike hyaluronan receptor homologues with heparin reduced the accumulation of fCNT in liver, spleen and lymph node. Heparin has been reported to be a substrate for Stabilin-1 and Stabilin-2 (refs 27-29) and both receptors are expressed on the specialized sinusoidal endothelium of liver, spleen and lymph [29][30][31] . Heparin competition in vivo resulted in a decrease in liver, spleen and lymph clearance of fCNT versus heparin null controls (P ¼ 0.017, 0.0025 and 0.09, respectively) with concomitantly more fCNT in the blood compartment of the heparin treated mice (P ¼ 0.00091) as shown in Fig. 3a,b, respectively. LSEC isolated from livers strongly expressed mRNA for both Stabilins and in contrast to Kupffer cell expression ( Fig. 3c and Supplementary Fig. 9). Isolated cell phenotype was confirmed by FACS analysis (Supplementary Fig. 10). Receptor specificity for fCNT was confirmed in vitro using CHO cells stably transfected with full-length human Stabilin-1 or Stabilin-2 expression constructs. Confocal microscopy showed punctate cellular structures in which fCNT (AF488-positive) co-stained with each anti-Stabilin-1 and anti-Stabilin-2 antibody, respectively  Supplementary Figs 11 and  12). Bound or internalized fCNT was quantified using FACS that showed significant clearance (Fig. 3f,g) versus CHO-transfected with an empty vector (negative control); acLDL binding served as positive control, (Supplementary Fig. 13a,b). Furthermore, the binding of fCNT was significantly reduced in these Stabilinexpressing CHO cells in the presence of a challenge from excess heparin in vitro ( Supplementary Fig. 13c,d). Organelle trafficking of fCNT in the mouse liver was monitored as a function of time. Punctate anti-AF488 staining patterns were associated with EEA-1 stained early endosomes at 5 min after injection; with both GM130 and Giantin stained Golgi compartments at 5 min through 1 day; and Lamp1 stained lysosomes at 40 min and 1 day (Fig. 3h-j and Supplementary Fig. 14).
Hepatic biocompatibility of fCNT. Biocompatibility of functionalized nanotubes in mice was examined as a function of time (1, 3, 7 or 30 days post injection) using terminal deoxynucleotidyl transferase dUTP nick end labelling (TUNEL) and cleaved caspase-3 (CC3) IF staining of immunocompetent mouse liver versus untreated controls. Few DNA fragmentation (Fig. 4a) or apoptotic ( Fig. 4b) events were noted in either the fCNT-treated or untreated control mouse livers. Further, no T cell infiltration was detected by CD3 staining (Fig. 4c). No morphological differences between treated and control mouse liver was observed by hematoxylin and eosin (H&E) staining (Fig. 4d). The normalized number of CD3positive T cells remained stable throughout the time course of the study (P40.01 compared with untreated control for all time points) and was not different compared to untreated control (Fig. 4e). The AF488-fCNT signal decreased with time ( Fig. 4f). Human hepatocyte microspheres were exposed to fCNT (15 and 30 mg l À 1 ) for 3 days in vitro and remained viable at concentrations exceeding those used in vivo. No morphological or apoptotic differences between treated and control spheres was observed by H&E, TUNEL or CC3 staining ( Fig. 4g-l and Supplementary Fig. 15).
Pharmacokinetics and toxicology of fCNT in non-human primates.
Towards clinical application of this fCNT platform, the fCNT pharmacokinetic and toxicologic profiles were evaluated in naive non-human primates using PET-computed tomography (PET/ CT) imaging, blood chemistry, haematological analyses and histopathology. The fibrillar nanomaterial exhibited similar blood clearance kinetics, tissue biodistribution, and renal elimination in a pair of 5 kg cynomolgus monkeys (Macaca fascicularis) as compared to mice. A 1 mg kg À 1 dose of [ 86 Y]fCNT was administered intravenously and had a blood half-life of 6-7 min. The majority of the dose was rapidly eliminated in the urine with a fraction accumulated in the liver and kidneys (SUV was 1 and 16, respectively). Figure 5a is a fused PET/CT image of [ 86 Y]fCNT in a non-human primate showing liver and kidney distribution. Time-activity curves for blood washout and hepatic clearance were equivalent to what was observed in mice (Fig. 5b).
