Research ArticleBioengineering

Predicting therapeutic nanomedicine efficacy using a companion magnetic resonance imaging nanoparticle

See allHide authors and affiliations

Science Translational Medicine  18 Nov 2015:
Vol. 7, Issue 314, pp. 314ra183
DOI: 10.1126/scitranslmed.aac6522

Particle prediction

One particle, it seems, can predict the behavior of another. Thankfully, this is not the beginning of a lesson on quantum physics; instead, it is the basis for potentially designing targeted clinical trials in nanomedicine, by knowing if a tumor is likely to respond to a particular therapeutic nanoparticle. Miller et al. hypothesized that if a tumor readily takes up magnetic nanoparticles (MNP), it will also accumulate other nanoparticles carrying a deadly payload. The authors injected MNPs and a fluorescent version of the therapeutic nanoparticles into mice and followed their biodistribution using imaging. Both types of nanoparticles had similar pharmacokinetics and uptake in tumor-associated host cells owing to the enhanced permeability and retention effect. In mice with human tumors, Miller and colleagues found that the tumors with high MNP uptake were significantly more responsive than those with medium or low uptake to nanoparticles delivering chemotherapeutics. Thus, MNPs can be used as companion imaging agents during nanomedicine trials to predict the therapeutic effect of their nanosized counterparts.


Therapeutic nanoparticles (TNPs) have shown heterogeneous responses in human clinical trials, raising questions of whether imaging should be used to identify patients with a higher likelihood of NP accumulation and thus therapeutic response. Despite extensive debate about the enhanced permeability and retention (EPR) effect in tumors, it is increasingly clear that EPR is extremely variable; yet, little experimental data exist to predict the clinical utility of EPR and its influence on TNP efficacy. We hypothesized that a 30-nm magnetic NP (MNP) in clinical use could predict colocalization of TNPs by magnetic resonance imaging (MRI). To this end, we performed single-cell resolution imaging of fluorescently labeled MNPs and TNPs and studied their intratumoral distribution in mice. MNPs circulated in the tumor microvasculature and demonstrated sustained uptake into cells of the tumor microenvironment within minutes. MNPs could predictably demonstrate areas of colocalization for a model TNP, poly(d,l-lactic-co-glycolic acid)-b-polyethylene glycol (PLGA-PEG), within the tumor microenvironment with >85% accuracy and circulating within the microvasculature with >95% accuracy, despite their markedly different sizes and compositions. Computational analysis of NP transport enabled predictive modeling of TNP distribution based on imaging data and identified key parameters governing intratumoral NP accumulation and macrophage uptake. Finally, MRI accurately predicted initial treatment response and drug accumulation in a preclinical efficacy study using a paclitaxel-encapsulated NP in tumor-bearing mice. These approaches yield valuable insight into the in vivo kinetics of NP distribution and suggest that clinically relevant imaging modalities and agents can be used to select patients with high EPR for treatment with TNPs.


Nanoscale platforms have been developed to improve drug delivery, particularly in oncology, where controlled drug release can mitigate chemotherapeutic toxicities and where structural properties of solid tumors are thought to enhance nanomedicine accumulation (1). Multiple factors including aberrant vascular architecture and basement membrane disruption contribute to the enhanced permeability and retention (EPR) effect, thought to facilitate accumulation of therapeutic nanoparticles (TNPs) (2, 3). Several nanotherapeutics have been clinically approved for treatment of various solid cancers, including liposomal doxorubicin (Myocet), PEGylated liposomal doxorubicin (Doxil and Caelyx), NP albumin-paclitaxel (nab-paclitaxel, Abraxane), and SMANCS [poly(styrene-co-maleic acid)-conjugated neocarzinostatin], whereas others are undergoing clinical trials (4, 5). Many such TNPs have the potential to increase efficacy by enhancing plasma and target tissue drug exposure [area under the curve (AUC)] and/or reduce toxicities by mitigating adverse effects associated with harmful solvents and high-peak drug concentrations (Cmax) of conventional intravenous drug formulations (5).

There is a lack of conclusive data establishing the superior clinical impact of TNPs compared with standard treatments (6), and it is hypothesized that this is largely due to substantial variation in EPR from patient to patient and even across sites within individual patients (3). For instance, modest correlation has been observed between tumor microvasculature and highly variable tumoral accumulation of Caelyx in patients (7), suggesting that EPR factors may substantially contribute to clinical efficacy. Consequently, several treatment strategies aim to therapeutically augment EPR effects in patients for improving nanotherapeutic efficacy, for example, by stimulating vasodilation through heat, nitric oxide induction, and prostaglandins; by stimulating hypertension through angiotensin II; or by degrading extracellular matrix through collagenase (6). Whereas these approaches may potentiate EPR effects, they may also complicate the clinical development of nanotherapeutics. More recently, targeted NPs have entered human clinical trials for small interfering RNA delivery (8) and for small-molecule drug delivery (9). It is expected that these targeted TNPs may improve clinical outcome, in part by directing NP uptake more specifically to tumor cells once reaching the tumor microenvironment (10). Nonetheless, EPR variability continues to be a potential barrier for maximal clinical impact. Thus, one key translational challenge has been to better match patients to novel TNP therapies on the basis of physiological determinants of the EPR effect.

