Imaging angiogenesis in an intracerebrally induced model of brain macrometastasis using αvβ3‐targeted iron oxide microparticles

Abstract Brain metastasis is responsible for a large proportion of cancer mortality, and there are currently no effective treatments. Moreover, the impact of treatments, particularly antiangiogenic therapeutics, is difficult to ascertain using current magnetic resonance imaging (MRI) methods. Imaging of the angiogenic vasculature has been successfully carried out in solid tumours using microparticles of iron oxide (MPIO) conjugated to a Arg‐Gly‐Asp peptide (RGD) targeting integrin αvβ3. The aim of this study was to determine whether RGD‐MPIO could be used to identify angiogenic blood vessels in brain metastases in vivo. A mouse model of intracerebrally implanted brain macrometastasis was established through intracerebral injection of 4T1‐GFP cells. T2*‐weighted imaging was used to visualise MPIO‐induced hypointense voxels in vivo, and Prussian blue staining was used to visualise MPIO and endogenous iron histologically ex vivo. The RGD‐MPIO showed target‐specific binding in vivo, but the sensitivity of the agent for visualising angiogenic vessels per se was reduced by the presence of endogenous iron‐laden macrophages in larger metastases, resulting in pre‐existing hypointense areas within the tumour. Further, our data suggest that peptide‐targeted MPIO, but not antibody‐targeted MPIO, are taken up by perivascular macrophages within the macrometastatic microenvironment, resulting in additional nonspecific contrast. While pre‐MPIO imaging will circumvent the issues surrounding pre‐existing hypointensities and enable detection of specific contrast, our preliminary findings suggest that the use of antibodies rather than peptides as the targeting ligand may represent a preferable route forward for new angiogenesis‐targeted molecular MRI agents.


| INTRODUCTION
Metastasis, or secondary spread from a primary tumour to distant sites, is one of the enduring challenges in the treatment of cancer. The detection and treatment of brain metastases remains an important clinical challenge. Thus, there is a need for the development of targeted therapies to treat brain metastases, together with more accurate strategies for measuring treatment response. Anti-angiogenic agents, predominantly targeting VEGF and VEGFR2, have been and are currently being evaluated in clinical trials for the treatment of human brain metastasis. 1 Angiogenesis is crucial for tumours to grow beyond a few millimetres in size 2 and, thus, represents an important therapeutic target.
Despite showing great promise in initial preclinical studies, the use of antiangiogenic drugs in the clinic has had mixed results. There are many reasons for these discrepancies, including an inability to stratify patients by angiogenesis levels, an inability to monitor treatment response using conventional imaging, and the activation of alternative signalling pathways mediating drug resistance. Angiogenesis is mediated by many different factors and pathways, one of which is the integrin α v β 3 . Also known as the vitronectin receptor, integrin α v β 3 is expressed at high levels on angiogenic endothelial cells 3,4 and some types of tumour cells, [5][6][7] but only at very low levels on the normal vasculature. Integrin α v β 3 has a specific binding site in the peptide motif Arg-Gly-Asp (RGD) 8 and, consequently, RGD peptides are used in a wide range of blocking and targeting applications, including small molecule drugs and molecular imaging. Antiangiogenic drugs targeting α v β 3 such as Cilengitide and Etaracizumab have shown promise in preclinical studies, but have not been successful in clinical trials, particularly in the brain and in metastatic cancer. [9][10][11][12][13][14][15] It is important to note that none of these clinical trials have attempted to stratify patients by α v β 3 expression levels, which might be expected to predict response. In the only trial that attempted to image α v β 3 using radiolabelled Etaracizumab, the imaging was unsuccessful, with only subtle localisation to the tumour bed, which the authors proposed was caused by the instability of 99m Tc labelling. 12 These findings highlight the need for sensitive, specific, and reproducible clinical imaging of the angiogenic vasculature in tumours, and brain metastases in particular.
Molecular MRI strategies that have previously been used to image angiogenesis include the conjugation of anti-α v β 3 antibodies or various RGD peptides to paramagnetic gadolinium-containing liposomes, [16][17][18][19][20] perfluorocarbon-based nanoparticles, [21][22][23][24] or lipoprotein cores. 25 However, the clearance rates, specificity, and signal-to-noise ratio (SNR) of these particles remained low, leading to the investigation of targeted iron oxide-based agents. Targeting of ultrasmall superparamagnetic iron oxide nanoparticles (USPIO) to α v β 3 has been reported, and in some studies with sufficient SNR to be detected on a clinical scanner. 26,27 In preclinical studies, this approach has been suggested to enable differentiation between integrin α v β 3 expressing angiogenic blood vessels and integrin α v β 3 overexpressing tumour cells, 26 as well as prediction or measurement of response to treatment. 28,29 Despite the increased sensitivity compared with gadolinium-based molecular imaging agents, however, the propensity of these nanoparticles to extravasate into tissue remains a significant confound, as the contrast produced is no longer specific to their endothelial target. [30][31][32][33] Moreover, nanosized particles have a relatively long half-life in the circulation ($16.5 h), which is a limitation to imaging of target-specific binding. 33 By contrast, microparticles of iron oxide (MPIO) are significantly larger than USPIO, exhibit a much shorter blood half-life ($1 min), [33][34][35] and are obligate intravascular agents. Consequently, MPIO have major advantages over USPIO for molecular imaging of endovascular targets, with greater contrast effects because of their higher iron content, and high sensitivity for target-specific binding owing to an absence of background contrast effects from circulating particles. For this reason, we have previously developed a cyclic RGD conjugated MPIO construct (RGD-MPIO) and have demonstrated its potential for imaging angiogenic tumour vasculature in solid tumours. 35 Thus, to address the unmet need for a specific and sensitive method for monitoring tumour angiogenesis, the primary aim of the current study was to determine whether RGD-MPIO in conjunction with MRI enables detection of angiogenic vasculature in an intracerebrally implanted model of brain macrometastases.