The nanotubes were also biocompatible as shown by acute and chronic liver toxicity evaluations of blood chemistry and haematological indices. Biomarkers indicative of hepatic injury (alkaline phosphatase, lactate dehydrogenase, alanine aminotransferase, aspartate aminotransferase and albumin) and WBC counts were changed little relative to baseline values ( Fig. 5c-h) and agreed with published ranges for similarly aged male animals 32 Fig. 2c with the yellow arrow) revealed that the Lyve1 (red) marker entirely circumscribed the anti-AF488 (green) signal (scale bar, 4 mm). (e) The liver VE did not accumulate fCNT (white solid arrow) while the fenestrated LSEC did accumulate fCNT (yellow asterisks). The liver vascular endothelium was stained with CD31 (red), LSEC marked with Lyve1 (white) and CD31, fCNT (appended with AF488) were located with anti-AF488 stain (green) and nuclei with DAPI (blue) (scale bar, 10 mm).
was scored normal. These pharmacokinetic and toxicological fCNT profiles in primates suggest that the favourable pharmacological properties of fCNT should translate in humans.

Discussion
Nanoparticles, in general, are severely limited by untoward hepatic clearance and lack of renal clearance. While the bulk of fCNT are renally cleared [6][7][8][9][10][11] , the next most prominent organ contributing to their accumulation and elimination is the liver. Data reported herein show the unexpected result that liver processed fCNT either by receptor-mediated LSEC scavenging or hepatobiliary elimination. Hepatic biocompatibity of these nanomaterials was confirmed in rodents and primates and is now explained by a combination of specific LSEC scavenging and intact biliary clearance. Hepatic and renal processing and  Fig. 13). Organelle trafficking of intracellular fCNT (anti-AF488 (green)) was observed within LSEC in mouse tissue in punctate patterns and was shown through co-staining to be associated with different intracellular organelles as a function of time. 3D-rendered confocal images of the (h) early endosome (EEA-1, red) at 5 min after injection; (i) Golgi compartments (Giantin, red) at 40 min; and (j) lysosomes (LAMP1, red) at 1 day. A complete time course of these representative data is shown in Supplementary Fig. 14  Images of human hepatocyte microspheres exposed to 30 mg l À 1 fCNT for 1 day in vitro remained viable. No morphological or apoptotic differences were observed by H&E (g,h), TUNEL (i,j) or cleaved caspase-3 (k,l) staining of untreated control hepatocyte microspheres versus fCNT-treated hepatocyte microspheres, respectively. Scale bar, 50 mm in all images. The complete 3 day treatment dataset and positive control tissues are shown in Supplementary Fig. 15.
clearance of this fibrillar nanomaterial are advantageous biological properties. Given the important role of the liver in mammalian physiology and homoeostasis, the absence of fCNT in parenchyma was an interesting and potentially significant finding that correlated with the observed biocompatibility of these covalently functionalized nanomaterials 6  fCNT are scavenged by LSEC via a Stabilin receptor-mediated process. Our physicochemical characterizations indicated that fCNT is soluble and well dispersed in physiologic milieu and behaved like a fibrillar macromolecule rather than particulates which would be expected to be phagocytosed by Kupffer cells 38 . Accordingly, we observed no evidence of Kupffer clearance of fCNT using discrete cell markers to interrogate and unequivocally identify each sinusoidal cell phenotype. Furthermore, we confirmed that Kupffer cells did not express either Stabilin receptor which explained the lack of fCNT clearance by this resident macrophage. The hypothesis that macromolecular fCNT remain soluble and individualized in vivo was supported by this study and previous reports [6][7][8]11 . In direct contrast, non-covalently modified SWCNT (for example, dispersed with surfactant or polyethylene glycol (PEG)) exist only for a brief interval in the blood before displacement of the solubilizing agent by serum protein 17 . This disadvantage of non-covalent dispersal is an existing methodological problem that makes our covalently functionalized CNT study relevant and impactful moving forward. Surfactant-and polyethylene glycol-dispersed nanocarbon suspensions are inherently unstable and readily aggregate in vivo, rendering these non-soluble materials susceptible to macrophage opsonophagocytosis 21 . In addition, since surfactant-dispersed nanomaterials were unable to efficiently clear renally, a greater majority of the injected dose accumulates in liver presumably in macrophages [17][18][19][20][21][22] . The untoward behaviour of non-covalently modified nanotubes in vivo is in direct contrast to soluble covalently functionalized fCNT that are rapidly filtered by the kidneys and exhibit only minimal accumulation in liver, spleen and lymph [6][7][8][9][10][11][12][13] .