The U.S. Food and Drug Administration (FDA) has approved the carboxymethyl dextran–coated magnetic NP (MNP) ferumoxytol (Feraheme) for treatment of iron deficiency. Ferumoxytol and other related MNPs have been used with magnetic resonance imaging (MRI) to visualize and estimate vascular permeability, NP retention, and phagocyte infiltration in both cancer (11, 12) and inflammation (13). Consequently, ferumoxytol has potential as a quantifier of EPR and thus a means of patient stratification. Despite clinical introduction several years ago and several studies for different indications (11, 14), relatively little is known regarding how these MNPs distribute in different tumor compartments and cell types, how distribution is related to EPR effects, how MNP distribution correlates with TNP distribution, and whether MNP imaging can be used to stratify patients according to preferable tumor uptake of TNPs.

The goal of this study was to understand TNP distribution in vivo and determine whether MNPs can be used as companion particles for predicting therapeutic efficacy. We used high-resolution microscopic imaging in live tumor-bearing mice, which allows single-cell quantification of NP uptake in specific cell populations (tumor versus host) at a resolution superior to MRI (15, 16). Results from these in vivo imaging studies as well as prospective MRI in mice demonstrate the feasibility of using MNPs as surrogate markers of intratumoral nanomedicine transport, particularly by labeling NP circulation in the tumor microvasculature and accumulation in macrophages within the tumor mass. We validate these findings in various orthotopic and syngeneic cancer models and further provide a computational framework to parse measurements for predictive modeling of TNP transport and single-cell uptake in vivo.


Magnetic and therapeutic NPs exhibit related intratumoral pharmacokinetics

Using intravital imaging, we first studied the intratumoral pharmacokinetics (PK) of ferumoxytol to see if this MNP behaved similarly to a model TNP (9, 17, 18). Poly(d,l-lactic-co-glycolic acid)-b-poly(ethylene glycol) (PLGA-PEG) polymeric TNPs are an attractive drug delivery platform for several reasons, including controlled drug release, tunable physical properties, extended plasma half-lives (t1/2), safety, and biodegradability. A fluorescent version of a model PLGA-PEG polymeric NP (λex = 488 nm) (fig. S1) was co-injected with the MNP ferumoxytol-VT680XL (λex = 630 nm) (Fig. 1A), and both NPs were simultaneously tracked in subcutaneous HT1080 human fibrosarcoma xenografts in nude mice (Fig. 1B and fig. S2, A and B). MNPs distributed throughout the entire tumor microcirculation with an initial t1/2 plasma, tumor of 70 min (Fig. 1C), which is consistent with ear imaging measurements in non–tumor-bearing mice (initial t1/2 plasma, ear = 71 min; fig. S3, A and B) and previous studies in non–tumor-bearing rats (initial t1/2 plasma = 67 min) (19) and which would scale to a terminal t1/2 plasma of 10 to 14 hours in humans by allometric predictions (11). The TNPs exhibited similar initial plasma half-lives in both the tumor microvasculature (Fig. 1C; initial t1/2 plasma, tumor = 55 min) and ear vasculature in healthy animals (initial t1/2 plasma, ear = 56 min; fig. S3, A and B), with initial kinetics approximately in the range of other clinically relevant polymeric and liposomal formulations (9, 17) such as PEGylated liposomal doxorubicin [initial t1/2 plasma of 0.8 to 2.2 hours in rats (20, 21) and an initial t1/2 plasma of 5.2 hours in humans for Doxil at 20 mg/m2 (22)].

Fig. 1. High-resolution intravital imaging of ferumoxytol and polymeric NPs show similar intratumoral behavior.

(A) Fluorescently labeled ferumoxytol (MNP) and PLGA-PEG (TNP) were co-injected intravenously into mice for real-time imaging. (B) Time-course measurement of intratumoral MNP and TNP distribution within a live xenograft mouse model of fibrosarcoma transgenically expresses membrane-localized red fluorescent protein/mApple (HT1080-membrane-mApple). Scale bar, 50 μm. (C) PK and tumor tissue uptake were quantified for MNPs and TNPs, normalized to concentration (Ct) as a fraction of initial vascular concentration (C0). Data are means (thick lines) ± SD (shading; n ≥ 7 tumor areas across n = 3 animals). (D) In the same tumor model as in (B) and (C), contrast-enhanced images show perivascular host cells (green) 10 min after NP injection, distinguishable by cellular morphology, perivascular location, lack of tumor-specific mApple, and MNP accumulation but lack of TNP uptake at early time points. Scale bar, 50 μm. (E) Zoomed-in MNP/TNP distribution within a perivascular host cell (arrows). Note the accumulation kinetics within minutes. Scale bar, 14 μm. (F) Perivascular host cells take up MNP more rapidly than TNP. Data are means (thick lines) ± SD (shading; n = 3 tumors; n > 50 cells).

Pixel-by-pixel correlation showed colocalization between MNPs and TNPs, particularly at early time points when the NPs were mostly confined to circulating in the tumor microvasculature, with MNPs successfully labeling >95% of the vasculature accessible to TNPs (Fig. 1B). Single-injection control experiments for each NP confirmed no fluorescence bleed-through (fig. S2, C and D) and demonstrated that MNP injection at the imaging dose does not affect tumor accumulation of subsequently injected TNPs (fig. S3, C and D).

Once reaching the tumor microvasculature, MNPs accumulated rapidly (within minutes) in perivascular host cells that closely neighbored or extended cytoplasmic processes to tumor capillaries (Fig. 1, D and E). Phagocytic perivascular macrophages influence vessel permeability and cancer intravasation during metastasis; thus, to further study these cells in the context of EPR effects, we imaged several metastases of human ovarian cancer after NP administration. MNPs again accumulated within minutes in the metastases (fig. S4). Polymeric TNPs were also taken up by perivascular host cells within the tumor, albeit at a lower level and more slowly (Fig. 1, E and F, and fig. S4). We also imaged MNP distribution in tumor-associated host cells in fractalkine Cx3cr1GFP/+ reporter mice, which have green fluorescent protein (GFP)–positive macrophages. MNPs accumulated exclusively within these GFP+ host leukocytes (fig. S5, A and B). Global expression of membrane-targeted tdTomato in all host cells of the Cx3cr1GFP/+ mice enabled simultaneous visualization of endothelium near tumor xenografts, revealing MNP uptake, especially in host leukocytes adjacent to tumor microvasculature (fig. S5, C and D).