| Preparation of peptide and antibody-conjugated MPIO
Cyclic RGD conjugated MPIO were prepared using 1-μm diameter Dynabead MyOne carboxylic acid MPIO (65011, Fisher Scientific, UK). MPIO were washed in MES buffer and resuspended, then incubated with EDC in a shaking platform. MPIO were washed with MES buffer, then incubated for 24 h with 10 mM cyclic RGD or control scrambled Arg-Asp-Gly (RDG) peptides ( Figure S1). The MPIO were then washed and stored in phosphate-buffered saline (PBS) at 4 C on a shaking platform.
Antibody-conjugated MPIO were prepared as described by Zarghami et al. 33  2.2 | In vitro binding of cyclic RGD peptides to mouse integrin α v β 3 The binding efficiency of various RGD peptides has been systematically measured previously for human integrin α v β 3 , 36 but not for mouse integrin α v β 3 . Thus, we tested the binding of three different RGD cyclic peptides, and one control scrambled RDG peptide (structures shown in Figure S1), to mouse integrin α v β 3 . These peptides were conjugated separately to MPIO, as described above.
To measure the binding of RGD-MPIO to mouse integrin α v β 3 under flow conditions, glass capillaries were mounted onto a narrow-gauge syringe with dry-cure glue. First, each glass capillary was filled with a mixture of o-Xylene, Triethoxy-(3-glycidyloxypropyl)silane (TEGOPS) and N,N-Diisopropylethylamine (DIPEA; ratio 1000:320:10), incubated for 17 h at 80 C with phosphorus pentoxide, then washed with o-xylene and methanol, and dried for 4 h at room temperature under vacuum conditions. Subsequently, capillaries were filled with 200 ng/mL mouse integrin α v β 3 (7889-AV-050, R&D Systems, Abingdon, UK) and incubated at room temperature for 24 h. The remaining functional groups were quenched with 10 mM ethanolamine for 24 h at room temperature. Control capillaries were filled with 200 ng/mL bovine serum albumin (BSA) and incubated as above. Capillaries were then washed and stored filled with PBS at 4 C.
Functionalised capillaries were mounted under a Nikon TE 2000-U microscope, and RGD-MPIO or control RDG-MPIO were resuspended in 30% sucrose in PBS to mimic the viscosity of blood. The sucrose-MPIO solution was infused through the capillaries, using a syringe pump, at 20 μL/min to mimic hydraulic shear conditions in blood vessels. The volume of solution and infusion time were recorded to calculate the flow rates retrospectively. Photomicrographs were taken across eight separate fields-of-view (FOVs) along the capillary, and the number of bound MPIO per FOV was manually counted.
Blocking experiments were also carried out to test the specificity of binding of RGD-MPIO. HUVEC-C cells were cultured in Media 199 supplemented with 10% fetal calf serum, penicillin, and streptomycin (100 μg/mL). Once confluent, cells were stimulated to express integrin α v β 3 by stimulation for 20 h with S-nitroso-n-acetylpenicillamine (SNAP). Blocking was carried out by adding RGD or control RDG peptide to the media to a final concentration of 10 μM, and incubating for 1 h. Media were removed, cells rinsed with PBS, and fresh media added. One microgram of RGD-MPIO or control RDG-MPIO was then added to each well, incubated in a shaking incubator for 40 min at 37 C and 2500 rpm. Cells were rinsed five times with PBS to remove any unbound MPIO, then fixed with 4% paraformaldehyde. Images of 10 FOVs per condition were taken using an Olympus IX-71 inverted microscope at 40 Â magnification, and bound MPIO counted by an observer blinded to condition. To study the angiogenic vasculature of brain metastases, an intracerebrally implanted mouse model of breast cancer brain macrometastasis was used. 4T1 tumours naturally metastasise to the brain when a primary tumour is implanted into the mammary fat pad, 38,39 and brain metastases can also be induced haematogenously by intracardiac or intracarotid injection of 4T1 cells, which is considered to be a close representation of the natural route of seeding to the brain. However, intracardiac injection also seeds 4T1 cells to lung, liver, and bone, and mice become moribund before macroscopic metastases develop within the brain. Consequently, to study larger, neo-vascularised brain metastases, the intracerebral route of induction is necessary. We have previously demonstrated that intracerebral injection of metastatic breast cancer cells leads to tumour growth that is very similar to that found when metastases are induced via intracardiac injection, as well as to human brain metastasis growth. 40 Female BALB/c mice (6-10 weeks old) were anaesthetised with 3% isoflurane in 30% O 2 : 70% N 2 O and focally microinjected with 5000 4T1-GFP metastatic murine mammary carcinoma cells 41 into the left striatum. Mice were randomly allocated to timepoint and treatment groups. Following imaging, mice underwent transcardial perfusion-fixation at weekly timepoints 1-5 weeks after tumour induction; and the brains were collected for histology.