The vascular endothelial termini in the liver, spleen and marrow are tortuous sinusoids. The hepatic sinusoids interface between the blood supply and hepatocytes and mediate scavenging and transport of blood-borne solutes. LSEC are fenestrated endothelium possessing numerous open fenestrae, without diaphragm or basement membrane 30,[36][37][38][39][40] . A sieve plate morphology and high endocytic capacity support the unique role of LSEC in solute trafficking and active scavenging of a range of macromolecules and colloids that escape Kupffer phagocytosis. Punctate anti-AF488 staining of LSEC organelles indicated fCNT trafficking through the early endosome, Golgi, and lysosome compartments and supported receptor-mediated endocytosis as the mechanism of clearance 25,26 . Furthermore, we now provide evidence that Stabilin-1 and Stabilin-2 mediated endocytic clearance of fCNT in vivo. The Stabilins are scavenger receptors that recognize and clear a range of molecular substrates (for example, hyaluronan, dermatan sulfate and acetylated low density lipoprotein) 38 . Heparin clearance remains controversial notwithstanding strong evidence for Stabilin-1 and Stabilin -2 binding 27-31 countered by reports that heparin is not a ligand for Stabilin-2 (refs 42,43). Our new data definitely suggest that heparin did compete with fCNT clearance by LSEC in vivo and with both Stabilin-1-and Stabilin-2-expressing CHO cells in vitro. Heparin also reduced the bidodistribution of fCNT in spleen and lymph tissue.
Stellate cells reside in the hepatic perisinusoidal space of Disse intimately positioned between LSEC and hepatocytes 41 . The paracrine secretion of vascular endothelial growth factor by stellate cells and hepatocytes sustains the LSEC population and promotes autocrine production of nitric oxide by LSEC 30 . It has been shown that LSEC play an important role in maintenance of stellate cell quiescence and prevent their activation and loss of the vascular endothelial growth factor paracrine effect 44 . While LSEC were the predominant target for fCNT, there was rare evidence of fCNT in stellate cells (Supplementary Fig. 16). Because one function of the LSEC is to guard these cells and prevent activation 30,44 , occasionally fCNT may have been taken-up by stellate cells. It is not clear how or if the stellate cell recognized fCNT, but the literature on these perisinusoidal cells is still developing.
The vascular endothelial architecture serves as the primary conduit to distribute fCNT in vivo but there was no evidence that these cells accumulated this nanomaterial (Fig. 2e). This particular result calls attention to the differential functionality and receptor expression of these two endothelial cell types with continuous or discontinuous fenestrated cytoplasm. Cultured endothelial cells were observed to accumulate SWCNT, albeit under non-physiologic conditions over a prolonged time 45 , however, our PET imaging and IF studies have not identified any fCNT accumulation in the VE 6,8,11 .
Intact fCNT cleared the liver via secreted bile ( Fig. 1f and Supplementary Figs 2 and 3). The presence of intact radiolabeled fCNT in the bile, intestines and faeces strongly supported a mechanism of hepatobiliary clearance. The epithelial bile duct cells that comprise these channels did not show any fCNT accretion (Supplementary Fig. 8). We propose that a portion of the fCNT in the sinusoid, not endocytosed by LSEC, diffused from the Disse space into the biliary canaliculi and subsequently mixed with bile. Because we observed no evidence of fCNT in hepatocytes and the bile contained intact nanomaterial, it is difficult to reconcile a mechanism whereby hepatocytes mediated transport of fCNT from blood to bile. Evidence does exist for a permeable barrier permitting pigments and cellular debris to bypass hepatocyte processing and transit directly from blood to bile 46 ; scanning electron microscopy has shown 100 nm zones between the space of Disse and the bile canaliculi that were interpreted as sites for molecular diffusion 47 . These permeable Disse/canalicular junctions have been utilized to effect retrograde, non-viral gene therapy to the liver via infusion from the biliary tree 48 . Conventional views of hepatocyte-mediated elimination of blood-borne solutes into bile have overlooked this interesting diffusion process across the Disse/canalicular junction 49 .