Although the initial plasma kinetic time scale was about 1 hour for both MNPs and TNPs, extended circulating NP half-life and the EPR effect drove gradual accumulation in tumor tissue over the course of 24 hours (fig. S6, A to C). TNP accumulation in tumor cells increased by nearly 20-fold from 3 to 24 hours after administration (fig. S6C), despite the fact that plasma levels had significantly declined by this time (Fig. 1C). An increase in TNP accumulation was also seen in host cells (fig. S6C). These results underscore that intratumoral NP accumulation through EPR effects continued over the course of 24 hours, despite a faster initial phase of plasma clearance.

NPs colocalize with tumor cells at a macroscopic but not single-cell level

Low-magnification tumor imaging showed substantial accumulation and colocalization between NPs and the bulk tumor mass 24 hours after treatment (Fig. 2A), which likely explains MRI observations of enhanced MNP accumulation in various cancers (2, 6, 14). At the cellular level, although host cells within the tumor microenvironment substantially accumulated both NPs (Fig. 2A), tumor cell uptake was considerably lower and slower (Fig. 1F): 3 hours after injection, >90% of both NPs were associated with host cells rather than tumor cells (fig. S6C). Although MNPs underestimate the amount of TNPs taken up by tumor cells, they do label host cells, such as leukocytes, that accumulate TNPs with >85% accuracy (fig. S6C). An orthotopic model of disseminated metastatic ovarian cancer showed similar patterns of colocalization and predominant accumulation in host rather than tumor cells (fig. S6, D and E). MNPs also colocalized with a model phosphatidylcholine/cholesterol liposomal formulation (mean diameter ± SEM, 171 ± 15 nm; n = 9) in tumor-associated host cells (fig. S7A). There was no significant difference in spatial colocalization according to pixel-by-pixel correlation when using five alternative fluorophores (fig. S7B).

Fig. 2. Multiscale spatial colocalization between MNP and TNP.

(A and B) NP tumor uptake in a live xenograft model, 24 hours after injection with MNP. For low (×4) and high (×40) magnification, scale bar denotes 500 and 50 μm, respectively. Intravenous co-injection with either TNP (A) or a free, unencapsulated fluorescent derivative of docetaxel (B) was imaged after vascular clearance. (C) MNP and TNP colocalization improves at lower spatial resolution, but MNP and docetaxel colocalization does not. Microscopy images (A and B) were computationally down-sampled to reduce spatial resolution, and pixel-by-pixel Pearson’s correlations (ρ) between MNP/TNP and MNP/docetaxel intensities were calculated across a range of pixel resolutions. Data are means (thick lines) ± SE (n = 3 animals and >400 images).

We next computationally modeled how our high-resolution microscopy results would apply to more clinically relevant imaging modalities, such as MRI, which has substantially lower spatial resolution. We decreased the microscopy resolution by down-sampling and then calculated the correlation between MNP and TNP fluorescence intensities as they varied from pixel to pixel. Although MNPs and TNPs exhibited modest high-resolution correlation (ρ = 0.2) 24 hours after injection, correlation increased nearly threefold (ρ = 0.55) at spatial resolutions typical of clinical MRI (Fig. 2C). This trend was not evident when comparing MNPs to the spatial distribution of free docetaxel, which was not encapsulated in an NP, demonstrating that increased MNP/TNP colocalization at lower spatial resolution is not simply an artifact of all injected compounds (Fig. 2, B and C). Overall, these data show that EPR effects, largely influenced by NP uptake in host cells, contribute to selective MNP and TNP accumulation within the bulk tumor mass, and suggest that imaging MNPs at a lower MRI resolution will still be able to predict TNP accumulation.

MNPs and TNPs are primarily taken up by tumor-associated macrophages

We next used flow cytometry and histology to quantitatively map NP distribution to immunologically defined cell populations within the bulk tumor mass. To better understand tumor interactions with the immune system, we used a syngeneic immunocompetent model of non–small cell lung cancer based on the subcutaneous implantation of Kras mutant p53−/− (KP) cells derived from autochthonous lung tumors in a genetically engineered mouse model (23). Leukocytes (CD45+ cells) comprised about one-third of all cells in the tumor (Fig. 3A). Similar to our observations in xenograft models of fibrosarcoma (Fig. 1 and fig. S6, B and C) and ovarian cancer (fig. S6D), host phagocytes (macrophages and neutrophils) accumulated more MNPs and TNPs than did tumor cells (Fig. 3, B to F). Although phagocytosis of tumor cells by host macrophages may complicate both flow cytometric and imaging analyses, imaging data suggest that this population represents <5% of all cells analyzed and therefore has minimal impact on the median-based statistics calculated here. CD45 host cell populations, which include endothelial cells and tumor-associated fibroblasts, did not accumulate substantial levels of any NPs (Fig. 3B), consistent with a report that fibroblasts limit rather than enhance intratumoral NP accumulation (24).

Fig. 3. MNP and TNP colocalize to tumor-associated macrophages in a syngeneic cancer model.