| Animal models
Brains were cryoprotected in 30% w/v sucrose, frozen in isopentane and dry ice, and 10-μm cryosections were cut through the striatum.
To induce focal blood-brain barrier (BBB) breakdown in one cerebral hemisphere, female BALB/c mice were anaesthetised and injected as described above with 1 μg of cytokine-induced neutrophil chemoattractant 1 (CINC-1) protein (453-KC-050/CF, R&D systems) in 1 μL of saline in the left striatum, as described previously. 42 Mice underwent imaging 5 h after injection. Following imaging, mice were transcardially perfusionfixed, and brains were collected and sectioned as above.

| Magnetic resonance imaging
4T1-GFP brain macrometastases were induced as described above, and each mouse was imaged once using RGD-MPIO at Days 7 (n = Experiments were performed using a 9.4-T Agilent DDR MRI spectrometer (Agilent Technologies, Santa Clara, CA, USA) with a 160 mm horizontal bore, and an internal diameter 100 mm/outer diameter 155 mm shim and gradient system (Agilent Technologies, Santa Clara, CA, USA) with a maximum strength of 400 mT/m. A 26 mm internal diameter volume transmit-receive birdcage RF coil (Rapid Biomedical GmbH, Rimpar, Germany). Mice were anaesthetised with 1%-3% isoflurane in 30% O 2 :70% N 2 and positioned in the volume coil using a homebuilt cradle. Mouse temperature was monitored and maintained at 37 ± 0.5 C with a rectal probe and heating blanket feedback system. 43 Respiration was monitored using a pressure balloon. Following this, 30 μL of gadodiamide (Omniscan, GE Healthcare, UK) was administered via the tail vein, and a post-gadolinium T 1 -weighted SEMS scan was acquired, as described previously, 5 min after gadolinium administration to assess BBB breakdown.

| MRI data analysis
The images were reconstructed for individual echoes of MGE3D using the square root of sum of squares (SqrtSOS) algorithm by an in-house MATLAB script. A segmentation mask for the striatum and any adjacent tumour-bearing areas in the ipsilateral striatum, along with a contralateral striatum mask, was created for each mouse using ITK-SNAP software. 44 For the contralateral (normal) striatum, the mean signal intensity and standard deviation were calculated in MATLAB. Any voxel with a signal intensity of more than three standard deviations below the mean of the contralateral striatum was designated as hypointense. The total number of hypointense voxels in both the tumour-bearing and contralateral striatum was recorded. As the tumours contained naturally hypointense areas in the pre-MPIO images, all data are reported as the difference in hypointense voxels between the post-contrast and pre-contrast scans (i.e., MPIO-induced hypointense voxels). In a number of tumours from both targeted and control groups (n = 3 RGD-MPIO, n = 5 control RDG-MPIO), the hypointense voxels in the tumour-bearing striatum of the pre-contrast image appeared to be greater than those in the post-contrast image. As a result, an apparently negative difference in postcontrast minus pre-contrast hypointense voxels was found. Images were inspected to ensure no artefacts were present that would affect image quality, or any differences in SNR. In the absence of any technical basis for this observation, it was determined that MPIO-induced hypointense voxels in these mice fall within the noise range of this technique (47 ± 31 voxels, $0.2% of contralateral striatum volume). Consequently, for further analysis, those animals were assigned a value of zero MPIO-induced hypointense voxels rather than a negative value, which is not possible.
Microvessel density (CD31 positive area fraction) was quantified for the area encompassing the metastatic foci in the injected striatum, and also for the contralateral striatum, as the percentage of area covered using the 'Positive Pixel Count 2004-08-11' algorithm in ImageScope (Leica Biosystems). The parameters used were moderately stained pixel intensity, between 202 and 185, and strongly stained pixel intensity, less than 10.

| Perls' Prussian blue staining
Perls' Prussian blue staining was carried out on brain tissue sections to visualise ferric (+3) iron, including MPIO. Sections were washed in distilled water, then incubated in equal parts 20 mg/mL potassium ferrocyanide and 2% v/v HCl in distilled water for 15 min. Sections were then washed in distilled water, followed by counterstaining with nuclear fast red for 2 min. Sections were then dehydrated and mounted with coverslips.
Where double staining for molecular markers and Perls' Prussian blue was performed, the immunohistochemistry protocol described above was carried out to the point of visualisation with DAB, and the Prussian blue staining protocol was then applied as above. Following Prussian blue staining, sections were either counterstained with nuclear fast red (Iba-1, VCAM-1) or proceeded directly to dehydration and coverslip mounting (CD31). To analyse receptor-specific uptake in RAW264.7 macrophages, cells were seeded onto six-well plates, as above. Cells were pretreated for 1 h with 0.1, 1 or 10 μL of 10 mM RGD or control RDG peptide; 0.1, 1 or 10 μg VCAM-1 or IgG antibody. Following this pretreatment, 1 μg of the same MPIO as the blocking agent was administered. Cells were incubated, washed, fixed, and examined by microscope, as above.