The spleen and lymph nodes also accumulated fCNT and clearance was connected to the expression of both Stabilin receptors in these tissues [28][29][30][31] . The implications are exciting because one can now predict fCNT accumulation based on receptor recognition. Lymph node clearance of fCNT has not been reported until now and only discovered on the basis of lymph expression of Stabilin receptors. Based on our identification of Stabilin-mediated endocytosis of fCNT, we looked for and confirmed accumulation in lymph nodes. The specific accumulation of fCNT by these organs was observed in biodistribution studies ( Fig. 1f and Supplementary Fig. 2) and blocked in heparin competition studies (Fig. 3a). In parallel with the liver, the specialized splenic sinusoidal endothelium (SSEC) also internalized fCNT while splenic macrophage clearance was not observed (Supplementary Fig. 17). The splenic sinusoids are tortuous VE termini lined with SSEC. The SSEC differ from LSEC in that they exhibit continuous cytoplasm and disorganized basement membrane 30 . The spleen is another important reticuloendothelial tissue and it paralleled the liver in the cell types that localized fCNT. Lymphatic endothelial cell accumulation of fCNT now warrants further investigation.
These data contribute significantly to our understanding of the hepatic processing of fCNT in vivo. The pharmacokinetic and safety profile of fCNT in a non-human primate strongly proposes extrapolation to humans. The predominant hepatic cell type that accumulated this fibrillar nanocarbon was a professional scavenger endothelial cell in the sinusoid that performed rapidly and at high capacity. Mouse LSEC has 14 ± 5 fenestrae per mm 2 (humans have 15-25 per mm 2 ) with diameters of 99±18 nm (humans, 50odo300 nm) 36 . This is an avid, dedicated mammalian scavenger cell and in combination with intact biliary elimination of fCNT has yielded a very favourable biological outcome for soluble fibrillar nanocarbon in animal models. Biocompatibility has been confirmed in several model systems and now these results explain the action of the host on fCNT and add further weight to arguments for use in man.
Potential therapeutic application of fCNT in humans could focus on the critical role of LSE in numerous liver disorders (for example, ischemia-reperfusion injury, sinusoidal obstructive syndrome, nonalcoholic steatohepatitis, and pseudocapillarization in the aging population) 30 . fCNT could be envisioned as a molecular platform to specifically target the LSEC and deliver therapeutic interventions (for example, small molecule drugs or RNA interference) to treat LSEC-based disorders. Hepato-targeted gene therapy strategies are being explored 50,51 and fCNT could be included as a selective delivery platform based on our favourable pharmacokinetic and toxicologic data and the ability of fCNT to strongly bind, deliver and unload siRNA in vivo 52,53 . Macrophages that express Stabilin-1 (not liver-resident Kupffer cells) should accumulate fCNT via receptor-mediated processes and suggests further investigation 54 . Also, now that lymph node clearance of fCNT has been described, applications for lymph targeting strategies could be explored.
Dynamic PET imaging to investigate pharmacokinetics. Dynamic imaging was performed with the microPET Focus 120 (CTI Molecular Imaging) for the mouse model (#, NCr/nu/nu, Taconic) as described 8 . Dynamic PET/CT imaging was performed with a Siemens Biograph mCT PET/CT system for the non-human primate model (#, Macaca fascicularis, Charles River) and described in the Supplementary Information. For all in vivo experiments, housing and care were in accordance with the Animal Welfare Act and the Guide for the Care and Use of Laboratory Animals. The animal protocols were approved by the Institutional Animal Care and Use Committee at MSKCC.
Biodistribution and elimination. Mice (#, NCr/nu/nu, Taconic) received an IV injection of [ 111 In]fCNT containing 0.04 mg of SWCNT construct and 74 kBq (0.002 mCi) of 111 In per mouse via the retroorbital sinus. The animals were placed into 4 groups of 3-5 mice per group. Each group was sacrificed with CO 2 aspiration at 1, 3, 24 h and 7 days. Tissue samples (blood, heart, kidneys, muscle, bone, lung, stomach, liver, spleen, lymph nodes (superficial cervical, axillary, brachial, renal inguinal and lumbar), bile, small intestine (consisting of the duodenum, jejunum and ileum), contents of the small intestine, large intestine (consisting of the caecum and colon), contents of the large intestine), and faeces were collected, weighed and counted using a g-counter (Packard Instrument, Co.) with a 315-435 keV energy window. Standards of the injected formulation were counted to determine the %ID g À 1 .