(A and B) Flow cytometric analysis of intratumoral cellular composition (A) and single-cell NP distribution (B) in KP subcutaneous xenografts cotreated with TNP and MNP for 24 hours. Cellular NP uptake was quantified by fluorescence intensity after subtracting the autofluorescence of each population and normalized to the highest average NP uptake (macrophage). (C) Cumulative NP uptake across total cell populations within the bulk tumor mass, normalized such that NP uptake across all cell populations, weighted by their relative frequency, sums to 1. Data are means ± SEM (n = 12). (D and E) KP xenografts were excised 24 hours after MNP and TNP cotreatment, stained with hematoxylin and F4/80 (brown), and imaged at ×10 (D) and ×40 (E) magnification. Scale bars, 100 μm (D); 50 μm (E). (F) Adjacent tumor sections were immunostained for EpCAM to label tumor cells and for F4/80 to label macrophages.

NP spatiotemporal mapping quantifies EPR effects

Computational modeling was used to quantify the observed kinetic processes involved in intratumoral NP accumulation and retention; to compare differences among NPs; to predict in vivo behavior in different tumor models; and to rank the relative contributions of individual parameters, such as vessel permeability, macrophage content, cellular uptake rates, and interstitial diffusion. We used an approach based on finite-element analysis that incorporated spatial NP diffusion and heterogeneous NP uptake at the single-cell level (Fig. 4A). Reaction/diffusion parameters were computationally inferred for each type of NP (table S1), and comparison of these parameter sets for each NP allowed for the derivation of a quantitative normalization factor that corrected for differences in their kinetics.

Fig. 4. Quantitative finite-element analysis describes single-cell reaction-diffusion processes of the EPR effect.

(A) Overview of computational modeling and optimization. (i) In HT1080 xenografts, automated morphological criteria identify vessels (green/red masking), and early MNP accumulation (white) identifies macrophage (as in Fig. 1D), which were computationally segmented with manual optimization. (ii) The finite-element mesh was generated on the basis of image segmentation. (iii) Change in concentration over time of free NP (dC/dt) and bound NP (dB/dt), along with boundary conditions describing NP flux across vessel walls [D(dC/dr)], and vessel NP concentrations over time (CP) were integrated across the finite-element mesh. (iv) Parameters were iteratively optimized by fitting model results to time-lapse imaging data. PDE, partial differential equation. (B) Model-fitting validation (green and yellow bars) and spatial correlation between MNPs and TNPs, with and without nonlinear PK correction based on finite-element modeling (gray and black bars). Correlation data are means ± SEM (n > 200 regions; P value determined by permutation test). (C) Parametric sensitivity analysis showing modeling parameters that most sensitively influence total NP accumulation within the bulk tumor at 2 hours after injection. Bmax, maximum NP cellular uptake; kbind, NP uptake rate; P, vessel permeability. Data are medians ± interquartile range (IQR) (n = 5; *P = 0.01, pooled two-tailed t test). (D) Example images and corresponding modeling show heterogeneous MNP accumulation in tumor regions with few (n = 18; left) and many (n = 98; right) phagocytes. Scale bars, 50 μm. (E) Finite-element modeling predicted increased MNP accumulation in the high-TAM tumor region (D), measured as average MNP concentration in tumor tissue outside of vessels.

Nonlinear PK correction, when applied to MNP images, improved the spatial correlation between MNPs and TNPs by ~300% and thus greatly increased the accuracy of MNPs in predicting intratumoral TNP levels (Fig. 4B). The upper accuracy limit for the computational framework was determined by measuring how well the model fit the original training image data set (Fig. 4B, green and yellow bars); encouragingly, the nonlinear PK correction enabled the spatial correlation between MNPs and TNPs to reach this limit. Thus, although MNP and TNP kinetics differed, they overlapped in spatial distribution, particularly among host phagocytes.

To assess the relative importance of different EPR factors in intratumoral NP accumulation, we performed a parametric sensitivity analysis for each NP type by locally adjusting individual modeling parameters (±25%), simulating NP behavior with the new parameter sets, and recording the resulting impact on bulk tumor NP accumulation (including tumor cells and host phagocytes but excluding vasculature) at 2 hours after injection. This analysis revealed that extracellular volume fraction in the tissue, ε, and systemic plasma half-life of the NPs, t1/2 plasma, were the two most important factors governing tumor uptake 2 hours after injection (Fig. 4C), suggesting that cellular uptake was limited at this early time point.

Because MNPs and TNPs have multiple distinct parameters that interact with each other, local changes in reaction rates can affect accumulation of each type of NP differently. MNPs were highly sensitive to macrophage uptake capacity (Bmax,,MΦ), kinetics (kbind,,MΦ), and density (macrophages per tumor tissue area), whereas TNPs were not, largely because MNP uptake far outstripped TNP uptake at the early time point of 2 hours (Fig. 4C). Macrophage density has been previously reported as varying widely across tumor types and patients, often correlating with clinical outcome (25). There was heterogeneity in macrophage density even within different regions of single tumors [Fig. 4D, top images; coefficient of variation (CV), 100% across 6 tumors]. We independently tested the computational model on such heterogeneous regions and accurately captured the significant effects of variable macrophage density on NP accumulation (Fig. 4, D and E).