| Endogenous iron-laden macrophages
To determine whether endogenous iron-laden macrophages were present in the 4T1-GFP brain macrometastasis model across the tumour timecourse, additional cohorts of mice were sacrificed at Days 7 (n = 4), 14 (n = 4), 21 (n = 2), 28 (n = 4) or 35 (n = 3) and the brains were taken for immunohistochemical analysis. These animals had no prior exposure to iron oxide contrast agents.
A further cohort of animals underwent MRI at a single timepoint, Day 28 (n = 6), to determine whether endogenous iron in tumourassociated macrophages produces nonspecific hypointense voxels in precontrast images. A T 2 *-weighted MGE3D dataset was acquired, but no MPIO administered, and a gadolinium-enhanced T 1 -weighted scan was also performed. Either single staining with Prussian blue or double staining for Iba-1 and Prussian blue was carried out on brain tissue sections, as described above, and iron-laden macrophages manually counted.

| Statistical analysis
Statistical analysis was carried out using Prism version 8.1.2. Normally distributed groupwise data were analysed using one-way ANOVA with Holm-Sidak's post-hoc multiple comparison tests. Where only two groups exist, an unpaired Student's t-test with Welch's correction was used.
Groupwise data that were not normally distributed were analysed using a Kruskal-Wallis one-way ANOVA with Dunn's post-hoc multiple comparison tests. Paired groupwise data (i.e., comparing ipsilateral and contralateral values across timepoints) were analysed using two-way repeated measures ANOVA with Sidak's multiple comparison tests.
Correlations were analysed using the Spearman r test. Linear regression lines were compared using the sum-of-squares F-test. To determine the binding efficacy of various RGD peptides to the mouse integrin α v β 3 , the number of RGD-MPIO binding events in α v β 3 or control (BSA-coated) glass capillaries was compared under flow conditions ( Figure S2). Comparison of binding events between α v β 3 and BSA capillaries showed an effect of both capillary (two-way ANOVA, p < 0.0001) and cyclic RGD (two-way ANOVA, p < 0.001). Tukey's multiple comparison post-hoc tests showed significant differences between BSA-and α v β 3 -coated capillaries for the RGD-1 ( p < 0.05), RGD-2 ( p < 0.001), and RGD-3 ( p < 0.001) cyclic RGD peptides ( Figure S2). The control RDG peptide showed no significant difference in binding to BSA-or α v β 3 -coated capillaries ( Figure S2). For the α v β 3 capillary, the RGD-2 peptide showed significantly higher binding than both the control RDG peptide (one-way ANOVA, p < 0.005; Bonferroni's multiple comparison post-hoc test, p < 0.001) and the other RGD peptides, RGD-1 (post-hoc test, p < 0.001) and RGD-3 (post-hoc test, p < 0.05). Thus, peptide RGD-2 was used for further experiments with RGD-MPIO, and peptide RDG was used as a negative control scrambled peptide for RDG-MPIO.
In blocking experiments, to assess the specificity of RGD-MPIO binding to α v β 3 on endothelial cells in culture ( Figure S3), significantly higher binding was observed for RGD-MPIO to SNAP-stimulated HUVEC cells than either unstimulated cells (Welch's ANOVA with Dunnett's multiple comparison test, p < 0.001) or cells pretreated with an RGD block ( p < 0.01). No differences were found for control RDG-MPIO binding to stimulated cells, with or without an RGD block, compared with unstimulated cells. Together, the above findings demonstrate the specificity of RGD-MPIO binding to α v β 3 .
3.2 | Assessment of angiogenesis and α v β 3 expression in a mouse model of brain macrometastasis For assessment of vessel density, correlation analysis showed that both the CD31-stained area and number of CD31 positive vessels increased more rapidly over time in the tumour-bearing striatum than in the contralateral striatum (sum-of-squares F-test, p < 0.05 for both; Figure 1A,B).