Competition binding in vivo to identify the LSEC receptor. Mice (# and~, balb/c, Taconic) were placed into two groups of five mice. Each mouse in one group received an intraperitoneal injection of 0.90 ml of 20 g l À 1 heparin (Sigma, 4180 USP units per mg) in normal sterile saline (NSS) at time 0 and then 30 min later those mice received a IV bolus of 0.10 ml of 20 g l À 1 heparin followed immediately by 0.01 mg of [ 86 Y]fCNT in 0.10 ml via retroorbital sinus IV injection. The mice in the control group (n ¼ 5) received only a 0.01 mg dose of [ 86 Y]fCNT in 0.10 ml via retroorbital sinus IV injection. At 1 h post injection of the radiolabeled fCNT, the mice were euthanized and the blood, liver, spleen and lymph nodes were harvested, weighed, and counted using a g-counter.
Immunofluorescence microscopy. Mice (#, NCr/nu/nu) received 0.01 mg of fCNT in 0.10 ml via retroorbital sinus IV injection. Mice were euthanized at 1, 3, 5, 20, 40, 60, 180 min, 1, 3, 7 or 30 days and the liver, kidneys and spleen collected for IF analyses. Controls included naive tissue (no construct was injected) to determine baseline autofluorescence, hydrolyzed-AF488 dye that was not conjugated to SWCNT (at 1 and 60 min post injection), and isotype-control staining IgG (nonspecific IgG as primary antibody). The immunofluorescent staining was performed in the Molecular Cytology Core Facility of Memorial Sloan-Kettering Cancer Center using Discovery XT processor (Ventana Medical Systems). Double and triple staining experiments were performed sequentially 55 . Specific details of the IF staining, microscopy and imaging are described in the Supplementary Information.
Isolate LSEC and Kupffer cells from mouse liver. LSECs and Kupffer cells were extracted from mouse liver by collagenase perfusion (Liberase TM, Roche), via the portal vein 56 . LSEC were purified using anti-CD146 immunomagnetic beads (Miltenyi Biotec, Bergisch-Gladbach, Germany) and Kupffer cells similarly isolated using an anti-F4/80 biotin and streptavidin immunomagnetic beads (Miltenyi Biotec) 57 . The identity and purity of the two populations was analysed by flow cytometry, using Fluorescein isothiocyanate-labelled check reagents (Miltenyi Biotec), and by PCR analyses. Expression of the Kupffer cell marker Emr1 and LSEC marker Lyve1, along with Stab1 and Stab2 was performed by RT-qPCR.
Liver was examined for apoptosis and immune cell infiltration after fCNT. Murine liver sections (~, balb/c, Taconic) were incubated in primary antibody solutions (rabbit anti-Alexa488 antibody (Molecular Probes) at 0.5 mg ml À 1 ; rabbit anti-CD3 antibody (DAKO) at 1.2 mg ml À 1 ; rabbit anti-Cleaved Caspase-3 antibody (Cell Signalling) at 0.1 mg ml À 1 ; or TdT-biotin-dUTP (Roche) for TUNEL staining. Representative snapshots were taken from the scanned images and analysed for signal intensity or counts, normalized to tissue area, using image analysis software Metamorph (Molecular Devices, PA). Statistics (Student's t-tests) were performed and graphs were made using Prism6 (GraphPad, LaJolla, CA).
In vitro endocytosis assays. CHO cells were stably transfected with empty vector, full-length human Stabilin-1 and Stabilin-2 expression constructs were generated as described 31,58,59 . For endocytosis, fCNT labelled with AF488 were added in concentration 15 mg l À 1 in serum free F12 medium for 30 min, 37°C. AcLDL-AF488 (Life Technologies) in concentration 5 mg l À 1 was used as a positive control for endocytic activity. Quantification of bound/internalized fluorescent ligands was performed with FACSCanto II flow cytometer (BD Biosciences) according to standard protocols. Data were visualized and analysed using FlowJo 7.6.5 software. For detection of fCNT endocytosis by confocal microscopy CHO cells were cultured on coverslips, fixed in PFA and stained using guinea pig anti-Stabilin-1 (clone GP2, self-produced) or mouse anti-Stabilin-2 (clone 3.1) abs as described 31 . fCNT were visualized using rabbit anti-AF488 antibody. Secondary antibodies were donkey anti-guinea pig or donkey anti-mouse labelled with Cy3, and donkey anti-rabbit-AF488 (all from Dianova). Confocal microscopy analysis was performed using Leica TCS SP8 microscope. Data were acquired and analysed with Leica Confocal software. CHO cells were pre-incubated with heparin (1 mg ml À 1 ) in serum free F12 medium for 30 min 37°C followed by addition of fCNT for another 30 min without changing medium. Endocytosed fCNT were detected by flow cytometry as described above.