MRI before treatment predicts extent of tumor cell DNA damage after TNP administration

To test whether MNP MRI would help to select animals for preclinical trials, we performed several prospective experiments to mimic a clinical scenario. A cohort of mice bearing subcutaneous human fibrosarcoma (HT1080) tumors was imaged by MRI before and after intravenous administration of ferumoxytol MNP to measure total tumor accumulation. MRI at 1 hour after ferumoxytol administration was used to assess tumor microvascularization accessible to circulating TNPs, and MRI at 24 hours labeled cellular uptake within the bulk tumor mass. Change in average T2 mapping of the tumor region (ΔT2) between 1 and 24 hours indicated highly heterogeneous MNP accumulation across the cohort (CV, 750%). We used this metric to stratify mice into “low,” “medium,” and “high” MNP categories (Fig. 5, A and B).

Fig. 5. MRI quantifies heterogeneous MNP accumulation and predicts initial TNP response.

(A) Representative cross-sectional T2 images of HT1080 tumors accumulating low and high intratumoral MNP, with pseudocolor overlays indicating ΔT2 within the tumor region. (B) Heat map shows T2 mapping averaged over the entire area of each tumor, which stratified a subset of tumors as low, medium, or high MNP. (C) Experimental design for using MNP MRI to predict paclitaxel (PTX)–loaded TNP response in HT1080 xenografts. s.c., subcutaneous; i.v., intravenous. (D) Fluorescence-activated cell sorting (FACS) analysis shows drug response in tumors exhibiting either low, medium, or high MNP uptake, as determined by MRI. Tumors with high MNP showed greater populations with abnormally low DNA content (sub-G1 cells) and elevated heterogeneous levels of DNA damage response, determined by deviation (σ) in γH2A.X staining across the population. Data are medians ± IQR (two-tailed t test). a.u., arbitrary units. (E) Representative FACS data for DNA content (using a DNA-intercalating dye) and DNA damage response (using γH2A.X immunostaining) in low-MNP and high-MNP tumors. (F) Representative FACS data for DNA content (using a DNA-intercalating dye) and apoptosis (TUNEL staining) in tumor cells from a high-MNP tumor. Contour lines and colors (E and F) denote single-cell distribution density.

Paclitaxel-loaded TNPs (fig. S8, A to C) were then administered, with a paclitaxel dose of 3 mg/kg, and tumors were analyzed for subsequent treatment responses (Fig. 5C). The different imaging groups had vastly different therapeutic responses, measured by cell cycle distribution and DNA damage (Fig. 5, D to F). Compared to tumors with medium MNP uptake, we found that high-MNP tumors contained substantially more cells with abnormally low DNA content (“sub-G1” population) and elevated DNA damage response, as measured by γH2A.X staining (Fig. 5, D and E). These sub-G1 cells were mostly viable, with only 2 to 10% (IQR over all tumors) of the sub-G1 cells being apoptotic (Fig. 5F), and have been implicated as a key feature in paclitaxel action (26). There was no significant difference in apoptosis between low-MNP and high-MNP tumors (P = 0.3, one-way analysis of variance). In contrast, low-MNP tumors exhibited substantially lower levels of DNA damage response compared to the medium-MNP tumors (Fig. 5, D and E). These data indicate that MRI can preliminarily stratify tumors by response to TNP treatment.

MNP distribution predicts disease progression and TNP accumulation

MNP MRI could be used to predict longitudinal disease progression after TNP treatment. A total of 33 subcutaneous HT1080 tumors were imaged before and after ferumoxytol administration to quantify MNP uptake. To induce greater heterogeneity of tumoral MNP concentration (similar to what is observed clinically), half of the cohort was pretreated with systemic liposomal clodronate to reduce tumor-associated macrophages (TAMs) before NP administration and MRI. As done in a previous experiment in Fig. 5, MRI of MNP uptake was quantified by calculating ΔT2 between 1 and 24 hours after administration of MNP. Tumors were stratified into low-, medium-, and high-MNP groups; paclitaxel-encapsulated TNPs (3 mg/kg) were administered intravenously immediately after MNP MRI; and tumor sizes were measured daily until excised for further analysis. MNP accumulation predicted tumor growth after TNP treatment: the low-MNP tumors grew twofold faster than the medium-MNP tumors, whereas the high-MNP tumors did not increase in size (Fig. 6A).

Fig. 6. MNP predicts longitudinal TNP response and accumulation of TNP payload.

(A) Tumor progression in HT1080 tumors ranked according to low, medium, and high MNP as measured by MNP MRI. Data are means ± SEM (total n = 33). (B) In orthotopic 4T1 breast cancer tumors, MNP prediction of TNP-encapsulated docetaxel, as determined by fluorescence of excised tumors 1 day after MNP/TNP injection. (C) MNP prediction of unencapsulated solvent-docetaxel accumulation as determined by fluorometry, using the same tumor model as in (B). (D) EGFR expression–based prediction of TNP-encapsulated docetaxel accumulation, using the same tumor model as in (B). TNP-docetaxel was co-injected with fluorescent tumor-targeting (α-EGFR) antibody, which stratified tumors into low, medium, or high expression groups. In all graphs, data are medians ± IQR. P values were determined by two-tailed t tests. n.s., not significant.

We next investigated the degree to which MNPs predict accumulation of the TNP chemotherapeutic payload itself. For these experiments, we used a syngeneic, orthotopic model of invasive breast cancer consisting of 4T1 mouse mammary carcinoma cells implanted into the mammary fat pads of immunocompetent BALB/c mice. To quantitatively and sensitively detect drug accumulation within tumors, we loaded TNPs with a fluorescent docetaxel derivative (fig. S8D). MNPs and docetaxel-encapsulated TNPs were administered intravenously, and 20 hours later, tumors were excised and analyzed for TNP accumulation. Tumor MNP levels correlated significantly with accumulation of the TNP therapeutic payload, such that high-MNP tumors exhibited about 25-fold higher levels of docetaxel compared to low-MNP tumors (Fig. 6B).