| Imaging angiogenesis in vivo using RGD-MPIO
Having demonstrated the presence of α v β 3 positive angiogenic vessels in this model of brain macrometastasis, the ability of the chosen RGD-MPIO to detect these vessels in vivo was assessed across a timecourse. Qualitatively, increasing numbers of hypointensities were apparent in the tumour-bearing striatum on T 2 *-weighted images following intravenous RGD-MPIO injection across the timecourse studied ( Figure S4), with 80% of mice showing MPIO-induced hypointensities by Day 35 (Figure 2Ai-ii). By contrast, few hypointensities were observed in mice injected with RDG-MPIO at the early timepoints (Days 7-21; Figure S5), but became more evident at Days 28 and 35 (57% of mice; Figures 2Aiv and S5). Some hypointense voxels were also apparent in the precontrast T 2 *-weighted MGE3D images in the tumour-bearing striatum (Figures 2A, S4, and S5).
Quantitative analysis of MPIO-induced hypointensities in mice injected with RGD-MPIO showed significantly more in the tumour-bearing versus contralateral striatum at Day 35 (two-way paired ANOVA, p < 0.05; post-hoc Sidak's test, p < 0.05; Figure 2B). However, in mice injected with the control RDG-MPIO, a significant difference was also observed between the tumour-bearing and contralateral striatum at Day 35 (twoway paired ANOVA, p < 0.05; post-hoc Sidak's test, p < 0.05; Figure 2C). In mice receiving the PBS control injection (no tumour), very few hypointense voxels were detected in mice receiving MPIO in either the ipsilateral or the contralateral striatum ( Figure 2B,C). At face value, these findings suggest that nonspecific retention of MPIO occurred in tumour-bearing mice. However, tumour growth and progression were found to be variable over the timecourse, with some tumours at the later timepoints showing marked BBB breakdown and gadolinium enhancement ( Figure 2Aii, iv), while others had an intact BBB (Figure 2Ai, iii). Consequently, to determine why significant uptake of control RDG-MPIO was observed at the Day 35 timepoint, gadolinium-enhancing and nonenhancing tumours were separated and pooled across timepoints for further analysis (Table S1).
In the pooled group of nonenhancing tumours (n = 7 RGD-MPIO, n = 13 control RDG-MPIO), mice injected with RGD-MPIO showed significantly increased MPIO-induced hypointense voxels in the tumour-bearing hemisphere (one-way ANOVA, p < 0.005; Figure 2D Figure 2E). Moreover, the number of hypointense voxels in the control RDG-MPIO group was greater than the F I G U R E 1 (A) Quantitation of CD31 total stained area as a percentage of tumour area in the tumour (blue circles) and the contralateral striatum (red squares), plotted against time with 95% confidence intervals. A significant increase in blood vessels in the tumour is evident after Day 12 (linear regression sum-of-squares F-test, p < 0.05; Spearman's r, tumour regression: 0.75, p < 0.0001; contralateral striatum regression: 0.74, p < 0.0001). (B) Comparison of microvessel density (CD31 positive vessels) in the tumour (blue circles) and the contralateral striatum (red squares), plotted against time with 95% confidence intervals. A significant increase in blood vessels in the tumour is evident after Day 13 (linear regression sum-of-squares F-test, p < 0.05; Spearman's r, tumour regression: 0.56, p < 0.01; contralateral striatum: 0.51, p < 0.05).  4). Data for these figures had a lognormal distribution, and no outliers were found to be present (ROUT test). Bars represent mean ± standard deviation; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. PBS, phosphate-buffered saline. group given RGD-MPIO (Holm-Sidak's post-hoc test, p < 0.001). In both groups the number of hypointensities observed was considerably greater than that for the nonenhancing tumours. Quantitation of single MPIO associated with blood vessels showed significantly more RGD-MPIO bound per mm 2 of endothelium than control RDG-MPIO (t-test, p < 0.01; Figure 3J). Double staining for CD31 and Prussian blue indicated that more than 90% of iron-laden macrophages were within 25 μm of the blood vessel lumen, with a median distance of 10 μm (Figure 3K), suggesting that they are likely to be perivascular macrophages.

| Effect of MPIO targeting moiety
Because retention of MPIO by perivascular macrophages has not previously been observed in studies using antibody-targeted MPIO to study brain pathology, 33,47,49-51 we next sought to determine whether peptide-targeted MPIO are taken up by perivascular macrophages preferentially and could explain, at least in part, the observed iron aggregations in the current study. showed significantly increased MPIO-induced hypointense voxels in the tumour-bearing hemisphere (one-way ANOVA, p < 0.005) than both the contralateral hemisphere and the mice administered control RDG-MPIO (black bars; n = 13) in the tumourbearing hemisphere. (E) In mice with gadolinium-enhancing tumours, significantly increased MPIO-induced hypointense voxels were observed in the tumour-bearing striatum compared with the contralateral striatum for both RGD-MPIO (white bars; n = 10) and RDG-MPIO (black bars; n = 3) (one-way ANOVA, p < 0.0001). However, in mice receiving control RDG-MPIO, the volume of MPIO-induced hypointense voxels was also significantly greater than those administered RGD-MPIO. Number of MPIO-induced hypointense voxels, presented as postcontrast minus precontrast hypointense voxels for all data. Bars represent mean ± standard deviation; post-hoc Holm-Sidak's tests. *p < 0.05, ***p < 0.001. MGE3D, multigradient echo three-dimensional; MPIO, microparticles of iron oxide; RDG, Arg-Asp-Gly peptide, scrambled control; RGD, Arg-Gly-Asp peptide, targeting integrin α v β 3 .
or after 1 h of blocking with VCAM-1 or IgG antibodies for VCAM-MPIO and IgG-MPIO, respectively ( Figure S6E,F). These results indicate that uptake by macrophages in culture is neither receptor nor ligand-type dependent.
Next, accumulation of antibody-conjugated MPIO in perivascular macrophages in vivo was assessed in 4T1-GFP macrometastases at Day 28 ( Figure 4A). Mice were injected with either targeted VCAM-MPIO or control IgG-MPIO ( Figure 4B,C). VCAM-1 has previously been shown to be upregulated on the vascular endothelium in close association with brain metastases. 47,49,52,53 In the mice receiving VCAM-MPIO, MPIOinduced hypointense voxels were observed in the tumour-bearing, but not the contralateral, striatum ( Figure 4D). Within this cohort of mice a significant variation in tumour size was evident, with one animal exhibiting a relatively small, nonenhancing tumour (0.5 mm 3 ), while the other two had much larger, contrast-enhancing tumours (5.5 and 18.2 mm 3 , respectively). In the mouse with the smaller tumour, only a relatively small number of hypointense voxels was detected, resulting in a large spread in effect size across the group ( Figure 4D) and, consequently, the difference between hemispheres did not reach statistical significance. Nevertheless, it is clear that all of the animals injected with VCAM-MPIO showed more hypointense voxels in the tumour-bearing hemisphere than in the contralateral hemisphere. Numbers of hypointense voxels were considerably higher than observed following RGD-MPIO administration. In mice receiving the control IgG-MPIO, no MPIO-induced hypointense voxels were observed in either striatum, despite the presence of large (2.9 and 8.6 mm 3 , respectively) contrast-enhancing tumours ( Figure 4D). section), with no significant differences between groups. These numbers of iron-laden macrophages were also not significantly different to those observed at the same timepoint in mice injected with peptide-targeted MPIO (17.8 ± 16.2 per tissue section; Figure S7). Moreover, further analysis indicated that all such macrophages were located in areas exhibiting hypointensity on T 2 *-weighted images prior to MPIO injection ( Figure 4B). This observation, together with the lack of MPIO-induced hypointensities observed in IgG-MPIO injected mice, suggests that these iron-laden macrophages do not contain MPIO, but rather endogenous iron. Double staining for Prussian blue and VCAM-1 revealed that, in addition to iron-laden macrophages, single MPIO were present bound to VCAM-1 positive blood vessels in mice imaged with VCAM-MPIO ( Figure 4F), but not in mice imaged with control IgG-MPIO.