Using the same 4T1 tumor model, we performed a control experiment to determine how effectively MNPs predict tumoral accumulation of free docetaxel. For this control, we used a silicon-rhodamine fluorophore to generate a spectrally distinguishable fluorescent docetaxel derivative that exhibits similar PK properties as the BODIPY-labeled drug (fig. S8E). After MNP and unencapsulated solvent-based docetaxel injection, tumors were excised and analyzed for drug and MNP uptake. Although MNPs somewhat predicted the accumulation of unencapsulated solvent-based docetaxel, the difference between the low-MNP and high-MNP tumors was more modest (2.8-fold) (Fig. 6C) compared to when docetaxel was encapsulated in TNPs (Fig. 6B).

As another control experiment, we investigated whether intratumoral TNP could be accurately stratified by accumulation of an antibody targeting epidermal growth factor receptor (EGFR), which has been used for imaging tumor burden, selecting treatments, and monitoring response in patients with EGFR-overexpressing cancers, including those receiving docetaxel and other chemotherapeutics (27). With the EGFR-expressing 4T1 tumor model (28), we stratified excised tumors into low, medium, and high antibody groups. Although the anti-EGFR antibody could predict docetaxel TNP accumulation for high versus low and high versus medium groups, the difference between low and high was modest (3.3-fold) (Fig. 6D). The greater order of magnitude of effect for correlation between MNPs and TNPs suggests that NP-specific EPR factors play a dominant role in governing heterogeneous TNP tumor accumulation. The poor correlation between antibody labeling (EGFR expression) and TNP accumulation imply that the anti-EGFR antibody is a poor predictor of tumor response to TNP treatment. Together, these results demonstrate that MNPs are effective predictors of TNP payload accumulation within tumors and, in turn, therapeutic efficacy.


TNPs distribute differently across tumors in different patients. A central question in nanomedicine is whether imaging could be used to identify patients with higher predisposition to TNP accumulation and, in turn, efficacy (5, 9). Answering this question could aid in the decision of whether to actively target TNPs or to let them accumulate “passively” within a tumor (3). These issues are at the core of understanding how to best exploit EPR effects (6) for clinical applications, how to design better TNPs, and how to alter key physiologic parameters to maximize distributions to and within tumors.

We hypothesized that an FDA-approved carboxymethyl dextran–coated MNP (ferumoxytol) can be used to predict TNP behavior by measuring its intratumoral distribution and kinetics across different tumor compartments. To define these compartments, we relied on intravital imaging capable of resolving intracellular details (15). Our study uncovered several findings, paving the way for companion particles in predictive nanomedicine. MNPs and TNPs, despite being of different sizes and composition, colocalized to a high degree, especially in the circulating vascular phase, at the macroscopic level (that is, resolutions used in MRI), and in phagocytic host cells. MNP accumulation within the bulk tumor was significantly influenced by host phagocyte content and surprisingly rapid peritumoral host cell uptake within minutes. For both NPs, tumor cell uptake was much slower than expected but was greater for TNPs than MNPs despite the larger size of the TNP. For translation to a therapeutic setting, the heterogeneity of intratumoral TNP accumulation could be predicted and measured by MRI and correlated with responsiveness to TNP treatment. In contrast, intratumoral MNP only modestly correlated with soluble therapeutic accumulation, and a common tumor-targeting antibody (anti-EGFR) performed poorly in predicting TNP accumulation. Thus, MNP imaging effectively captured EPR factors that concordantly governed both NPs yet did not equally extend to solvent-based formulations or antibodies.

Understanding individual EPR factors is essential for predicting and eventually tailoring NP behavior. EPR effects of vascular permeability have been studied extensively in the past (2, 3, 24, 29); however, these studies have been limited by spatiotemporal resolution and do not sufficiently account for heterogeneous behavior at the single-cell level, particularly in TAMs. We not only studied the multiple EPR factors but also used high-resolution time-lapse microscopy to reveal their highly dynamic character. Vascularization and permeability drove PK at early time points (t < t1/2 plasma) for both NPs, and rapid cellular uptake contributed to early EPR effects for MNPs, particularly in perivascular regions where capillary-associated phagocytes accumulated MNPs within minutes. TNP uptake was minimal at this early stage. In contrast, after 24 hours (t >> t1/2 plasma), EPR effects were dominated by cellular NP uptake, particularly in TAMs.

TNPs are typically engineered to release chemotherapeutic payloads at prescribed and tunable rates. The relative therapeutic importance of tumor vascularity, vessel permeability, interstitial fluid content and pressure, extracellular matrix, and phagocyte infiltration will ultimately depend on TNP physicochemical properties including size, shape, payload release kinetics, and transport properties of the released drug. When keeping the TNP physicochemical properties constant, as done in our study, both slow and rapid payload release will influence intratumoral distribution: cellular NP uptake will dominate in the former, whereas vascular permeability and extracellular volume fraction will dominate in the latter. Although we did not extensively investigate different TNP formulations, our study lays the groundwork for understanding how heterogeneous cell populations and the EPR within the tumor microenvironment affect drug delivery, and demonstrates that MNPs may be used to predict the behavior of NPs with differing physicochemical properties.

We used computational modeling to provide a framework for quantifying and comparing physiological effects that govern tumoral NP accumulation. This work extends previous modeling of drug transport (15) to nanotherapeutics, which required the explicit modeling of NP uptake at single-cell resolution. The largest differences in NP behavior were all related to heterogeneous cellular NP uptake. For macrophages, the MNP uptake rate was about sixfold faster than for TNP, and maximum uptake levels were threefold higher, which is not surprising considering the differences in surface modifications between MNPs and TNPs. Future studies should extend this approach to better capture three-dimensional behavior; link NP kinetics with drug payload release and cellular response; more closely examine differences in tumoral penetration between MNPs and TNPs; and address additional heterogeneity in parameters related to interstitial pressure, convection, diffusion through fibrotic tissue, and pH.

Several of our findings have direct implications for the effective design of nanotherapeutic clinical trials. Heterogeneous tumor vascularization is a recognized clinical feature that can be detected using various angiography modalities and that can affect both drug delivery (29) and overall survival (30). Immediate MNP MRI showed the tumor vasculature accessible to TNPs, which will be especially important in the context of therapeutics that affect vascular structure, such as targeted antiangiogenics, vascular dilators, and hyperthermic induction. MNP imaging after 24 hours revealed peritumoral cell populations that accumulated high levels of TNPs. Macrophage content represents a critical yet highly variable component of the overall EPR effect. Recent work has highlighted the extensive heterogeneity of phagocyte populations within tumors and its significant effects on drug response and clinical outcome (25). Moreover, several therapeutics directly target TAMs, for instance, by blocking colony-stimulating factor 1 receptor; therefore, assessing phagocyte content will be especially important for selecting patients to receive these drugs and to monitor their response.

To progress clinically, more human-representative disease models, such as patient-derived xenografts, genetically engineered autochthonous mouse models, and larger animals, should be used to study EPR effects in metastatic lesions, to examine the correlation between MNP uptake and long-term efficacy, and to see if MNPs can indicate which cancer types may be more responsive to nano-based drug delivery. Ultimately, we have provided a high-resolution description of dynamic EPR effects governing nanomedicine transport and demonstrated the potential of MNPs and imaging to predict TNP efficacy in the clinic—a major stepping stone toward translation of new nanomedicine and for eventually selecting patients for nanotherapeutic trials.


Study design

The hypothesis was that MNP would predict accumulation and efficacy of TNP in tumors. Imaging studies and drug accumulation measurements were designed to measure the kinetics and colocalization between MNP and TNP within tumor tissue; tumor caliper measurements, along with flow cytometry measurement of DNA damage, cell cycle, and apoptosis were designed to assess the correlation between MNP accumulation and drug response efficacy. All experiments were performed with at least two independent replicates (specified in figure legends). Data collection methods were predetermined for all experiments, and animals were assigned randomly to treatment groups. No outliers were excluded. Cohort sizes in the experiments using the 4T1 model (Fig. 6) were informed by a power analysis based on MRI data in the HT1080 model (Fig. 5), using the measured intragroup heterogeneity in MNP uptake (CV, 60%), an intergroup difference of 80% between low- and high-MNP groups, and an objective power (1 – β) of 0.95 calculated with available software (31). Researchers were blinded to groups for MNP and anti-EGFR antibody uptake during nonimaging experiments in Figs. 5 and 6, including caliper measurement, because prespecified stratification procedures were not performed until after all data had been collected.

TNP synthesis

All polymeric TNPs were synthesized by nanoprecipitation, were characterized by size and surface charge using dynamic light scattering (Malvern Zetasizer), and were freshly prepared before each experiment. Synthesis details are given in Supplementary Methods.

Animal and cell models

All animal research was performed in accordance with the guidelines from the Institutional Subcommittee on Research Animal Care. All experiments were performed using female mice that were 5 to 7 weeks old at the start of the experiment. For experiments with human HT1080 tumors, 2 × 106 cells were subcutaneously implanted into nu/nu mice; 2 to 3 weeks later (once tumors reached about 8 mm in diameter), imaging experiments were initiated. For 4T1 tumors, 0.5 million cells were implanted into the mammary fat pads of BALB/c mice; about 10 days later, imaging and therapeutic agents were intravenously injected, and tumors were excised for analysis the following day. For ovarian cancer imaging, 107 A2780CP cells were injected intraperitoneally into nu/nu mice, and experiments were performed about 6 weeks later with evident ascites or tumor masses. For imaging experiments using KP1.9 cells, 106 cells were subcutaneously implanted into C57Bl/6 background animals (all JAX), including Cx3cr1GFP/+ and Cx3cr1GFP/+ R26mT-mG/+ dual-reporter mice, both containing GFP+ monocytes, macrophages, and dendritic cells. The mouse and human cell lines HT1080, A2780CP, 4T1, and KP1.9 are described in Supplementary Methods.

Intravital microscopic imaging

Intravital microscopy was performed on an Olympus FV1000 confocal multiphoton imaging system using a XLUMPLFLN 20× water immersion objective (numerical aperture, 1.0; Olympus America) with 2× digital zoom. Images were scanned sequentially using 405-, 473-, 559-, and 635-nm diode lasers with a DM405/473/559/635-nm dichroic beam splitter; emitted light was collected using combinations of beam splitters (SDM473, SDM560, and/or SDM 640) and emission filters BA430-455, BA490-540, BA575-620, and BA655-755 (all Olympus America).

Dorsal window chamber imaging was performed following previously described procedures (16); briefly, 2 million HT1080-membrane-mApple cells in 50 μl of phosphate-buffered saline (PBS) were injected under the fascia of nu/nu mice (Cox7, Massachusetts General Hospital) 30 min after surgical chamber implantation and were imaged 2 weeks later.