| Endogenous iron-laden macrophages
The above findings led us to investigate whether the larger iron aggregations observed were indeed endogenous iron rather than MPIO, because they correlated spatially with pre-existing hypointensities. Thus, mice bearing 4T1-GFP tumours, but with no exposure to iron oxide contrast agents, were assessed for the presence of endogenous iron-laden macrophages. Low levels of endogenous iron-laden macrophages were observed in the tumour tissue across all timepoints ( Figures 5A,B and S7) and were histologically indistinguishable from the iron-laden macrophages observed in the previous studies; no statistically significant differences were evident between timepoints. When endogenous iron-laden macrophage levels were compared with tumour size, no significant correlation was found ( p > 0.2, Pearson's coefficient R 2 = 0.08; Figure S8A). T 2 *-weighted images from 4T1-GFP tumours at the Day 28 timepoint ( Figure 5C) were compared with Prussian blue-stained histological sections. The tumours showed varying levels of pre-existing hypointense voxels in the tumour-bearing striatum (720 ± 570; Figure S8B). By contrast, the contralateral striatum exhibited very few pre-existing hypointense voxels (57 ± 21; Figure S8B), and a significant difference was evident between hemispheres (paired t-test, p < 0.05). The levels of pre-existing hypointense voxels in this study are similar to the levels of precontrast hypointense voxels observed in the 4T1-GFP-bearing animals subsequently imaged with RGD-MPIO or control RDG-MPIO. Spatially, the endogenous iron-laden macrophages correlated closely with pre-existing hypointense voxels in the T 2 *-weighted images ( Figure 5C,D). Double staining for Prussian blue and the macrophage/microglia marker Iba-1 showed colocalisation of iron aggregations and Iba-1 staining ( Figure 5E). These data suggest that there is a baseline level of endogenous iron-laden macrophages that underlie the precontrast hypointensities observed in the above studies.

| MPIO penetration of the BBB
Given that the numbers of iron-laden macrophages did not appear to differ significantly between groups, irrespective of targeting moiety or indeed the presence of MPIO, we next tested the hypothesis that peptide-targeted MPIO are able to simply extravasate across the BBB when frank breakdown is evident. To this end, a CINC-1-induced model of BBB breakdown was used. As nonspecific accumulation of MPIO had not previously been observed in models of brain macrometastasis when targeted with antibodies rather than peptides, 52 both peptide-and antibodytargeted MPIO were again evaluated. BBB breakdown was confirmed in all animals using gadolinium-enhanced imaging ( Figure S9A). In mice across all groups, a small number of hypointense voxels post-MPIO injection was evident in the CINC-1-injected striatum, but not in the contralateral striatum ( Figure S9B,C), although the difference between hemispheres did not reach statistical significance. Histological examination of brain tissue, however, revealed only occasional single MPIO and a small number (fewer than two per section) of iron-laden macrophages ( Figure S6D,E). No differences were found between the different MPIO used. These data suggest that neither peptide-nor antibody-targeted MPIO cross a permeable BBB per se to any significant degree.