Magnetic resonance imaging

Mice were anesthetized by isoflurane inhalation and placed in a birdcage radio frequency coil with an inner diameter of 38 mm. Mice were scanned for a baseline T2 value before MNP injection, using a 4.7-T MRI system (PharmaScan, Bruker BioSpin). Without removal of the mouse from the coil, mice were scanned after intravenous MNP injection (20 mg Fe/kg body weight), and a third scan was performed 24 hours later. T2 values were calculated by fitting of a standard exponential relaxation model to the data averaged over the tumor region of interest on each slice, using Osirix software. The quantitative probe accumulation was calculated as follows: [ln (T21h post-MNP/T224h post-MNP)], where T21h post-MNP indicates the T2 value 1 hour after MNP injection (vascular phase) and T224h post-MNP indicates the T2 value 24 hours after MNP injection (cell accumulation phase).

Computational image analysis

Intravital microscopy images were analyzed using either MATLAB (MathWorks) or ImageJ and were preprocessed using background subtraction based on data acquired immediately before NP injection. Vascular half-life calculations, finite-element analysis, and other details are described in Supplementary Methods.

MNP prediction of drug uptake, response, and progression

Three-week-old subcutaneous HT1080 tumors were imaged before and 1 and 24 hours after intravenous ferumoxytol injection, once tumors reached an average diameter of 8 mm. To generate heterogeneity in EPR effects (Fig. 6A), tumors were evenly split into two groups of equally distributed tumor sizes, receiving either clodronate liposomes (5 mg/ml) or PBS liposomes as a vehicle control, with 150 μl administered intraperitoneally 3 days before imaging and 100 μl administered intravenously 24 hours before imaging (ClodLip BV). Tumoral MNP uptake was quantified as [T21h post-MNP − T224h post-MNP] and weighted by MNP plasma half-life for each treatment group to control for residual MNP in circulation (32). To measure tumor volume (V = 4/3πr3), caliper measurements were performed by two blinded researchers. Results were categorized into three groups defined by boundaries that maximized the statistical significance in differential drug response and tumor progression as measured by DNA damage response or tumor volume, respectively. Magnetic resonance images (Fig. 5A) were selected on the basis of their representativeness for each group and also for tumors being of roughly equal anatomical position and size. For correlation analysis between tumor uptake of MNPs, TNP, free drug, and EGFR-targeting antibody, we used the orthotopic 4T1 breast cancer syngeneic mouse model, as described in Supplementary Methods.

Statistical analysis

Statistical analyses were performed using GraphPad Prism, MATLAB, and Microsoft Excel. Measurement uncertainties throughout are denoted by error bars and shading as indicated in figure legends. All statistical tests were two-tailed with testing level thresholds of α = 0.05. As a predefined procedure, stratifications by MNP or antibody uptake (Figs. 5 and 6) were performed using thresholds that maximized the statistical difference (using t tests) in drug response, tumor progression, or docetaxel uptake.



Fig. S1. Fluorescent NP synthesis and characterization.

Fig. S2. Imaging single-cell kinetics of MNP distribution with no TNP/MNP fluorescence spectral bleed-through.

Fig. S3. MNP matches TNP plasma kinetics and does not influence TNP uptake.

Fig. S4. MNP and TNP uptake by perivascular host cells in ovarian cancer.

Fig. S5. MNP uptake in tumor-associated Cx3cr1+ host cells.

Fig. S6. Imaging cytometric analysis of single-cell NP distribution kinetics.

Fig. S7. MNPs colocalize with liposomes in tumor-associated cells.

Fig. S8. Characterization of paclitaxel-loaded TNP and fluorescent derivatives.

Table S1. Optimized finite-element method modeling parameters and reference values.

References (3338)


Acknowledgments: We thank A. Zaltsman, D. Pirovich, M. Sebas, O. Kister, N. Sergeyev, K. King, and Y. Iwamoto for technical assistance; S. Lippard and Y. Zheng for helpful discussions. Funding: This work was supported in part by the NIH (R01CA164448, U54-CA151884, 5P50CA086355, and HL084312), DoD (PC140318), and the David H. Koch–Prostate Cancer Foundation Award in Nanotherapeutics. M.A.M. was supported by T32 CA 79443. C.P. is in part supported by Deutsche Forschungsgemeinschaft (DFG) PF809/1-1. Author contributions: M.A.M., S.G., O.C.F., and R.W. developed the concept; M.A.M., S.G., C.P., M.J.P., O.C.F., and R.W. designed the experiments; M.A.M., S.G., N.K., and S.B. synthesized and characterized PLGA-PEG NPs; M.A.M. synthesized and characterized MNP, docetaxel-TNP, and antibody-dye conjugate; M.A.M., C.P., and C.E. performed FACS experiments; M.A.M. and R.H.K. performed intravital imaging experiments; M.A.M., G.W., O.C.F., and R.W. designed or performed therapeutic MRI experiments; M.M.S. designed and synthesized docetaxel-dye conjugates; M.A.M., A.M.L., and K.S.Y. generated the cell lines; M.A.M., S.G., O.C.F., and R.W. wrote the paper; all authors analyzed the results and edited the manuscript. Competing interests: In compliance with institutional guidelines, O.C.F. discloses his financial interest in BIND Therapeutics, Selecta Biosciences, and Blend Therapeutics, which develop NP medical technologies but did not support this study. The other authors declare that they have no competing interests. Data and materials availability: All cell lines were obtained through material transfer agreements. Requests for collaboration involving materials used in this research will be fulfilled provided that a written agreement is executed in advance between Brigham and Women’s Hospital or Massachusetts General Hospital and the requesting parties.
View Abstract

Navigate This Article