| DISCUSSION
In this study, we have demonstrated that expression of the integrin α v β 3 increases over time in this model of brain macrometastasis and appears to correlate with increased vascularity and vessel permeability. These data suggest that integrin α v β 3 has potential as a biomarker of angiogenesis in brain macrometastasis, and led to testing of the α v β 3 -targeted RGD-MPIO contrast agent. However, subsequent findings revealed a number of confounds that may reduce the sensitivity for detection of α v β 3 -specific binding using this peptide-targeted MPIO, including (i) the presence of endogenous iron-laden macrophages, particularly in larger brain metastases, which yield significant baseline hypointensities on T 2 *-weighted images; and (ii) nonspecific retention of peptide-targeted MPIO within the tumour microenvironment. By contrast, antibody-conjugated MPIO, despite specific binding to their endothelial target VCAM-1, showed negligible in vivo uptake by tumour macrophages and we propose, therefore, that the use of antibodies rather than peptides as the targeting ligand may represent a preferable route forward for new angiogenesis-targeted agents.
This study was a proof-of-concept study aiming to determine whether RGD-MPIO could detect expression of integrin α v β 3 associated with tumour angiogenesis in vivo. In the model of directly induced 4T1-GFP brain macrometastasis used, endothelial α v β 3 expression was found across the timecourse studied, but increased significantly at later timepoints together with increased microvessel density. Consequently, this was considered to be a suitable model to assess imaging of α v β 3 as a biomarker of tumour angiogenesis in brain macrometastasis. Increased microvessel density and integrin α v β 3 expression are both features of human breast cancer brain metastasis. It has been shown in both clinical samples 54 and preclinical studies 55 that breast cancer brain metastases have a higher microvessel density than their respective primary tumours. Breast cancer brain metastases have also been shown to express integrin α v β 3 , 56,57 so it is likely that if a successful integrin α v β 3 imaging agent is developed it will be clinically relevant.
In our study, T 2 *-weighted imaging showed that pre-existing hypointensities could be observed prior to MPIO injection. Subsequent investigation of mice with tumours, but no exposure to iron oxide contrast agents, revealed endogenous iron-laden macrophages were present in the tumour tissue throughout the tumour timecourse, which correlated spatially with the pre-existing MRI hypointensities. This finding is supported by the results of Leftin et al., who have previously used T 2 *-weighted imaging to quantify macrophages in a mouse model of brain metastasis. 58 Although the presence of iron-laden macrophages has not been observed in the micrometastatic stages of tumour development in the brain (< 500 μm diameter), it is likely that endogenous iron-laden macrophages will be present at some level in most, if not all, brain tumours once they are more established. Thus, for any study using iron oxide-based contrast agents, it will be important to quantify any pre-existing hypointense voxels prior to administration of contrast.
Nevertheless, despite the presence of natural hypointensities, specific α v β 3 -targeted binding and MPIO-induced hypointensities could be observed with RGD-MPIO when only non-enhancing tumours were considered, with no retention of the control RDG-MPIO. Thus, while the background endogenous iron may reduce the sensitivity of this method, it does not necessarily preclude it, and imaging of α v β 3 positive angiogenic vessels may be possible in brain tumours if confounding factors can be eliminated. Moreover, this approach may be combined with complementary methods, such as arterial spin labelling and dynamic contrast-enhanced MRI, to provide additional functional information on angiogenesis. 59 In gadolinium-enhancing tumours, however, retention of both the RGD-MPIO and the control RDG-MPIO was observed, resulting in large volumes of MPIO-induced hypointensities. This was an unexpected result, as the control RDG-MPIO had shown only very low levels of binding in vitro both here and in a previous study. 35 These data appeared to suggest that the MPIO, irrespective of targeting peptide, were nonspecifically retained within the brain in the presence of frank BBB breakdown (as detected by gadolinium enhancement). Analysis of the histology from these animals showed few control RDG-MPIO bound to the endothelium, indicating that the MPIO-induced hypointensities in these animals do not result from binding to α v β 3 . Further histological analysis showed large numbers of iron-laden macrophages and/or microglia, closely associated with blood vessels, leading to the suggestion that the MPIO may be taken up by perivascular macrophages. While the majority of iron detected was visible inside macrophages, single endothelium-bound MPIO were also present in mice injected with RGD-MPIO, with significantly more endothelium-bound MPIO present compared with those given control RDG-MPIO. In accord with the above findings in non-enhancing tumours, these data support the notion that specific binding of RGD-MPIO to angiogenic blood vessels is present.
Anecdotally, unconjugated MPIO have also been found to be retained in tumour-bearing striatum, suggesting that the nonspecific uptake of MPIO is specific to tumour. However, caution is required in the interpretation of these results, because only a single mouse was tested and the surface of the unconjugated MPIO has different properties and charge to the peptide-and antibody-conjugated MPIO. Nonspecific uptake of MPIO has not been previously observed in our experience using antibody-targeted MPIO. Therefore, antibody-conjugated MPIO targeting endothelial VCAM-1 were also tested in the 4T1-GFP model to determine whether MPIO with different targeting moieties behave differently in vivo.
MPIO-induced hypointense voxels were observed in the tumour-bearing striatum in mice imaged with VCAM-MPIO, but not in the contralateral striatum, nor in mice imaged with control IgG-MPIO. Further, histological analysis showed marked endothelial VCAM-1 expression at the margins of the tumours, and single MPIO bound to VCAM-1 positive blood vessels. These data are in accord with previous studies in the micrometastatic stages of the 4T1-GFP model 47,49 and other macrometastasis models, 52 showing that antibody-conjugated MPIO are capable of sensitive and specific imaging of VCAM-1 positive blood vessels. However, neither of the two animals imaged with IgG-MPIO showed any MRI-detectable hypointensities or macrophage uptake by immunohistochemistry, in line with our previous studies. Nor was macrophage uptake of MPIO evident in any of the VCAM-MPIO-injected animals. Thus, across a total cohort of five mice injected with antibody-conjugated MPIO, no nonspecific hypointensities were observed and no macrophage uptake was evident, despite the presence of large, contrast-enhancing tumours in four out of five animals. These data support the conclusion that there is no nonspecific retention of antibody-targeted MPIO. These findings are in contrast to the peptide-targeted MPIO findings discussed above, where animals with large, contrast-enhancing tumours showed significant nonspecific retention of both RGD-MPIO and control RDG-MPIO.
Together, the above findings suggest that peptide-targeted MPIO, but not antibody-targeted MPIO, are nonspecifically retained within the tumour microenvironment, and that this is particularly evident when the BBB is compromised. Given the presence of iron-laden macrophages observed in tumour tissue, we first assessed whether the numbers of these increased in animals injected with peptide-targeted MPIO compared with either those given antibody-targeted MPIO or animals with no exposure to MPIO. We found, however, that late-stage tumours (Day 28-35 timepoints) had similar levels of Prussian blue-stained macrophages, regardless of the targeting moiety, and similar levels were also present in tumours with no MPIO exposure. These findings indicated that there is no increase in the absolute number of iron-laden macrophages in the peptide-MPIO-injected animals specifically.
As the MPIO-induced hypointensities were only observed in tumours with frank BBB breakdown, we next considered whether peptidetargeted MPIO had simply extravasated into the parenchyma. It is generally believed that MPIO with a diameter of 1 μm are unable to cross the BBB as they are too large to extravasate from the circulation, and are used in preference to iron oxide nanoparticles for this reason. To directly test this assumption, we used a model of CINC-1-induced BBB breakdown. In accord with the above premise, however, only occasional single MPIO and iron-laden macrophages were observed in the CINC-1-injected hemisphere, indicating that neither peptide-nor antibody-targeted MPIO cross a permeable BBB per se to any significant degree.
A previous clinical study from Iv et al. showed a correlation between macrophage density in high grade glioma with R2* measurements following ferumoxytol administration, 60 suggesting that the nonspecific retention of peptide-targeted MPIO in the 4T1-GFP model may be related to the tumour microenvironment, rather than BBB breakdown per se, and that factors within the tumour microenvironment may prime macrophages/microglia to take up MPIO across a compromised BBB. Studies by Leftin et al. demonstrated that iron-laden macrophages in breast cancer brain metastases are pro-inflammatory polarised specifically. 58 Because pro-inflammatory macrophages exhibit increased phagocytic activity, we hypothesise that pro-inflammatory macrophages within the tumour microenvironment, that are already iron-laden and have access to the blood through close association with the BBB, may be predisposed to take up peptide-targeted MPIO, and to a greater degree when the BBB is compromised. Such uptake would lead to an increase in the volume of observed hypointensities, and could explain the nonspecific hypointensities seen when using control RDG-MPIO. Currently, endogenous iron and MPIO cannot be distinguished using Prussian blue histology. Future studies may be able to use fluorescently or otherwise labelled beads to distinguish between endogenous iron and MPIO.
Nonspecific MPIO-induced hypointensities have not been observed in any previous studies using MPIO to study brain pathology. 33,47,[49][50][51]61 However, all previous studies have used an antibody as the targeting moiety compared with the peptides used in this study. It is unclear why peptide-rather than antibody-targeted MPIO may be more readily phagocytosed by perivascular macrophages, leading to nonspecific contrast effects, but our current data suggest that antibody targeting may be preferable to peptide targeting under these circumstances. At the time of study design, no antibodies targeted to mouse integrin α v β 3 and suitable for in vivo use were commercially available. Therefore, a direct comparison of peptide and antibody targeting moieties was not possible in this study, but warrants further study in the future.

| CONCLUSIONS
While specific binding of the α v β 3 -targeted molecular imaging agent RGD-MPIO was observed, a number of confounds, including pre-existing hypointensities and nonspecific retention of peptide-targeted MPIO, need to be overcome before this approach can progress. Little evidence of iron retention in the brain parenchyma, either in macrophages or as bound MPIO, was found in areas of frank BBB breakdown induced by CINC-1 in the absence of a tumour. We propose, therefore, that the nonspecific retention of peptide-targeted MPIO is related to uptake by primed perivascular macrophages/microglia within the tumour environment, rather than BBB breakdown per se. While low levels of endogenous iron-laden macrophages may reduce the sensitivity of the molecular MRI results, precontrast imaging would enable differentiation between endogenous macrophages and specific contrast. Moreover, antibody-conjugated MPIO enable sensitive and specific imaging of endothelial targets, with no apparent nonspecific retention, and thus show significant advantages over peptide-targeted MPIO. These findings have important implications for the emerging field of molecular MRI, suggesting that antibodies rather than peptides may be preferable as the targeting ligand.